Whenever you partition poll results by age, sex, political affiliation, or whatever; the margin of error increases for each of the subgroups. The margin of error depends on the size of the sample. If you divide by (say) sex, then the sample size is reduced by (about) half & the margin of error increases a lot. If the M.O.E. is (say) +/4% for a sample of 600; it increases to about +/- 6% for a sample of 300. The results by age, or by sex, etc. could vary substantially within the M.O.A. Try it yourself here:
http://americanresearchgroup.com/moe.html
Good point, it could be an outlier. I know the statistics formula. When Bush ran the first time I memorized it. :-}