Robust clustered standard errors

[Ellinor Hjelvik]

Dear all,

Is there a way I can do both robust and clustered standard errors in ivprobit/xtprobit/xtlogit?
As in using both the vce(robust) and the vce(cluster varid) alternatives at the same time.

Thank you for your help.

[Jesse Wursten]

I might be mistaken, but the cluster() standard errors are by definition "robust". If you define x = 1 for every observation, cluster(x) should produce the robust standard errors.

[Ellinor Hjelvik]

Thank you for your answer. But when you do a normal probit for cross-section or pooled data you have the option to run "probit , ro cluster(id)". Why should the clustering for ivprobit/xtprobit/xtlogit be different?

And this might be due to my lack of knowledge, but what do you mean by "defining x=1 for every observation"? I know that the variance is set to 1 in probit modelling, hence there model should automatically be homoskedatic. Which is confusing as to the option of running a ivprobit/xtprovit with vce(robust) in the first place. Can anyone explain this?

[Jesse Wursten]

probit, robust cluster(id) will return the same results as probit, cluster(id). Whether this is a matter of laziness (the options should be exclusive) or intentional (most people refer to clustered standard errors as cluster-robust standard errors) I don't know. If my memory serves me right, robust standard errors are identical to clustered standard errors if you have just one cluster (this is what I meant with the x = 1). Or vice versa, clustered standard errors are robust standard errors extended to a situation with multiple clusters.

sysuse xtline1.dta, clear
gen y = runiformint(0,1)
probit y calories, r cluster(person)
probit y calories, cluster(person)

[Ellinor Hjelvik]

Ok, thank you for your help