Categorical Data Analysis

Social science data, when quantified, can be subject to a variety of statistical analyses, the most common of which is regression analysis. However, the requirements of a continuous dependent variable and of other regression assumptions make linear regression sometimes a less desirable analytic tool because a lot of social science data are categorical.

Social science data can be categorical in two common ways. First, a variable is categorical when it records nominal or discrete groups. In political science, a political candidate in the United States can be a Democrat, Republican, or Independent. In sociology, one's occupation is studied as a discrete outcome. In demography, one's contraceptive choice such as the pill or the condom is categorical. Education researchers may study discrete higher education objectives of high school seniors: university, community college, or vocational school.

Furthermore, a variable can take on an ordinal scale such as the Likert scale or a scale that resembles it. Such a scale is widely used in psychology in particular and in the social sciences in general for measuring personality traits, attitudes, and opinions and typically has five ordered categories ranging from most important to least important or from strongly agree to strongly disagree with a neutral middle category. An ordinal scale has two major features: There exists a natural order among the categories, and the distance between a lower positioned category and the next category in the scale is not necessarily evenly distributed throughout the scale. Variables not measuring personality or attitudes can also be represented with an ordinal scale. For example, one's educational attainment can be simply measured with the three categories of “primary,” “secondary,” and “higher education.” These categories follow a natural ordering, but the distance between “primary” and “secondary” and that between “secondary” and “higher education” is not necessarily equal. It is apparent that an ordinal variable has at least three ordered categories. Although there is no upper limit for the total number of categories, researchers seldom use a scale of more than seven ordered categories.

The examples of categorical data above illustrate their prevalence in the social sciences. Categorical data analysis, in practice, is the analysis of categorical response variables.