Clustering standard errors at state level

[Celine Li]

Hello,

I am having a problem with clustering. I am regressing a dummy variable on a state-level variable using unbalanced panel data.

My code is:

probit dummy_y state_x i.industry_fe i.year_fe, vce(cluster state)

And I find the following results:
When I adjust standard errors using state clustering, the variable of interest state_x, is statistically insignificant. But when clustering by the firm or using robust standard errors, I find a strongly significant coefficient on state_x. What would be the cause of these results?

Thank you in advance.

[Carlo Lazzaro]

Celine:
welcome to this forum.
As per FAQ, please post what you typed and what Stata gave you back (via CODE delimiters, please). Thanks.

[Jeff Wooldridge]

As suggested by Carlo, it would be helpful to see the output. Generally, when you cluster you are admitting that the usable variation in the data (the state_x variable) is at a more aggregated level than your unit of observation. My guess is you have a lot more firms than states. Unfortunately, clustering at the state level is arguably the valid way to cmpute the standard errors. You are, in effect, assuming you have 50 independent pieces of information (assuming 50 states) rather than many more if you cluster at only the firm level. You can't justify not clustering at all because you have a panel data set. You can argue that you're conditioning on the policy assignment and cluster only at the firm level, but that's not very satisfying.