Blog
Why Your 800 Was Lying About Your Training Paces
Coaches told us the paces felt too fast when they came off an 800 PR. They were right, and the athletes' own race results said so by exactly how much.
Why Your 800 Was Lying About Your Training Paces
Over the last few weeks several athletes told us the same thing: the target paces felt too fast. Not unreachable, but consistently a step beyond where the workout should have sat. The complaint had a pattern worth paying attention to — it showed up when the paces were anchored to an 800 PR, and it mostly went away when they were anchored to a mile.
That is a very specific piece of feedback, and specific feedback can be checked. We went and checked it. The athletes were right, and their own race results told us by exactly how much.
How a Race Becomes a Training Pace
Every pace rslts shows you comes from one race result. The system picks a race, then converts it into other distances and into training zones.
The conversion uses Riegel’s formula, which is the standard way to turn a performance at one distance into a predicted performance at another:
predicted_time = known_time × (new_distance / old_distance) ^ exponent
Everything depends on that exponent. Riegel published 1.06, and 1.06 is what almost every calculator on the internet still uses, including ours until this week.
The trouble is where that number came from. Riegel fitted it on adult road racers over distances from 1500m up. Our athletes are neither adults nor road racers, and the exponent gets applied to an 800 anyway.
Checking It Against Real Results
Plenty of athletes hold PRs at more than one distance. That makes the formula testable: take their 800, predict their 5K, and compare the prediction against the 5K they actually ran.
The predictions were not close.
| Predicting the 5K from a… | athletes | median error at 1.06 |
|---|---|---|
| 800 | 114 | 12.5% too fast |
| 1600 | 122 | 8.2% too fast |
Every quartile was too fast, not just the median. This was not noise around a correct number — the whole distribution sat on the wrong side of reality.
It also explains the shape of the complaint precisely. The error is not really about the 800; it is about how far the formula has to reach. Going from an 800 to a 5K is a 6.25× stretch. From a 1600 it is 3.1×. Same broken exponent, roughly half the damage — which is exactly why the mile “felt more correct” without feeling right.
One athlete made the cost concrete. A ~2:00 800 runner whose actual mile is 4:30 was being handed mile pace around 4:11. Nineteen seconds per mile. Their own two PRs imply an exponent of 1.165, nowhere near 1.06.
The Same Error, Backwards
Here is the part we did not expect. If the exponent is too low, long predictions come out too fast — but short predictions come out too slow. And they did:
| Predicting from a 5K… | athletes | error at 1.06 | error at 1.135 |
|---|---|---|---|
| down to an 800 | 114 | 14.3% too slow | 0.3% |
| down to a 1600 | 122 | 8.9% too slow | 0.0% |
So the old model was wrong in both directions at once. It pushed athletes too hard on tempo and threshold work, and let them off too easily on short reps. A 10:00 3200 runner was being given 400m reps at 1:06 when their real 400 speed is closer to 0:57.
Fitting the exponent to our own athletes’ results gives 1.135, and it lands nearly every conversion within about 1% — in both directions.
What About Altitude?
The first theory anyone offers is altitude, and it is a reasonable one. Our athletes race at roughly 4,500 feet, and altitude penalises a 5K far more than an 800. That would push the exponent up exactly the way we measured.
We cannot prove it. Race elevation is not recorded on any result in our data, and effectively the entire population races at the same altitude, so there is nothing to compare against. Altitude and the simple fact that developing athletes have more speed than aerobic base would both produce this signal, and we cannot separate them.
What we can say is that it does not change the fix. The exponent is fitted to the athletes actually using the system, and it absorbs whatever combination of causes is behind it. If you coach at sea level, treat 1.135 as ours rather than universal.
The Second Problem
Fixing the exponent still left something out of place: the 800 was winning the anchor selection far too often. For 41.6% of athletes with results at two or more distances, the 800 was chosen as the race everything else got derived from.
That comes from a different formula. Training zones use Daniels’ VDOT, which scores a race partly by how long it lasted. Its model of what fraction of your VO2max you can hold reports a value above 100% for any race under about 11 minutes. For a two-minute 800 it returns 119%, which is not a statement about physiology so much as a formula being evaluated somewhere it was never defined.
The practical effect: an 800 scores 4–8% higher than an equally good 1600 from the same athlete. So the 800 wins, and it wins with an inflated number that then inflates every easy, moderate, and threshold pace derived from it.
Two changes, then. The exponent is now 1.135. And when an athlete has any race of 1500m or longer, that race anchors their paces; the 800 is used only when it is genuinely all they have raced.
What Changes For You
- Threshold, tempo, and easy paces get slower if your zones were anchored to an 800. This is the change the feedback asked for.
- Short reps get faster. 400s and 800s off a distance anchor were too conservative and are now closer to what athletes can actually run.
- Your anchor race may change to a longer one you have already run. Some athletes will find a race they thought was worse is actually their strongest mark — the old formula simply could not see it.
- Race predictions are unchanged. The race plan engine never extrapolated this far; it only ever reaches from 60% of the target distance, which is well inside where 1.06 measures fine. It was already calibrated separately and we left it alone.
If you have a threshold saved from before this change, re-save it to pick up the new numbers.
Why We Are Writing This Down
The honest summary is that we shipped a well-known formula outside the range it was fitted for, and it took athletes telling us the paces felt wrong to go and measure it.
That is worth saying plainly, because “the standard calculator says so” is not evidence. The athletes who pushed back were carrying better information than the model was — and the data to prove it was already sitting in their own results.
If a pace looks wrong to you, tell us. It is a faster route to a correct model than any amount of confidence in a formula.