MCAT content library / Research & data / Cross-sectional and longitudinal studies: separate age, cohort, and change
Recover who was measured and when before interpreting an age difference, a changing mean, or a claim about development.
About 14 minutes, including the worked examples and checks.
Useful foundations
On this page
- Concept: a difference between people is not yet a change within a person
- Pattern: identify age, cohort, and period
- Application: read two designs without forcing them to agree
- Rule: compare the same people before interpreting a changing mean
- Use a stronger comparison to test the explanation
- Check your reasoning
- Continue learning
- Sources
Concept: a difference between people is not yet a change within a person
A cross-sectional study measures different participants at one period. A longitudinal study measures the same participants repeatedly. This distinction determines what a time-related comparison can show. If adults aged 20, 40, and 60 complete an attention task today, the study compares age groups. It has not observed the 20-year-olds become 40-year-olds. If the same people complete the task twenty years later, their paired observations describe change over that interval.1
Neither design automatically establishes a mechanism. A cross-sectional age difference can reflect age-related processes, differences between birth cohorts, differential selection, or some combination. A longitudinal change can reflect aging, a change in the historical period, practice with the task, or events that occurred during follow-up. The design clarifies what was observed; the causal explanation still requires evidence.
Pattern: identify age, cohort, and period
Age is the participant’s position in the lifespan. A birth cohort groups people born during a defined period, who may share exposures such as schooling or technology. A period effect is a change associated with the time of observation, such as a new testing platform introduced for everyone. These possibilities are linked: at a fixed date, age determines birth cohort; for one cohort followed forward, age increases as calendar time advances. Collecting more observations does not magically make the factors independent.2
Suppose older participants perform worse on a computer-based task. The observation might fit a change in attention, a cohort difference in experience using the interface, or both. A second measure that reduces interface demands could help distinguish those explanations. Merely relabeling the result an “aging effect” adds a conclusion the comparison has not supplied.
Application: read two designs without forcing them to agree
All data and scenarios on this page are original teaching examples. Higher attention scores indicate better performance. One investigation samples three age groups in the same year. Another follows a separate set of participants from age 20 to age 40, retaining the same scoring method.
| Age group | Mean score |
|---|---|
| 20 years | 80 |
| 40 years | 70 |
| 60 years | 60 |
| Observation | Mean score |
|---|---|
| Age 20 | 80 |
| Age 40 | 76 |
The cross-sectional difference between ages 20 and 40 is −10 points. The longitudinal change among retained participants is −4 points. These results answer different questions and need not conflict. The first compares people with different ages and cohort histories at one date. The second follows a particular cohort through two periods. It would be incorrect to replace the paired −4 with the cross-sectional −10 simply because the age interval is the same.
The paired result supports “mean performance declined by four points among those observed at both times.” It does not by itself show that biological aging caused the decline or that every participant declined by four points. Some participants could improve while others decline. A mean change should be interpreted alongside the distribution of individual changes, the testing conditions, and the people missing from follow-up.
Rule: compare the same people before interpreting a changing mean
Attrition is loss of participants during follow-up. It can change the composition of the observed group. Consider 100 participants at baseline: 60 have scores of 90 and 40 have scores of 60. Their mean is (60 × 90 + 40 × 60)/100 = 78. At follow-up, 40 of the higher-scoring participants and 10 of the lower-scoring participants remain. Assume, for this illustration, that no person’s score changes.
| Group | Baseline count | Retained count | Score at both observations |
|---|---|---|---|
| Higher baseline score | 60 | 40 | 90 |
| Lower baseline score | 40 | 10 | 60 |
The follow-up mean is (40 × 90 + 10 × 60)/50 = 84. Comparing the original sample’s 78 with the follow-up sample’s 84 suggests a six-point improvement even though nobody improved. The retained participants had a baseline mean of 84 as well. Their paired mean change is zero. The problem is the change in who contributes to the average, not a calculation error.
To detect this threat, ask for baseline characteristics of retained and lost participants, retention rates by relevant subgroup, and the method used to handle missing outcomes. Similar observed baseline scores cannot prove that the missing outcomes are harmless: people may differ on unmeasured causes of later performance. More initial recruitment increases the eventual number observed, but it does not necessarily remove selective dropout.1
Use a stronger comparison to test the explanation
Following several birth cohorts across overlapping ages can reveal whether a pattern repeats across groups or changes with the observation period. For example, if separate cohorts all show a score drop immediately after a testing-platform change, a period-linked explanation deserves attention. If the same age-related pattern appears under stable measurement in multiple cohorts, the explanation becomes more consistent with an age-related process. Such designs improve the available comparisons; they still require explicit assumptions to disentangle age, cohort, and period.
Practice effects deserve the same care. Repeating a familiar task may improve scores even when the underlying ability is stable or declining. Alternate forms or a comparison group without prior task exposure can address that specific explanation. A randomly assigned intervention embedded within a longitudinal study addresses a different question: whether the intervention changes the trajectory under the trial conditions.
On an MCAT passage, annotate each observation with three labels: who, when, and whether the same people were measured again. Then choose the conclusion matching that unit of comparison. A statement about group differences, a statement about within-person change, and a statement about why that change occurred are three distinct claims.
ACTIVE PRACTICE · LONGITUDINAL RESEARCH
Change Who Remains—Without Changing Anybody’s Score
The original cohort starts with 60 participants scoring 90 and 40 scoring 60 (mean 78). At follow-up, 40 higher-scoring and 10 lower-scoring participants remain; no individual’s score changed.
Retained sample: participants. Follow-up cross-sectional mean: . Difference versus all original participants: .
Mean within-person change among retained participants: 0.0 points in this explicit no-change model.
Teal: high initial score; gold: low initial score. The original 100-person baseline mean remains 78 regardless of the retention sliders.
Check your reasoning
Choose an answer before opening its feedback. Each option explains the reasoning behind that choice.
Check 1
In the first application, which conclusion is directly supported by the paired follow-up data?
A. All participants lost four attention points as they aged.
Review this choice. A four-point mean change does not imply that every individual changes by that amount.
B. Attention necessarily declines ten points between ages 20 and 40.
Review this choice. Ten points is the cross-sectional contrast between different people, not the paired change.
C. Biological aging caused the entire decline in the retained cohort.
Review this choice. Period changes and practice or other influences remain possible; the paired comparison alone does not isolate biological aging.
D. The retained participants had a mean score four points lower at follow-up.
Correct. The same retained participants have means of 80 and 76, supporting a mean change of −4.
The principle to carry forward
Match the conclusion to the observed unit: the same participants at two times.
Check 2
In the selective-retention example, what is the mean within-person change among retained participants?
A. −6 points
Review this choice. No retained participant changed score under the stated assumption. The −6 value reverses an inappropriate comparison of different sample compositions.
B. 0 points
Correct. The retained group scored 84 at both observations, so its mean paired change is zero.
C. +6 points
Review this choice. The apparent +6 compares all 100 participants at baseline with only 50 retained participants later.
D. +14 points
Review this choice. The retained average of 84 should not be compared with the unrelated cross-sectional age-40 value of 70.
The principle to carry forward
A changing sample average can reflect changing membership even when every observed individual remains stable.
Check 3
Several cohorts show an abrupt score decrease in the same calendar year, when the task switches to a new interface. Which follow-up most directly tests the measurement-based period explanation?
A. Compare old and new interfaces in randomly assigned participants from each age group.
Correct. Randomizing interface exposure within age groups creates a direct comparison of the proposed measurement change while limiting age imbalance.
B. Increase the number of participants tested only with the new interface.
Review this choice. A larger new-interface sample can estimate that condition more precisely but supplies no direct old-interface contrast.
C. Replace each participant with a different person of the same age next year.
Review this choice. Replacing people changes the design and does not isolate the effect of the interface.
D. Report the change separately for participants with high baseline scores.
Review this choice. Stratification by baseline score may reveal heterogeneity, but it does not directly compare the two measurement interfaces.
The principle to carry forward
Test the proposed source of change with a comparison that actually varies that source.
Continue learning
Related lessons
- Sampling and generalizability: trace who enters the estimate
- Reading graphs: axes, scales, and the comparison that matters
Browse research & data · MCAT content library
MCAT preparation and roadmap · Aim High, Score Higher
Sources
1. OpenStax Psychology 2e — Approaches to Research
2. Statistics Canada — Gaining and Losing Literacy Skills Over the Lifecourse