Wild cluster bootstrap in Stata 18

[Katherine Adams]

Hello,

I struggle to get the standard errors after

wildbootstrap xtreg

in Stata 18. The help section says that r(table) would give one a matrix containing the coefficients with their standard errors, test statistics, p-values, and confidence intervals. Running

mat list r(table)

indeed gives me a table with the coefficients, t-statistics, p-values, confidence intervals, as well as something called eform (which is always 0). However, the table does not contain the standard errors. Does anyone know how to get the standard errors then?

Thank you.

[George Ford]

The distribution need not be symmetric/normal. So there is no SE in the normal way we thing about it.

[Kevin Duncan]

Thanks for your question Katherine and George for his answer. I am crafting a table of regression results based on WCB. Typically, journals prefer that standard errors are reported in parens under coefficients. Would you recommend reporting p-values instead of standard errors? Or t-statistics? CIs would be useful, but space-consuming and awkward with numerous variable. Thanks for any advice, Kevin

[George Ford]

You can report p-vals or the both sides of the critical t-values.

[Kevin Duncan]

Thanks for the advice George. I was leaning toward using p-values. This is my first experience with WCB and noticed something odd about the t and p-values. For example, one of the coefficients has a t-stat of 2.10 with a corresponding p-value of 0.084. Since there are over 600 degrees of freedom, I would expect this p-value to be less than 0.05. Another t-stat is -3.17 with p-value of 0.022. Again, I would think this p-value would be less than 0.01. In general the WCB t and p-values do not align as they do with the corresponding OLS results. Do you have any idea why this is the case? Thanks again, Kevin

[George Ford]

The probabilities do not come from the normal distribution. The t-distribution is constructed from the bootstrap, so the probability is telling you where the t-value sits in the empirical distribution of the t's.

[Kevin Duncan]

Thanks once again. This is very helpful.