Fixed and Random Zeros
Not infrequently, one encounters a table in which a number of the cell counts are O’s. These cells sometimes cause problems when fitting log-linear models. Recall that, in our discussion of logistic regression in Section VII.3, we had a 2 × 200 table that was riddled with 0 counts. Except for the fact that some asymptotic results did not hold, the zeros caused no problems. Cells with zero counts merely have the potential to cause problems.
KeywordsSaturated Model Zero Count Marginal Total Full Table Iterative Proportional Fitting
Unable to display preview. Download preview PDF.