Estimated Group-level Propensity Scores

[bsaul]

Users would like to access the estimated propensity scores in addition to the estimated weights.

[BarkleyBG]

Perhaps a simple way to do this would be to add alpha=0.5 to the allocations, and then internally take those weights (and the group sizes) and peel away the IPW from the pi term?

[bsaul]

I’d rather have something more elegant. I don’t know what it is though.

On May 9, 2017, at 12:58 PM, Brian G. Barkley notifications@github.com wrote:

Perhaps a simple way to do this would be to add alpha=0.5 to the allocations, and then internally take those weights (and the group sizes) and peel away the IPW from the pi term?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/bsaul/inferference/issues/1#issuecomment-300231036, or mute the thread https://github.com/notifications/unsubscribe-auth/AEqCLjXuoaFI8esVILQMEAkA3tXHBM6Yks5r4JsjgaJpZM4FBbIg.

[BarkleyBG]

I think there's two options:

1. Add an extra call to wght_matrix() in the ipw_interference() function with allocation 1. (that would work, no?)
2. Create a helper function that takes the output from interference() and calculates the IPW (so long as user supplies 0.5 in allocation. (although I guess this is the same as what I suggested above)

I could add one of these as a feature, if youd like

[bsaul]

I think I prefer option 1. Perhaps do this if the user specifies to keep_ps = TRUE?

On May 10, 2017, at 11:06 PM, Brian G. Barkley notifications@github.com wrote:

I think there's two options:

Add an extra call to wght_matrix() in the ipw_interference() function with allocation 1. (that would work, no?) Create a helper function that takes the output from interference() and calculates the IPW (so long as user supplies 0.5 in allocation. (although I guess this is the same as what I suggested above) I could add one of these as a feature, if youd like

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/bsaul/inferference/issues/1#issuecomment-300670134, or mute the thread https://github.com/notifications/unsubscribe-auth/AEqCLh43syUJRzxOTzD07MP14GFvL8Gnks5r4ntBgaJpZM4FBbIg.

[BarkleyBG]

Seems reasonable. I'll have it default to TRUE.