Zack, Matthew M. wrote: > It doesn't matter how you construct the matrix of marginal totals, > TABLE, as long as > they reproduce the marginal totals of interest. One simple way to do > so is to assume a model of independence among rows and columns. That > would imply that each cell in TABLE would equal the following > expected value: > > expected cell value > in row_i and column_j = (row_i marginal total)*(column_j marginal > total)/(Table total) > > For the cell in row=1 and column=1, the expected cell value would > equal the following: > > (1412)*(3988)/18324 = 307.3, or 307 if rounded to the nearest > integer. > > You can then calculate the values for other 23 cells in this 8-by-3 > matrix, TABLE. Finally, you can modify these values if their rounding > errors cause the marginal totals of this table > to disagree with those you ultimately want.

Ah, yes. I'm with you now. I was obviously looking for complications that weren't there.

Although I can't see why you shouldn't be able to just specify a vector of the marginal totals, I should be able to manage using the technique you suggest.

Thanks for the advice.