01 · The Question
Do You Need to Compare Something to Answer Your Research Question?
Researchers sometimes begin designing a study and immediately look for two groups: users versus non-users, intervention versus control, online versus face-to-face, high performers versus low performers.
But what if the research question does not actually require a comparison?
A study can describe a population, estimate a prevalence, explore experiences, examine an association, develop a prediction model, or investigate a process without necessarily dividing participants into groups. Conversely, some questions cannot be answered meaningfully without comparing conditions, groups, time points, or alternative exposures.
The issue is not whether comparison makes a study look more rigorous. It is whether the comparison is necessary to answer the question you are asking.
02 · The Short Answer
Comparison Is Necessary Only When the Question Depends on a Contrast
In Brief
Your research question requires a comparison when answering it depends on estimating or interpreting a difference between meaningful conditions, groups, exposures, interventions, or time points.
If your objective is instead to describe, estimate, characterize, explore, or model something without needing such a contrast, adding a comparison may be unnecessary. The question should determine the design, rather than the design convention determining the question.
03 · What You Need to Know
Start With the Information Needed to Answer the Question
Almost any study can contain comparisons if you look hard enough. That does not mean every possible comparison deserves to become part of the research design.
A more useful approach is to ask: what evidence would constitute an answer to the research question?
Some questions are inherently comparative
Consider the question:
Do students taught using method A achieve higher examination scores than students taught using method B?
The phrase “higher than” makes the contrast explicit. An examination score from method A alone cannot answer the question because the claim is defined relative to method B.
The same logic applies to questions involving concepts such as:
higher or lower than;
more or less effective than;
better or worse than;
different from;
increased or decreased relative to another condition;
changed from an earlier time point.
These questions require some form of comparison because the quantity of interest is itself a contrast.
Many important research questions do not require comparison groups
Now consider:
What proportion of university students use generative AI for academic writing?
The researcher may estimate the proportion from an appropriate sample. There is no inherent need to divide students into two groups simply to make the study comparative.
Similarly, a researcher could ask how teachers experience the implementation of an institutional AI policy, what barriers researchers encounter when sharing data, how accurately a model predicts student attrition, or how frequently a particular practice occurs.
These are legitimate research questions. Their quality does not depend on manufacturing a second group.
Research question
Is an explicit comparison required?
Why?
What proportion of students use generative AI for coursework?
Not necessarily
The target is a prevalence or proportion.
How do doctoral students experience the peer-review process?
Not necessarily
The target is an experience or phenomenon rather than a difference between predefined groups.
Is AI literacy associated with responsible AI use?
Not necessarily as separate groups
The target is an association between variables.
Can learning analytics predict course completion?
Not necessarily as treatment groups
The primary target is predictive performance.
Do students receiving intervention A perform better than those receiving intervention B?
Yes
The target is a difference between conditions.
Did achievement improve after the intervention compared with an appropriate alternative?
Yes
The claim depends on a contrast across conditions and potentially time.
Association does not always require dividing people into groups
Researchers sometimes assume that examining a relationship means categorizing participants. A continuous variable is then split into “high” and “low” groups simply to enable a group comparison.
That may discard useful information and is not automatically necessary.
If your question concerns whether study time is associated with examination performance, for example, both variables can remain quantitative. The analysis can examine how they vary together rather than artificially classifying students as “high-study” and “low-study.”
This distinction follows from the broader difference among correlation, association, prediction, and causation . A relationship between variables and a difference between predefined groups are not synonymous research objectives.
Causal questions usually contain a comparison, even when it is not obvious
Suppose the question is:
Does providing automated formative feedback improve student achievement?
There is no explicit “compared with” in the sentence, but the question is still comparative. “Improve” means achievement under the feedback condition must be considered relative to achievement under some relevant alternative.
That alternative might be no feedback, usual teaching practice, another feedback method, delayed feedback, or some other well-defined condition.
This reflects the counterfactual logic of causal inference. Asking whether an intervention caused an outcome means asking what would have happened under an alternative condition. The comparison may be implicit in the wording, but conceptually it is still there.
A comparison does not automatically mean you need a conventional control group
Once researchers recognize that a question requires comparison, another assumption often follows: there must be an intervention group and a control group.
Not necessarily.
The appropriate comparison depends on the question. You might compare two active interventions, different naturally occurring exposure groups, repeated measurements from the same participants, different periods, different settings, or outcomes against another scientifically justified reference.
This is why deciding whether you actually need a control group is a separate methodological decision.
Between-group comparison is only one form of comparison
Imagine that you want to determine whether a brief training program changes participants' research-data-management knowledge.
One design could compare participants who receive the training with different participants who do not. Another could measure the same participants before and after training. More sophisticated designs could combine between-group and repeated measurements.
These designs answer related but not necessarily identical questions and make different assumptions. Choosing between between-subjects and within-subjects comparisons should therefore follow from the estimand, intervention, outcome, possible carryover or period effects, and practical constraints.
A before-and-after difference is a comparison, but it may not isolate the cause of change
A single-group pretest-posttest study compares the same group's outcome before and after an intervention. It can therefore establish whether the measured outcome changed over that interval.
What it cannot automatically establish is why the change occurred.
Other events, maturation, changes in measurement, regression toward the mean, secular trends, or other influences may also produce change. A contemporaneous comparison condition can sometimes help researchers distinguish an intervention-related change from changes that would have occurred anyway.
A baseline measurement and a comparison group therefore solve different design problems. Having one does not automatically substitute for the other.
The comparison must correspond to the claim you want to make
“Does it work?” is incomplete unless “work compared with what?” has a defensible answer.
An intervention compared with no intervention addresses a different question from the same intervention compared with established practice. Similarly, comparing a new teaching method with an unusually weak alternative may demonstrate a difference without answering whether the new method is preferable to what educators would realistically use instead.
Comparator selection is therefore substantive, not decorative. Methodological guidance on intervention trials emphasizes that different comparators address different research questions, and comparator choice can influence both the interpretation and usefulness of the findings.
04 · A Practical Example
When Adding a Second Group Changes the Question Rather Than Improving It
Hypothetical Example
Studying faculty members' use of generative AI
A researcher wants to investigate how frequently university faculty use generative AI for teaching activities and which teaching tasks they use it for.
Original question How do faculty members use generative AI in their teaching? The study needs evidence about patterns and forms of use. A suitably designed descriptive study could answer this without constructing a second group.
An unnecessary comparison The researcher decides that the study needs to look “more rigorous” and divides participants into faculty aged 40 and below versus those older than 40. The age comparison was not required by the original question and has now introduced a different research objective.
A genuinely comparative question If theory or prior evidence instead motivates the question “Do patterns of generative AI use differ between early-career and senior faculty?”, then comparison becomes integral to the question. The researcher must now define those groups appropriately and consider factors that could complicate interpretation of any observed difference.
A causal question If the question becomes “Does AI-literacy training increase responsible generative AI use?”, a meaningful alternative condition is needed because the researcher wants to attribute a change to the training rather than merely describe participants after receiving it.
The lesson is subtle but important. Comparison does not inherently strengthen the first question. It creates another question.
That additional question may be worthwhile, but it should have a substantive rationale rather than being added because comparative analyses appear more sophisticated.
06 · What This Means for You
Let the Research Question Tell You Whether a Contrast Is Needed
Before adding groups, conditions, or time points, complete a simple sentence:
To answer my research question, I need to know ______.
If the blank can be completed with a single quantity, description, experience, pattern, relationship, or predictive performance measure, you may not need an explicit group comparison. If the answer necessarily involves a difference between alternatives, you do.
A simple decision framework
If your question asks what exists, how often it occurs, or how participants experience something
Do not add a comparison unless it serves a separate, substantively justified question.
If your question asks whether variables are related
Determine whether the relationship can be examined directly without artificially converting variables into groups.
If your question asks whether groups, conditions, treatments, settings, or time points differ
Build the required comparison explicitly into the design.
If your question asks whether an intervention or exposure causes an outcome
Define the relevant alternative condition and consider whether the design can support that causal contrast.
Once you determine that a comparison is necessary, the next question is not simply “Where can I find another group?” You need to determine what the comparator should represent, how units enter the conditions, whether the groups differ before the study, and whether those differences threaten the interpretation.
Those choices often matter more than the mere presence of a comparison.
07 · A Quick Checklist
Before Adding a Comparison to Your Research Design
Before creating groups or conditions, check:
Can you state exactly what quantity, experience, relationship, prediction, difference, or effect your research question asks you to estimate?
Would the research question remain answerable if there were only one group or condition?
If you need a comparison, can you explain what each condition represents substantively?
Is the proposed comparison motivated by the research question rather than by the availability of convenient demographic categories?
Have you avoided categorizing continuous variables solely to manufacture comparison groups?
If you are making a causal claim, have you defined the relevant alternative or counterfactual condition?
If you are comparing existing groups, have you considered whether they differ in consequential ways before the exposure or intervention?
Can you explain how the chosen comparison allows you to answer the original question rather than a different one?
11 · Cite this Guide
How to Cite This Guide
This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.
Recommended (Field Guide)
APA
MLA
Chicago
Copy Citation
