The first 5 km costs more than it feels
The first 5 km of a marathon feels easy at almost any pace: you’re tapered, the crowd is loud and goal pace feels slow. Split data from more than 1.7 million recreational finishers show how expensive that feeling can be, and also that holding back too much has a cost of its own.
Pacing matters on the day, but it can only spend what your training has built. For nearly all runners, the biggest gains come from running more, consistently, in the months before the race. How you distribute effort over 42.2 km is a second-order refinement on top of that, worth getting right once the running is in place.
Key takeaways
In 1,724,109 recreational marathoners (elites under 2:30 for men and 3:05 for women excluded), the best average finish times came from runners who ran the first 5 km at their own average race pace.
Starting 10% faster than average race pace went with finishes about 37 minutes slower on average. Starting 10% slower went with about 29 minutes slower.
More than 68% of runners started at least 5% faster than their average pace, and fast starters were far more likely to slow sharply in the second half.
A fast final 2.2 km also went with slower finishes, which the author reads as a sign of overly cautious pacing earlier on.
The data are observational, contain no training information, and measure pace against each runner’s own average, which is partly the outcome being studied.
What the data cover
Barry Smyth of University College Dublin published the analysis in the Journal of Sports Analytics in 2018. He collected public results with 5 km splits from 64 races in 11 cities, including Berlin, Boston, Chicago, London, New York and Tokyo, between 2005 and 2016.
The runners were recreational. Smyth excluded 10,302 elites (men under 150 minutes, women under 185 minutes) and 7,395 walkers slower than seven hours. Of the 1,724,109 who remained, 35% were women and 51% were aged 40 or older. Average finish times by city ranged from about 234 minutes in Boston to about 305 minutes in Los Angeles. The results contain no information on training, weekly volume, experience or goal time.
Smyth compared each runner’s pace in a segment with their own mean race pace (MRP) for the whole marathon. The start was 0–5 km and the finish was the 2.195 km from 40 km to the line.
What a fast start costs
The best average finish, 233 minutes, belonged to runners who ran the first 5 km at their MRP. A start 10% faster than MRP went with an average finish about 37 minutes slower, and a start 10% slower with about 29 minutes slower.
Fast starts were the norm. More than 68% of runners ran the first 5 km at least 5% faster than their MRP, against 9% who started at least 5% slower. A third of runners ran the opening 5 km as their fastest segment of the race, and those runners went out about 12% faster than their average.
Smyth links the cost to hitting the wall, defined as a second-half slowdown beyond 25%, 33% or 50%. Few runners who started near or slower than MRP hit the wall by any definition. At the 33% threshold, 44% of runners who started 20% faster than MRP did. Men were more likely than women to hit the wall at any given start pace.
To put the percentages in running terms with a made-up example: a runner aiming for 4:00 needs 5:41 per kilometre. A start 5% faster is about 5:24 per kilometre, and 10% faster is about 5:07.
Does it hold across ability levels?
Smyth split runners into three finish-time bands: 3–4 hours (34% of finishers), 4–5 hours (40%) and 5–6 hours (17%). Fast and slow starts carried a cost in all three. Runners in the 4–5 hour band who started 15% faster than MRP finished almost 4% slower than even starters in the same band, or more than 10 minutes. An even start was best for 3–4 hour finishers, while the 4–5 hour band did slightly better starting a little slower and the 5–6 hour band a little faster.
He also looked at 476,584 runners who appeared in three or more races, and treated each runner’s fastest race in the data set as their best. On average, best races started about 3% faster than MRP and other races about 6% faster. Of runners who started at MRP, 36% ran their best time there, and 77% of best times came from starts within 5% of MRP.
The fast-finish paradox
At the other end, just over 58% of runners finished slower than their MRP, about 21% at it and 19% faster. Slow finishers ran their second half about 15% slower than the first, even finishers about 7% slower and fast finishers about 5% slower. True negative splits were rare.
Fast finishes did not go with fast times. Runners who covered the last 2.2 km 10% faster than their MRP finished, on average, 22 minutes slower than those who finished at MRP, while slow finishers paid a smaller cost. Smyth reads a strong finish as evidence that a runner left too much in the tank. Men were more likely to finish slowly (65% against under 47% of women) and less likely to finish fast (16% against 26%), which fits women pacing more cautiously overall.
The level of runner matters here. Smyth notes that research on elite athletes (Hanley, 2015) suggests tactical mid-race slowdowns and fast finishes do produce better times for them, so these findings apply to recreational runners racing the clock.
How far to trust the data
Smyth argues for a causal link between start pace and finish time: the association is strong, it rises with the size of the pacing error, and it holds across ability bands, cities and repeat runners. Several limits remain.
Relative pace is partly the outcome. MRP comes from the finish time itself. A runner who collapses late has a slower MRP, which makes their opening 5 km look faster relative to it. Some of the fast-start effect is built into the measurement. This is kaizen’s reading of the method, which the paper doesn’t address directly.
The ability bands use the outcome too. Grouping runners by finish time means grouping them by the result being explained.
There is no training data. Weekly volume, fitness, experience and goal time are all missing. Runners who trained more and set realistic goals could both pace more evenly and finish faster, and the study had no way to adjust for that.
Best times are only the best in the data set. They are not necessarily lifetime personal bests.
Conditions vary. Course profile and weather affect splits. Smyth reports that city-by-city and race-by-race analyses gave the same findings, which reduces this concern.
Where the target pace comes from
Even pacing only works if the target is right. A fast start is often a symptom of a goal the training doesn’t support, and generic pace tables make that goal easy to pick.
A second study gives a training-based view. Barry Smyth and Daniel Muniz-Pumares (Medicine & Science in Sports & Exercise, 2020) calculated critical speed from logged training for more than 25,000 recreational marathoners in Dublin, London and New York, averaging about 230 minutes and logging roughly 34–47 km a week. Runners who covered 2–16 km faster than 94% of their critical speed were more likely to slow by more than 25% late in the race: 26.0% versus 20.5% of men and 15.6% versus 9.6% of women. Weekly volume was reported but not adjusted for in the pacing analysis, so fitter, higher-volume runners may account for part of the pattern. The kaizen blog’s critical speed explainer covers that study in more detail.
Pace tables are generic by design. kaizen calculates your fitness from your recent running and past races, so the finish time it shows reflects the training you have actually done. The weekly training target it sets in the build-up comes from the overall intensity of your real running, from pace and elevation, and recalculates as each run arrives. A goal grounded in that fitness gives an even pace a fair chance of holding.
Using this on race day
The findings point to a plain plan for recreational runners in roughly the 3–6 hour range the data cover:
Run the first 5 km close to goal pace, within a few per cent, and treat anything much quicker as borrowed time.
Plan to hold that pace through halfway and beyond, with no time banked early and no big finish saved for the end.
Expect the second half to be a little slower. In this data set even finishers still slowed by about 7%.
Sources and uncertainty
Smyth 2018, Journal of Sports Analytics. 1,724,109 recreational finishers from 64 races in 11 cities, 2005–2016; public 5 km splits; no training data; observational, with an argued but unproven causal case.
Smyth and Muniz-Pumares 2020, Medicine & Science in Sports & Exercise. More than 25,000 recreational marathoners averaging about 230 minutes; critical speed from unverified training efforts; observational; not adjusted for volume.
Hanley 2015. Cited in Smyth 2018 for elite pacing; not reviewed directly here.
No randomised trial here assigned recreational runners to pacing strategies, and neither study could separate pacing from the fitness and goal-setting behind it.
The bigger picture
The split data say a steady start protects the race, and a steady start is easiest when goal pace sits within what your training has built. For almost every marathoner, more running, done consistently through the build-up, moves the finish time far more than any pacing plan. Pacing well makes sure you get what that running earned.
Where kaizen fits
If you’re spending months chasing seconds over 42.2 km, kaizen calculates your fitness from your recent running and past races, so you can set a goal pace your training supports. It then sets the weekly training target that goal requires, built from the pace and elevation of the running you actually do, and recalculates as each run arrives. You choose the sessions. See your fitness in kaizen before you pick your marathon pace.




