I may be able to help with one part of your question, and I want to expand upon a second. As for the cell chi squared values, I am not aware that you see any p values for the individual cells. The cell chi squared value is computed as the cell deviation from the expected (under the null hypothesis of no association) squared and divided by the expected number. These cell chi squared values are then summed for the hypothesis test, which does give a p value. So, if your chi squared test shows a significant association, then I don't think there is any way to map that back into the individual cells that cause that - it is a property of all the cells together (of course, looking at how the row/column percentages vary suggests where the largest differences lie).
Your other question about weights relates to a question I posed last week, which nobody responded to. It appears that weights and frequencies are interchangeable in the fit Y by X platform for nominal variables. Of course, the statistical properties (such as confidence intervals) are greatly affected. If the weights are such as those found in Census Bureau surveys, the confidence intervals are artificially narrow when using the weights (as weights or frequencies, since they have the same impact). But the descriptive statistics, I believe, are more accurate when the weights are used than if they are left out. What puzzles me (and was my question last week) is that the weight variable is ignored in the Analyze Distribution platform. Why? And, why is it used in Fit Y by X but not in Analyze distribution. In Analyze distribution, if you put the weight variable as a frequency, then it affects (properly) the results, but is ignored if used as a weight variable. So, I am confused about the differential treatment of weights and frequencies in the two platforms.
