Panel Data Regression

Models
Kabul University
Economics Faculty
MFB program

Presenter: Hassib
Supervisor: Dr. Prof. Ajmal Arian
What is Panel Data

 To data collected on multiple subjects (individual, firms and countries) over multiple
points in time, allowing analysis of both individual and time-related effects.
 Providing more information and potentially more reliable estimates compared to purely
cross sectional or time-series data.
 Combining Cross sectional time series data
 Commonly used in economic phenomena such as: effects of policies, market trends, and
individual behavior over time.
 Other names: pooled data, combination of time series and cross-section data, micro panel
data, longitudinal data, longitudinal data and cohort analysis.
Some of the well-known panel data sets are:

 Panel Study of Income Dynamics (PSID): University of Michigan, tracking and survey of
American families and their descendants since 1968.
 National Longitudinal Survey of Youth(NLSY): Bureau of Labor Statistics that follows
the lives of several cohorts of American born between 1957 and 1984.
 European Community Household Panel(ECHP): Longitudinal survey in European
countries from 1994 to 2001.
 German Socio-Economic Panel (SOEP) : Longitudinal survey of German’s households in
various aspects of social and economic life since 1984.
 Panel Study of Belgian Households ( PSBH): Based on Belgian households on
socioeconomic changes over time.
Why Panel Data?
Advantages of Panel data rather than cross-sectional and time-series
data.
 Estimate heterogeneity
 consequence of combination of Cross-sectional and Time-series is efficient data
 panel data are better suited to study the dynamics of change
 Measuring and finding effects in panel data is much better.
 Enabling to study more complicated models
 Minimize biase
 Enriching empirical analyze
To illustrate an example
Example
Example
Example

Pooling all the 80 observations, we can write the Grunfeild Investment function:
Estimation of Panel Data Regression Model

 The fixed effects approach (Less Square Dummy Variable) LSDV

 The random effects approach (Error component model)ECM
Estimation of Panel Data Regression Model: The
fixed effects approach

For estimating of intercept, slope coefficient, and error term; there are several possibilities:
 Assume that the intercept and slope coefficients are constant across time and space and the
error term captures differences over time and individuals.
 2. The slope coefficients are constant but the intercept varies over individuals.
 3. The slope coefficients are constant but the intercept varies over individuals and time.
 4. All coefficients (the intercept as well as slope coefficients) vary over individuals.
 5. The intercept as well as slope coefficients vary over individuals and time.
All Coefficients Constant across Time and
Individuals
Slope Coefficients Constant but the Intercept Varies across
Individuals: The Fixed Effects or Least-Squares Dummy Variable
(LSDV) Regression Model

 Individually model for constant slope coefficient across firms is:

Yit = β1i + β2X2it + β3X3it + uit
 Deferential intercept dummies model is:
Yit = α1 + α2D2i + α3D3i + α4D4i + β2X2it + β3X3it + uit
 LSDV is also known Covariance Model and X1 and X2 are covariate
Yit = −245.7924 + 161.5722D2i + 339.6328D3i + 186.5666D3i + 0.1079X2i + 0.3461X3i
se = (35.8112) (46.4563) (23.9863) (31.5068) (0.0175) (0.0266)
t = (−6.8635) (3.4779) (14.1594) (5.9214) (6.1653) (12.9821)
R2 = 0.9345 d = 1.1076 df = 74
Time Effect and Slope Coefficients Constant but the Intercept Varies over
Individuals As Well As Time

We can introduce time dummies(Time Effect) as like dummy variable.

 Let to take previous example:
Yit = λ0 + λ1Dum35 + λ2Dum36+· · ·+λ19Dum53 + β2X2it + β3X3it + uit

For doing this we combine Time effect and covariance models,

Yit = α1 + α2DGMi+ α3DUSi+ α4DWESTi+ λ0 + λ1Dum35+· · ·+λ19Dum53 + β2X2i+β3X3i
+ uit
A Caution on the Use of the Fixed Effects, or
LSDV Model
 First, if you introduce too many dummy variables, as in the case of model, you will run up
against the degrees of freedom problem.
 Second, with so many variables in the model, there is always the possibility of
multicollinearity, which might make precise estimation of one or more parameters difficult.
 Third, invariant variables such as, sex, color and ethnicity, LSDV might not be able to
identify impact of time-invariant variables.
 Fourth, we have to think carefully about the error term uit. All the results we have
presented so far are based on the assumption that the error term follows the classical
assumptions, namely, uit ∼ N(0, σ2). Since the i index refers to cross-sectional
observations and t to time series observations, the classical assumption for uit may have to
be modified.
ESTIMATION OF PANEL DATA REGRESSION MODELS:
THE RANDOM EFFECTS APPROACH

 When dummy variable do not pretend sufficient knowledge of about true model, and
disturbance term. The approach which is suggested is called Error Components Model
(ECM) or Random Effect Model (REM).
 The basic idea is started by this below equation:
Yit = β1i + β2X2it + β3X3it + uit
Instead of B1i as fixed, we assume this is a random variable with a mean value of B1(no i): β1i
= β1 + εi i = 1, 2, . . . , N
FIXED EFFECTS (LSDV) VERSUS
RANDOM EFFECTS MODEL
 One of the challenges for a researcher is to chose FEM or ECM.
 FEM is appropriate when εi and the X’s are correlated. On the other hand, while they are uncorrelated
ECM is appropriate.
 If T (the number of time series data unite) is large and N (the number of cross sectional data unite) is
small, there is little deference in estimation of parameters through FEM and ECM. Thus, selecting
approach would be based on computational convenience, and FEM may preferable.
 If T is small and N is large, parameters that are estimated by two approaches would be significant
deference. Statistical inference: If we believe cross sectional unites are not random drawing from large
sample; FEM is appropriate. And if possible, ECM is appropriate.
 If the individual error component εi and one or more repressor are correlated, then the ECM estimators
are biased, whereas those obtained rom FEM are unbiased.
 If N is large and T is small, and if the assumptions underlying ECM hold, ECM estimators are more
efficient than FEM estimators.
Tests for choosing FEM and CEM

 Hausman Test: The test statistic developed by Hausman has an asymptotic χ2 distribution.
If the null hypothesis is rejected, the conclusion is that ECM is not appropriate and that we
may be better off using FEM, in which case statistical inferences will be conditional on the
εi in the sample.
 Johnston and DiNardo : there is no simple rule to help the researcher navigate past the
Scylla of fixed effects and the Charybdis of measurement error and dynamic selection.
Although they are an improvement over cross-section data, panel data do not provide a
cure-all for all of an econometrician’s problems
Some concluding comments regarding to panel
data regression model
 Hypothesis testing with panel data.
 Heteroscedicity and autocorrelation in ECM.
 Unbalanced panel data
 Dynamic panel data models in which the legged value(s) of the regression (Yit) appears as
an explanatory variable.
 Simultaneous equations involving panel data.
 Qualitative depends variables and panel data.