Some half baked possibly "fun" questions to ask with a refined version of this model in the future:
-it would be cool to take a team from any season, and see how they might perform against another team from another season. I know the game has changed drastically over the years, but still, would at the very least be interesting to see how two historic teams matchup. Or for a player like LeBron, it would be neat to do some head-to-head comparisons of all the team compositions he's had throughout the years.
-I wonder if we might be able to use the model to find similar teams or very closely "clustered" teams, which we could maybe use to increase the amount of data we have for each team. One helpful application of this might be at the beginning of the season after a few games. Another might be to have a focus on playoff teams only, which naturally suffers from a small base size.
Though with all that said, I'm very much happy right now with this more "mundane" question you're working on as I'm getting more and more familiar with this stuff. It's like a paint-along with Bob Ross, but a model-along. :) Great stuff!
Theta_defense_bar is listed in the parameters but it doesn’t have any priors set. And I don’t see anywhere in the rest of the model that it’s connected to anything. Is that intentional or am I missing something? I’m getting high rhat on it and low (6) Neff, and I’m thinking maybe that’s why?
How are you thinking about measuring accuracy/performance and the effect of iterating?
Some half baked possibly "fun" questions to ask with a refined version of this model in the future:
-it would be cool to take a team from any season, and see how they might perform against another team from another season. I know the game has changed drastically over the years, but still, would at the very least be interesting to see how two historic teams matchup. Or for a player like LeBron, it would be neat to do some head-to-head comparisons of all the team compositions he's had throughout the years.
-I wonder if we might be able to use the model to find similar teams or very closely "clustered" teams, which we could maybe use to increase the amount of data we have for each team. One helpful application of this might be at the beginning of the season after a few games. Another might be to have a focus on playoff teams only, which naturally suffers from a small base size.
Though with all that said, I'm very much happy right now with this more "mundane" question you're working on as I'm getting more and more familiar with this stuff. It's like a paint-along with Bob Ross, but a model-along. :) Great stuff!
Theta_defense_bar is listed in the parameters but it doesn’t have any priors set. And I don’t see anywhere in the rest of the model that it’s connected to anything. Is that intentional or am I missing something? I’m getting high rhat on it and low (6) Neff, and I’m thinking maybe that’s why?
What are your thoughts on modeling team-specific home advantage effects? Denver and Utah could pop out with their elevation