24 of 40 at 1x, 21 at 2x a week later: how fast to watch a lecture
I watch recorded conference talks at 1.75x. Most of the people I train do the same with internal course videos, and the speeds they reach for most often are 1.25x and 1.5x 1. Four years of experiments and three meta-analyses now cover that habit. Comprehension survives moderate speeds better than I expected. The ability to use the material afterwards does not hold up as well.
What the research shows.
The best-known experiment comes from Dillon Murphy's group at UCLA. They had 231 undergraduates watch two lectures of 13 to 15 minutes, one on the Roman Empire and one on real estate appraisals, at 1x, 1.5x, 2x or 2.5x, then tested them on 40 questions immediately and again a week later 2.
Immediate scores: 26 of 40 at 1x, about 25 at both 1.5x and 2x, about 22 at 2.5x. A week later every group had lost ground. 24 at 1x, 21 at 1.5x, 21 at 2x, 20 at 2.5x. The authors described the cost up to 2x as minimal and placed the clear decline past that point.
That single study drove most of the "2x is free" coverage. The pooled evidence is less generous.
Cheng, Pastore and Ritzhaupt combined 18 effect-size comparisons from 7 experiments on time-compressed multimedia. At 1.4 to 1.5x, learning dropped by g = 0.21, with a confidence interval running from 0.05 to 0.38 that excluded zero. At 1.8 to 2.0x the estimate grew to g = 0.36, though that interval was wide enough to contain zero 3.
Tharumalingam and colleagues ran the bigger analysis in 2025: 110 effect sizes from 24 studies of lecture video, aggregated with robust variance estimation. The same shape came out. Faster playback lowers test scores, the cost stays small and often statistically unreliable at 1.5x and below, and it reaches statistical significance as speed approaches 2x. None of the moderators they tested, test type and lecture duration included, changed that picture 4.
Transfer degrades more than recall.
Huang, Du and Yang pooled 12 experiments on accelerated instructional video and separated the outcomes. Acceleration raised cognitive load (g = 0.59), lowered retention test scores (g = 0.41), lowered transfer test scores (g = 0.50) and lowered learner satisfaction (g = 0.24) 5.
The ordering is the useful part. A retention test asks you to reproduce what the speaker said. A transfer test asks you to apply it to a problem you have not seen. Speed costs more on the second, which fits the cognitive load figure: compressed speech leaves less spare capacity for the work of tying a new idea to something you already hold. For a course you have to pass a quiz on, 1.5x is close to free. For a tool you have to operate on Monday, speed costs you the part you watched the video for.
The time you save only counts if you spend it.
Murphy's group also tested rewatching. Two viewings at 2x, back to back, gave no advantage over one viewing at 1x. Rewatching at 2x immediately before the test did beat a single 1x viewing a week earlier 2. The gain came from the second pass landing close to the test, which makes the timing of a rewatch worth more than its speed.
A 45-minute talk at 1.5x gives you back 15 minutes. Those 15 minutes are worth more than the 3 points of delayed recall you traded for them, as long as they go into a retrieval pass or a second viewing. Watch three talks instead and you have converted the saving into more material you will forget.
The protocol.
1. Cap new material at 1.5x. The pooled cost there sits near 0.2 SD and often cannot be distinguished from zero 34. Set it as the default in your player and stop fiddling.
2. Drop to 1x for the part you will have to operate. When the speaker reaches the configuration step, the formula, the exception, the thing you will repeat yourself, put it back to 1x until that section ends. Transfer is the outcome that degrades most (g = 0.50) 5, and it is the only one that matters for a step you have to execute.
3. Never pass 2x. Against the 1x group, 2.5x cost 4 of 40 questions immediately and 4 of 40 again a week later 2. The two 2025 meta-analyses put the reliable costs in that same region 45.
4. Put the banked minutes into a blank page. After a 45-minute talk at 1.5x, spend 10 of your 15 saved minutes writing the 5 claims and the one number you will use, from memory, before you open anything else.
5. Schedule the rewatch next to the moment of use. Rewatch at 2x the day before you have to apply the material rather than the day after you first saw it 2.
6. Watch at 1x once for anything you will teach. In my own trainings I have to answer questions the speaker never anticipated, and that is a transfer task. The 20 minutes I would save are the cheapest thing I can give up.
My rule since reading this: 1.5x for context, 1x for anything I will have to run myself, and the saved quarter of an hour goes into a page I write from memory before the next video opens.
Sources.
- Tharumalingam, T., Risko, E.F. (2025). Altering the playback speed of recorded lectures as a learning technique: Examining student practices, motivations, and beliefs. Journal of Research on Technology in Education. doi:10.1080/15391523.2025.2459130
- Murphy, D.H., Hoover, K.M., Agadzhanyan, K., Kuehn, J.C., Castel, A.D. (2022). Learning in double time: The effect of lecture video speed on immediate and delayed comprehension. Applied Cognitive Psychology, 36(1), 69-82. doi:10.1002/acp.3899
- Cheng, L., Pastore, R., Ritzhaupt, A.D. (2022). Examining the Accelerated Playback Hypothesis of Time-Compression in Multimedia Learning Environments: A Meta-Analysis Study. Journal of Educational Computing Research, 60(3), 579-598. doi:10.1177/07356331211043535
- Tharumalingam, T., Roberts, B.R.T., Fawcett, J.M., Risko, E.F. (2025). Increasing Video Lecture Playback Speed Can Impair Test Performance: A Meta-Analysis. Educational Psychology Review, 37(35). doi:10.1007/s10648-025-10003-9
- Huang, Du, Yang (2025). Facilitating or hindering learning: a meta-analysis of acceleration on video learning. Frontiers in Psychology, 16, 1427609. doi:10.3389/fpsyg.2025.1427609