Watching Lectures at 2x Speed: What It Costs You

Almost everybody does it. In a survey of 123 undergraduates run alongside one of the main experiments in this area, 85% reported watching lecture videos at quicker than normal speeds.

And almost everybody is uneasy about it. In the same survey, 91% said they believed watching at normal or 1.5x speed was better for learning than 2x or 2.5x. So students are doing something they suspect is bad for them and doing it anyway, which is usually a sign that the honest answer is more interesting than the folk wisdom on either side.

It is. The evidence does not say speeding up lectures is fine. It does not say it wrecks your learning. It gives you a fairly precise dose-response curve, and the curve has a bend in it.

The number that matters

Tharumalingam, Roberts, Fawcett and Risko (2025) meta-analysed 110 effect sizes from 24 studies of learning from lecture videos, comparing accelerated playback against a 1.0x control on identical content tests. The pooled effect was g = -0.28, 95% CI [-0.40, -0.16], p < 0.001 - a real cost, small on average.

But the average is the wrong thing to look at, because speed was a significant continuous moderator (Q_M(1) = 56.82, p < 0.001; b = -0.62, p < 0.001). Broken out by speed:

Their summary: "for speeds at 1.5x and below this effect is very small (and statistically non-significant)... As lecture speed approaches 2x and beyond, the costs are statistically significant."

At 1.5x the cost is not distinguishable from zero. At 2x it is real but modest. At 2.5x it is large - a -0.86 effect is not a rounding error, it is most of a standard deviation.

If you want a single practical line from this article: 1.5x is close to free, 2x costs you something, 2.5x costs you a lot.

The study everyone cites, and where it disagrees

Murphy, Hoover, Agadzhanyan, Kuehn and Castel (2022) is the paper that put this question on the map, and its findings are more optimistic than the meta-analysis.

Experiment 1. 231 undergraduates watched two lectures - one on real estate appraisals, one on the history of the Roman Empire - at 1x, 1.5x, 2x or 2.5x, then took 20-question comprehension tests immediately and again a week later. No pausing, no note-taking.

There was a main effect of speed (F(3, 202) = 3.98, p = .009, eta-squared = .06), but when the pairwise comparisons were run, only 1x versus 2.5x survived (p_bonf = .004, d = .24). Every other pair came out at p_bonf > .385, d < .13.

So Murphy found no significant cost at 2x. The meta-analysis, pooling 49 effect sizes at that speed, found one. Both are right about what they measured; the meta-analysis has far more data at 2x and is the better guide. Murphy's own conclusion - "minimal costs incurred by increasing video speed from 1x to 1.5x, or 2x speed, but performance declined beyond 2x speed" - is the more generous reading of a single well-run study, and it is worth knowing that the cumulative evidence has since landed slightly harsher.

Immediate versus durable learning: the answer is not what you would guess

The intuitive worry is that speeding up produces knowledge that evaporates - fine on the quiz, gone by the exam.

Murphy's Experiment 1 tested exactly this and found the opposite of what the worry predicts. There was a large main effect of delay (F(1, 202) = 98.96, p < .001, eta-squared = .33) - everyone forgot a lot in a week, dropping from .60 to .54 on average. But speed and delay did not interact (F(3, 202) = .78, p = .504, eta-squared = .01). The forgetting curves ran parallel.

In that dataset, watching faster did not selectively damage durable retention. It cost you a bit at the start and you kept your reduced amount about as well as anyone else kept theirs.

This is a single experiment, and it is the only direct test of the question we could find. Every other study in this literature that we verified - the older-adults work, the 2026 Frontiers study, the 2026 working-memory study - tested immediate performance only. Treat the finding as encouraging and unreplicated, not as settled.

The trade you are actually making

Murphy's second experiment asked the question students actually face: given a fixed amount of time, is it better to watch once at normal speed or twice at double speed?

Experiment 2A - both viewings back to back, tested immediately. 106 undergraduates. Comprehension: .64 for once at 1x versus .63 for twice at 2x (t(104) = .16, p = .876, d = .03). No difference at all. Interestingly, the twice-at-2x group expected to do worse (.51 versus .61, t(104) = 2.50, p = .014, d = .49) and did not.

Experiment 2B - the second viewing moved to just before the delayed test. 110 undergraduates. Here the 2x group watched at the start, waited a week, then rewatched at 2x immediately before testing. The 1x group watched once and waited a week with no rewatch. Result: .63 versus .55 (t(108) = 2.85, p = .005, d = .54), favouring twice at 2x.

Read those two together, because the popular retelling collapses them. Watching twice at 2x beat watching once at 1x only when the second viewing was spaced a week later and placed just before the test. Cramming both viewings together produced no advantage over a single normal-speed viewing.

The finding here is not really about speed at all. It is about spacing. Speed bought the time; spacing is what turned the time into a benefit.

Who this does not apply to

Murphy, Hoover and Castel (2023) ran three experiments with 1,020 participants across younger adults (18-25) and older adults (mostly 64-94). The pattern was consistent and one-sided.

In the combined analysis, speed interacted with age (t(1016) = 3.86, p < .001, beta = .24): "younger adults' performance was similar regardless of speed (t(1016) = -.45, p = .656, beta = -.02) while older adults' performance decreased at faster speeds (t(1016) = -6.13, p < .001, beta = -.26)."

Their conclusion is blunt: "while younger adults can watch videos at faster speeds without significant consequences, we advise against older adults watching at faster speeds."

There is a second group the evidence flags. Chung, Nam and Hong (2026), using eye-tracking with 38 participants split by working-memory capacity, found recall significantly impaired for the lower working-memory group at every speed from 1.25x upward (z = -2.15 at 1.25x, rising to z = -3.37 at 2x), while the higher-capacity group showed no significant effect. That is 19 people per group and immediate recall only, so treat it as a signal rather than a finding - but it points the same way as the age result. The averages in this literature are hiding people for whom the cost arrives much earlier.

Does it feel worse? Yes. Does that track the cost? No.

Rodriguez Orejuela, Arango Espinal and Osorio-Andrade (2026) ran 328 participants through a 16-minute lesson at 1.0x, 1.5x or 2.0x and measured both comprehension and perceived clarity.

Comprehension showed no significant difference at either speed (2.0x versus 1.0x: beta = -0.019, t = 0.277, p = 0.782). Perceived clarity fell sharply at 2.0x (beta = -0.311, t = 6.164, p < 0.001). Their phrasing: playback speed "operates as an asymmetric exchange that preserves retention but increases the perceived cost of processing."

This is a useful corrective in an unusual direction. The standard worry about study shortcuts is that they produce an illusion of learning - it feels easy, you feel fluent, you learn less. Here the dissociation runs the other way: 2x felt substantially worse while scoring the same. Murphy found something similar, with students consistently under-predicting their performance in the fast conditions.

So your discomfort at 2x is real, and it is not a reliable readout of what you retained. That cuts both ways: it means you should not stop at 2x just because it feels hard, and it means you should not assume you are fine at 2.5x just because you have got used to the sound.

Two honest caveats on this study: the comprehension result is a null (R-squared = 0.022), not a demonstration of equivalence, and it tested immediate performance only.

What none of this is

It is not study advice. Nothing here says speeding up lectures is a good learning strategy, and the meta-analysis is titled Increasing Video Lecture Playback Speed Can Impair Test Performance for a reason.

What the evidence supports is narrower and more useful: the cost is a function of speed, it is near zero up to 1.5x, it is real and modest at 2x, and it becomes large past that. Time saved is only a gain if you spend it on something - and the one condition in this literature where speeding up clearly beat normal speed was the one where the saved time bought a spaced second viewing rather than an early finish.

Everything in this article was measured on recorded lectures with a single speaker: 12- to 16-minute videos on appraisals, Roman history and digital marketing, tested with multiple-choice and recall questions, mostly on undergraduates. It does not tell you about a two-hour seminar, a discussion, a language class, or a lecture in your second language. The meta-analysis found no moderation by lecture duration, subject or test format, which is mildly reassuring, but it was drawing on the same narrow band of materials.

If the real problem is that hours of study are not converting into anything, speed is not the variable to adjust - see studying for hours and remembering nothing and laptops in the classroom.

A note on why this article has no keyword

The obvious search term for this subject is playback speed. We are not targeting it. Its search results are dominated by browser extensions and software vendors - the top result is a speed-controller extension in the Chrome Web Store - which means the people searching it want a tool, not an answer. The query that matches this article, watching lectures at 2x speed, returns a US search volume of 20.

We would rather publish the article that answers the question than target the keyword that does not.