18.1 Overview

Two types of data must be distinguished:

  • Pooled data: Combination of multiple cross-sectional data over time
    • Two or more different observational units over time
    • Grades in an economics class based on students’ concentration combined from multiple semesters
    • American Community Survey (ACS)
  • Panel data: Repeated measurement on the same individual \(i\) over time \(t\).
    • Individual units can be people, states, firms, counties, countries, etc.
    • National Longitudinal Survey (NLSY79): To access the data: Accessing Data > Investigator > Begin searching as guest.
    • Necessary adjustments of standard error due to correlation across time.

There are some necessary assumptions about linear panel models.

  • Regular time intervals
  • Errors are correlated
  • Parameters may vary across individuals or time
  • Intercept: Individual specific effects model (fixed or random)

Note that the General Social Survey (GSS) is not a panel data set because different respondents are questioned every year. Besides the National Longitudinal Survey mentioned above, here are some additional examples of panel data sets:

  • Panel Study of Income Dynamics (PSID): Data on approximately 5,000 families on various socioeconomic and demographic variables
  • Survey of Income and Program Participation (SIPP): Interviews about economic condition of respondents

Panel models have the advantage that they take into account heterogeneity among observational units, e.g., firms, states, counties. Those models also contribute to the better understanding on the dynamics of change for observational units over time. Panel data combines cross-sectional data with time series data leading to more complete behavioral models.

  • Balanced versus unbalance panel: A balanced panel has the same number of time-series observations for each subject or observational unit, whereas an unbalanced panel does not.
  • Short versus long panel: A short panel has a larger number of subjects or observational units than there are time periods. A long panel has a greater number of time periods than observational units.

There are three types of regression models presented in this chapter:

  • Pooled Ordinary Least Square model
  • Fixed Effects Panel Data Model
  • Random Effects Panel Data Model