Over the past few weeks there have been a few PTD developments, though none have been quite big enough to warrant a substack post (besides the surprise consultations from Spain and Portugal - more here). I do try to avoid swamping your inbox.

So here’s a quick round-up instead. Five PTD stories you may have missed, moving from the broadly relevant and more likely to be of wider interest, to the more niche and “probably only for the die hards” stories.

Let’s dig in.

Story 1: EU Commission Publishes PTD FAQ Guidance

Back in July, the Commission published an FAQ on the PTD.

Guidance is usually the flesh on the bones of legislation. It clarifies what things actually mean in practice and can help make new laws workable.

The Commission’s FAQs guidance sits in a bit of no man’s land. The PTD has to be implemented, and doesn’t take effect until it has been (besides in some particular circumstances - more on that here).

Given this, the FAQs shouldn’t be relied on as conclusive and determinative evidence of what’s right and what’s not. But could be referred to as something to help work out how to interpret local laws, especially in the absence of any local guidance to the contrary.

So, now that we’ve set the scene and provided a bit of context, let’s take a look at some of the more useful points from the FAQs.

The potentially useful stuff

A bit question employers have been grappling with - especially in the face of the wide definition of pay in the PTD - is what pay elements to include.

The FAQs say that universal or unconditional benefits (lunch vouchers and company phones, for example), and statutory social benefits, are generally outside the definition. So if something is made available to everyone, the Commission says to ignore it. Finally, a bit of clarity.

And there’s also a bit of useful guidance on what to do with GDPR and the small groups problem - how do you comply with requests when personal data might be revealed. The Commission says Member States may restrict direct access, meaning that any request can instead go through labour inspectorates, equality bodies or worker representatives.

Finally, there is some commentary on market rates. the FAQs confirm they are not a magic wand. They can only justify a pay difference where the employer can show that the factor is objective, gender-neutral, proportionate and used consistently. The PTD switches the pay setting process from “what is the market rate for this job” to “what is the rate we pay for roles at this level? Now are there legitimate and proportionate market based reasons why we should pay more”.

Story 2: France’s Pay Transparency Bill goes to Conseil des Ministres on 9 September

Next up: France.

France has finally put a date on it. At the CFDT conference, Minister of Labour Jean-Pierre Farandou said the pay transparency bill will go to the Conseil des ministres on 9 September 2026.

That gives use a bit more of an idea of the timeline for France implementing. The parliamentary debate is expected to start in the second half of 2026, with a hope for a vote before the presidential election (in April).

There had been suggestions of France implementing in January 2027, but now that looks less and less likely.

Story 3: Bulgaria’s response to its PTD Consultation

Bulgaria’s Ministry of Labour and Social Policy has now published its full response to the public consultation on amendments to the Protection against Discrimination Act (this is the bill that will transpose the PTD). I covered the original draft in a previous post. That draft is now going to change and the consultation sets out the direction of travel (but a new draft is yet to be published).

The Ministry did move on several points. On categorisation, Several respondents pointed out that the draft said “complexity” and “difficulty” where the Directive says “skills” and “effort.” The Ministry accepted the point and will bring the terminology into line to match the PTD. Workers representatives will also be given clearer roles, with clearer rules around Joint Pay Assessments.

The Ministry did not move on everything though and held the line on a few important points including no broadening of the definition of pay, no delays or transition periods, and no expansion to reporting for smaller employers (under 100 employees).

Bulgaria response to PTD consultation

1.15MB ∙ PDF file

Download

Download

Story 4: Moldova’s new PTD transposition, with 1 January 2027 implementation

Moldova isn’t even in the EU, yet it’s doing the work to earn its place, and the Ministry of Labour and Social Protection has published draft law (source here).

Moldova joins Montenegro as one of a small but growing number of non-EU countries proactively aligning with the PTD as part of EU accession processes.

Moldova’s plans are to implement the PTD faithfully, with little in the way of creative solutions to its challenges. It goes down the route of state run calculations (like Lithuania), rather than employee run (like Bulgaria).

Story 5: Iceland’s new pay gap law is interesting but isn’t PTD implementation

Iceland’s new law is now in force. Iceland’s framework instead rests on equal pay certification, which has been in place since 2018 and is one of the world’s most advanced pay equity frameworks. The new law expands and updates that system rather than throwing it out and becoming PTD aligned.

The two regimes part company in a few important places:

• Certification rather than disclosure. Iceland relies on equal-pay certification audited by a third party; the PTD takes a transparency-and-reporting route.

• Recruitment rules may be missing. No bans on asking about pay history. No requirement for publication of pay ranges.

Will Iceland eventually implement the PTD?

Maybe.

Although Iceland is not in the EU, it is in the EEA. EEA countries will (probably) have to implement the PTD eventually, though the process is slower as it has to be incorporated into the EEA Agreement. And, just a few days ago, Iceland rejected a plan to restart EU accession negotiations.

Iceland has been ahead on pay equity for years, yet it is choosing not to align with the PTD - at least for now. Its certification-based model may be more rigorous in some respects, but the philosophy is different: Iceland starts with certification, while the PTD starts with transparency.

Iceland still wants to get to the same place - equal pay for men and women - but, arguably, this new legislation suggests it doesn’t think the PTD is the best way of getting there.

Tom’s thoughts

So, what now?

The pace will start to pick up. It’s September. Politicians and lawmakers are back from summer recesses and holidays, and the pressure to implement continues.

My guess: we’ll start to see substantive movement in those places where we’ve seen little/no movement so far. And those places with drafts will push hard to get them finalised.

I think Spain and Portugal will be the next big economies to finalise their laws over the next couple of months, and I’m expecting we’ll see at least a draft from Germany before Christmas.

I went on holiday last week. I figured August must be a good month to go, without missing out on big PTD updates.

How wrong I was.

Busting the myth that all of Europe closes in August, both Spain AND Portugal came out with big developments: drafts from both countries.

Let’s dig in.

Subscribe now

Spain and Portugal both publish drafts

Spain published its draft Royal Decree on 3 August. Then Portugal published its draft proposal for a law two days later.

Both are still drafts, both are open for consultation, and both shouldn’t be treated as the final answer. But they’re both worth reading now because they give a really big indication on what things are likely to look like:

• Spain is, broadly, trying to build the new requirements into its existing pay transparency machinery (it has some quite detailed obligations around job evaluations already).

• Portugal is going down a different track and (like a few other countries such as Lithuania) by putting a lot of emphasis on what happens with the pay gap data: who collects it, who processes it, who can see it, what happens when the numbers don’t look right, and what the employer has to do next.

Short consultation periods suggest desire to “get it done” rather than “done right”

As I said at the start, nothing ever happens in August. Politicians, civil servants, lawmakers (and everyone in between) all tend to go on holiday, so no one was expecting this.

Spain’s draft is open to public consultation until 24 August - a pretty short turnaround for a major piece of legislation. Although the Spanish Government had already carried out an earlier public consultation between 24 April and 8 May, that wasn’t in relation to specific proposals. Things have moved on a bit now and there is more meat on the bones.

Portugal takes a similar approach. Their draft was published in the Boletim do Trabalho e Emprego on 5 August, with interested parties having just 20 days to comment. The Government is also asking the Assembleia da República to give the proposal priority and urgency.

So while both countries are *technically* carrying out a public consultation, in practice the scope of feedback might be pretty limited. That means final legislation might be harder for the government to get right. Consultations are useful ways for governments to iron out the potential problems in draft legislation. Shorter consultations (especially done in August) mean fewer responses and less feedback.

Arguably, consulting for 20 days in August doesn’t really demonstrate a burning desire to obtain or listen to feedback. It demonstrates a desire to just get things done (especially now the deadline has passed), and less about getting it done right.

Spain: the Pay Transparency Directive meets an existing pay transparency system

Spain is one of the (relatively few) countries coming to the PTD with an existing regime.

Categorisation

Spain isn’t throwing out the baby with the bathwater.

The draft does not require employers to abandon existing classifications or build a new categorisation system from scratch. It uses existing professional groups, categories, pay levels and classification systems as the starting point.

But those categories must be capable of withstanding an equal-value analysis based on objective, gender-neutral criteria, including competencies, effort, responsibility and working conditions (broadly replicating the PTD’s requirements).

The employee information request is one of the more immediate changes

As currently drafted, the right to information will come into force as soon as the legislation is passed. There is no plan for a phased transition like in some other countries. It’ll be limited to one request a year, while employers will need to provide both annual and hourly rate averages.

The range in Spain falls mainly on…

There’s a pretty big gap in the Spanish draft.

There is nothing on the provision of pay ranges in recruitment.

The draft is explicitly a partial transposition, and the impact assessment makes clear that the proposal is concentrating on the elements being dealt with through this regulatory instrument.

So expect another consultation and draft dealing with this point.

Portugal has a different feel

The Portuguese proposal feels much more focused on what happens after the employer has produced the data.

Portugal proposes a data-processing entity which would process remuneration information, retain information covering the previous four years, publish information on pay disparities and proportions of workers, and provide information annually to the relevant statistical authorities.

Portugal is building a machine around the calculations.

Gender pay gap reporting obligations will apply to employers with 50+ employees (eventually)

The final point to discuss is a bit of gold plating from both Spain and Portugal.

The PTD mandates pay gap reporting for employers with 100+ workers. But both Spain and Portugal are going further and proposing reporting for employers with 50 or more.

In Portugal, employers with 250 or more will report annually, while those with 50–249 will report every 3 years. It’ll be phased in (in line with the PTD), but with employers in the 50–99 range coming into the reporting regime from 7 June 2031.

Spain’s proposal similarly extends the reporting regime to employers with 50–99 workers from 2031.

What caught my eye: two countries, two different approaches

What’s interesting here isn’t the 50+ threshold. It wasn’t unexpected (especially in relation to Spain). But it’s the difference in approach.

Spain is trying to make the PTD work within its existing pay transparency framework. Portugal is building more infrastructure around the data and what happens once the numbers are produced.

Both approaches can get you to broadly the same place: implementation of the PTD. But they create very different compliance environments.

The interesting bit is seeing which model other countries end up following. But maybe the most interesting bit will be seeing which works best.

The government has just launched a 15-week consultation on equal pay and pay discrimination. Tucked away within its 58 pages is the UK’s first concrete proposal resembling the EU Pay Transparency Directive: requiring employers to disclose salary information before interview.

But many of the Pay Transparency Directive’s other core measures are missing.

• Ban on asking about salary history? No.

• Right to information? Nope.

• Requirement to publish pay-setting criteria? Still no.

The consultation never mentions the EU Pay Transparency Directive by name.

Instead, it presents these proposals as the next stage of a domestic reform process that began with the government’s 2025 call for evidence. Yet for anyone familiar with Directive 2023/970, the parallels — and the omissions — are difficult to ignore.

In this post, I’ll explain what’s in the consultation, and what might happen next.

(The equal pay reforms relating to race and disability are a significant part of this package, but they sit outside the pay transparency story and I’ve kept them to a single section here).

The story so far: the 2025 call for evidence (CfE)

The Office for Equality and Opportunity launched its call for evidence on equality law last year on 7 April 2025, as one of several evidence-gathering exercises feeding into the draft Equality (Race and Disability) Bill.

Pay transparency was one of nine listed areas, alongside the prevalence of pay discrimination on the basis of race and disability, making the right to equal pay effective for those groups, outsourcing, and the proposed Equal Pay Regulatory and Enforcement Unit.

The call for evidence document was specific about what “pay transparency” could mean in practice. It described measures involving employers providing the specific salary or salary range of a job in the advert or before interview, not asking candidates about their salary history, and publishing or providing employees with information on pay, pay structures and criteria for progression.

It also drew a direct link between these measures: preventing employers from asking about salary history, the document noted, would force them to set their own parameters for pay negotiation, while publishing salary information with any confidence would first require employers to undertake a proper job evaluation process.

In other words, the government’s own “pre-consultation” thinking treated the salary history question and the job evaluation question as connected pieces of the same reform, not separate asks.

What the evidence actually said

The analysis of the call for evidence responses was carried out by Fortia Insight, an independent research consultancy commissioned by government.

Of 176 submissions received, 147 were included in the final analysis after duplicates and ineligible responses were removed, with 59% from organisations, 19% from individuals, 14% from academics or researchers, and 8% classed as other.

On pay transparency specifically, the findings gave the government a reasonably strong evidence base to work with. A minority of respondents from trade unions and academia pointed to evidence that salary history bans, published pay bands and equal pay audits have reduced gender pay gaps elsewhere.

One academic respondent cited a US study finding that where employers were barred from asking about salary history, job changers earned 5–6% more than people in similar roles, with women’s pay rising by 8–9%. The same respondent referenced a separate 2020 study finding that California’s salary history ban narrowed the gender pay gap, with particular benefit to women over 35 and mothers.

On race and disability, evidence was described as more limited, but some respondents supported extending mandatory pay gap reporting, banning salary history inquiries, including pay bands in job adverts, and introducing a “right to information” mechanism, specifically to reduce discrimination on those grounds too. Separate government-commissioned research into pay discrimination also concluded that transparency measures were needed to address what it called the “awareness gap”, and specifically raised the introduction of a right to information as a way of letting workers get comparator information.

The evidence wasn’t entirely one-sided, either. Some respondents, particularly employer representatives, warned that publishing salary bands could increase wage costs, reduce the scope for performance-related incentives, and create internal tension if existing staff felt underpaid by comparison. The government was not looking at a one-sided evidence base, and the consultation’s caution has some grounding in that.

What’s actually being proposed

Against that backdrop, the consultation’s pay transparency chapter is fairly narrow.

The centrepiece is a proposed statutory requirement for all employers to publish pay information in job adverts, or, where there is no advert, to give the candidate that information in writing before interview. The government proposes to introduce this through Regulations, meaning the detail — for example, how wide a salary range needs to be, or whether there would be any sort of “good faith” or reasonableness requirement — would be set out later in regulations rather than fixed now.

The stated aim is to keep the burden on employers as low as possible, avoiding new reporting requirements, and the consultation notes that most employers already publish pay ranges voluntarily, so a statutory floor would mostly be codifying existing good practice rather than demanding something new.

Two supporting measures sit alongside this.

First, a reinstated statutory questionnaire procedure for pay discrimination disputes only, reviving a mechanism that existed under the Equality Act 2010 (until it was repealed in 2013 during the Coalition government war on “red tape”), allowing claimants to ask a respondent standard questions about pay before or during proceedings. The new version would be optional rather than mandatory, though a tribunal could still draw an adverse inference from an evasive answer or a failure to respond, and the government says it wants the questionnaire redesigned to be simpler and more targeted than its predecessor.

Second, a proposal to narrow the existing exceptions to the tribunal’s duty to order an equal pay audit once it has found a breach, removing three of the four current grounds for exemption and leaving only the exception for employers who have completed a qualifying audit in the previous three years.

What’s not on the table this time

Two measures that were live questions in the 2025 call for evidence don’t appear as proposals in the consultation at all.

The first is a ban on employers asking candidates about their salary history. This was explicitly described as a pay transparency measure in the call for evidence, attracted supportive international evidence in the responses that came back, and is one of the more operationally simple asks available to policymakers — it requires employers to stop asking a question, rather than build new infrastructure. It doesn’t feature among the consultation’s proposals, questions, or discussion of options considered.

The second is the right to information, giving workers a route to request pay information about comparators doing the same or equivalent work. This too was raised in the call for evidence responses, including specifically as a tool for addressing race and disability pay gaps, and separate government-commissioned research had already flagged it as a way of closing the pay information gap for workers. It isn’t part of this consultation either.

There may be a reasonably coherent internal logic here, even though the consultation doesn’t spell it out explicitly. A right to pay information is only as good as the comparator data behind it, and the consultation is candid that many employers don’t currently operate a non-discriminatory job evaluation scheme.

Introducing a right to information without first requiring employers to have a valid job evaluation scheme in place risks generating pay comparisons that aren’t actually comparing equal work.

A VERY narrow targeted fix: the job evaluation order

One further proposal deserves its own mention, not least because even the government’s own figures suggest its practical reach will be limited.

The consultation proposes that a tribunal which finds an equal pay breach must also order the employer to implement a non-discriminatory job evaluation scheme, if an up-to-date one isn’t already in place.

The logic is sound: an equal pay audit is only as reliable as the job classifications underneath it, and without a proper evaluation scheme those classifications may not reflect a genuinely objective assessment of relative job value.

The scale point, though, comes straight from the consultation itself. It notes that, to the government’s knowledge, only one equal pay audit has ever been ordered by an employment tribunal, because virtually every case to date has fallen within one of the existing exceptions. Even after those exceptions are narrowed, the government’s own assessment is that the job evaluation order “would affect only a very small number of employers”.

How this lines up against the EU Pay Transparency Directive

The comparison is best set out plainly, one element at a time, rather than as a scorecard.

On pay range disclosure before interview, the UK and the EU are heading in a similar direction. The PTD requires employers to give applicants information on the initial pay or its range early enough to allow an informed negotiation, based on objective, gender-neutral criteria. The UK consultation’s core proposal — publish in the advert, or provide in writing before interview if there’s no advert — sits comfortably within that same basic model.

On salary history questions, the two diverge. The PTD prohibits employers from asking applicants about their current or previous pay as part of the same recruitment-transparency package. The UK consultation doesn’t propose this at all, despite it being raised and supported in the evidence gathered specifically to inform this process.

On right to information, the gap is wider still. The PTD gives workers the right to request information on their own pay level and on average pay levels by sex for categories of workers doing the same or equivalent work, and requires employers to notify staff annually of that right. Nothing equivalent appears in the UK consultation.

On transparent, gender-neutral criteria for pay-setting and progression, the PTD requires employers to make those criteria easily accessible to workers. Again, this isn’t part of the current UK proposals, though the government’s planned replacement of the 2016 statutory code of practice for equal pay may end up touching on similar ground indirectly.

And on gender pay gap reporting and joint pay assessments — the PTD’s requirement for employers with 100 or more workers to report pay gap data, with a mandatory joint assessment where an unexplained gap of 5% or more is identified — there is no UK equivalent proposed here at all. The consultation is about equal pay and pay discrimination law, not gender pay gap reporting, which remains a separate regime in the UK. The government has made changes to UK gender pay gap reporting - equality action plans will be mandatory from April 2027 - and nothing new is being proposed.

Put together, of the PTD’s main pillars, the UK consultation proposes something close to one of them, in a form left largely to future regulations rather than fixed now.

This is definitely NOT implementation of the PTD. It is implementation of one idea that just happens to be also contained in the PTD.

Race and disability equal pay

The consultation’s other major strand — broadly levelling up protections against pay discrimination on the basis of race and disability toward parity with the existing sex-based equal pay scheme, including two narrowly defined scenarios for hypothetical comparators and new judicial discretion to extend time limits — is a substantial reform in its own right and deserves separate treatment another time.

For present purposes, the point to flag is simply that the government has explicitly sequenced this behind the pay transparency and enforcement measures, saying those improvements need to land first before protections are broadened.

Why the caution — the government’s own reasoning

Taking the government’s stated logic at face value for a moment, before drawing conclusions: The consultation frames itself as “Phase 1: Fixing the System”, built on the premise that the current equal pay framework is too complex, costly and slow to bear any further weight yet. It cites a volume of over 10,000 equal pay claims a year (though how many of those are different cases, and how many are multi-party claims with hundreds or thousands of claimants).

Equal pay cases can often take a decade or more to resolve, and a system that leaves both claimants and employers in prolonged uncertainty. On that basis, the government’s argument is that transparency and enforcement need to be fixed first, with broader protections — race and disability parity, and by extension any wider transparency measures — to follow once that foundation is in place.

That reasoning holds up reasonably well as an explanation for sequencing the race and disability reforms behind Phase 1. It’s a harder argument to extend to the salary history ban and the right to information specifically, since neither requires tribunal reform, a functioning Equal Pay Regulatory and Enforcement Unit, or any of the other institutional plumbing the consultation is trying to fix first. A salary history ban, in particular, is a rule about what employers may ask at interview — it doesn’t obviously depend on fixing the tribunal backlog to be workable now.

Where this leaves things

Across Europe, North America and parts of Asia-Pacific, governments have increasingly moved towards greater pay transparency over the last decade.

For now, the UK has clearly chosen an incremental path, rather than going with the global flow: adopting one element of pay transparency while leaving most of the EU Pay Transparency Directive's architecture untouched.

Incremental though it may be, this is a genuinely important step for the UK, and the recruitment pay-disclosure proposal, once implemented, will bring UK practice measurably closer to the PTD.

The consultation is open for 15 weeks and can be accessed here.

A few quick answers, for reference

Does the UK’s new consultation ban asking about salary history? No. It isn’t proposed, discussed, or put forward as a question for consultation, despite being raised as a pay transparency measure in the 2025 call for evidence.

What’s the main difference between the UK proposals and the EU Pay Transparency Directive? The UK is consulting on pay range disclosure before interview, which mirrors one PTD requirement. It is not proposing a salary history ban, an individual right to information, transparent pay-setting criteria, or gender pay gap reporting with joint pay assessments, all of which are PTD requirements.

How many equal pay audits have UK employment tribunals ordered? One, according to the consultation’s own figures, because almost every case has fallen within one of the existing statutory exceptions.

How many responses did the 2025 call for evidence receive? 176 submissions were received, of which 147 were included in the final analysis.

Is the UK implementing the EU Pay Transparency Directive? No. The consultation adopts one proposal that resembles the Directive — salary disclosure before interview — but leaves most of its other core provisions untouched.

Sometimes, when renovating a property, you don’t gut the entire house in one go. You take it a room at a time. Finish the kitchen, then come back months later for the bathroom, and then maybe tackle the dodgy wiring you inherited. For a while, the place looks inconsistent, and a bit untidy maybe. But it remains perfectly liveable. It still functions.

Arguably, that is roughly what is happening with the Pay Transparency Directive in Finland and Estonia. Neither country has delivered a full transposition. Both have delivered a working part, while the rest of the engine waits its turn.

Finland has finally sent its implementing bill to Parliament. Estonia has finalised law, but only for the provisions it always described as “the easy half”.

Read together, the two stories say something useful about where Pay Transparency Directive implementation is heading across the EU. We’re now over a onth past the supposed deadline for implementation. Recruitment-stage transparency is becoming a settled expectation, while the harder architecture - job evaluation, reporting and joint pay assessments - is still being built one bolt at a time.

More broadly, this sort of pattern is becoming difficult to ignore. Across Europe, governments increasingly appear to be implementing the Directive in the order of political and operational difficulty: the recruitment-stage obligations first, with the more complex pay equity infrastructure following later.

Finland’s Pay Transparency Directive bill reaches Parliament

Finland’s government submitted HE 129/2026 vp to the Eduskunta on 9 July 2026, signed by Prime Minister Petteri Orpo and Minister Karoliina Partanen.

This is a government bill rather than another round of consultation, which matters because Finland has been circling this legislation for well over a year. The original target for entry into force was 18 May 2026, later slipping past the EU’s own Pay Transparency Directive transposition deadline of 7 June 2026.

The bill now on the table proposes entry into force on 1 January 2027.

Finland is not remotely alone in missing the deadline. They are firmly in the “delayed but progressing” camp (alongside most of the EU).

What distinguishes Finland is that the bill is now genuinely substantive, not a discussion draft.

It amends:

• the Equality Act (609/1986)

• the Ombudsman for Equality Act

• the Incomes Register Act

• the Statistics Finland Act

• the Fine Enforcement Act

The substance itself sticks closely to the Directive’s floor rather than building above it, consistent with the current government’s stated approach of implementing at minimum level.

The headline points:

• Pre-employment transparency. Employers must give candidates the starting salary or salary range for a role (doesn’t have to be in the job ad), cannot ask about salary history, and must keep job advertisements and titles gender-neutral. All very much in line with the PTD.

• Pay criteria disclosure. All employers must explain, in an accessible form, the criteria used to determine pay and pay levels. The obligation to explain pay progression criteria specifically is limited to employers with 50 or more staff, Finland using the national discretion the Directive allows rather than extending the duty further.

• Right to pay information. Employees can request their own pay level and the average pay of comparable colleagues, with a written answer due within two months (other countries have set shorter deadlines, but not Finland).

• Gender pay gap reporting. Employers with 100 or more staff face reporting obligations, but the mechanics are distinctly Finnish: data flows through the Incomes Register and Statistics Finland, which does much of the calculation before reaching the Ombudsman for Equality.

• Joint pay assessments. Required where an unjustified pay gap of 5% or more shows up within a worker category.

• Sharper enforcement. The existing conditional fine (uhkasakko) is extended to cover more of the new obligations, a new standalone non-compliance fee (laiminlyöntimaksu) is introduced specifically for employers who fail to submit a joint pay assessment, and the limitation period for pay discrimination compensation claims is extended from two to three years.

One detail deserves particular attention because it changes the practical implementation timetable.

The bill’s transitional provisions push the first gender pay gap reporting deadline to 7 June 2028, covering 2027 data, for employers with 250 or more employees and those with 150 to 249 employees.

Earlier commentary on the draft had assumed a 2027 deadline covering 2026 data. The clock has therefore moved a full year later than the timeline being discussed as recently as February 2026, buried inside the transitional provisions rather than announced as a headline change.

If you are building an internal project plan around Finland’s reporting dates, the number to use is 2028, not 2027.

The bill also was not a rubber stamp exercise.

Consultation responses show genuine friction. Trade union bodies, including Akava, argued that the proposed equal-value provisions fall below the Directive’s minimum standard, while employer organisations pushed back on the new non-compliance fee as unnecessary and duplicative of existing sanctions.

Neither side got everything it wanted, which is usually a reasonable sign that a minimum-level implementation has landed roughly where it was always expected to.

Finland Draft

6.62MB ∙ PDF file

Download

Download

Estonia partially implements the Pay Transparency Directive

Readers who followed the earlier instalments of this story will remember the arc.

In April, Estonia said no to the Directive in its current form, with officials suggesting it might be cheaper to absorb EU fines than transpose in full.

Within a week, that hardened “no” softened into something closer to “not no, but not yes either,” as the government began progressing legislation covering the lower-friction elements of the Directive while leaving the harder architecture unresolved.

That legislation is now law (available here)

The Riigikogu passed the omnibus amendment to the Employment Contracts Act on 17 June 2026.

It delivers exactly the package flagged back in April, now with specific statutory hooks attached:

• A new ban on employers asking job seekers about their previous or current pay, inserted as section 11(2¹) of the Employment Contracts Act.

• A new duty to give job seekers the pay or pay range for the role in writing, in particular through the job advertisement before the interview, including any relevant collective agreement pay terms, at section 11(2²).

• A new prohibition on employers preventing employees from disclosing their own pay, added to the list of prohibited conduct at section 28(2).

• A freestanding codification of the equal pay principle itself: employers must ensure equal pay for equal or equivalent work between men and women. This sits alongside, rather than inside, the recruitment-stage mechanics, and is a broader statement of principle than the Directive’s minimum procedural requirements strictly demand.

The law is an “omnibus instrument”, making changes to various other pieces of law too.

Commencement

Commencement is phased, and the phasing is worth reading carefully.

Only a specific numbered set of provisions (points 9 to 11 of the Employment Contracts Act amendments) is deferred until 1 January 2027.

The salary history ban, the pay range disclosure duty, the pay secrecy ban and the new equal pay codification are not among the deferred provisions.

They therefore take effect on the law’s general commencement date rather than in 2027, confirming that Estonia has genuinely prioritised the easier implementation measures rather than merely promising to get to them later.

The harder architecture is a different story.

Job evaluation frameworks, mandatory gender pay gap reporting and joint pay assessments remain unaddressed.

Estonia’s Economy Minister formally asked the European Commission in April 2026 for a two-year postponement of the reporting obligations until 2028. The Commission did not grant it.

In place of a mandatory reporting engine, Estonia continues to rely on the voluntary Palgapeegel (”Pay Mirror”) tool.

That is a genuinely different model from the state-run compliance systems being built in Lithuania and Slovakia, which are designed to compel disclosure rather than invite it. A small provision buried in the Occupational Health and Safety Act amendment reinforces the point: gender pay gap indicators will not be stored in the relevant government database.

It is an easy line to miss, but it is a tell about Estonia’s overall direction. Estonia is deliberately avoiding the kind of centralised data infrastructure that mandatory reporting regimes elsewhere in the EU depend upon.

What these two developments tell us about the Pay Transparency Directive

Finland and Estonia have taken different legislative routes, but they are moving in the same direction.

Both missed the 7 June 2026 transposition deadline. But neither has stopped moving, which already puts them in a meaningfully different position from Sweden, which has postponed transposition altogether without a confirmed legislative timetable.

More importantly, both countries have prioritised the parts of the Pay Transparency Directive that are easiest to implement operationally. Recruitment-stage transparency is arriving first. The more demanding parts of the Directive are following later.

That sequencing is not a coincidence, and it is not unique to Finland and Estonia.

Considered more broadly, they point towards something more significant.

Across Europe, governments increasingly appear to be implementing the Directive in two phases. The first phase introduces the obligations that employers can incorporate into existing recruitment processes relatively quickly: salary range disclosure, salary history bans and restrictions on pay secrecy clauses. The second phase tackles the structural reforms that require new systems, methodologies and regulatory infrastructure, including equal value job evaluation, mandatory gender pay gap reporting and joint pay assessments.

That distinction is becoming one of the defining features of Pay Transparency Directive implementation across the EU. The greatest divergence between Member States is no longer emerging around recruitment transparency. It is emerging around the infrastructure needed to measure, assess and remedy pay inequality.

What HR, legal and reward teams should be watching

Finland’s reporting timetable has moved again. The first gender pay gap reporting deadline is now 7 June 2028 for the largest employers, a year later than many earlier commentaries anticipated. Project plans should be built around the bill’s transitional provisions rather than its headline commencement date.

Estonia has moved beyond political signalling. The salary history ban, pay range disclosure duty and prohibition on pay secrecy clauses are now legislative requirements rather than policy proposals. Employers recruiting in Estonia should treat these obligations as current compliance requirements.

Estonian implementation remains incomplete. Mandatory gender pay gap reporting, joint pay assessments and job evaluation frameworks have yet to be implemented. Organisations operating across multiple Member States should not assume Estonia has fully transposed the Directive simply because recruitment-stage obligations are now in force.

Read the transitional provisions as carefully as the commencement date. Finland provides perhaps the clearest example so far of how the practical compliance timetable can differ materially from the date printed on the front page of the legislation.

Look beyond individual countries. The wider implementation pattern is becoming increasingly clear. Recruitment transparency is converging across Europe. The greatest differences between Member States are now emerging in the mechanisms designed to deliver longer-term pay equity.

What to watch next

The next phase of implementation will be considerably more revealing than the first.

Salary ranges, salary history bans and pay secrecy provisions are comparatively straightforward legislative changes. They fit comfortably within existing employment law frameworks and require relatively modest changes to employer processes.

The harder questions come afterwards.

How will Member States define work of equal value? How prescriptive will they become on job evaluation methodologies? Will reporting systems rely on employer calculations, centralised government infrastructure or some combination of both? How aggressively will national regulators enforce joint pay assessment obligations where unexplained pay gaps emerge?

Those are the questions that will determine whether Member States ultimately produce genuinely comparable pay transparency regimes or simply a shared set of recruitment rules sitting on top of very different domestic systems.

Finland and Estonia have both taken another step towards implementing the Pay Transparency Directive. The more interesting question now is not who reaches the finish line first, but whether the next phase of implementation produces greater convergence, or exposes just how differently Member States intend to deliver the Directive’s underlying objective of pay equity.

Most discussions about pay gap analysis focus on the regression model.

Over the last few articles in this series, I’ve looked at how to build a regression model, how to test whether it’s robust, and how to interpret the results.

All of those discussions assume one thing: that the data going into the model actually measures what you think it measures.

Often, that’s a big assumption. And not one that’s always well-founded.

In practice, the quality of your data often matters much more than the sophistication of your statistics.

A beautifully specified regression cannot rescue poor data.

And under the EU Pay Transparency Directive (PTD), where organisations will increasingly rely on statistical analysis to understand and explain pay gaps, that really matters.

What we mean by “bad data”

When people hear “bad data”, they often think of obvious mistakes:

• missing data

• duplicate records (employees who leave and rejoin in the same year)

• incorrect dates (the classic being dd/mm/yyyy dates being stored as mm/dd/yyyy dates, and vice versa)

• typing errors (even just a few blank spaces can be enough to throw things off)

Those are definitely annoying problems, but they are not usually the most important ones. They can be cleaned up and dealt with. It’s pretty tedious work, but essential.

From a statistical perspective though, bad data usually means something more subtle.

It means a variable does not accurately measure the thing it is supposed to represent.

We call this measurement error: the recorded value differs, to some extent, from the underlying concept you are actually trying to measure.

That difference may be small. It may be unavoidable. But if it is systematic, it can materially affect the conclusions your regression reaches.

Let’s think about tenure and its inclusion in your regression because you believe experience influences pay.

If your tenure variable records years with the company, but many employees joined after spending years doing the same job elsewhere, have you really measured experience?

Not quite.

You’ve measured something related to experience (it collineates), but not experience itself.

The regression doesn’t know the difference. It’s doing what you told it to do. It simply assumes the variable means what you told it it means.

Random versus systematic errors

Not all data problems behave in the same way.

Broadly speaking, statisticians distinguish between random errors and systematic errors.

Random errors occur unpredictably. Perhaps a handful of employees have incorrect start dates because of historic migration from an old HR system. Or bonus payments have occasionally been allocated to the wrong reporting period.

These mistakes add noise. They make relationships harder to detect, but they do not consistently push the analysis in one direction.

Systematic errors are more problematic.

These occur when a variable consistently fails to measure what you intend. Imagine an international employer using company service as a proxy for experience. In countries with low employee turnover, this may be a reasonable approximation. In countries where external hiring is common, it may be a very poor one.

The variable is still populated. It is still internally consistent.

But it is consistently measuring the wrong thing.

Those are often the hardest problems to spot because the data looks perfectly clean.

Measurement error: when important variables appear unimportant

Regression analysis rests on an assumption that explanatory variables have been measured accurately.

When they have not, something slightly counter-intuitive often happens.

The model starts to underestimate their importance.

Imagine two employees.

One has twenty years’ experience but joined the organisation last year.

The other has spent ten years with the organisation since graduating.

If your model uses company service as a measure of experience, those two employees appear to have one year and ten years of experience respectively.

The variable no longer reflects reality.

As a result, the relationship between experience and pay appears weaker than it actually is.

The regression may estimate that experience has a much smaller effect on pay than it actually does.

Statisticians refer to this phenomenon as attenuation bias.

Remember the practical implication:

Poorly measured variables often look less important than they really are.

That has an important consequence for PTD analysis.

If legitimate drivers of pay appear weaker than they actually are, more of the observed pay gap remains unexplained.

In other words, poor measurement can increase the apparent unexplained pay gap without anything about employees’ pay actually changing.

This also affects the gender coefficient itself.

Suppose performance genuinely explains part of the observed pay gap, but performance is measured poorly. Because the model cannot fully attribute pay differences to performance, some of that variation remains elsewhere. In practice, that often means the estimated gender coefficient is larger than it would otherwise have been.

The regression has not uncovered more evidence of gender-related differences. It has simply failed to explain those differences using the information available.

Common examples in PTD analysis

This issue appears surprisingly often in pay gap work.

Tenure

Length of service is easy to obtain.

Relevant experience is much harder.

Those are not necessarily the same thing.

Performance ratings

Many organisations include performance as a control variable.

But performance ratings themselves are imperfect measurements.

Some managers use the full rating scale.

Others rarely award the highest scores.

Calibration practices differ between departments and countries.

A performance rating may therefore measure manager behaviour as much as employee performance.

Market supplements and allowances

Many employers pay additional allowances for:

• hard-to-fill roles

• scarce skills

• geographical differences

• shift work

• on-call responsibilities

They are generally (but not always) permitted as objective reasons for pay. PRactically, these will be genuine drivers of pay.

If they are omitted, inconsistently recorded or grouped together under broad payroll categories, the model loses part of the explanation for why employees are paid differently.

This is a problem employers are dealing with right now. How do you take into account market forces into a regression model? It’s the most important reason why there is a difference in two equal value (but very different) jobs in the same category, such as sales/tech and finance/HR, but how do you take that into account in a model? What is the actual data you feed in?

Working hours

Part-time working is generally straightforward.

Contracted hours are usually available.

Actual working patterns can be more complicated.

Employees working regular overtime, additional shifts or variable hours may have pay that is only partially explained by contracted hours alone.

Missing variables are not the same as missing data

There is another distinction that often gets overlooked.

Missing observations are not the same as missing variables.

Missing observations occur when information is absent.

Perhaps performance ratings are unavailable for some employees.

Missing variables are different.

They occur when an important factor has not been measured at all.

Suppose individual performance genuinely influences pay decisions, but no reliable performance data exists.

Or market scarcity premiums are negotiated individually but never recorded in HR systems.

Or professional qualifications materially affect pay but are not captured centrally.

The regression cannot control for factors it cannot see.

This is different from poor measurement. A poorly measured variable gives the model an imperfect version of the truth. A missing variable gives it no opportunity to account for that source of pay variation at all.

Instead, their influence is absorbed elsewhere in the model.

Often, that means the unexplained component becomes larger than it would otherwise have been.

Statisticians describe this as omitted variable bias.

If an important driver of pay has not been measured, the regression cannot account for it.

A practical data audit before running your model

Most organisations spend considerable time discussing model specification.

Relatively few spend the same amount of time auditing the variables themselves.

Before building a regression model, it is worth asking a few simple questions.

Is the variable complete?

Are there substantial gaps?

Is it defined consistently?

Would two HR teams in different countries record it in the same way?

Does it actually measure what I want it to measure?

Or is it merely a convenient proxy?

Would another reasonable analyst define this variable in the same way?

If not, your results may depend as much on judgement as they do on statistics.

Is it comparable across jurisdictions?

Definitions of bonuses, allowances and working patterns can differ significantly across multinational organisations.

Would I be comfortable explaining this variable to an employee representative?

If not, it probably deserves further scrutiny before it enters the model.

Key takeaways

• A sophisticated regression cannot compensate for poor quality variables.

• Random errors increase noise and make conclusions harder to identify; systematic errors are worse and can lead to the wrong conclusions altogether.

• Variables often measure proxies rather than the concepts we actually care about.

• Poor measurement can make genuine drivers of pay appear less important than they really are.

• Poor data quality can change the estimated gender coefficient, not just the explanatory variables.

• Missing variables are fundamentally different from missing observations.

• Data quality should be audited before model specification, not afterwards.

Closing

Regression models are powerful tools. They’ll continue to be a big part of every pay equity analysis for years to come.

But they are only as good as the information they are given.

The temptation is to spend hours debating transformations, interaction terms and statistical significance.

Those decisions matter.

But they are only worth debating once you are confident the underlying variables actually represent the concepts you want to analyse.

Organisations often assume that better statistics produce better answers.

More often, better data does.

Because the best regression in the world cannot compensate for variables that were never measured properly.

A month ago, I wrote about Greece’s Pay Transparency Directive draft - the consultation version that appeared on 3 June 2026 and gave stakeholders two weeks to respond.

At the time, many thought enactment was months away.

They were all (sort of) wrong.

On 6 July 2026, Law 5316/2026 was signed by the President and published in the Official Gazette.

Greece’s PTD transposition: what happened

Here is how fast this moved:

• 3 June 2026: Draft published, public consultation opens

• 7 June 2026: EU transposition deadline passes — Greece misses it

• 17 June 2026: Consultation closes

• Early July 2026: Parliamentary committee and plenary vote

• 6 July 2026: Presidential signature, publication, law enacted

Thirty-three days from consultation opening to enacted statute. Nineteen days from consultation close to presidential signature. The Greek Parliament was still sitting in early July; the government pushed the bill through in the final days before summer recess rather than waiting until October.

Greece is the first southern European member state to implement the Pay Transparency Directive

Only four EU Member States (Slovakia, Italy, Lithuania, and Malta) met the 7 June 2026 deadline. Greece is the first of the latecomers to actually finish. More significantly, it is the first southern or Mediterranean EU state to get the Pay Transparency Directive over the line, ahead of France, Spain, and Germany — none of which have published final legislation.

What changed between the draft and the final law?

Almost nothing, but with one meaningful exception.

The draft said the law would take effect on publication in the Official Gazette. The final law splits commencement:

• Immediate effect (6 July 2026): structural provisions such as scope, definitions, institutional designations, the Ombudsman’s role

• 1 November 2026: all operational substance: pay transparency obligations, right to information (RTI), pay gap reporting, joint pay assessments, remedies, burden of proof, victimisation protections

That four-month run-up is the news for employers operating in Greece. Greek employers have until November before the transparency obligations bite.

Everything else from my earlier analysis — the Ombudsman centrality, the mediated disclosure for small groups, the refusal rights for abusive requests, the CBA presumption — is intact and unchanged.

Greece’s gender pay gap is one of the lowest in the EU

Here is the part that rarely gets mentioned in the transposition coverage.

According to Eurostat’s 2024 data, the EU average unadjusted gender pay gap is 11.1%. Greece’s figure has consistently been among the lowest in the EU (3–4%). On the headline metric that the Pay Transparency Directive is designed to address, Greece barely registers as having a problem.

But Greece has the second-worst gender employment gap in the entire EU: 18.8 percentage points in 2024 (ie a much larger percentage of men are employed than women). Only Italy is worse. Fewer than 60% of working-age Greek women are in employment, compared with more than 77% of men.

The paradox: relatively few women work in Greece, but those who do earn close to what men earn. The low pay gap partly reflects a selection effect. The women who participate in the Greek labour market tend to be in higher-skilled, better-paid roles. A large cohort of women simply is not in the formal workforce at all.

So is this legislation solving the right problem? Maybe not for the Greek case. Pay transparency regimes are designed to surface gaps in earnings. They do not address gaps in participation. Greece is legislating for the EU metric, which is reasonable (it is a Directive transposition, after all) but the structural gender inequality in the Greek labour market sits somewhere this law cannot reach.

That is not a criticism. It is an observation about what transparency frameworks can and cannot do.

What this tells use about the new phase of transposition

The pace is accelerating. Political pressure from the Commission may be working. Greece demonstrates that even countries that miss the deadline by days, not months, can still move quickly once political will exists.

The substance of the Greek law is faithful to the Directive, lightly employer-friendly in its protective features, and institutionally centralised around the Ombudsman — exactly as I described in my earlier piece on the draft. The structural thesis holds: wherever transparency creates a tension, Greece inserts an intermediary to manage it.

The news here is not a major change in direction, but the speed. And the question is whether any of it will move the needle in a country where the bigger gender problem is not pay, it is participation.

A regression can run cleanly and still be wrong to trust.

This piece covers how to check whether a PTD regression is behaving as it should: reading residual plots for funnel, curve and cluster patterns, using Cook’s Distance to spot employees who are quietly driving the result, and — new for this series — checking those same things inside a decomposition, where a single influential observation in one subgroup’s regression can shift the explained/unexplained split that everyone actually cares about.

You run the regression. The coefficients appear. The gender coefficient is statistically significant. Nothing crashes, no warnings appear, and it’s tempting to treat that as the end of the process.

It isn’t.

Part #8 in this series looked at robustness: whether your conclusion survives being poked. Change the specification, drop a variable, try a different control set, and see if the gender coefficient holds its shape. That’s an external question, one about how the answer behaves when the model around it changes.

This piece is about something narrower, and arguably more basic. Before you start swapping specifications, is the one model in front of you behaving the way a regression is supposed to? Are its errors random, or is there a pattern hiding in them? Is the whole result quietly resting on one or two employees? These are internal questions. They concern the model you’ve already chosen, not the alternatives you haven’t tried yet.

Both matter. They’re not the same question though, and treating them as interchangeable is how PTD regressions end up looking rigorous without being properly checked.

Residuals: what the model couldn’t explain

Every regression produces a prediction for each employee: what their pay should be, given grade, tenure, location, performance, and whatever else sits in the model. The gap between that prediction and their actual pay is the residual.

Predicted £52,000, actual £55,000. Residual of +£3,000. Actual £49,000 instead: residual of −£3,000.

Residuals are everything the model failed to explain. That’s not a flaw. No regression explains every pound of pay, and it shouldn’t try to. The real question isn’t whether residuals exist, it’s whether they look like noise or whether they look like a pattern the model has missed.

Spreadsheets full of residual values don’t tell you much on their own.

Plotted, with predicted values along the bottom and residuals up the side, they tell you a great deal, usually within seconds of looking at the chart.

Three patterns to recognise

A funnel. Residuals sit tightly together at lower predicted salaries and fan out wider as predicted pay increases. This is heteroscedasticity, the model’s errors growing alongside the pay itself. It’s common in reward data because senior pay is genuinely lumpier: bonuses, LTIPs, retention awards and individually negotiated packages all sit on top of base salary in ways junior pay bands rarely see. A funnel doesn’t invalidate the regression. It’s a prompt to think about robust standard errors, or whether senior and junior populations need separating out.

A curve. The residuals bow rather than scatter. This usually means a relationship you’ve modelled as a straight line isn’t actually straight, early-career pay growth that’s rapid and then flattens being the classic example. The regression is trying to draw a straight line through something that bends. A curved pattern points towards a transformation, a polynomial term, or a different functional form. It isn’t a verdict that the model is broken.

Clusters. Instead of one cloud, you see two or three distinct groups of points. This is usually the model trying to explain several different pay systems with a single equation: corporate, retail and engineering populations, say, each with its own labour market and progression logic. Clustering is often the clearest signal that an important variable, such as business unit, job family or geography, has been left out.

None of these patterns comes with an instruction manual telling you exactly what’s wrong. What they do is tell you where to look next, which is most of what a diagnostic is for.

Influential observations, in brief

Part #8 covered this conceptually: one heavy brick can hold up a structure that looks stable everywhere else. The mechanics behind that idea are worth having to hand, so here they are without repeating the argument.

Leverage describes how unusual an employee’s characteristics are, an outlying combination of tenure, grade and location, say, regardless of what they earn.

Influence is different: how much the regression’s coefficients would move if that employee were removed. An employee can have high leverage and almost no influence, or ordinary leverage and outsized influence, if their pay sits a long way from what the model expects.

Cook’s Distance combines the two into a single number per employee: roughly, how much the fitted model shifts when that person is dropped. There’s no universal cut-off, but a common rule of thumb flags anything above 4/n (n being your sample size) as worth attention.

To give a very simplistic example at the companywide level (rather than category level): say your organisation has 3,500 employees, almost all earning between €30,000 and €120,000, plus a female CEO on €1.8 million. Drop her from the regression and the gender coefficient moves noticeably. That doesn’t mean she should be excluded. She’s a genuine employee and belongs in the dataset. It means the result is sensitive to one exceptional observation, and that’s exactly the kind of thing worth disclosing rather than having someone else discover it first.

Diagnosing a decomposition

Everything above assumes a single regression. Part #10 introduced decomposition, Oaxaca-Blinder, which splits the pay gap into an explained portion (differences in grade, tenure, and so on) and an unexplained portion (what’s left once those are accounted for). That structure changes what diagnostics need to check.

A standard decomposition runs two regressions, one for each group being compared, and uses the coefficients from both to build a counterfactual: what would the female population earn, on average, if paid according to the male pay structure? That counterfactual is where the explained/unexplained split comes from, and it means the diagnostics need to be run on both underlying regressions, not just a pooled one.

This matters more than it sounds. If one regression, say the female-employee model, has a single influential observation pulling one coefficient around, that coefficient feeds directly into the counterfactual calculation. The result isn’t just a slightly noisy overall model. It’s a shift in how much of the gap gets classified as explained versus unexplained, which is usually the number everyone actually cares about.

Run leverage and influence checks separately within each group’s regression, not just on the combined dataset. An employee who looks unremarkable in the pooled model can be one of only a handful of observations driving a subgroup’s coefficient, particularly where one group is small. If you’ve read part #4 on small samples, this is where that discussion and this one meet. Small subgroups and influential-observation risk tend to arrive together, not separately.

For any decomposition, check Cook’s Distance within each group’s regression before trusting the explained/unexplained split, not just the top-line gender coefficient.

When diagnostics disagree

A residual plot can look clean, no funnel, no curve, no clusters, while Cook’s Distance flags one employee as carrying an outsized share of the result. That isn’t a contradiction so much as two different questions being answered separately. The residual plot asks whether the model has systematically missed something. Cook’s Distance asks whether one observation is doing more work than the rest. A model can pass the first test and still depend heavily on the second.

This is why experienced analysts rarely lean on a single diagnostic. Each answers a different question, and the answers need reading together rather than in isolation. There’s no equivalent of an MOT with one pass-or-fail light. Judgement still does most of the work; diagnostics just make sure that judgement is informed rather than guessed.

The 60-second PTD regression diagnostic

Here’s a quick five point aide:

1. Do the residuals look broadly random, or is there a funnel, a curve, or clustering in the plot?

2. Is the relationship genuinely linear, or would a transformation better reflect how pay actually moves?

3. Have you checked Cook’s Distance (or similar) for influential observations, and if you’ve run a decomposition, checked it separately within each group’s regression?

4. Would the headline conclusion survive a reasonable sensitivity check?

5. Could you explain, plainly, why this model is the right one, not to another statistician but to an employee rep? Remember - they are the key audience for PTD analysis.

If all five hold up, you’ve gone further than most PTD regressions ever do.

A couple of common questions

Does a failed diagnostic mean the regression is wrong? No. It means the result deserves a closer look before you rely on it. Most diagnostic flags point to something worth investigating or disclosing, not something that invalidates the analysis outright.

Should influential employees, like a high-earning CEO, be removed from the model? Generally no. If they’re genuine employees, they belong in the dataset. The correct response to an influential observation is usually a sensitivity check and transparent reporting, not deletion.

Which diagnostic matters most if I only have time for one? The residual plot. It’s the fastest to run and the most likely to reveal a genuine problem with the model’s structure within seconds of looking at it.

Final thoughts

A regression gives you an answer. Diagnostics tell you how much weight that answer can bear. They won’t certify a model as correct; nothing does that. What they do is narrow the gap between “the software ran without error” and “there’s a defensible basis for what it concluded.”

Under the PTD, that gap is where most of the real scrutiny will land. Not on whether you ran a regression, but on whether you can show you looked hard enough at what it was actually telling you.

# ---------------------------------------------------------
# Regression diagnostics: 3 clean assumption violations
# Each plot shown separately (not side-by-side)
# ---------------------------------------------------------

set.seed(42)

n <- 300

# =========================================================
# 1. FUNNEL (heteroscedasticity only)
# =========================================================

tenure <- runif(n, 0, 25)

# keep mean structure VERY clean and monotonic
base_salary <- 30000 + tenure * 1500

# CRITICAL FIX:
# variance depends ONLY on tenure (not salary, not grade)
noise_funnel <- rnorm(n, 0, sd = 500 + (tenure^2) * 25)

salary_funnel <- base_salary + noise_funnel

model_funnel <- lm(salary_funnel ~ tenure)

plot(fitted(model_funnel), resid(model_funnel),
     main = "Funnel shape (heteroskedasticity)",
     xlab = "Predicted salary (£)",
     ylab = "Residual (£)",
     pch = 16,
     cex = 0.7,
     col = rgb(0.1, 0.3, 0.8, 0.35))

abline(h = 0, col = "red", lwd = 2, lty = 2)
grid()


# =========================================================
# 2. CURVE (non-linearity only — strong U-shape)
# =========================================================

years_service <- runif(n, 0, 30)

# Strong quadratic curvature + constant variance
true_salary <- 28000 +
               1600 * years_service -
               45 * (years_service^2) +
               rnorm(n, 0, 1200)

model_curve <- lm(true_salary ~ years_service)

plot(fitted(model_curve), resid(model_curve),
     main = "Curved shape (non-linearity)",
     xlab = "Predicted salary (£)",
     ylab = "Residual (£)",
     pch = 16,
     cex = 0.7,
     col = rgb(0.8, 0.2, 0.2, 0.35))

abline(h = 0, col = "red", lwd = 2, lty = 2)

# makes curvature visually obvious
lines(lowess(fitted(model_curve), resid(model_curve)),
      col = "black", lwd = 3)

grid()

# readline("Press [enter] for next plot")


# =========================================================
# 3. CLUSTERS (omitted variable only)
# =========================================================

business_unit <- sample(c("Corporate", "Retail", "Engineering"),
                        n, replace = TRUE)

tenure2 <- runif(n, 0, 20)

unit_effect <- ifelse(business_unit == "Engineering", 18000,
                ifelse(business_unit == "Corporate", 8000, 0))

salary_cluster <- 24000 +
                  unit_effect +
                  tenure2 * 500 +
                  rnorm(n, 0, 1800)

model_cluster <- lm(salary_cluster ~ tenure2)

cols <- c("Corporate" = "darkgreen",
          "Retail" = "orange",
          "Engineering" = "purple")

plot(fitted(model_cluster), resid(model_cluster),
     main = "Clusters (omitted variable)",
     xlab = "Predicted salary (£)",
     ylab = "Residual (£)",
     pch = 16,
     cex = 0.7,
     col = cols[business_unit])

abline(h = 0, col = "red", lwd = 2, lty = 2)

legend("topright",
       legend = names(cols),
       col = cols,
       pch = 16,
       cex = 0.8)

grid()

One of the EU Pay Transparency Directive’s greatest strengths is also one of its greatest frustrations.

Rather than prescribing exactly how employers should achieve equal pay, the Directive sets out broad principles. Employers must establish objective, gender-neutral pay structures. They must ensure work of equal value is rewarded equally. They must be able to objectively justify pay differences.

But beyond those principles, it says surprisingly little.

What exactly is a “pay structure”? How should employers assess work of equal value? When can market forces justify a pay difference? And what does an “objective, gender-neutral” pay system actually look like in practice?

Those questions have largely been left to Member States.

Slovakia has now become one of the first countries to attempt some answers.

Following its implementation of the Pay Transparency Directive (Directive (EU) 2023/970 of 10 May 2023), the Slovak Ministry of Labour has published a 106-page guidance document covering everything from job evaluation and pay structures to objective justification and recruitment practices. Slovakia transposed the Directive through Act No. 76/2026, which took effect on 7 June 2026—the Directive’s transposition deadline. Unlike much of the guidance published elsewhere in Europe, this isn’t simply a restatement of the legislation. It is a serious attempt to explain what compliance might look like in practice.

That’s significant.

Not necessarily because every employer should follow it. And certainly not because it changes the law. But because it provides one of the clearest indications yet of how at least one Member State expects employers to operationalise some of the Directive’s most ambiguous concepts.

There is another reason this guidance deserves attention.

Government guidance on employment law is often disappointing.

Too often it simply paraphrases legislation that readers have already struggled through themselves. It tells employers what the law says, but offers very little help with how to comply with it.

The Slovak guidance is different.

It contains practical examples. It explores difficult concepts such as work of equal value. It discusses the role of market forces, seniority and performance. It devotes substantial attention to job evaluation methodology. Even where you disagree with some of its conclusions, it is difficult to argue that it hasn’t engaged seriously with the practical challenges employers face.

That doesn’t mean it is beyond criticism.

Some parts arguably go further than the Directive itself. Other parts leave important questions unanswered, particularly around reporting calculations and implementation. And because the guidance is not legally binding, there are limits to how much weight employers should place on it.

Which brings us to the first question most employers will ask.

How much weight does the guidance actually carry?

One of the most refreshing aspects of the guidance is that it is remarkably candid about its own status.

Rather than presenting itself as an authoritative interpretation of the legislation, the Ministry expressly states that the document is intended to support implementation, creates no new legal obligations, and is not legally binding. It also acknowledges that only the courts can ultimately determine how the legislation should be interpreted.

That disclaimer matters.

An employer does not become non-compliant simply because it has chosen a different approach from the one described in the guidance. Nor does the document create a new legal test for equal pay.

But it would be a mistake to dismiss it for that reason.

Official guidance frequently occupies an important middle ground between legislation and case law. It does not determine the outcome of litigation, but it often influences the way legislation is understood and applied in practice. Labour inspectorates, equality bodies and claimant representatives are all likely to read this document closely. It is difficult to imagine a labour inspector assessing an employer’s approach to job evaluation without at least considering a methodology published by the Ministry responsible for implementing the legislation.

Whether courts will attach significant weight to the guidance is a more nuanced question.

A court is unlikely to conclude that an employer has breached the legislation simply because it has departed from the Ministry’s preferred methodology. The Directive itself deliberately permits flexibility, and the guidance expressly accepts that alternative approaches exist.

Equally, however, the guidance may still prove persuasive. If a claimant argues that an employer’s pay structures are inadequately documented, or that its job evaluation process lacks objective criteria, a court may well look to the Ministry’s guidance as one indication of what a reasonable implementation looks like. That is very different from saying the guidance is legally binding, but it is equally different from saying it is irrelevant.

Perhaps the greatest influence will be outside the courtroom.

Labour inspectorates, equality bodies and claimant lawyers are all likely to use this document as a practical benchmark when assessing whether an employer has taken a sufficiently objective approach. For that reason alone, it deserves careful attention.

The more interesting question, however, is not whether employers have to follow the guidance. It is whether the guidance itself always reflects what the Directive requires.

Nowhere is that question more interesting than in the section on job evaluation.

Job evaluation takes centre stage in Slovakia’s PTD guidance

By far the largest section of the guidance is devoted to job evaluation.

That isn’t surprising. Equal pay ultimately depends on being able to answer one deceptively simple question: are these jobs of equal value?

Everything else in the Directive flows from that.

You cannot build objective pay structures without understanding how roles compare. You cannot identify equal pay risks if you cannot assess work of equal value. And if challenged, employers need to be able to explain why two apparently different jobs attract different levels of pay.

The guidance therefore places job evaluation at the heart of PTD compliance.

That emphasis is entirely consistent with the Directive.

Where the discussion becomes more interesting is how the guidance suggests employers should go about it.

What does the Directive actually require?

The Directive does not require employers to undertake a particular form of job evaluation.

Instead, it requires pay structures to be based on objective, gender-neutral criteria such as skills, effort, responsibility and working conditions. It also requires Member States to make analytical tools and methodologies available to support employers in assessing work of equal value.

That distinction matters: the Directive requires objective outcomes, but it does not prescribe a single route for achieving them.

In practice, employers have a range of options available.

Some organisations use detailed analytical points-factor job evaluation systems. Others use job classification or career frameworks. Some adopt hybrid approaches that combine elements of both.

Provided the approach is objective, gender-neutral and capable of demonstrating why roles are grouped together, there is nothing in the Directive itself that says one methodology is inherently superior to another.

Slovakia has a clear favourite: analytical points-factor evaluation

Although the guidance acknowledges that different methodologies exist, there is little doubt about which approach it prefers.

The analytical points-factor method dominates the discussion.

The guidance explains in considerable detail how employers should identify evaluation factors, allocate weightings, score roles and document the resulting assessments. It is, in effect, a practical manual for implementing a points-factor job evaluation system.

There are good reasons why governments like analytical job evaluation.

When implemented well, it provides a structured and transparent way of comparing jobs that appear very different on the surface. It creates an audit trail showing why roles received particular scores. And because each role is assessed against the same predefined criteria, it can help reduce the influence of unconscious bias or historical assumptions about the relative value of different types of work.

These are precisely the kinds of objectives the Directive is trying to achieve.

The question is whether employers should infer from the guidance that points-factor job evaluation is now expected.

Has the guidance raised the perceived compliance burden?

Legally, probably not.

Practically, perhaps.

The guidance is explicit that it is not legally binding and that employers remain free to adopt other methodologies.

Yet it is difficult to ignore the practical effect of publishing more than a hundred pages of guidance, much of which is devoted to explaining one particular methodology.

That does not make alternative approaches non-compliant.

It does, however, raise an interesting question about expectations.

The Directive deliberately preserves flexibility. The guidance, perhaps inevitably, narrows the conversation towards a single preferred methodology.

Is analytical job evaluation always the right answer?

Not necessarily.

One of the risks with detailed government guidance is that employers begin to confuse the recommended approach with the required approach.

For some organisations, analytical points-factor evaluation will be entirely appropriate.

Large public sector employers, manufacturers and organisations with well-defined job families have been using such systems successfully for many years. Where there are thousands of employees and a diverse range of operational roles, the investment can be justified.

For others, the picture is less clear.

A 250-person software company with relatively few job families faces a very different challenge from a multinational manufacturer employing thousands of people across dozens of occupational groups. Likewise, a professional services firm with an established career framework may already have many of the building blocks needed to demonstrate objective pay structures without undertaking a full analytical evaluation of every role.

The Directive itself recognises proportionality in a number of places.

That is why it is difficult to conclude that a detailed points-factor exercise should become the default expectation for every employer, regardless of size, complexity or existing governance arrangements.

Ultimately, the legal question is not whether an employer has allocated 340 points rather than 327 points to a particular role.

It is whether the employer can demonstrate that its pay structures are based on objective, gender-neutral criteria, that work of equal value is rewarded equally, and that any differences can be objectively justified.

Analytical job evaluation is one way of achieving those objectives—but it is not obviously the only way.

That is perhaps the most interesting feature of the Slovak guidance.

In attempting to answer one of the Directive’s biggest practical questions, it also raises another.

Has Slovakia simply provided employers with a helpful implementation manual, or has it inadvertently encouraged them to believe that one methodology enjoys a privileged status that the Directive itself never intended to confer?

There may not yet be a definitive answer.

But it is a question that employers (and perhaps eventually the courts) are likely to return to.

Pay structures: much more than salary bands

One of the more interesting aspects of the guidance isn’t what it says about job evaluation. It’s what it says about pay structures.

Recently, a recurring question has been:

What exactly is a “pay structure”?

The Directive requires employers to have “pay structures” ensuring equal pay for equal work or work of equal value, based on objective, gender-neutral criteria.

But it never really explains what a pay structure is.

Many employers have therefore assumed that the answer is straightforward: salary bands.

The Slovak guidance takes a much broader view.

Rather than treating pay structures as simply a set of salary ranges, it describes them as the framework through which pay is determined across the organisation. Job evaluation, objective criteria, pay progression, remuneration policies and governance all become part of the same system.

That feels much closer to the underlying purpose of the Directive.

After all, salary bands alone don’t explain why people are paid what they are. Two employers could have identical pay ranges while reaching them in completely different ways. One might use robust, objective criteria. The other might rely almost entirely on managerial discretion.

The Directive is interested in the former, not merely the existence of the bands themselves.

This is likely to become an increasingly important distinction as employers prepare for compliance.

Organisations that have focused solely on designing salary ranges may discover that they also need to document the decision-making process sitting behind them. How are roles grouped? What determines progression? Which factors influence starting salary? When are market supplements available? Who approves exceptions?

Those are all questions about pay structures; salary bands are simply one output.

Objective justification becomes much more practical

The guidance is equally helpful when discussing objective justification.

The Directive permits differences in pay where they are based on objective, gender-neutral criteria. That principle is well established.

The difficulty has always been working out what that means in practice.

The Slovak guidance begins to answer that question.

Rather than listing abstract legal principles, it discusses the kinds of explanations employers routinely rely upon and considers the circumstances in which they may, or may not, be justified.

Perhaps the most useful example concerns market forces.

Market forces as an objective justification

Employers often explain pay differences by referring to recruitment difficulties or competitive labour markets. Anyone involved in reward will recognise the conversation.

“We had to pay more to hire this individual.”

Sometimes that’s entirely true.

The guidance doesn’t reject that argument.

Instead, it raises the evidential bar.

Market forces may justify differences in remuneration, but employers should be able to demonstrate that those market pressures genuinely existed, that the response was proportionate, and that similar situations are treated consistently.

That strikes a sensible balance.

The Directive was never intended to prevent employers responding to commercial realities. Labour markets differ. Skills shortages exist. Recruitment pressures fluctuate.

What it does seek to prevent is vague reliance on “market rates” becoming an all-purpose explanation for pay disparities that cannot otherwise be justified.

The same theme appears elsewhere in the guidance.

Length of service as an objective justification

Length of service, for example, is not treated as an automatic justification for higher pay. Instead, the guidance suggests that experience should only matter where it genuinely contributes to improved performance, capability or value to the organisation.

Again, that feels consistent with the underlying purpose of the Directive.

Pay differences should reflect objective differences in work or contribution, not simply historical practice.

Performance as an objective justification

Performance receives similar treatment.

The guidance does not discourage performance-related pay. Rather, it emphasises that performance criteria themselves should be objective, transparent and consistently applied.

That may sound obvious.

In reality, many organisations continue to rely on performance assessments that contain a significant degree of managerial discretion. Where those assessments influence pay, employers should expect increasing scrutiny over how those judgements are reached and documented.

Some conclusions on justification

Taken together, these examples reveal something important about the guidance.

It shifts the discussion away from legal theory and towards operational governance.

The question is no longer simply whether an employer can think of a reason for a pay difference.

It is whether that reason can be evidenced, applied consistently across the organisation and demonstrated to be genuinely objective.

That is a much more demanding exercise.

Small examples that may have wider consequences

The guidance also contains a number of smaller observations that are easily overlooked but could prove influential.

One concerns salary ranges in recruitment.

The Directive requires applicants to receive information about the initial pay or pay range for a role before interview. It says little, however, about how broad that range can be.

The Slovak guidance attempts to answer that question.

It suggests that employers should not use excessively wide salary ranges and gives the example of a €1,000–€7,000 monthly range as one that would undermine the purpose of pay transparency.

That example is not legally binding.

Nevertheless, it is one of the first practical attempts by a Member State to define what meaningful salary transparency might actually look like.

Whether other countries take a similar approach will be worth watching.

The guidance also adopts a broad interpretation of remuneration itself.

It reminds employers that remuneration can encompass a wide range of financial and non-financial benefits, including company cars, accommodation, mobile phones, travel benefits, study support and other advantages provided in connection with employment.

Again, that should not come as a surprise to anyone familiar with EU equal pay law, and the definition of pay in the PTD itself.

It serves as a useful reminder that equal pay assessments rarely stop at basic salary.

As employers begin preparing for the Directive’s reporting and information obligations, ensuring that all relevant elements of remuneration have been identified may prove just as important as analysing pay itself.

What the guidance still doesn’t answer

For all its strengths, the guidance also illustrates an important point.

Implementing the Pay Transparency Directive isn’t just a box ticking compliance exercise. It’s a status quo-shattering operational problem.

That means many of the most difficult questions are not questions of legal interpretation at all. They are questions of methodology, governance and data.

Perhaps inevitably, some of those questions remain unanswered.

Reporting methodology

The Directive’s reporting obligations have received enormous attention over the last two years.

Yet the Slovak guidance says relatively little about how employers should actually perform the calculations.

That’s understandable. Many of the reporting obligations will not apply for several years, and some technical questions may ultimately require further guidance at European level.

Nevertheless, employers are already grappling with practical issues such as:

How should part-year employees be treated? Should bonuses be allocated according to payment date or performance period? How should overseas assignments be handled? How should different working patterns be annualised? How should employers approach currency conversion within multinational groups?

These questions may not be particularly exciting.

But they are exactly the kinds of issues that determine whether two employers applying the same legislation ultimately produce comparable results.

For now, the guidance offers little assistance.

Proportionality

The guidance addresses proportionality within the methodology itself, noting, for example, that smaller employers can simplify the approach and use fewer sub-factors, but it says much less about the proportionality of overall implementation effort.

One of the strengths of the Directive is that it applies to organisations of vastly different sizes.

A multinational manufacturer employing 40,000 people presents very different implementation challenges from a technology company employing 250.

Yet much of the guidance inevitably describes what good practice looks like, rather than exploring how different employers might reasonably arrive at that destination.

That matters because implementation is not cost-free.

Designing objective pay structures, documenting evaluation criteria, consulting employee representatives and maintaining governance arrangements all require significant investment.

For many employers, the question is not whether to do these things, but how to do them proportionately.

The guidance leaves that judgement largely to employers themselves.

Looking beyond compliance

There is another omission, although whether it should be described as one is perhaps debatable.

The guidance focuses primarily on demonstrating compliance.

It says much less about using pay transparency as a diagnostic tool.

Many employers will want to go beyond simply identifying whether pay gaps exist. They will want to understand why they exist.

That is where statistical techniques such as regression analysis and decomposition become valuable. They help distinguish differences that are associated with role, experience or location from those that remain unexplained. They can also help organisations prioritise remediation efforts before reporting obligations begin.

The Directive does not require these techniques.

Nor should it.

But as employers become more sophisticated in their approach to pay transparency, these kinds of analytical tools are likely to become increasingly important. They represent the next stage of maturity beyond legal compliance.

Final thoughts: what employers should take from Slovakia’s guidance

It is easy to criticise government guidance. I’m guilty of this myself.

Too often government guidance can be late, and only muddy the water further.

Other guidance tries to answer every conceivable question and ends up creating rigid obligations that the law itself never intended.

The Slovak guidance largely avoids both extremes.

It is practical, thoughtful and clearly written. It engages with the difficult questions that employers are actually asking, particularly around job evaluation and objective justification. Even where readers disagree with its conclusions, there is considerable value in seeing how one Member State has approached the Directive’s more ambiguous concepts.

That said, employers should resist the temptation to treat it as the definitive interpretation of the Pay Transparency Directive.

It isn’t.

The guidance itself makes that clear.

Its strongest recommendations, particularly its clear preference for analytical points-factor job evaluation, should be understood as one possible route to compliance, not necessarily the only one. Employers should continue to start with the Directive itself, then consider whether their chosen approach delivers the objective, gender-neutral outcomes that the legislation requires.

Perhaps the most interesting question is what comes next.

If Slovakia remains unusual in publishing detailed implementation guidance, this document will be an interesting case study.

If other Member States follow suit, however, it may prove to be something more significant: the first glimpse of how the Directive evolves from a set of legal principles into day-to-day regulatory expectations.

That will be worth watching.

Because over the next few years, compliance with the Pay Transparency Directive is unlikely to be shaped by legislation alone.

It will increasingly be shaped by how governments, and ultimately courts, choose to fill in the gaps the Directive deliberately leaves behind.

Thanks to Dušan Nitschneider for sending the guidance over!

Slovak Ptd Methodology

1.45MB ∙ PDF file

Download

Download

Subscribe now

We tend to start PTD regression work with a fairly clean objective: estimate what the gender pay gap looks like once you account for the obvious differences inside a category of equal value work.

That category might be a grade, a job level, or some other internal grouping. The idea is that people inside it are broadly comparable in terms of job value.

So we run a regression. We control for tenure, performance, job family, location, sometimes other variables depending on how the model is specified.

And we get a result.

Usually a single coefficient. An adjusted gap. A percentage figure that looks like it answers the question.

But it doesn’t really answer the question people are trying to ask inside PTD analysis.

Because underneath that adjusted figure, there is still something simpler sitting there:

the actual difference in average pay between men and women inside the same category.

Regression doesn’t remove that. It just reframes it.

And at some point, especially when you are asked to explain the gap rather than just estimate it, that distinction starts to matter.

What regression is actually doing inside a PTD category

It helps to be precise here.

Inside a category, regression is answering something like:

if men and women had the same observable characteristics, would their pay still differ?

That is a conditional comparison. It holds things constant in order to isolate a relationship.

That is useful. But it also quietly changes the object you are looking at.

Because the category itself doesn’t disappear. It still contains structure: different mixes of tenure, different internal roles, different positioning within the same band.

And those differences are part of how the overall gap is formed.

Regression doesn’t separate that structure. It compresses it into a single adjusted estimate.

So what you end up with is not really an explanation of the gap. It is a cleaned comparison of outcomes under assumptions of equivalence.

That is often enough for modelling.

It is not always enough for explanation.

Why decomposition enters the picture

At some point, the question shifts.

Instead of asking:

is there still a gap after adjustment?

you start asking:

how is the gap inside this category actually being produced?

That is where decomposition methods, particularly Oaxaca-Blinder, become relevant.

They do something different from regression. Not more advanced, just different in intent.

Instead of adjusting the gap away, they break it apart.

Model selection does not disappear here — it becomes more important

It is easy to think that once you move from regression into decomposition, model selection (covered in part #3) becomes less central.

But it’s even more important.

Because Oaxaca-Blinder decomposition does not replace regression. It builds directly on it. Oaxaca-Blinder is just another way of doing regression.

In Oaxaca-Blinder, you are not just selecting one model. You are selecting two linked regression specifications:

the first is the standard wage model estimated separately for each group:

• which variables are included (tenure, job family, location, performance)

• how pay is modelled (log wages vs levels)

• how categories and interactions are specified

This is still standard regression model selection. It determines what “observable characteristics” actually mean in the decomposition.

But decomposition introduces a second modelling decision that is easy to miss.

Once you have estimated separate wage equations, you need a way to construct the counterfactual that decomposes the gap. This requires choosing a benchmark set of coefficients used to value differences in characteristics.

In practice, this is usually one of:

• the male wage equation

• the female wage equation

• a pooled wage equation estimated on the combined sample

This choice does not change the overall pay gap. But it does change how that gap is split into “explained” and “unexplained” components, because it changes the weights used to translate characteristics into pay units in the counterfactual.

So model selection here is not about defining a single “true” wage structure. It is about choosing the reference point for a hypothetical comparison.

In other words:

regression determines what relationships you estimate
decomposition determines how those relationships are used to construct a counterfactual breakdown of the gap

And both remain genuine modelling choices.

This is why model selection does not disappear in decomposition. It becomes layered: one layer for estimation, one layer for interpretation.

Why Oaxaca-Blinder cannot be used everywhere

It is also worth being clear about something that is often missed in applied PTD work.

Oaxaca-Blinder decomposition is not always available as a stable or meaningful option.

The reason is structural rather than technical.

Because once you move from a single regression to a decomposition, you are no longer estimating one model. You are effectively estimating two separate wage equations — one for each group inside the category — and then comparing them.

That creates an immediate constraint:

both groups must be large enough, and stable enough, for separate regressions to be meaningful.

Inside PTD categories, that is not always the case. I’ve covered this in part #4.

Some grades or job levels may contain:

• relatively small numbers of women

• relatively small numbers of men

• highly uneven distributions across subgroups

• or sparse combinations of controls (e.g. tenure × job family × location)

In those situations, splitting the sample into two regressions compounds the small-sample problem rather than solving it.

You are no longer dealing with one estimation problem. You are dealing with two, each with a reduced sample size, and then a third step that depends on the stability of both.

This matters because decomposition relies on coefficient stability.

If the underlying regressions are noisy, the decomposition will inherit that noise and often amplify it in the unexplained component.

So in practice, there is a threshold effect:

below a certain sample size or structural density, Oaxaca-Blinder becomes less informative than a single well-specified regression

This is why decomposition tends to work best in PTD analysis when:

• categories are sufficiently large

• internal composition is reasonably balanced

• and the regression specification is stable across subgroups

Where those conditions fail, the model selection decision often shifts back toward simpler regression approaches, because they are more robust to sparse data.

So decomposition is not a replacement for regression.

It is a refinement — but only when the data structure supports it.

What Oaxaca-Blinder decomposition is doing

At its core, Oaxaca-Blinder decomposition splits a difference in average outcomes into two components:

• differences in characteristics (composition)

• differences in returns to those characteristics (structure)

It does this using a counterfactual construction.

You don’t just observe the gap. You rebuild what one group’s pay would look like under another group’s structure.

So you are repeatedly asking versions of:

• what would women earn if they had the same characteristics as men?

• what would women earn if their characteristics were rewarded in the same way as men’s?

These are not real-world states. They are constructed comparisons.

But they allow a single observed gap to be broken into interpretable parts.

And that is the key difference from regression.

Regression isolates a relationship. Decomposition reconstructs the gap.

The explained component (within-category structure)

Once you do that decomposition, part of the gap is usually attributed to differences in characteristics inside the category.

This is the explained component.

In simple terms, it answers:

are men and women distributed differently within the same category?

Even within a single grade or job level, that can happen.

You might see one group slightly more concentrated in higher-tenure roles, or in roles at the top end of the band, or in business units that sit differently within internal pay structures.

None of that breaks the idea of equal value work. But it does mean the category is not internally uniform.

So the explained component is capturing composition inside the category.

It is important not to overread the word “explained” here. It does not mean justified. It simply means the gap is consistent with observable differences in characteristics.

It is a way of saying:

part of the gap is coming from how people are distributed inside the structure.

The unexplained component (what remains after structure is aligned)

After accounting for differences in characteristics, there is usually a residual.

This is the unexplained component.

It comes from comparing individuals who look similar on observable characteristics, inside the same category, and still finding a difference in pay.

At that point, you are no longer talking about composition. You are in the space of pay structure itself.

The unexplained component can reflect several things:

• differences in how characteristics are rewarded

• unobserved variables not captured in the model

• structural features of pay setting

• or simply limits of measurement and categorisation

But the most precise way to think about it is narrower than people often make it:

it is the part of the gap that cannot be accounted for by differences in observed characteristics inside the category

It is not a causal claim. It is not a diagnosis. It is the remainder after a structured counterfactual exercise.

Why this is not the same as regression

It is easy to think regression and decomposition are doing similar work. They use similar inputs. They often sit in the same workflow.

But they are answering different questions.

Regression asks:

if we hold characteristics constant, what is the difference between men and women inside this category?

That is a conditional expectation problem. It reduces variation to isolate a relationship.

Decomposition asks something different:

how is the total average difference inside this category actually made up?

That is a structural partitioning problem. It preserves the gap and splits it into components.

So regression simplifies. Decomposition reconstructs.

One is about isolating a clean estimate. The other is about explaining how a difference is assembled.

The reference group issue (why results can shift)

There is a quiet but important technical detail in decomposition that affects interpretation.

When you construct the decomposition, you have to choose a reference structure for returns to characteristics.

That could be:

• male wage structure

• female wage structure

• or a pooled structure

This choice changes how the gap is split between explained and unexplained components.

The total gap stays the same. But the internal breakdown does not.

That is known as the index number problem.

In practice, it means:

decomposition results are not purely mechanical — they depend on the benchmark used to define “returns”

It is one of those issues that doesn’t matter much in early analysis, but becomes important when you start using decomposition for explanation rather than illustration.

How results tend to behave inside PTD categories

Once applied inside PTD categories, decomposition results tend to fall into a few patterns.

Sometimes the result is fairly clean.

A small explained component. A small unexplained component. Both pointing in the same direction as the total gap.

That usually suggests the category is relatively internally consistent, and that most variation is compositional rather than structural.

But more often, things are less tidy.

You might see:

• explained larger than the total gap

• unexplained negative while the total gap is positive

• components pulling in opposite directions entirely

Those patterns are often where decomposition becomes most useful.

Because they show something regression cannot:

that different mechanisms inside the category can be operating in opposite directions at the same time.

Composition might be pushing the gap one way, while pay structure is pulling it back the other way.

Regression compresses that into a single coefficient. Decomposition leaves the tension visible.

What decomposition is not doing

It is worth being explicit about this, because decomposition is easy to overinterpret.

It is not:

• a test of discrimination

• a causal model of pay setting

• a fairness metric

• or a policy conclusion

What it is doing is more limited, but also more practically useful in PTD work:

it takes a single observed difference and shows what internal structure must exist for that difference to hold.

Nothing more than that.

But also nothing less.

Why this matters under the PTD

The PTD changes the level at which pay analysis operates.

You are no longer only describing company-wide pay gaps. You are explaining gaps inside categories of equal value work.

That shifts the question.

It becomes less about:

is there a gap?

and more about:

what is the structure of that gap inside this category?

Decomposition is useful here because it separates two things that regression keeps blended:

• who sits where inside the category

• how those positions are rewarded

Once you separate those, the category stops being a single number and starts behaving like a system.

And that is usually where the real explanation begins.

A worked example: reading an Oaxaca-Blinder output in R

It is easier to understand decomposition when you map it directly onto what the software actually returns.

Take a PTD category, for example Grade 6 roles in a fictional company.

In R, using a standard Oaxaca-Blinder implementation, the output might look like this:

Group 1 (Men) mean log wage:     3.21  
Group 2 (Women) mean log wage:   3.17  

Difference:                      0.04  

Explained:        0.055  
Unexplained:     -0.015  

So the first thing to notice is that the model starts from a very simple object:

the difference in average (often log) wages inside the category

That 0.04 is the raw gap. Everything else is a decomposition of it.

Step 1: the total gap (the only observed object)

The raw difference is:

0.04 log points (≈ +4%)

Inside this category, men earn more on average than women.

This is not modelled. It is directly computed from the data.

Decomposition explains how it is formed.

Step 2: the explained component (composition evaluated through model coefficients)

The explained part is:

0.055

In R terms, this is computed as something like:

(X̄_male − X̄_female) × β_reference

So what it is actually doing is:

• take differences in average characteristics inside the category

• weight them using a chosen coefficient structure

• translate them into a wage effect

So when we say:

explained = +5.5%

the correct interpretation is:

differences in observed characteristics inside the category, evaluated at the reference wage structure, would generate a +5.5% gap

In plain PTD language:

the way men and women are distributed across tenure, roles, and position within the same grade would, on its own, imply a larger gap than we actually observe

So composition is pushing the gap in a stronger direction than the observed outcome.

Step 3: the unexplained component (coefficient / structure effects)

The unexplained part is:

−0.015

In R terms, this comes from:

X̄ × (β_male − β_female)

So this is not “remaining noise”. It is:

• differences in estimated returns to characteristics

• differences in pay structures conditional on observables

• and any unobserved effects captured by the model

So the correct interpretation is:

conditional on observed characteristics, the structure of wage coefficients generates a −1.5% effect, favouring women

In plain language:

once we compare like-for-like individuals inside the category, the remaining pay structure offsets part of the gap in the opposite direction

This is why the total is only +4%.

Because:

• composition pushes up (+5.5%)

• structure pushes down (−1.5%)

Step 4: why this is not a contradiction

The identity is mechanical:

0.04 = 0.055 + (−0.015)

So there is no inconsistency. What looks like a contradiction is actually:

two different mechanisms producing offsetting effects inside the same category

Regression would typically compress this into a single adjusted coefficient.

Decomposition keeps the internal tension visible.

Step 5: what the R output is really telling you

If you translate the full output into plain English, it becomes:

Inside this grade, men and women differ in their observable characteristics in a way that would normally predict a larger pay gap than we actually see. However, once we compare like-for-like individuals, the remaining pay structure offsets part of that gap in the opposite direction.

That is the full story.

Everything else is just the decomposition machinery that produces it.

Final technical note (important for credibility)

One key detail in the R implementation is that:

the decomposition depends on the choice of reference coefficients (male, female, or pooled model)

So while:

• total gap is invariant

the split between:

• explained

• unexplained

is not uniquely defined.

So the correct framing is always:

“this is a decomposition under a specified reference wage structure”

not a universal partition.

Closing

Regression tells you whether a difference exists inside a category once you adjust for observable factors.

Decomposition tells you how that difference is actually constructed.

One simplifies the category to isolate a relationship.

The other reconstructs it to explain a structure.

And under the Pay Transparency Directive, that distinction becomes central.

Because the question is no longer just whether a gap exists inside a category of equal value work.

It is how that gap is built — and what internal structure is producing it.

In earlier articles in this series, we covered model specification, model selection, and robustness testing in Pay Transparency Directive (PTD) regression analysis.

Those pieces focused on how regression models are built in practice: which variables are included, how PTD categories of equal work are defined, and how sensitive results are to different specifications.

This article starts after that point.

Because once a regression model has been estimated, a more difficult question appears:

what does the result actually mean in a PTD context?

If a regression shows an unexplained pay gap within a category of equal work, has discrimination been identified? Or has the model simply reached the limits of what it can explain?

That distinction matters because the PTD shifts the focus away from reporting differences in pay and towards explaining them.

Statistical analysis sits directly in the middle of that shift.

Statistics answer “what”. The PTD asks “why”.

In pay transparency discussions, it is easy to blur the distinction between pay gaps and pay discrimination.

They are related concepts, but they are not the same.

A pay gap is a statistical difference in average pay. It tells us that one group earns more than another within a defined population or category.

Pay discrimination is a conclusion about causation. It is a judgement about why that difference exists.

That distinction is not semantic. It is structural.

Consider a PTD category of equal work where men earn an average of €70,000 and women earn €60,000.

There is a €10,000 gap within the category.

At this point, the data tells us only that a difference exists. It does not tell us why.

That gap may reflect differences in tenure, role allocation within the category, performance distribution, negotiation outcomes, organisational history, or discriminatory decision-making.

The raw statistical output itself cannot distinguish between these explanations.

This is where regression analysis enters the picture.

As discussed throughout this series, regression is used to estimate conditional differences in pay after controlling for observed variables such as tenure, location, and performance ratings. In practice, an employer might find that a €10,000 gap reduces to €1,000 after controls are applied.

At that point, the model is not describing a single cause. It is describing what remains unexplained by the variables included.

Under the PTD, that distinction becomes important. Employers are increasingly expected not only to report differences within categories of equal work, but to explain them. Regression helps describe structure. It does not complete the explanation.

What an “unexplained pay gap” actually means

The term “unexplained pay gap” is often treated as if it has a stronger meaning than it actually does.

In regression terms, an unexplained gap is simply the portion of the pay difference that remains after controlling for the variables included in the model.

Suppose a PTD regression estimates that women earn 7% less than comparable men within a category after controls.

That result does not establish discrimination.

It establishes that the model has not accounted for the full difference in pay.

There are several possible reasons for this:

• relevant variables may be missing from the model

• pay-relevant differences may not be fully measured

• the category structure may still contain heterogeneity

• organisational processes may be generating differences not captured in the data

• discriminatory effects may be present

The regression output cannot separate these explanations.

It only reports that a residual difference remains.

This is why, under the PTD, an unexplained gap is better understood as a trigger for further investigation rather than a conclusion.

It signals that the current explanation is incomplete.

It does not determine what the missing explanation is.

Why regression does not prove causation

This distinction between explanation and causation is central to how statistical models should be interpreted.

Regression analysis estimates relationships between variables conditional on the model specification. It does not identify causal mechanisms on its own.

A simple example illustrates the point.

A regression model would detect a relationship between the two variables.

But the relationship is not causal. Both are driven by a third factor: temperature.

The same issue arises in pay analysis.

A relationship between gender and pay within a PTD category does not, on its own, explain why that relationship exists.

Regression coefficients are conditional associations. They show how outcomes vary given the variables included in the model.

They do not show what caused those outcomes.

This matters under the PTD because the Directive requires employers to explain pay differences within categories of equal work. Statistical models can identify where differences persist. They cannot, on their own, establish why they persist.

The missing variable problem inside PTD categories

One of the key limitations of regression analysis is omitted variable bias.

This occurs when a factor that affects pay is not included in the model.

Consider a PTD category containing senior HR business partners.

Men in the category earn more on average than women. A regression is run controlling for tenure, location, and performance ratings. A gap remains.

At first glance, this may appear to indicate an unexplained gender effect.

But the result depends on whether all relevant pay drivers have been included.

If specialist skills, responsibility for major accounts, or internal role complexity are not properly captured, part of the observed difference may be incorrectly attributed.

The model is not wrong. It is incomplete.

This is a key constraint in PTD analysis. Categories of equal work are necessarily broad. Within them, pay may still vary for legitimate reasons that are difficult to measure precisely.

A regression model can only operate within the structure it is given.

When regression hides the issue

There is also a less intuitive limitation.

Sometimes the issue does not sit within PTD categories. It sits between them.

Imagine an organisation where pay within each category is tightly controlled. A regression analysis finds little or no gender pay gap within categories.

On that basis, the organisation may conclude that pay equity is broadly achieved.

However, further analysis shows that women are promoted into higher-paying categories less frequently than men.

The regression result is accurate within its scope. But the scope itself is limited.

This is important under the PTD because the Directive focuses on categories of equal work. But organisational inequality can also arise through movement between categories.

Statistical models do not automatically capture that dynamic unless it is explicitly included in the analysis.

A clean within-category result does not necessarily imply a clean overall system.

Can statistical analysis be used as evidence under the PTD?

Yes.

Statistical analysis is likely to play a significant role in PTD compliance and pay equity assessment.

It can identify pay differences within categories, support internal review processes, and inform responses to employee information requests.

But it does not replace explanation.

A regression result showing a 4% unexplained gap within a category does not answer the key question on its own.

It identifies where further questions should be asked:

• how are starting salaries determined within the category?

• how are progression decisions made?

• are performance ratings applied consistently?

• how are allowances and supplements allocated?

The value of statistical analysis under the PTD is not that it provides conclusions. It is that it helps locate where explanation is needed.

What this means for Pay Transparency Directive compliance

The PTD changes the centre of gravity in pay analysis.

Historically, organisations focused on measuring pay gaps.

The Directive moves the focus towards explaining pay differences within categories of equal work.

Regression analysis supports that shift. It helps structure comparisons and control for observable differences.

But it does not resolve the underlying question of justification.

If a model identifies an unexplained gap, the appropriate response is not immediate conclusion. It is further investigation.

That investigation sits outside the regression model. It focuses on organisational processes: how pay is set, how decisions are made, and how those decisions vary across people performing the same work.

Closing

Statistical analysis is effective at identifying patterns in pay data within PTD categories of equal work.

It is less effective at explaining why those patterns exist.

That limitation is not a flaw in the method. It is a reflection of what statistical models are designed to do.

Under the Pay Transparency Directive, that distinction becomes central.

The Directive does not treat statistical output as an answer. It treats it as a starting point for explanation.

Regression can identify where differences remain after controls.

It cannot determine whether those differences are justified.

Statistics answer what happened.

The PTD requires employers to explain why.

In earlier articles in this series, we covered model specification and model selection in Pay Transparency Directive (PTD) regression analysis.

That work focused on how you build a gender pay gap regression model in the first place:

• which variables you include (tenure, role, location, performance ratings)

• how you define within-grade (category) comparisons under the PTD framework

• and how different regression specifications change estimated pay gaps

In other words, those articles were about construction.

You were deciding how to build the statistical machine.

This article starts after that point.

Because once a regression model is built, a more uncomfortable question appears:

what happens if the model is slightly wrong?

That is where robustness checks and sensitivity analysis come in.

They sit at the core of serious pay equity analysis, equal pay modelling, and PTD pay transparency reporting, because they determine whether your conclusion is stable or fragile.

Subscribe now

Specification vs robustness: why this distinction matters in pay gap analysis

It is easy to treat model specification, model selection, and robustness checks as variations of the same thing. In practice, they are not.

Think of it like building a bridge (or a jenga tower).

• Model specification (from earlier articles) is the design phase
  You decide what the structure looks like: which pieces exist, what carries weight, how it is assembled

• Robustness and sensitivity analysis (this article) is the stress testing phase
  You test what happens under wind, load, vibration, and imperfect assumptions

Both use the same statistical tools you’ll be familiar with by now: regression models, control variables, estimation techniques. But they answer different questions.

One asks:

“What model should we use to estimate the gender pay gap within PTD categories?”

The other asks:

“If we change reasonable assumptions, does the pay gap conclusion still hold?”

That distinction is central to regression-based pay equity analysis.

Because a model that only works under one very specific set of assumptions is not yet a defensible explanation. It is just a fragile fit.

Why robustness matters in Pay Transparency Directive (PTD) analysis

Under the PTD framework, the key analytical unit is the category (grade or job level of equal value).

So when we estimate a within-grade gender pay gap using regression analysis, we are not simply describing pay differences.

We are making a structured claim about inequality inside a defined group of comparable roles.

That claim must survive reasonable scrutiny.

In practice, that means testing whether the estimated gender pay gap within a grade remains stable when we:

• change regression specifications (linear vs log-linear pay models)

• adjust control variables (tenure, job family, location, performance ratings)

• modify sample definitions (outlier removal, minimum group size thresholds)

• or apply different standard error approaches (robust or clustered standard errors)

This is not just tidying up. It’s not just statistical hygiene.

It is part of defensible pay transparency reporting.

Because employee representatives are the ones asking whether your model is elegant:

“If we look at this differently, does the explanation still hold?”

Fragile results: when the pay gap depends on the model

A fragile result is one that changes materially under reasonable analytical adjustments.

In practice, this might look like:

• a 6% within-grade gender pay gap disappears when tenure is added as a control variable

• the estimated gap flips direction when switching from raw pay to log-transformed pay

• the result is driven almost entirely by one subgroup within a category (for example, a single department or job family)

• statistical significance disappears when using robust or clustered standard errors

In regression terms, this often signals sensitivity to:

• omitted variable bias (important factors not included in the model)

• functional form assumptions (how variables are mathematically structured)

• sample composition effects (who is included in the dataset)

• or influential observations (a small number of data points driving the result)

A useful way that I like to think about it is this:

a fragile model is like a structure balanced on uneven ground. It holds until you shift perspective slightly

That does not automatically make it wrong. But it does mean the explanation is not yet stable enough for PTD-level scrutiny.

Robustness checks: testing whether the result survives stress

Robustness checks are structured tests that evaluate whether a regression result holds under different reasonable modelling choices.

They sit within the broader toolkit of regression diagnostics, model validation, and statistical inference testing.

I think there are three core types in gender pay gap analysis.

1. Specification robustness (does the regression model change the story?)

Specification robustness tests whether the estimated gender pay gap within PTD categories depends on how the regression model is built.

This typically involves:

• adding or removing control variables (tenure, performance ratings, role level)

• changing functional form (linear vs log-linear wage models)

• introducing interaction terms (e.g. gender × grade effects)

• using alternative estimation methods (OLS vs robust regression techniques)

The key question is:

does the estimated within-grade pay gap remain directionally stable across reasonable model specifications?

If yes, the result is structurally robust.

If not, the result is highly specification-dependent.

To return to the metaphor:

you are testing whether the bridge still stands when you change the materials slightly

2. Sample robustness (is the result driven by a small part of the data?)

Sample robustness tests whether the estimated pay gap is sensitive to who is included in the dataset.

This includes:

• removing outliers in pay distributions

• applying minimum sample size thresholds within grades (PTD categories)

• excluding recent hires or short-tenure employees

• testing whether a small subgroup dominates the estimate

This links directly to statistical concepts such as:

• leverage effects in regression

• heteroskedasticity in wage data

• and influential observations in small samples

This matters especially in PTD analysis, because:

• senior grades often have small sample sizes

• categories can be unevenly distributed by gender

• and individual observations can disproportionately affect results

A useful metaphor here:

you are checking whether one unusually heavy brick is holding up the entire structure

3. Subgroup robustness (is the effect consistent within categories?)

Subgroup robustness tests whether the within-grade gender pay gap is consistent across internal divisions.

For example:

• departments or business units

• job families within the same PTD category

• geographic locations

• tenure bands

This connects to interaction effects and heterogeneous treatment effects in regression modelling.

The key question is:

is the gender pay gap a consistent feature of the category, or a localised pattern within it?

This distinction is critical in PTD analysis.

Because:

• a consistent effect suggests structural inequality within a grade

• a localised effect suggests a more specific organisational driver

Metaphorically:

it is the difference between a pattern across the whole map, and a stain in one corner of it

Sensitivity analysis: how much does the result move when assumptions change?

Where robustness checks are discrete (change X → observe outcome), sensitivity analysis is continuous.

It examines how the estimated pay gap responds when assumptions are gradually adjusted.

This includes:

• shifting category definitions (PTD grade boundaries)

• varying inclusion thresholds for employees

• adjusting pay definitions (base pay vs total compensation vs bonus-included pay)

• progressively adding or removing control variables in regression models

In statistical terms, this relates to:

• model uncertainty

• specification uncertainty

• and stability of regression coefficients under perturbation

A useful framing:

robustness asks “does it change?”
sensitivity asks “how much does it change, and how quickly?”

A good result does not need to be invariant.

But it should not collapse under small, reasonable changes in assumption.

A practical workflow for robustness in PTD pay gap analysis

In applied pay equity regression analysis, a typical workflow looks like:

1. Baseline model
   Estimate within-grade gender pay gap using a standard regression model

2. Specification testing
   Vary controls, functional forms, and estimation methods

3. Sample stress testing
   Remove outliers, adjust category definitions, test minimum group sizes

4. Subgroup analysis
   Test consistency across departments, job families, and tenure bands

5. Sensitivity mapping
   Track how the estimated pay gap changes across specifications

6. Synthesis for reporting
   Translate statistical variation into a clear explanation for employee representatives

This final step is often the most important.

Because in PTD work, the output is not just a coefficient.

It is an explanation of structure.

Communicating robustness to employee representatives

This is where statistical analysis becomes organisationally meaningful.

Avoid:

• “the model is statistically significant”

• “robust standard errors were applied”

• “multiple regression specifications were tested”

Instead, translate into structural language:

• “we tested whether the result changes under reasonable alternative assumptions”

• “the within-grade gender pay gap remains broadly consistent across specifications”

• “where variation occurs, it is driven by identifiable subgroups or data sensitivity”

The goal is not statistical precision.

It is interpretability under scrutiny.

Think of it as translating from regression output into a story about organisational structure.

What counts as a robust result?

There is no single threshold in equal pay statistical analysis that defines robustness.

But in practice, robust results typically show:

• consistent direction of the gender pay gap across specifications

• stable magnitude under reasonable model changes

• identifiable drivers of any variation

• and persistence within PTD categories (grades)

A simple classification helps:

• Strong robustness: stable across all reasonable checks

• Moderate robustness: some variation, but a consistent underlying pattern

• Weak robustness: high sensitivity to specification or sample changes

Importantly:

robustness is not about eliminating uncertainty — it is about understanding it

Common pitfalls in robustness and sensitivity analysis

Even technically strong analysts often make the same mistakes:

• running robustness checks but not integrating them into the narrative

• selectively reporting only stable specifications

• treating robustness as an appendix exercise rather than a core part of inference

• ignoring instability instead of explaining its structural causes

• over-interpreting fragile regression coefficients

From a PTD perspective, the real failure is not technical.

It is communicative:

failing to turn statistical sensitivity into a defensible explanation of pay structure

Closing: robustness is where analysis becomes defensible

Robustness checks and sensitivity analysis are sometimes presented as technical validation steps.

In Pay Transparency Directive analysis, they are something more important.

They are the mechanism that turns a regression output into a defensible explanation of pay differences within categories of equal work.

The question is not whether your model is sophisticated.

It is whether your conclusion still holds when someone reasonably asks:

“What changes if we look at it differently?”

That is the only robustness test that ultimately matters.

France has published a revised draft of its Pay Transparency Directive legislation.

And unlike some of the more incremental revisions emerging across Europe, this one contains several genuinely significant developments that will shape how the regime operates in practice.

The headlines are easy to spot.

The draft introduces:

• new criminal sanctions of up to two years’ imprisonment and a €7,500 fine for certain repeated equal pay breaches;

• exclusion from public procurement for employers sanctioned for specified pay transparency failures;

• a new employer defence against abusive information requests;

• broader use of statistical and hypothetical comparators in equal pay claims; and

• a more granular and phased implementation timetable, particularly for smaller employers.

Those changes will attract most of the immediate attention.

Before getting into the detail, it is worth noting that this is the second major French PTD draft in recent months (my thoughts on that are here).

France increasingly appears to view categorisation not as an HR classification exercise, but as the operational foundation of Pay Transparency Directive compliance. It’s gone far beyond the minimum PTD requirements.

But before turning to that broader shift, it is worth unpacking the most immediately consequential (and most headline grabbing) changes.

Criminal sanctions: equal pay breaches move into criminal law territory

The June draft introduces a clearer criminal enforcement layer that was largely absent from the March version.

Where the same employer commits equal pay offences against multiple individuals, penalties may now reach:

• two years’ imprisonment, and

• a €7,500 fine.

Criminal sanctions are unlikely to affect most employers in day-to-day compliance terms.

But they matter structurally.

They signal that certain equal pay breaches are now being treated not only as regulatory failures, but as conduct with a criminal law dimension.

That shift is likely to influence both enforcement posture and public perception of risk.

Public procurement exclusion: compliance now carries a commercial consequence

The revised draft also introduces exclusion from public procurement procedures for certain pay transparency breaches.

Where employers are sanctioned for specified failures, they may be excluded from public contracts for one year.

This is a meaningful expansion of enforcement tools.

Across multiple EU regulatory domains, access to public procurement is increasingly used as a behavioural lever.

France now explicitly imports that mechanism into pay transparency enforcement.

For organisations with public sector exposure, the practical significance may lie less in fines and more in loss of market access.

France joins Greece on “abusive requests”: an emerging implementation pattern

One of the more interesting additions is also one of the least prominent.

The June draft introduces an explicit defence against abusive requests for information or explanations.

Employers are not required to respond where requests are abusive, particularly due to:

• their number, or

• their repetitive or systematic nature.

At first glance, this appears procedural.

However, it is notable that France is not acting in isolation.

The recent Greek draft on Pay Transparency Directive implementation contains a similar mechanism allowing employers to refuse abusive or repetitive requests.

This does not yet establish a formal EU-wide trend.

But two in two weeks does suggest a converging implementation concern:

how transparency rights should operate when exercised repeatedly, systematically, or at scale.

Italy did something altogether different but thematically similar: a cap on the number of requests (can only be used once in 12 months).

Member States arriving at similar safeguards independently is not determinative of a trend — but it is analytically significant.

It may reflect the first signs of legislators building operational friction controls into transparency regimes.

Beyond the headlines: France is building a system around categorisation

Once headline enforcement changes are set aside, a deeper structural pattern becomes visible.

The June draft significantly expands the definition of work of equal value, anchoring it in a broader set of objective, gender-neutral criteria, including:

• professional knowledge and qualifications;

• skills derived from experience;

• soft skills;

• responsibilities;

• working conditions; and

• physical or nervous load.

This is important not only because the definition is broader, but because it is now clearly designed to be operationalised through categorisation systems.

Categorisation is becoming the backbone of the regime

The March draft already established a hierarchical structure for categorising employees performing work of equal value.

The June draft refines and strengthens that structure.

In summary, categorisation may be established through:

• company-level agreement (preferred route);

• branch-level agreement;

• application of a branch agreement following failed company negotiations; or

• unilateral employer categorisation following CSE consultation (for a limited period).

This creates a more flexible but also more consequential system.

France quietly expands the evidential toolbox for equal pay claims

The June draft also expands the evidential basis available in equal pay disputes.

Where no direct comparator exists, claimants may rely on:

• statistical evidence; and

• hypothetical comparators.

This matters most in contexts where traditional comparators are structurally limited:

• senior leadership roles;

• small organisations;

• highly specialised job functions; and

• occupationally segregated environments.

The practical effect is to reduce reliance on strict comparator availability as a gating issue in litigation.

What is a hypothetical comparator?

A hypothetical comparator in an equal pay case is where a claimant argues they were paid less than they should have been by comparing their situation not to a real colleague doing the same job, but to a fictional employee who would be treated differently in a comparable role or scenario. Instead of pointing to an actual named comparator, the court is asked to assess how a notional person of the opposite sex (or in a comparable position) would likely have been paid if they had been doing substantially similar work or work of equal value. It matters because it allows equal pay claims to proceed even where there is no obvious real-world comparator, which is especially important in workplaces where roles are unique, highly specialised, or gender-segregated.

Enforcement is becoming sharper — but not uniformly harsher

The enforcement architecture is evolving in two directions simultaneously.

On the “hardening” side:

• criminal sanctions;

• procurement exclusion;

• expanded evidential tools.

On the “softening / balancing” side:

• abusive request defence;

• refined privacy-based disclosure limits;

• movement of some breaches into lower fixed-fine categories.

The result is not a uniformly stricter regime.

It is a more selective enforcement architecture, targeting systemic or repeated breaches more explicitly while introducing procedural safeguards elsewhere.

A subtle wording shift that may matter later

One small but recurring change appears throughout the draft.

References to “objective and non-sexist criteria” have largely been replaced with:

“objective criteria and not based on sex”.

This may ultimately prove purely stylistic. However, it also more closely aligns with standard EU anti-discrimination drafting language.

In interpretive terms, alignment with established legal vocabulary can matter in borderline cases. And this is particularly important in the murky grey period during which the PTD isn’t yet implemented.

Final thoughts

The June draft still leaves important questions unresolved.

Key thresholds, indicators, and operational details remain dependent on future decrees. The legislation must also complete its passage through the Conseil d’État and Parliament.

But the direction of travel is increasingly coherent.

The March draft largely established the architecture of France’s Pay Transparency Directive implementation.

The June draft begins to operationalise it.

France PTD draft - June 2026

454KB ∙ PDF file

Download

Download

Most Member States implementing the EU Pay Transparency Directive are doing so in broadly familiar ways.

Draft legislation. Consultation periods. Phased timelines that push meaningful impact into 2027 or later.

Malta has taken a different route.

There is no extended consultation cycle (at least not publicly). No long policy runway. No staged implementation narrative.

Instead, the law was published on 5 June 2026 and entered into force on 7 June 2026.

Alongside Italy, Slovakia and Lithuania, Malta is among the first wave of full implementing jurisdictions.

The legislation leaves almost no time between publication and application.

But the more interesting feature is not simply how quickly the law arrived.

It is that the legislation creates immediate operational obligations for employers while also containing a number of unusual enforcement and transitional provisions.

The result is a regime that is more complex than a simple “early implementation” story might suggest.

Immediate effect, but a more complex enforcement picture

Much of the attention naturally focuses on the right to information.

That is where Malta is genuinely fast.

Workers may request:

• their own pay level; and

• average pay levels for workers performing the same work or work of equal value, broken down by sex.

Employers are expected to respond within 8 days (likely calendar days, given that the Regulations expressly use the term “working days” elsewhere when that is intended).

That deadline is unusually short by European standards.

However, it should not be confused with the ultimate enforcement position.

The legislation also creates roles for worker representatives and the Equality Body where information is not provided, although the precise interaction between those mechanisms is not always clear from the drafting.

What is clear is that a failure to provide the required information within 45 days of the original request may trigger enforcement consequences.

In practice, will employers treat 45 days as the real deadline? The operational challenge remains ensuring that requests can be answered quickly enough that disputes never progress further.

The right to information is operationalised — but important questions remain

The more relevant point is how the right is structured in practice.

Employers must disclose:

• individual pay level; and

• average pay for the relevant category.

Malta also addresses a point that has been uncertain in several early PTD implementations.

For initial application, the reference dataset is explicitly tied to 2026 pay information.

That provides more certainty than some other emerging regimes.

It does, however, create practical questions where requests are made part-way through the year. Because the legislation relies on concepts such as gross annual pay and corresponding gross hourly pay, employers may need to determine how incomplete-year data should be treated.

In practice, this may become particularly relevant where bonuses or other variable pay elements have not yet been paid when the request is made.

What Malta has done is reduce one area of uncertainty. It has not removed the need for employers to develop consistent internal approaches to pay calculations.

Pay structures move into the compliance sphere

Malta’s approach to pay structures is less prescriptive than some other emerging PTD regimes, but it still moves pay architecture firmly into the compliance sphere.

The Regulations require employers to maintain pay structures capable of ensuring equal pay for equal work or work of equal value. The criteria used must be objective and gender-neutral, while transparency provisions require written pay-setting criteria to be maintained and made accessible to workers.

In practice, this means many employers will need to move beyond informal manager discretion and ensure that pay-setting criteria are documented somewhere employees can actually access them.

For organisations with mature reward frameworks, that may be relatively straightforward.

For others, it could expose gaps that have never previously carried legal significance.

Categorisation becomes a live compliance issue

Malta largely follows the Directive’s framework for assessing work of equal value, relying on skills, effort, responsibility, working conditions and other job-related factors.

The Regulations also require objective and gender-neutral criteria to be agreed with worker representatives where such representatives exist.

Worker representatives are defined broadly and may include trade union representatives, elected representatives and other representatives acting in defence of workers’ interests.

In practice, this means employers will need to consider whether existing job evaluation methodologies, competency frameworks and role-classification approaches can be clearly mapped against those factors.

Categorisation therefore becomes more than an HR exercise.

It becomes part of the legal framework supporting equal pay compliance.

Recruitment processes move into scope

Malta’s recruitment transparency provisions are broadly consistent with the Directive.

Applicants must receive information about the initial pay level or pay range, together with any relevant collective bargaining provisions, before the recruitment process concludes. Questions about salary history are prohibited, while job advertisements and recruitment processes must remain gender-neutral.

In practice, this means pay transparency compliance is not solely a reward issue.

Hiring managers, recruiters and talent acquisition teams will also need clear guidance on what information must be provided and when.

Existing collective agreements receive significant protection

One of the most unusual features of the Maltese legislation concerns existing collective agreements.

The Regulations provide that collective agreements already in force when the legislation commenced are deemed compliant with the new regime. During their lifetime, employers are largely protected from claims arising from terms contained within those agreements.

That protection is not permanent. It falls away once the collective agreement expires, and claims relating to breaches of the agreement itself remain possible.

It is a significant transitional measure.

In practice, employers operating under collective agreements may face a materially different risk profile from employers relying solely on internal pay structures.

It is also one of the aspects of the Maltese implementation most likely to attract scrutiny from a Directive compliance perspective.

Pay transparency and pay secrecy do not align perfectly

Malta prohibits contractual restrictions that prevent workers disclosing their pay for the purpose of enforcing equal pay rights.

Employers may, however, require information relating to other workers’ pay to be used only for purposes connected with exercising those rights.

The more interesting issue is what the legislation does not expressly say.

The Directive protects workers from being prevented from disclosing their pay for the purpose of enforcing equal pay. The Maltese drafting focuses more narrowly on disclosures made to authorities, worker representatives and trade unions.

Malta’s narrow drafting raises questions about how far workers are protected when discussing pay directly with colleagues in order to explore potential equal pay concerns.

Whether that distinction proves meaningful will likely depend on how the legislation is interpreted and applied.

Immediate obligations do not always mean stronger enforcement

While Malta creates immediate operational obligations for employers, several aspects of the legislation are notably employer-friendly.

Existing collective agreements receive substantial transitional protection. Compensation rules may limit recovery beyond lost pay unless discrimination is both wilful and vexatious. Financial penalties are relatively modest compared with those emerging elsewhere in Europe.

There are also questions around the scope of comparator provisions and the threshold a claimant must satisfy before the burden of proof shifts.

In practice, this means the legislation combines rapid compliance obligations with a number of safeguards that may limit enforcement exposure in certain circumstances.

The result is a framework that is more nuanced than either a purely worker-friendly or employer-friendly characterisation would suggest.

Final thought

Malta will often be described as one of the first Member States to implement the Pay Transparency Directive in full.

That is true, but it is only part of the story.

The more interesting feature is that the legislation creates immediate operational obligations while simultaneously incorporating a number of significant transitional protections and enforcement limitations.

For employers, the challenge is therefore not simply understanding the law.

It is determining whether recruitment processes, pay-setting practices, job architecture and information-request procedures are already capable of operating under the new rules.

In Malta, that question is no longer theoretical.

It is already live.

The new law is available here.

One of the more unintuitive consequences of the Pay Transparency Directive (PTD) is that employers may start to see statistically significant gender effects in places where no real gender effect exists.

That sounds wrong at first. If something is “statistically significant”, surely it must mean gender is influencing pay.

Not quite.

The issue sits in how PTD-style analysis is actually run. Employers are not looking at pay in one unified model. They are required to break the workforce into categories and run separate analyses within each.

That usually means multiple regression models, each operating on a smaller slice of the organisation.

A company with 20 grades is effectively running 20 different statistical tests. Each one is valid. The complication is what happens when you look at them together.

Why PTD Analysis Turns One Question into Many

What the regression is actually doing

At category level, each model is trying to answer a fairly simple question:

once we account for relevant factors, is there evidence that gender is associated with pay differences in this group?

That sounds precise. And in a single model, it usually is.

But PTD analysis doesn’t stop at one model. It repeats the same exercise across multiple parts of the organisation, often with relatively small sample sizes once the data is split.

And this is where interpretation starts to shift. Not because the model is wrong, but because the information becomes more fragmented.

You are no longer looking at one estimate. You are looking at many.

Statistical Significance in Pay Gap Analysis

What it actually means (and what it doesn’t)

Statistical significance is often treated as a signal of importance. In practice, it is much narrower.

A result is usually labelled “significant” if it would be unlikely under a specific assumption of no effect, often using a 5% threshold.

That does not mean:

• the effect is large

• the effect is meaningful

• or that gender is genuinely driving pay differences

It simply means: the observed pattern would be unusual if there were no difference at all.

That distinction matters more than it tends to in pay equity discussions.

Why the 5% threshold becomes fragile in practice

A 5% threshold sounds reassuringly strict. But it only really behaves well when you run a small number of tests.

If you run many tests, the logic changes. Even when no real effect exists, some results will cross that threshold just by chance.

For a single regression, that is usually manageable.

For PTD-style reporting, it becomes harder to ignore.

The Multiple Comparisons Problem in PTD Reporting

Why repetition changes what you see

Take a simple example.

Imagine an employer with 20 grades and no underlying gender effect anywhere in the organisation. Pay differences exist, but they are fully explained by legitimate structural factors.

Now run a regression in each grade.

Even in that fair system, you would still expect a small number of “significant” results to appear just by chance.

Not because something is wrong.

But because statistical testing is designed to allow for a small rate of false positives.

When you repeat that test 20 times, those small probabilities accumulate.

You don’t need anything unusual in the data for that to happen.

Why PTD makes this more visible

The PTD pushes exactly this kind of structure:

• more granular categories

• more segmentation

• more reporting of sub-results

That improves transparency. But it also increases the number of statistical outputs that can be misread in isolation.

Why Category Results Can Be Misleading

The pattern problem

Suppose an employer sees the following:

• Grade 2: significant

• Grade 3: not significant

• Grade 4: significant

• Grade 5: not significant

The natural reaction is to start building explanations.

Something must be happening in Grade 2. Something must be happening in Grade 4.

And sometimes that is right.

But the same pattern can also arise in a system with no structural gender effect at all.

That is the uncomfortable part: the output does not distinguish between “real structure” and “statistical noise”.

It only tells you what is unlikely under a particular assumption.

Fragile Inference in Small PTD Samples

Why slicing the data changes the reliability

The deeper issue is not just multiple testing.

It is that category-level PTD regressions often involve small samples once the workforce is split.

That creates:

• unstable coefficients

• wider uncertainty

• higher sensitivity to small changes in data or specification

In that environment, results can move around quite easily.

A “significant” result in one grade may reflect a genuine structural issue. Or it may reflect nothing more than noise amplified by a small sample.

The output alone doesn’t tell you which.

Why This Trade-Off Is Built Into the PTD

None of this is a flaw in the Directive.

One of its main purposes is to force organisations to look beyond overall averages. Aggregate figures can hide variation that matters.

But that shift inevitably creates a different problem: more granular analysis produces more uncertain inference.

So you get better visibility, but noisier signals.

That trade-off is unavoidable.

How to Interpret Statistical Significance Properly

The question people usually ask

When a category comes back as significant, the instinctive question is:

why is this happening here?

That is understandable, but it can lead you too quickly into story-building.

A more useful question

A better starting point is:

how unusual is this pattern, given how many tests we have run and how the data has been split?

That reframes the result. You are no longer treating it as a standalone finding. You are placing it in context.

What stronger signals tend to look like

When there is a real structural issue, you usually see it:

• across multiple related grades or roles

• consistently across different model specifications

• and reflected in meaningful effect sizes, not just significance labels

Random variation does not behave like that. It is scattered, inconsistent, and rarely stable when you look at it from different angles.

Four Practical Checks for PTD Results

When reviewing category-level outputs, four questions tend to be useful:

1. How many tests are we actually looking at?
More tests mean more chance findings.

2. Do results cluster or appear in isolation?
Structure tends to cluster. Noise tends not to.

3. Do findings survive reasonable model changes?
Instability is a warning sign.

4. Is the effect size meaningful, or just statistically detectable?
These are not the same thing.

The Key Misinterpretation Risk

The biggest mistake in PTD analysis is treating statistical significance as proof of a structural gender issue.

That is not what it represents.

A regression tells you whether the data is consistent with a difference after controls. It does not tell you why that difference exists, or whether it reflects a systemic problem.

And once you start running many regressions on fragmented data, some significant results become expected, even in a fair system.

The challenge is recognising that distinction in practice.

Final Takeaway

The Pay Transparency Directive forces a more detailed view of pay structures than most organisations have previously worked with.

That is a good thing. It surfaces variation that aggregate numbers can hide.

But it also introduces a less comfortable reality: more detailed analysis produces more ambiguous signals.

Some statistically significant gender effects will reflect genuine structural issues. Others will be artefacts of repetition, fragmentation, and chance.

The difficulty is that they often look identical when you first see them.

The job of interpretation is not to react to significance in isolation, but to understand what remains once you step back and look at the full pattern.

On 7 June 2026, the deadline for Member States to implement the EU Pay Transparency Directive (“PTD”) expires.

In practice, that moment does not switch the PTD into full national law where implementation is missing.

What it does instead is shift the legal environment into a transitional phase where EU law starts to influence outcomes through interpretation, litigation strategy, and State liability principles.

From that point onwards:

• Public sector employers can, in some cases, face direct reliance on certain PTD provisions (“vertical direct effect”, and there’s more on this below - stay with me!).

• National courts must interpret domestic law, where possible, in line with the Directive (Von Colson, Marleasing).

• Private employers are generally not directly bound by the Directive itself, but are affected through interpretation of existing national law and equal pay frameworks (Marshall).

• The PTD becomes practically relevant in litigation across jurisdictions such as Germany, France, Spain, Sweden, Belgium, Estonia and the Netherlands, even where transposition is incomplete.

The important shift is not legal activation, but legal influence.

What the EU Pay Transparency Directive actually is

The Pay Transparency Directive is part of the EU’s wider equal pay framework. Its purpose is straightforward: reduce gender pay gaps by forcing greater transparency around how pay is set, disclosed, and justified. It’s really about giving people everything they need to enforce a fundamental EU right: the right to equal pay.

In practice, it affects three core areas of employment law:

• how pay is communicated during recruitment

• how employees access information about pay structures

• how employers must justify pay differences and report gaps

It sits alongside existing EU equal pay principles but strengthens them by increasing procedural transparency and evidential accessibility.

What changes if a Member State misses the deadline?

After 7 June 2026, a State that has not implemented the PTD is in breach of EU obligations. That much is clear.

But the legal consequences do not stop there.

Three things happen almost immediately:

• First, national courts remain bound by EU law principles when interpreting domestic legislation.

• Second, employees (and their lawyers) begin testing whether specific Directive provisions can be relied upon directly against public bodies.

• Third, litigation strategy starts to shift, even before formal national legislation catches up.

The Directive does not disappear into the background. It starts to operate indirectly through existing legal structures.

Does the Directive become law automatically?

No. There is no automatic conversion of an unimplemented directive into national legislation (EU Regulations, like the GDPR, are part of national law automatically upon coming into force).

However, EU law prevents Member States from treating non-implementation as a legal shield.

Instead, enforcement operates through established doctrines developed by the Court of Justice of the European Union, including:

• Van Duyn and Ratti (direct effect principles)

• Marshall (limits on horizontal direct effect)

• Von Colson and Marleasing (consistent interpretation)

• Francovich (State liability for non-implementation)

These cases collectively ensure that failure to implement does not neutralise the Directive’s legal relevance.

How direct effect actually works in PTD cases

Vertical direct effect

Some provisions of the Directive may be invoked directly against the State or public-sector employers if they are sufficiently clear, precise, and unconditional.

This includes bodies such as:

• government departments

• NHS-style public healthcare systems

• universities

• local authorities (councils)

• other state-controlled employers

In those contexts, the Directive can move from being aspirational to operational much earlier than private-sector implementation.

Horizontal direct effect

The position is more limited for private employers.

As a general rule, EU directives cannot be relied upon directly in claims between private parties.

That means employees usually cannot bring a standalone claim against a private employer solely on the basis of the unimplemented Pay Transparency Directive.

That principle remains firmly embedded in EU case law, particularly Marshall.

But this does not mean private employers are unaffected.

The mechanism that matters most in practice: consistent interpretation

Where direct enforcement is not available, EU law still requires national courts to interpret domestic legislation, so far as possible, in line with the Directive.

This principle, developed in Von Colson and Marleasing, is where most practical impact occurs.

In employment disputes, it means courts may:

• read existing equal pay provisions more expansively

• resolve ambiguity in favour of transparency

• align evidential expectations with EU objectives

• reshape interpretation of comparator and justification tests

This is often subtle rather than explicit. But it is real.

What actually changes in employment litigation after 7 June 2026?

The shift is not sudden. It accumulates through four channels.

1. Courts begin absorbing the Directive into interpretation

Judges do not ignore EU-level obligations simply because transposition is late.

Where domestic law is open-textured, courts increasingly lean toward interpretations consistent with pay transparency objectives.

Not because they are required to apply the Directive as statute law, but because EU interpretive obligations require alignment where possible.

2. Equal pay disputes begin to shift in framing

In labour courts and employment tribunals, arguments increasingly draw on:

• EU equal pay principles

• transparency expectations embedded in the Directive

• evidential imbalance arguments in pay disputes

Even where not decisive, these arguments influence how cases are structured and assessed.

3. Public sector employers face earlier exposure

The most immediate legal risk sits in the public sector.

Because vertical direct effect can apply, employers such as public universities or state healthcare systems may find that Directive-based arguments are raised directly in disputes sooner than in private-sector litigation.

4. Non-implementation becomes context, not irrelevance

Once the deadline passes, courts are aware that the State is in breach of EU obligations.

That background matters. It does not decide cases on its own, but it shapes judicial sensitivity to EU objectives.

How this plays out in real disputes

To make this concrete, consider a few typical scenarios.

In a Spanish or German labour court, an employee challenging unequal pay will likely see arguments framed not only through domestic equal pay law, but also through EU transparency principles. Courts will not treat those principles as irrelevant background noise.

In recruitment disputes, particularly where salary information was withheld, national courts may interpret existing anti-discrimination frameworks with greater emphasis on transparency obligations.

In public sector disputes, employees may go further and attempt to rely directly on Directive provisions, particularly where domestic law is silent or ambiguous.

In jurisdictions with strong works council systems such as Germany, France, or the Netherlands, employee representatives may use the Directive as leverage even before full implementation, particularly around information requests and pay structuring discussions.

Can employees sue for failure to implement the Directive?

Yes, but not against their employer. There is a separate route of potential liability against the State itself.

Under the Francovich doctrine, Member States may be liable in damages where:

• the Directive confers identifiable rights

• those rights are sufficiently clear

• failure to implement causes loss

This is not automatic, but it is well established in EU law and remains a live possibility in cases of significant non-implementation.

There is a decent chance the PTD meets those three Francovich requirements.

What employers should take from this

The most important practical point is that the Directive starts to operate as a legal reference point even before full transposition.

Not as binding statute everywhere, but as something that shapes:

• how claims are framed

• how courts interpret ambiguity

• how evidence is assessed in pay disputes

• how regulators and employee representatives approach transparency issues

The strategic risk is assuming that absence of national legislation equals absence of legal exposure.

That assumption becomes weaker the longer implementation is delayed.

Key legal concepts (quick reference)

EU Pay Transparency Directive (PTD)
EU legislation strengthening pay transparency, reporting obligations, and equal pay enforcement across Member States.

Vertical direct effect
Ability to rely on EU directive provisions against the State or public bodies.

Horizontal direct effect
Direct enforceability against private employers — generally not available for directives.

Consistent interpretation (indirect effect)
Obligation on national courts to interpret domestic law, so far as possible, in line with EU directives (Von Colson, Marleasing).

Francovich liability
State liability for failure to implement EU directives causing loss.

Frequently asked questions

Does the Pay Transparency Directive apply before implementation?
Not as fully transposed law. But it can still influence outcomes through interpretation and, in some cases, direct effect against public bodies.

Can private employers be sued under the Directive?
Not directly in most cases. However, domestic law may be interpreted in light of it, indirectly affecting private employers.

What happens if a country misses the deadline?
The State is in breach of EU law. Courts must still interpret domestic legislation consistently with the Directive, and some provisions may be enforceable against public authorities. The EU might bring action against States that haven’t implemented on time (they did this a few years ago).

What is vertical direct effect?
It allows individuals to rely on certain EU directive provisions against the State or public-sector employers.

Can courts use the Directive even if it is not implemented?
Yes. Courts are required to interpret national law, where possible, in line with EU directives.

What this means for private sector employers in practice

From June onwards, many employers are likely to receive employee requests for:

• pay information;

• comparator data;

• salary ranges; or

• explanations for pay differences,

This will happen even in countries where the Pay Transparency Directive has not yet been implemented.

Legally, employers in those jurisdictions could simply refuse such requests if no domestic disclosure obligation exists yet. In short “ask us again later”.

But the more difficult question is not always:

“Can we say no?”

It is:

“Should we?”

For multinational employers, this becomes particularly complicated.

Many organisations already have:

• PTD-compliant processes going live in some EU countries;

• centralised pay transparency procedures;

• HR guidance and manager training already prepared.

That creates obvious tension where one country has implemented the Directive and another has not.

Employees will increasingly assume the new rights already apply across the business. Inconsistent approaches between jurisdictions may quickly become visible internally.

Refusing requests too aggressively may create employee relations, litigation, or works council issues later. While there are “brownie points” for implementing transparency voluntarily and in advance of legislation, going live early may expose incomplete pay structures or inconsistent decision-making before remediation work is finished.

For private sector employers, this is an operational and strategic judgement call.

Final thought

The expiry of the implementation deadline does not create a moment of legal silence.

It creates a messy period of murky grey legal water, where EU law begins to influence outcomes through interpretation, litigation strategy, and institutional pressure rather than direct statutory application. The longer the PTD goes unimplemented, the murkier it gets.

The practical question is no longer whether the Directive is implemented. It is how quickly it begins to shape legal reasoning in practice. And in many jurisdictions, that process begins immediately after the deadline passes.

Most countries implementing the EU Pay Transparency Directive are, broadly speaking, asking the same question: how far do we actually have to go?

Greece seems to be asking something slightly different: what happens when transparency creates more problems than it solves?

That question sits underneath almost every interesting choice in Greece’s draft legislation.

And it leads, repeatedly, to the same institutional reflex.

If something becomes uncomfortable, or ambiguous, or potentially risky for either side, the answer is usually not to leave it there. It is to route it somewhere else.

For Greece, that somewhere is the Ombudsman.

Greece will miss the deadline

At the time of writing, the legislation is still a draft. The consultation came out yesterday (3 June) and runs for 2 weeks (until 17 June).

The EU Pay Transparency Directive had a transposition deadline of 7 June 2026.

Obviously, Greece is going to miss it.

Once enacted, the law will take effect on publication in the Official Gazette.

That matters, because there is already some confusion between that moment and the end of the first reporting period, which runs to 31 December 2026 for larger employers.

Those are not the same thing, even though they are sometimes treated as if they are.

Salary ranges: technically flexible, practically interesting

One of the most closely watched parts of any PTD implementation is simple.

Do salary ranges have to go in job adverts?

Greece says:

No. Applicants must receive pay information before interview (or before contract if there is no interview).

The right to information, and the moment things get awkward

The Directive gives workers the right to ask for average pay levels by gender for people doing the same or equivalent work.

Greece implements that right. Then immediately adds a pressure valve. Employers get a right too.

The right to refuse.

Employers can refuse requests that are “manifestly disproportionate or abusive”, especially if they are repetitive. That is not in the Directive.

If that happens, the worker does not just accept it. They escalate to the Ombudsman, who decides whether the request stands.

On paper, this is about preventing abuse of process.

In practice, it introduces a second layer of judgement into something the Directive tried to make fairly direct.

There is also a quieter implication. The right exists, but it is no longer purely bilateral. It now has a supervisor.

No other country has done something like this.

The information problem nobody quite solves

There is a gap in both the Directive and the Greek draft that becomes obvious once you try to operationalise it.

When someone requests average pay information, what time period are you supposed to use?

• Last year?

• Last twelve months?

• Current payroll snapshot?

Nobody says.

The Greek draft specifies the reference period for reporting (previous calendar year).

But it stays silent on individual information requests.

This is one of those details that looks minor until you try to build a system around it.

Then it becomes unavoidable.

At some point, if employers are to avoid guesswork, guidance will have to fill the gap.

Privacy, but with intermediaries

The most structurally interesting part of the Greek approach is how it deals with very small groups. There is no minimum group size threshold for rights to information.

If a category is so small that disclosure could identify an individual, the information is not simply withheld, but redirected.

Worker representatives, the Labour Inspectorate, or the Ombudsman receive it instead.

They then tell the worker whether there is a potential issue, without disclosing identifiable pay data.

The confidentiality tension

Then comes the counterweight.

Workers who receive pay information are placed under a statutory duty of confidentiality, and breaches can lead to damages claims by the employer.

This goes way beyond the Directive and is potentially seriously undermining the right to information, and the whole purpose of the PTD.

It creates an uncomfortable structure.

Because the system is designed to surface pay differences that may indicate discrimination. But Greece introduces legal risk around discussing those differences (while at the same time also banning pay secrecy clauses).

There is an obvious tension.

The pattern becomes clearer: someone else decides

The further you go into the draft, the more a pattern emerges.

When something is difficult, ambiguous, or potentially sensitive, the decision is rarely left entirely with one party.

It is pushed outward.

• Refused requests are reviewed by the Ombudsman.

• Small-group disclosures go through intermediaries.

• Enforcement sits with institutional bodies rather than direct party-to-party escalation.

• Even reporting sits centrally with the Ombudsman platform rather than being purely employer-driven.

It starts to feel less like a binary employer–worker framework and more like a triangular one. There’s always a third wheel

• Employer.

• Worker.

• And the institution in the middle.

Enforcement is built around that same logic

The Ombudsman becomes both equality body and monitoring authority.

There are additional specialist roles and technical infrastructure built around that function.

The Labour Inspectorate also gains a dedicated equal pay role.

It is a fairly substantial institutional build-out compared to some other Member States.

But it is consistent with the same underlying instinct.

If pay transparency creates disputes, route them somewhere structured.

Somewhere procedural.

Somewhere that can absorb conflict without immediately turning it into direct confrontation between employer and employee.

The practical questions employers are already asking

A few things are becoming operationally important quite quickly.

Do salary ranges have to appear in job adverts?

No.

They must be disclosed before interview or before contract, but the advert itself is only one possible channel.

How long do employers have to respond to information requests?

Two months.

That aligns with the Directive.

Can employers refuse requests?

Yes, in limited cases where they are manifestly abusive or repetitive.

But that refusal can be challenged via the Ombudsman.

What happens in very small categories?

There is no minimum group size.

But disclosure may be routed through worker representatives or public bodies if individuals would otherwise be identifiable.

What period is used for information requests?

Not specified.

Which is itself significant, because it will need to be operationalised somewhere.

The quieter policy split inside the draft

Greece is doing something slightly more complex than simply “transposing” the Directive.

On one hand, it expands structural requirements:

• pay structures become more formalised;

• enforcement capacity is strengthened;

• reporting infrastructure is centralised;

• remedies are detailed and robust.

On the other hand, it introduces controls that limit friction:

• refusal rights for abusive requests;

• confidentiality obligations around shared data;

• mediated disclosure in small groups.

So you get both expansion and containment at the same time.

Which is why it does not feel purely pro-employer or pro-worker.

It feels like a system designed around managed trust rather than assumed trust.

Tom’s final thought

The most interesting thing about the Greece Pay Transparency Directive draft is not any single provision.

It is the pattern they form when you step back.

Whenever transparency creates a potential problem, Greece does not usually resolve it by choosing one side over the other.

It inserts a process.

An intermediary.

A referee.

An institution that sits between the parties and absorbs the awkwardness.

That makes the system more structured.

Arguably more robust.

But also more mediated than in some other Member States.

And that, ultimately, is what the Greek draft reveals about pay transparency more broadly.

This was never just about publishing information.

It is about deciding who gets to sit in the middle when that information becomes uncomfortable.

The consultation is here and documents are below.

Greece Impact Assessment

747KB ∙ PDF file

Download

Download

Greece Draft Law

453KB ∙ PDF file

Download

Download

One of the more quietly loaded words in the Pay Transparency Directive is also one of the simplest.

Objective.

It is also where a lot of pay gap analysis quietly becomes fragile.

The PTD does not prohibit pay differences between men and women. It prohibits pay differences that cannot be explained by objective, gender-neutral factors.

On paper, that sounds straightforward.

In practice, it turns into a question of whether your explanation will survive scrutiny.

And that scrutiny does not usually come from your model.

It comes from employee representatives deciding whether they accept your assumptions in the first place.

Subscribe now

The real constraint: defensibility under challenge

In most organisations, pay differences are explained using familiar factors. For example:

• experience

• performance

• qualifications

These are not controversial in themselves. They are standard components of pay structures, and have been accepted by the ECJ.

But under the PTD, the issue is not whether these factors exist.

It is whether they can be defended as genuinely explanatory when the data is examined.

A typical challenge might be very direct:

“We don’t accept that these variables are actually objective explanations for the gap.”

And once that claim is made, the analysis shifts.

Because at that point, it is no longer enough to show a model that fits the data.

You have to justify why the inputs to the model should be trusted at all.

That is where a lot of analyses quietly fail. Not at the level of results, but at the level of assumptions.

Why classical tests are often not enough

The instinct at this stage is to reach for standard tools: t-tests, ANOVA, chi-squared tests.

These are all really great methods of comparing differences in means.

But they test whether group differences are statistically significant under certain assumptions.

In PTD contexts, those assumptions become the weak point:

• distributional shape (normality)

• equal variance

• independence

• sample size stability

In large, clean datasets, these assumptions are often reasonable approximations.

In workplace datasets, they are frequently difficult to defend in a meaningful way.

And under when scrutinised under the PTD magnifying glass, “difficult to defend” often becomes “not accepted”.

Why randomisation tests fit this problem better

Randomisation tests avoid that dependency on assumptions.

Instead of assuming a statistical model of the world, they build one from the data itself.

The logic is simple:

If gender has no relationship to a variable like experience, then reshuffling gender labels should not meaningfully change the observed differences.

So you:

• calculate the observed difference

• randomly shuffle gender labels

• recalculate the difference

• repeat many times

• compare the observed result to the resulting distribution

What you get is a direct, data-based benchmark of what “no relationship” looks like in that organisation.

No theoretical distribution.

No structural assumptions about normality.

Just the observed data points, after mixing them up loads of times.

Why this matters in PTD environments

There is a second reason this approach fits the PTD context.

It is not just statistical.

It is interpretive.

Because the real test is not whether a method is theoretically valid, but whether it survives discussion with people who did not build it.

Employee representatives do not engage with sampling theory or asymptotic assumptions.

They engage with a much simpler question:

“Why should we believe this result reflects reality rather than modelling choices?”

Randomisation tests are easier to defend in that environment because they are easier to explain without distortion.

You are not asking anyone to accept an abstract probability model.

You are showing them what happens when the data is repeatedly reshuffled.

That difference matters more than it first appears.

A more realistic example

Consider a small workforce.

Men (8 employees):
16, 8, 14, 9, 15, 10, 13, 7

Women (7 employees):
13, 9, 12, 8, 11, 10, 9

Average experience:

• Men: 11.5 years

• Women: 10.29 years

Difference: 1.21 years

There is a visible gap.

Not large, but not nothing either.

And that is often exactly the kind of pattern that creates disagreement in practice. An employee rep might jump straight to “this isn’t objective - it’s biased towards men”.

But the immediate question should not be whether a difference exists.

It is whether that difference is structurally meaningful or just the result of variation in a small, uneven dataset.

A permutation test shows that differences of this size are not unusual under random assignment.

In this case, p ≈ 0.25 (approx.)

So the more accurate interpretation is not that there is no difference.

It is that the difference is not distinguishable from noise in this dataset.

Or, to put it in terms that an employee rep might listen to:

“While there’s a difference of means, there’s no evidence of bias”.

What this does and does not do

Randomisation tests do not:

• determine objective justification

• replace regression analysis

• prove discrimination or its absence

• resolve legal disputes

They do something narrower.

They test whether an observed difference is meaningfully unusual, relative to what would be expected under random assignment, without relying on distributional assumptions that may be hard to defend in practice.

In PTD contexts, that narrowness is often the point.

Because the challenge is rarely about statistical complexity.

It is about whether the reasoning can survive scrutiny.

Key concepts

Randomisation test (permutation test)
A method that evaluates whether an observed difference is unusual by comparing it to outcomes generated by randomly reshuffling group labels within the observed dataset.

Objective justification (PTD context)
The requirement that pay differences must be explained by objective, gender-neutral factors.

Classical statistical tests
Methods such as t-tests or ANOVA that rely on assumptions about distributional form and variance structure.

Employee representatives
Stakeholders who get the power under the PTD to challenge whether analytical assumptions and explanatory variables are valid and acceptable as explanations of differences in gender pay gaps.

What employers should take from this

Most PTD analysis focuses on modelling outcomes.

But many of the real challenges arise earlier.

They arise when explanatory narratives are turned into assumptions that underpin the model itself.

Randomisation tests are useful because they operate at that preliminary stage. They can gie a real boost of assurance to the subsequent regression analysis.

Randomisation tests help ensure that what enters a model is at least defensible before it is relied upon.

When employee representatives look at pay gap results, one of the most common reactions is:

“This doesn’t feel right. It looks like a handful of very high salaries are affecting the gap.”

That instinct is usually pointing at something real.

Under the Pay Transparency Directive (PTD), this issue becomes more important, because analysis is not just about organisation-wide averages. It is about within-group pay differences, within job categories, within grades, and within roles that are supposed to be comparable.

And that is exactly where outliers matter most.

Not because they are automatically wrong.

But because they can materially change how a group-level pay difference is interpreted.

1. What an “outlier” actually means in pay gap analysis (in plain terms)

An outlier is simply:

a pay value that is unusually high or low compared to others in the same comparison group.

The key phrase is “same comparison group.”

Under PTD, that usually means:

• people doing broadly similar work

• grouped into job categories or equal value clusters

So a salary that looks extreme in the organisation overall might not be extreme at all within its own job category.

This distinction is where most misunderstandings start.

2. Why outliers matter more under PTD (the structural reason)

Under older pay gap approaches, you could often rely on organisation-wide averages.

PTD changes that.

It forces analysis into structured groupings:

• job categories

• equal value work groupings

• comparable roles within those categories

So the question becomes much more specific:

“Do a small number of individuals distort the pay picture within a defined job group?”

That is a very different question from “what is the average pay gap overall?”

And it is a more sensitive one.

Because within groups, sample sizes are often small — especially at senior levels.

3. Why senior pay is the main source of distortion

In most organisations, senior roles have three important features:

• there are very few of them

• they are paid significantly more than other roles

• they often include variable pay (bonuses, incentives, equity)

That combination creates a structural effect:

a small number of individuals can disproportionately influence the results for an entire job category.

In practice, I’ve seen cases where:

• a single executive or senior leader meaningfully shifts the estimated gap in their entire group

• removing just a handful of high earners changes the interpretation of a “gap” from moderate to small (or vice versa)

This is not an error in the data.

It is a feature of how pay is structured.

4. How outliers actually distort PTD-style results

You don’t need statistical terminology to understand the mechanism. Three simple effects explain most of what happens.

4.1 They pull the average within a job group

Most pay gap figures rely on averages.

So if a group looks like this:

• most employees earn £50k–£70k

• one or two people earn £200k+

Then the average no longer reflects a “typical” employee.

It reflects a blend of:

• typical pay

• and exceptional senior pay

So employee reps may reasonably ask:

“Does this average reflect what most people in this group actually experience?”

4.2 They can shift adjusted model results

When we adjust pay gaps for factors like:

• performance

• tenure/experience

• geography

we are effectively trying to isolate structured differences.

But if a small number of extreme values exist in a group, they can:

• pull the model’s estimate towards themselves

• exaggerate or dampen the estimated gender difference

• create instability in small job categories

In plain terms:

the model starts reflecting a few individuals rather than the group pattern.

4.3 They make results fragile in small groups

This is the PTD-critical point.

Many job categories are small, especially at senior levels.

So when:

• 10–20 people define a group

• and 1–2 of them are extreme earners

Then the result becomes sensitive.

Meaning:

small changes in inclusion or treatment of those individuals can change the reported gap.

That is not a statistical curiosity — it is a structural feature of PTD data.

5. The most important PTD principle: always analyse within groups first

A common mistake is to identify outliers across the whole organisation.

But under PTD, that is not the relevant comparison.

You should always ask:

“Is this pay level unusual compared to others doing similar work in the same group?”

Not:

“Is this unusual in the organisation overall?”

Because PTD is fundamentally about comparability within job categories, not global dispersion.

In practice, many apparent “outliers” disappear once you apply the correct grouping.

And equally, some hidden outliers only become visible once you do.

6. A simple PTD example (how interpretation changes)

Imagine a job category with 12 employees:

• 10 employees earn between €55k–€65k

• 2 senior employees earn €150k and €180k

Now consider the interpretation:

Without context

The average pay looks high, and the gap may appear wider than expected.

Within PTD grouping logic

You realise:

• most employees are tightly clustered

• the category contains a small senior subset with structurally different pay

So the key question becomes:

“Is this a single comparable group, or two different pay structures forced into one category?”

That is often the real issue behind outliers in PTD analysis.

7. How to think about outliers in practice (what I actually do)

In applied PTD analysis, I use a simple sequence.

Step 1: Check whether it is a data error

Ask:

• is this plausible?

• could it be duplication, mis-entry, or misclassification?

If yes, it should be corrected.

Step 2: Check whether it reflects a different pay structure

Ask:

• is this executive pay?

• is it bonus-heavy or equity-driven?

• does it follow a different compensation logic from the rest of the group?

If yes, it may not be directly comparable within the group.

Step 3: Check it within the job category (the PTD step)

Ask:

• is it still extreme compared to peers doing similar work?

This is the key interpretive test.

Step 4: Test whether it changes the conclusion

In practice, I always compare:

• results including all data

• results with extreme values capped (to reduce distortion)

• and sometimes results excluding the top extremes (as sensitivity testing only)

Then I ask a simple question:

does the conclusion change in a meaningful way?

If it doesn’t, the result is stable. If it does, the result needs careful explanation.

8. What to do about outliers (PTD decision logic)

There is no single correct answer. But there is a defensible logic.

Keep them when:

• they reflect real, structured pay (especially senior roles)

• they are part of normal compensation design

• results remain stable when included

Adjust them when:

• they dominate the interpretation of a job category

• they make within-group comparisons misleading

• they obscure the typical pay experience

Remove them only when:

• they are clearly non-comparable (e.g. one-off payments, misclassified data)

• or represent a different pay system entirely

• or you think it would be accepted by employee representatives (let’s not forget - they are your target audience for this).

In most PTD contexts, removal should be rare and explicitly justified.

9. The key insight employee representatives should take away

This is the central message:

Outliers are not just statistical noise. They are often a reflection of how pay is actually structured at the top of an organisation.

That means they can indicate:

• how senior roles are rewarded

• whether pay progression is compressed or stretched

• whether bonuses are concentrated in a small group

• how uneven pay structures are within job categories

So the real interpretive question is not:

“Should we remove these values?”

It is:

“What are these values telling us about how pay actually works within this group?”

10. What this means when reading PTD pay gap outputs

For employee representatives reviewing results, the key practical takeaways are:

• small groups are highly sensitive to a few high earners

• senior pay often drives most within-group variation

• averages can misrepresent typical experience in a group

• results should always be tested for stability

• interpretation matters as much as calculation

And critically:

A pay gap figure is not a fixed truth. It is a summary of a structure that can be shaped by a small number of individuals.

11. Practical checklist (what to look for in any PTD analysis)

When reviewing results, ask:

• Are extreme values concentrated in senior roles?

• Do they sit within or outside job categories?

• Do they materially change the gap when adjusted?

• Are they consistent with known pay structures (bonus, equity, commission)?

• Is the conclusion stable under reasonable alternative treatments?

Closing thought: outliers are a structural question, not just a statistical one

The biggest shift under PTD is this:

You are not trying to eliminate complexity from pay data.

You are trying to understand and explain to employee reps whether that complexity reflects:

• a data issue

• or a genuine feature of how pay is structured

Outliers sit exactly on that boundary.

Handled carefully, they improve understanding of how pay systems actually work. Handled poorly, they can distort how fairness is interpreted entirely.

The exceptions that prove the rule? Or the exceptions that undermine it?

That’s up to you.

The Netherlands has now published its draft legislation implementing the EU Pay Transparency Directive (“PTD”), together with an extensive explanatory memorandum and advisory materials setting out the government’s thinking, policy objectives, and implementation strategy.

And compared to several other recent PTD drafts across Europe, the Dutch approach feels strikingly restrained.

The government repeatedly emphasises a philosophy of “zuivere implementatie” — essentially “clean” or “pure” implementation. The idea is simple: transpose the Directive faithfully, avoid unnecessary gold plating, and do not create additional administrative burdens unless genuinely required.

For once, the legislative mechanics broadly match the marketing.

Because while many Member States claim “minimum implementation” before quietly adding extra deadlines, procedural layers, or expanded obligations, the Dutch draft stays relatively close to the PTD baseline in most major areas.

There are still some notable Dutch-specific choices. A few are quite important. But the overall tone is pragmatic, technocratic, and operationally focused rather than politically performative.

The most interesting part is probably this.

The Dutch government openly acknowledges that one major element of the implementation will be late.

Not “challenging”. Not “under review”. Not “subject to ongoing consultation”.

Late.

By legislative standards, that level of candour is refreshing.

The Netherlands is pursuing actual minimum implementation

A recurring theme throughout the draft bill and explanatory memorandum is minimising unnecessary burden on employers.

That philosophy appears repeatedly in the choices the government has made:

• reporting is not extended below 100 employees;

• employers under 50 are exempted from pay progression transparency obligations;

• flexible “reasonable period” wording is retained rather than replacing it with rigid deadlines;

• optional restrictions on information requests are largely avoided;

• reporting is linked to existing payroll infrastructure where possible.

In other words, the Dutch government appears to have looked at the PTD and asked:

“What is the minimum workable system required to comply?”

Rather than:

“How many additional processes can we fit into this thing before employers notice?”

That immediately distinguishes the Dutch draft from several other implementations currently emerging across Europe.

The Netherlands is openly delaying part of implementation

This is the most unusual aspect of the Dutch draft.

Under the PTD, employers with 150+ employees are supposed to begin reporting by 7 June 2027.

The Netherlands says that will not happen.

Instead, the first reporting deadline for employers with 150 or more workers will be 7 June 2028, covering the 2027 calendar year.

And the government is extremely candid about why.

Its position is essentially:

• the reporting methodology still needs to be finalised;

• software systems need to be developed;

• the monitoring infrastructure must be operational first;

• employers cannot realistically comply before those systems exist.

Legally, this is difficult to dress up as anything other than non-compliance with the Directive timetable. The explanatory materials more or less concede that point.

Operationally, though, the Dutch position is fairly understandable.

A lot of PTD implementation debates across Europe currently involve a strange collective pretence that large-scale reporting systems can simply materialise into existence because a Directive says so.

The Dutch government seems far more focused on whether the system can actually function in practice.

Smaller employers get a meaningful exemption

One of the clearest examples of the Netherlands sticking closely to the PTD minimum concerns pay progression transparency.

Article 6(2) PTD allows Member States to exempt employers with fewer than 50 workers from obligations relating to pay progression criteria.

The Netherlands exercises that option.

So employers below 50 employees will not need to provide workers with easy access to progression criteria and pay-setting structures.

That contrasts with countries like Bulgaria, which chose not to use the exemption despite also claiming “minimum” transposition.

For smaller employers, this matters more than it might initially sound.

Many smaller organisations simply do not currently operate highly formalised progression systems capable of neat disclosure. Requiring them to construct one purely for compliance purposes could have created disproportionate burden very quickly.

The Dutch government explicitly frames the exemption as an administrative burden reduction measure.

Dutch employers will not face rigid procedural deadlines everywhere

Another major contrast with some other PTD drafts is the Dutch preference for flexibility over fixed procedural timetables.

Throughout the draft legislation, the Netherlands generally preserves the PTD’s softer wording around “reasonable periods” rather than replacing it with hard deadlines.

For example:

• workers must receive pay information within a reasonable period, and no later than two months;

• corrective action following unjustified pay differences must occur within a “reasonable period”;

• clarification requests relating to reporting data also follow the same two-month framework.

There is no Dutch equivalent of Bulgaria’s increasingly crowded compliance calendar involving 31 January notices, 14-day response periods, and fixed remediation deadlines.

The overall Dutch approach feels much more principles-based than deadline-driven.

That may create slightly less procedural certainty in some areas, but it also gives employers more operational flexibility.

The Dutch government has made a very deliberate GDPR trade-off

This is probably the single most interesting policy choice in the entire draft.

Under Article 12(3) PTD, Member States can introduce safeguards restricting access to certain pay information where disclosure could reveal an individual colleague’s pay.

Some countries are using intermediary mechanisms here. For example, information may be routed through worker representatives or equality bodies instead of directly to employees.

The Netherlands is not doing that.

Instead, workers receive the information directly, even in situations where category sizes are so small that the requesting worker can effectively work out another individual’s pay.

And the explanatory memorandum openly acknowledges this consequence.

The Dutch government’s position is essentially:

• yes, this creates privacy implications;

• yes, re-identification may occur;

• but equal pay transparency outweighs that concern.

The safeguard is therefore not an access restriction but a use restriction.

In other words, workers may only use the information for enforcing equal pay rights.

That is a very conscious policy decision.

It also creates a genuinely interesting tension between GDPR-style data minimisation principles and the PTD’s underlying transparency objectives. Other Member States may ultimately land in very different places on that balance.

No limit on information requests

The PTD itself does not impose any cap on how often workers may request pay information.

Neither does the Dutch draft.

In fact, the explanatory memorandum explicitly states that the Directive provides no basis for restricting the number of requests workers may make.

That matters because some Member States are already experimenting with limitations or procedural gating mechanisms (looking at you Italy).

The Dutch approach is comparatively straightforward:

if the Directive does not permit a restriction, the Netherlands generally does not add one.

The Netherlands quietly gold plates a few things anyway

Despite the overall “minimalist” philosophy, the draft still contains several genuinely stronger provisions.

Works councils get real influence

Under the PTD, joint pay assessments must occur “in cooperation with” worker representatives.

The Netherlands goes a little further.

Dutch works councils receive an instemmingsrecht — effectively a consent right — over both:

• the pay evaluation process;

• the method used to remedy unjustified differences.

That is materially stronger than mere consultation.

But it also fits neatly within the Netherlands’ existing works council culture and governance framework.

For multinational employers unfamiliar with Dutch employee participation structures, this could become operationally significant. A pay remediation exercise that might sit largely within HR and legal functions elsewhere could require much more formal employee representative engagement in the Netherlands.

Board-level sign-off on reporting accuracy

The Dutch draft also requires the board of directors to confirm the “faithfulness” (getrouwheid) of reporting information after consulting the works council.

That language is very deliberate.

The terminology is borrowed from Dutch audit and accounting concepts and gives the reporting process a distinctly governance-oriented flavour.

This is not just HR compliance. It starts pushing pay reporting closer to formal corporate governance architecture.

The Netherlands is trying to make PTD reporting operationally workable

One of the most pragmatic features of the draft is its attempt to build reporting around existing payroll infrastructure.

The explanatory materials repeatedly stress the importance of avoiding duplicate systems and minimising administrative overhead.

Key choices include:

• leveraging existing payroll tax reporting chains (loonaangifteketen);

• using FTE-based calculations for workforce size;

• deferring detailed technical specifications to secondary legislation;

• phasing implementation around actual system readiness rather than theoretical deadlines.

This may ultimately make the Dutch system more adaptable than some more rigid implementations emerging elsewhere.

The trade-off is that several important technical details remain unsettled for now.

The Dutch government thinks most employers are not starting from zero

Unlike some governments framing the PTD as a radical transformation exercise, the Dutch materials repeatedly suggest that much of the underlying infrastructure already exists.

The Netherlands already has:

• an established equal treatment framework;

• a relatively mature works council system;

• widespread use of job evaluation methodologies;

• existing anti-discrimination enforcement architecture through the Netherlands Institute for Human Rights;

• relatively developed payroll and labour reporting systems.

The PTD is therefore presented less as a revolutionary reform and more as an extension of systems already operating in practice.

That probably explains some of the government’s confidence in taking a more minimalist legislative approach.

The practical questions Dutch employers are likely asking now

Will salary ranges need to appear in job adverts?

No.

Like many other PTD drafts, the Dutch legislation appears to require candidates to receive pay information before interview and before salary discussions take place.

But the draft does not appear to expressly require salary ranges to physically appear inside job advertisements themselves.

That distinction matters enormously operationally.

A requirement to disclose pay before interview is not necessarily the same thing as mandatory salary advertising.

Across Europe, this is becoming one of the major practical PTD battlegrounds.

What happens where reporting reveals a 5% gap?

The Dutch approach here tracks the Directive fairly closely.

A joint pay evaluation is only triggered where:

• a 5% or greater gap exists;

• the employer cannot objectively justify it;

• the issue has not been remedied within six months.

Importantly, there is no automatic front-loaded obligation to submit formal justifications immediately upon identifying a 5% gap.

That is a materially lighter-touch approach than some other drafts currently emerging across Europe.

How aggressive will enforcement actually be?

The Dutch Labour Inspectorate (Arbeidsinspectie) will enforce the regime through:

• compliance orders;

• administrative fines;

• penalty payments;

• publication of enforcement decisions.

But the government repeatedly stresses that early enforcement will focus heavily on guidance, awareness, and helping employers understand their obligations.

The tone is cooperative rather than punitive.

The government is saying “just do your best - we know this is going to be hard, but we’re not going to come charging in heavy handed if you get things wrong”.

The Dutch implementation feels unusually coherent

A lot of PTD implementation drafts across Europe currently feel internally conflicted.

Governments say they want minimum implementation while simultaneously adding layers of procedural rigidity, fixed deadlines, expanded obligations, and domestic policy objectives.

The Dutch draft feels more intellectually consistent.

The government says it wants “zuivere implementatie”.

And, for the most part, that is what the legislation delivers.

There are still some Dutch-specific additions:

• four-year data retention;

• stronger works council involvement;

• board-level reporting sign-off;

• highly transparent worker information rights.

But overall, the Netherlands stays relatively close to the Directive’s architecture and policy logic.

Ironically, the most significant deviation is also the one the government is most transparent about:

the implementation timetable itself.

Final thought

The Dutch draft is probably the clearest example so far of a Member State treating PTD implementation primarily as an operational governance exercise rather than a political branding exercise.

The overall tone is calm, technical, and unusually candid (or rather, typically Dutch). There’s no beating around the bush.

The government appears far more concerned with whether the system can function in practice than with theatrically declaring itself “fully compliant” on paper.

That may ultimately prove quite sensible.

Because across Europe, the hardest part of the PTD is increasingly looking less like legal transposition and more like institutional execution:

• reporting infrastructure;

• data quality;

• governance processes;

• remediation frameworks;

• employee relations;

• defensible pay architecture.

The Dutch government seems to understand that already.

Even if it also means admitting, rather bluntly, that part of the timetable is slipping by a year.

Relevant documents

Advies Afdeling Advisering Raad Van State En Nader Rapport

112KB ∙ PDF file

Download

Download

Voorstel Van Wet

77KB ∙ PDF file

Download

Download

Memorie Van Toelichting

913KB ∙ PDF file

Download

Download

Bulgaria has now published its draft legislation implementing the EU Pay Transparency Directive (“PTD”), alongside a substantial package of supporting documents explaining the government’s rationale, objectives, and impact assessment.

For anyone tracking the broader EU Pay Transparency Directive Bulgaria debate, the government’s message is remarkably consistent:

this is intended to be a minimum implementation exercise.

The accompanying ministerial report, motives document, and impact assessment repeatedly emphasise that Bulgaria does not intend to go beyond the Directive’s requirements or create unnecessary burdens for employers.

And, to be fair, in several major areas Bulgaria does follow the PTD framework quite closely. The reporting thresholds mirror the Directive almost exactly. Employers with fewer than 100 employees remain outside the formal reporting regime. The phased introduction broadly follows the EU timetable.

But once you move beyond the headline positioning, the picture becomes more interesting.

Because while Bulgaria describes its approach as minimalist, the draft legislation contains a number of provisions that, in practice, go further than the PTD strictly requires. Some are relatively technical. Others could materially affect how employers manage compliance, pay governance, and remediation exercises.

The result feels less like “pure minimum transposition” and more like a fairly rigid procedural system built around the PTD.

And the kicker is this: they’re planning to implement on time for 7 June.

Bulgaria’s overall approach to the PTD

The government has chosen to implement most of the Directive through amendments to Bulgaria’s anti-discrimination framework rather than embedding everything directly into the Labour Code.

That is not accidental.

The explanatory documents repeatedly stress that the PTD is, fundamentally, an anti-discrimination instrument rather than simply an employment transparency law. Pay transparency is framed as the mechanism, while gender equality enforcement is framed as the objective.

That legislative philosophy matters because it shapes the overall tone of the draft.

Rather than positioning the reforms primarily as an HR reporting exercise, the Bulgarian approach places the Commission for Protection against Discrimination (“CPD”) at the centre of enforcement and monitoring. The draft significantly expands the CPD’s role beyond reactive complaint handling into proactive oversight, methodology development, technical guidance, monitoring, and data aggregation.

The CPD is clearly being prepared for a much bigger role than it historically had.

The government thinks Bulgaria has a structural transparency problem

The accompanying documents are unusually candid about the government’s view of the current legal framework.

Existing equal pay protections are described as largely “declarative” in nature. The government’s position is essentially that Bulgaria already prohibits discrimination formally, but lacks the mechanisms necessary to detect or challenge it effectively in practice.

Several themes appear repeatedly throughout the impact assessment:

• lack of proactive pay transparency;

• lack of comparative pay information rights;

• weak institutional monitoring;

• widespread pay confidentiality practices;

• low employer readiness for PTD compliance.

One particularly striking point is the government’s reliance on a consultancy assessment suggesting Bulgarian companies currently score only 1.06 out of 3 for readiness to comply with the Directive’s requirements.

That number quietly explains quite a lot about the draft’s structure.

The legislation repeatedly tries to reduce ambiguity by introducing concrete deadlines, formal procedural obligations, and specified response periods. Bulgaria appears less interested in flexibility than in making sure the machinery actually works.

The practical PTD questions Bulgarian employers are likely to focus on first

For many employers, the most important issues are not actually the reporting thresholds or the drafting philosophy.

They are the operational questions sitting underneath the Directive itself.

And several of those questions remain only partially answered by the Bulgarian draft.

Will salary ranges need to appear in job adverts?

The Bulgarian draft follows the PTD requirement that candidates must receive information about the initial pay level or pay range before interview and before pay discussions take place.

But, like the Directive itself, the draft does not appear to expressly require salary ranges to be physically included inside the job advertisement itself.

That distinction matters.

In practice, there is a major operational difference between:

• “the salary must appear in the advert”, and

• “the salary must be provided before interview”.

Those are not the same compliance model.

The Bulgarian wording currently appears closer to the second approach.

What counts as “pay” and “pay level”?

One of the most important PTD issues across Europe is that the Directive uses concepts like:

• “pay”

• “pay level”

• “categories of workers”

• “average pay levels”

That matters because the meaning can shift depending on the obligation involved.

For example:

• the “pay range” disclosed pre-interview may not necessarily be constructed identically to reportable remuneration calculations;

• equal pay comparisons may involve broader remuneration analysis than recruitment disclosures;

• reporting metrics may require inclusion of additional remuneration elements that are not operationally relevant in hiring discussions.

The Bulgarian draft largely mirrors the PTD architecture here rather than materially redefining these concepts domestically.

So a lot of these arguments will probably end up being fought through guidance, litigation, and eventually CJEU interpretation.

Will the ban on pay secrecy clauses actually change much?

Potentially yes, particularly because the Bulgarian government explicitly identifies pay confidentiality clauses as widespread practice.

The draft prohibits contractual restrictions preventing employees from discussing pay and reinforces this prohibition in both the anti-discrimination legislation and the Labour Code.

Importantly, the Labour Code wording goes further than simply banning clauses in employment contracts. It extends the prohibition across:

• collective agreements;

• internal remuneration rules;

• employer acts more broadly.

In theory, many employers could deal with this relatively quickly by issuing communications confirming that existing secrecy provisions will no longer be enforced.

But the harder issue is usually cultural rather than contractual. Just because people can talk about pay, it doesn’t mean that they will.

How will “equal value” work in practice?

The PTD requires employers to assess work based on objective criteria such as:

• skills;

• effort;

• responsibility;

• working conditions.

The Bulgarian draft broadly follows the Directive’s structure here rather than introducing a new approach. No surprises here.

How will employee representatives work under the Bulgarian system?

The Bulgarian draft appears designed to integrate the PTD into existing worker representation structures rather than creating an entirely new representative regime specifically for pay transparency.

In practice, trade unions and existing employee representation mechanisms will probably absorb much of the representative function.

That matters because the effectiveness of PTD processes often depends heavily on how developed employee representation structures already are. In countries with strong works council traditions, the PTD may slot into existing governance relatively smoothly. Elsewhere, it may be much messier.

A lot of employers across Europe still do not actually know who their eventual PTD representatives will be in practice.

Will the pay history ban matter in Bulgaria?

Potentially quite a lot.

The PTD prohibits employers from asking candidates about prior pay history. The practical effect is often larger than it first sounds because salary history has historically functioned as an anchoring mechanism in negotiations.

For many employers (not just in Bulgaria), the real change is behavioural:

• recruiter habits;

• hiring manager conversations;

• negotiation culture;

• application form design.

In some businesses, those practices are deeply embedded.

How will Bulgarian pay gap reporting actually work?

One important practical question is whether Bulgaria intends to centralise reporting infrastructure (like in Lithuania) or leave calculation and publication obligations primarily with employers themselves.

The Bulgarian draft currently appears closer to a more employer-driven model.

Employers themselves appear responsible for:

• calculating the required indicators;

• submitting reporting data;

• retaining supporting information;

• preparing mandatory justifications where gaps exceed 5%.

Reports must be submitted by 7 June of the relevant reporting year.

The reports are submitted to the CPD rather than simply published informally on a company website.

Which enforcement bodies will actually enforce the PTD in Bulgaria?

The key enforcement actor appears to be the Commission for Protection against Discrimination (CPD). The Bulgarian draft significantly expands its role.

The CPD is not positioned merely as a complaints body. The legislation gives it a nice set of new teeth: broader monitoring, methodological, assistance, and oversight functions.

That includes:

• monitoring compliance;

• collecting and publishing information;

• developing methodological tools;

• supporting employers;

• aggregating complaint data;

• overseeing aspects of reporting and justification processes.

The draft also provides for financial sanctions, including increased penalties for repeat or continuing violations.

Whether the CPD ultimately has sufficiently strong teeth will probably depend less on the legislation itself and more on:

• resourcing;

• staffing;

• technical expertise;

• enforcement appetite;

• political backing.

The most important areas where Bulgaria goes beyond the PTD minimum

The Bulgarian government insists the draft does not go beyond the Directive unnecessarily.

But here’s where the advertised product doesn’t quite match up with what’s in the box.

Several provisions almost certainly do.

No exemption for employers with fewer than 50 employees

Under Article 6(2) PTD, Member States are allowed to exempt employers with fewer than 50 employees from obligations relating to pay progression criteria.

Bulgaria has chosen not to use that exemption.

This is probably the clearest substantive example of Bulgaria going beyond the minimum PTD baseline.

For smaller employers, this could become more operationally significant than the government’s impact assessment suggests. Many smaller businesses simply do not currently have formalised progression structures capable of easy disclosure.

Mandatory justification for any 5% pay gap

This may end up mattering more in practice than the reporting thresholds themselves.

Under the PTD itself, employers may attach explanations to reported gender pay gaps. The Directive does not automatically require formal justification every time a threshold is exceeded.

The Bulgarian draft takes a much harder line.

Where reporting identifies a pay gap of 5% or more, employers must prepare and submit a justification to the CPD.

Automatically.

Even where the gap may ultimately prove objectively justified.

That shifts the compliance burden forward considerably. Instead of waiting for escalation into a joint assessment process, employers may effectively need defensible explanations prepared from the moment reporting is produced.

Many employers will discover the difficult part is not reporting. It is discovering whether their pay systems can survive scrutiny once explanations become mandatory.

Bulgaria likes hard deadlines

One recurring feature of the draft is its preference for fixed dates and defined response periods where the PTD itself is more flexible.

Examples include:

• annual employee notification of pay information rights by 31 January each year (get this on your post Christmas “to do” list now);

• pay reporting submissions by 7 June (the PTD’s birthday);

• 14-day deadlines for clarification responses (VERY short);

• mandatory correction of unjustified pay differences within one year following joint assessment.

The PTD itself often uses softer language such as “reasonable period”.

Bulgaria generally does not. It likes things nice and clear.

From a legal certainty perspective, employers may actually welcome this. Ambiguity creates litigation risk. Clear deadlines at least create operational clarity.

But the trade-off is reduced flexibility.

The four-year data retention requirement

The draft introduces a minimum four-year retention obligation for category-level pay data.

That sounds technical. It probably will not stay technical for long.

Once employers begin retaining detailed remuneration data for extended periods, questions around GDPR governance, audit trails, data integrity, and evidential consistency become much more important.

Additional collective bargaining disclosure obligations

The draft also inserts additional transparency obligations into Bulgaria’s collective bargaining framework.

Employers engaged in collective bargaining will need to provide unions with information about:

• average pay levels by job level;

• rules governing basic remuneration;

• additional remuneration structures.

That goes beyond the PTD’s minimum transparency framework and reinforces the broader Bulgarian approach of embedding pay transparency into existing labour relations structures rather than treating it as a standalone exercise.

One area where Bulgaria may actually be more restrictive than the PTD

Interestingly, the Bulgarian draft is not uniformly stricter.

Under one Labour Code provision, additional salary information requests appear limited to once per year (like in Italy).

The PTD itself does not expressly impose such a cap.

That said, the practical significance may ultimately be limited because the broader right under the anti-discrimination legislation appears not to contain the same restriction.

Still, the coexistence of two partially overlapping information-right regimes could create confusion in practice, particularly in the early stages of implementation.

The government is trying very hard to reassure employers

A recurring theme across all the supporting documents is reassurance.

The government repeatedly emphasises that:

• obligations are proportionate;

• implementation is phased;

• burdens on SMEs are limited;

• technical assistance will be available;

• costs are manageable.

The impact assessment estimates compliance costs for smaller employers at below approximately €132 annually per enterprise, drawing on Eurofound comparator data.

Which, frankly, is ludicrous. The burden is much heavier than that.

Whether employers agree with those assumptions once implementation begins is another question entirely.

Particularly because some of the hardest PTD work is not reporting itself. It is the upstream governance exercise employers often discover they need first:

• job architecture;

• grading consistency;

• documentation quality;

• pay rationale defensibility;

• historical anomaly remediation.

Those exercises are rarely quick and cheap, even where reporting obligations themselves appear administratively modest.

What Bulgarian employers should probably focus on now

At this stage, the Bulgarian draft still remains draft legislation. The text may evolve during the legislative process.

But several themes already look relatively clear.

1. Bulgaria is building quite a rigid compliance system

Even where Bulgaria stays broadly within PTD boundaries, the draft consistently prefers concrete obligations, procedural specificity, and institutional oversight.

2. Smaller employers are not entirely escaping transparency obligations

Even employers outside formal reporting thresholds may still face significant obligations around pay criteria transparency, recruitment transparency, and information rights.

3. Documentation quality is going to matter

The mandatory justification requirement for 5% pay gaps significantly increases the importance of defensible pay rationale documentation.

For many organisations, the difficult question will not be:

“Do we have a gap?”

It will be:

“Can we explain it coherently under scrutiny?”

4. The CPD is becoming a much more important actor

The draft transforms the CPD from a largely reactive discrimination body into a proactive monitoring and guidance institution.

That institutional shift may end up being one of the most important long-term consequences of the Bulgarian implementation model.

Final thought

Bulgaria’s draft PTD legislation is interesting precisely because of the contrast between its messaging and its mechanics.

Politically and rhetorically, the government presents itself as pursuing restrained, minimum-compliance transposition.

Operationally, though, the draft often chooses precision over flexibility, procedural structure over discretion, and institutional oversight over employer autonomy.

Some of that is probably sensible.

The government clearly believes the existing Bulgarian framework lacks the mechanisms necessary to detect and challenge unequal pay effectively. It’s being unusually clear about its equal pay failings.

From that perspective, tightly specified procedural obligations are not accidental overreach. They are the implementation strategy.

The more important question now is probably not whether Bulgaria has technically gold plated certain provisions.

It is whether employers, regulators, courts, and the CPD can make this framework work in practice, particularly in a market the government itself believes is only marginally prepared for the Directive’s arrival.

And that may ultimately matter far more than whether the legislation is labelled “minimum transposition” or not.

All documents relating to the draft can be found here.

Share