When we think about only one observation, we could assume that the bet data provides more information about the rider's probability of success with that particular trick than the outcome data that the observation provides. But when we expand that to think about multiple observations, the outcome data from all of them can give a much clearer picture of what the rider's probability of success with that particular trick actually is.

The issue is that you need multiple (or many) observations to make the outcome data be more useful than the bet data.

Here I'm thinking that we can use the bet data as priors and the outcome data as data that's still incoming. This presents some problems, as, since this data is coming in over time and is trying to infer characteristics about the system that was in play during that whole time period, it's not clear which observation we should use to set the prior from its bet data or how to use all the available observations' bet data at once (this would likely be somewhat better, but

  1. there's missing data
    1. it's unclear whether/how we should impute values where data is missing
  2. there's semi-missing data
    1. reality of recording data in the field led to weird thing of conveniently repeating values
      1. should these be treated a missing or as slightly better than wholly missing data (again, how?)
  3. how should I do this? (I've been thinking about first averaging the bets->probability values across all observations we have them for)