There were a couple questions about changes in fitness and effects of age:
On age: my conjecture is that when we select runners based on whether they exceeded a performance criterion, we are biasing the sample towards people who are in their prime, since a random runner is more likely to exceed the criterion when they are in their prime. Then when we follow the runners' performances forward, we are biased towards people who are past their prime. Obviously this will not be true for every runner in the sample. I'm talking about an average effect, and at this point, it's a conjecture that could be studied further.
On fitness changing over time: The published model includes a latent "ability" for each runner. This ability does not change over time, which does not account for the fact that runners' fitness changes over time. We tried using a model in which their abilities were allowed to change over time. This model has some nice features. For one, I captures the time-varying fitness. Also, suppose we had two runners that switched from non-vaporfly shoes to vaporfly shoes, but the first runner's marathons were 3 years apart, whereas the second runner's marathons were 6 months apart. It seems logical that the second runner's performances are more valuable than the first, because their fitness was less likely to change by a large amount in 6 months, rather than the first runner's fitness over 3 years. The time-varying fitness model allows us to put more weight on the second runner's performances when estimating the vaporfly effect.
However, the problem was that there was at least one runner who ran marathons in different shoes within a few weeks. The model was putting too much weight on this runner's performances, so we opted to go with the simpler model with unchanging abilities.