03 · What You Need to Know
Description Is a Research Purpose, Not a Preliminary Version of “Real” Research
A research question tells you what the study seeks to learn. Sometimes the answer involves an association between variables. Sometimes it involves a difference between groups. Sometimes the aim is prediction or causal inference.
And sometimes the thing you genuinely need to know is simply what is happening.
Methodological literature explicitly recognizes descriptive research questions as a form of quantitative research question alongside comparative and relationship questions. Descriptive questions seek to describe characteristics, behaviors, phenomena, or other features of a population rather than necessarily examining relationships among variables.
This distinction is important because a good research question should follow the knowledge gap rather than a hierarchy in which every study is expected to progress toward an “effect.”
What does a descriptive research question ask?
A descriptive question asks for an account or estimate of a phenomenon without necessarily asking why it occurs or whether it is related to something else.
In quantitative research, descriptive questions might ask:
- What proportion of undergraduate students use generative AI for academic writing?
- How frequently do university students use AI tools when preparing assignments?
- What are the levels of academic self-efficacy among first-year students?
- What types of cybersecurity practices are used by small businesses?
Each question can be answered with numerical evidence, but none inherently requires a relationship between an independent and dependent variable.
This is consistent with published methodological guidance that distinguishes descriptive quantitative questions from comparative and relationship questions.
Descriptive does not mean “no variables”
A descriptive quantitative study may measure one or many variables. What makes the research question descriptive is not the number of variables in the dataset but what the study asks you to do with them.
Suppose a survey collects students' age, year level, AI use, academic self-efficacy, study habits, and GPA. The existence of several measured variables does not automatically make the research question relational.
If the question is:
“How frequently do undergraduate students use generative AI for academic writing?”
the primary purpose is descriptive.
If the question becomes:
“Is frequency of generative AI use associated with GPA among undergraduate students?”
the purpose has changed. You are now asking about a relationship.
Descriptive question
What exists, how much, how often, how common, or what characteristics are present?
Relationship question
Are two or more measured characteristics associated, and if so, how?
Prevalence questions are descriptive
One of the clearest examples is prevalence.
“What proportion of first-year university students report food insecurity?” is a meaningful empirical question. Answering it requires decisions about the target population, sampling, measurement, and analysis, but it does not require the researcher to identify a predictor of food insecurity.
The resulting estimate may be valuable for needs assessment, resource planning, policy development, hypothesis generation, or subsequent analytical research.
Adding a relationship simply because the study “needs two variables” can distort the purpose. If the knowledge gap concerns prevalence, prevalence is what the research question should ask.
Descriptive questions can characterize distributions, not just percentages
Description extends beyond asking how common something is. A study may describe a distribution, pattern, profile, behavior, practice, characteristic, or trend.
For example:
“How much time do undergraduate students spend using generative AI for academic tasks during a typical week?”
The answer might include measures such as the mean, median, range, or distribution of reported usage. Another study might describe the types of academic tasks for which students use AI or the frequency with which different uses occur.
Those analyses remain descriptive until the research question asks whether the observed characteristics differ across groups, relate to another variable, predict an outcome, or can be attributed to some cause.
Descriptive studies can still compare categories informally, but formal comparisons change the question
Suppose you report AI-use frequencies separately for first-, second-, third-, and fourth-year students. Stratified descriptive results can help readers understand the sample or population.
But if the research question asks:
“Do rates of generative AI use differ by year level?”
you now have a comparative question. The analysis must address that comparison appropriately rather than merely present separate percentages.
This distinction matters because researchers sometimes begin with a descriptive objective and gradually accumulate exploratory comparisons until the manuscript appears to answer questions that were never clearly formulated in advance.
A well-defined research question helps determine what data should be collected and analyzed, and methodological guidance cautions against developing the primary question retrospectively from patterns found in the data.
Description can be the necessary evidence before explanation
Researchers sometimes dismiss descriptive studies because they do not explain why a phenomenon occurs. Yet explanation can be difficult to pursue responsibly when the basic contours of the phenomenon are unknown.
Before asking why students abandon an educational technology, for example, it may be useful to know how frequently abandonment occurs, when it occurs, which features are commonly discontinued, and what usage patterns look like.
Description can therefore establish an empirical baseline, reveal patterns worth investigating, identify neglected populations, support subsequent hypothesis development, or show that an assumed problem is less common than expected.
That does not make every descriptive study important. A descriptive question still needs relevance. Collecting easily obtainable statistics without a meaningful knowledge gap can produce a perfectly answerable study that contributes very little. The distinction between answerability and contribution remains part of deciding whether a research question is worth turning into a study.
Descriptive research is not automatically simple research
“Just describe it” can conceal substantial methodological difficulty.
If you want to estimate the prevalence of generative AI use among university students, whom exactly do you want the estimate to represent? How will students be sampled? What counts as generative AI use? Does checking grammar count? What about generating an outline? What period should students recall? Will nonresponse systematically distort the estimate?
A descriptive statistic can be easy to calculate. Producing a defensible description of a target population may not be.
This is particularly important for survey and cross-sectional research. STROBE, the major reporting guideline for observational epidemiological studies, covers cross-sectional studies and emphasizes transparent reporting of what was planned, done, found, and interpreted. The STROBE initiative also explicitly notes that its recommendations concern reporting rather than prescribing how observational studies must be designed or conducted.
A descriptive question should specify what is being described
“What is happening with AI in universities?” is technically seeking description, but it is far too open to determine what evidence would answer it.
A more useful question might be:
“What proportion of undergraduate students report using generative AI for assessed written assignments?”
or:
“What academic tasks do undergraduate students report using generative AI to complete?”
The question should identify the population or cases and the phenomenon or characteristic sufficiently for the study to be designed. This does not require placing every methodological detail in the sentence. The broader issue of which components belong explicitly in a research question depends on what information is necessary to remove consequential ambiguity.
Descriptive quantitative questions do not require a hypothesis
If the study is genuinely asking for an unknown quantity or distribution, there may be no meaningful directional hypothesis to test.
Consider:
“What percentage of university students use generative AI at least weekly for academic work?”
Inventing the hypothesis “More than 50% of students use generative AI” adds a threshold unless there is a theoretical, practical, or prior-evidence reason why 50% matters. The descriptive estimate may be the result the study actually needs.
Published guidance distinguishes research questions from hypotheses and recognizes descriptive quantitative questions as a legitimate question type. Hypotheses are formal predictions about expected outcomes or relationships and are not required merely to give every research question a predicted answer.
Descriptive does not mean causal
A descriptive study can reveal striking patterns. Those patterns do not automatically explain themselves.
Suppose a survey finds that students who report frequent generative AI use also report lower confidence in independent writing. A purely descriptive study might report the distributions of both characteristics. If researchers formally examine their association, the analysis becomes relational.
Even a demonstrated association would not by itself establish that AI use caused lower confidence. Students with lower confidence might be more likely to use AI, another factor might influence both, or the measures might capture something else entirely.
The language of the question should therefore match the inference the study can support. This becomes particularly important when researchers use terms such as “impact,” “influence,” and “effect” in observational questions.
Qualitative research can also be descriptive
“Descriptive” is not exclusively a quantitative label. Qualitative methodological literature recognizes descriptive qualitative questions alongside contextual, explanatory, exploratory, phenomenological, ethnographic, grounded-theory, and other forms.
A qualitative question might ask:
“How do undergraduate students describe their use of generative AI during academic writing?”
The intended answer is not a prevalence estimate. Instead, the researcher seeks a detailed account of practices or experiences.
This is one reason the distinction between qualitative and quantitative research questions should not be reduced to whether the study is “descriptive.” Both can describe, but they may describe using different forms of evidence and for different epistemic purposes.
Descriptive, comparative, relational, and causal questions make different promises
| Question type |
What it asks |
Example |
What the answer can primarily establish |
| Descriptive |
What exists, how common it is, or what it is like |
What proportion of students use generative AI weekly? |
A frequency, distribution, characteristic, pattern, or account |
| Comparative |
Whether groups or conditions differ |
Does weekly AI use differ between undergraduate and graduate students? |
A difference or lack of evidence for a difference under the study design |
| Relational |
Whether measured characteristics are associated |
Is frequency of AI use associated with writing self-efficacy? |
An association and its estimated direction or magnitude |
| Causal |
Whether changing one factor changes an outcome |
Does access to an AI writing assistant affect subsequent writing performance? |
A causal effect only when the design and assumptions support causal inference |
Moving from one row to another is not simply making the study “better.” It changes what the study promises to establish and often changes the design, sampling, measurement, analysis, and assumptions required.
You do not need to add a relationship just to make the question look advanced
A common progression in student research goes something like this: first describe a phenomenon, then compare groups, then correlate variables, then add predictors, because each step appears more statistically impressive.
That sequence is backwards if the research problem only requires description.
Methods should serve the question. If the unanswered question is how common a practice is, a defensible prevalence estimate is more useful than a poorly justified regression model attached to the same dataset.
Research does not earn extra methodological credit for using the largest menu in the statistical software.
Ask whether description is sufficient for the knowledge gap
The strongest reason to move beyond description is not that descriptive research is inferior. It is that description may not answer the actual problem.
If policymakers already know that teacher turnover is high but need to understand which working conditions are associated with leaving, another prevalence estimate may add little. If the prevalence is unknown in the population of interest, however, establishing it could be essential.
The literature review therefore matters. It helps determine whether the unanswered question is descriptive, comparative, relational, explanatory, or something else.
Your question should stop where the knowledge gap stops. Do not ask for an effect when you only need a description, and do not settle for description when the problem genuinely requires evidence about relationships or causation.
