RESEARCH GUIDE
Longitudinal Study Design Guide
Waves, Attrition and Time-Varying Measures
A longitudinal study design measures the same units on several occasions so that change within them can be described and explained. Most of its quality is fixed before the first wave: the number and spacing of occasions, the measures that change over time and the plan for dropout.
A longitudinal study collects repeated measurements on the same people, schools, firms or other units, which lets a researcher separate stable differences between units from change within them. That separation is what a single survey cannot provide, and it is why repeated observations are correlated and need models that respect the clustering of occasions within units. The design is only as good as its schedule and its retention: waves spaced badly miss the process of interest, and attrition that is related to the outcome distorts it. For longitudinal projects, OnlineSPSS, a specialist statistical consultancy for academic, clinical and professional research, works on the wave schedule, the missing-data strategy and the repeated-measures models, while the research question stays with the researcher.
FORMS OF LONGITUDINAL STUDY
Types of Longitudinal Studies by Schedule and Units
The forms differ in how many occasions are measured, how closely they are spaced and what is held constant, and each one suits a different kind of change and a different level of funding.
Panel Studies A panel follows the same sample at fixed intervals, usually with a modest number of waves, and suits questions about stability and change in attitudes, income or health.
Birth and Age Cohorts A cohort is followed from a shared starting event, such as birth or enrollment, which allows long trajectories but requires years of funding and sustained participant contact.
Experiments With Follow-Up A randomized or quasi-experimental study measures participants before and after an intervention and again later, which shows whether an effect persists or fades.
Intensive Longitudinal Designs Diary and momentary assessment studies collect many short measurements per person, which suits processes that change within days and produces data with dense within-person structure.
Accelerated Cohort-Sequential Designs Several age groups are each followed for a short time with overlapping ages, which covers a long age span in a short study but depends on the assumption that the groups share one trajectory.
Record-Based Longitudinal Data Administrative or health records provide repeated observations without recruitment, with wave timing set by when contacts occurred and not by the researcher.
DESIGN COMPARISON
Longitudinal Panels, Repeated Surveys and Time Series
Designs that involve time answer different questions, and the table separates following the same units from sampling anew each round and from observing one aggregate series.
| Feature | Longitudinal panel | Repeated cross-sections | Time series |
|---|---|---|---|
| Units observed | The same units on every wave | Different units each round | One aggregate observed many times |
| Individual change | Estimated within each unit | Not available | Not applicable |
| Typical model | Mixed model, GEE or growth curve | Comparison of independent samples | Regression with serial dependence |
| Number of occasions | A few to many, set by design | Few, set by survey schedule | Usually many |
| Main cost | Retention and repeated contact | New sampling at each round | Long and consistent record needed |
| Main threats | Attrition and mistimed waves | Changing methods between rounds | Autocorrelation and structural breaks |
KEY PLANNING DECISIONS
Choices That Cannot Be Repaired After the First Wave
Once data collection is under way the schedule, the instruments and the contact strategy are difficult to change, so these choices are made while the plan can still be adjusted.
Number and Spacing of Waves Match the interval to the pace at which the process changes, and plan enough occasions for the shape of change that the intended model has to describe.
Choosing the Time Metric Decide whether change is modeled against age, time since baseline or wave number, because the choice shapes both the schedule and how mistimed assessments are handled.
Keeping Instruments Stable Use the same measures with the same wording and response format at every wave, since a changed instrument confounds real change with change in measurement.
Time-Varying and Fixed Measures Identify which variables are recorded once and which at each wave, and capture the date of every assessment so that exposure order can be reconstructed.
A Retention Plan Collect several ways to reach participants, schedule follow-up contact and ask for reasons for dropout, which also gives the information that later missing-data models rely on.
Sample Size for Within-Person Change Power depends on the number of waves, the correlation between a person's repeated measures and the expected loss, so it is calculated for the planned model and not for a single comparison.
COMMON ISSUES
Attrition, Timing and Other Threats Across Waves
Attrition in longitudinal studies is the most frequent threat, but several other problems are built into the data structure, and each is checked in the design before models are fitted.
Selective Attrition Participants who drop out often differ in the outcome or its predictors, so attrition is described by wave and group and its likely effect is examined, not assumed to be random.
Complete-Case Default Analyzing only people with every wave discards information and can bias results, while methods that use all available observations rest on explicit assumptions about why data are missing.
Mistimed Assessments When the real interval between waves differs from the planned one and the analysis treats them as equal, the estimated change rate is distorted for those whose visits slipped.
Practice and Reactivity Repeated testing can improve scores through familiarity, and repeated questioning can change behavior, so some of the observed change may reflect the measurement itself.
Time-Varying Confounding A variable that confounds later exposure may itself be changed by earlier exposure, and ordinary adjustment can then remove part of the effect or introduce bias.
Age, Period and Cohort Mixed In one study the effects of age, calendar period and birth cohort cannot be fully separated without assumptions, so interpretations of trends are stated with that limit.
REPORTING
Reporting Standards for Longitudinal Studies
Reporting should make the schedule and the losses visible, and the checklist depends on whether the study observed people or allocated an intervention.
Observational longitudinal studies in health research are usually reported with the STROBE checklist for cohort designs, randomized trials with follow-up with CONSORT, and panel studies in psychology and the social sciences often with the APA Journal Article Reporting Standards. Whichever standard applies, readers need the number of participants at each wave, the timing of assessments, the reasons for loss and how missing data were handled. A flow diagram or table of retention by wave and group is the clearest way to show attrition.
STATISTICAL METHODS
Methods Used to Analyze Longitudinal Data
Repeated observations on the same units are correlated, so the model depends on the outcome type, the number and timing of waves and the pattern of dropout.
Linear Mixed Models (Multilevel Modeling)
Individual trajectories with random effects
Generalized Estimating Equations (GEE)
Population-averaged repeated outcomes
Latent Growth Curve Modeling (LGM)
Growth as latent intercepts and slopes
Cross-Lagged Panel Models
Reciprocal relations across waves
Missing Data Analysis
Dropout and incomplete waves
Longitudinal & Multilevel Analysis
All models for repeated and nested data
ANALYSIS SERVICES
Statistical Support for a Longitudinal Project
Most longitudinal problems are easier to prevent than to repair, so support usually begins with the schedule and retention plan and continues through modeling, reporting and an outside review.
Research Planning & Study Design
The wave schedule, the time metric, the sample size for change and the missing-data strategy, written down before the first wave is collected.
Quantitative Data Analysis
The mixed, marginal or growth models for your repeated measures, with dropout handled under stated assumptions and change estimated with its uncertainty.
Results Reporting & Interpretation
Retention tables, trajectory figures and methods wording prepared to the checklist your journal requires, with change interpreted for your field.
Independent Statistical Review
Finished repeated-measures models are checked for the time metric, the handling of dropout and mistimed visits, and whether change is read within persons or between them.
FREQUENTLY ASKED QUESTIONS
Questions About Longitudinal Study Design
Waves, dropout and missing occasions raise the most questions in repeated-measures projects, and the last two answers cover confidentiality and the assessment step.
How many waves does a longitudinal study need?
The number of waves depends on the shape of change you want to estimate: two waves give only a difference, while three or more are generally needed to describe a trajectory and more again for nonlinear change. The spacing should reflect the pace at which the process unfolds, and the plan should state the model the waves are meant to support. Extra waves are useful only if the sample is retained.
What is the difference between a panel study and a cohort study?
Both follow the same people. A cohort is defined by a shared starting event and is usually followed for outcomes, while a panel study design samples a population and re-interviews it at regular intervals to track change in many measures. The terms overlap in practice, and what matters analytically is the clustering of repeated observations and the missing-data pattern.
How much attrition is too much?
There is no threshold that makes results safe or unsafe. What matters is whether dropout is related to the outcome and to the variables used to explain it. A study with substantial loss can be informative if the reasons are recorded and sensitivity analyses are run, while a small loss concentrated in one group can still bias a comparison.
Can missing waves be filled in or ignored?
Neither is a default. Methods such as mixed models and multiple imputation use all observed data and are valid when missingness can be explained by what was recorded, an assumption the data cannot confirm, so its consequences are explored through sensitivity analyses. Simple carrying forward of the last value is generally discouraged, because it invents stability that was not observed.
How do you protect my research data and confidentiality?
Longitudinal data link records across time and so can be especially identifying, and your data and project details are handled confidentially. Project information is not published, and researcher names and institutions are not used publicly without permission. An NDA is available where required and can be arranged before sensitive materials are shared. Read more about confidentiality and data security.
What happens when I request an assessment, and how long does it take?
You share your research question, design and relevant materials, such as a protocol, an analysis plan or a dataset, and a statistically trained Research Coordinator reviews the request and responds the same day. The analysis, deliverables, timeline and one fixed project fee are then confirmed before any work begins, so the timeline follows from the agreed scope.
Still have a question? Find more answers in our full FAQ.
NEXT STEP
Talk Through
Your Longitudinal Study
Share the wave schedule, the retention so far and the data or protocol. We confirm how dropout and timing will be handled and agree the scope before any analysis starts.
No Commitment Required
Strictly Confidential
Fixed Project Fee
In-House PhD-Level Statisticians
Same-Day Response
Serving Researchers Since 2014