Home/Insights/Methodology reference/Longitudinal research design
Longitudinal research design
Repeated measurement on the same people. It can separate change within a person from differences between people — provided the people stay.
What repetition buys
Measuring the same individuals at multiple points allows each person to act as their own comparison, which removes stable individual differences as a source of confounding. It supports claims about trajectory, timing and order that a single snapshot cannot.
It also distinguishes age effects from cohort effects: a difference between 40-year-olds and 70-year-olds measured once could be ageing or could be the era each group grew up in. Following people resolves it.
Attrition is the governing constraint
Participants drop out, and they rarely drop out at random. Those who leave are often those doing worst, which biases the remaining sample toward better outcomes and can reverse a finding.
Report attrition at every wave, compare leavers with stayers on baseline characteristics, and state the assumption your analysis makes about missingness. Complete-case analysis assumes data are missing completely at random, which is usually implausible; mixed models and multiple imputation assume missing at random, which is weaker but still an assumption that should be named.
Analysis has to respect the structure
Repeated observations on one person are correlated. Analysing them as independent inflates the sample size and produces confidence intervals that are too narrow and p-values that are too small.
Mixed-effects models and generalised estimating equations exist for this. Which is appropriate depends on whether the question is about individuals or population averages.
Other threats
Testing effects, where repeated measurement changes the thing measured. Time-varying confounding, where a confounder is also affected by earlier exposure, which standard adjustment handles incorrectly.
In practice
Fix, flag or query.
The distinction that keeps an edit defensible: what an editor may change silently, what must be shown to the author, and what only the author can answer.
- Report: attrition at each wave, and how leavers differed.
- Never: analyse repeated measures as independent observations.
- State: the missing-data assumption your model relies on.
Reporting standards
- STROBE Statement, cohort checklist
- Reporting standards for longitudinal and repeated-measures analysis
Reporting guidelines named here are hosted by the EQUATOR Network. Writing to the checklist for your design before submission is the cheapest available improvement to a methods section.
Related