The latest version of boottest lets you do Anderson-Rubin testing of hypotheses that constrain all the coefficients on instrumented variables--e.g., impact of treatment is zero. By default it bootstraps the distribution of the test using Davidson and MacKinnon's (2010) extension of the wild bootstrap to IV models. boottest could already test linear constraints on the parameters in an IV model using the regular Wald test and Davidson and MacKinnon's bootstrap. When applicable, the Anderson-Rubin test runs much faster (and also tests for instrument validity along with testing the null).
A new wrapper called "artest" makes classical (non-bootstrapped) Anderson-Rubin testing convenient.
Examples:
Also, a new syntax lets you specify null hypotheses directly in the command line rather than via the "constraint define" command. And you can specify multiple hypotheses and request corrections for multiple hypothesis testing.
"bootest [a b] [c d], madjust(bonferroni)" tests the joint hypothesis that the coefficients on a and b are zero and, separately, the joint hypothesis that the coefficients on c and d are zero, and requests the Bonferroni correction. "bootest [(a=0) (b=0)] [(c=0) (d=0)], madjust(bonferroni)" does the same.
Install with "ssc install boottest, replace".
--David
A new wrapper called "artest" makes classical (non-bootstrapped) Anderson-Rubin testing convenient.
Examples:
Code:
use http://www.stata.com/data/jwooldridge/eacsap/mroz.dta ivregress 2sls lwage exper expersq (educ = fatheduc motheduc) boottest, ar // bootstrapped A-R test of zero impact artest // classical A-R test
"bootest [a b] [c d], madjust(bonferroni)" tests the joint hypothesis that the coefficients on a and b are zero and, separately, the joint hypothesis that the coefficients on c and d are zero, and requests the Bonferroni correction. "bootest [(a=0) (b=0)] [(c=0) (d=0)], madjust(bonferroni)" does the same.
Install with "ssc install boottest, replace".
--David
