Types of variables in research: with examples

variablenoun

any characteristic, number or quality that can take different values in a study, such as age, an exam score or a blood pressure reading.

That definition is from Editage's guide to variables, and the key word is different. If every clerk had the same blood pressure, there would be nothing to study.

Variables are sorted in two different ways, and students often mix them up:

• By the role they play in the study: independent, dependent, control, extraneous, confounding, mediating and moderating.
• By the kind of data they hold: categorical or quantitative, and within those, nominal, ordinal, discrete, continuous, interval or ratio.

Every variable has one type of each kind. Blood pressure is Devansh's dependent variable by role, and a continuous ratio variable by data. The role shapes your design; the data type decides your statistics.

Every study that looks at cause and effect has these two at its centre. Editage defines the independent variable as the one a researcher changes or categorises to see its effect on another variable, the presumed cause. The dependent variable is the outcome you measure, the presumed effect.

If you cannot tell which is which, ask: what am I changing, or grouping people by? That is the independent variable. What am I measuring? That is the dependent variable. Then test them in a sentence: "yoga changes blood pressure" makes sense, but "blood pressure changes whether a clerk did yoga" does not.

Naming these two is also the first step in writing a hypothesis, which our guide on how to write a hypothesis explains step by step.

Blood pressure moves for many reasons besides yoga: a cup of tea, a rushed walk from the bus stand, a Monday morning. Editage calls any factor other than the independent variable that could affect the outcome an extraneous variable.

You deal with extraneous variables in one of two ways. You hold them constant, or you measure them so you can allow for them in the analysis. A variable you deliberately hold constant is a control variable: Editage's example is giving every participant the same room, lighting and quiz time.

Dr. Pant asked Devansh to list everything that could move a blood pressure reading. Then they decided what to fix:

1. Same time of day.Every reading taken between 9 and 10 in the morning, before the branch opens.
2. Same machine.One digital monitor and the same size of cuff for everyone.
3. Same routine before the reading.Five minutes of rest, sitting, with no tea in the half hour before.
4. Measure what cannot be fixed.Age, weight and whether a clerk takes BP medicine, recorded at the start.

You cannot make every clerk the same age, but if you record age, you can check later whether it explains part of the result.

A confounding variable is the most dangerous kind of extraneous variable. The York St John University study skills factsheet defines it as a variable that is correlated with the independent variable and affects the dependent variable. Their example: when sales of iced coffee go up, so do cases of heat stroke. Iced coffee is not causing heat stroke. Temperature is pushing both up.

Devansh had a confound in his own plan. He had meant to let clerks choose whether to join the yoga group. Dr. Pant pointed out that clerks who volunteer for morning yoga may already eat less salt and walk more. If their blood pressure falls, is it the yoga, or the kind of person who signs up for yoga?

Random assignment, matching groups on likely confounds, or measuring them and adjusting in the analysis are the usual answers. Which ones you can use depends on your research design. If you can do none of these, say so in your limitations, because a confound left unexplained is exactly what an examiner is likely to ask about.

These two sound alike but do different jobs. Both involve a third variable, and each answers a different question.

The factsheet points out a common slip: people call a mediator a confound. The difference is direction. A confounder changes both the independent and dependent variable. A mediator is changed by the independent variable, and then it changes the dependent one.

If it explains the route, it is a mediator. If it changes the size of the effect, it is a moderator.

Devansh decided to measure stress at the start and end, to test it as a mediator, and to compare clerks under and over 50, to test age as a moderator. If you draw these links as boxes and arrows, you already have the start of a conceptual framework.

Now the second way of sorting: by the kind of data a variable holds. Editage describes qualitative or categorical variables as ones that describe a quality or category, summarised with frequencies or percentages, and quantitative variables as ones expressed in numbers, analysed with statistics such as means and standard deviations.

A quick test: if "half" of it makes sense, it is continuous. Half a kilo makes sense; half a session attended does not.

The psychologist S. S. Stevens set out these four scales in a paper called "On the Theory of Scales of Measurement" in Science in 1946, and they are still the best-known way of sorting data. GraphPad's guide gives a clear summary of each, and of the calculations that make sense at each level.

The difference between interval and ratio is easy to check. GraphPad notes that 10 degrees Celsius is not twice as hot as 5 degrees, because zero on that scale does not mean no heat. A weight of 4 grams, though, is twice as heavy as 2 grams.

Devansh's stress questionnaire used five-point answers, from "never" to "very often". Each answer is ordinal. Statisticians still debate whether a total of many such answers can be treated as interval and averaged, so follow what your field accepts, and say what you did.

You can find an average blood pressure, but not an average blood group. Our guide on qualitative vs quantitative research explains how this fits the wider choice of method.

Dr. Pant's red circle was really asking for one table. This is the order Devansh built it in.

1. Start from your research question.The thing you change or compare is the independent variable; the thing you hope to see change is the dependent variable.
2. List everything else that could move the outcome.Ask your guide and read similar studies. These are your extraneous variables.
3. Decide what to do with each one.Hold it constant, measure it, or design it out with random assignment.
4. Look for mediators and moderators.Include one only if your theory gives you a reason.
5. Write an operational definition for every variable.Say exactly how it will be measured, with the tool and the unit.
6. Note the type of data for each.Nominal, ordinal, interval or ratio. This decides which tests you can run.

Editage calls the fifth step operationalisation: defining a variable in specific, measurable terms. "Stress" is an idea; "the total score on a ten-item stress questionnaire" is a variable. Here is part of Devansh's final table:

This table belongs in your methodology chapter, often under a heading such as "variables of the study". Our guide on how to write a research methodology shows where it fits. Department formats vary, so check yours.

Devansh took the new table to Dr. Pant a week later. There was no "etc." on it. She read it twice and drew a small tick beside every row.