For data analysis, the researcher assigns a numerical value to each nominal category. For example, “male” may be assigned a value of 1, whereas “female” may be assigned a value of 2. The assignment of numbers is purely arbitrary, and no mathematical functions or assumptions of magnitude or ranking are implied or can be performed. Many survey analyses, conducted by telephone, mail, or face-to-face, use nominal level questions. The intent of these surveys is to describe the distribution of responses along these discrete categories. This level of measurement therefore is also referred to as “categorical” because the assignment of a nominal number denotes category membership.

Ordinal

The next level of measurement is ordinal. This level involves the ranking of phenomena. Ordinal means “order” and thus can be remembered as the numerical value that assigns an order to a set of observations. Variables that are discrete and conceptualized as having an inherent order at the theoretical level can be operationalized in a rank-order format. Variables operationalized at the ordinal level have the same properties as nominal categories in that each category is mutually exclusive of the other. In addition, ordinal measures have a fixed order so that the researcher can rank one category higher or lower than another.

The assignment of a numerical value is symbolic and arbitrary, as in the case of nominal variables, because the distance or spacing between each category is not numerically equivalent. However, the numbers imply magnitude; that is, one is greater than the other. Because there are no equal intervals between ordinal numbers, mathematical functions such as adding, subtracting, dividing, and multiplying cannot be performed. The researcher can merely state that one category is higher or lower, stronger or weaker, or greater or lesser.

Many scales useful to health and human service researchers are composed of variables measured at the ordinal level. For example, the concept of self-rated health is ranked from 1 = very poor to 5 = excellent. Although the numbers imply a state of being in which 5 is greater than 1, it is not possible to say how much greater the distance is between responses. Also, the difference in magnitude between 5 (excellent) to 4 (very good) may be perceived differently from person to person. Thus, you must be careful not to make any assumptions about the degree of difference between numerical values.

We want to raise a caution here to remind you that ordinal data must denote mutually exclusive phenomena. A common mistake is the inclusion of the extreme points of an interval in more than one category. For example, let’s say a questionnaire asks us about our age using categories as 1–10, 10–20, 20–30, and so on. What category do you check if you are 10, 20, or 30 years old? These numbers incorrectly appear in two categories. The investigator should have specified mutually exclusive intervals, such as 1–10, 11–20, 21–30, and so on.

Interval

The next level of measurement shares the characteristics of ordinal and nominal measures but also has the characteristic of equal spacing between categories. This level of measurement indicates how much categories differ. Interval measures are continuous variables in which the 0 point is arbitrary. Although interval measures are a higher order than ordinal and nominal measures, the absence of a true 0 point does not allow statements to be made concerning ratios. However, the equidistance between points allows the researcher to say that the difference between scores of 50 and 70 is equivalent to the difference between scores of 20 and 40.

In the social and behavioral sciences, there is considerable debate as to whether behavioral scales represent interval levels of measurement. Typically, such scales have a Likert-type response format in which a study participant responds to one of four to seven categories, such as strongly agree, agree, uncertain, disagree, or strongly disagree. Researchers who accept this type of scaling as an interval measure argue that the distance between “strongly agree” and “agree” is equivalent to the distance between “disagree” and “strongly disagree.” Others argue that there is no empirical justification for making this assumption and that the data generated should be considered ordinal. In the actual practice of research, many investigators assume such scales are at the interval level to use more sophisticated and powerful statistical procedures that are only possible with interval and ratio data. You should be aware that this issue continues to be controversial among experimental-type researchers. If you choose this type of scaling, think about the most purposive assignment of numeric levels to your response set.

Ratio

Ratio measures represent the highest level of measurement. Such measures have all the characteristics of the previous levels and, in addition, have an absolute 0 point. Income is an example of a ratio measure. Instead of classifying income into ordinal categories, as in the previous example, it can be described in terms of dollars. Income is a ratio measurement because someone can have an income of 0, and we can say that an income of $40,000 is twice as high as an income of $20,000.

Experimental-type researchers usually strive to measure a variable at its highest possible level. However, the level of measurement also reflects the researcher’s concepts. If political affiliation, gender, and ethnicity are the concepts of interest, for example, nominal measurement may be the most appropriate level because magnitude does not typically apply to these concepts.

Likert-Type Scale

In the Likert-type scale, the researcher develops a series of items (usually between 10 and 20) worded favorably and unfavorably regarding the underlying construct that is to be assessed. Respondents indicate a level of agreement or disagreement with each statement by selecting one of several response alternatives (usually five to seven). Researchers may combine responses to the questions to obtain a summated score, examine each item separately, or summate scores on specific groups of items to create subindices. In developing a Likert-type format, the researcher must decide how many response categories to allow and whether the categories should be even or odd in number. Even choices force a positive or negative response, whereas odd numbers allow the respondent to select a neutral or middle response.