search

Mixtape-Sessions / Causal-Inference-2

Causal Inference II Mixtape Session taught by Scott Cunningham
171 stars 135 forks source link

Sant’Anna and Zhao weights are different #2

Closed pedrohcgs closed 7 months ago

pedrohcgs commented 7 months ago

Just want to mention that the weights mentioned here are not equivalent to thosenin Sant’Anna and Zhao’s (2020) .

SZ use Hayek-type weights while those here are Horwitz-Thompson ones. SZ weights are guaranteed to sum up to one in finite samples, while those here are not.

This matters for stability when overlap is not perfect.

kylebutts commented 7 months ago

Thanks Pedro. Let me make sure I understand this correctly. I was going off of this estimator: https://github.com/pedrohcgs/DRDID/blob/master/R/drdid_panel.R. Is this not the Horwitz-Thompson weights?

kylebutts commented 7 months ago

Ahhhhh, ok I think I follow @pedrohcgs. I was basing off of ipw_did_panel which I believe uses Horwitz-Thompson weights. You are recommending std_ipw_did_panel with the main change being: https://github.com/pedrohcgs/DRDID/blob/8a1c09f932b142e6c3b8281cd3177ab302c583d2/R/std_ipw_did_panel.R#L103 https://github.com/pedrohcgs/DRDID/blob/8a1c09f932b142e6c3b8281cd3177ab302c583d2/R/ipw_did_panel.R#L11

pedrohcgs commented 7 months ago

Yes. Abadie used the H-T type pscores weights, though.

SZ is all about Hayek weights

Pedro H. C. Sant'Anna https://psantanna.com https://psantanna.com

Warning: This email may contain confidential or privileged information intended only for the use of the individual or entity to whom it is addressed. If you are not the intended recipient, please understand that any disclosure, copying, distribution, or use of the contents of this email is strictly prohibited.

On Sun, Mar 17, 2024 at 14:50 Kyle F Butts @.***> wrote:

Ahhhhh, ok I think I follow @pedrohcgs https://github.com/pedrohcgs. I was basing off of ipw_did_panel which I believe uses Horwitz-Thompson weights. You are recommending std_ipw_did_panel with the main change being:

https://github.com/pedrohcgs/DRDID/blob/8a1c09f932b142e6c3b8281cd3177ab302c583d2/R/std_ipw_did_panel.R#L103

https://github.com/pedrohcgs/DRDID/blob/8a1c09f932b142e6c3b8281cd3177ab302c583d2/R/ipw_did_panel.R#L11

— Reply to this email directly, view it on GitHub https://github.com/Mixtape-Sessions/Causal-Inference-2/issues/2#issuecomment-2002576604, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABE734ZGX5CYFM7KT7OJA6DYYXQYJAVCNFSM6AAAAABE2LWWDKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDAMBSGU3TMNRQGQ . You are receiving this because you were mentioned.Message ID: @.***>