Elementary Statistical Concepts: Data, Variables, and Measurement Scales

Open any newspaper and you’ll see the word “statistics” used two different ways in the same article. One sentence might report unemployment statistics for the quarter, and the next might explain what statistics tells us about the economy. Both uses are correct, and understanding the difference is the first step toward making sense of any data-driven field, from economics to psychology to market research.

The two meanings of statistics 🔗

In everyday use, the word statistics refers to numerical facts themselves: the number of new businesses registered last year, average rainfall in a district, or the literacy rate of a state. This is the plural sense of the term, treating statistics as a collection of figures.

In the singular sense, statistics is a scientific method. It covers the entire process of collecting, organising, analysing, and interpreting data so that valid conclusions can be drawn. This is what separates statistics from casual information. A rumour or an opinion is also a kind of “data,” but it hasn’t been gathered systematically or checked for accuracy. Statistical data, by contrast, follows a defined process at every stage, which is why researchers, policymakers, and businesses rely on it to make decisions rather than guesses.

Classifying variables: qualitative vs quantitative 🔗

Before any analysis begins, you need to know what kind of variable you’re dealing with. A variable is simply a characteristic that can differ from one subject to another, and every variable falls into one of two broad categories.

Qualitative variables place subjects into categories rather than assigning them a number. Gender, religious preference, blood group, and marital status are all qualitative because they describe a quality, not a quantity. These variables are also called categorical, and their categories are usually recorded as words or labels rather than numbers.

Quantitative variables, on the other hand, are numerical and can be ranked or ordered. Age, height, weight, income, and marks scored in an exam are quantitative because each value carries a magnitude that can be compared meaningfully with another. This distinction matters well beyond definitions. A correlation coefficient makes sense for two quantitative variables, while a chi-square test suits two categorical ones, so the type of variable you’re working with directly decides which statistical technique is even valid to use.

Quantitative variables: discrete and continuous 🔗

Quantitative variables split further into two types, and the difference comes down to how the values are obtained.

Discrete variables are countable. They take specific values such as 0, 1, 2, or 3, and nothing in between makes sense. The number of children in a family is a classic example, since you can’t have a fractional child. Other examples include the number of hospital admissions in a month or the number of patents filed by a company, where whole numbers are the only possibility.

Continuous variables can take an infinite number of values within a given interval. Temperature, weight, and height are continuous because they are obtained through measurement rather than counting, and they can include fractions and decimals. A person’s height could be 165 centimetres, 165.4 centimetres, or 165.42 centimetres, depending on how precisely you measure it. This is also why continuous data such as body mass, blood pressure, and cholesterol levels require a measuring instrument, while discrete data simply requires a count.

How data is collected: time series, cross-section, and pooled data 🔗

Beyond classifying individual variables, it helps to know how an entire dataset was assembled, because this shapes what kind of analysis is appropriate.

Time series data consists of observations of the same variable recorded at different points in time. Quarterly GDP figures, monthly inflation rates, and daily stock prices are all time series, because the same measure is tracked as time moves forward. The order of observations matters here, since each value is often connected to the one before it.

Cross-section data is collected at a single point in time but across many different subjects. India’s decennial Census, which records population, literacy, and housing data for every household in the country during a fixed enumeration period, is a well-known example. Here, time is held constant while the subjects vary.

Pooled data combines both dimensions. It brings together time series and cross-sectional elements in a single dataset, so you get multiple subjects measured across multiple time periods.

Panel data: a special kind of pooled data 🔗

A specific form of pooled data, called panel data, tracks the exact same cross-sectional units repeatedly over time. If you recorded the GDP of India, Brazil, and Kenya every year from 2015 to 2020, you would have panel data because the same three countries are being followed across the same set of years. This is different from simply combining unrelated cross-sections and time series, because panel data lets researchers study both how a unit behaves over time and how different units compare to each other, in a single framework.

Nominal and ordinal measurement scales 🔗

Even after you know whether a variable is qualitative or quantitative, you still need to know its level of measurement, because this determines exactly how precise the data really is and which statistical operations you can perform on it.

Nominal scale 🔗

The nominal scale classifies data into mutually exclusive categories with no inherent order. The subject a teacher teaches, for instance, History, Mathematics, or Economics, is nominal because none of these categories is “higher” or “lower” than another. Gender, blood group, and mother tongue are nominal for the same reason. You can count how many fall into each category, but you cannot rank or average them meaningfully.

Ordinal scale 🔗

The ordinal scale allows categories to be ranked, but it doesn’t tell you the precise size of the gap between them. Ranking guest speakers at a college seminar as superior, average, or poor is ordinal, since you know the order but not exactly how much better “superior” is than “average.” Class rank, customer satisfaction levels, and income brackets (low, middle, high) are all ordinal measurements for the same reason: order exists, but consistent, measurable distance between categories does not.

Interval and ratio measurement scales 🔗

Interval and ratio scales sit higher on the measurement hierarchy because they provide more precise, numerically meaningful information than nominal or ordinal data.

Interval scale 🔗

The interval scale has meaningful, equal differences between units, but it lacks a true zero point. Temperature measured in Fahrenheit or Celsius is the standard example: the difference between 20°C and 30°C is the same as the difference between 30°C and 40°C, but 0°C doesn’t mean “no temperature,” it’s simply an arbitrary reference point. IQ scores work the same way. A score of 0 wouldn’t mean “no intelligence,” which is precisely why you can add and subtract interval data but cannot form meaningful ratios from it.

Ratio scale 🔗

The ratio scale has everything the interval scale has, plus a true, meaningful zero. This is the highest level of measurement, and it’s what allows you to form valid ratios between two values. Height, weight, and income are all ratio variables. Zero income genuinely means no income, and because of that true zero, it’s accurate to say that someone earning 200 units is earning exactly twice as much as someone earning 100 units. That kind of comparison simply isn’t possible with interval or lower-level data.

Why this classification actually matters 🔗

None of this is just vocabulary for an exam. Every statistical technique you’ll encounter later, from calculating an average to running a regression, comes with assumptions about the type of variable and scale of measurement it needs. Treat ordinal data as if it were ratio data, and your averages will be misleading. Treat continuous data as discrete, and you’ll lose precision that could matter for a real decision. Getting this classification right at the start is what makes every conclusion that follows trustworthy.

What do you think? Think about the last form you filled out, whether for college admission, a survey, or an online purchase. Which of the questions on it were nominal, and which were asking for ratio-level data? And if you were designing that form yourself, would you have collected any of that information differently to make it more useful for analysis?

References

1. https://gradcoach.com/nominal-ordinal-interval-ratio/
2. https://pmc.ncbi.nlm.nih.gov/articles/PMC5958489/
3. https://www.mayo.edu/research/documents/data-types/doc-20408956
4. https://censusindia.gov.in/census.website/en/data
5. https://analystprep.com/cfa-level-1-exam/quantitative-methods/time-series-data-vs-cross-sectional-data/
6. https://www.statology.org/levels-of-measurement-nominal-ordinal-interval-and-ratio/