Retour
Longitudinal Survey: Types, Examples and How to Run One

Matthieu Saussaye

A longitudinal survey collects data from the same people, or the same defined group, at several points in time to measure how things change. Instead of one snapshot, you get a sequence. That lets you see who changed, by how much, and what happened first, which a one-off survey cannot show.
The essentials:
Repeated measurement of the same variables, at planned intervals called waves.
Three main designs: panel (same individuals), cohort (people sharing a starting experience, such as a sign-up month), and trend (fresh samples from the same population, technically a repeated cross-section).
Why it matters: it separates individual change from population change and establishes time order, which is the first condition for any causal claim.
The main cost: attrition. People drop out between waves, and the ones who leave are rarely random.
Market research uses: customer panels, onboarding cohorts, brand trackers, employee journeys, product diaries and churn studies.
Panel, cohort and trend surveys compared
"Longitudinal" is used loosely. These are the designs people mean, and they answer different questions.
Design | Who is measured each wave | Shows individual change? | Typical market research use | Main risk |
|---|---|---|---|---|
Panel | Exactly the same people | Yes | Customer or employee panel measured quarterly | Attrition, panel conditioning |
Cohort | People who share a starting event (same people, or a sample of them) | Yes, if the same people are followed | Users who signed up in January, tracked at 1, 3 and 12 months | Cohort effects confused with time effects |
Trend (repeated cross-section) | A new sample from the same population | No, population-level only | Annual brand health tracker | Sample differences between waves |
Rotating panel | Part of the sample is replaced each wave | Yes, for the overlap | Large trackers that want both trend and individual change | More complex weighting |
Retrospective | One sample asked about past points | Approximate only | Quick look at how a decision unfolded | Recall error |
If you only need population trends, a repeated cross-section is cheaper and avoids attrition. Choose a true panel when the question is about individuals: do the customers who were unhappy in Q1 leave by Q3?
Famous longitudinal surveys, and what they teach
Long-running public studies show both the power and the cost of the design. Figures below are from the list of longitudinal studies on Wikipedia:
Study | Design | Started | Sample | Lesson for market research |
|---|---|---|---|---|
National Longitudinal Survey of Youth 1979 (NLSY79), US | Cohort | 1979 | 12,686 respondents born 1957-64 | A single cohort, followed long enough, explains careers and income in a way no snapshot can |
Panel Study of Income Dynamics (PSID), US | Household panel | 1968 | Listed at about 70,000 individuals | Following households (and their new members) keeps a panel representative as families split |
British Household Panel Study, UK | Household panel | 1991 | 5,500 households, about 10,000 individuals | Modeled on PSID; shows panel designs travel across countries |
1970 British Cohort Study, UK | Birth cohort | 1970 | About 17,000 | Fix the cohort definition precisely: here, babies born in one week of April 1970 |
The US General Social Survey shows the hybrid route: a repeated cross-section since 1972 that, per its Wikipedia entry, added rotating panel samples between 2008 and 2014. Many commercial trackers do the same: a fresh sample for the trend line, plus a re-contacted subset for individual change.
When a longitudinal survey is the right choice
You need to know who changed, not just whether the average moved. A stable 70% satisfaction score can hide 15% of customers getting happier and 15% getting unhappier.
Time order matters. Did the drop in perceived value come before the cancellation, or after the price increase?
You track a journey. Onboarding, first purchase, renewal. See also our guide to customer satisfaction metrics for which scores to track at each stage.
You evaluate an intervention. Measuring the same people before and after a product change, a training program or a campaign gives a cleaner comparison than two unrelated samples.
When not to: if the decision is due next month, if the population turns over quickly (event visitors), or if a single well-designed cross-section answers the question.
How to run a longitudinal survey in 7 steps
State the change you want to measure. "Change in perceived value between month 1 and month 6 for new subscribers" defines the population, the variable and the timing.
Choose the design. Panel, cohort or trend, using the table above. Decide now whether new members will join over time.
Set the wave schedule. Match the interval to how fast the thing changes. Weekly for an onboarding diary, quarterly for satisfaction, annually for brand attitudes.
Lock a core questionnaire. Keep core questions word-for-word identical across waves, in the same order and with the same scales. A changed wording breaks the series. Add rotating modules for topical questions.
Recruit for retention, not just for wave 1. Explain the commitment upfront, collect more than one contact channel, and over-recruit to absorb dropout. Sourcing options are in how to find survey respondents.
Link records safely. Use a persistent pseudonymous ID so answers can be joined across waves without storing names next to responses. Get consent for re-contact.
Analyze change, then report it. Compare the same individuals across waves (transition tables, change scores, survival curves) and report how attrition might affect the result.
Example core questions for a customer panel
Core questions must stay identical every wave. Keep them few, so later waves stay short and retention stays high. A reusable core set:
Relationship status
"How long have you been using [brand]?" (asked once, then carried forward)
"In the last 30 days, how often did you use [product]?"
"Are you currently using any alternative to [product]?"
Tracking scores
"Overall, how satisfied are you with [product]?" (1-5, labelled)
"How likely are you to recommend [brand] to a friend or colleague?" (0-10)
"How easy is it to get what you need from [product]?" (1-7)
"How likely are you to renew when your plan ends?" (1-5)
Change and reasons
"Compared with the last time we asked, is [product] better, the same or worse for you?"
"What has changed for you since then?" (open)
"What is the one thing we should fix first?" (open)
Rotate topical modules around this core: a pricing module one wave, a new-feature module the next. Ratings follow the conventions in our Likert scale guide; keep the same labels and the same number of points across every wave.
Attrition and other longitudinal pitfalls
Attrition is the defining weakness. The Wikipedia overview of longitudinal studies notes that they are time-consuming and expensive and cannot avoid an attrition effect, and that a shrinking sample biases what remains. In customer research this is acute: the least engaged customers are the first to stop answering, so later waves look happier than reality.
Pitfall | What it does to your data | Mitigation |
|---|---|---|
Attrition | Remaining sample skews engaged and satisfied | Short waves, incentives per wave and a completion bonus, reminders, weight for dropout, report retention per wave |
Panel conditioning | Being surveyed changes behavior or answers (people notice the brand more) | Compare with a fresh cross-sectional sample; rotate part of the panel |
Instrument drift | Small wording changes create fake trends | Freeze core questions; test changes in parallel before switching |
Cohort vs period confusion | You cannot tell if change comes from age, tenure or the calendar | Track several cohorts started at different times |
Mode changes | Switching from phone to online shifts answers | Keep mode constant, or run both in parallel for one wave |
Analyzing longitudinal survey data
The point of the design is individual change, so analyze at the individual level first:
Transition tables. Cross-tab wave 1 answer against wave 2 answer for the same people. A table showing how many promoters became passives or detractors (see our Net Promoter Score guide) is far more actionable than two NPS figures.
Change scores. Wave 2 minus wave 1 per respondent, then compare the change between segments.
Survival analysis. Time until an event such as churn or first repeat purchase, and which earlier answers predict it.
Lagged models. Does wave 1 satisfaction predict wave 2 spending once wave 1 spending is controlled for? This is the closest a survey gets to causal evidence without an experiment.
Example: a 3-wave onboarding panel
A software company surveys each new customer cohort at day 7, day 30 and day 90. Core questions: ease of setup (1-7), perceived value (1-5), likelihood to renew (0-10), and an open question, "What nearly made you give up?" At day 90 it joins answers with renewal data. The value is in sequences: customers who rated setup low at day 7 but recovered by day 30 behave differently from those who never recovered, and the open answers explain what the recovery depended on.
Follow the same people without losing them
Longitudinal studies fail on two things: respondents stop answering, and open questions return too little to explain the change you measured. SmartInterview keeps each wave short and conversational, by voice or text and in the respondent's language, and an AI follows up when an answer is vague, so a short wave still yields a reason. Open answers are coded into the same themes wave after wave, which makes change in the "why" comparable over time, not only change in the scores.
Use it for a recurring customer panel, an onboarding cohort or an employee pulse. Start free.
Frequently Asked Questions
What is a longitudinal survey in simple terms?
It is a survey that asks the same people, or the same defined group, the same questions several times over a period. Comparing the answers shows how opinions or behaviors change, and in what order.
What are the three types of longitudinal studies?
The three types usually listed are panel, cohort and retrospective studies. Panel studies follow the same individuals; cohort studies follow people who share a starting experience; retrospective studies look back using existing records or recall. Trend studies on fresh samples are often grouped with them.
What is the difference between longitudinal and cross-sectional surveys?
A longitudinal survey measures the same people repeatedly; a cross-sectional survey measures different people once. Longitudinal designs show individual change and time order. Cross-sectional designs are faster and cheaper but only describe one moment.
How long does a longitudinal survey last?
As long as the change you study takes. Market research panels often run from a few weeks (onboarding diaries) to a few years (customer panels). Public research cohorts such as NLSY79, started in 1979, have run for decades.
What is the biggest disadvantage of a longitudinal survey?
Attrition. Participants drop out between waves, and those who leave tend to differ from those who stay, which biases later results. Cost and the time to get results are the other main drawbacks.
Can a longitudinal survey prove causation?
Not on its own, but it gets closer than a snapshot. It establishes that one thing happened before another and lets you control for earlier values. Confounding factors can still explain the link, so experiments remain the stronger test.

