Nominal, Ordinal, Interval, or Ratio? Why Measurement Level Changes the Analysis

Start With the Comparisons You Want to Make

A useful way to understand measurement level is to ask progressively stronger questions:

These questions correspond broadly to nominal, ordinal, interval, and ratio measurement. Each successive level supports additional numerical relationships, although the familiar four-level classification should not be treated as an automatic test-selection algorithm. (Lovric, 2011; Sekaran & Bougie, 2016).

Nominal: Categories Without an Inherent Order

A nominal variable places observations into distinct categories. The categories tell you whether observations are alike or different with respect to the measured characteristic, but they do not establish a meaningful ranking among those categories. (Lovric, 2011; Meier et al., 2014; Sekaran & Bougie, 2016).

Realistic examples include:

• Survey research: preferred source of news—television, social media, newspaper, radio, or other.
• Health research: treatment group—usual care, medication, or behavioral intervention.
• Education research: degree program—PhD, EdD, or PsyD.
• Business research: customer acquisition channel—search, referral, social media, or direct sales.

The categories may be represented by words or numerical codes, but the codes do not create quantitative magnitude. Coding acquisition channels as 1 = search, 2 = referral, 3 = social media, and 4 = direct sales does not imply that direct sales represents “more” acquisition channel than search. The numbers function as category labels. (Sekaran & Bougie, 2016; Adams & Lawrence, 2018).

Useful summaries for nominal data

Frequencies and percentages directly describe how observations are distributed across nominal categories. The mode can identify the most frequent category. Means and medians are not meaningful summaries of nominal codes because nominal measurement supplies neither quantitative distances nor an ordering required to locate a middle category. (Meier et al., 2014; Adams & Lawrence, 2018).

For example, a health study could report that 42% of participants were assigned to usual care, 31% to medication, and 27% to behavioral intervention. Averaging arbitrary treatment codes would add no interpretable information.

Ordinal: Categories With Meaningful Order

An ordinal variable retains the categorical structure of nominal measurement but adds a meaningful ranking. You can determine that one observation represents more or less of the measured characteristic than another, but the scale does not establish the quantitative size of the difference between successive ranks. (Meier et al., 2014; Lovric, 2011; Sekaran & Bougie, 2016).

Examples include:

• Survey research: satisfaction categories from very dissatisfied to very satisfied.
• Health research: ordered symptom severity such as mild, moderate, and severe.
• Education research: class rank or an ordered achievement category.
• Business research: customers ranking several product features from most to least important.

The distinction between nominal vs ordinal is therefore not whether numbers appear in the dataset. It is whether the categories themselves possess a substantively meaningful order. (Lovric, 2011; Adams & Lawrence, 2018).

Suppose a survey records satisfaction as 1 = very dissatisfied, 2 = dissatisfied, 3 = neutral, 4 = satisfied, and 5 = very satisfied. The coding preserves an obvious ordering. But the coding alone does not demonstrate that the change from 1 to 2 represents exactly the same amount of change in satisfaction as the change from 4 to 5. Ordinal measurement establishes order without establishing equal quantitative distances. (Meier et al., 2014; Sekaran & Bougie, 2016).

Useful summaries for ordinal data

Frequencies and percentages preserve the category structure, while the median can describe the middle ordered response. Meier et al. (2014) and Adams and Lawrence (2018) distinguish these summaries from a mean, which requires treating numerical differences as meaningful. (Meier et al., 2014; Adams & Lawrence, 2018).

This distinction matters in survey reporting. Saying that the median patient-rated symptom severity was “moderate” respects the ordering of the categories without claiming that the numerical distance between categories is known.

Interval: Differences Become Meaningful

An interval scale adds another property: equal numerical differences represent equal differences in the measured characteristic. The scale therefore supports statements about the magnitude of differences, not merely category membership or rank. (Lovric, 2011; Sekaran & Bougie, 2016).

Temperature measured on a scale with an arbitrary zero provides the classic logic. A one-unit difference represents the same numerical interval wherever it occurs on the scale, but zero does not necessarily represent the complete absence of the underlying characteristic. Consequently, differences can be interpreted while ratios need not be. (Lovric, 2011; Sekaran & Bougie, 2016).

Interval measurement supports summaries such as the arithmetic mean and standard deviation because numerical distances are being treated as quantitatively meaningful. Adams and Lawrence (2018) likewise connect interval and ratio measurement to quantitative descriptive summaries, while emphasizing that the distribution should also be considered when deciding how best to describe the data. (Sekaran & Bougie, 2016; Adams & Lawrence, 2018).

A caution for rating scales

Ratio: Meaningful Differences and a Meaningful Zero

A ratio scale has the ordering and equal-interval properties of interval measurement plus a meaningful zero point. The zero represents the absence of the measured quantity in the sense required by the scale, allowing ratios between values to have a meaningful interpretation. (Lovric, 2011; Sekaran & Bougie, 2016).

Examples can include:

• Health research: elapsed treatment time.
• Education research: number of completed assignments.
• Business research: units sold or revenue measured from a meaningful zero.
• Survey or administrative research: number of service contacts during a defined period.

If one business unit sold 200 units and another sold 100 units under the same definition and period, the first sold twice as many units. That ratio interpretation depends on a meaningful zero: zero units means no units were sold. Ratio measurement therefore supports comparisons that would not be justified merely because an interval scale contains the numerical value zero. (Sekaran & Bougie, 2016).

For ratio variables, arithmetic means, standard deviations, and other quantitative summaries can be meaningful, subject to the distribution and purpose of the summary. Sekaran and Bougie (2016) also identify ratio-specific possibilities such as ratio comparisons because the zero point permits proportional statements. (Sekaran & Bougie, 2016; Adams & Lawrence, 2018).

The Measurement-Level Matrix

Coding Numbers Onto Categories Does Not Upgrade the Measurement Level

Suppose an education study records instructional format as:

• 1 = in person
• 2 = hybrid
• 3 = online

The software sees numbers, but the research variable remains nominal if those formats are simply alternative categories. Replacing their names with 1, 2, and 3 does not create an ordering or equal distances. Sekaran and Bougie (2016) explicitly describe nominal numerical codes as convenient category labels without intrinsic quantitative value, and Adams and Lawrence (2018) similarly note that nominal categories could be assigned very different numbers without changing what was measured. (Sekaran & Bougie, 2016; Adams & Lawrence, 2018).

The same principle applies to ordered categories. Coding health status as 1 = mild, 2 = moderate, and 3 = severe preserves the ordering, but it does not establish that the difference between mild and moderate is quantitatively identical to the difference between moderate and severe. The variable remains ordinal unless the measurement process provides a defensible basis for interpreting equal distances. (Lovric, 2011; Meier et al., 2014).

Why Measurement Level Changes Statistical Analysis

Measurement level matters because statistical calculations make claims about relationships among values.

Calculating a mode requires only identifying which value occurs most frequently, so it can be meaningful even for nominal categories. A median additionally requires meaningful ordering. A mean uses numerical distances and therefore requires a quantitative interpretation that simple category membership or ranking does not provide. (Meier et al., 2014; Lovric, 2011).

The same issue carries into inferential analysis. If a business researcher wants to compare the mean revenue of two groups, a mean-based procedure may match the outcome and research question because revenue can be treated quantitatively. If an education researcher instead wants to compare students on an ordered achievement category, a rank-based or ordered-response approach may better preserve the measurement structure. The choice still depends on the complete design and intended inference rather than measurement level alone. (Lovric, 2011; Sekaran & Bougie, 2016).

Similarly, a nominal predictor can legitimately appear in an analysis of a quantitative outcome when it is represented as a categorical predictor rather than treated as though its category codes were a continuous numerical scale. Meier et al. (2014), for example, discuss dummy-variable regression as a way of incorporating categorical information into regression analysis. (Meier et al., 2014).

A Practical Decision Sequence

Before searching for the “right statistical test,” work through the measurement question in this order:

1. Define what the variable represents. Is it a category, an ordered category, or a quantitative measurement?
2. Ask what comparisons are justified. Can you distinguish categories, rank observations, compare numerical differences, or interpret ratios?
3. Examine zero. If ratios matter, does zero genuinely represent the required absence of the measured quantity?
4. Choose descriptive summaries that preserve those properties. Do not calculate a mean simply because the data column contains numbers.
5. Define the research question and estimand. Are you comparing category frequencies, ranks, means, relationships, or something else?
6. Choose candidate statistical methods. Use measurement level to rule out interpretations the data cannot support, then consider the study design and the assumptions of the candidate model.
7. Check diagnostics and assumptions. Measurement level is one part of method selection, not a replacement for examining independence, distributions, variance structure, or other conditions required by the analysis. (Meier et al., 2014; Lovric, 2011; Sekaran & Bougie, 2016).

Common Mistakes

Bottom Line

The four levels of measurement in statistics are useful because they describe progressively stronger kinds of information.

With nominal measurement, you can distinguish categories. With ordinal measurement, you can also rank them. With interval measurement, numerical differences become meaningful. With ratio measurement, a meaningful zero additionally permits proportional or ratio comparisons. (Lovric, 2011; Sekaran & Bougie, 2016).

A category code is still a category code, an ordered category does not automatically have equal intervals, and the appearance of zero does not automatically create a ratio scale. Those decisions determine what your descriptive summaries and statistical models can legitimately mean. (Meier et al., 2014; Lovric, 2011; Sekaran & Bougie, 2016; Adams & Lawrence, 2018).

References

Adams, K. A., & Lawrence, E. K. (2018). Research methods, statistics, and applications (2nd ed.). SAGE Publications.

Lovric, M. (Ed.). (2011). International encyclopedia of statistical science. Springer. https://doi.org/10.1007/978-3-642-04898-2

Meier, K. J., Brudney, J. L., & Bohte, J. (2014). Applied statistics for public and nonprofit administration (9th ed.). Cengage Learning.

Sekaran, U., & Bougie, R. (2016). Research methods for business: A skill-building approach (7th ed.). John Wiley & Sons.