Testing Independent Relationships

An example of data from a study of disfl uencies in speech is shown in Table 1, which shows the parts of speech of 335 words following three types of disfl uencies. Both the part of speech and the disfl uency markers represent categorical variables. Table 1 shows that 30 uh’s were followed by a noun, 20 uhm’s were followed by a verb, etc. One question a researcher may be interested in exploring is whether there is a correlation between the kind of disfl uency produced—the variable in the rows—and the part of speech of the word following the disfl uency—the variable in the columns. An exploratory glance at the data suggests that uh mostly precedes conjunctions while silences most precede nouns, but an actual statistical test is required to determine (a) whether the distribution of the parts of speech after the disfl uencies is in fact signifi cantly different from chance, and (b) what preferences and dispreferences this data set refl ects. The most frequent statistical test used to analyze two-dimensional frequency tables such as Table 1 is the chi-square test for independence.