The way I ordered the output of estfun.DirectAdjusted() was liable to change upon different calls when no by argument was provided to cov_adj() and the fallback ordering variables--the unit of assignment variables in the Design object--contained NA's. This resulted in vcovDA() spitting out different estimates from call to call. I fixed this bug, which necessitated changing the order of the estimating equations matrix (not consequential in and of itself). The ordering is provided in the documentation. I beefed up testing, and made sure this new procedure works when the cluster argument is invoked in vcovDA().
The way I ordered the output of
estfun.DirectAdjusted()was liable to change upon different calls when nobyargument was provided tocov_adj()and the fallback ordering variables--the unit of assignment variables in theDesignobject--contained NA's. This resulted invcovDA()spitting out different estimates from call to call. I fixed this bug, which necessitated changing the order of the estimating equations matrix (not consequential in and of itself). The ordering is provided in the documentation. I beefed up testing, and made sure this new procedure works when theclusterargument is invoked invcovDA().