Timeline for How to handle regression where there are groups of dependent variables
Current License: CC BY-SA 4.0
13 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 12, 2022 at 19:36 | comment | added | radio23 | "This would however work only as long as the sum of all other players doesn't overcome such threshold." Is there a way to impose this constraint so it doesn't overcome this threshold, this is the issue I am facing as I don't know how to add this as a constraint | |
| Jul 12, 2022 at 10:38 | comment | added | Federico Tedeschi | The problem is that you should introduce dependence between the number of minutes each player is playing. In that case, dependence is expressed by the sum of minutes being $240+5*n$, where $n$ is the number of OTs. One way could be to drop one model (and calculate minutes as difference from $240+5*n$). This would however work only as long as the sum of all other players doesn't overcome such threshold. | |
| Jul 12, 2022 at 1:45 | comment | added | radio23 | I still don't understand with your approach how I would be able to add in the constraint on the sum of the outputs, I have a model for each player specifically but when I take the sum of all of the teams players for a given game some of the values are less than 240 @FedericoTedeschi | |
| Jul 6, 2022 at 20:00 | comment | added | Federico Tedeschi | @radio23: of course you don't know ahead of time whether the match will go to OT. That's why I think this should be modeled. There are models that are two steps: for example, the zero-inflated Poisson model, where you first model zeros then, in case the result from the first model is negative, you perform a Poisson model. Also, if you don't model your outcome like that, you could end out with a sum of minutes corresponding to impossible matches (for example, 245 minutes). To prevent this issue, you could ignore possible OTs, so that the sum is always 240. | |
| Jul 6, 2022 at 17:20 | history | edited | radio23 | CC BY-SA 4.0 |
edited body
|
| Jul 6, 2022 at 17:19 | comment | added | radio23 | @FedericoTedeschi I would not know ahead of time if a game went to overtime, this also doesn't constrain the outputs as a group to make sure the sum of the group is 240 minimum which is 48 minutes * 5 players on the court so the sum of the outputs should always be at least 240 | |
| Jul 6, 2022 at 7:18 | review | Close votes | |||
| Jul 21, 2022 at 3:03 | |||||
| Jul 6, 2022 at 7:08 | comment | added | Federico Tedeschi | I think the constraints must include the non-negativity of all values. I think the "real" model is more complicated, since you should first build a count variable for the number of overtimes and then, given the number of OTs, you'd have the constraint both on the total time, and on the time per player (in case of no OTs, not above 48). I think the number of minutes of the last player could be removed from the equation and calculated as difference. Still, there should be the constraint that such number is between 0 and (in case of no OT) 48. | |
| Jul 6, 2022 at 7:00 | comment | added | user2974951 | @YairDaon The constraints are on the outputs, not inputs/coefficients. | |
| Jul 6, 2022 at 6:59 | comment | added | Yair Daon | Does this answer your question? Linear regression with constrained coefficient | |
| Jul 6, 2022 at 6:58 | comment | added | Yair Daon | You can look at this question: stats.stackexchange.com/questions/79059/… | |
| S Jul 6, 2022 at 6:08 | review | First questions | |||
| Jul 6, 2022 at 9:17 | |||||
| S Jul 6, 2022 at 6:08 | history | asked | radio23 | CC BY-SA 4.0 |