I just noticed that -estat ovtest- always produces the traditional, nonrobust version of RESET even if the -vce(robust)- or vce(cluster id)- options are used in -regress-. This leads to logically inconsistent analysis where one uses robust or cluster-robust inference for everything except specification tests. Why should the original version of the test always be used when it's known how to easily make it robust? RESET is easily set up as a variable addition test, and therefore one can use the same robust statistics as when testing for significance of omitted variables use the -test- option. As far as calling RESET -ovtest-, I've already complained about that because it makes users think it's testing for omitted variables when it is a pure functional form test.
Admittedly, it's easy to carry out the fully robust RESET in a few lines, but many don't know that. So they rely on -estat ovtest-, never suspecting that they are using a flawed test.
Admittedly, it's easy to carry out the fully robust RESET in a few lines, but many don't know that. So they rely on -estat ovtest-, never suspecting that they are using a flawed test.
Comment