Hi,

I am using survey level data for the US which is divided at the county level, and I have multiple observations per county (multiple residents in each county responded to the survey). I matched this data with data for Covid cases, and data for number of hospitals.

I want to run an IV regression with probit where the dependent variable is whether the respondent voted for Trump (1 = yes, 0 = no). I control for some of the variables in the survey, and I unstrument Covid cases with number of hospitals.

My regression is of the form: ivprobit 2020vote (covid=hospitals) control1 control2 control3 i.county, vce(cluster county)

Normally, if there was only one observation per county I know I could not include county-level fixed effects in my regression. Yet, in this case there are multiple observations for each county so I included county-level fixed effects. However, if I run a regression my variable of interest is dropped because of collinearity, and it says that some counties "predict failure perfectly".

Is it because I am adding the fixed effects at the county level? Should I change it to state level? Any help would be super appreciated

Many thanks,

Joan