03 · What You Need to Know
The Research Topic Can Stay the Same While the Question Changes
Start by Separating the Topic From the Research Question
A topic identifies the general phenomenon you want to investigate. A research question specifies what you want to know about that phenomenon.
“Generative AI and student learning” is a topic. “Is generative AI use associated with academic self-efficacy among university students?” is a research question. “Does within-student generative AI use change across the first year of university, and are those changes associated with subsequent changes in academic self-efficacy?” is another.
The topic remains similar. The evidence required to answer the questions does not.
This distinction explains why researchers should not choose cross-sectional or longitudinal design solely from the topic. Ask what relationship to time is built into the question.
A Cross-Sectional Version Usually Asks About What Exists at a Defined Period
Suppose the broad question concerns academic stress and social-media use among doctoral students.
A cross-sectional version might ask:
“Is frequency of social-media use associated with academic stress among doctoral students during the current academic year?”
The researcher measures the variables within a defined period and examines their relationship. Depending on sampling, the study may also estimate the prevalence of particular characteristics or compare groups.
The design can provide useful evidence about the pattern that exists. It does not directly reveal whether an individual's social-media use changed before that person's stress changed, or whether either variable increased or decreased over time.
A Longitudinal Version Makes Time Part of the Question
The same topic could produce a longitudinal question:
“How do social-media use and academic stress change across doctoral training, and is earlier social-media use associated with subsequent stress?”
Now the researcher needs repeated observations. The question explicitly concerns change and temporal ordering, neither of which can be observed directly from one cross-section.
Longitudinal designs commonly use repeated measurements to investigate change over time, and those observations within the same participant are correlated. Appropriate analysis therefore needs to recognize that repeated values from one person are not independent observations.
The Design Changes the Meaning of “Difference”
Imagine that final-year doctoral students report greater academic confidence than first-year students in a cross-sectional survey.
You can state that the groups differed at the measurement period. You cannot directly conclude that doctoral students generally become more confident as they progress through their programs.
Why? The first-year and final-year groups contain different people. They may differ because of selection, attrition, prior experience, cohort effects, changes in admissions, different institutional conditions, or other factors.
If the same students are followed from first year to final year, researchers can examine within-person change directly. That longitudinal evidence answers a different question.
Cross-sectional comparison
Are different people or groups different at the measured period?
Longitudinal comparison
Do the same people, units, or defined populations change across measured periods?
The Design Also Changes What “Association” Means
Suppose AI use and academic confidence are positively associated in a cross-sectional dataset. That association describes how the two measured variables vary together within the study's temporal frame.
A longitudinal study can ask additional questions. Does earlier AI use predict later confidence after accounting for relevant prior measurements? Do changes in AI use correspond with changes in confidence? Are the trajectories different for different groups?
These are not simply more sophisticated analyses of the same cross-sectional question. They are different temporal formulations of the relationship.
Longitudinal Design Can Clarify Temporal Ordering but Does Not Automatically Establish Causality
If an exposure is measured before an outcome, the temporal order is clearer than when both are measured simultaneously. This can strengthen causal reasoning because a proposed cause should precede its effect.
It does not follow that the earlier variable caused the later one. Confounding, selection, time-varying factors, measurement error, and alternative causal structures may remain.
The broader distinction among experimental, quasi-experimental, and observational research still determines how exposure or intervention status arises and which causal interpretations are defensible.
Cross-Sectional Research May Be Exactly What the Question Requires
The ability to study change does not make longitudinal research inherently superior.
Suppose a university needs to know the current prevalence of faculty generative AI use to plan professional development for the next semester. Following the same faculty members for three years would not improve the answer to that immediate descriptive question. It would answer additional questions that decision-makers did not necessarily ask.
Cross-sectional research can also be valuable for identifying patterns worthy of subsequent investigation, estimating current conditions, comparing groups, and examining associations efficiently.
A simpler design is not methodologically deficient merely because a more elaborate design exists.
Longitudinal Research Is Worthwhile When the Added Temporal Information Matters
Longitudinal research becomes particularly valuable when the question contains concepts such as change, development, trajectory, incidence, persistence, transition, prediction over time, delayed outcomes, or temporal sequence.
These concepts cannot usually be answered adequately by a single cross-sectional measurement.
However, longitudinal research also introduces attrition, missing observations, participant burden, repeated-measurement effects, longer study duration, and more complex analysis. The decision should therefore depend on whether temporal information materially changes the answer.
The Same Question Wording Can Hide Different Inferential Intentions
Consider the apparently simple question:
“What is the relationship between research self-efficacy and publication productivity among academics?”
That wording does not specify time. A cross-sectional study might measure current self-efficacy and recent publication productivity concurrently. A longitudinal study might measure self-efficacy at baseline and publication outcomes during subsequent years.
The resulting evidence would support different interpretations even though the research question appears almost identical on paper.
This is why research questions should communicate important temporal relationships when those relationships matter. Words such as “subsequent,” “over time,” “change,” “during follow-up,” or explicit measurement periods can clarify what evidence is required.
Measurement Timing Should Follow the Question, Not Be Added Later
Once the longitudinal version of a question is chosen, researchers still need to decide how many observations are required and when they should occur.
Two measurements can show change between two observed occasions. More measurements may be needed to characterize trajectories, distinguish temporary from sustained patterns, or identify pre-existing trends. The appropriate schedule depends on what the timing of data collection needs to reveal.
Simply converting a one-time survey into “longitudinal research” by repeating it at an arbitrary later date may not provide the temporal evidence the revised question actually requires.
Cross-Sectional and Longitudinal Are Not the Only Dimensions That May Change
Moving from a cross-sectional to a longitudinal design can create additional design decisions. Will data be collected prospectively or reconstructed from existing records? Will the same participants be followed? How will attrition be handled? Are exposures observed or assigned? How many sites are needed?
The distinction between prospective and retrospective research, for example, addresses a different temporal issue. A longitudinal dataset may be created prospectively through follow-up or retrospectively from records containing repeated observations.
Research design is therefore better understood as a set of interacting decisions than as a single label.
04 · A Practical Example
One Topic Can Produce Several Legitimate Research Questions
Hypothetical Example
Generative AI use and academic self-efficacy among university students
A researcher is interested in whether students' use of generative AI is related to their academic self-efficacy. The broad topic does not dictate a single temporal design.
Cross-sectional question Is frequency of generative AI use associated with academic self-efficacy among university students during the current semester?
What it establishes The study can estimate the concurrent relationship between measured AI use and self-efficacy within the defined period, subject to the sampling, measurement, and analytical limitations of the design.
Longitudinal question Is generative AI use measured at the beginning of the academic year associated with subsequent changes in academic self-efficacy across the year?
What it adds Repeated measurements permit researchers to investigate within-student change and establish that the earlier measurement preceded later measured outcomes.
The longitudinal study is not simply the cross-sectional study with extra questionnaires. The inferential target has changed from a concurrent association to a temporally structured relationship involving subsequent change.
Now consider a different objective: “What proportion of students currently use generative AI at least weekly?” For that question, longitudinal follow-up may add little. The cross-sectional design is not a compromise. It is the design that matches the question.
