Panel Study: Long-Term Data for Long-Term Insight

A panel study is a type of longitudinal study that collects data from the same people (the panel) at several points in time, called waves. Because each person is measured more than once, researchers can see who changed, by how much and in what order, instead of comparing different people at different times.

Example: a company surveys the same 600 customers when they sign up, after six months and after a year, and tracks how each customer's satisfaction and usage change.

Panel study vs other research designs

A panel study is also different from an online panel, a pool of people who have agreed to take surveys. Drawing a fresh sample from an online panel for each survey is a trend design, not a panel study.

Well-known panel studies

• Panel Study of Income Dynamics (PSID), United States: following the same families since 1968.
• German Socio-Economic Panel (SOEP): running since 1984.
• Understanding Society, United Kingdom: started in 2009 and absorbed the earlier British Household Panel Survey (from 1991).

In business, panel designs are used to follow the same customers through onboarding, the same employees through a reorganization, or the same users before and after a product change.

What a panel study shows that a snapshot can't

Suppose 480 customers answered both waves of a satisfaction survey. In wave 1, 300 of them were satisfied (62.5%); in wave 2, 340 were (70.8%). Two separate snapshots would show only a net gain of 40 people.

The panel shows more: 70 customers moved from not satisfied to satisfied and 30 moved from satisfied to not satisfied. In total 100 people changed their answer, and you can look at what happened to each group between the waves, such as a support case, a price change or a new feature they started using.

This is the main advantage: individual change, the order of events, and the ability to control for stable traits of each person (fixed-effects models do exactly that).

Limitations

• Panel attrition: people drop out between waves, and if those who leave differ from those who stay, results become biased.
• Panel conditioning: answering the same questions repeatedly can change how people answer, or even how they behave.
• Locked-in questions: to compare waves you must keep wording and scales the same, even when you would like to improve them.
• Time and cost: results about change arrive only after several waves.

How to run a panel study

1. Define what change you want to measure and how often it is likely to happen. That sets the interval between waves: weeks for a product launch, years for careers.
2. Recruit the panel from your target population, ideally with probability sampling, and recruit more people than you need to allow for attrition.
3. Get consent for repeated contact and explain how answers will be linked across waves (see informed consent).
4. Run the baseline wave and freeze the core questions and scales.
5. Link each person's answers with a stable ID, not with names in the data file.
6. Keep people engaged between waves: reminders, short surveys, sharing what you learned.
7. Analyze change per person: transition tables, change scores, fixed-effects or growth-curve models.

Common mistakes

• Calling repeated surveys of different people a panel. Without the same respondents, you can't measure individual change.
• Losing the link between waves because no ID was stored, or because the ID changed format.
• Rewording questions between waves, which makes change in answers impossible to separate from change in the question.
• Comparing wave averages that include different people. Compare the people who answered both waves, and report attrition separately.

Running a panel study with online surveys

Give each panelist a personal link that carries their ID, for example ?id=P0457, and capture it as a hidden variable (from the Basic plan). Each wave's export then has the same ID column, and the waves can be joined in a spreadsheet or statistics package. Use a code instead of a name or email, and keep the key that maps codes to people separately from the answers. Tell panelists their answers are confidential and linked by a code rather than calling the study anonymous.