At the suggestion of James MacKinnon, I've added an option to boottest that lets you bootstrap the distribution of a coefficient rather than the corresponding t/z statistic. This is theoretically inferior, but Alwyn Young advocates it in his recent paper, "Consistency without inference: instrumental variables in practical applications." He presents evidence that the bootstrap-c is more reliable in IV applications in economics.
I'm not advocating for or against the method at this point--just offering it as an option.
Example:
I'm not advocating for or against the method at this point--just offering it as an option.
Example:
Code:
webuse nlsw88 ivregress 2sls wage ttl_exp collgrad (tenure = union), cluster(industry) boottest tenure, ptype(equaltail) seed(987654321) stat(c) gridmin(-2) gridmax(2)
