Variables in research: 5 types explained with qualitative and quantitative examples

A variable in research is any characteristic, attribute, or condition that can take on different values across participants, settings, or time points. The 5 main types are: independent variables (the presumed cause), dependent variables (the measured outcome), control variables (held constant to isolate effects), confounding variables (hidden factors that distort results), and moderating/mediating variables (which shape how or when effects occur). In qualitative research, participant attributes such as gender, role, or sector serve a similar function to experimental variables.

What is a variable in research?

A variable is simply something that varies. In a study, variables are the characteristics, conditions, or attributes you observe, manipulate, or measure in order to understand a relationship or phenomenon.

In a clinical trial testing whether a new drug reduces blood pressure, the drug dose is a variable. So is the patient's baseline blood pressure, age, smoking status, and diet. Some of these you control deliberately; others you measure and monitor; others may confuse your results if you overlook them.

The concept of "variable" originates in quantitative research, where you can assign numerical values to characteristics (blood pressure measured in mmHg, income measured in pounds sterling) and analyse relationships statistically. But variables also appear in qualitative research, though in a different form. We will come back to this distinction throughout the post.

Understanding which type of variable you are dealing with is one of the most practically useful skills in research design. Get it right and you know what to measure, what to hold constant, and what might be quietly distorting your findings.

What are the 5 main types of research variables?

The five types most researchers encounter are: independent, dependent, control, confounding, and moderating/mediating. Here is how they differ:

Variable type	Role in the study	Question it answers	Example
Independent	The presumed cause; what you manipulate or observe	"What changes?"	Teaching method (traditional vs. flipped classroom)
Dependent	The measured outcome	"What do we measure?"	Student test scores
Control	Held constant to isolate the relationship	"What do we keep the same?"	Class size, socioeconomic background
Confounding	An unmeasured third factor that distorts results	"What might be misleading us?"	Student prior attainment
Moderating/mediating	Shapes when or how the effect occurs	"For whom, or through what mechanism?"	Student motivation (moderator); increased engagement (mediator)

Independent variables

The independent variable (IV) is the factor you manipulate or vary to observe its effect on something else. It is the presumed cause in a cause-and-effect relationship.

In an experiment, you actively control the IV. In an observational study, you observe it without manipulating it. Either way, it is the variable you treat as the input.

Examples:

A nutritionist studying weight loss randomly assigns participants to a low-carbohydrate or low-fat diet. The diet type is the IV.
An HR team compares employee satisfaction between departments with flexible working arrangements and those without. Flexible working is the IV.
A psychologist tests whether sleep deprivation affects cognitive performance by having some participants sleep 8 hours and others sleep 4 hours. Sleep duration is the IV.

In quantitative vs qualitative research, the independent variable is most explicitly defined in experimental quantitative studies. In qualitative research, the concept is looser, but it still has an analogue, as we will discuss below.

Dependent variables

The dependent variable (DV) is the outcome you measure. It "depends" on what happens with the independent variable.

Examples:

In the diet study, weight loss in kilograms after 12 weeks is the DV.
In the HR study, scores on a validated job satisfaction survey are the DV.
In the sleep study, reaction time and accuracy on a cognitive task are the DVs.

A single study can have multiple dependent variables. A researcher studying a new classroom intervention might measure both academic performance and student wellbeing as separate DVs.

Control variables (extraneous variables)

Control variables are factors that might influence your dependent variable but are not the focus of your study. You hold them constant (or account for them statistically) to ensure they do not confuse your results.

In the diet study, a researcher might control for:

Baseline body weight (by measuring it before the study begins)
Physical activity levels (by asking all participants to maintain their usual exercise routine)
Age and sex (by matching participants across conditions)

Without controlling these factors, it becomes impossible to say whether any weight loss differences between groups are caused by the diet or by other characteristics of the participants.

Control variables are sometimes called extraneous variables, particularly when they have not yet been identified or controlled. Once you identify an extraneous variable and actively control it, it becomes a control variable.

Confounding variables

A confounding variable is more troublesome than a simple extraneous variable. A confounder is related to both your independent variable and your dependent variable, which means it can produce a spurious association: a relationship that looks real but is actually produced by the confounder.

The classic example: studies once found that people who carry lighters are more likely to develop lung cancer than people who do not. But carrying a lighter is not a cause of lung cancer. Smoking is the confounder: it causes lung cancer, and it also explains why someone would carry a lighter. The lighter-cancer association is entirely explained by the confounding effect of smoking.

In research, confounders are particularly dangerous because they can make you believe that X causes Y when the real cause is Z. Methods for managing confounders include:

Randomisation: randomly assigning participants to conditions so confounders are distributed evenly across groups
Stratification: analysing results separately within subgroups (e.g. smokers and non-smokers)
Statistical control: including confounders as covariates in regression models

Identifying likely confounders before you begin is a core part of good research design.

Moderating and mediating variables

Moderating and mediating variables both describe the relationship between an IV and a DV, but they do so in fundamentally different ways.

A moderating variable changes the strength or direction of the relationship between IV and DV. It answers the question: "For whom, or under what conditions, does this effect occur?"

Example: A study finds that social media use is associated with reduced wellbeing. But this relationship is stronger for adolescents than for adults. Age moderates the relationship: the same IV has different effects depending on the moderator.

A mediating variable explains the mechanism through which the IV affects the DV. It sits inside the causal chain: X causes M, and M causes Y.

Example: A management training programme (IV) increases team productivity (DV). The mediator is improved communication skills: the training improves communication (M), and better communication drives productivity (Y). Without the mediating variable, you know the effect exists but not why.

The distinction matters for how you design your study and what conclusions you can draw. Moderators tell you about boundary conditions; mediators tell you about mechanisms. Both are worth identifying when you are reviewing literature or planning data collection.

How do variables work in quantitative vs qualitative research?

In quantitative research, variables are typically numerical or categorical and are defined before data collection begins. A survey might measure age (numerical), employment status (categorical), and job satisfaction (rated on a 1-10 scale). Statistical analysis then tests relationships between these variables.

The logic is relatively linear:

Identify your IV and DV
Define control variables
Identify potential confounders
Collect data in a structured form
Run statistical tests

In qualitative research methods, the picture is different. Most qualitative research is inductive rather than deductive: you begin with open-ended data collection and allow concepts to emerge from the data. You are not usually testing a hypothesis about the relationship between two predetermined variables. Instead, you are trying to understand the texture and meaning of people's experiences.

This means that the formal language of "independent variable" and "dependent variable" does not map neatly onto most qualitative work. A thematic analysis of 30 interviews about career transitions is not testing whether one variable causes changes in another. It is exploring what career transitions feel like, what shapes them, and what meanings participants attach to them.

And yet, variables are not absent from qualitative research. They appear in two important ways.

What are variables in qualitative research specifically?

Participant attributes as variables

The most common form of variables in qualitative research is participant attributes: the background characteristics of the people you have interviewed or observed. These include:

Demographics: age, gender, ethnicity, socioeconomic background
Role and experience: job title, years of experience, seniority level
Context: organisation type, sector, country or region
Situational factors: whether a participant experienced a specific event (redundancy, illness, promotion)

These attributes function similarly to independent variables in quantitative research. You might not have set them up experimentally, but they can explain meaningful differences in what participants say.

For example, in a study of healthcare worker burnout, you might find that participants who work in A&E describe a qualitatively different kind of pressure than those in outpatient departments. The work setting is acting as an explanatory variable, even though you never randomised participants to settings or manipulated anything.

This is what researchers often call purposive sampling: deliberately recruiting participants from specific categories (by role, experience level, or organisational context) so that you can observe whether those attributes matter. See our guide on qualitative research sample size for more on how to structure purposive samples.

Emergent variables in qualitative data

In qualitative research, variables can also emerge from the data rather than being predetermined. Through coding qualitative data, you may identify attributes that distinguish one set of participants from another, attributes you had not anticipated before you began.

A researcher studying how startups approach failure might begin coding for themes and discover that "how founders talk about early mentorship" consistently separates those who recover and try again from those who exit the sector entirely. Mentorship quality becomes a variable that was not in the original research design but emerges as a meaningful explanatory factor.

This emergent quality is one of the distinctive strengths of qualitative data. The researcher is not limited to variables they thought to measure in advance.

The key difference

In quantitative research, variables are defined a priori and measured precisely. In qualitative research, participant attributes may be planned (purposive sampling), but interpretive variables tend to emerge from the data itself. The mixed methods research approach often bridges both logics: collecting qualitative data to understand mechanisms while using quantitative variables to test whether those mechanisms operate at scale.

How to use participant metadata as variables in qualitative analysis

This is where the concept of research variables becomes practically useful for qualitative researchers who are moving beyond manual analysis.

When you collect qualitative data across multiple participants, each transcript or document carries a set of attributes: who the participant is, what context they were in, when the interview took place. If you treat these attributes as structured metadata, you can use them to systematically compare what different groups of participants are saying.

For instance, in a study of employee experience across 60 interviews, you might attach the following metadata to each transcript:

Department (HR, Engineering, Sales, Customer Success)
Tenure (under 1 year, 1-3 years, 3-5 years, 5+)
Seniority (individual contributor, manager, director, executive)
Location (UK office, US office, remote)

Once themes are identified across the full dataset, you can then filter and cross-tabulate by these metadata variables to ask: do junior employees talk about career development differently from senior ones? Are burnout themes concentrated in particular departments or tenures?

This kind of analysis is described in detail in our post on discovering themes using metadata variables. In Skimle, metadata fields are attached to each document at the point of upload, and the platform then lets you slice your thematic analysis by any combination of those fields, effectively giving you the equivalent of a pivot table for qualitative insights. You can read more about adding metadata to your documents and analysing your data by metadata in the Skimle documentation.

This approach is particularly powerful when your research question is inherently comparative: you want to understand not just what themes appear, but which participant attributes explain where and why they appear. If you work in academic research, our page on how Skimle fits academic workflows explains how this works in practice.

The analogy with experimental variables is useful here. In a quantitative experiment, you might control for participant role by including it as a covariate. In structured qualitative analysis with metadata, you do something conceptually similar: you ensure that role (or tenure, or sector) is recorded consistently so you can interrogate it analytically, rather than having it be an invisible source of noise in your data.

Common mistakes researchers make with research variables

1. Confusing correlation with causation

The most common error is treating a correlation between an IV and DV as evidence of causation without accounting for confounders. Two variables moving together does not mean one causes the other. The spurious relationship between carrying lighters and lung cancer is a famous example, but the same logic applies in social science, business research, and policy analysis.

Always ask: what other factor could explain both the IV and the DV? That is your potential confounder.

2. Failing to operationalise variables clearly

"Employee satisfaction" is not a variable. "Score on the Utrecht Work Engagement Scale, version 3" is a variable. Operationalisation means specifying exactly how you will measure a concept. Without this, different researchers may measure different things and arrive at contradictory findings.

This applies in qualitative research too. If your research question involves "resilience" or "leadership quality", you need to think carefully about what evidence in your data would count as an instance of each construct.

3. Ignoring moderating variables

A finding that holds "on average" across a sample may not hold for specific subgroups. If your intervention increases productivity in large teams but decreases it in small teams, a single average effect masks the more important moderating effect of team size. Reporting the average without investigating moderation can lead to misguided policy decisions.

4. Treating participant attributes as background noise

In qualitative research, demographic and contextual attributes are often reported in a table and then largely ignored in the analysis. This is a missed opportunity. If sector, tenure, or role consistently explains differences in what participants say, those attributes are analytically important variables, not just demographic metadata.

5. Over-controlling in quantitative research

Including too many control variables (particularly variables that are themselves consequences of your IV) can introduce bias. This is known as "collider bias" or "over-adjustment." Not every variable that correlates with your DV needs to be controlled for. Constructing your variable model from a clear causal theory, rather than throwing in every available covariate, produces more reliable results.

Frequently asked questions

What is the difference between a variable and a constant in research?

A variable takes on different values across participants, time points, or conditions. A constant stays the same. In an experiment where all participants receive the same questionnaire instructions, the instructions are a constant. The responses to the questionnaire are variables. Constants are often deliberately introduced to remove a source of variation; control variables are constants you have created by design.

Can qualitative research have independent and dependent variables?

Qualitative research rarely uses the language of independent and dependent variables because most qualitative work is exploratory rather than hypothesis-testing. However, the underlying logic applies. A participant attribute (such as organisational role) that you use to compare groups of respondents functions similarly to an independent variable, and the themes or experiences you find in each group function as the "outcome" being explained. In qualitative work, this comparison is interpretive rather than statistical.

What is the difference between a confounding variable and a control variable?

A control variable is one you have deliberately held constant or measured and adjusted for. A confounding variable is one you have either not measured or failed to account for. The same factor can be a potential confounder (if ignored) or a control variable (if addressed). Good research design aims to identify confounders in advance and convert them into control variables through measurement and adjustment.

What is the easiest way to remember the difference between moderators and mediators?

A useful mnemonic: moderators are about the circumstances (for whom, when, under what conditions), while mediators are about the mechanism (through what process, why). Moderators sit beside the relationship between IV and DV; mediators sit inside it. If you remove the mediator, the effect should weaken or disappear. If you vary the moderator, the size or direction of the effect changes.

How do participant metadata variables work in qualitative analysis tools like Skimle?

Metadata variables in tools like Skimle are structured attributes attached to each document in your project: things like department, interview date, participant role, or country. Once you have run your thematic analysis across the full dataset, you can filter and cross-tabulate by these attributes to see whether specific themes are concentrated in particular participant groups. This functions as a qualitative analogue to subgroup analysis in quantitative research. See adding metadata and metadata analysis in the Skimle documentation for practical setup guidance.

Want to bring structure to your qualitative data analysis? Try Skimle for free and experience how structured metadata and AI-assisted coding can turn a pile of transcripts into clear, auditable findings, with every insight traceable back to the source.

Continue reading:

About the authors

Henri Schildt is a Professor of Strategy at Aalto University School of Business and co-founder of Skimle. He has published over a dozen peer-reviewed articles using qualitative methods, including work in Academy of Management Journal, Organisation Science, and Strategic Management Journal. His research focuses on organisational strategy, innovation, and qualitative methodology. Google Scholar profile

Olli Salo is a former Partner at McKinsey & Company where he spent 18 years helping clients understand the markets and themselves, develop winning strategies and improve their operating models. He has done over 1000 client interviews and published over 10 articles on McKinsey.com and beyond. LinkedIn profile