There are many other types of tests that can be achieved when you associate categorical
variables. For instance:
Homogeneity of odds ratios tests: If you have more than two categorical variables associating at the same time, then
obviously the analysis becomes more complex. Typically, a core two-way analysis is
split by other variables. For instance, in the textbook example Size x License forms
the core categorical association. Now, say we add a third categorical variable regarding
customers, such as stock exchange Listed versus Unlisted. Now, SAS would create two
Size x License contingency tables, one for listed customers and one for unlisted.
Each could be examined separately, but what if you wish to assess the similarities
or differences between these two listed and unlisted tables? As illustrated in
Figure 15.10 Homogeneity test for differences in contingency tables below, there are tests (e.g. the “Homogeneity of the Odds Ratios” test) that explicitly
test whether these split contingency tables are roughly similar or very different.
Agreement tests: Contingency tables can be used to test the extent to which different raters assess
the same objects. For instance, two influential automotive journalists may each rate
a set of cars on an ordinal scale “Great,” “Good,” “Average,” “Below average,” or
“Terrible.” You could test the extent to which they agree by arranging their ratings
into a two-way table. Similar tests to those discussed above would apply.
Categorical measurements over time: If categorical measures are made at several periods in time, tests can be made to
see whether or not the responses differ over time.
Noninferiority and superiority tests: There are specific tests to compare certain categories. A classic application of
these tests is in pharmaceutical testing. To get a new drug to market, one of the
tests you will typically do is to show statistically that its outcomes are not significantly
worse than some benchmark – perhaps the benchmark provided by other drugs already
on the market. This is a noninferiority test. Say that you are examining drug side effects of a new cholesterol medication.
You might set out to show that your drug does not have significantly worse side effects
than other competitor drugs. Your contingency table would compare drug types (one
categorical variable) against numbers of reported side effects (the second categorical
variable). Superiority, on the other hand, seeks to show clear advantages of your category over others.
You may seek to show that your cholesterol drug has statistically significantly fewer
side effects than your competitors. Contingency table tests exist for each of these
situations.
As noted in the previous section, there are many SAS procedures other than PROC FREQ
which specialize in associations of categorical data. Perhaps foremost among these
is PROC CATMOD, which allows for advanced modeling of these situations, and which
should be the next stop for many readers interested in this area once the essential
skills mentioned here have been conquered.