The best way to test as a whole would be to do a likelihood ratio test. Since Model A is completely nested in Model B, the difference in the -2 log likelihoods should be asymptotically distributed as chi-squared with df=number of additional parameters. ODS OUTPUT is your friend for getting the likelihoods into datasets where you can manipulate the values and do the calculations. Oh, and I forgot this, everything between Model A and B has to really, really be exactly the same--no changes in PSUs or sampling weights.
Steve Denham