We are a group of friends who decided to stop arguing about running advice and actually test it. 25 of us lined up at the Lyngby Half Marathon, GPS watches on, training logs ready, and let the data do the talking.
25 friends who trained differently, ran the same course, had very different days. Click any name to see their profile.
We built a regression model on training load, blended with each runner's own stated target time. How well did the data know them?
Bars = actual finish time, coloured by archetype. Triangles = model prediction (blend of OLS regression on training load and each runner's own stated target time).
21.1 km through Lyngby: a rolling northern loop, a vicious climb at km 14, then the long run home. Select a runner to see their pace heatmap.
The claim: the biggest trap in a half marathon is starting too fast. Race-day excitement pushes you to go out harder than planned; the first 5 km feels effortless. But running above your target pace burns through energy reserves early, and by km 15–17 your legs give out and you "blow up": pace drops sharply as you struggle to the finish. Experienced runners know this trap and deliberately hold back early, saving energy for a strong second half. We measured pacing discipline using a split ratio: your average pace per km in the second half ÷ your first half. A ratio of 1.0 = perfectly even. Above 1.0 = you slowed down (burned out). Below 1.0 = you ran the second half faster, a negative split, the gold standard of race execution. The prediction: experienced runners cluster near or below 1.0; less experienced runners skew above it.
Prior: Haney & Mercer (2011) found that recreational runners with more experience show significantly smaller positive splits; experience directly improves in-race pacing judgement.
Each dot = one runner. Values > 1.0 = slowed down in 2nd half. Yellow line = even split.
Compare strategies: try selecting one runner from each archetype; notice how the Experienced start conservatively and stay consistent, while Believers spike early then fade. · Friends in sync: select Cristina + Marta to see two Grinders who ran almost stride for stride, with nearly identical pacing curves, same split ratio (0.97).
Finding: The cluster averages (◆ markers on the chart) tell the story clearly. Experienced runners averaged 0.978: they ran the second half faster, a disciplined negative split. Grinders averaged 0.981, almost perfectly even, the textbook controlled effort. Believers averaged 1.035, showing the classic blow-up pattern: go out ambitious, fade hard. The three groups land in exactly the predicted order. Our data supports this myth clearly: experience and discipline translate directly into smarter pacing.
The claim: the km 13–16 section of Lyngby is a sustained uphill grind, the kind that forces most recreational runners to slow dramatically or even walk. Climbing requires significantly more power output from your legs than flat running, and your heart has to work harder to deliver oxygen. If your training only ever happened on flat roads, your body has never learned to manage that extra demand efficiently. Hill-trained runners have repeatedly stressed those exact muscle groups and cardiovascular pathways, so the climb is a familiar challenge rather than a shock. The prediction: hill-trained runners will show a smaller heart-rate spike (less cardiovascular stress) and a smaller pace drop (less performance loss) through km 13–16.
Prior: Billat et al. (2003) showed hill-specific training improves VO2max utilisation on inclines and reduces the cardiac cost of uphill running.
Lines = avg pace per km. Shaded = killer climb (km 13–16). Red = heart rate (right axis).
Pace increase (slowdown) during km 13–16 vs km 1–12 baseline. Sorted by impact. Cyan = hill-trained.
Finding: The left chart shows the group average pace and heart rate through the climb. Hill-trained runners (solid lines) maintain pace better and show a smaller HR spike; their bodies handle the ascent more efficiently. Non-trained runners drop pace more sharply and push heart rate higher, and many never fully recovered in the final 5 km. The right chart makes it individual: it shows each runner's pace slowdown during the climb versus their flat baseline. Nearly every hill-trained runner (cyan) has a smaller drop than the non-trained group, and the effect is consistent, not driven by a single outlier. Our data supports this myth.
The claim: the intuition seems bulletproof: run more, get fitter, race faster. But for amateur runners, the reality is murkier. Fitness isn't just about how many kilometres you log: recovery quality, training structure, experience, and even natural ability all play a role. A runner doing 50 km/week of slow jogging may not beat one doing 25 km/week of purposeful, varied training. We measured the Pearson correlation between self-reported weekly km during the build-up and actual race finish time to see just how strong (or weak) the volume-speed relationship really is in our group.
Prior: Barandun et al. (2012) found that weekly training volume is a significant predictor of half-marathon race time in recreational runners, but explains only part of the variance; training quality and running economy account for much of the rest.
Finding: There is a moderate negative correlation (r ≈ −0.55); higher weekly volume does tend to mean faster finish times, so the myth is partially supported. But look at the scatter: the spread within any given volume bracket is huge. The most telling pattern is across archetypes: Experienced runners (blue) finish faster than Grinders and Believers even at similar or lower weekly km. Volume matters, but experience and training quality are at least equally important. More running helps; blindly running more does not guarantee a faster race.
Training volume has a shadow side, in theory. But looking at our 5 injured runners, there is no clear pattern: they are spread across the full volume range, from low-mileage to high-mileage weeks. Injury does not respect the weekly km counter. This likely reflects how multifactorial injuries are: sudden load spikes, poor recovery, biomechanics, and bad luck all play a role that a single number cannot capture.
5 runners with no injury data excluded · Hover for names
The claim: interval training means alternating between hard efforts and recovery jogs, for example, 8 × 400 m at race pace with 90 seconds rest in between. Unlike easy base running, it forces your body to sustain a high percentage of maximum effort repeatedly, which builds both raw speed and the ability to control pace under fatigue. The theory is that this race-day pacing control translates directly to smarter splits: interval-trained runners should be less likely to blow up, because they've already practised what hard effort feels like and learned to calibrate it. We compared finish times and split ratios (even/negative vs positive splits) between runners who included intervals in their build-up and those who only ran easy.
Prior: Helgerud et al. (2007) showed that high-intensity interval training significantly improves VO2max and running economy, allowing athletes to sustain a higher percentage of maximal pace for longer.
■ Interval trainers ■ No intervals · Positive = missed target · Negative = beat it
Finding: Interval trainers finish on average 7 minutes faster and show a higher proportion of even and negative splits, suggesting the benefit is not just raw speed but genuine pacing awareness. However, there is a key caveat: in our group, interval training strongly correlates with higher weekly volume and more race experience. A partial correlation controlling for both still shows a positive signal, but with n = 25 we cannot isolate the effect cleanly. Take this as a strong indicator: interval training likely helps, but the effect is probably a combination of speed, volume, and experience working together rather than intervals alone.
After 25 GPS tracks, 4 myths, and a lot of km. Here is what we actually found.
How did our group perform relative to a typical half marathon field? We compare against the Copenhagen Half Marathon, one of Scandinavia's largest races, with thousands of finishers across all ages and levels.
Copenhagen HM representative sample (n=500). Our group n=25. Scaled for comparison. Note: our group is 23-26 years old (prime running age), while the Copenhagen field is open to all ages, with many participants in their 30s and 40s.
Context: Our group median (≈ 109 min) is about 6 minutes faster than the Copenhagen median (115 min). This gap is expected: we are a self-selected, motivated cohort aged 23–27, all of whom tracked training for weeks leading up to race day. The Copenhagen field is open to all ages and abilities, with a large contingent of casual and first-time runners in their 30s and 40s.
Top performers: Célien, Pablo Bauri and Oriol would rank in roughly the top 6–8% of a Copenhagen field, competitive amateur territory. At the other end, Coline and Carlos finish right in the heart of the typical recreational field, ahead of roughly 35–45% of participants.
Spread: Our IQR is roughly 20 min vs Copenhagen's ~33 min. This is a self-selection effect: we all trained together, shared advice, and ran the same course. We converged more than a random urban field would.
Copenhagen HM reference: approximate distribution from publicly reported aggregate results (~2019 edition). Sample n ≈ 500 used for histogram comparison. Percentile estimates use a normal approximation (μ = 115.4, σ = 23.8 min).
The full analysis: raw data pipeline, statistical methodology, chart design decisions, and every line of code that produced this website.
The explainer notebook walks through the complete pipeline from raw GPX files and survey responses to the charts you see here. It covers data cleaning, exploratory analysis, the OLS prediction model, all four myth tests with bootstrapped confidence intervals, and a full mapping of the site's design onto the Segel & Heer narrative visualisation framework.
Open Explainer NotebookOpens as a standalone page on this site. Source .ipynb on GitHub.