I am in charge of weighting survey data of teens in Utah. I have been working my way through the Stata Press publication "Survey Weights: A Step-by-Step Guide to Calculation" and have not been able to find supplementary material regarding the command svycal rake. I don't want to collapse more categories than necessary, but I am not sure what guidelines to use when deciding I have too few elements in a category. I am raking on income and religion. When checking for too few units, do I look at the cells formed by crossing religion and income (eg: there are only 15 agnostic/low income units) or do I look at the individual variables (eg: there are only 67 agnostics)? How few is too few?

Below is the command I am using. The calibration totals for income were recovered from census data and the totals for religion were recovered from a private organization. 'wt' is my base weight.

Code:
svycal rake i.income i.rel [pw=wt], gen(rake_wt) totals(_cons=262720 1.income=138716.16 ///
0.income=124003.84 1.rel=172288.77 0.rel=90431.23)