I'm trying to run a fixed effects regression with a binary dependent variable. I have repeated observations of individuals within a state over time so I'm specifying something along the lines of:
xtset individual_id time_var
xtlogit dep_var ind_vars i.time_var, fe i(state_var)
However, what I am ultimately interested in is the impact of state-level interventions. As a result, I want to include cluster-robust standard errors at the state level. Unfortunately, xtlogit only allows for bootstrapped (and not clustered) standard errors--and will not let me specify a cluster variable for those bootstrapped standard errors.
Based on documentation it seems to be the case that the commands below should perform a cluster bootstrap as clustering is assumed in a panel setting.
xtset individual_id time_var
xtlogit dep_var ind_vars i.time_var, fe i(state_var) vce(bootstrap, ....)
My question is, is the cluster variable for that bootstrap identified based on the cluster identifier in the xtset command (in my case, clustered by individual ID)--or--would the cluster variable be based on the grouping specified in the regression (in my case, clustered on state).
Any thoughts/insights/etc... would be much appreciated.
xtset individual_id time_var
xtlogit dep_var ind_vars i.time_var, fe i(state_var)
However, what I am ultimately interested in is the impact of state-level interventions. As a result, I want to include cluster-robust standard errors at the state level. Unfortunately, xtlogit only allows for bootstrapped (and not clustered) standard errors--and will not let me specify a cluster variable for those bootstrapped standard errors.
Based on documentation it seems to be the case that the commands below should perform a cluster bootstrap as clustering is assumed in a panel setting.
xtset individual_id time_var
xtlogit dep_var ind_vars i.time_var, fe i(state_var) vce(bootstrap, ....)
My question is, is the cluster variable for that bootstrap identified based on the cluster identifier in the xtset command (in my case, clustered by individual ID)--or--would the cluster variable be based on the grouping specified in the regression (in my case, clustered on state).
Any thoughts/insights/etc... would be much appreciated.