Why You Type Faster Some Days: What 3 Million Nights Show

You take a typing test on Tuesday morning and get 54 words per minute. You take the same test on Thursday afternoon and get 63. Nothing about your ability changed in two days, and the difference is larger than most people improve in a month of practice. It is also larger than the margin many people are tested against, which matters if you are sitting a typing test for a job on a scheduled day.

This is the single most misread phenomenon in typing measurement. People conclude they are inconsistent, that their practice is not working, or that one of the tests must be broken. The actual explanation is that typing is a motor skill, motor performance runs on biological rhythms, and nobody measures those before taking a test.

This article covers what the research says about how much performance swings across a day, the sleep finding that surprises everyone, the smaller factors that stack on top, and how to get a number that does not move around — which is the only kind worth quoting or tracking.

The size of the effect

Start with the headline figure, because it reframes everything that follows. Research on circadian rhythms in performance estimates the range of variation attributable to time of day alone at roughly 10 to 30 percent of the daily mean.

Apply that to typing. A typist with a genuine 60 WPM ability is not producing 60 all day. They are moving somewhere within a band that might run from the low fifties to the mid sixties depending purely on when the clock says they are testing — before sleep, warm-up or anything else is considered.

Set that against how much people actually improve with practice — a few words per minute per month for someone already competent — and the implication is uncomfortable. Day-to-day noise is frequently larger than a month of genuine progress, which means comparing today's score against last Tuesday's tells you almost nothing about whether your practice is working.

The sleep finding nobody expects

The most striking evidence here comes from a study that is unusually well suited to this question, because it measured performance the same way a typing test does.

In Harnessing the Web for Population-Scale Physiological Sensing, Althoff, Horvitz, White and Zeitzer analyzed 3 million nights of sleep against 75 million interaction tasks, using the speed of keystroke and click interactions as the performance signal. That is not a proxy for typing performance — it is essentially the same measurement, at a scale no laboratory could reach.

Two findings matter here. Two consecutive nights of less than six hours of sleep were associated with measurable decreases in performance. And the degradation persisted for around six days afterwards.

That second number is the one that surprises people. The intuitive model is that a bad night costs you the next day and a good night's sleep resets it. The data says otherwise: the effect of a short-sleep episode outlasts the tiredness by the better part of a week, which means a rough weekend is still influencing your Thursday score whether or not you feel it.

Read the bottom row against the rest. Skill is the only permanent entry, and it changes slowly. Everything above it moves fast and moves your score, which is why a single measurement tells you about your circumstances at least as much as about your ability.

The shape of a typing day

Circadian effects on motor performance are well documented and reasonably consistent in shape, even though individuals differ in where their peak falls.

Manual dexterity, grip strength and tapping speed show a low point in the morning and rise through the day toward a peak in the late afternoon or evening. Cognitive performance follows a related pattern, with the poorest results early in the morning and late at night, and the best in the afternoon. The attention literature, including work collected in Circadian Rhythms in Attention, describes the same broad curve.

Two features of that curve are worth noting. The early-morning trough is real and substantial, which is why testing at seven in the morning produces a figure that flatters nobody. And there is often a dip in the early afternoon before the later peak, which many people experience as the post-lunch slump.

The important caveat is chronotype. Whether you are a morning or evening person shifts this whole curve by hours, and the sleep research specifically identified chronotype as one of the factors influencing measured performance. General advice about the best time to test is a population average, not a prescription — your own peak is discoverable by testing at different times for a week.

The warm-up effect is the biggest thing you control

Of everything on this page, one factor is both large and entirely within your control, and almost nobody accounts for it.

The first typing run of any session is measurably slower than the second. This holds for nearly everyone regardless of skill level, and the mechanism is straightforward — motor patterns take a short period to fully engage, and the first attempt is spent doing that engagement.

The consequence for measurement is severe, because people rarely compare like with like. A cold first attempt last month against a warm fourth attempt today is the single most common way people convince themselves they have improved dramatically. Add an observer and a second mechanism appears entirely, examined in why skilled typing collapses when it is being assessed. The reverse comparison convinces them they have got worse.

• Always discard the first run of a session — treat it as the warm-up it already is.
• If you must record one run, take three and use the middle one rather than the first.
• Do not compare a warmed-up score against a cold one from a previous session.
• The effect is largest for people who type less often, because the patterns are less readily engaged.

This single habit removes more measurement noise than anything else available, which is why it appears in every protocol on this site. The broader set of reasons two measurements of the same typist disagree is in what makes two sites disagree about the same typing.

The smaller factors that stack

Beyond sleep, clock and warm-up, several smaller influences each move the number by a modest amount — and because they stack, they collectively account for a good share of the remaining variation.

Hand temperature is the most underrated. Finger dexterity depends on the temperature of muscles and tendons, and cold hands are measurably stiffer and slower. Anyone typing in a cold room, or immediately after coming in from outside, is starting from a handicap that a minute of warming would remove.

Accumulated typing matters too. Testing at the end of a day spent writing produces a lower figure than testing at the start, because the hands and the attention have already done their work. This is distinct from within-session fatigue and it is why the same person tests differently before and after a working day.

The point of that list is not to supply excuses. It is that a score produced under three or four of those conditions is not measuring the same thing as a score produced under none of them, and treating the two as comparable is what produces the confusion this article is about.

Worth noting how the items interact rather than simply adding up. Under-slept and cold and untested-today is not three small penalties; sleep debt makes the warm-up take longer, cold hands raise the error rate, and a raised error rate costs more time than the errors themselves because each one has to be corrected. The bad days feel disproportionately bad because the factors genuinely compound.

The same works in reverse, which is why the good days feel remarkable. A rested afternoon on a familiar keyboard after a warm-up run stacks several favorable conditions at once, and the resulting score can sit near the top of your range. That is the number people remember and quote — and it is exactly the one they cannot reproduce on demand, because reproducing it requires all the conditions to align again.

Effort, tension and the harder-is-slower trap

One source of variation is self-inflicted and worth separating from the biological ones, because the fix is immediate.

Trying harder usually makes typing slower. The physical expression of effort is tension — raised shoulders, gripped hands, forced rhythm — and every one of those degrades fine motor control. Errors rise, and since each error costs the time to make it plus the time to correct it, effective output falls even when the raw pace looks faster.

This is why people sometimes record their best scores on days they were not trying, and their worst on days they sat down determined to beat a target. The determination produced the tension that produced the errors. Relaxed typing is faster typing, and the habits that generate unnecessary tension are catalogued in the mistakes that quietly cost you accuracy.

Getting a number that does not move

All of this has one practical purpose: producing a figure you can quote, reproduce and track. The method follows directly from the mechanisms.

If you want a baseline now, take a one-minute test after a warm-up run and note the conditions beside the score. Where the resulting figure sits against everyone else is in what counts as an average typing speed, and where your history is stored is covered in how this site records and stores every run.

Why runs in one sitting tell you less than you think

A natural response to variable scores is to take more tests. That helps, but only if the tests are spread out — and the reason is worth understanding because it changes how you should sample.

Five runs taken in ten minutes share almost every condition on the list above. Same sleep debt, same time of day, same hand temperature, same warm-up state, same keyboard, same accumulated fatigue. They are not five independent measurements of your typing; they are five looks at one state.

Statistically, that means averaging them narrows the noise from within-session randomness and does nothing at all about the between-day variation, which is the larger component. You end up with a very precise measurement of how you type on Tuesday afternoon after four hours of email, which is not the quantity you wanted.

The practical rule follows directly. If you are comparing two things — two keyboards, two passage settings, two techniques — test them back to back in one sitting, because holding your state constant is exactly what you want. If you are establishing your speed, spread the runs across days, because sampling different states is exactly what you want. Using the wrong sampling strategy for the question is a common and invisible error.

Why this matters for tracking progress

The practical consequence of everything above is a rule about time horizons, and it is the reason many people abandon practice that was actually working.

If daily variation is larger than weekly improvement — and for anyone already competent it usually is — then a day-to-day comparison is measuring noise. Someone practicing diligently might genuinely gain two words per minute over a fortnight while their individual scores swing by ten, which means the improvement is invisible in the day-to-day data and clearly present in the fortnightly medians.

People who compare today against yesterday therefore experience their progress as random. Some days it looks like they have gained five words per minute and some days like they have lost eight, and the reasonable conclusion from that pattern is that the practice is not doing anything. It usually is; the measurement is simply too noisy to show it at that resolution.

• Compare weekly or fortnightly medians, never individual sessions.
• Expect a flat or negative week occasionally even while genuinely improving.
• Watch accuracy and consistency alongside speed, since they often move first.
• If a month of medians is flat, that is a signal worth acting on. A bad Tuesday is not.

A genuinely flat month is a different matter and does mean something — usually that practice has become repetition rather than learning, which is a mechanism rather than a mood. That case is covered in why automaticity stops progress, and what to change is in a practice routine that actually moves the number.

What variation is not

Two distinctions are worth drawing clearly, because both get confused with the day-to-day swing discussed here.

The first is the consistency score. That measures how evenly your speed was distributed within a single run — whether you typed at a steady pace or in bursts and pauses. It is a technique signal and it is measured inside one test. Day-to-day variation happens between tests and is mostly a state signal. A typist can have excellent within-run consistency and still swing widely across a week.

The second is a genuine plateau. Variation is scores bouncing around a stable level; a plateau is that level not moving over months. They feel similar week to week and they have entirely different causes and fixes. The way to tell them apart is resolution — variation disappears when you average across weeks, and a plateau does not.

Which is really the theme of this whole article. Almost every confusion about typing scores comes from reading a measurement at the wrong resolution: judging skill from one session, judging progress from one week, or comparing two numbers produced under conditions that were never comparable. The typing is usually fine. The reading of it is what needs work.

There is a final point worth making about how this changes the experience of practicing, because it is the part that keeps people going. Once you accept that a single score is you plus your circumstances, a bad session stops carrying any information about your ability — and therefore stops being discouraging. It becomes what it actually is, which is one sample from a range, taken under conditions you can now name.

That reframing is worth more than any of the individual mechanisms above. People abandon typing practice because a run went badly and they read it as evidence, when the honest interpretation was that they slept poorly on Sunday, it was their first attempt, and it was eight in the morning. None of those facts say anything about whether the practice is working.

So take the measurement seriously and take any single instance of it lightly. Fix the conditions when the number matters, ignore the daily noise when it does not, and judge your progress on a timescale long enough for the signal to be larger than the variation. That is the whole discipline, and it is the difference between a number that tells you something and a number that just moves.