Every research project, survey, or scientometric study begins with the same raw material: data. But before you can calculate an average citation count or compare the productivity of two journals, you need to know what kind of data you are actually holding. A number is not always a quantity, and a category is not always just a label. Misreading the nature of your data is one of the most common reasons statistical analysis goes wrong. This guide breaks down the main types of data and the four measurement scales, so you can match the right tool to the right information from the very start.
Table of Contents
Why understanding data types comes first
The type of data you collect directly shapes the statistical methods, charts, and conclusions available to you. Understanding data types is the first step in choosing the correct statistical test and the most suitable way to display your findings. Get this step right, and the rest of your analysis rests on solid ground. Get it wrong, and you risk calculating an average that means nothing or running a test the data cannot support.
At the broadest level, data falls into two families: quantitative data, which is numerical, and qualitative data, which is categorical. Within these families sit finer distinctions that determine exactly what you can and cannot do with your numbers.
Types of data
The first question to ask about any variable is simple: does it represent a measurable quantity, or does it describe a quality or category? That single distinction separates the two main types of data.
Quantitative data
Quantitative data consists of numerical values that represent measurable quantities, and these values can be subjected to mathematical operations and statistical analysis. Think of the number of books issued in a month, the page count of a thesis, or the time a user spends in a digital library. Not every number is quantitative, though. A PIN code or a student enrolment number looks numerical, but it does not express a quantity. These are identifiers, and adding or averaging them would be meaningless.
Quantitative data splits further into two important sub-types: discrete and continuous.
Discrete data is counted, not measured. It takes whole-number values that cannot be broken into fractions. Discrete data can be measured and counted as finite values assigned to whole numbers that cannot be decomposed into decimals. The number of students in a class, the number of articles published by an author, or the number of libraries in a district are all discrete. You cannot have 2.7 students or 4.5 published papers. A useful clue is that discrete variables often begin with the phrase “the number of.”
Continuous data is measured and can take any value within a range, including fractions and decimals. Body mass, height, and blood pressure are classic examples of continuous variables. The time spent reading, the weight of a book, or the temperature in an archive room can all be measured with ever-greater precision. The more accurate your instrument, the finer the value you can record. Continuous data forms a smooth distribution along a continuum rather than landing on separate, distinct points.
Qualitative data
Qualitative data, also called categorical data, describes a quality or attribute rather than a quantity. It places things into named groups. The subject of a journal, the gender of a survey respondent, the language of a manuscript, or the department a researcher belongs to are all qualitative. You can count how many items fall into each category, but the categories themselves are descriptions, not measurements.
Qualitative data is itself divided into two kinds based on whether the categories have a natural order. When categories have no inherent ranking, the data is purely categorical. When the categories follow a sensible sequence, such as letter grades or satisfaction levels, the data is ordinal. This is where the discussion of data types naturally flows into the four measurement scales, because those scales describe these distinctions with more precision.
The four measurement scales
In 1946, psychologist Stanley Smith Stevens proposed a framework that has shaped statistics ever since. Stevens developed four scales of measurement, and analysts continue to use them because how you record your data affects what you can learn from it. The four scales, often abbreviated as NOIR, are nominal, ordinal, interval, and ratio. They are arranged in increasing order of the information they provide, and each level builds on the properties of the one before it.
Nominal scale
The nominal scale simply names categories without any quantitative value or order. The word “nominal” comes from the Latin for “name,” which captures its purpose exactly. Examples include the type of document (book, journal, thesis), the language of publication, or a researcher’s field of study. The categories are mutually exclusive, and none of them carries numerical significance. You can label them with numbers for convenience, but those numbers are just tags.
Because nominal data carries no order or magnitude, the statistical options are limited. For nominal data, only the mode and frequency counts are appropriate. You can find the most common category, but calculating an average makes no sense. Pie charts and bar charts are the usual ways to display this data.
Ordinal scale
The ordinal scale also names groups, but those groups can be placed in a meaningful order. The catch is that the distances between the ranks are not equal or known. A library user satisfaction survey with responses of “poor,” “fair,” “good,” and “excellent” is ordinal. You know “excellent” ranks above “good,” but you cannot say the gap between them is the same as the gap between “poor” and “fair.” Likert scales, education levels, and competition rankings all fall here.
Since the order matters but the intervals do not, a major limitation of the ordinal scale is that the distance between measurements is not always equal. For this reason, the median and mode are suitable measures of central tendency for ordinal data, while the mean is generally avoided.
Interval scale
The interval scale is numeric and adds a crucial property: the differences between values are equal and meaningful. The classic example is temperature measured in Celsius. The difference between 20 and 30 degrees is exactly the same as the difference between 30 and 40 degrees. However, the interval scale has one important catch. A distinctive feature of the interval scale is that it does not have a meaningful or true zero, which is chosen arbitrarily. Zero degrees Celsius does not mean an absence of temperature. Calendar years are another example. Because the zero is arbitrary, you can add and subtract interval values, but you cannot say that 40 degrees is “twice as hot” as 20 degrees.
Ratio scale
The ratio scale has everything the interval scale offers, plus a true and meaningful zero. This zero represents a genuine absence of the quantity being measured. Ratio scale variables can be meaningfully added, subtracted, multiplied, and divided. The number of citations an article receives, the page count of a document, age, weight, and income are all ratio data. Zero citations truly means no citations, so it is valid to say one paper has twice as many citations as another. This makes the ratio scale the richest of the four, supporting the widest range of statistical operations.
Why choosing the right scale matters
The whole point of identifying your scale of measurement is that it tells you which statistical operations are legitimate. The mean can only be used on interval and ratio levels of measurement because it requires equal spacing between adjacent values. Apply the mean to ordinal data, such as averaging “agree” and “strongly agree,” and the result has no real meaning. This is one of the most frequent errors in student research projects.
A simple rule helps here: statistical measures that apply to lower scales also apply to higher scales, but not the other way around. The mode works for every scale. The median works for ordinal, interval, and ratio data. The mean is reserved for interval and ratio data only. As the scales climb from nominal to ratio, the toolbox of valid techniques grows wider.
The same logic governs which charts you should use and which tests you can run. Nominal data suits proportion tests and frequency tables. Ordinal data fits non-parametric tests that work on ranks. Interval and ratio data open the door to powerful parametric methods like correlation and regression. In a scientometric study comparing journal impact, for instance, citation counts are ratio data, so you have full freedom to compute means, growth rates, and ratios. But if you classified those same journals by broad subject area, you would be back to nominal data and limited to counts and proportions.
This is why thoughtful researchers decide on their measurement scales before collecting a single value. Depending on the shape of the distribution and the level of measurement, only one or two measures of central tendency may be appropriate. Recognising your data type early prevents the painful discovery, halfway through analysis, that your data simply cannot support the conclusions you hoped to draw.
What do you think? Looking at a research project you have worked on, can you correctly classify each variable as nominal, ordinal, interval, or ratio? And how might your conclusions change if you realised a “number” you had been averaging was actually ordinal data all along?
References
- https://pmc.ncbi.nlm.nih.gov/articles/PMC5958489/
- https://math.libretexts.org/Courses/Los_Angeles_City_College/STAT_C1000/01:_Introduction_to_Statistics/1.04:_Types_of_Data
- https://guides.library.harvard.edu/c.php?g=1324910&p=9889267
- https://www.mayo.edu/research/documents/data-types/doc-20408956
- https://statisticsbyjim.com/basics/nominal-ordinal-interval-ratio-scales/
- https://www.mathematicalpsychology.com/Ordinal_Interval_Ratio_Scales
- https://www.statisticshowto.com/probability-and-statistics/statistics-definitions/nominal-ordinal-interval-ratio/
- https://arxiv.org/pdf/2504.03359
- https://www.cdlsworld.org/xwiki/bin/view/WaihonaPedia/observationDataTypes
- https://www.scribbr.com/statistics/central-tendency/
- https://www.scribbr.com/statistics/levels-of-measurement/
Leave a Reply