Nominal-Level Measurement

Nominal data are one of the four levels of measurement described in 1946 by S. S. Stevens, a Harvard psychologist. The four levels are nominal, ordinal, ratio, and interval data, and all have specific definitions of their characteristics. It is important to identify the type of data being collected in a research study, so that the correct type of statistical analysis is performed. This entry describes the unique characteristics of nominal data and outlines the data analysis techniques permissible to use with these results.

Nominal data are considered the most crude or simplest of the four levels of measurement. Nominal data are also called categorical, labeled, or nonranked information because the value given functions only to delineate each individual result and to allow the researcher to place similar values into categories. Nominal data refers to a discrete type of information, in which the results are neither measured nor ordered, but simply allocated into distinct categories according to some sort of arbitrary organizing scheme. One category is not considered to be higher or lower than the others. Nominal scales are considered qualitative classifications and are not treated as continuous.

Nominal data are classified as discrete and are analyzed using the binomial class of statistical tests. Nominal data have three characteristics that differentiate them from ordinal, ratio, and interval data: (1) There is no ordering of the different categories, (2) there is no measure of distance between values, and (3) the categories can be listed in any order without influencing the relationship between and among the categories.

It is not possible to conduct arithmetic, statistical, or logical operations on nominal data because the numerical value has meaning only as an identifier rather than an integer. A person’s home address is an example of a number that functions only as a nominal value. A street number of “100 Grove” carries no significance except to identify the building that has been designated “100 Grove Street.” The number “100” does not declare that this home is 100 feet, or 100 miles, from a clear landmark or that the home is 100 times more comfortable than other homes in the area. The number “100” is simply a way to identify the specific home, much as we might use “green house with big tree near the park” to identify a specific home.

It is possible to measure the number of occurrences in each nominal category and calculate a [Page 1149]frequency count for that category. Some nominal variables are dichotomous or binary, defined as only two categories or levels. Examples of dichotomous nominal results include such things as a surgical outcome (dead or alive), a smoker (yes or no), an epidemiological status (healthy or ill), or a functioning status (on or off).

Nominal data are important for researchers, as they often provide key descriptive information about the subjects and their features. Common measure nominal variables include gender, race, ethnicity, marital status, nationality, language, biological species, and religious preference.

Clustering is a data mining technique that has been used successfully with nominal variables. Clustering is the grouping of a set of values in such a way that those in the same cluster are homogeneous or similar to each other. By the same reasoning, objects that belong to different clusters or categories are dissimilar to each other—a phenomenon known as separation—and the distance or dissimilarity measure between clusters can be calculated. Distance calculations such as simple matching—Russell–Rao, Jaccard, Dice, Rogers–Tanimoto, and Kulczynski—can be completed. Distance measures such as Yule, Sokal-Sneath-c, and Hamann can be calculated for binary data.

...