Screens don’t affect comprehension; scrolling does.

When it comes to writing lecture notes, studies show that writing on paper is more beneficial for learning outcomes than writing on a tablet. But is the same true for reading lecture notes?

I’m not talking about re-reading your own hand-written notes. I’m talking about reading those copious prose notes that your lovely lecturers have spent many painstaking hours constructing and typing up to accompany their teaching.

Or indeed a textbook, I suppose. Because let’s face it, the textbook that you want is almost universally the same as the one that everybody else wants, and so - with the exception of one lucky person (who probably graduated a decade ago, but never returned the book) - you have to make do with the e-book version from the university library. I don’t know why I’m making this sound like a bad thing. Those e-books are basically magic. Being able to access any textbook from the university library at the click of a button, from anywhere, at any time, is frankly genius. But will it support your learning objectives as well as the real thing?

It seems that might depend on which digital device you use to read it.

A recent review* of the literature examined the relative effect on reading comprehension from using different digital devices compared to plain old paper. The devices that were considered were tablets, computers (laptops and desktops), smartphones and e-readers. The review considered results from 56 different studies, containing 79 different comparisons. And on the face of it, their results suggest that paper is probably your best bet. I’ve summarised these results in the figure below.

The yellow points indicate the mean (average) of the results for each type of digital device relative to reading from paper. The error bars indicate a 95% confidence interval, i.e. 95% of those results for each device would fall in that range. These values are something called a Hedge’s-g score, which is basically a measure of the difference between two results, in units of their combined standard deviation. For anyone who is familiar with a Cohen’s-d score, this is just a sample-size corrected version.

Anything below the zero line is worse for reading comprehension than a paper copy. So in principle these results suggest that using any digital device will have a detrimental effect on your reading comprehension compared to reading the same material from a paper copy. The most significant effects are for computers and smartphones, which both come off rather badly. Tablets and e-readers are not so bad; they have confidence intervals that overlap the “no difference from paper” line, which means that in some circumstances you wouldn’t see a difference in comprehension at all. Overall though, you can see that on average all four digital options sit below that line.

The authors also calculate a second measure for each device, which they call a netrank P-score. It’s a frequentist version of the Surface Under the Cumulative RAnking curve (SUCRA) and quite a nice statistic. Formally it's the mean of the one-sided p-values testing whether an option is superior to each of its competitors, i.e. the probability that one option is better than each of the other options, computed pairwise and accounting for the uncertainty in each, and then averaged across them all. The upshot of using it in this case is that you get a value that can be used for ranking the different digital devices and the paper option in terms of their helpfulness for comprehension. That’s a bit different to simply looking at their Hedge’s-g scores, because it takes into account the spread in those scores as well as the mean. Those results are shown in the figure below.

In this case that means that the computer and e-reader end up being ranked equivalently. The e-reader has a slightly worse average Hedges’s-g score than the computer, but because its confidence interval overlaps the “no difference from paper” line that nudges it up a bit in the ranking. The smartphone comes out at the bottom of the ranking.

However… if you dive into these results a little deeper things start to look a bit different.

Scroll down

The authors of the review went further and separated the studies into those that required the participants to scroll through the reading material, and those that didn’t. The “no scroll” option could mean that either the text was short enough to fit on a single screen, or that they used pagination. You know, like when you tap on the side of the e-reader to advance a page.

If you split the results in the literature by scroll vs. no scroll then a clear pattern emerges. For all of the digital devices where scrolling is an option, the negative effect on comprehension relative to paper is dominated by the scrolling component of the results. You can see this illustrated in the figure below. There’s no scrolling option for the e-reader, that’s why it only has one data point.

If you take away the comparisons that require scrolling then none of the digital devices are significantly worse for reading comprehension than paper, i.e. it’s not the digital devices that are bad for reading comprehension, it’s the scrolling.

In terms of their ranking this changes things a bit: see the figure below. The e-reader and smartphone are now ranked last. However, given the magnitude of the Hedge’s-g score in each case, none of them are really a bad option if you’re not scrolling. So the ranking itself is splitting hairs a bit.

In reality of course, for reading either lecture notes or an e-book, you’re going to need to scroll on a smartphone. Equally if your notes or textbook are likely to contain figures or diagrams then an e-reader is also not an option. So when it comes to reading online lecture notes or e-books, your best options are either a tablet or computer - just remember to have the pagination setting enabled so that you don’t have to scroll.

There’s also a take home here for the lecturers: it’s probably a good idea to provide your notes in a form that supports pagination. That’s no problem for people who create pdfs or similar, but if you’re using markdown or html to render online notes it might be worth converting to another format.

[*] Virginia Clinton-Lisell and Christine Litzinger, Decoding digital reading: a network meta-analysis of comprehension across devices, Education and Information Technologies, 2026, 31, 1611–1643, https://doi.org/10.1007/s10639-025-13843-8 [online article]

But the modern breakfast cereal, with its flashy advertising, irritating jingles and high sugar content, would be anathema to John Harvey Kellogg, the man who originally presented bland cereal-based breakfasts as religious morality in a bowl.

The idea that breakfast plays a significant role in childhood education has had such a huge impact that it led to multiple public health interventions calling for increased breakfast consumption around the world. However, there is now a growing body of literature pointing out that these early influential studies ignored other socio-economic factors. It turns out that socio-economic status is not only a big predictor of academic performance, it’s also a predictor of breakfast consumption. So the early studies effectively fell into the trap, either unwittingly or willingly, of conflating correlation with causation. Which is probably why several later studies that examined the effect of school breakfast programs, where all children were provided with breakfast, found that these seemed to have no effect.

In a large Australian study¹ following approximately 2000 school children over several years, the effect of skipping or not skipping breakfast on longer term academic performance was found to be very small - and even smaller when socio-economic factors were taken into account. The largest effect seen in the data was around numeracy, where breakfast skippers were found to perform slightly worse on average than breakfast eaters in national assessments at age 10-11. However, the size of the breakfast-based effect was more than 3 times smaller than the average difference between the grades of children who had parents with a university degree and those with parents who had not continued into further education after school.

The same thing has been seen in studies of adult cohorts, interest in which has increased significantly since the popularity of intermittent fasting has grown in recent years. In a 2025 systematic review² of the difference in cognition amongst more than 3000 fasted and non-fasted adults, with a median fast duration of 12 hours (basically a bit more than overnight), no significant difference in cognitive performance was found between participants who had fasted and those who had not.

The review also broke this result down into the effect on different types of cognitive process: attention, working memory etc. These results are summarised in the figure below, where a shift to the right indicates that fasting had a negative effect and a shift to the left indicates that fasting had a positive effect. None of the shifts are very significant, statistically speaking. The largest effects they found in either direction were for cognitive flexibility (-0.23) and inhibitory control (+0.16), but even those are not considered very strong.

The authors of the review concluded that short-term fasting (up to ~24 hours) had very little effect on cognitive performance. Moreover, if the fast was predominantly overnight then the effect was minimised even further.

Which means… breakfast is the least important meal of the day?

Feeling Hungry

It’s probably unsurprising then that while undergraduate students who skip breakfast consistently self-report that their day-to-day academic performance is poorer than those who don’t skip breakfast, studies³ show that there is no equivalent difference seen in their formal grades.

There’s an interesting parallel to be drawn here. We might assume that students who skip breakfast find it more difficult to focus and pay attention in their classes (this would not be supported by the short-term fasting analysis, but let’s just run with it for a moment), which is why they self-report poorer performance. However, from equivalent sleep studies, we can see that such consistent impacts on attention and information processing should also result in poorer long-term performance, and furthermore those effects would also be expected to accumulate over the course of an academic year. That is what has been observed in studies looking at sleep deficits. But the same does not seem to be true for breakfast skipping. So perhaps something else is going on here?

Somewhat unsurprisingly, recent research⁴ looking at the connections between blood glucose, hunger and mood found that both blood glucose levels and feeling hungry are linked to lower (sadder) moods. Could it be that the mood-related impact of skipping breakfast has an effect on the self-reporting of day-to-day academic performance? i.e. students who skipped breakfast perceive themselves as having poorer academic performance because their mood is lower?

This is very possible. The connection between someone’s mood and their personal state, known as their self-concept, is well-studied. In one study⁵ from Argentina, researchers used pieces of classical music that were considered happy or sad to temporarily induce the equivalent mood in their participants. They then asked the participants to describe themself in words.

Happy music: Four seasons: Spring I Allegro and Spring III Alegro, by Vivaldi; The Nutcracker: Trepak and Waltz of the Flowers, by Tchaikovsky; and Serenade #9: Finale, by Mozart

Sad music: Four seasons: Autumn, Vivaldi’s adagio; Adagio in G Minor, by Albinoni; and Violin Concerto: Adagio di Molto, by Sibelius.

The study analysed the words the participants used, not only individually, but also by looking at the relationships between the words. Their results showed that mood affects in-the-moment judgments of self-concept, i.e. lower (sadder) moods resulted in more negative judgements of a participant’s self, and happier moods resulted in more positive judgements.

This was not simply a result of more positive words being used, but also fewer negative words. Their results are summarised in the figure below.

Interestingly, one of the key (central) words that came up in the positive mood group during this experiment was “Enthusiastic”. The same word didn’t even place in the top 15 for the negative mood group. I like this. It implies that mood, and - by extension in this context - breakfast, may also contribute to improving motivation.

So, in conclusion, while skipping breakfast is unlikely to have any direct impact on your academic performance, it is likely to make you feel less good about yourself (temporarily).

On the other hand, eating breakfast is likely to make you feel more positive about yourself and more motivated.

[1] Smith et al., Skipping breakfast among 8-9 year old children is associated with teacher-reported but not objectively measured academic performance two years later, BMC Nutrition, 2017, 3, 86, doi:10.1186/s40795-017-0205-8 [online article]

[2] Bamberg, C., & Moreau, D., Acute effects of fasting on cognitive performance: A systematic review and meta-analysis, Psychological Bulletin, 2025, 151, 1147–1169, https://doi.org/10.1037/bul0000492 [online article]

[3] Raza, S., Akhtar, P., Memon, F., Muhammad Rahimoon, Munwar Us Salam, & Rahimoon, U., Association of Breakfast Skipping with Academic Performance among Undergraduate Students, Indus Journal of Bioscience Research, 2024, 2, 660-665. https://doi.org/10.70749/ijbr.v2i02.247 [online article]

[4] Kaduk K, Kaeber M, Kühnel A et al., Glucose levels are associated with mood, but the association is mediated by ratings of metabolic state, eBioMedicine, 2025, 124, 106035, https://doi.org/10.1016/j.ebiom.2025.106035 [online article]

[5] PE Flores Kanter, LA Medrano, H Conn, Does Mood Affect Self-Concept? Analysis through a Natural Semantic Networks Based Approach. Int J Behav Res Psychol, 2015, 3, 114-120, http://dx.doi.org/10.19070/2332-3000-1500022 [online article]

But what if we could make our sleep even more effective?

A recent study* by a team in Germany investigated whether doing some intense exercise before sleeping might improve people’s memory the next day. To test this, their participants did 20 minutes of high intensity interval training (HIIT; see below for details) exercise in the evening, finishing about 3 hours before their normal bedtime, and then slept for approximately 6 - 8 hours. In the morning they were shown some material to memorise, then given a reaction speed test, followed by a memory test on the earlier material. The participants also repeated the reaction speed and recall tests 24 hours later.

The idea was to test whether the pre-sleep exercise had any effect on the participants’ capacity to encode the test material in their memory, i.e. to see if it had improved their ability to memorise things.

The researchers found that the participants who did the HIIT training performed better on the subsequent memory tests than both a control group (who watched TV for an hour), and better than a second exercise group who did moderate intensity continuous training (MICT) for 60 minutes instead of HIIT. Both the HIIT and MICT groups did their exercise on an erg bike (fancy stationary bike) …while watching the same TV show as the control group.

Even more interesting, the most significant differences in performance on the memory tests were seen for participants with generally lower memory performance overall, i.e. the HIIT exercise was most helpful for those participants who struggled most with encoding the information. This effect is illustrated in the figure above where you can see that the performance of the HIIT group on a memory test was more enhanced for the low performers, when compared with the control group and the MICT group.

After their exercise, all of the participants were required to wear a sleep monitor device (think Fitbit). Surprisingly, there was no significant difference observed in the sleep quality of the participants. Unfortunately the level of detail available from the monitors wasn’t sufficient to determine if the participants were experiencing different durations of REM (rapid-eye-movement) or non-REM sleep. This would have been useful because it’s the non-REM sleep in particular that has been associated with renewing the brain’s capacity to encode information. So while the study showed that the HIIT exercise produced an improvement in memory, it doesn’t tell us why. However, the implication is that doing some HIIT exercise before bed basically helps our brain to refresh its ability to encode new information while we sleep.

When considered in combination with the difference in effect between the “low performers” and the “high performers” this also suggests that even if some people have a lower ability than others to renew their capacity for encoding new information while they sleep exercise can be used to mitigate this disparity. If you are one of the lucky people whose capacity renews more efficiently then doing some HIIT before bed won’t help as much, but it’s also not going to hurt.

Exercise and memory

But is the sleep even required to obtain this gain in memory performance? The German study seems to be the first of its kind to examine the effect of exercise before sleeping, but there are certainly other studies in the literature that demonstrate the beneficial effects of high-intensity exercise on memory.

For example, a 2023 study** from the University of Northumbria looked at the effect of running on memory performance. In that study, participants were shown some material to memorise, then they went for a 25 minute interval run. Their performance on subsequent memory tests, both immediately after the run and then again 24 hours later, was significantly better than for a control group of participants who walked for 25 minutes instead.

A key difference between this study (runners) and the previous one (cycling), is the order in which the exercise was performed with respect to the point when the participants were given the material to memorise. For the cyclists, the exercise happened first (they were given the material the morning after), but for the runners the exercise happened second (they ran after being given the material to memorise). A 2024 review*** of many many different studies on exercise and memory found that overall the literature seems to support a larger effect from exercising after memory encoding compared with exercising before memory encoding, but the difference is small.

A key element of post-encoding exercise (i.e. exercising after learning) is that participants shouldn’t think about the material they had just learned while exercising. This seems to be because the cellular consolidation happening in our brains that strengthens synaptic connections and has been proposed to lead to improved memory retention is expected to be an automatic and opportunistic process that can actually be disrupted by trying to re-process the same information. Again, this is why sleep is so important - it doesn’t come with any distractions to disrupt the process.

So why is high-intensity exercise also thought to be beneficial? Does it provide the same distraction-free state as sleep? At some level, perhaps, but in sensory terms it’s much more disruptive. Instead, it has been proposed that high intensity exercise results in increased blood flow to the hippocampus, and there is also some emerging evidence that it produces an increase in the production of brain-derived neurotrophic growth factor (BNDF), which is again important for those synaptic connections and is consequently implicated in the success of memory consolidation.

Exercise and timing

The timing of both pre- and post-learning exercise (with the exception of the German sleep study) seems to that learning is enhanced when it is combined with another experience within a reasonable (~2 hour) time window. However, there also seems to be a second useful window, approximately 11 hours after the initial learning, which is associated with promoting long-term retention of information****. Exercising for 30 minutes during this later window has also been shown (in rats, not humans…) to enhance memory performance for up to seven days.

Consequently, it may seem that timing is quite intricately connected to these results. However, the reality is that, with a busy undergraduate timetable, it’s never going to be possible to schedule in HIIT sessions to accompany each piece of teaching.

So what’s the best way to think about this?

Well, 20 mins of HIIT while watching TV in the evening sounds like a pretty minimal place to start. The benefits of this are expected to manifest while you are asleep.

If you can also muster up the time to throw in another 30 minute exercise session at some point during the day then that may provide an additional boost via increased hippocampal blood flow and enhanced BNDF production.

And, of course, there are also supposed to be other benefits associated with a bit of exercise aside from improved learning. So it’s really a win-win.

HIIT training (cycling): a 3 minute warm up, then 3 x 3 min cycling at 90% maximum effort separated by 2 min cycling at 25% maximum effort, then a 4 minute cool down.

HIIT training (running): a 12 minute run, alternating (using lamp-posts) between a high perceived rate of exertion (RPE) and a moderate RPE, followed by 1 minute of walking rest, followed by another 12 minute interval run.

[*] Ramirez Butavand, D., Nagel, J., Feld, G.B. & Simon Steib, High intensity exercise before sleep boosts memory encoding the next morning, Nature Scientific Reports, 2025, 15, 20388, https://doi.org/10.1038/s41598-025-07880-z, [online article]

[**] Makepeace R, Craig M, Higher intensity exercise after encoding is more conducive to episodic memory retention than lower intensity exercise: A field study in endurance runners, PLoS ONE, 2024, 19, 9, https://doi.org/10.1371/journal.pone.0308373, [online article]

[***] Qazi AS, Schmid D, Gridley N, Lambourne K, Daly-Smith AJ and Tomporowski PD, The effects of acute exercise on long-term episodic memory: a systematic review and meta-analysis, Frontiers of Cognition, 2024, 3, 1367569, doi:10.3389/fcogn.2024.1367569 [online article]

[****] Karine Ramires Lima, Ana Carolina de Souza da Rosa, Gabriela Cristiane Mendes Gomes, Gabriela Jaques Sigaran, Anna Cecilia Perretto, Pamela Billig Mello-Carpes, Acute exercise performed during the late consolidation phase improves memory persistence by hippocampal protein synthesis and catecholamine modulation, Pharmacology, Biochemistry and Behavior, 2024, 245, 173893, https://doi.org/10.1016/j.pbb.2024.173893, [online article]

A 2023 study* of ~500 first year college students in the USA recorded an average of 6 hours and 37 minutes sleep per night, well below the recommended guideline amount of 8 to 10 hours per night. It seems sensible to ask, does this deficit have an impact on the learning outcomes for those students?

Perhaps unsurprisingly, the answer seems to be yes. Although the effect is not enormous. Looking at the correlation between the number of hours of sleep that students recorded (using a wrist monitor - think Fitbit) and their final grades, the authors found a positive association. Consistently getting more sleep was associated with getting a higher grade.

You might wonder whether this result simply reflected the students’ own level of self-regulation, i.e. students who put more effort into making sure they sleep well are also the same students who put more effort into making sure they study well, but in fact a positive correlation held even when the study took account of student’s existing grades. However, while the correlation is positive, it’s also pretty weak: each extra hour of sleep per night only equated to about 0.07 GPA points, which is approximately 1% difference in a percentage grade.

However, there did seem to be a sleep duration threshold in the results. Students who averaged fewer than 6 hours sleep per night did notably worse than students who averaged 6+ hours per night. This is illustrated in Figure 1 in the left hand plot. Moreover, comparing final grades (Spring) to grades from the first semester (Fall) showed a significant dip in performance over the course of the year for students sleeping <6hrs, which wasn’t present for students who got more than 6 hours sleep a night. For those students, the effect of missing out on so much extra sleep had accumulated over the academic year. This is illustrated in Figure 1, where the plot on the right shows the difference between students’ mid-year (Fall Term) results and their final year results (Spring Term). A negative value means that their mark decreased, and a positive value means that it increased.

Other variables didn’t seem to have much of an impact. For example, it didn’t matter whether students went to sleep at 10pm or 2am (…and most students went to sleep at 2am). Likewise, the same study found that demographic differences had no measurably significant effect.

The key message seems to be: get more than 6 hours sleep a night.

It would be remiss of me not to note that other studies** in the literature, which also used Fitbit sleep tracking, have found much stronger positive correlations between student grades and sleep duration. However, these studies tend to have smaller sample sizes and do not control for individual student GPAs. I’m not ignoring those results - I think they all feed into a single message, which is that getting some extra sleep is not going to hurt your learning outcomes. In fact, every one of these studies seems to contradict the idea that studying all night will lead to better learning outcomes.

Sleep and memory

Why does a sleep deficit affect the learning process? The effect seems to be a combination of the way in which the cognitive (i.e. brain) functions associated with attention, memory and decision-making are impacted.

Studies on brain activity during sleep (admittedly on rats, not students) have suggested*** that the synchronised “sharp-wave” ripples of neuron activity in the brain associated with the transfer of information from short-term to long-term memory don’t only happen during the day, but also while we sleep. Disrupting these ripples, which have been described as a key cognitive biomarker for memory and understanding, led to impaired performance in later tests. Notably, a few subsequent nights of uninterrupted sleep wasn’t enough to completely reset the disrupted ripples for the rats - the memory impairment persisted even when the lost sleep had been regained.

This result seems to support the idea that a consistent lack of sleep affects your ability to internalise the new information that you are learning, because its transfer into your long term memory is disrupted.

Staying up all night cramming before an exam may not be the answer either. In another recent study**** (this time on humans), fMRI scans of participants who were deprived of sleep showed that activity in the parts of their brains associated with working memory, attention and information processing such as the Cerebellum, Insula and Thalamus (i.e. important bits for taking tests) were significantly suppressed relative to the same regions in the brains of participants who were not sleep deprived. Sleep deprived participants took 20% longer to answer test questions and scored ~5% lower on average than their well-rested equivalents.

So don’t be fooled into thinking that you need to stay up all night studying. If you’re tired then call it a night and hit the hay.

[*] J. David Creswell et al., Nightly sleep duration predicts grade point average in the first year of college, Proceedings of the National Academy of Sciences - Psychological and Cognitive Sciences, 2023, 120, 8, https://doi.org/10.1073/pnas.2209123120 [online article]

[**] Yao Zhao & Haoyu Zhou, Functional data analysis of college students’ sleep patterns and their relationships with academic performance and social networks: A four-year longitudinal study, PLoS One, 2026, 21, 7, https://doi.org/10.1371/journal.pone.0351120 [online article]

[**] Kana Okano, Jakub R. Kaczmarzyk, Neha Dave, John D. E. Gabrieli and Jeffrey C. Grossman, Sleep quality, duration, and consistency are associated with better academic performance in college students, Nature Science of Learning, 2019, 4, 16, https://doi.org/10.1038/s41539-019-0055-z [online article]

[***] Bapun Giri, Nathaniel Kinsky, Utku Kaya, Kourosh Maboudi, Ted Abel & Kamran Diba, Sleep loss diminishes hippocampal reactivation and replay, Nature, 2024, 630, 935, https://doi.org/10.1038/s41586-024-07538-2 [online article]

[****] Anping Ouyang et al., Napping restores working memory and brain activation impaired by sleep deprivation, IBRO Neuroscience Reports, 2025, 19, 709–717, https://doi.org/10.1016/j.ibneur.2025.09.013 [online article]

1. Your brain on ink [link] - far and away my most popular article.

   How handwritten note-taking consistently outperforms typed notes for learning outcomes.

2. Attention is all you need [link]

   How to stay focused in lectures, and why your notes are more than just a revision aid

3. Fast forward [link]

   How much should you speed up a lecture podcast?

4. Goldilocks AI [link]

   Can AI assistance in note-taking help with learning outcomes?

5. Noti-vation [link]

   Does your note-taking style have an impact on motivation?

Low noise levels and natural lighting improve focus and support well-being

It is a truth universally acknowledged, that an academic in possession of a laptop, must be in want of a table with a power socket. Or words to that effect.

In a recent study* mapping how students chose their study spots in a Chinese university library, the single most important factor was the proximity of a power socket.

This reality of modern education has been driven by a number of factors, not least that most learning resources are now provided to students online, so the use of electronic devices for personal study has also risen. Moreover, students increasingly study outside classrooms, using a variety of informal spaces - libraries, cafés, residences - to support their own learning practices. So the pure functionality of those spaces to accommodate that learning is something that has evolved in recent years. However, beyond that base functionality, what is it about study spaces that students value the most - and what do they consider most helpful for how they can be enabled to learn?

Another recent study**, this time from the University of Manchester, looked at the factors that affect students’ ability to focus on their work when studying in different places. The results are summarised in the figure below.

Perhaps unsurprisingly, the biggest factor that students cited was noise. “People talking” was voted the most distracting source of noise, which makes it perhaps unsurprising that open spaces were also cited as a source of distraction. There was also a strong preference for study spaces that provided natural light, followed by those that had muted lighting schemes. Bright lights were viewed as least favourable to studying.

The study also examined whether the responses were consistent between neurotypical and neurodivergent students. Overall, neurodiverse students found all factors more affecting, but there were also notable enhancements in the distraction due to open spaces, and light, compared to responses from neurotypical students.

Neurodiverse students also cited cluttered or busy spaces containing more visual elements/stimuli as being a distraction, although when asked to rank their priorities for a study environment they ranked the level of clutter as least important. Noise and light were ranked as most important.

The authors report that these results are consistent with other studies looking at environmental factors influencing learning practices. The key message is that a quiet place, with some degree of privacy and a good amount of natural light, is the probably the best option for maintaining focus.

Affordance theory

Given these considerations, is there a way that we can design a single ideal space for learning? The answer seems to be no, and the reason is that students will typically choose a given study environment based on the specific task they are planning to undertake. Moreover, recent studies using (e.g.) student photo diaries*** show that they will also adapt that space to serve their individual needs. This resonates with the idea that autonomous motivation is important for learning, i.e. that students are active participants, rather than passive receptors.

So, rather than try and create a single perfect study space, it is better to consider the learning opportunities that different environments afford. This is an example of affordance theory, which was first introduced by the ecological psychologist James J. Gibson in 1979. Gibson proposed that people perceive an environment primarily in terms of the opportunities for action it offers, rather than as a collection of objects with abstract properties.

The upshot of this theory is that different environments represent different opportunities to different people depending on their needs and capabilities, i.e. for an adult a chair may be something to sit on, while for a child it could be a climbing frame. This is a great explanation for the popularity of cafés as study spots: they tend to offer individual tables, a low-noise environment, and also often a good amount of natural light. The availability of caffeinated drinks probably doesn’t hurt either.

The low-noise, rather than silent, aspect of cafés is also possibly a positive. Students in both the University of Manchester study** and the Shanghai Jiao Tong University library study* noted that they often preferred a low-level of background noise to complete silence and isolation.

Consequently, to a student, a café represents not just a coffee provider, but an environment that affords the opportunity for focused study. The online nature of learning resources, and the ubiquity of digital devices, also means that students are able to adapt the basic café environment to meet their needs.

Study spaces and well-being

The design of a study space is important for more than just learning effectively - it can also influence a student’s individual well-being and motivation. The study** from the University of Manchester, highlighted that as well as being distracting, excess noise and bright lights also caused increased anxiety, especially for neurodiverse students.

This idea that where you study outside the formal learning spaces of lecture theatres and labs can influence your well-being is not a completely new one, but actual empirical studies are only now starting to emerge.

Moreover, measuring an individual’s well-being is not a trivial thing. But two of the same key factors that seem to influence measures of well-being are those that also improve focus: low noise levels and natural light.

[*] Zhan Zhang, Alima Adalaiti, Yi Sun, Haodong Zhang, Xiaoyu Rong, Yunji Hua, Linjun Lu, Mechanisms of environmental stimuli on learning efficacy: An empirical study of university library spaces, The Journal of Academic Librarianship, 2026, 52, 103243, https://doi.org/10.1016/j.acalib.2026.103243, [online article]

[**] Maria Mercè Canal, Nida Yagmur Yeserdir, Emily Hynds and Tristan Pocock, Designing sensory-inclusive study spaces: comparing environmental preferences and barriers for neurodivergent and neurotypical university students, Frontiers of Education, 2026, 11, 1813799, doi: 10.3389/feduc.2026.1813799 [online article]

[***] Mike Frazier, Qiyang Lin, Garam A. Lee, Kui Xie, From space to place: A Visual Grounded Theory analysis of how college students shape study environments, Learning in Context, 2026, https://doi.org/10.1016/j.lecon.2025.100016, [online article]

Short term difficulty can be a desirable thing for long-term learning

There are a few interesting concepts that have come up multiple times in the literature as I’ve been writing these posts. One of these is the perception gap between what students often think is most useful for learning vs. what the evidence shows is actually most useful. Here are three quick examples:

1. Automated AI vs. Minimal/Intermediate AI for Notetaking

   Students using an Automated AI notetaker in lectures had the perception that it was much more beneficial for their learning than either a direct transcription of the notes or even using AI summaries to support personal note-taking. However, this perception was not supported by test data - students using the Automated AI notetaker did worse in the course assessment than students using the other methods, even when they were permitted to use their notes in the test.

   [Read the article]

2. Re-reading of lecture material vs. Retrieval Practice

   When asked to compare the impact of re-reading lecture material vs. AI-supported formative quizzes, students ranked re-reading of the material as much more helpful. In fact, when preparing for a delayed test, students even ranked the AI-supported retrieval practice as detrimental to their learning. However, the students who used the AI-supported retrieval practice got better test scores by ~10% on average than the students who re-read the material.

   [Read the article]

3. Blocked vs. Interleaved problems.

   Students self-reported that weekly problem sets which interleaved questions on different topics from a lecture course were less helpful for learning compared to alternative problem sets that organised questions into blocks by topic. However, in surprise tests at the course mid-point and at the end of the course, students who had completed the interleaved problems scored 20% higher than students who had completed the blocked problems.

   [Read the article]

Metacognitive illusions

A common theme among these three examples is that the method that was (incorrectly) perceived to be less helpful was the more difficult choice: writing your own notes is more difficult than letting AI produce them automatically; answering quiz questions is more difficult than re-reading lecture notes; students score more poorly on problem sheets with interleaved topics than those with blocked problems. In the short term all of these things are more difficult, but in the long term they all lead to better results.

This kind of mismatch between perception and reality is more common than you might think. It’s even got a name: metacognitive illusion. In this case, the illusion is that people positively associate the ease of doing something with its anticipated effect on their learning, when in fact some level of difficulty can be a desirable thing.

One consequence of this metacognitive illusion is that when students are asked to judge their own learning they often equate the easier option with the most effective one, and that can lead them to make choices that are ultimately less beneficial for learning outcomes. The psychological impact of the short term difficulty is to make it feel like you are learning less, when in reality you are learning more …and when your intuition is no longer a reliable measure that makes it difficult to make informed choices.

Desirable Difficulty

But how can you judge when the level of difficulty is just right? There will obviously be limits on quite how difficult something can be and still produce that positive long term effect. It’s another Goldilocks situation. It’s also true that there are different kinds of difficulty, that result in different types of cognitive load. Not all of these difficulties are desirable when it comes to learning, and recognising which is which can be helpful. Fortunately, cognitive load theory (CLT) summarises them quite nicely.

You have:

Intrinsic load This is the cognitive load associated with the intrinsic difficulty of the topic you’re learning. It’s tightly tied to what you know already, because the intrinsic load for someone learning something from scratch will be quite different from the intrinsic load for someone who already has relevant expertise in the topic.

Extraneous load This is basically the bad cognitive load; the one you want to minimise. It comes from all those things that make it harder to interpret what you’re looking at: distractions, confusing presentation, additional irrelevant information, etc.

Germane load This is the most beneficial type of cognitive load for learning. It’s the bit associated with constructing understanding. Germane load has some overlap with intrinsic load, because it will depend on existing background knowledge. Some researchers lump them together, suggesting that germane load is part of the overall intrinsic load. Either way, I think it’s useful to be explicit about what this piece of mental effort does, because it’s basically a map to new knowledge.

There are suggestions emerging in the literature about how we can use the framework of CLT to understand when we’re being faced with desirable difficulty, i.e. things that feel more difficult but ultimately lead to improved learning outcomes, vs. undesirable difficulty, i.e. extraneous load or excessive intrinsic load.

Ultimately, these seem to be variations on Self-Regulated Learning practices. The key to self-regulated learning is dedicating some time to self-reflection, evaluating whether the choices you’ve made in your learning practice are actually leading to the outcomes that you want - and adapting them if not. Integrating this self-reflection with CLT is in some ways quite straightforward. For example, CLT provides a framework to hold your choices up against and assess whether there are elements that map to undesirable difficulties.

However, recognising desirable difficulties is still not completely trivial, because the cadence of your self-reflection might not match the time-scales associated with the longer term retention that characterises these desirable difficulties. For the moment, it’s probably better to try and have an awareness of which learning practices are already known to have desirable difficulty, e.g. retrieval practice, spacing and interleaving, and to recognise that just because something is more difficult it’s not necessarily less useful for learning.

Which should hopefully offer some reassurance when those problem sheets are driving you crazy.

Wesley Pyke , Johan Lunau and Amir-Homayoun Javadi, Does difficulty moderate learning? A comparative analysis of the desirable difficulties framework and cognitive load theory, Quarterly Journal of Experimental Psychology, 2025, Vol. 78(10) 2181­ –2195, DOI: 10.1177/17470218241308143

Anique B. H. de Bruin, Felicitas Biwer, Luotong Hui, Erdem Onan, Louise David, Wisnu Wiradhany, Worth the Effort: the Start and Stick to Desirable Difficulties (S2D2) Framework, Educational Psychology Review, 2023, 35, 41, https://doi.org/10.1007/s10648-023-09766-w

Asher Koriat, Easy comes, easy goes? The link between learning and remembering and its exploitation in metacognition, Memory & Cognition, 2008, 36 (2), 416-428, doi: 10.3758/MC.36.2.416

We all do it.

Cramming for exams is a little like carb-loading before a big race. It might give you a bit of a boost, but ultimately, consistent attention to your nutrition during training is what really pays off.

The advantages of consistently staggering, or spacing, your learning over the full duration of a lecture course is one of the most well-documented and robustly supported practices in the education literature. So much so in fact that the question in the literature seems not to be whether you should space your learning, but rather how you should space your learning to get the best outcomes.

For example, if you are going to do a problem sheet after a lecture, should that problem sheet group together questions by topic in blocks, or should it interleave questions on different topics, potentially even including questions on topics from previous lectures?

In a recent study* on undergraduate physics students, the authors tested exactly this distinction across an 8 week lecture course. The course had three lectures a week and each lecture was accompanied by a problem sheet. Half the class were given weekly problem sheets with blocks of questions, and half had problem sheets with interleaved questions, including interleaved topics from earlier lectures. During the lecture course, students were asked (i) how difficult they found the problem sheets for each lecture, and (ii) whether they thought the problem sheets were helping them learn. The students doing the interleaved problems found them more difficult on average, and also thought they were not as helpful for learning. This more negative perception was reflected in their marks on individual problem sheets, which were on average lower for the students in the interleaved group.

However… when the researchers ran surprise tests on the course material after 4 weeks and again at the end of the course, the students who had completed the interleaved problems scored 20% higher than the students who had completed blocked problems. For the second surprise test at 8 weeks, the difference was > 20%.

I find this difference so massive that I struggle to believe it. To rationalise it I think it’s important to note that these results were for surprise tests, where students had no opportunity to cram beforehand. The actual median marks per test were pretty low - especially for the second surprise test after eight weeks. For the formally scheduled mid-term exam after 4 weeks, students did cram (based on a exit survey the study did) and that additional cramming halved the difference in grades between the two groups of students, although the interleaved group still did marginally better. Unfortunately the final midterm at 8 weeks was affected by campus closures due to the pandemic so most students didn’t take it and so we don’t know what the outcome would have been.

Interleaving practice problems on different topics seems to have a positive benefit for learning outcomes.

For other research studies** that base their results on formal examinations the improvement in the mean mark for interleaved vs. blocked problems seems to be somewhere in the range of 2 - 5%, but it varies across subjects. There has been a suggestion in the literature that an interleaved approach is more beneficial when (i) it provides opportunities to integrate knowledge from previously seen topics with learning on newer objectives, which helps more with courses such as mathematics and physics where new material tends to build on older material, (ii) where problems are typically short answer formats or calculations, rather than multiple-choice, and (iii) where feedback on the problems is provided.

So while the principle of blocked vs interleaved problems seems simple, there are a lot of potential confounders that can affect the extent to which it is effective. For me the take away message from the literature seems to be: don’t worry too much about it, but… maybe mix up the order in which you do practice problems every so often because on balance it’s likely to have a positive effect.

[*] Joshua Samani and Steven C. Pan, Interleaved practice enhances memory and problem-solving ability in undergraduate physics, Nature Science of Learning, 2021, 6, 32, https://doi.org/10.1038/s41539-021-00110-x [online article]

[**] Campbell R. Bego et al., Single‑paper meta‑analyses of the effects of spaced retrieval practice in nine introductory STEM courses: is the glass half full or half empty? International Journal of STEM Education, 2024, 11, 9, https://doi.org/10.1186/s40594-024-00468-5 [online article]

Using AI-generated quiz questions for retrieval practice leads to improved learning outcomes.

I recently learned the term “zombie concept”. It describes a theory in the literature that, no matter how many times someone kills it with empirical evidence, keeps coming back again. Examples of zombie concepts in the education literature are things like learning styles, i.e. describing students as visual learners or auditory learners etc., and growth mindset training, which to me basically sounds like trying to get students to manifest better learning. The supposed benefit of neither example is robust to impartial testing, but they continue to re-emerge time after time.

Interestingly, the kind of concepts that seem to survive in these zombie loops are generally those that promote passive and comfortable practices, which give a bit of a warm fuzzy feeling. Because they feel good, this misleads people to believe that they are more beneficial than they truly are, i.e. there is a perception gap between the reality of their impact and the degree of help that they actually provide.

A similar example of a largely passive learning practice that has a perception gap is re-reading of lecture material. In this case, it’s not that re-reading material has zero benefit, but rather there is a perception gap in terms of its effectiveness when compared to more active forms of learning - such as formative quizzes. Possibly because those quizzes feel more uncomfortable in the moment.

Using low-stakes quizzes (whether provided by the lecturer or self-created) to pull information out of memory is a variant of something known more generally as retrieval practice or active retrieval. The basic premise is that testing yourself is not a neutral process, i.e. every time you actively test yourself (whether in a formative or summative setting) it leads to improved retention. Unlike zombie concepts, retrieval practice performs consistently well under empirical testing. Notably though, in controlled studies comparing active retrieval practice to passive re-reading of lecture material, students self-report that re-reading of the lecture material is more beneficial for their learning. Even though their test marks say something different.

For those who have read my article on AI notetaking, you’ll notice a recurring pattern here: someone’s perception of what is most useful does not always match their learning outcomes. AI can be particularly misleading in this respect, because the efficiency it provides can seem very useful …masking the fact that it can be problematic in other ways.

Cognitive off-loading

One of the concerns about using AI in an educational setting is that it can result in cognitive off-loading, where information is being parsed by the AI and not by the student. For obvious reasons, this does not help someone to learn things themself.

A correlation study* in China following more than 25,000 students found that AI adoption lowered their monthly (closed-book) exam marks by 20%, even though their homework scores increased by the same amount. The drop in test marks took about six months post-AI-adoption to reach its nadir, but sadly the long term effects were even more pronounced, with similarly low exam scores affecting the entrance exams for the students’ next level of education.

This study is one of a number I’ve seen recently that reports the negative effects on learning outcomes associated with general, undirected, use of generative AI. But are there ways in which generative AI can be used to help accelerate or improve learning? When I asked this question previously in the context of note-taking, it did turn out that there was a sweet spot. It seems that another sweet spot may be the use of generative AI to support retrieval practice.

AI-supported retrieval practice

One of the barriers to retrieval practice is that there needs to be some structure around it, even if that structure is minimal. In its most basic form, active retrieval can look as simple as trying to write down everything that you remember about the content of a particular lecture or seminar. More typically these days, it might involve formative quizzing either during or after a lecture. However, not all lecturers provide these kinds of quiz and generating quiz questions from scratch can be challenging for students who are not yet familiar with the material. So this seems like an excellent opportunity for directed use of generative AI.

In a recent experiment**, researchers from the University of Arizona split students into two groups: one where the students used generative AI to produce retrieval practice questions from the class material and tested themselves, and a second group where the students simply re-read the material directly. Individual students then self-reported how much they thought the method they had used had helped them to understand the material. As you might guess at this point, the students ranked re-reading of the material as much more helpful. In fact, when preparing for a second test 2 days later students even ranked the AI-supported retrieval practice as detrimental to their learning.

But, surprise, surprise… the students who used the AI-supported retrieval practice got better test scores by ~10% on average than the students who re-read the material. The one exception to this was for a multiple-choice test immediately after the class, where all students got the same scores on average.

Something I noticed is that the 10% gain found by the University of Arizona study is consistent with a similar study that I wrote about previously. In that case, the researchers looked at the difference between test scores from students who did short formative quizzes after a lecture, and those who were given review slides instead. The comparative impact of the quizzes on average test marks was also approximately +10%, relative to reading review slides. The underlying principle in both studies is the same: active retrieval practices, such as quizzes, lead to improved learning outcomes when compared to passive learning practices such as re-reading notes, or viewing review slides.

As a student, the benefit of generative AI is that you no longer have to rely on your lecturer to provide the quizzes!

[Disclaimer: The University of Arizona research is only presented as a poster at this point, which means that the detail is somewhat limited and the methodology hasn’t been peer-reviewed yet. I’ll keep an eye out for the paper and update this article when it appears. Likewise, I do not have access to the full paper on the Chinese study, so the cited results come from the abstract. Links below.]

[*] D. Strömberg, V. Lei & Y. Wu, The Generative AI Learning Penalty: Evidence from Chinese Secondary Education, Centre for Economic Policy Research (CEPR) Discussion Paper No. 21577, 2026 [online abstract]

[**] Jonathan G. Tullis, Using GenAI To Support Retrieval Practice, International Teaching-Learning Conference Lilly Conference, May 2026, [online poster]

Peer-learning is the practice where students learn from other students rather than just from their lecturers. This can be a self-organised and informal social activity that students develop organically during their time at university, but increasingly it’s also something that is organised directly by individual academic departments following a structured model.

Typically, the peer-assisted learning programs offered by universities are optional, with students choosing whether or not to participate. But are they beneficial? With so many calls on your time as a student, is joining a peer-learning group going to be a sensible investment? The evidence says… Yes.

Structured peer-learning consistently improves learning outcomes

What does the data say

In a recent meta-analysis*, researchers from the University of Kwa-Zulu Natal (UKZN) compiled results from more than 1000 individual studies on peer-learning. I’ve focused here on the subset of those studies that looked specifically at the Supplemental Instruction (SI) model of peer-learning, which is more often referred to as PASS or Peer-Assisted Study Sessions. This is the kind of peer-learning model we have operating in my own university department.

PASS is a “near-peer” form of peer-learning, which means that both the tutor and the tutees are at the same level, i.e. undergraduate, but might differ in seniority by one or two years. Near-peer models are expected to be successful because the tutees share the same kind of subject context and language as the tutees, so they can explain things in an understandable and accessible way. This makes me feel old.

The UKZN meta-analysis included 21 different SI/PASS studies: 14 from the USA, 2 from South Africa, 2 from Australia, and one each from Canada, Hong Kong and Sweden. There’s a huge variation in these studies. The smallest one considered a sample of 70 students, while the largest included 11,738 students! That’s Study 19 - the error bar is so small you can’t even see it in the figure below.

Ok, straight away, just from looking at the data you can see that the global signal here is going to be positive, because there are literally no negative effects among the individual studies. You might even start to wonder if these results are biased by people omitting to publish negative results. It seems not. The authors of the meta-analysis tested this and concluded that the sample was not biased by publication choices.

However, since the studies are very heterogeneous you can’t just take the average of all the reported outcomes to find an overall effect. These studies contain different student populations, different settings, different course types, and different implementations of the peer-learning model. The authors of this meta-analysis calculated that the degree of variance due to real inter-study differences was about ~82%, which is a very good indication that the scatter you can see in the results is not just due to sampling effects.

In this kind of case, deciding if the studies are probing a global underlying effect requires a slightly more involved calculation than simply taking an average. That’s what the random-effects (RE) model does. You can think of it as estimating the global signal in the data while taking into account the individual differences. Using a random effects model, the meta-analysis found that the probability of the PASS schemes having zero global positive effect on student outcomes was 0.01%. That’s the same probability that you can guess a stranger’s 4-digit PIN in a single try.

Overall, the authors found that the actual global effect of participating in a PASS scheme was +0.40 (+/-0.10). But what does that mean? Put simply, that positive effect would shift the distribution of grades for a course where the average grade was originally 65% with a standard deviation of 10% (blue curve) to one where the average grade was approximately 3 - 5% higher (pink curve). If the pass mark for that course was 40%, this same shift would reduce the number of course failures by 70%.

You can see this illustrated in the figure below.

Zone of Proximal Development

One of the often cited reasons for the success of peer-learning is how it aligns with a piece of socio-cultural theory known as the Zone of Proximal Development, or ZPD for short. The idea behind ZPD is that what we can learn next is bounded; there are concepts that are so far beyond our current level of understanding that even with help we wouldn’t be able to learn them. Exactly what we can learn lies in a zone between this outer boundary and an inner core of things we know already. That zone is the ZPD: where we can be helped to learn new things. The theory posits that the key to growing our core of independent ability is guidance. Interactions in the ZPD guide new learning before our understanding is internalised to become part of our core knowledge. As our core grows, so does our ZPD.

The fact that near-peer learning sits firmly in the Zone of Proximal Development is the general argument for why things like SI/PASS work. However, it’s a bit hand-wavy. Possibly because Lev Vygotsky, who originally conceived ZPD, died young, leaving the theory under-developed, open to interpretation …and possibly to mis-translation**.

Irrespective of their interpretation, the results of the 2023 UKZN meta-analysis* are consistent with a number of other parallel studies looking at peer-assisted and active learning approaches. The size of the global effect even seems to come out in the same ball park each time. That’s easy to interpret: peer-learning is a good time investment.

[*] Bengesai, A. V., Amusa, L. B., & Dhunpath, R., A meta-analysis on the effect of formal peer learning approaches on course performance in higher education, 2023, Cogent Education, 10(1) https://doi.org/10.1080/2331186X.2023.2203990 [online article]

[**] Peter Smagorinsky, Deconflating the ZPD and instructional scaffolding: Retranslating and reconceiving the zone of proximal development as the zone of next development, Learning, Culture and Social Interaction, 2018, 16, 70, http://dx.doi.org/10.1016/j.lcsi.2017.10.009, [online article]

Structured note-taking can increase student motivation

Motivation

Motivation is an internal self-determined drive to both initiate and to sustain the activities that help us achieve a particular goal. It’s a key component of maintaining a self-regulated learning practice because it supports the activities that underpin the self-regulation cycle: forethought, practice and self-reflection. We are motivated to make choices, to be consistent in the execution of our pre-determined plan, to take time to reflect on whether those choices are working for us. Consequently, the literature** has consistently demonstrated that personal motivation is positively associated with better learning outcomes and improved well-being.

In short, motivation is great. If you have some.

And therein lies the problem. Motivation is not something that you can simply turn on and off at the flick of a switch. Nor is it just one unified thing: motivation can be divided up into different types. For example, waking up one morning and rushing to put the bins out because you’re motivated to start your day in an organised manner is quite different from waking up and rushing to put the bins out because your mum shouts at you to do it. In the first case your motivation is internal, personal, self-determined; in the second case, the motivation is external, imposed, mastigophobic (yes that is a word).

Self-determination theory, an influential theory of human motivation, says that motivation exists as a continuum, with the lowest degree of self-determination at one end of the scale, and the highest degree of self-determination at the other.

The implication of this is that two people could have exactly the same amount of motivation to do something, but for wildly different reasons. Going back to the bins example - the same amount of effort is put into the action in each case, i.e. there is an equivalent level of motivation to perform the task, because the bins get taken out either way; what differs is the type of motivation in each case. And this difference in motivational type has consequences. In the first case, there is positive feedback from completing the task. You’ve crushed a personal goal, who doesn’t love that? But in the second case, the feedback is neutral or potentially negative: you’ve been made to do something by someone else. Lingering guilt, residual annoyance, or simply avoidance of potentially negative consequences are not things that generally make you feel good.

So it’s not simply how much work you put into something that determines the outcome. The reason that you do it matters as well. Or equivalently, it’s not just how much motivation you have to complete a task that matters, it’s also the type of motivation.

Beyond simply feeling good, the literature suggests that it is those autonomous motivations, intrinsic and identified, that are associated with more positive learning outcomes, with some studies linking autonomous motivations directly to longer term retention of information. The controlled types of motivation, external and injected, do not have equivalently positive associations with learning - in fact some studies have suggested that they are actually detrimental to learning outcomes.

When it comes to building motivation for a particular subject or lecture course this means that it’s not only important to increase the level of motivation, but also to instil the correct type.

Pro-note and proud

In a recent study*, researchers looked at whether the way in which students took their notes during lectures had an effect on their intrinsic motivation - as well as their test performance.

The students were divided into four groups, each of which used a different note-taking method: Cornell, Parallel, Digital and Sentence.

• In the Cornell method of note-taking a student basically divides their page into two columns. They write their detailed notes in the right-hand, larger, column, and then annotate those notes by highlighting where key concepts are introduced, or where they have questions, on the narrower left-hand side column. Following the lecture, they use space at the bottom of the page to summarise the lecture in their own words. It reminds me a bit of the Tufte style of textbook, where key concepts are pulled out and highlighted in the margin parallel to the main text. The idea behind the Cornell method is that it requires the student to interact with their notes in a way that reinforces their learning, and there’s certainly some evidence in the literature that it does work well.

• The Parallel method is simpler, it is basically a print out of the lecture slides, shrunk down so that you can fit maybe two or three per page, on one side of a sheet, with space to write notes next to each one. Part of the rationale for this method is that students don’t have to split their attention between looking at the lecture slides on the screen and writing their notes. Because the two are presented together, it increases focus and reduces cognitive load. A couple of recent meta-analyses have supported the idea that this kind of guided note-taking improves the quality of students’ notes.

• Digital note-taking is something that I’ve discussed previously. In this study it seems that the digital note-taking was mixed between students using tablets and those using laptops to take notes. There was no restriction on what note-taking software package was used (Goodnotes, OneNote etc.) or whether students used a template for their notes.

• The final note-taking approach they used was the Sentence method, where students take their notes as a series of numbered single sentences chronologically during a lecture. The main advantage of this method is expected to be that students don’t have to think about structuring their notes during the lecture, so it reduces cognitive load. This is expected to be most beneficial in a fairly fast moving lecture environment. In the study the group of students using the Sentence method was designated as the control group, but I don’t think that is really correct, because they also had to adhere to a specific methodology. So I’ll not refer to it as such here.

The students in this study used their randomly designated note-taking method over a five-week lecture course. They were tested on the course content one week after the end of the course, and then again four weeks later to see how well they had retained the information.

The researchers measured the students’ motivation twice during the study: once at the beginning of the lecture course and once at the end. They did this using a set of questions developed in a previous work designed to assess intrinsic motivation. The difference between the students’ scores was used as a measure of how much their learning motivation had increased over the course of the lectures.

I’ve summarised their results in this figure to emphasise the behaviours that the author found to have statistical significance in the data. The students using the Cornell method of note-taking got the highest marks in their tests, both immediate and retention; and the students using the Cornell method of note-taking expressed the largest increase in motivation over the course of the lectures. Students using the Parallel note-taking method were not far behind those using the Cornell method.

The students using the Sentence method got the lowest test scores and expressed the lowest increase in motivation - in fact their motivation decreased over the course of the lectures. Those using the Digital note-taking method were a little more motivated, but achieved equally low test scores.

Note that while it may appear that there was also a difference in the degree of retention between the note-taking methods, the author concluded that this was not statistically significant. However, their results did suggest that the strongest predictor of retention was motivation across all of the note-taking methods.

Structure and Self-motivation

From the perspective of self-determination theory, the Cornell method is a note-taking approach that provides more opportunity for additional personal choice (self-determination) in both the content and organisation of the notes. That additional degree of autonomy is consistent with higher intrinsic motivation, which supports the empirical results of the study. While the Digital note-taking method allows for completely unbounded autonomous input, its lack of clear structure may be associated with higher cognitive load. In practice, both the Digital and Sentence methods tend to result in students taking simple verbatim transcriptions of lectures, which are less effective for learning.

This interpretation is also broadly consistent with findings on the use of AI note-taking that I have written about previously. A high level of AI assistance (Automated AI) in note-taking reduces the level of self-determination involved in the note-taking process and results in poorer learning outcomes. Similarly, verbatim transcriptions (Minimal AI) also resulted in poorer results. The most beneficial level of AI was found to be intermediate, where students could incorporate AI summaries into their own notes, a task that arguably allowed for a higher level of autonomy.

The four types of note-taking examined in this study are in no way a comprehensive set of all possible approaches. There are a multitude of different note-taking structures in the literature: concept maps, boxing notes, charting notes, SQ3R notes, Zettelkasten …the list goes on.

However, in this study the most successful note-taking structure was the Cornell method, both in terms of learning outcomes and motivation. For anyone wanting to try out the Cornell method, I have created my own version of the template that you can download here.

More templates on Etsy

[*] Richard Ryan & Edward Desi, Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being, American Psychologist, 2020, 55, 68–78, DOI:10.1037110003-066X.55.1.68 [online article]

[**] Mesut Yıldırım, The effects of note-taking methods on lasting learning: the role of motivation and cognitive load, Frontiers in Psychology - Educational Psychology, 2026, 16:1697151, doi:10.3389/fpsyg.2025.1697151 [online article]

First Thoughts are the everyday thoughts. Everyone has those. Second Thoughts are the thoughts you think about the way you think. People who enjoy thinking have those.

Terry Pratchett, A Hat Full of Sky

When I started writing these articles, I gave them a purpose: to pull out the positive and helpful elements from the academic literature that can make undergraduate learning more effective and more enjoyable. To create a positive guide to independent learning.

But what does it mean to be an independent learner?

In this article I’m revisiting that question, and considering it in the framework of self-regulated learning. The concepts associated with this framework go a long way to encapsulating the key elements of independent learning - and self-regulated learning is widely regarded as the single most important practice that students can develop to support their personal learning goals.

Self-regulated learning, or SRL, is built around six pillars:

In practice, these pillars are enacted in a kind of feedback loop: you start by setting yourself some goals, choosing some strategies and mapping out a time-management schedule; then you do the things in that plan. After some time, you actively take stock of whether this has worked or not, and if not, why not? Adjust, and go again. In Zimmerman’s model* of self-regulated learning these steps are called the Forethought Phase, the Performance Phase and the Self-Reflection Phase.

I find that the concept of self-regulated learning resonates with Terry Pratchett’s definition of Second Thoughts. To paraphrase: self-regulated learning is the learning you do about the way you learn. People who enjoy learning do that. Third thoughts are obviously best restricted to post-graduate research.

Unsurprisingly, self-regulated learning is almost universally associated with improved learning outcomes. It’s a method of understanding how you can maximise the resources that are provided in order to personalise your learning.

For example, I’ve been writing quite a lot about getting the most out of undergraduate lectures recently. In response, several people have commented that they struggle to learn from lectures directly; they prefer problem-based learning. As someone who also depends heavily on problem-based learning, this resonates with me. So, for myself, a potentially effective task strategy in a self-regulated learning cycle would be to focus on problem sheets or worked examples, rather than primarily reviewing lecture notes. That doesn’t mean I wouldn’t attend the lectures (see my previous article on the benefits of in-person attendance), but in terms of my time management I would probably prioritise setting aside time blocks to work on problem sheets after each lecture, and deprioritise reviewing lecture recordings. My goal would be to complete each problem sheet within (say) 2-3 days of the lecture. Using my own lecture courses as an example, I would then self-evaluate this strategy based on how well I was doing in the course workshops. Help-seeking and environment structuring (I feel a Pinterest board in my future here) would be parallel considerations.

Of course, I know that I do better with problem-based learning because, well, I’ve had quite a while to work this out. But as a student transitioning from a largely passive learning environment at school into a university system where active learning becomes an independent responsibility, how do you develop your own self-regulated learning practice? What are some of the things you can do to accelerate the process?

Developing a self-regulated learning practice

It seems that it’s actually very easy to become a self-regulated learner. A simple initial intervention to promote self-reflection about different learning resources could be all you need.

This was demonstrated in a study** that took ~200 students on an introductory course at a university in the USA and split them randomly into two groups. The first group were treated as a control group and just completed the course as they would do normally. The second (SRL) group were given an exercise that prompted them to deliberately consider the format of the mid-term and final exams for the course, and to decide which learning resources would best help them study. They were asked to explicitly write down why they thought that those particular resources would be most useful, and also to make a specific, but realistic, plan for how they were going to use the selected resources to study. The same experiment was also repeated across a second cohort of another ~200 students.

The results were really startling. For both of the cohorts, and all of the exams, a significant improvement in grades was observed for the second (SRL) group.

You can see this in the following figure. The green bars are the first (control) group and the red bars are the second (SRL) group.

Obviously self-regulation is something that can be refined over time, but it seems that even a cold start from some simple directed self-reflection is enough to make a difference!

Based on the description in the paper**, I’ve created my own version of the SRL prompting exercise. You can download it here - feel free to share.

More templates on Etsy

[*] Barry Zimmerman, Becoming a Self-Regulated Learner: An Overview, THEORY INTO PRACTICE , 2002, 41, 2, https://doi.org/10.1207/s15430421tip4102_2 [online article]

[**] Patricia Chen, Omar Chavez, Desmond C. Ong and Brenda Gunderson, Strategic Resource Use for Learning: A Self-Administered Intervention That Guides Self-Reflection on Effective Resource Use Enhances Academic Performance, Psychological Science, 2017, 28, 774­, https://doi.org/10.1177/0956797617696456, [online article]

What kind of learner are you when it comes to planning your semester?

In 2014, researchers from the University of Saskatoon in Canada performed a clustering analysis* on the data from their lecture capture system to see if different patterns of engagement appeared among their students. Depending on how students accessed lecture recordings, they divided them into five groups: high activity learners, just-in-time learners, early learners, deferred learners and minimal activity learners.

The descriptions of each group are summarised in the figure:

The analysis found that the high-activity learners, i.e. those who accessed recordings regularly throughout the teaching period, did better in their assessments (mid-term and final) by roughly 10 - 15%, compared with all the other types of learning behaviour.

However, the largest grouping in the Saskatoon categorisation was the minimal activity learners. Possibly because at that point in time online resources weren’t as much of a standard offering in lecture courses, so students didn’t engage with them as much. These days, of course, lecture recordings are a key component of undergraduate teaching. So have things changed, or are there still differing patterns of engagement?

Are the different ways that students engage with lecture recordings associated with different learning outcomes?

More recently, researchers from Imperial College London performed a similar analysis** using the data from their lecture capture system. They were considering this exact question and wanted to find out how students were accessing lecture recordings and whether those patterns were correlated in any way with their final grades.

Students who get higher grades watch lecture recordings consistently during the semester, not just for revision.

The study doesn’t say exactly how many students were included in their numbers, but they do say that it includes 17 undergraduate modules from the first and second years of their Biochemistry, Biology, Chemistry, Mathematics and Physics degrees. So we can assume it’s quite a large number.

A key thing to note about the kind of approach in this study is that it’s not possible to separate correlation from causation, i.e. if a student who watched more lectures got a better grade that doesn’t necessarily mean that it was the additional lecture watching that was responsible for the better grade, it could be that the student was already a strong student for other reasons and that was why they watched more lectures. Nor do the authors try to account for this by moderating their results using student GPA or similar (perhaps they didn’t have this information). So, these are just straight-up observational results. Caveats aside, they’re still pretty interesting.

When to watch

One very clear pattern in their data is that students who get higher grades tend to watch lecture recordings more during the semester than those who get lower grades.

You can see this from the figure below:

In particular, it seems that the students who got the lowest grades tended to try and cram their lecture recordings in during the revision period and didn’t spread them out over the semester.

The Imperial College study doesn’t go as far as the Saskatoon clustering analysis in terms of grouping students into different patterns during the teaching period, but the more detailed access patterns in their data showed that when a lecture recording was accessed during the teaching period, it was typically very shortly after the lecture itself, indicating that those students who didn’t wait for the revision period were mostly exhibiting a regular pattern of engagement, consistent with the high-activity learners in the Saskatoon groupings.

How much should you watch?

The other interesting pattern that emerges from the Imperial College study** is associated with the amount of each lecture that differently achieving students watched.

Higher performing students do not re-watch full lectures

They found that the students who got the highest grades watched the least amount (in terms of time) of each individual lecture recording. Whereas students who got the lowest grades, or failed the course, watched the most.

You can see this in the following figure:

Because this was an observational study, there are multiple possible interpretations for this behaviour. The first thing we should probably rule out is an attendance issue. There is now repeated evidence in the literature that having lectures recorded doesn’t affect actual in-person attendance (see my previous article), so it’s unlikely that this group of students needed to watch more lecture content simply because they didn’t turn up to the original lecture - but of course not impossible.

A more likely explanation is that higher performing students tend to be selective about which parts of a lecture they watch. In fact, there is evidence in the literature that simply watching full lecture recordings might be detrimental to learning outcomes*** (caveat: unless you did miss the original lecture). Taking a targeted approach to watching lecture recordings is thought to be more beneficial - and is certainly more efficient.

Note that Lecture 15 in this specific figure (which shows a Statistics course) was considered to be a particularly challenging lecture; hence the peak in content volume being re-watched across all grade groups. Lecture 15 also marked a change point for the lowest grade group, who stopped watching as much lecture content after this lecture - presumably because they were so disheartened. The other grade groups obviously also found this lecture challenging, but they continued to engage consistently afterwards. It’s possible that this difference in behaviour is also related to timing, i.e. re-watching a particularly challenging lecture during the revision period might have a disproportionately negative effect on motivation compared with re-watching it during the teaching semester.

Because of the nature of these studies, we can’t make causal conclusions about watching lecture recordings, but what we can say is:

• Students who get higher grades tend to watch the lecture recording shortly after the lecture itself, and don’t wait for the revision period.

• Students who get higher grades don’t watch the full lecture recording, they likely only target the parts that fill in gaps in their understanding.

[*] Christopher Brooks, Graham Erickson, Jim Greer, Carl Gutwin, Modelling and quantifying the behaviours of students in lecture capture environments, Computers & Education, 2014, 75, 282 http://dx.doi.org/10.1016/j.compedu.2014.03.002 [online article]

[**] Moira Sarsfield and John Conway, What can we learn from learning analytics? A case study based on an analysis of student use of video recordings, Research in Learning Technology, 2018, 26, 2087 http://dx.doi.org/10.25304/rlt.v26.2087 [online article]

[***] Leonardo Martin, Caitlin Mills, Sidney K. D’Mello, and Evan F. Risko, Re-Watching Lectures as a Study Strategy and Its Effect on Mind Wandering, Experimental Psychology, 2018, 65, 297 https://doi.org/10.1027/1618-3169/a000412 [online article]

This even has a name now: “cognitive off-loading”. Helpful if you’re trying to speed up drudge work, but perhaps unhelpful if you’re trying to learn something…

That said, surely there is a happy medium where AI can be used to help accelerate or improve learning? In a recent paper*, the authors seem to have identified just this middle ground when it comes to AI-assisted note-taking.

I’m a big fan of note-taking. I’ve previously written about how it can be used to reduce mind-wandering during lectures, and about how hand-written notes are associated with improved learning outcomes when compared to typed note-taking. So if AI can help to make it even more effective then that seems like a win to me.

An intermediate amount of AI assistance for note-taking seems to support improved learning outcomes from lectures.

In the study* the authors defined three different levels of AI assistance for note-taking during a lecture:

1. Minimal AI : a direct transcript of the lecture was provided in real-time.

2. Intermediate AI : summary blocks of the lecture were provided in real-time.

3. Automated AI : structured notes were provided every ~90 seconds.

In each case, students were allowed to select, copy and paste AI-generated information into their own text editor along side their own notes. The same 30 participants rotated through different levels of AI assistance across three different ~10 minute online lectures. After each lecture the students then took a test on the lecture content, twice. The second time they took the test they were allowed to revise their original answers with the help of their notes.

The results of these tests showed that use of Intermediate AI assistance improved learning outcomes the most in both tests …and the use of Automated AI assistance produced the worst test scores.

You can see this in the figure below:

Based on a set of survey questions, designed to probe the students’ perception of the different levels of AI assistance, the authors determined that the Automated AI was perceived to be most optimal for reducing “extraneous cognitive load”, defined as the cognitive burden required by the manual effort of note-taking. And this was supported by the “click” data from the note-taking interface, which showed that 30% of students didn’t add any additional notes themselves when using the Automated AI assistance (compared with 0% and 7% in the Minimal and Intermediate AI cases, respectively).

This isn’t necessarily a black mark against the Automated AI. In fact the kind of additional cognitive burden covered by the “extraneous cognitive load” label is sometimes cited in the literature as an argument against the efficacy of note-taking in lectures**, where it is suggested that it can reduce students’ capacity for engaging with the lecture material. However, I have yet to find any empirical studies that support this hypothesis.

From the same survey questions, the authors also measured each level of AI assistance against “germane cognitive load”, which is the beneficial cognitive effort associated with learning, understanding and internalising knowledge. In this case they found that both Automated AI and Intermediate AI were perceived to support this kind of cognitive load. Interestingly, students thought that reading the notes produced by the Automated AI was most beneficial for their understanding of the lecture material; however, this perception was not supported by the test data - those students did worse, even when they were permitted to use their notes!

Use AI to make summaries, not notes.

So (in summary) a little bit of AI goes a long way, and making your own notes - perhaps with a little bit of AI help - is still the best way to improve learning outcomes.

[*] Xinyue Chen, Kunlin Ruan, Kexin Phyllis Ju, Nathan Yap and Xu Wang, More AI Assistance Reduces Cognitive Engagement: Examining the AI Assistance Dilemma in AI-Supported Note-Taking, Proceedings of the ACM on Human-Computer Interaction, 9, 7 (2025), https://doi.org/10.1145/3757632 [online article]

[**] Annie Piolat, Thierry Olive and Ronald T. Kellogg, Cognitive Effort during Note Taking, Applied Cognitive Psychology, 19, 291–312 (2005), https://doi.org/10.1002/acp.1086 [online article]

Before we get into it, let’s firstly take a moment to ask: what is the evidence for that estimate of (oh, let’s just say) ~15 minutes?

In 2016, Neil Bradbury from the University of Chicago asked exactly this question*. He discovered that the magic number of ~15 minutes was repeated again and again in teaching guides, but that there seemed to be no originating reference for it in the literature. In fact, the one paper that was occasionally cited barely mentioned student attention at all! Furthermore, the few reliable empirical studies on student attention that he could find didn’t seem to support the figure of 15 minutes either.

The conclusion? Congratulations, your attention span is probably longer than you think! Although, equally, Bradbury’s review did suggest that the primary cause of attention deficit in lectures is …boring lecturers. So perhaps don’t celebrate too soon.

There is no empirical evidence that students are only able to concentrate for 15 minutes at a time in a lecture.

So the 15 minute attention span is a myth. While it’s helpful to know this, it doesn’t immediately mean that everyone can focus perfectly for 45 - 60 minutes. What the primary studies that Bradbury did find back in 2016 showed was that students’ attention went up and down over the course of a lecture, as you might expect.

This expectation has of course motivated a range of measures to help students maintain focused attention in lectures; however, these are largely structural and include things like short breaks, formative quizzes, etc. And, as a student, these kinds of intervention are basically out of your hands. So what’s the one most important thing that you can do to help maintain your focus? It seems to be note-taking.

Note-taking is the most widely evidenced measure for maintaining focus during lectures.

Using note-taking to maintain focus

First things first: taking notes and paying attention are not necessarily synonymous; however, there does seem to be consistent evidence in the literature that taking notes during a lecture (whether recorded or in-person) helps to prevent your mind wandering.

Multiple recent studies have compared the degree of “mind-wandering” during lectures between students who were taking notes and those who were just watching the lecture. In one recent study**, students were asked to rate the degree to which their mind was wandering at different times during a lecture. The students who were taking notes reported a systematically lower degree of mind-wandering at all points compared to those who were simply listening to the lecture - even though both groups reported feeling an equivalent amount of boredom at each point during the lecture.

The experiment ranked the degree of mind-wandering on a (self-reported) 7 point scale, with a score of 7 indicating that the student’s mind was “not on task at all”, and a score of 1 indicating that the student’s mind was “completely on task”. Note: they have these limits reversed in the caption for their original figure, but I think that must be a typo given their subsequent discussion of the results.

This is how the results looked:

While it’s clear that students’ minds wandered progressively more and more as the lecture went on, the effect on those taking notes seems to have been constrained below the initial level of mind-wandering experienced by listen-only students for the entire duration of the lecture. The same study** also found that the students who took notes did better on a subsequent multiple choice test about the lecture material. I think it’s worth pointing out that this was a controlled experiment, where the students were allocated to their group (listening or note-taking), and there was therefore no individual choice made to take notes or not that might have biased the outcomes.

A similar pattern was also seen in another recent study*** that looked at the effect of mind-wandering across multiple lectures. The same effect was seen: students progressively experienced more mind-wandering both within individual lectures and as a function of lecture order, i.e. students’ minds wandered more during the third lecture than the second, and more during the second lecture than the first. However, students who were taking notes experienced lower overall mind-wandering than those who were not. Importantly, this study*** mixed up the order in which individual students watched the three lectures - avoiding the “boring lecturer bias” identified by Neil Bradbury. Once again, students’ mind-wandering rates were also found to be negatively correlated with their performance in a subsequent test, i.e. lower mind-wandering was associated with better test scores.

In conclusion: those notes you keep taking in your lectures? They’re probably helpful in more ways than you thought.

[ If you found this interesting, you might also want to take a look at my previous article about the relative impact of taking handwritten notes vs typed notes on learning outcomes. ]

[*] Neil Bradbury, Attention span during lectures: 8 seconds, 10 minutes, or more?, Advances in Physiological Education, 40: 509–513, (2016), doi:10.1152/advan.00109.2016. [online article]

[**] Emily Krysten Spencer-Mueller and Mark J Fenske, Note-taking for the win: Doodling does not reduce boredom or mind-wandering, nor enhance attention or retention of lecture material, Quarterly Journal of Experimental Psychology, 77, 1780­ –1796, (2024), doi:10.1177/1747021823122 [online article]

[***] Sarah Shi Hui Wong and Stephen Wee Hun Lim, Take Notes, Not Photos: Mind-Wandering Mediates the Impact of Note-Taking Strategies on Video-Recorded Lecture Learning Performance, Journal of Experimental Psychology: Applied, 29, 124–135, (2023), doi:10.1037/xap0000375 [online article]

One thing that has evolved is the way in which students take notes. There was a point a few years ago (probably pre-pandemic) when I frequently found myself facing a wall of laptop lids during my lectures. So I’m quite pleased that typed notes seem to have fallen out of favour somewhat at this point, with a majority of students preferring to take hand-written notes instead - although often on tablets rather than directly on paper.

But does it make a difference how you take your lecture notes? Hand-written or typed? The answer seems to be: yes. Although the literature is divided on the significance of the difference.

Handwritten notes are associated with improved learning outcomes compared to typed note-taking.

In a recent paper* the authors made a combined (meta) analysis of 24 previous studies that compared handwritten note-taking to typed note-taking, and the relative effect on learning outcomes. They found that in the majority of cases (88%), handwritten note-taking had a positive effect on achievement relative to typing notes.

The authors calculated that the mean size of that positive effect would shift the distribution of grades for a course where the average grade was originally 75% with a standard deviation of 10% (blue curve) to one where the average grade was approximately 2 - 3% higher (pink curve). See their figure here:

There were also some other interesting conclusions:

• It doesn’t matter whether the assessment is conducted immediately after the lecture or at a later time. The positive effect of hand-written notes is consistent.

• It doesn’t matter what kind of assessment it is (factual, conceptual, or both). The positive effect of hand-written notes is consistent.

• Reviewing notes before assessment has a big effect and was a key factor influencing the positive advantage of hand-written notes over typed note-taking, in spite of both mediums being reviewed.

But why does hand-writing your notes lead to improved learning outcomes?

In another recent work** that looked at experimental results from 30 different neuroimaging studies, the authors tried to answer exactly this question. They found that, while there was overlap in which parts of the brain were used during both hand-writing and typing, there were also key differences. In particular, handwriting primarily activated the motor complexes (responsible for parsing and executing motor commands) and the visuo-spatial complexes (for letter shapes and orthographic memory), while typing activated the areas of the brain associated with linguistic processing and working memory. Overall, typing activated fewer areas of the brain associated with language, but more that were associated with procedural memory, i.e. remembering the positions of keys on the keyboard.

They summarised their findings in this lovely figure:

The authors suggested that handwriting connected motor activities more closely with cognitive processes than typing did, which led to enhanced activity in areas of the brain associated with forming memories. This connection was enhanced when the writing was done on paper - rather than on a tablet - suggesting that the sensory feedback from the paper was also a key factor.

So sometimes the old methods are the best. Just make sure to keep your notes for review!

[*] Abraham E. Flanigan, Kenneth A. Kiewra, Jordan Wheeler, Tiphaine Colliot, & Junrong Lu, Typed Versus Handwritten Lecture Notes and College Student Achievement: A Meta‑Analysis, Educational Psychology Review, 36, 78 (2024), https://doi.org/10.1007/s10648-024-09914-w, [online article]

[**] Marano, G.; Kotzalidis, G.D.; Lisci, F.M.; Anesini, M.B.; Rossi, S.; Barbonetti, S.; Cangini, A.; Ronsisvalle, A.; Artuso, L.; Falsini, C.; et al. The Neuroscience Behind Writing: Handwriting vs. Typing—Who Wins the Battle? Life, 15, 345 (2025), https://doi.org/10.3390/life15030345, [online article]

I think this is a great idea. In fact, I’m planning to do this myself next semester and provide the summary alongside the prose notes.

But it’s also possible to go one step further and create an audio-only version of the summary. This might seem like overkill, but there’s some emerging evidence that it might be a useful tool for supporting neurodiverse students.

Results suggest that AI-assisted audio-learning may mitigate the negative effects of ADHD symptoms on learning outcomes.

In a recent study*, the authors looked at the effect of audio-only summaries (7 - 15 minutes) of course reading material. The scripts for the audio summaries were AI generated from the reading material itself. For those who might be interested in replicating the process, there’s a very nice description in the paper of how the AI prompt was constructed.

A key finding in the study* was that students who listened to the audio summaries spent more time actually reading the course material than the students who didn’t listen to the summaries. Their course grades were also higher.

Another interesting result was that the students who spent most time listening to the audio summaries were able to mitigate the symptoms of ADHD, a common form of neurodiversity. Note that the amount of time students spent listening to the recording seems to be important: of course not everyone listened to everything, and the effect was only seen for the students who engaged for longer than the median listening time (right hand plot; green line).

You can see this in their figure here:

I actually find this division a bit strange. Why would the mitigation depend on the median listening time? That makes the result cohort dependent. I would have preferred to see this analysis done in terms of the percentage of total audio duration listened to, which would be more generally applicable. From the numbers in the paper, the median listening time seems to be ~60% of the total audio.

I also find the result a bit weak. The authors report that even for the students who listened for an above median time (right hand plot in the figure), there is no statistically significant overall difference between the control group (those who had no audio) and the audio-assisted group, which I interpret as the data being drawn from the same distribution, but that there is a statistically significant difference between the two groups as a function of ADHD symptom. Given the small number of data samples in the figure at high and low symptom severity, I don’t find this result very convincing.

The degree of ADHD symptom severity in the above figure was based on a self-reporting scale (ASRS-18), rather than formal diagnoses. This was done because there were only a small number of students (23) in the study who had formal diagnoses of ADHD. Because of the small number, the authors report that their results using data from the formally diagnosed students alone aren’t statistically significant. However, they still include the analysis in their supplementary material:

Even though they’re not statistically significant, these supplementary results do also suggest that audio-assisted learning is beneficial for mitigating the effect of ADHD symptoms on learning outcomes.

So in summary, I think that the evidence presented in this study* isn’t strong. But it is probably suggestive enough for me to go ahead and create audio versions of my lecture summaries next semester.

ADHD is a complex form of neurodiversity that seems to manifest differently across individuals; however, it’s very encouraging that new AI tools can be used to experiment on an individual basis to try and find personal mitigation strategies.

[*] Nanda R. Jafarian & Anne-Wil Kramer, AI-assisted audio-learning improves academic achievement through motivation and reading engagement, Computers and Education: Artificial Intelligence, 8, 100357 (2025). https://doi.org/10.1016/j.caeai.2024.100357 [online article]

Lecture-based learning comes from more than content - it’s also about community.

9am on a Monday morning and 5pm on a Friday afternoon. These are the two lecture slots that lecturers never want to see on their teaching timetable. They are almost always a guarantee of poor attendance - and there’s nothing more depressing than giving a lecture to an empty room.

We get it. We don’t love those times either. We understand that they’re difficult. We know how tempting it must be to say “Well, I can always watch the podcast…”

And that’s fine. Maybe it’s just too much that day, or perhaps other parts of your life simply got in the way. That’s one of the beauties of lecture capture - the flexibility it can allow when things get complicated.

More generally though, going to lectures in person has been shown to be correlated with learning outcomes - even more so than watching lecture podcasts. In a large modern study* it was found that lecture attendance was positively correlated with course grade. Watching lecture podcasts was also positively correlated, but not as strongly. Interestingly, the correlations were most pronounced for students in the first and second years of their degree.

I’ve summarised these results here:

Furthermore, in-person lecture attendance gave the biggest grade boost to the students who were struggling most in terms of their average performance across other courses, rather than those who were doing consistently well.

Before lecture capture existed, that same association between attending lectures in person and course grades was found to be even more pronounced**. But these days, if a modern lecture podcast covers the same course content, then why should attendance still be the strongest factor?

Ok, I know what you’re thinking: surely it just means that attending the lecture and watching the podcast are also correlated. But no! The same modern study* showed that lecture attendance and recording use were basically uncorrelated. Which should also offer some reassurance to lecturers who worry that students will deliberately not come to lectures because they can watch the podcast.

So why is lecture attendance so beneficial? It can’t simply be a question of content, because the podcast has identical content. Some researchers have argued that consistently attending lectures requires students to be more effective at self-regulation, i.e. intentionally controlling their day to day actions in order to achieve their learning goals, and it’s this improved self-regulation that ultimately leads to better learning outcomes.

The ultimate test of self-regulation? Consistently turning up for a 9am Monday lecture 😂

Personally, I prefer the alternative theory that lectures are a social event and that the sense of community they create among students is actively beneficial for learning. Rather than simply being a happy bonus, this sense of community has been suggested as a key underlying factor that causes in-person attendance at lectures to result in tangibly improved learning outcomes. I couldn’t find any studies that quantitatively analysed the community hypothesis directly; but the idea that in this digital world, where we are increasingly given the tools to do things by ourselves, it makes me happy to think that lectures still bring us together. It is also a reminder that independent learning does not necessarily mean learning alone.

Caveat: the study* that I’ve cited here was conducted pre-pandemic. It’s of course possible that the experience of doing so many things exclusively online during the pandemic has altered students’ ability to learn effectively from lecture recordings. If anyone can point me at a more recent study that examines the same variables it would be really interesting to compare those results.

[*] Emily Nordmann, Colin Calder, Paul Bishop, Amy Irwin & Darren Comber, Turn up, tune in, don’t drop out: the relationship between lecture attendance, use of lecture recordings, and achievement at different levels of study, Higher Education 77, 1065–1084 (2019), https://doi.org/10.1007/s10734-018-0320-8 [online article]

[**] Credé, M., Roch, S. G., & Kieszczynka, U. M., Class attendance in college: a meta-analytic review of the relationship of class attendance with grades and student characteristics. Review of Educational Research 80(2), 272–295 (2010), https://doi.org/10.3102/0034654310362998 [online article]

Those low-stakes un-assessed mini-quizzes that lecturers interleave with their online course material? They’re not only there for fun. Mini-quizzes have repeatedly been shown to improve student performance - and they just might make you happier too.

Quiz like no-one’s watching (because they aren’t)

It probably comes as no surprise that having short formative quizzes after each section of material in a lecture course is expected to be beneficial in terms of learning outcomes. No lecturer is going to make extra work for themselves without a good reason. However, even though the impact of quizzing is well recognised, people are still really divided on why it works.

It could be that those little quizzes help students to focus their attention better on the lecture material, or it could be that breaking up lecture material with quizzes forces context changes in the brain that improve recall, or it could be that regularly answering quiz questions helps students to formulate improved strategies for encoding and retrieving information. The jury is still out.

In a recent study* using a large cohort of students, this positive effect was clearly demonstrated once again by comparing the difference in test scores between students who did a short formative quiz after each segment of an online lecture, and those who were shown review slides instead. The students who did the quizzes performed almost 10% better in subsequent assessed tests on the same lecture material.

You can see this from the figure in their paper (control group):

The study also looked at one of the big downsides of online learning: online distractions.

There were two interesting outcomes from their research: (i) moderate/medium levels of distraction (provided by funny memes) didn’t really affect student performance, regardless of whether the students did quizzes or not, and (ii) high levels of distraction (TikTok videos) were detrimental for the students who did the quizzes, but beneficial for the students who didn’t - which was totally unexpected.

So with respect to the question of whether to quiz or not, the answer is clear: doing those mini-quizzes is likely to help you get a better mark in your course. As long as you’re not watching TikTok at the same time.

Even better still - it could make you feel happier while doing it: another recent study** has suggested that doing regular low-stakes quizzes reduces student anxiety. This reduced anxiety was also found to be positively correlated with improved academic performance, but it’s not clear whether students were less anxious because they were doing better, or doing better because they were less anxious. Either way, it seems like a good thing.

[*] Jason C. K. Chan, Dahwi Ahn, Karl K. Szpunar, Zohara Assadipour & Hymnjyot Gill, In-lecture quizzes improve online learning for university and community college students, Nature Communications Psychology 3, 54 (2025). https://doi.org/10.1038/s44271-025-00234-5 [ online article ]

[**] Joshua Nosa Edokpayi, Assessing the impact of frequent quizzes on student performance: reflections from a rural-based institution, Discover Education 4, 480 (2025). https://doi.org/10.1007/s44217-025-00930-8 [ online article ]

In a recent paper*, the authors synthesised the results from 24 different studies on the effect of increasing the playback speed of recorded lectures on assessment performance. They found that watching lectures at up to 1.5 x original speed had a negligible effect on assessment performance, but beyond that point, i.e. if you speed up the playback by more than 1.5, performance started to drop off.

You can see this from the figure in their paper:

Personally, I wouldn’t have bothered to fit a trend line to these data. I think what you’re looking for is simply how much the distribution of effect deviates systematically from zero. At a playback speed of 1.5 x default, the distribution is still pretty much centred on zero (there’s a slight mean offset of -0.09) and it appears quite symmetric. For a playback speed of 2.0 x default, the distribution is clearly displaced from zero (mean offset: -0.36), and moreover there’s a long (if sparsely sampled) tail towards much poorer performances (see that one data point below -2.0 on the y-axis?).

The paper helpfully contextualises these performance effects: if the typical assessment outcomes have a mean mark of 75% and a mark spread of +/- 20%, then an offset of -0.09 would translate to a reduction in marks of 2%, and an offset of -0.36 would translate to a reduction of 7%.

For a 45 minute lecture, using playback at 1.5 x original speed will save you 15 minutes. What will you do with that 15 minutes?

[*] Tharumalingam, T., Roberts, B.R.T., Fawcett, J.M. et al. Increasing Video Lecture Playback Speed Can Impair Test Performance – a Meta-Analysis. Educ Psychol Rev 37, 35 (2025). https://doi.org/10.1007/s10648-025-10003-9 [ online article ]

Subscribe now

Share