Why Study Design Depends on How Participants Are Selected
Suppose researchers want to understand whether a health condition is related to a past exposure, or whether a school program is related to later achievement. The variables alone do not identify the study design. A key question is how people entered the study: were they sampled at one point in time, grouped by exposure and followed for an outcome, or selected because they already had the outcome? The answer determines which comparisons the design can support.
All three designs in this tutorial are observational: researchers record exposures and outcomes as they occur or have occurred, rather than assigning participants to receive an exposure. As in the earlier material on observational studies and randomized experiments, a measured association by itself does not show that an exposure caused an outcome. Here the focus is how to recognize three ways of organizing an observational study.
In these definitions, an exposure is a characteristic, experience, or circumstance whose relationship to an outcome is being studied. It need not be a harmful substance; it could be an educational program or a health behavior. The outcome is the event or condition of interest, such as developing an illness or reaching a learning benchmark. The study design is determined by the sampling and measurement plan, not by whether the exposure sounds beneficial or whether measurements are collected using records, surveys, or examinations.
Three Designs, Three Directions of Selection
| Design | How participants are selected | Exposure and outcome information | A central limitation or use |
|---|---|---|---|
| Cross-sectional | Sampled from a defined population without selection based on the outcome or exposure | Both are measured at or around one time | Describes how common characteristics are at that time; often cannot establish which came first |
| Cohort | People are identified as a group, with exposure status determined before outcome follow-up | Outcome occurrence is observed over a period | Can compare outcome risks when the groups and follow-up denominators are known |
| Case-control | Selected based on outcome status: cases have the outcome, controls do not | Earlier exposure status or history is compared between the groups | Can be efficient for uncommon outcomes; the sampled case fraction is set by selection, not population risk |
A quick classification method is to ask first whether the researchers selected people because they had the outcome. If so, the design is case-control. If not, ask whether an identified group is observed for new outcomes over a period; if so, it is a cohort study. If exposure and outcome are assessed together in a population sample, it is cross-sectional. These are guiding questions: the actual procedures matter more than a study's label.
Cross-Sectional Studies: A Snapshot
A cross-sectional study takes a snapshot of a population at a specified time or during a short, defined period. Researchers may record both a possible exposure and an outcome for every sampled person. For example, they might survey students this month about current participation in tutoring and administer a reading assessment during the same period.
Because participants are not selected based on whether they have the outcome, the study can describe the proportion of the sampled population with that outcome at the time of measurement. This is often called prevalence: the proportion of people in a defined population who have a condition at a specified time. A cross-sectional study can also show whether the measured exposure and outcome vary together. But if the measurements do not establish which came first, it may be unclear whether the exposure preceded the outcome. Students with weaker reading skills, for example, might be more likely to receive tutoring, rather than tutoring necessarily preceding the difference in skills.
Worked Example: A One-Time Student Survey
A hypothetical district selects 300 students from its middle schools in October. On the same week, researchers record whether each student currently attends an after-school study group and whether the student meets a reading benchmark. Of the 300 students, 180 meet the benchmark. Classify the design, and calculate the sample proportion meeting the benchmark.
Classification. The students are sampled from the district without being selected according to their benchmark status or study-group attendance. Both the possible exposure (study-group attendance) and the outcome (meeting the benchmark) are measured in the same period. This is a cross-sectional study.
The observed proportion meeting the benchmark is
Thus, 60.0% of the sampled students met the benchmark in October. This describes the sample at that time; extending the estimate to all district students would require the sample to represent that population. The design can compare benchmark proportions for students with and without study-group attendance, but the simultaneous measurements do not by themselves show that attendance came before achievement.
Cohort Studies: Start With a Group, Observe Outcomes
In a cohort study, researchers identify a group of people and determine their exposure status, then observe the group for a defined period to see which participants develop the outcome. A cohort might consist of students entering a school year, patients enrolled in a health system, or workers at a set of workplaces. Participants are not selected because they already have the outcome of interest; the outcome is recorded as follow-up proceeds.
If the number of people who develop the outcome and the number initially at risk are known for each exposure group, researchers can calculate each group's risk over the specified follow-up period. Here, risk means the proportion of people initially at risk who develop the outcome during that period. Different follow-up durations or incomplete outcome information can complicate comparisons, so the time period and who remains under observation must be considered.
Worked Example: Following a Health Cohort
In a hypothetical study, researchers enroll 200 adults without a particular illness. At enrollment, 80 report regular exposure to workplace dust and 120 do not. Over the next two years, 12 adults in the exposed group and 6 in the unexposed group develop the illness. Classify the design and calculate the two observed risks.
Classification. The study begins with adults who do not have the outcome, records exposure status, and observes subsequent illness over two years. Participants were not chosen because they developed the illness, so this is a cohort study.
For the exposed group, the observed two-year risk is
For the unexposed group, it is
In this hypothetical cohort, 15.0% of exposed adults and 5.00% of unexposed adults developed the illness over two years. These calculations use the number developing the outcome divided by the number in the corresponding group who were followed from enrollment. The observed difference is an association; because exposure was not assigned by the researchers, these results alone do not establish that dust caused the illness.
Cohort studies are not limited to research that begins collecting data before outcomes occur. Researchers can sometimes use existing records to identify an earlier cohort and determine what outcomes followed. The defining feature for this tutorial is the cohort structure: exposure status identifies the groups, and the outcome is determined for those groups over a period. Whether information was collected prospectively or assembled retrospectively is a separate distinction.
Case-Control Studies: Start With the Outcome
A case-control study begins by selecting people who have the outcome (cases) and people who do not (controls). Researchers then determine whether, or how often, each group had a particular exposure in the past. For example, cases might be patients with a rare disease, while controls are patients without it. The comparison asks whether the exposure history differs between these outcome-defined groups.
This design can be practical when an outcome is uncommon: researchers can find cases directly instead of following a very large population in the hope that enough people develop the condition. However, the numbers of cases and controls are determined by the study's selection plan. Consequently, the fraction of sampled participants who are cases is not generally the outcome risk in the source population. A direct risk calculation needs a meaningful denominator representing the population at risk, which the case-control sample does not provide merely by counting its selected cases and controls.
Worked Example: Looking Back From a Rare Outcome
A hypothetical research team selects 60 patients with a rare illness as cases and 120 patients without it as controls. From records, the team finds that 24 cases and 18 controls had a specified medication exposure before the relevant time. Classify the study and compare the exposure proportions.
Classification. Participants were selected according to illness status, and their earlier medication exposure was then assessed. This is a case-control study.
Among cases, the sample exposure proportion is
Among controls, it is
The sampled cases had a higher proportion with the recorded exposure: 40.0%, compared with 15.0% of sampled controls. This is a comparison of exposure histories within the selected groups. It does not mean that 40.0% of exposed people in the population developed the illness: the study deliberately selected 60 cases and 120 controls, so its case fraction is set by design.
A Short Justification: Why Selection Changes What Can Be Calculated
The distinction between cohort and case-control designs follows from their denominators. In a cohort, the researchers know how many people with each exposure status were initially followed. If 12 of 80 exposed people develop an outcome, the observed risk is \(12/80\). The starting group of 80 supplies the denominator for that exposure-specific risk.
In a case-control study, by contrast, the researchers choose a number of cases and controls. Suppose they select 60 cases and 120 controls: the resulting case fraction is \(60/(60+120)=1/3\). That fraction comes from the sampling plan. Changing the plan to select 60 cases and 240 controls would change the sample fraction to \(60/(60+240)=1/5\), even if the source population had not changed. Therefore the sample case fraction cannot, by itself, estimate population risk.
This does not make case-control studies uninformative. Comparing exposure histories between cases and controls can reveal an association worth investigating. It means only that the study's outcome-based sampling must be respected when deciding which quantities the sample can directly estimate.
Classifying Ambiguous Scenarios
Study descriptions sometimes use words such as “follow-up,” “records,” or “survey,” which are not enough to determine the design on their own. Focus instead on the selection rule and the timing of the exposure and outcome information.
If people were selected because they have or do not have the outcome, classify the study as case-control.
If researchers determine exposure status and then observe outcome occurrence over a period, classify it as a cohort study.
If a population sample is assessed around one time without outcome-based selection or outcome follow-up, classify it as cross-sectional.
A school district might use attendance records to identify students who missed many days in an earlier term and compare their later graduation status. If it selects students by earlier attendance and determines later outcomes for that group, the structure is cohort-like. If instead it selects graduates and nongraduates first, then compares their attendance histories, it is case-control. The same two variables can appear in both studies; changing the rule used to select participants changes the design.
Common Mistakes and Why the Design Matters
- Calling every study using past records case-control. Records may be used in a cohort study if researchers identify an earlier group by exposure status and determine later outcomes. The selection rule, not whether data are old, is decisive.
- Calling every study that measures two variables cross-sectional. A study that follows an exposure-defined cohort and records later outcomes is not a one-time snapshot, even if it measures both variables.
- Treating the case fraction in a case-control sample as population risk. The researchers choose how many cases and controls to include. Use the design's selected groups to compare exposure histories, not to infer risk from the case fraction alone.
- Assuming a cohort study proves causation. Cohorts establish the order of exposure and subsequent outcome more clearly than simultaneous measurements do, but observational exposure groups may differ in other ways. A cohort association is not, by itself, proof of a causal effect.
- Assuming a cross-sectional association reveals direction. If exposure and outcome are recorded together, the study may not show which occurred first. Describe the observed relationship without claiming an unsupported temporal sequence.
Check Your Understanding
For each scenario, identify the design and briefly explain which feature determines your answer.
- A health department surveys a population sample once and records both current smoking status and current asthma status. What design is this, and what timing question might remain unclear?
- Researchers enroll people without a disease, group them by a measured workplace exposure, and record new diagnoses over five years. What design is this?
- A research team selects patients with a particular injury and patients without it, then compares their earlier use of a safety device. What design is this?
- Why can a cohort study calculate an observed outcome risk from its exposure-group denominators, while the case fraction in a case-control sample does not directly estimate population risk?
- A school study selects students by whether they met a benchmark, then compares earlier tutoring histories. What design is this, and why?