As I was doing some cross-tabs recently to generate county- and PUMA-level findings with respect to “persons-per-household”, in which I (for purposes of tracking households) cross-tabbed households categorized by three broad age groups (young adults householder, prime working-age householder, and senior householder), by 10 income categories (<$10K, $10K-14K, $15K-$24K, $25K-$29K , . . . ., $150K-$199K, >$200K), by tenure (Own versus Rent), and select number of household variables of interest (such as “not over-crowded living situation”, “over-crowded”, and “severe over-crowding”). (I did the same cross-tabbing to generate findings on the number of persons in households exhibiting the above cross-tabbed traits). Since I was slicing and dicing the cross-tabbed variables in such as granular fashion, I decided to report findings in situations where I had at a minimum 25 cases.
Question: Is there a “rule-of-thumb” as to minimum number of cases whose findings should be reported? I realize if I utilized the replicate weights (which would allow me to also report m.o.e.) I could report findings even for situations where less than 25 cases share the same cross-tabbed variables of interest to me – but, since I use Excel I prefer to not do this. So, I’ve used the “minimum” number of cases as a short-hand approach to focusing only on findings that I might have some level of comfort in. I came up with “25” as the cut-off purely by random.