Speed Reading: What the Evidence Says About the Trade-Off

Speed reading is one of the few areas where psychology has something close to a settled position, and where that position is almost entirely absent from the results you get when you search for it.

The position is this: there is a trade-off between speed and comprehension, and no known technique removes it. What is genuinely disputed is where the trade-off starts to bite - and there is one large study suggesting it starts later than the strongest version of the claim implies. Both halves are below.

The baseline

Skilled adult readers of English who have been through higher education typically read at 200 to 400 words per minute. That range comes from the major review of this literature - Rayner, Schotter, Masson, Potter and Treiman (2016), published in Psychological Science in the Public Interest, which is the standing scientific reference on the question.

The claims made by speed-reading products usually start at around 600 wpm and go up from there, sometimes into the thousands. The gap between 400 and 1,000 is where the argument lives.

Why the eye cannot simply go faster

Reading is not a smooth sweep across a line. Your eyes make short jumps - saccades - and pause between them on individual words, taking in the text during the pauses. Two features of that process turn out to matter enormously.

Preview. While fixating one word, you extract partial information about the next one from the edge of vision. This is why reading is faster than looking at words one at a time.

Regressions. Skilled readers move backwards constantly - roughly ten to fifteen per cent of eye movements - usually without noticing. Rereading a phrase that did not parse is a core part of comprehension, not a failure of it.

Most speed-reading techniques work by attacking one or both. Instructions to stop subvocalising, to use a finger as a pacer, or to take in whole lines at once are all attempts to reduce fixations. And RSVP - rapid serial visual presentation, the technology behind apps that flash one word at a time in a fixed position - removes both by construction. There is no parafovea to preview from, and no way to go back.

The regression experiment

The cleanest test of what removing regressions costs was run by Schotter, Tran and Rayner (2014), published in Psychological Science.

Forty undergraduates read sentences while an eye tracker followed them. In the critical condition, a trailing mask blanked each word once the eyes had passed it - so the text was still read normally, at the reader's own pace, but going back was impossible. This isolates exactly what RSVP takes away.

Comprehension fell. Readers were significantly less accurate in the trailing-mask condition than in normal reading: b = -0.92, z = 5.09, p < .001.

The important part is what the effect was not confined to. The obvious prediction is that rereading only matters for tricky garden-path sentences. But the interaction with sentence ambiguity did not reach significance (b = 0.70, z = 1.90, p = .06), and the cost persisted even when comparing trials on which readers had not, in fact, made a regression (b = -0.75, z = 4.03, p < .001).

Rayner and colleagues' review reports the resulting comprehension figures starkly: around 75% accuracy when rereading was allowed, against roughly 50% - chance performance - when it was prevented.

Being able to go back appears to matter even when you do not use it.

The RSVP app studies, which disagree with each other

This is where the honest picture gets more interesting than the simple one.

Benedetto and colleagues (2015) tested Spritz directly. Sixty participants, half reading with the app at a nominal 250 wpm, half reading traditionally at their own pace, with eye tracking.

Comprehension type	Spritz	Traditional	Statistic
Literal	60% (SD 19)	72% (SD 14)	F(1, 55) = 7.30, p = .009, ηp² = .12
Inferential	44% (SD 17)	48% (SD 18)	p > .250 (not significant)

A twelve-point deficit on stated facts, and no reliable difference on inference. The authors attributed the cost to suppressed preview and the absence of regressions.

Di Nocera, Ricciardi and Juola (2018) ran a larger study - 210 students - comparing a self-paced control at about 226 wpm against RSVP at 250, 300, 350, 400 and 450 wpm, tested with 49 inferential true/false items.

They found no significant difference at 250, 300 or 350 wpm. Degradation appeared at 400 and 450. The overall effect was F(5, 204) = 7.65, p < .001, ηp² = .15, with the 300-versus-400 comparison at F(1, 68) = 6.37, p = .014.

These two results are in genuine tension, and it would be dishonest to present only the first. Note that they measured different things - Benedetto found the deficit in literal comprehension while Di Nocera tested inferential - so neither is a replication of the other, and neither settles it.

The defensible reading is narrower than "RSVP destroys comprehension." It is that costs appear at higher speeds and on detail retention, and that at rates close to ordinary reading, the format may be roughly neutral. Which also means that at rates close to ordinary reading, it is not speed reading.

A fourth study - Acklin and Papesh (2017), titled "Modern speed-reading apps do not foster reading comprehension" - reaches the same directional conclusion. We could not obtain its full text and so are printing no figures from it.

What the review concludes

Rayner and colleagues' assessment of RSVP specifically, in their own words:

"Although immediate comprehension may be successful with single sentences presented using RSVP speeds well beyond typical reading rates, scaling up to full text passages yields substantial comprehension costs."

And on the general question, as the review's authors put it in the accompanying release: the evidence "demonstrates that there is a trade-off between speed and accuracy - as readers spend less time on the material, they necessarily will have a poorer understanding of it."

What actually makes people read faster

The review does not end at "you cannot." It identifies what does work, and it is not an eye technique:

"The way to maintain high comprehension and get through text faster is to practice reading and to become a more skilled language user (e.g., through increased vocabulary). This is because language skill is at the heart of reading speed."

That is a mechanistic claim, not encouragement. Reading rate is limited less by how fast the eyes move than by how fast words are recognised and integrated. A reader with a larger vocabulary identifies more words automatically, needs fewer regressions, and gets more from preview. The bottleneck is language, and language responds to reading a great deal - slowly, over years.

What this does not say

It does not say skimming is useless. Skimming works, and everybody does it. The review is explicit that it runs two to four times faster than reading with substantially lower comprehension. That is a legitimate trade when you are looking for whether a document is worth reading. It stops being legitimate when it is sold as reading.

It does not say RSVP has no use. On a small screen, one word at a time is a reasonable presentation format. The evidence argues against the speed claim, not the format.

It does not say your reading speed is fixed. It says the gains come from the slow route.

It is not about audiobooks. That is a modality question with a different answer, covered separately.

The practical version

If you want to get through more text, the evidence supports three things and rules out a fourth.

Read more. It is the only intervention with a mechanism behind it, and it compounds.

Skim deliberately, and know you are doing it. Deciding in ninety seconds whether a paper is worth an hour is a genuinely valuable skill. It is not the same skill as reading the paper.

Protect your ability to go back. The Schotter finding is the most actionable result here: comprehension suffered even on trials where readers did not regress, simply because they could not. Formats that remove the option - RSVP apps, autoscrolling text - take something away whether or not you were going to use it.

Do not buy the technique. There is no published evidence that any commercial speed-reading method delivers a large speed increase at maintained comprehension, and there has been enough testing that its absence is informative.

Reading rate is one of several places where the intuitive fix turns out to cost more than it saves - see the cognitive offloading guide for the wider picture, screen reading versus paper for what the surface does, and do audiobooks count as reading for the same question about modality rather than rate.