Timeline for Heterogeneous Treatment Effects - Interpretable Methods?
Current License: CC BY-SA 4.0
4 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Mar 22, 2020 at 17:49 | vote | accept | Parseltongue | ||
| Sep 30, 2021 at 16:38 | |||||
| Feb 6, 2020 at 22:56 | comment | added | MartinQLD | You can look at something like the interaction frequency (number of times a particular moderator appears in a tree with the treatment) - the tidytreatment package that will show that. The interaction frequency is only a rough guide to how important a moderator is, as it doesn't say anything about the magnitude of the splits. If it's a particular moderator it's a matter of manipulating both treatment and moderator simultaneously to get estimates of the treatment effect conditional on the moderator. You can see this paper academic.oup.com/poq/article/76/3/491/1893905 for examples. | |
| Feb 5, 2020 at 14:30 | comment | added | Parseltongue | Thanks for this. Just want through bartCause, and it doesn't seem like it has methods for interpreting the relative effect of the "confounders" on the lift of the treatment (something akin to variable importance, but with clearer interpretation). Are there straightforward ways to get at the effect of the moderators on the treatment effect with BARTs? What approaches do you recommend? | |
| Feb 5, 2020 at 4:59 | history | answered | MartinQLD | CC BY-SA 4.0 |