Handling possessions with substitutions in them

[rainmantrail]

I'm curious how others have handled possessions in their PBP databases which have substitutions in them. When aggregating the PBP data to create 'stints' for which players are on the court during a given possession, this leads to having possessions with multiple lineups in them, one for the players on the court before and one for after the substitution occurs. If I were to group the data by which players are on the court, I would end up with duplicated possessions. One solution would be to simply suspend lineup changes untli after the possession ends, even if they included points scored by one of the players that was substituted in. Another option would be to do the reverse, overwriting the current players for a possession with those who were subbed in. Or we could use some logic to try to determine if points were scored by any of the players that were swapped in/out and overwrite the lineups for that possession based on that.

Alternatively, we could aggregate the lineups by using durations of stints as opposed to possessions, but then that wouldn't allow us to compare across eras very well, and would need to be adjusted for pace of play in some other way.

[J.E.]

I can't give you a perfect solution but, as a general rule of thumb:

- if the sub occurs at the same time stamp as a shooting foul, and "player makes X of X FT" ends the possession, having the original player in there makes more sense
- if the sub is in the "middle" of a possession (clock runs before and after the sub) - but there are no points scored before the sub - having the substituted-in player as "on the court" seems more intuitive

The truly problematic cases are
SCORE (but possession continues, maybe through And1-fouls)
SUB
SCORE again

but these are rare

[rainmantrail]

I can't give you a perfect solution but, as a general rule of thumb:

- if the sub occurs at the same time stamp as a shooting foul, and "player makes X of X FT" ends the possession, having the original player in there makes more sense
- if the sub is in the "middle" of a possession (clock runs before and after the sub) - but there are no points scored before the sub - having the substituted-in player as "on the court" seems more intuitive

The truly problematic cases are
SCORE (but possession continues, maybe through And1-fouls)
SUB
SCORE again

but these are rare

Thanks J.E., this is along the lines of what I was thinking too. It would get us more than 90% of the way there. And like you said, the remaining problematic cases would be pretty rare. I think for the And-1 fouls, I would want to keep the original player in the lineup if they were taken out during the FT attempt, as they would have had the greater impact on those points scored. I guess that would leave us with just the rare cases where a player gets subbed in before a FT, then the shooter misses the FT, and there's an offensive rebound extending the possession. But in the limit, it seems like these cases should almost even themselves out if players are just as likely to be subbed in as they are subbed out, which probably isn't exactly true, but it's close enough. And any affect this would have on the RAPM values would be very minimal. Probably to at least the 2nd decimal point I would guess.