Panel Data Analysis Using Stata

Sebastian T. Braun
University of St Andrews
Course Outline 1-1

Course Objectives

To provide a concise introduction to applied panel data

analysis.
To review core theoretical methods of panel data analysis and
apply these methods hands-on.
To learn how to analyze (microeconometric) panel data using
the statistical software Stata.

Panel Data
Course Outline 1-2

Recommended Readings

The applied part of the course will draw heavily on Chapter 8 of

Cameron, A. Colin and Pravin K. Trivedi (2010). Applied
Microeconometrics Using Stata. Stata Press.

Recommended introductory textbooks that provide an introduction

to panel data analysis are:
Wooldridge, Jeffrey M. (2015). Introductory Econometrics.
Cengage Learning Services, 5th edition.
Kennedy, Peter (2008). A Guide to Econometrics. John Wiley
& Sons, 6th edition.

Panel Data
Course Outline 1-3

Course Material

You find the slides on my homepage:

https://sebastiantillbraun.wordpress.com/teaching/

Panel Data
Overview 2-1

Overview

1. Course Outline X
2. Introduction
3. Panel Data Management
4. Regression Analysis
5. Hypothesis Testing
6. Extensions
7. Outlook: Advanced Panel Data Analysis

Panel Data
Introduction 3-1

What is Panel Data?

A cross-section (of people, firms, countries, etc.) is observed

over time.
Panel data provides observations on the same units in several
time periods (unlike independently pooled cross sections).
Panel data often consist of a very large number of
cross-sections over a small number of time periods.

Panel Data
Introduction 3-2

What Advantages Do Panel Data Offer?

Panel data allows us to...

...examine issues that cannot be studied using either time
series or cross-sectional data.
...deal with unobserved heterogeneity in the micro units.
...analyze dynamics with only a short time series.
...increase the efficiency of estimation.

Panel Data
Introduction 3-3

What Advantages Do Panel Data Offer?

Panel data allows us to...

...examine issues that cannot be studied using either time
series or cross-sectional data
...deal with unobserved heterogeneity in the micro units.
...analyze dynamics with only a short time series.
...increase the efficiency of estimation.

Panel Data
Panel Data Management 4-1

Getting Started...

We now consider data from the Panel Study of Income Dynamics.

You can install the relevant files from within Stata. Type:
. net from http://www.stata-press.com/data/mus
. net install mus
. net get mus
You can also download the data from
www.stata-press.com/data/mus.html.

Panel Data
Panel Data Management 4-2

The Dataset

Open the data set:

. use "mus08psidextract.dta", clear
The data set contains information on 595 individuals (the cross-sectional
units) over 7 years (1976-1982).
The total number of observations is thus 595 × 7 = 4165.
There are no missing observations (so the data set is balanced).

Panel Data
Panel Data Management 4-3

describe the Data

. use "mus08psidextract.dta", clear
(PSID wage data 1976-82 from Baltagi and Khanti-Akom (1990))

. describe

Contains data from mus08psidextract.dta

obs: 4,165 PSID wage data 1976-82 from
Baltagi and Khanti-Akom (1990)
vars: 22 26 Nov 2008 17:15
size: 295,715 (99.7% of memory free) (_dta has notes)
-------------------------------------------------------------------------------
storage display value
variable name type format label variable label
-------------------------------------------------------------------------------
exp float %9.0g years of full-time work
experience
wks float %9.0g weeks worked
occ float %9.0g occupation; occ==1 if in a
blue-collar occupation
ind float %9.0g industry; ind==1 if working in
a manufacturing industry
fem float %9.0g female or male
union float %9.0g if wage set be a union contract
ed float %9.0g years of education
lwage float %9.0g log wage
id float %9.0g
t float %9.0g
tdum1 byte %8.0g t== 1.0000
exp2 float %9.0g
-------------------------------------------------------------------------------

Panel Data
Panel Data Management 4-4

Panel Data Organization

Panel data is usually organised in the so-called long form, with

each observation a distinct individual-time pair.
In our case, the cross-section (panel) and time variables are id
and t, respectively.

Panel Data
Panel Data Management 4-5

Panel Data Organization (ctd.)

. * Organization of dataset
. list id t lwage exp union occ in 1/14, clean

id t lwage exp union occ

1. 1 1 5.56068 3 0 0
2. 1 2 5.72031 4 0 0
3. 1 3 5.99645 5 0 0
4. 1 4 5.99645 6 0 0
5. 1 5 6.06146 7 0 0
6. 1 6 6.17379 8 0 0
7. 1 7 6.24417 9 0 0
8. 2 1 6.16331 30 0 1
9. 2 2 6.21461 31 0 1
10. 2 3 6.2634 32 1 1
11. 2 4 6.54391 33 0 1
12. 2 5 6.69703 34 0 1
13. 2 6 6.79122 35 0 1
14. 2 7 6.81564 36 0 1
Panel Data
Panel Data Management 4-6

Panel Data Organization (ctd.)

Inform Stata about the panel and time variables id and t by

typing:
. xtset id t
You can now use the time-series operators of Stata (L., D.,...)
and all the xt commands.

Panel Data
Panel Data Management 4-7

xtdescribe the data

. * Panel description of dataset

. xtdescribe

id: 1, 2, ..., 595 n = 595

t: 1, 2, ..., 7 T = 7
Delta(t) = 1 unit
Span(t) = 7 periods
(id*t uniquely identifies each observation)

Distribution of T_i: min 5% 25% 50% 75% 95% max

7 7 7 7 7 7 7

Freq. Percent Cum. | Pattern

---------------------------+---------
595 100.00 100.00 | 1111111
---------------------------+---------
595 100.00 | XXXXXXX

Panel Data
Regression analysis 5-1

Estimating the Union Wage Premium

We now use the panel data to estimate the union wage

premium.
In our case, the premium measures the degree to which wages
are higher if set by a union contract.
In general, the vast empirical literature on the issue finds that
union bargaining increase wages above the market rate.
We will see how panel data can be used to overcome some of
the difficulties associated with estimating the wage premium.
We restrict the analysis to men (drop if fem == 1 )!

Panel Data
Regression analysis 5-2

The Basic Linear Panel Model

yit = α + xit β + ai + it (1)

i denotes the cross-sectional unit and t the time period.

yit is the dependent variable.
α is a common intercept.
xit are explanatory variables.
ai are unobserved individual-specific (fixed) effects.
it is an error term.

Panel Data
Regression analysis 5-3

The Basic Linear Panel Model

wit =α + xit β + γUnionit +ai + it (2)

i denotes the cross-sectional unit and t the time period.

wit is the log of the hourly wage.
α is a common intercept.
Unionit indicates whether wage is set by a union contract.
ai are unobserved individual-specific (fixed) effects.
it is an error term.

Panel Data
Regression analysis 5-4

The Unobserved Fixed Effect

wit = α + xit β + γUnionit + ai +it (3)

The ai captures all unobserved, time constant factors that

affect wit .
The unobserved fixed effect is specific to an individual and
does not vary over time.
The unobserved fixed effect is unknown to the researcher.
Examples: ability, ambition...

Panel Data
Regression analysis 5-5

Pooled OLS

wit = α + xit β + γUnionit + ai + it . (4)

How should one estimate the parameter of interest, γ, given

our seven years of panel data?
One possibility is just to ‘pool’ the data and use OLS.
Do this in Stata using exp, exp2, wks, ed, ind and occ as
additional controls.

Panel Data
Regression analysis 5-6

Pooled OLS Estimates

. ******* 2. POOLED OLS
.
. * Pooled OLS with incorrect default standard errors
. regress lwage exp exp2 wks ed union ind occ

Source | SS df MS Number of obs = 3696

-------------+------------------------------ F( 7, 3688) = 229.48
Model | 215.291596 7 30.7559423 Prob > F = 0.0000
Residual | 494.284928 3688 .134025197 R-squared = 0.3034
-------------+------------------------------ Adj R-squared = 0.3021
Total | 709.576524 3695 .192036948 Root MSE = .36609

------------------------------------------------------------------------------
lwage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
exp | .0384824 .002442 15.76 0.000 .0336945 .0432703
exp2 | -.0006084 .0000539 -11.29 0.000 -.000714 -.0005027
wks | .0047247 .0012331 3.83 0.000 .0023071 .0071422
ed | .0635993 .0028509 22.31 0.000 .0580098 .0691887
union | .1204051 .0138341 8.70 0.000 .0932818 .1475284
ind | .0431938 .0126986 3.40 0.001 .0182968 .0680908
occ | -.150339 .016286 -9.23 0.000 -.1822695 -.1184086
_cons | 5.24959 .0780379 67.27 0.000 5.096589 5.402592
------------------------------------------------------------------------------

Panel Data
Regression analysis 5-7

What is Wrong with Pooled OLS?

Let us re-write the linear basic panel model as follows:

wit = α + xit β + γUnionit + (ai + it ) (5)

= α + xit β + γUnionit + vit . (6)

where vit = ai + it is referred to as the composite error term.

Panel Data
Regression analysis 5-8

Problem 1: Serially Correlated Errors

If w is overpredicted in one year for a given person, then it is

likely to be overpredicted in other years.
The composite error vit = ai + it is serially correlated even if
it is i.i.d. with a variance of σ2 .
A worker that is more able today will be more able tomorrow
and we thus have:

Cor (vit , vis ) 6= 0 for t 6= s if ai 6= 0. (7)

We can easily show this in Stata...

Panel Data
Regression analysis 5-9

Problem 1: Serially Correlated Errors (ctd.)

.
. * Autocorrelations of residual
. quietly regress lwage exp exp2 wks ed union ind occ

. predict uhat, residuals

. corr uhat L1.uhat

(obs=3168)

| L.
| uhat uhat
-------------+------------------
uhat |
--. | 1.0000
L1. | 0.8599 1.0000

Panel Data
Regression analysis 5-10

Problem 1: Serially Correlated Errors (ctd.)

Each additional observation for a given person provides less

than an independent piece of new information.
With serially correlated errors, standard errors are thus biased.

Panel Data
Regression analysis 5-11

Solution: Cluster-Robust Standard Errors

Calculate cluster-robust standard errors that allow for

correlation within clusters (cross-sections).
Cluster-robust standard errors only require that errors are
independent between cross-sections.
Use the vce(cluster ) option in Stata...

Panel Data
Regression analysis 5-12

OLS with Cluster-Robust Standard Errors

. * Pooled OLS with cluster-robust standard errors
. regress lwage exp exp2 wks ed union ind occ, vce(cluster id)

Linear regression Number of obs = 3696

F( 7, 527) = 46.65
Prob > F = 0.0000
R-squared = 0.3034
Root MSE = .36609

(Std. Err. adjusted for 528 clusters in id)

------------------------------------------------------------------------------
| Robust
lwage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
exp | .0384824 .0047986 8.02 0.000 .0290556 .0479092
exp2 | -.0006084 .0001087 -5.60 0.000 -.0008219 -.0003948
wks | .0047247 .0018448 2.56 0.011 .0011005 .0083488
ed | .0635993 .0062134 10.24 0.000 .0513931 .0758054
union | .1204051 .027477 4.38 0.000 .0664272 .174383
ind | .0431938 .0254036 1.70 0.090 -.006711 .0930986
occ | -.150339 .0321478 -4.68 0.000 -.2134925 -.0871856
_cons | 5.24959 .1434456 36.60 0.000 4.967795 5.531386
------------------------------------------------------------------------------

Panel Data
Regression analysis 5-13

Problem 2: Omitted Variable Bias

wit = α + xit β + γUnionit + vit (8)

We must assume that vit = ai + it is uncorrelated with

Unionit for OLS to consistently estimate γ.
So even if it is uncorrelated with Unionit , pooled OLS is
biased and inconsistent if ai and Unionit are correlated.
The resulting heterogeneity bias is caused from omitting a
time-constant variable.

Panel Data
Regression analysis 5-14

Problem 2: Omitted Variable Bias (ctd.)

Why should ai and Unionit be correlated?

Unobserved factors that affect wages may also affect workers’
selection into the covered sector.
Wage standardization policy of unions might be most
appealing to workers with low underlying earnings potential.
Unionised employers might pick workers from the queue, as
not all workers who desire union employment can find union
jobs.

⇒ Unionit might be positively or negatively correlated with ability.

Panel Data
Regression analysis 5-15

The Fixed Effects Model

The fixed effects model allows the unobserved effects to be

correlated with the explanatory variables.
In fact, it uses a transformation to remove the unobserved
effect prior to estimation.

Panel Data
Regression analysis 5-16

The Fixed Effects Transformation

Consider our basic linear panel model:

yit = α + xit β + ai + it . (9)

For each cross-section, average this equation over time:

yi = α + x i β + ai + i , (10)

where the average of some variable z is given by

z i = T −1 T
P
t=1 yit .

Panel Data
Regression analysis 5-17

The Fixed Effects Transformation (ctd.)

Now subtract equation (10) from (9) to get rid of the fixed effect:

(yit − yi ) = (α − α) + (xit − x i )β + (ai − ai ) + (it − i )

= (xit − x i )β + (it − i ). (11)

Because ai has been eliminated, OLS leads to consistent estimates

of β even if xit is correlated with ai !

Panel Data
Regression analysis 5-18

Re-Estimate the Union Wage Premium...

Use xtreg, fe in STATA to re-estimate the union wage premium

using the fixed effects model:

(wit − wi ) = (xit − x i )β + γ(Unionit − Unioni ) + (it − i ). (12)

What does your estimate of γ suggests about the correlation

between Union and ability?

Panel Data
Regression analysis 5-19

Fixed Effects Estimates

. ******* 3. FIXED EFFECTS ESTIMATOR (WITHIN ESTIMATOR)
.
. * Within or FE estimator
. xtreg lwage exp exp2 wks ed union ind occ, fe
note: ed omitted because of collinearity

Fixed-effects (within) regression Number of obs = 3696

Group variable: id Number of groups = 528

R-sq: within = 0.6558 Obs per group: min = 7

between = 0.0228 avg = 7.0
overall = 0.0427 max = 7

F(6,3162) = 1004.25
corr(u_i, Xb) = -0.9223 Prob > F = 0.0000

------------------------------------------------------------------------------
lwage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
exp | .1149389 .0026801 42.89 0.000 .1096841 .1201938
exp2 | -.0004347 .0000584 -7.44 0.000 -.0005491 -.0003202
wks | .0004693 .0006576 0.71 0.476 -.0008201 .0017587
ed | (omitted)
union | .0316998 .0159769 1.98 0.047 .0003736 .063026
ind | .0182395 .0160431 1.14 0.256 -.0132165 .0496954
occ | -.0113013 .0146455 -0.77 0.440 -.0400169 .0174144
_cons | 4.600501 .044283 103.89 0.000 4.513675 4.687328
-------------+----------------------------------------------------------------
sigma_u | 1.047369
sigma_e | .1534777
rho | .97897847 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(527, 3162) = 33.82 Prob > F = 0.0000

Panel Data
Regression analysis 5-20

Caveats of the Fixed-Effects Estimator

The fixed effects estimator uses the time variation in y and x

within cross-sectional units only.
It discards variation across cross-sections (between variation).
It does not allow us to estimate the coefficients of
time-invariant regressors (gender, education...).
Differenced regressors may be more susceptible to
measurement error.
Does not solve the problem of time-varying omitted variables.

Panel Data
Regression analysis 5-21

Regression analysis 5-21

Fixed Effects Estimator (by David Bell)
Fixed Eects Estimator (by David Bell)

60
A
50

40

30

20
B
10
B A
0
-5 0 5 10 15 20

Individual 1 Individual 2 Individual 3

Individual 4 Linear (Individual 1) Linear (Individual 3)
Linear (Individual 2) Linear (Individual 4)

Panel Data
Regression analysis 5-22

Within- and Between-Variation

The STATA command xtsum decomposes the overall variation in a
2 + s 2 ):
variable as follows (where sO2 ≈ sW B

2 1 XX
Within: sW = (xit − x i )2 , (13)
NT − 1 i t
1 X
Between: sB2 = (xi − x )2 , (14)
N −1 i
1 XX
Overall: sO2 = (xit − x )2 . (15)
NT − 1 i t

Use xtsum (and possibly xttrans) to assess the relative importance

of between and within variation in the data.
Panel Data
Regression analysis 5-23

xtsum the Data

. * Panel summary statistics: within and between variation
. * Notice: The min and max columns give the min and max of x_it for overall,
x^bar_i for
> between and x_it-x^bar_i+x^bar for within
. xtsum id t lwage exp wks ed union tdum1

Variable | Mean Std. Dev. Min Max | Observations

-----------------+--------------------------------------------+----------------
id overall | 297.7292 171.7455 1 594 | N = 3696
between | 171.8851 1 594 | n = 528
within | 0 297.7292 297.7292 | T = 7
| |
t overall | 4 2.000271 1 7 | N = 3696
between | 0 4 4 | n = 528
within | 2.000271 1 7 | T = 7
| |
lwage overall | 6.729774 .4382202 5.01728 8.537 | N = 3696
between | .3656286 5.518704 7.813596 | n = 528
within | .2420119 4.835235 8.674519 | T = 7
| |
exp overall | 20.21402 10.99381 1 51 | N = 3696
between | 10.81909 4 48 | n = 528
within | 2.000271 17.21402 23.21402 | T = 7
| |
wks overall | 46.96374 4.983413 5 52 | N = 3696
between | 3.165729 31.57143 51.57143 | n = 528
within | 3.850824 12.39232 63.82089 | T = 7
| |
ed overall | 12.84659 2.822298 4 17 | N = 3696
between | 2.824592 4 17 | n = 528
within | 0 12.84659 12.84659 | T = 7
| |
union overall | .3833874 .4862772 0 1 | N = 3696
between | .4597488 0 1 | n = 528
within | .1594985 -.4737554 1.24053 | T = 7
| |
tdum1 overall | .1428571 .3499745 0 1 | N = 3696
between | 0 .1428571 .1428571 | n = 528
within | .3499745 0 1 | T = 7

Panel Data
Regression analysis 5-24

xttrans union

. xttrans union, freq

if wage |
set be a | if wage set be a
union | union contract
contract | 0 1 | Total
-----------+----------------------+----------
0 | 1,890 66 | 1,956
| 96.63 3.37 | 100.00
-----------+----------------------+----------
1 | 63 1,149 | 1,212
| 5.20 94.80 | 100.00
-----------+----------------------+----------
Total | 1,953 1,215 | 3,168
| 61.65 38.35 | 100.00

Panel Data
Regression analysis 5-25

xttrans ed
. * Transition probabilities for a variable
. xttrans ed if ed>=12, freq

years of | years of education

education | 12 13 14 15 16 17 | Total
-----------+------------------------------------------------------------------+----------
12 | 1,116 0 0 0 0 0 | 1,116
| 100.00 0.00 0.00 0.00 0.00 0.00 | 100.00
-----------+------------------------------------------------------------------+----------
13 | 0 144 0 0 0 0 | 144
| 0.00 100.00 0.00 0.00 0.00 0.00 | 100.00
-----------+------------------------------------------------------------------+----------
14 | 0 0 264 0 0 0 | 264
| 0.00 0.00 100.00 0.00 0.00 0.00 | 100.00
-----------+------------------------------------------------------------------+----------
15 | 0 0 0 66 0 0 | 66
| 0.00 0.00 0.00 100.00 0.00 0.00 | 100.00
-----------+------------------------------------------------------------------+----------
16 | 0 0 0 0 480 0 | 480
| 0.00 0.00 0.00 0.00 100.00 0.00 | 100.00
-----------+------------------------------------------------------------------+----------
17 | 0 0 0 0 0 396 | 396
| 0.00 0.00 0.00 0.00 0.00 100.00 | 100.00
-----------+------------------------------------------------------------------+----------
Total | 1,116 144 264 66 480 396 | 2,466
| 45.26 5.84 10.71 2.68 19.46 16.06 | 100.00

Panel Data
Regression analysis 5-26

Within and Between R 2

Stata’s xtreg command calculates the following three R 2 measures:

n o
Within R 2 : ρ2 (yit − y i ), (xit − x i )β̂ , (16)
n o
Between R 2 : ρ2 y i , x i β̂ , (17)
n o
Overall R 2 : ρ2 yit , xit β̂ , (18)

where ρ2 (x , y ) denotes the squared correlation between x and y .

Panel Data
Regression analysis 5-27

LSDV and First-Difference Estimators

There are two other estimators that also allow the unobserved
fixed-effect to be correlated with the regressors:

1. Least-squares dummy variables (LSDV) estimator

2. First-difference (FD) estimator

Both estimators are also widely used in practice but share the
caveats of the fixed effects estimator.

Panel Data
Regression analysis 5-28

The Dummy Variables Regression

The LSDV regression considers the unobserved effects ai as

parameters to be estimated.
It directly estimates yit = α + xit β + ai + it adding a dummy
for each cross-sectional unit i.

Panel Data
Regression analysis 5-29

The Dummy Variables Regression

The LSDV regression gives us exactly the same estimate of β

as the fixed-effects estimator.
It does not allow us to estimate the coefficients of
time-invariant regressors (why?).
Use areg or reg to estimate the union wage premium using the
LSDV regression!

Panel Data
Regression analysis 5-30

The LSDV Estimates

. * LSDV model fitted using areg
. areg lwage exp exp2 wks ed union ind occ, absorb(id)
note: ed omitted because of collinearity

Linear regression, absorbing indicators Number of obs = 3696

F( 6, 3162) = 1004.25
Prob > F = 0.0000
R-squared = 0.8950
Adj R-squared = 0.8773
Root MSE = .15348

------------------------------------------------------------------------------
lwage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
exp | .1149389 .0026801 42.89 0.000 .1096841 .1201938
exp2 | -.0004347 .0000584 -7.44 0.000 -.0005491 -.0003202
wks | .0004693 .0006576 0.71 0.476 -.0008201 .0017587
ed | (omitted)
union | .0316998 .0159769 1.98 0.047 .0003736 .063026
ind | .0182395 .0160431 1.14 0.256 -.0132165 .0496954
occ | -.0113013 .0146455 -0.77 0.440 -.0400169 .0174144
_cons | 4.600501 .044283 103.89 0.000 4.513675 4.687328
-------------+----------------------------------------------------------------
id | F(527, 3162) = 39.191 0.000 (528 categories)

Panel Data
Regression analysis 5-31

First-difference Estimator

Consider again our basic linear panel model

yit = α + xit β + ai + it . (19)

and lag it one period:

yi,t−1 = α + xi,t−1 β + ai + i,t−1 . (20)

Panel Data
Regression analysis 5-32

First-difference Estimator (ctd.)

Now subtract equation (20) from (19) to obtain the

first-differenced equation...

(yit − yi,t−1 ) = (α − α) + (xit − xi,t−1 )β + (ai − ai )

+(it − i,t−1 )
= (xit − xi,t−1 )β + (it − i,t−1 ) (21)

Because ai has been eliminated, OLS leads to consistent estimates

of β even if xit is correlated with ai !

Panel Data
Regression analysis 5-33

Re-Estimate the Union Wage Premium...

Now use Stata to re-estimate the union wage premium using the
model in first differences:

(wit − wi ) = (xit − xi,t−1 )β + γ(Unionit − Unioni,t−1 )

+(it − i,t−1 ). (22)

You should use the time-series operator for differences D...

Panel Data
Regression analysis 5-34

The First-difference Estimates

. ******* 5. FIRST DIFFERENCE ESTIMATOR
.
. sort id t

. * First-differences estimator
. regress D.(lwage exp exp2 wks ed union ind occ), noconstant
note: _delete omitted because of collinearity

Source | SS df MS Number of obs = 3168

-------------+------------------------------ F( 6, 3162) = 149.83
Model | 30.2606538 6 5.04344229 Prob > F = 0.0000
Residual | 106.438928 3162 .0336619 R-squared = 0.2214
-------------+------------------------------ Adj R-squared = 0.2199
Total | 136.699582 3168 .04315012 Root MSE = .18347

------------------------------------------------------------------------------
D.lwage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
exp |
D1. | .1192783 .006853 17.41 0.000 .1058416 .1327151
|
exp2 |
D1. | -.0005583 .0001491 -3.75 0.000 -.0008506 -.000266
|
wks |
D1. | -.0003622 .0006187 -0.59 0.558 -.0015752 .0008508
|
ed |
D1. | (omitted)
|
union |
D1. | .0162718 .0162282 1.00 0.316 -.0155471 .0480907
|
ind |
D1. | .0144281 .0168265 0.86 0.391 -.0185639 .0474201
|
occ |
D1. | -.0158051 .0147561 -1.07 0.284 -.0447376 .0131274
------------------------------------------------------------------------------

Panel Data
Regression analysis 5-35

Excursion: The Between Estimator

The antipode to the within estimator is the between estimator.

The between estimator uses only the cross-section variation in
the data.
To obtain the between model, average the basic linear panel
model:
y i = α + x i β + ai + i . (23)
The between estimator is simply the OLS estimator in this
model (xtreg, be in Stata).

Panel Data
Regression analysis 5-36

Excursion: The Between Estimator (ctd.)

The between estimator is only consistent if the error ai + i is

uncorrelated with x i .
Even if the between estimator is consistent, we have more
efficient estimators at hand (pooled OLS and RE).
The between estimator is rarely used in practice but is actually
an input into the RE estimator that we study now...

Panel Data
Regression analysis 5-37

The Random Effects Estimator

The FE / LSDV / FD estimators wipe out between variation

(i.e., variation across cross-sections).
If we believe that the ai ’s are not correlated with the
regressors, we should resort to more efficient estimators.
One possibility is pooled OLS with clustered-robust standard
errors.
A more efficient estimator is the Random Effects (RE)
estimator.

Panel Data
Regression analysis 5-38

The Random Effects Estimator (ctd.)

Consider again our basic linear panel model:

yit = α + xit β + ai + it . (24)

Now suppose that ai is purely random with mean 0 and

variance σα2 .
Thus, ai is assumed to be uncorrelated with the regressors.

Panel Data
Regression analysis 5-39

The Random Effects Estimator (ctd.)

Yet, the composite error vit = ai + it is still serially correlated.

Under the random effects assumption, Var (vit ) = σα2 + σ2 and
Cov (uit , uis ) = σα2 , s 6= t.
The intraclass correlation of the error is thus given by
σα2
ρv = Cor (vit , vis ) = . (25)
σα2 + σ2

Panel Data
Regression analysis 5-40

The Random Effects Estimator (ctd.)

The Random Effects Model is the OLS estimator in a model

transformed to have serially uncorrelated errors.
The transformed model is given by (see Wooldridge, 2002):

(yit − θ̂i yi ) = (1 − θ̂i )α + (xit − θ̂i x i )β + (1 − θ̂i )ai

+(it − θ̂i i ) (26)

where θ̂i is a consistent estimator of

q
θ =1− σα2 /(T σα2 + σ2 ), (27)

which is bounded by 0 and 1.

Panel Data
Regression analysis 5-41

The Random Effects Estimator (ctd.)

It is worth noting that the RE model is a weighted average of the

within and the between model:

(yit − yi ) + (1 − θ̂i )y i = (xit − x i )β + (it − i )

+(1 − θ̂i )(α + x i β + ai + i ). (28)

Panel Data
Regression analysis 5-42

The Random Effects Estimator (ctd.)

Before you apply the RE estimator in Stata using xtreg, re notice

that...
...for θ = 0, we obtain pooled OLS (which is an unweighted
average of the between and within model):

yit = α + xit β + ai + it . (29)

...for θ = 1, we obtain the within model:

(yit − yi ) = (xit − x i )β + (it − i ). (30)

Panel Data
Regression analysis 5-43

Random Effects Estimates

. ******* 6. RANDOM EFFECTS ESTIMATORS
.
. * Random-effects estimator
. xtreg lwage exp exp2 wks ed union ind occ, re theta

Random-effects GLS regression Number of obs = 3696

Group variable: id Number of groups = 528

R-sq: within = 0.6259 Obs per group: min = 7

between = 0.1823 avg = 7.0
overall = 0.1887 max = 7

Random effects u_i ~ Gaussian Wald chi2(7) = 2332.16

corr(u_i, X) = 0 (assumed) Prob > chi2 = 0.0000
theta = .80072752

------------------------------------------------------------------------------
lwage | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
exp | .0825226 .0030822 26.77 0.000 .0764816 .0885636
exp2 | -.0007411 .0000676 -10.96 0.000 -.0008737 -.0006086
wks | .0006109 .0008368 0.73 0.465 -.0010292 .002251
ed | .1057496 .0062768 16.85 0.000 .0934474 .1180518
union | .0646388 .0181938 3.55 0.000 .0289796 .100298
ind | .0013936 .0179769 0.08 0.938 -.0338405 .0366276
occ | -.0348985 .0176442 -1.98 0.048 -.0694806 -.0003165
_cons | 4.059829 .1030828 39.38 0.000 3.857791 4.261868
-------------+----------------------------------------------------------------
sigma_u | .28526616
sigma_e | .1534777
rho | .77551783 (fraction of variance due to u_i)
------------------------------------------------------------------------------
Panel Data
Regression analysis 5-44

Estimates of Variance Components

The composite error ai + it is referred to as ui + eit in the

Stata output.
Stata provides estimates of the std. dev. of ai (denoted
sigma_u) and it (denoted sigma_e).
The output rho is an estimate of the intraclass correlation
2
ρv = Cor (vit , vis ) = σ2σ+σ
α
2.
α 

For the RE model, the estimated θ̂i can be obtained using the
theta option.

Panel Data
Regression analysis 5-45

Comparison of Estimates

Now recap and store the different estimates by typing:

. quietly regress lwage exp exp2 wks ed union ind occ,
vce(cluster id)
. estimates store OLS_rob
. quietly xtreg lwage exp exp2 wks ed union ind occ, fe
. estimates store FE
. quietly xtreg lwage exp exp2 wks ed union ind occ, re
. estimates store RE
. estimates table OLS_rob FE RE, b se stats(N r2 r 2o r 2b
r 2w )

Panel Data
 . quietly xtreg lwage $xlist, fe

. estimates store FE
Regression analysis 5-46
. quietly xtreg lwage $xlist, re

. estimates store RE
Comparison of Estimates (ctd.)
. estimates table OLS_rob FE RE, ///
> b se stats(N r2 r2_o r2_b r2_w) b(%7.4f)

--------------------------------------------------
Variable | OLS_rob FE RE
-------------+------------------------------------
exp | 0.0415 0.1135 0.0869
| 0.0050 0.0025 0.0028
exp2 | -0.0007 -0.0004 -0.0008
| 0.0001 0.0001 0.0001
wks | 0.0076 0.0008 0.0010
| 0.0018 0.0006 0.0008
ed | 0.0704 (omitted) 0.1086
| 0.0060 0.0062
union | 0.1517 0.0321 0.0615
| 0.0275 0.0149 0.0170
ind | 0.0966 0.0182 0.0124
| 0.0259 0.0155 0.0172
occ | -0.1447 -0.0214 -0.0411
| 0.0318 0.0137 0.0164
_cons | 4.9149 4.5930 3.9015
| 0.1413 0.0400 0.0983
-------------+------------------------------------
N | 4165 4165 4165
r2 | 0.3205 0.6574
r2_o | 0.0491 0.1933
r2_b | 0.0289 0.1833
r2_w | 0.6574 0.6287
--------------------------------------------------
legend: b/se
Panel Data
Hypothesis Testing 6-1

Hypothesis Testing

In applied work, you may want to test whether...

...the cross-sectional intercepts differ at all.
...to use the random or fixed effects estimators in case
intercepts differ.

Panel Data
Hypothesis Testing 6-2

RE/FE or Pooled OLS?

There is an easy way to test whether the cross-sectional

intercepts are different from each other.
Just run xtreg, fe and calculate the corresponding dummy
variable estimates.
Do a simple F test to test whether the coefficients are
identical.
Stata actually reports this test below the regression output of
xtreg, fe.
If the coefficients do not differ, you can use pooled OLS.

Panel Data
Hypothesis Testing 6-3

RE/FE or Pooled OLS? (ctd.)

. * Within or FE estimator
. xtreg lwage exp exp2 wks ed union ind occ, fe
note: ed omitted because of collinearity

Fixed-effects (within) regression Number of obs = 3696

Group variable: id Number of groups = 528

R-sq: within = 0.6558 Obs per group: min = 7

between = 0.0228 avg = 7.0
overall = 0.0427 max = 7

F(6,3162) = 1004.25
corr(u_i, Xb) = -0.9223 Prob > F = 0.0000

------------------------------------------------------------------------------
lwage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
exp | .1149389 .0026801 42.89 0.000 .1096841 .1201938
exp2 | -.0004347 .0000584 -7.44 0.000 -.0005491 -.0003202
wks | .0004693 .0006576 0.71 0.476 -.0008201 .0017587
ed | (omitted)
union | .0316998 .0159769 1.98 0.047 .0003736 .063026
ind | .0182395 .0160431 1.14 0.256 -.0132165 .0496954
occ | -.0113013 .0146455 -0.77 0.440 -.0400169 .0174144
_cons | 4.600501 .044283 103.89 0.000 4.513675 4.687328
-------------+----------------------------------------------------------------
sigma_u | 1.047369
sigma_e | .1534777
rho | .97897847 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(527, 3162) = 33.82 Prob > F = 0.0000

Panel Data
Hypothesis Testing 6-4

Random Effects or Fixed Effects?

If the ai ’s are correlated with the regressors, only the FE

estimator is consistent.
If the ai ’s are uncorrelated with the regressors, both the FE
and the RE estimators are consistent but the RE estimator is
more efficient.
The reason is simply that the RE estimator uses both the
within and the between variation of the data.

Panel Data
Hypothesis Testing 6-5

Random Effects or Fixed Effects?

The Hausman test uses the fact that both estimators are
consistent under the null hypothesis of no correlation.
Under the null hypothesis, the two estimates of β should thus
not differ systematically.
We can implement the test in Stata using hausman.
For the technical details of the test, you may consult Greene
(2008).

Panel Data
Hypothesis Testing 6-6

Random Effects or Fixed Effects? (ctd.)

. ******* 8. HAUSMAN TEST
.
. * Hausman test assuming RE estimator is fully efficient under null hypothesis
. hausman FE RE, sigmamore

---- Coefficients ----

| (b) (B) (b-B) sqrt(diag(V_b-V_B))
| FE RE Difference S.E.
-------------+----------------------------------------------------------------
exp | .1134669 .086851 .026616 .0013417
exp2 | -.0004198 -.0007766 .0003567 .0000299
wks | .0008245 .0010262 -.0002017 .00012
union | .0321123 .0615498 -.0294375 .0084284
ind | .0182378 .0124311 .0058066 .0093685
occ | -.0214375 -.0410889 .0196515 .0059918
------------------------------------------------------------------------------
b = consistent under Ho and Ha; obtained from xtreg
B = inconsistent under Ha, efficient under Ho; obtained from xtreg

Test: Ho: difference in coefficients not systematic

chi2(6) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= 1578.13
Prob>chi2 = 0.0000

Panel Data
Hypothesis Testing 6-7

Fixed Effects or First Differencing?

For T = 2, the two estimators are the same.

For T ≥ 2, the estimators are not the same but both are
unbiased and consistent.
For large N, and small T , FE is more efficient if the
idiosyncratic error it is serially uncorrelated.
Yet, there is no easy way to choose between the two
estimators and both are widely used in practice.

Panel Data
Extensions 7-1

Extension: Two-way-effects Model

A standard extension of the basic linear panel model is to allow the

intercept to vary not only over individuals but also over time:

yit = α + xit β + λt +ai + it . (31)

The time effects λt shift the intercept over time and affect all
micro-units uniformly.
Examples: business cycle movements, common trend in
wages...
In short panels, we usually include a full set of time dummies
among the xit ’s.
Panel Data
Extensions 7-2

Extension: Two-way-effects Model (ctd.)

Re-estimate the wage equation including a full set of time

dummies.
You can either generate the dummies by hand or use the i.
option of Stata.

Panel Data
Extensions 7-3

FE Estimates with Time Dummies

. ******* 9. TWO-WAY-EFFECTS MODEL
.
. * Within or FE estimator with time dummies
. xtreg lwage exp exp2 wks ed union ind occ i.t, fe
note: ed omitted because of collinearity
note: 7.t omitted because of collinearity

Fixed-effects (within) regression Number of obs = 4165

Group variable: id Number of groups = 595

R-sq: within = 0.6605 Obs per group: min = 7

between = 0.0287 avg = 7.0
overall = 0.0493 max = 7

F(11,3559) = 629.58
corr(u_i, Xb) = -0.9089 Prob > F = 0.0000

------------------------------------------------------------------------------
lwage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
exp | .1117254 .0026174 42.69 0.000 .1065936 .1168571
exp2 | -.000401 .0000546 -7.35 0.000 -.000508 -.000294
wks | .0006692 .0005994 1.12 0.264 -.000506 .0018444
ed | (omitted)
union | .0288605 .0148838 1.94 0.053 -.0003212 .0580422
ind | .0198261 .0154044 1.29 0.198 -.0103763 .0500285
occ | -.0193217 .0137094 -1.41 0.159 -.0462008 .0075574
|
t |
2 | -.0081074 .0081694 -0.99 0.321 -.0241246 .0079098
3 | .0258101 .0077694 3.32 0.001 .0105772 .041043
4 | .0286159 .0076429 3.74 0.000 .013631 .0436009
5 | .0235754 .0077717 3.03 0.002 .0083381 .0388128
6 | .007083 .0081633 0.87 0.386 -.0089221 .0230881
7 | (omitted)
|
_cons | 4.613659 .0431638 106.89 0.000 4.52903 4.698287
-------------+----------------------------------------------------------------
sigma_u | 1.0261585
sigma_e | .15150311
rho | .97866713 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(594, 3559) = 29.82 Prob > F = 0.0000
Panel Data
Extensions 7-4

RE Estimates with Time Dummies

.
. * RE estimator with time dummies
. xtreg lwage exp exp2 wks ed union ind occ i.t, re

Random-effects GLS regression Number of obs = 4165

Group variable: id Number of groups = 595

R-sq: within = 0.6600 Obs per group: min = 7

between = 0.3420 avg = 7.0
overall = 0.4279 max = 7

Random effects u_i ~ Gaussian Wald chi2(13) = 7163.49

corr(u_i, X) = 0 (assumed) Prob > chi2 = 0.0000

------------------------------------------------------------------------------
lwage | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
exp | .0289861 .0025233 11.49 0.000 .0240404 .0339317
exp2 | -.0004355 .00005 -8.72 0.000 -.0005334 -.0003375
wks | .0009599 .0005954 1.61 0.107 -.0002072 .0021269
ed | .0735711 .0049521 14.86 0.000 .0638651 .0832771
union | .0500422 .0134164 3.73 0.000 .0237465 .0763379
ind | .0409469 .0136322 3.00 0.003 .0142283 .0676656
occ | -.0339167 .0129552 -2.62 0.009 -.0593084 -.0085249
|
t |
2 | .0756715 .0089433 8.46 0.000 .058143 .0931999
3 | .1931833 .0092131 20.97 0.000 .175126 .2112406
4 | .279892 .0096431 29.03 0.000 .260992 .2987921
5 | .3588786 .0101782 35.26 0.000 .3389297 .3788275
6 | .4269871 .0108193 39.47 0.000 .4057817 .4481926
7 | .5040351 .0115335 43.70 0.000 .4814297 .5266404
|
_cons | 5.055152 .0812668 62.20 0.000 4.895871 5.214432
-------------+----------------------------------------------------------------
sigma_u | .30410708
sigma_e | .15150311
rho | .80115798 (fraction of variance due to u_i)
------------------------------------------------------------------------------

Panel Data
Outlook: Advanced Panel Data Analysis 8-1

Outlook: Advanced Panel Data Analysis

Two major topics in applied panel data analysis were not covered:
1. Panel IV estimation
2. Dynamic panel estimation

Panel Data
Outlook: Advanced Panel Data Analysis 8-2

Panel IV estimation

The fixed effects estimator (and related estimators) has some

major shortcomings:
It does not allow us to estimate the coefficients of
time-invariant regressors.
It does not solve the problem of time-varying omitted
variables.

Panel Data
Outlook: Advanced Panel Data Analysis 8-3

Panel IV estimation (ctd.)

Suppose you want to estimate the effect of job training on worker

productivity:

lscrapit = α + γhrsempit + ai + it , (32)

using the dataset scrap.dta.

i now refers to a single firm.
lscrap is the log of the scrap rate of a firm (a productivity
measure).
hrsemp is hours of job training per employee.

Panel Data
Outlook: Advanced Panel Data Analysis 8-4

Panel IV estimation (ctd.)

The unobserved fixed effect may capture, e.g., the (average)

ability of a firm’s workforce.
As hrsempit may well be correlated with ai (why?), we
estimate the fixed effects model:
(lscrapit − lscrap i ) = γ(hrsempit − hrsemp i ) + (it − i ). (33)

Panel Data
Outlook: Advanced Panel Data Analysis 8-5

xtset and xtdescribe the Data

. use "scrap.dta", clear

. xtset fcode year

panel variable: fcode (unbalanced)
time variable: year, 1987 to 1988
delta: 1 unit

. xtdescribe

fcode: 410523, 410563, ..., 419483 n = 47

year: 1987, 1988, ..., 1988 T = 2
Delta(year) = 1 unit
Span(year) = 2 periods
(fcode*year uniquely identifies each observation)

Distribution of T_i: min 5% 25% 50% 75% 95% max

1 2 2 2 2 2 2

Freq. Percent Cum. | Pattern

---------------------------+---------
45 95.74 95.74 | 11
1 2.13 97.87 | .1
1 2.13 100.00 | 1.
---------------------------+---------
47 100.00 | XX

Panel Data
Outlook: Advanced Panel Data Analysis 8-6

FE Estimates
. use "G:\Lhre\Panel Data\Wooldridge\scrap.dta", clear

. xtset fcode year

panel variable: fcode (unbalanced)
time variable: year, 1987 to 1988
delta: 1 unit

. xtreg lscrap hrsemp, fe

Fixed-effects (within) regression Number of obs = 92

Group variable: fcode Number of groups = 47

R-sq: within = 0.1193 Obs per group: min = 1

between = 0.0160 avg = 2.0
overall = 0.0243 max = 2

F(1,44) = 5.96
corr(u_i, Xb) = 0.0294 Prob > F = 0.0187

------------------------------------------------------------------------------
lscrap | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
hrsemp | -.0097174 .0039812 -2.44 0.019 -.017741 -.0016937
_cons | .6737459 .064658 10.42 0.000 .5434363 .8040555
-------------+----------------------------------------------------------------
sigma_u | 1.4400308
sigma_e | .43425379
rho | .91664268 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(46, 44) = 20.80 Prob > F = 0.0000

Panel Data
Outlook: Advanced Panel Data Analysis 8-7

Panel IV estimation (ctd.)

The fixed effects estimator will still be biased if

hrsempit − hrsemp t is correlated with the time-varying error
it .
A firm might, for instance, increase productivity by hiring
more skilled workers and simultaneously reduce job training.
In that case, we have to resort to Panel IV estimation...

Panel Data
Outlook: Advanced Panel Data Analysis 8-8

The IV Idea

We have to find an instrument z that is...

1. ...correlated with the endogenous variable x
(hrsempit − hrsemp i ).
2. ...uncorrelated with the error term  (it ).

z x y

Panel Data
Outlook: Advanced Panel Data Analysis 8-9

Panel IV Estimation in Stata

In our data, the dummy grant that indicates whether a firm

received a job training grant by the state may provide a valid
instrument (under which assumptions?).
The xtivreg command allows us to combine the fixed effect
transformation with IV estimation.
Replace hrsempl with (hrsempl = grant) to instruct Stata
that hrsempl should be instrumented by grant.

Panel Data
Outlook: Advanced Panel Data Analysis 8-10

Correlation between grant and hrsemp

. xtreg hrsemp grant, fe

Fixed-effects (within) regression Number of obs = 92

Group variable: fcode Number of groups = 47

R-sq: within = 0.4836 Obs per group: min = 1

between = 0.0801 avg = 2.0
overall = 0.2143 max = 2

F(1,44) = 41.21
corr(u_i, Xb) = -0.0875 Prob > F = 0.0000

------------------------------------------------------------------------------
hrsemp | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
grant | 26.01751 4.053008 6.42 0.000 17.84921 34.18581
_cons | 6.787258 1.441732 4.71 0.000 3.881638 9.692877
-------------+----------------------------------------------------------------
sigma_u | 14.822833
sigma_e | 11.816449
rho | .61143602 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(46, 44) = 3.11 Prob > F = 0.0001

Panel Data
Outlook: Advanced Panel Data Analysis 8-11

Panel IV Estimation Using xtivreg

. xtivreg lscrap (hrsemp=grant), fe

Fixed-effects (within) IV regression Number of obs = 92

Group variable: fcode Number of groups = 47

R-sq: within = 0.0783 Obs per group: min = 1

between = 0.0160 avg = 2.0
overall = 0.0243 max = 2

Wald chi2(1) = 153.69

corr(u_i, Xb) = -0.0202 Prob > chi2 = 0.0000

------------------------------------------------------------------------------
lscrap | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
hrsemp | -.0154088 .0058563 -2.63 0.009 -.026887 -.0039306
_cons | .7397372 .0821938 9.00 0.000 .5786403 .9008341
-------------+----------------------------------------------------------------
sigma_u | 1.4405516
sigma_e | .44422418
rho | .91316478 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(46,44) = 19.87 Prob > F = 0.0000
------------------------------------------------------------------------------
Instrumented: hrsemp
Instruments: grant
------------------------------------------------------------------------------

Panel Data
Outlook: Advanced Panel Data Analysis 8-12

Dynamic panel estimation

Panel data enables us to estimate parameters of dynamic models

with lagged dependent variables such as:

yit = α + υyi,t−1 + xit β + ai + it . (34)

Panel Data
Outlook: Advanced Panel Data Analysis 8-13

Dynamic panel estimation (ctd.)

Dynamic models are usually estimated in first-differences so as to

erase the unobserved effect ai :

(yit − yi,t−1 ) = (xit − xi,t−1 )β + υ(yi,t−1 − yi,t−2 )

+(it − i,t−1 ). (35)

As the lagged dependent variable is still correlated with the error

term, IV estimation is required to obtain consistent estimates.

Panel Data
Outlook: Advanced Panel Data Analysis 8-14

Dynamic panel estimation (ctd.)

In practice, appropriate lags of the dependent variable are

used as instruments.
In our example, yi,t−2 − yi,t−3 might be an appropriate
instrument for yi,t−1 − yi,t−2 .
A widely used estimator for dynamic panel models is the
Arellano-Bond estimator (xtabond in Stata).

Panel Data
Appendix 9-1

Omitted Variable Bias: An Example

Suppose that a person’s wage wi is a function of his education edi

and his IQ ai :
wi = α + β1 edi + β2 ai + ui . (36)
As you do not have on IQ, you instead estimate:

wi = α + β1 edi + ũi . (37)

Panel Data
Appendix 9-2

Omitted Variable Bias: An Example (ctd.)

Now suppose that IQ is related to education through the following

model:
ai = γ + δ1 edi + i . (38)
Then the regression that you actually run can be written as:

wi = α + β1 edi + β2 (γ + δ1 edi + i ) + ũi

= (α + β2 γ) + (β1 + β2 δ1 )edi + (ũi + β2 i ) (39)

Panel Data
Appendix 9-3

Omitted Variable Bias: An Example (ctd.)

The estimated effect of education on wages is thus β1 + β2 δ1 .

Education and IQ are usually positively correlated, i.e., δ1 > 0.
IQ should also have a positive effect on wages, i.e., β2 > 0.
It thus follows that our estimated effect of education is too
large as β1 + β2 δ1 > β1 .

Panel Data
Appendix 9-4

Why does xtreg, fe Report an Intercept?

Stata actually fits the model:

(yit − yi + y ) = α + (xit − x i + x )β + (it − i + a + ), (40)

where z = N −1 z is the ‘grand’ mean of some variable z and Stata

imposes the constraint a = N1 N
P
i=1 ai = 0.

Notice that the slope estimate β is not affected by the

‘transformation’.

Panel Data