ivprobit with multiple endogenous variables, Wald test of exogeneity

[Cathy Antonakos]

I am using ivprobit to test an instrumental variable probit model with multiple (continuous) endogenous variables. The estimate of the p-value for the Wald chi-squared test of exogeneity displayed in the results does not match the p-value stored in e(p_exog). I believe the test of the endogenous variables is a test of joint endogeneity. Can someone explain this difference in the p-values? I have been storing the e(p_exog) in a matrix for later use, and want to be sure that I am capturing the correct p-value for the Wald test. (This difference in p-values does not occur with a single-mediator model.) Thanks.

[Cathy Antonakos]

This is a followup to my question about ivprobit with multiple endogenous variables described in the earlier post. Specifically, using -ivprobit- in Stata version 12.1 (never updated), a model with two endogenous variables appeared to store a 1 df p-value for the Wald chi-square test of exogeneity in e(p_exog). We surmised this by typing the chi-square values for several models into a chisquare calculator, and varied the df from 1 to 2. The p-values returned by the calculator using 1 df matched the p values stored in e(p_exog). The p-values returned using 2 df match the p values stored in e(p_exog) using -ivprobit- in Stata version 14.0. The version 12.1 results displayed under the regression table were correct (Wald chi-square, df, and p-value), but the p-value stored in returned estimates did not match the p-value displayed under the table. I no longer have access to the computer with version 12.1 installed, but using the same data and code I now appear to get the correct results with -ivprobit- in Stata version 14.0.