Panel Data
Also known as: longitudinal data, repeated observations.
Observations of the same units, such as products, firms or people, repeated over time, so change can be studied within each unit.
Synthetic dataExamples on this page use synthetic data created for illustration.
Definition
Panel data follow the same units, such as products, firms or people, across several points in time. Because each unit is observed repeatedly, you can study change within a unit and separate it from stable differences between units.
In plain terms
Three shapes of data are easy to confuse. A cross-section looks at many units once. A time series looks at one unit many times. A panel looks at many units many times. Only a panel lets you ask “what changed for product X?” separately from “how do products differ?”.
Panels bring their own problems: units join and leave, so the group measured in week 8 is not the group measured in week 1 (see Attrition Bias), identifiers must stay stable, and the way of measuring must stay the same.
Example
Three products, prices over three weeks. Product C is withdrawn after week 1.
| Product | Week 1 | Week 2 | Week 3 |
|---|---|---|---|
| A | 10 | 10 | 12 |
| B | 20 | 20 | 20 |
| C | 30 | — | — |
| Average of the products present | 20 | 15 | 16 |
The average falls from 20 to 16, which looks like a 20% price drop, yet no price fell: A rose 20%, B did not move, and C simply left. The change is a change in which products are present, not in prices.
Common misunderstandings
- “Panel data just means a lot of data.” It means the same units observed repeatedly.
- “A missing week is a zero.” It is a missing observation (see Missing Data).
- “The average across units each week measures change.” When the set of units changes, the average moves even if every unit stands still.
References
- Cheng Hsiao. Analysis of Panel Data . Cambridge University Press, 2014. ISBN 978-1107657632.
- Jeffrey M. Wooldridge. Econometric Analysis of Cross Section and Panel Data . MIT Press, 2010. ISBN 978-0262232586.