Dire Dawa University

College of Business and Economics

Department of Economics

Econometrics II

By Desalegn N.

1 October 23, 2023

Chapter Four: Introduction to Panel Data Regression Models

Outlines

 Introduction

 Estimation of Panel Data Regression Model : The Fixed Effects Approach

 Estimation of Panel Data Regression Model: The Random Effects Approach

Introduction
There are three types of data that are generally available for empirical analysis:

time series, cross section, and panel.

In time series data, we observe the values of one or more variables over a period of

time (e.g., GDP for several quarters or years).

In cross-section data, values of one or more variables are collected for several

sample units, or entities, at the same point in time (e.g., consumption level for 50

households in Dire Dawa city for a given year).

Introduction

In panel data, the same cross-sectional unit (say a family or a firm or state) is

surveyed over time.

Thus, panel data is a combination of cross-sectional and time series data. panel

data has both space and time dimensions.

Panel data/longitudinal data study over time about variables or group of subjects.
Advantage of Panel Data

A. Identification of parameters (robustness)

 Unobserved Heterogeneity

Endogenous regressors or measurement error

Individual dynamics

B. Efficiency of Parameter Estimators

Unobserved Heterogeneity
Since panel data relate to individuals, firms, states, countries, etc., over time, there

is bound to be heterogeneity in these units.

Panel Data take into account unobserved heterogeneity explicitly by allowing for

individual-specific variables.

It can better detect and measure effect that simply cannot be observed in pure

cross-section or pure time series data.

Unobserved Heterogeneity
As a result, omitted variable bias arises due to correlation of exclude variable with

the included variables is excluded from the model is eliminated.

For instance, The effect of innate ability on earnings can be better studied if we

include individual fixed effect, which cannot be captured by either cross-section or

time series data.

In addition firms' output depends on physical inputs denoted by x (observed in the

data) and on managerial ability denoted by M (unobserved in the data). M can be

account by panel data

Unobserved Heterogeneity

Yit = α + X’itβ + miβk+1 + εit------------(1),

Yit = α + X’itβ + Uit-------------(2), if m not include in the model

Thus, Uit= miβk+1 + εit as result estimating by OLS leads to biased and

inconsistent result.

However, Panel data offer the possibility to control for (eventually eliminate)

sources of heterogeneity (both observed and unobserved).

 Endogenous regressors or measurement error
Panel data will provide `internal' instruments for regressors that are endogenous

or subject to measurement error.

For instance, if Xit is correlated with unobserved individual specific factors (αi), it

can be argued that Xit-𝑋𝑖, where xi is the time-average for individual i , is

uncorrelated with αi and provides a valid instrument for Xit.

More generally, estimating the model under the Fixed effect assumption eliminates

αi from the error term and, consequently, eliminates all endogeneity problems

relating to it.
Identification of Individual Dynamics

By studying the repeated cross section of observations, panel data are better suited

to study the dynamics of change. This is because individual histories are observed

and can be included in the model.

Example: Spells of unemployment, job turnover, and labor mobility are better

studied with panel data.

Efficiency of Parameter Estimators

By combining time series and cross-section observations, panel data give more

informative data, more variability, less collinearity among variables, more degrees

of freedom and more efficiency.

Panel data gives efficient and accurate result even the observation of panel equals

to cross sectional and time series data.

This reasoned by panel data allow explanatory variables to vary over two

dimensions (individuals and time).

Estimation of panel Regression Model
Estimation of Panel Data Regression Model : The Fixed Effects Approach

In the case of panel data sets consisting time and cross-sectional dimension using

OLS will not result better result.

Thus, we need to have either Fixed or Random effect model.

Fixed effect model: is simply a linear regression model in which the intercept terms

vary over the individual units.

Yit = αi + X’itβ + εit-----------------------------(2)

Estimation of Panel Data Regression Model : The Fixed Effects Approach

where αi is individual intercept, also, we usually assumed that all X’it are

independent variables and assumed that X’it independent from εit.

Suppose firm's output (Yit) depends on the amount of labor (X1it) and capital

(X2it) for the period 2001-2022 for four firms: F1, F2, F3, and F4.

Yit = αi + β1X1it + β2X2it+ εit----------------------------(3)

Here, there are 4 cross-sectional units and 22 time periods. Thus, the number of

observation becomes 88.

Estimation of Panel Data Regression Model : The Fixed Effects Approach

In fixed effect model the intercept assumed to differ across individuals (here the

four firms), each individual's intercept does not vary over time; that is, it is time

invariant.

Since FEM: Estimate β while controlling for all sources of heterogeneity that are

individual specific and time invariant (like management quality).

This can be done by: dummy variable technique, particularly, the differential

intercept dummies.
Estimation of Panel Data Regression Model : The Fixed Effects Approach

Accordingly, Yit = α1i + α2D1+ α3D2 + α4D4 + β1X1it + β2X2it+ εit---------------(4)

where D2i = 1 if the observation belongs to F2, 0 otherwise; D3i = 1 if the

observation belongs to F3, 0 otherwise; and D4i = 1 if the observation belongs to

F4, 0 otherwise.

Here there is no dummy for F1 (the comparison region). i.e, α1 represents the

intercept of F1 and α2 , α3 , and α4, the differential intercept coefficients, tell by

how much the intercepts of F2, F3, and F4 differ from the intercept of F1.
Estimation of Panel Data Regression Model : The Fixed Effects Approach

Thus, specified model [Eq(4)] is known as Least Square Dummy Variable

(LSDV)/fixed effect.

But, if there exist many cross sectional unit using Least Square Dummy Variable

(LSDV) reduce degree of freedom. Also, it may lead to multicollinarity problem.

Alternative: within estimation.

Estimation of Panel Data Regression Model : The Fixed Effects Approach

Using within estimator (FEE), exactly the same estimator for β is obtained if the

regression is performed in deviations from individual means. That is

Yit − 𝑌𝑖= αi + (Xit − 𝑋𝑖) β + (εit − ε𝑖) -------------------------------(5)

equation 5 is within transformation and estimating equation 5 eliminate

individual effect and observed time invariant heterogeneity.

The estimators in equation 5 is known as within estimator or fixed effect estimator

Estimation of Panel Data Regression Model : The Fixed Effects Approach

Assumption of FEM:

Unbiasedness: All Xit are independent of all εit. E {(Xit − 𝑋𝑖) εit} = 0

Normality distributed: εit is normally distributed

Consistent
Estimation of Panel Data Regression Model : Random Effects Model

REM: is a panel model that treat αi as random variable with a mean value of “α1” (no

subscript i here).

REM: Yit = α1+ X’it β + Ut+εit--------------(6)

Accordingly the intercept value for an individual company can be expressed as:

αi = α1+ Ui

where Ui is a random error term with a mean value of zero and variance of constant

variance.
Estimation of Panel Data Regression Model : Random Effects Model

What we are essentially saying is that the four firms included in our sample are

drawing from a much larger universe. As result, selected firms have a common mean

value for the intercept (= α1) and the individual differences in the intercept values of

each region are reflected in the error term Ui.

To demonstrate the mechanics of the random effects estimator, define the time

. .
means of Yit and Xit as where: 𝑌𝑖 = 𝑡 𝑌𝑖𝑡 /𝑇and 𝑋𝑖 = 𝑡 𝑋𝑖𝑡 /𝑇.
Estimation of Panel Data Regression Model : Random Effects Model

Yit-ψ𝑌𝑖= α1(1-ψ)+ (Xit − ψ𝑋𝑖) β + Ui(1- ψ )+(εit − ψ ε𝑖)----------------------(6)

Where ψ=δ2 ε/δ2 ε+δ2 u

Estimating the Eq(6) by OLS is random effects model.

Fixed or random effects model?: Hausman Test

Treating αi as fixed or random brings difference in the estimates of the β

parameters.

If the individual effects αi and the explanatory variables are correlated, the

correlation can be handled by using the FEM, which essentially eliminates the αi

from the model, and thus eliminates any problems that they may cause and FEM is

appropriate.
Fixed or random effects model?: Hausman Test

In contrast, the REM may be preferred if one wants to make inference with respect

to the population characteristics. One way to formalize this is noting the following:

Thus, REM is appropriate when If the regressors and the individual effects are not

correlated.

The test used to choose between the two models is known as the Hausman test.

The null hypothesis of this test states that there is no correlation between

regressors and individual effects. So rejection of the null favors the FEM.