Binary Fuzzy Regression Discontinuity Design estimation using panel data

[Sujin Oh]

Dear Statalists,

I am trying to find the proper package for conducting fuzzy RDD using panel data. The model is constructed as such:

1. R* = theta_0 + theta_1*(x-c) + theta_2*Z + theta_3*X + u(error term)
2. R = 1[R*>0]
3. V = alpha + tau*R + beta*(x-c) + gamma*(x-c)*R + phi*Z + epsilon(error term)

here, the effect of R on V is what I want to estimate. Since R is endogenous, its probability grows as x increases and jumps around c, fuzzy rdd approach is chosen. X serves as the instrument variable and Z is a common control variable in both equations.

The point here is that V is also a binary variable(0 or 1). So, I want to conduct a logit or probit estimation for that. However, I couldn't find the way.

The syntaxes listed below were what I've tried, but I couldn't make it to find the syntax that meets three conditions needed: treatment variable(binary instrument variable; binary dependent variable; panel data.

a. xtivreg: dep var cannot be binary
b. etregress: dep var cannot be binary, not with panel data
c. xtlogit: cannot consider treatment effect
d. rdrobust with fuzzy option: not with panel data


Thanks a lot for your generous help in advance.

[Maxence Morlet]

Several options, one of them being

ivprobit

[Jeff Wooldridge]

I have some thoughts on this that I’ll share later.

[Sujin Oh]

Several options, one of them being

ivprobit

Thanks! But I'm afraid ivprobit syntax cannot reflect the feature of panel data.
Is it enough to put i.year in the regression to get fixed effect estimation?

Thanks again for your reply.