Power analysis for IV-regression with binary response variable

[Tarjei W. Havneraas]

Hi

Does anyone have references on how to perform power analysis for an instrumental variable regression with a binary response variable? Specifically, we want to estimate power for a model that will use IV-regression to estimate the effect of a drug prescription (X) on the probability of an event occurring (Y) with institutions’ prescription rate as an IV (Z).

I have read Austin Nichols “Causal inference for binary regression” on the topic (https://www.stata.com/meeting/chicag...11_nichols.pdf) and I know there is an online power calculator for IV analysis for continuous response variables (https://venexia.shinyapps.io/PharmIV/) that also provides Stata/R code from GitHub. However, I have not found similar references (or code) for performing power analysis in cases with binary response variable, so if anyone has any leads on this it would be much appreciated.

[Gabriel Schwartz]

Hi Tarjei - were you ever able to find a solution? I am in the exact same spot 9 years after you and hoping you still use Statalist, ha.

[Felix Bittmann]

Given that many people nowadays use linear models for binary outcomes, I wonder whether the linked shiny app is a sufficient approximation. Another option is to use a simulation based approach.