Using two clusters in logistic regression

[Em Mohamed]

Dear Statalister,

I want to use two clusters in logit regression.

I tried to do it, but it doesn’t work. A very simple

logit Y X1 X2 i.Year i.Industry, cluster (Firm State)

Is there any solution to do this?

Thanks in advance.

[Clyde Schechter]

If you mean that you want to cluster on the combinations of Firm and State:

egen firm_state = group(Firm State), label
logit Y X1 X2 i.Year i.Industry, cluster(firm_state)

Note: This assumes that many firms operate in more than one state. If firms are actually nested in states, then this approach is equivalent to just clustering on Firm, and I'm really not sure what you want to do in that case.

[Em Mohamed]

Thanks. But the combination now gives me number of clusters equal to the number of observations. I understand your point. For example, ifI want to cluster on firm and year, should I also create another variable? I mean any two clusters should I do the same? As the logit doesn’t support the multiple clusters.

[Em Mohamed]

If you mean that you want to cluster on the combinations of Firm and State:

egen firm_state = group(Firm State), label
logit Y X1 X2 i.Year i.Industry, cluster(firm_state)

Note: This assumes that many firms operate in more than one state. If firms are actually nested in states, then this approach is equivalent to just clustering on Firm, and I'm really not sure what you want to do in that case.

Thanks. But the combination now gives me number of clusters equal to the number of observations. I understand your point. For example, ifI want to cluster on firm and year, should I also create another variable? I mean any two clusters should I do the same? As the logit doesn’t support the multiple clusters.

[Andrew Musau]

See the logit2.ado file from https://www.kellogg.northwestern.edu...rogramming.htm. Otherwise, you could also use boottest from SSC or vcemway (also from SSC) to implement two-way clustering.