Multinomial logistic regression - There are ..% cells (i.e., dependent variable levels by subpopulations) with zero frequencies
I'm doing multinomial logistic regression in SPSS, but it gives me this error - There are ... ...% cells (i.e., dependent variable levels by subpopulations) with zero frequencies. I ran crosstables between the dependent variable and independent variables and in every dependent variable levels there are independent variable levels with frequencies. Some of them are quite small categories, but still, I have no missing values.
Is that a problem (can I still trust my model and forget about that error) and what could be causing it?
Thanks.
Look at the table "Observed and Predicted Frequencies". Pairwise crosstabulation, if that is what you did, does not show the counts in cells defined by all the predictors at once. Note also the Subpopulation count in the Case Processing Summary. That tells you the number of subpopulations defined by all the factors and covariates.
Also, quoting from an example in the help, "It also notes there are 2 cells with zero frequencies. This is important because the validity of the goodness-of-fit tests depends upon "large sample" properties of the test statistics; thus cells with zero frequencies can be a useful indicator for potential problems. Since there are few (only 4.2%) of these empty cells, you can probably safely use the results of the goodness-of-fit tests."