Zero Frequencies and Incomplete Tables
When several categorical variables are cross-tabulated to form a table of several dimensions, some cells often contain zero frequencies of response. Such a situation may arise in at least three ways. The zero may have occurred simply because the sample is not large enough and that combination of categories is not represented. A large enough sample would theoretically include such combinations. In the other two cases, the combination of categories is actually impossible or is excluded from the model for some theoretical reason. In the first case, the expected frequencies or fitted values should be positive. In the latter two cases, the expected frequencies for any model must be zero. In this section, we consider the first case, that of sampling zeroes. In subsequent sections, various possibilities of incomplete tables with structural zeroes will be covered.
KeywordsScale Parameter Migrant Behaviour Symmetry Variable Structural Zero Observe Time Interval
Unable to display preview. Download preview PDF.