Basic Statistics for Comparing Categorical Data From 2 or More Groups

Abstract

In both clinical research and quality improvement, it is commonplace to compare groups of patients (eg, treatment versus control, pre versus post, hospital A versus hospital B) on a variety of characteristics. These characteristics usually take the form of (1) continuous data with comparisons made with t tests (for normal distributions) or Wilcoxon rank-sum tests (for nonnormal distributions) or (2) categorical data. Continuous data are data that can take almost any numeric value within a given range and can be subdivided into smaller and smaller increments without losing the meaning associated with the data. Examples of continuous data commonly found in health care include age, height, weight, temperature, or cost. Categorical data, as the name suggests, can be put into nonoverlapping categories, groups, or classes. Some examples of categorical data that frequently occur in health care are gender, disposition, and skill level (eg, RN, LPN, AHT). Antibiotic receipt, chest radiograph receipt, or admission from the emergency department also qualify because they can be categorized into “yes” or “no” responses. As long as people cannot be classified in >1 group, you are likely dealing with categorical data. There is, however, a special type of categorical data that is treated a little differently from the data we discuss in this article, and it is somewhere between categorical and continuous. It too can be put into nonoverlapping categories, but the categories have a logical ordering or sequence. This is called ordinal data, and a Likert scale commonly used on surveys (1 = strongly disagree, 2 = disagree, 3 = neutral, 4 = agree, 5 = strongly agree) is an example. Although experts do not always agree on the best approach to analyze ordinal data, it generally requires a different approach from the categorical data that we discuss in this here. Knowing what type …