Different number of IVs for treatments in simultaneous analysis

[hadigilan]

Hi all,

very briefly:
Suppose we have three treatments in a simultaneous analysis. But we want to input only two instrumental variables for the first and third treatments. How can we do that?

Thanks for your response,

[PhilippBach]

Hi @hadigilan ,

thanks for your interest in our package!

Currently, we have only implemented the case that all instruments Z are used for all endogeneous treatments D in the multiple (endogenous) treatment case. In case you want to deviate from this, I'd suggest to implement each of the IV models separately, i.e., set up a data DoubleMLData backend and and estimate DoubleMLPLIV model separately for each of the treatment variables (say d1 and d3) separately. This should already give you valid IV-results, I think.

In case you want to also use DoubleML's methods for simultaneous inference, things are getting a bit more complicated. Either you'd have to manually merge the objects and/or modify the multiplier bootstrap for adjustment of p-values (p_adjust()) and joint confidence intervals (confint()); Or you'd try to go with bonferroni or p-value based approaches instead (like using stats::p.adjust() function for the resulting p-values) for valid simultaneous inference.

Hope that's helping you already a bit.

[hadigilan]

Thank you for the information! It is certainly helpful. Sincerely, Hadi

…

On Mon, 25 Apr 2022, 08:53 PhilippBach, ***@***.***> wrote: Hi @hadigilan <https://github.com/hadigilan> , thanks for your interest in our package! Currently, we have only implemented the case that all instruments Z are used for all endogeneous treatments D in the multiple (endogenous) treatment case. In case you want to deviate from this, I'd suggest to implement each of the IV models separately, i.e., set up a data DoubleMLData backend and and estimate DoubleMLPLIV model separately for each of the treatment variables (say d1 and d3) separately. This should already give you valid IV-results, I think. In case you want to also use DoubleML's methods for simultaneous inference, things are getting a bit more complicated. Either you'd have to manually merge the objects and/or modify the multiplier bootstrap for adjustment of p-values (p_adjust()) and joint confidence intervals ( confint()); Or you'd try to go with bonferroni or p-value based approaches instead (like using stats::p.adjust() function for the resulting p-values) for valid simultaneous inference. Hope that's helping you already a bit. — Reply to this email directly, view it on GitHub <#152 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ATOI6LAJCIORMZVNGHCA7FLVGZFQXANCNFSM5UDI7ZMA> . You are receiving this because you were mentioned.Message ID: ***@***.***>