Deep dive/Training science
About three percent of free speed is sitting in the last two weeks before your race. The pooled evidence says almost everyone gets the shape wrong, and in the same three directions.
Four pooled outcomes from 14 studies of 174 endurance athletes. Right of the line means the taper helped. Note the bottom row.
| Outcome | SMD | 95% CI | I² |
|---|---|---|---|
| Time to exhaustion | 1.28 | 0.43 to 2.12 | 64% |
| Time trial performance | -0.45 | -0.68 to -0.23 | 8% |
| Economy of movement | -0.47 | -1.06 to 0.12 | 57% |
| VO2max | 0.20 | -0.93 to 1.33 | 87% |
Published values, plotted unmodified. Every point and interval is the pooled estimate from reference 1 exactly as reported. The only transformation is a sign flip on the two time-based outcomes so that right always means better, which is standard direction harmonisation and changes no magnitude. The raw standardised mean differences are printed in the table underneath so the flip can be checked. I² is the reported between-study heterogeneity.
You have done the work. Sixteen weeks, maybe twenty. The long runs are in the bank, the intervals hurt and then stopped hurting quite so much, and there is nothing left to build. Two weeks out from the race, every session you do from here is either going to help or waste what you already have.
Most people treat this fortnight as a holding pattern. Ease off, stay loose, do not do anything stupid. It feels like the part of training where nothing happens.
It is the opposite. The taper is the single most concentrated performance intervention available to an endurance athlete, it is free, and the pooled literature puts the typical gain at about three percent. That is comparable to what a pair of carbon-plated shoes buys you, and unlike the shoes it does not cost anything or require permission from anyone.
It is also, by the numbers, the part of training people most reliably get wrong. Not randomly wrong. Wrong in three specific, predictable directions, all of which follow from the same instinct: that a taper means rest.
A taper is not rest. It is a high-intensity, low-volume block, and every part of that phrase is doing work.
What follows is built almost entirely on one meta-analysis, which is unusual for these pieces and worth stating up front. Wang and colleagues pooled 14 controlled studies covering 174 endurance athletes and, crucially, broke the results down by every decision a taper involves: how much volume you remove, over how many days, what you do with intensity, whether you cut sessions, and what shape the reduction takes. That breakdown is what makes this article possible, and its limits get a chapter of their own before the end.
Start with the strangest row in the whole dataset.
Go back to the chart at the top of this article and look at the bottom two rows.
Time trial performance improves. Time to exhaustion improves, and by a lot. And then VO2max does not move. The pooled estimate is 0.20 with a confidence interval running from −0.93 to 1.33, which is a statistically polite way of saying we have no idea and it is probably nothing. Running economy does not reach significance either.
Put those four rows together and you have the entire mechanism of tapering in one picture. You get faster. The engine does not get bigger. The cost of moving at a given speed does not measurably fall.
So where does the three percent come from?
It comes from removing something rather than adding it. Through a training block you accumulate two things at once: fitness, which is durable and decays slowly, and fatigue, which is large and decays quickly. What you can do on any given day is roughly the first minus the second. For most of a training block the second is substantial, which means your actual capability is permanently hidden underneath it.
A taper does not add to the first. It gets rid of the second. The three percent was always yours; you simply could not reach it while carrying the fatigue that earned it.
The taper does not make you fitter. It makes you visible.
This has a consequence that is worth being blunt about, because it is the most commonly hoped-for thing in endurance sport and the data does not support it. If VO2max does not improve during a taper, then a taper cannot rescue a training block that did not happen. There is nothing to uncover. You can only cash in what you already built, and if you built very little then a perfect taper reveals very little.
It is worth sitting with the VO2max result a moment longer, because it is doing more work than any other number here and it is easy to skate past.
Maximal oxygen uptake is the most-cited number in endurance sport. It is also, over a two-week window, extremely stable. It took months of training to move it, and it will take weeks of doing nothing to lose it. Two weeks of reduced volume is far too short to change it in either direction, and the pooled data confirms exactly that: no detectable change, with an interval wide enough to contain both a modest gain and a modest loss.
Which resolves the anxiety that stops most people tapering properly. The fear is always the same: if I stop training, I will lose fitness. The data says that over the taper window, you will not. The engine is not going anywhere. Detraining is real, but it operates on a timescale that a two-week taper does not reach.
Heterogeneity on that outcome is 87%, which is very high. The studies disagree substantially with each other. That is partly because VO2max is measured differently across labs and partly because it is the wrong instrument for detecting a two-week change.
The time trial estimate, by contrast, has heterogeneity of 8%. The studies almost entirely agree. The outcome closest to actually racing is also the one the literature is most confident about, which is a convenient thing to be true.
So: fatigue leaves, fitness stays, performance appears. Now the question that actually matters, which is how much to take off.
Here is where the shape appears. The relationship between how much volume you remove and how much faster you get is not a straight line, and it is not “more rest is better”. It is a hill, and you can fall off either side of it.
Four bands, four pooled estimates. Right is better.
Published values, plotted unmodified,from reference 1. Signs flipped so right means faster; raw SMDs are −0.25, −0.45, −0.77 and −0.21 respectively.
Cut twenty percent or less and nothing happens. The estimate is small and the interval sails through zero. This is the taper most people actually do, the one where you take an easy day and skip a session, and the pooled evidence says it does not work.
Cut 41 to 60 percent and you land on the peak. It is the largest estimate of the four, it is the most precise, and it is the only one significant at p = 0.001. Halving your training volume is the answer, and for most people that is a far more aggressive cut than they have ever made voluntarily.
And then keep going, past sixty percent, and the benefit disappears again. The estimate collapses to 0.21 and the interval reopens across zero. There is such a thing as too much rest, and it is not far past the optimum.
That last band is the one people find hardest to believe, because it contradicts the intuition that rest is a good you can never have too much of. But it follows directly from the mechanism in chapter two. You are trying to shed fatigue while keeping fitness. Below the peak you have not shed enough fatigue. Above it you have started to lose the thing you were trying to reveal.
The same hill shape appears in the duration data, and this time the far side of it does something more dramatic than fade out.
Watch the bottom row cross to the other side of zero.
Published values, plotted unmodified,from reference 1. Raw SMDs are −0.36, −1.47, −0.78 and +0.69. Note the very wide interval on the 8 to 14 day band: the estimate is the largest but also the least precise.
A week or less works, modestly. Eight to fourteen days produces the biggest effect in the entire analysis. Fifteen to twenty-one days still works. And past three weeks the point estimate lands on the wrong side of zero.
Be careful with that last row: the interval is wide and crosses zero, so the honest reading is not “tapering for a month makes you slower” but “there is no evidence it helps, and the estimate leans the wrong way”. Still, combined with the volume result, the direction is consistent. Both dimensions have an optimum, and both punish overshooting.
The two meta-analyses converge here in a way that is reassuring. Bosquet and colleagues, working from a different and larger set of 27 studies, independently landed on two weeks as the most efficient duration. Two separate reviews, different inclusion criteria, same answer.
If you take one thing from this article, take this one. It is the sharpest result in the dataset and it contradicts what nearly everybody does in race week.
Two arms. They do not merely differ in size.
Published values, plotted unmodified,from reference 1. Raw SMDs are −0.55 for maintained intensity and +0.25 for reduced intensity.
Hold your intensity through the taper and the pooled effect is +0.55, significant, with an interval comfortably clear of zero.
Reduce your intensity and the estimate is −0.25. Not smaller. The other side of zero.
The interval is wide and crosses zero, so this is not proof that easing off makes you slower. What it is, unambiguously, is the absence of any detectable benefit. Dropping intensity does not weaken the taper. On this evidence it abolishes it. You do all the resting, you lose all the sharpness, and the three percent does not arrive.
And yet reducing intensity is exactly what race week feels like it should be. Everything about the instinct says go gentle, keep it easy, do not risk anything. Mujika and Padilla reached the same conclusion twenty years earlier from a different direction: maintain intensity, reduce volume.
Your interval session still happens in the final ten days. The repetitions are at the same pace they were in peak training. What changes is that there are fewer of them, and the easy running around them shrinks a lot.
A useful test: if your taper week contains no running at or faster than race pace, you are not tapering. You are resting, and the pooled evidence says resting is the arm that does not work.
There is a third way to reduce training load, and it turns out to be the wrong lever too.
You can train less by making each session shorter, or by removing sessions from the week. Those sound equivalent. They are not.
Keeping the same number of sessions is the arm that reaches significance.
Published values, plotted unmodified,from reference 1. Raw SMDs are −0.53 for maintained frequency and −0.32 for reduced frequency.
Both arms point the right way, so this is a weaker finding than the intensity one. But only the maintained arm reaches significance, and it does so decisively at p = 0.0002. The reduced arm has an interval that crosses zero.
Mujika and Padilla put a number on the same idea: cut frequency by no more than 20 percent. Keep showing up nearly every day you were showing up before. Just do much less when you get there.
There is a plausible reason. Skills and neuromuscular patterns degrade faster than aerobic fitness, and daily contact with your race pace keeps the movement sharp. Removing whole days removes those contacts. Shortening sessions does not.
Four decisions down, one left: not how much you remove, but the path by which you remove it.
A step taper drops volume once and holds it flat. A progressive taper brings it down over the whole period. Within progressive tapers there are further shapes, and the literature names them: linear, exponential with slow decay, exponential with fast decay.
Both broad approaches work. The progressive one has the larger and more precise estimate, at +0.51 against +0.38, and Bosquet and colleagues went further and identified the exponential fast-decay form specifically as the most efficient.
Published values, plotted unmodified,from reference 1. Raw SMDs are −0.51 progressive and −0.38 step.
Set a length and a total cut, then compare the shapes. The panel underneath looks up the real pooled estimate for the settings you chose.
Volume falls quickly at the start, then flattens out into the race. The form Bosquet et al. identify as most effective.
Volume: a cut of 41 to 60% pools to +0.77, the largest and most reliable band in the analysis.
Length: 8 to 14 days pools to +1.47, the largest estimate in the analysis, though with a wide interval.
The curves are definitions, not data. They are the functional forms named in the tapering literature, drawn from their own equations, the way one would draw a parabola. The panel underneath contains no modelling at all: it reports the actual pooled effect size from reference 1 for whichever band your settings fall into, including where that estimate is null.
Compare the fast-decay curve against the step. Both arrive at the same volume on race day. The fast-decay version drops hard immediately and then flattens out, which means it spends most of the taper at low volume while still descending gently, whereas the step spends the whole period at one fixed level.
The practical difference is smaller than the other four decisions. If you get the volume, the length and the intensity right, the exact curve you take to get there is a refinement rather than a make-or-break.
Every subgroup analysis in the meta-analysis, switchable. This is the whole evidence base for the shape of a taper in one place.
The benefit peaks in the 41 to 60% band and disappears on either side of it.
Published values, plotted unmodified, from reference 1. These are the same five analyses shown separately in chapters four through eight, collected here so they can be compared directly.
There is a subgroup finding in the meta-analysis that reframes everything above, and it is easy to miss because it is stated in a single line.
Studies where athletes did a deliberate overload block before the taper produced significantly better results than conventional tapers. Tapering works better when you arrive into it more fatigued.
Which follows directly from chapter two. If the taper works by removing accumulated fatigue, then how much it can give you depends on how much fatigue there is to remove. Someone who trained comfortably for sixteen weeks has less to shed and therefore less to gain. Someone who pushed into genuine overreaching in the final build has more.
The taper is not a reward for the training block. It is the second half of it, and it only pays out in proportion to what the first half cost you.
This is the part to be careful with, and the meta-analysis does not give us the numbers to be precise about it. There is a real difference between functional overreaching, which a taper resolves into a performance gain, and genuine overtraining, which a two-week taper does not fix. The studies deliberately induced the first under supervision. Nothing here licenses doing the second to yourself and hoping the taper sorts it out.
Effect sizes are hard to feel. Mujika and Padilla give the percentage instead: typically about 3 percent, with a usual range of 0.5 to 6 percent.
Percentages are also hard to feel. So here is the multiplication.
Enter your current race time, or pick a reference one.
This is multiplication, not a prediction. It applies the published performance range to a time you supply, and says nothing about whether you personally will land inside that range.
Arithmetic on a published number.The percentages are the range reported in reference 3, applied by multiplication to a time you supply. There is no model here and no prediction about you specifically. It answers only the question “what would a gain of this size look like on this clock”.
Three percent of a three-thirty marathon is over six minutes. Three percent of a forty-five minute ten kilometre is more than a minute. These are not marginal gains. They are the difference between qualifying and not, between a personal best and a near-miss, and they are available in the two weeks when most people believe nothing is happening.
It is worth naming the comparison directly. Carbon-plated racing shoes generate something in the region of a few percent, and people spend hundreds on them and argue about whether they should be legal. A correctly shaped taper is in the same territory, costs nothing, and requires only that you do less of something you were already doing.
Three sources have been quoted throughout this article as though they agree. On one substantial point they do not, and papering over it would be dishonest.
Both meta-analyses put the optimal volume reduction at 41 to 60 percent. Wang and colleagues found the largest effect there and a null result above sixty. Bosquet and colleagues independently identified the same band.
Mujika and Padilla’s review recommends reducing volume by 60 to 90 percent. Their lower bound is the meta-analytic upper bound. Their upper bound is in territory the pooled data suggests is counterproductive.
How to hold that. The 2003 paper is a narrative review synthesising a broad literature including case studies of elite athletes, and elite athletes train at volumes where a ninety percent cut still leaves a substantial week. The meta-analyses pool controlled trials in mostly sub-elite athletes and weight them statistically. Those are different instruments answering slightly different questions, and the disagreement may be less about tapering than about who is tapering.
The defensible position: for most people, aim at halving your volume. If you are training twenty hours a week, the upper end of the range may apply to you, and you have coaching resources this article cannot replace.
Everything above rests on a small literature, and you should know how small before you rebuild your race preparation around it.
14 studies. 174 athletes. That is the entire pooled sample for the primary source of this article. Some of the subgroup analyses therefore rest on a handful of studies each, which is why several of the confidence intervals are so wide. The 8 to 14 day estimate of +1.47 has an interval running from +0.19 to +2.75, which spans everything from trivial to enormous.
Quality is moderate. The included studies score 5 to 6 on the PEDro scale. Blinding is close to impossible in a training study, since athletes know whether they are tapering.
The sample is narrow. Athletes aged 17 to 32, predominantly male, almost entirely running and cycling. Swimming, rowing, skiing and team sports are barely represented, and there is little in this data about masters athletes.
The bands are coarse.“41 to 60 percent” is a wide range presented as a precise recommendation. The analysis cannot tell you whether 45 or 58 is better, and treating the midpoint as an optimum reads more precision into the data than it contains.
Heterogeneity varies enormously. The time trial result is highly consistent across studies at I² of 8%. The VO2max result is at 87%, meaning the studies substantially disagree, which is a reason to hold the null result there more loosely than the performance results.
What survives all of that is still useful, because the findings that matter most are the ones that replicate. Two independent meta-analyses agree on two weeks and on 41 to 60 percent. Three sources agree on maintaining intensity. Those are the parts to act on. The finer-grained numbers are directional.
Six things, all of them taken directly from the estimates above.
Eight to fourteen days carries the largest pooled effect, and both meta-analyses converge there independently. Under a week works but works less. Past three weeks there is no evidence of benefit at all.
A 41 to 60 percent cut is the peak of the curve. This will feel like far too much, which is the point: the band below it is where most people taper and it produces no detectable benefit.
The single most important line in the data. Your hard sessions stay at the same pace. There are simply fewer repetitions. Reducing intensity is the one arm of the analysis that fails to show any benefit.
Shorten sessions rather than deleting them. Cut frequency by no more than about a fifth. The maintained-frequency arm is the one that reaches significance.
A gradual reduction beats a single step, and a fast initial drop that flattens toward race day is the specific shape identified as most efficient. This matters less than the four decisions above it.
The taper reveals fitness rather than creating it, and it pays out in proportion to the fatigue it clears. A taper on top of a thin block reveals a thin block. There is no version of these two weeks that substitutes for the sixteen before them.
The instinct that ruins tapers is that they are a rest. Everything in the pooled data says otherwise. Cut harder than feels responsible, stop cutting sooner than feels safe, and keep running fast the whole way in. The three percent was always yours. The taper is just how you get to see it.
Every forest plot in this article is published data, plotted unmodified. Six of the seven visuals are pooled effect sizes and confidence intervals taken directly from reference 1, with the only transformation being a sign flip so that right always means better. The raw values are printed alongside.
The one exception is the curve chart in chapter eight, which draws the four taper forms named in the literature from their own equations. Those are definitions rather than measurements, and the evidence panel underneath them reports real pooled estimates rather than modelled ones. The race-time tool is multiplication on a published percentage range. Nothing in this article is simulated.
Effects of tapering on performance in endurance athletes: a systematic review and meta-analysis
Used for: The primary source. Every pooled effect size and confidence interval plotted in this article comes from here: the four outcome estimates, and the moderator analyses by volume reduction, taper duration, intensity, frequency and taper type. Also the cohort details of 14 studies and 174 athletes aged 17 to 32, the PEDro quality range of 5 to 6, and the heterogeneity statistics.
Effects of tapering on performance: a meta-analysis
Used for: The earlier meta-analysis: 27 of 182 screened studies, an overall effect size of 0.59 plus or minus 0.33, and the conclusion that a two-week exponential fast-decay taper reducing volume by 41 to 60% without altering intensity or frequency is the most efficient strategy.
Scientific bases for precompetition tapering strategies
Used for: The performance range used in the race-time arithmetic: typically about 3%, usual range 0.5 to 6%. Also the recommendation to maintain intensity, cut frequency by no more than 20%, and the observation that progressive nonlinear tapers beat step tapers. This review's volume figure of 60 to 90% conflicts with the two meta-analyses, which chapter eleven addresses rather than smooths over.
Written for general interest and not medical or coaching advice. Effect sizes pooled across small studies describe averages, not individuals. If you are managing an injury or a medical condition, take advice from someone who knows your history.