STATA Commands for

Unobserved Effects Panel Data

John C Frain
21st February 2005

Contents
1 Introduction 1

2 Estimation using xtreg 9

3 Testing after xtreg 13

4 Prediction after xtreg 15

5 Faster estimation of alternative models using xtdata 15

6 More general error structures 18

1 Introduction
Panel data or cross-sectional timeseries are observations on a panel of i units or cases
over t time periods. Most panel data commands start with xt For an overview type
help xt. These notes present the annotated log of a STATA session demonstrating
the use of many of these commands. The data sets used are those used in the STATA
cross-sectional time series reference manual. This note should be regarded as an intro-
duction to that manual and to the STATA on-line help files which give comprehensive
descriptions of the facilities in STATA for cross-sectional time series analysis.
To obtain the optimum benefit from these notes I would recommend that one should
work through the STATA session with a copy of Wooldridge available for reference.
The emphasis here is on the implementation of the methods described in Chapter 10
of Wooldridge and no attempt is made to explain the theory set out there. Note the
different fonts used for comments (this font), instructions in comments (help xt) and
for computer input/output (help xt).

. help xt

---------------------------------------------------------------------------------
help for xt, iis, tis manual: [XT] xt
dialogs: iis tsset
---------------------------------------------------------------------------------

1
Cross-sectional time-series analysis

xt ... [, i(varname) t(varname) ... ]

iis [varname] [, clear]

tis [varname] [, clear]

Description

The xt series of commands provide tools for analyzing cross-sectional

time-series (panel) datasets:

help xtdes Describe pattern of xt data

help xtsum Summarize xt data
help xttab Tabulate xt data
help xtdata Faster specification searches with xt data

help xtline Line plots with xt data

help xtreg Fixed-, between- and random-effects, and

population-averaged linear models
help xtregar Fixed- and random-effects linear models with an AR(1)
disturbance
help xtgls Panel-data models using GLS
help xtpcse OLS or Prais-Winsten models with panel-corrected
standard errors
help xtrchh Hildreth-Houck random coefficients models
help xtivreg Instrumental variables and two-stage least squares for
panel-data models
help xtabond Arellano-Bond linear, dynamic panel data estimator

help xttobit Random-effects tobit models

help xtintreg Random-effects interval data regression models

help xtlogit Fixed-effects, random-effects, & population-averaged

logit models
help xtprobit Random-effects and population-averaged probit models
help xtcloglog Random-effects and population-averaged cloglog models

help xtpoisson Fixed-effects, random-effects, & population-averaged

Poisson models
help xtnbreg Fixed-effects, random-effects, & population-averaged
negative binomial models

help xtgee Population-averaged panel-data models using GEE

Each observation in a cross-sectional time-series (xt) dataset is an

observation on x for unit i at time t.

2
 iis is related to the i() option of the other xt commands. Command iis or
option i() sets the name of the variable corresponding to index i.

tis is similarly related to the t() option. Command tis or option t() sets
the name of the variable corresponding to index t.

Some xt commands use time-series operators in their internal calculations

and thus require that your data be tsset; see help tsset. For instance,
since xtabond uses time-series operators in its internal calculations, you
must tsset your data before using it. The particular help file will
indicate if tsset is required for the command.

Options

i(varname) specifies the variable name corresponding to index i. This must

be a single, numeric variable, although whether it takes on the values
1, 2, 3 or 1, 7, 9, etc., is irrelevant. (If the identifying variable
is a string, use egen’s group() function to make a numeric variable;
see help egen.)

t(varname) specifies the variable name corresponding to index t. This must

be a single, numeric variable.

clear removes the definition of i or t. For instance, typing "tis, clear"

makes Stata forget the identity of the t() variable.

Remarks

Once i() and t() have been specified, either by option or by the iis and
tis commands, they need not be specified again except to change the
variable’s identity.

iis and tis, without arguments, list the current name of the index
variable.

Example

An xt dataset:

pid yr_visit fev age sex height smokes

----------------------------------------------
1071 1991 1.21 25 1 69 0
1071 1992 1.52 26 1 69 0
1071 1993 1.32 28 1 68 0
1072 1991 1.33 18 1 71 1
1072 1992 1.18 20 1 71 1

3
 1072 1993 1.19 21 1 71 0

The other xt commands need to know the identities of the variables

identifying patient and time. You could type

. iis pid
. tis yr_visit

Also see

Manual: [XT] intro,

[XT] xt

Online: help for xtabond, xtcloglog, xtdata, xtdes, xtgee, xtgls,

xtintreg, xtivreg, xtline, xtlogit, xtnbreg, xtpcse, xtpoisson,
xtprobit, xtrchh, xtreg, xtregar, xtsum, xttab, xttobit; tsset

Load the data set nlswork.dta

. use nlswork, clear

. describe

Contains data National Longitudinal Survey.

Young Women 14-26 years of age
in 1968
obs: 28,534 18 Feb 2005 22:17
vars: 21
size: 1,055,758
-------------------------------------------------------------------------------
storage display value
variable name type format label variable label
-------------------------------------------------------------------------------
idcode int %8.0g NLS id
year byte %8.0g interview year
birth_yr byte %8.0g birth year
age byte %8.0g age in current year
race byte %8.0g 1=white, 2=black, 3=other
msp byte %8.0g 1 if married, spouse present
nev_mar byte %8.0g 1 if never yet married
grade byte %8.0g current grade completed
collgrad byte %8.0g 1 if college graduate
not_smsa byte %8.0g 1 if not SMSA
c_city byte %8.0g 1 if central city
south byte %8.0g 1 if south
ind_code byte %8.0g industry of employment
occ_code byte %8.0g occupation
union byte %8.0g 1 if union
wks_ue byte %8.0g weeks unemployed last year

4
ttl_exp float %9.0g total work experience
tenure float %9.0g job tenure, in years
hours int %8.0g usual hours worked
wks_work int %8.0g weeks worked last year
ln_wage float %9.0g ln(wage/GNP deflator)
-------------------------------------------------------------------------------
Sorted by: idcode year

To start one must set the indices i (units) and t (time). As already described this
can be done using the iis tis commands, i() t() options or the tsset command.
Examples of the commands follow.

. iis idcode

. tis year

.
. iis
i() is idcode

. tis
t() is year

. iis, clear

. iis
(i() has not been defined)

.
. tsset idcode year
panel variable: idcode, 1 to 5159
time variable: year, 68 to 88, but with gaps

. tsset
panel variable: idcode, 1 to 5159
time variable: year, 68 to 88, but with gaps

xtdes describes the participation pattern of panel data. We have 4711 women in
the survey. The maximum number of years over which any women is obsaerved is
15. the most common patern is participation in only the first year (136 or 2.89% are
observed in this patern). The bottom line of the table give the totals for participation
paterns not observed.

. xtdes

idcode: 1, 2, ..., 5159 n = 4711

year: 68, 69, ..., 88 T = 15
Delta(year) = 1; (88-68)+1 = 21
(idcode*year uniquely identifies each observation)

5
Distribution of T_i: min 5% 25% 50% 75% 95% max
1 1 3 5 9 13 15

Freq. Percent Cum. | Pattern

---------------------------+-----------------------
136 2.89 2.89 | 1....................
114 2.42 5.31 | ....................1
89 1.89 7.20 | .................1.11
87 1.85 9.04 | ...................11
86 1.83 10.87 | 111111.1.11.1.11.1.11
61 1.29 12.16 | ..............11.1.11
56 1.19 13.35 | 11...................
54 1.15 14.50 | ...............1.1.11
54 1.15 15.64 | .......1.11.1.11.1.11
3974 84.36 100.00 | (other patterns)
---------------------------+-----------------------
4711 100.00 | XXXXXX.X.XX.X.XX.X.XX

. xtdes, pattern(20)

idcode: 1, 2, ..., 5159 n = 4711

year: 68, 69, ..., 88 T = 15
Delta(year) = 1; (88-68)+1 = 21
(idcode*year uniquely identifies each observation)

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

1 1 3 5 9 13 15

Freq. Percent Cum. | Pattern

---------------------------+-----------------------
136 2.89 2.89 | 1....................
114 2.42 5.31 | ....................1
89 1.89 7.20 | .................1.11
87 1.85 9.04 | ...................11
86 1.83 10.87 | 111111.1.11.1.11.1.11
61 1.29 12.16 | ..............11.1.11
56 1.19 13.35 | 11...................
54 1.15 14.50 | ...............1.1.11
54 1.15 15.64 | .......1.11.1.11.1.11
49 1.04 16.68 | .........11.1.11.1.11
45 0.96 17.64 | ............1.11.1.11
43 0.91 18.55 | 1111.................
42 0.89 19.44 | ...1.................
40 0.85 20.29 | .....1.1.11.1.11.1.11
38 0.81 21.10 | ....11.1.11.1.11.1.11
38 0.81 21.91 | 111..................
34 0.72 22.63 | ..1111.1.11.1.11.1.11
31 0.66 23.29 | .................1...
30 0.64 23.92 | ..........1.1.11.1.11
29 0.62 24.54 | ...111.1.11.1.11.1.11

6
 3555 75.46 100.00 | (other patterns)
---------------------------+-----------------------
4711 100.00 | XXXXXX.X.XX.X.XX.X.XX

xtsum generalizes summarize by reporting means and standard for panel data. It
differs from summarize in that it decomposes the standard deviation into between and
within components.

. summ hours

Variable | Obs Mean Std. Dev. Min Max

-------------+--------------------------------------------------------
hours | 28467 36.55956 9.869623 1 168

. xtsum hours

Variable | Mean Std. Dev. Min Max | Observations

-----------------+--------------------------------------------+----------------
hours overall | 36.55956 9.869623 1 168 | N = 28467
between | 7.846585 1 83.5 | n = 4710
within | 7.520712 -2.154726 130.0596 | T-bar = 6.04395

. xtsum birth_yr /* Time invariant variable */

Variable | Mean Std. Dev. Min Max | Observations

-----------------+--------------------------------------------+----------------
birth_yr overall | 48.08509 3.012837 41 54 | N = 28534
between | 3.051795 41 54 | n = 4711
within | 0 48.08509 48.08509 | T-bar = 6.05689

xttab generalizes tabulate by performing one-way tabulations and by decompos-

ing counts into between and within components in panel data.

. summ msp

Variable | Obs Mean Std. Dev. Min Max

-------------+--------------------------------------------------------
msp | 28518 .6029175 .4893019 0 1

. tab msp

1 if |
married, |
spouse |
present | Freq. Percent Cum.
------------+-----------------------------------
0 | 11,324 39.71 39.71
1 | 17,194 60.29 100.00
------------+-----------------------------------

7
 Total | 28,518 100.00

. xttab msp

Overall Between Within

msp | Freq. Percent Freq. Percent Percent
----------+-----------------------------------------------------
0 | 11324 39.71 3113 66.08 55.06
1 | 17194 60.29 3643 77.33 71.90
----------+-----------------------------------------------------
Total | 28518 100.00 6756 143.41 64.14
(n = 4711)

xttrans is another generalization of tabulate. It reports changes in a single cate-

gorical variable over time.

. xttrans msp

1 if |
married, | 1 if married, spouse
spouse | present
present | 0 1 | Total
-----------+----------------------+----------
0 | 80.49 19.51 | 100.00
1 | 7.96 92.04 | 100.00
-----------+----------------------+----------
Total | 37.11 62.89 | 100.00

. xttrans msp, freq /* Does not normalize for missing time periods */

1 if |
married, | 1 if married, spouse
spouse | present
present | 0 1 | Total
-----------+----------------------+----------
0 | 7,697 1,866 | 9,563
| 80.49 19.51 | 100.00
-----------+----------------------+----------
1 | 1,133 13,100 | 14,233
| 7.96 92.04 | 100.00
-----------+----------------------+----------
Total | 8,830 14,966 | 23,796
| 37.11 62.89 | 100.00

.
. * Rectangularize the data
. fillin idcode year

8
. xttrans msp, freq

1 if |
married, | 1 if married, spouse
spouse | present
present | 0 1 | Total
-----------+----------------------+----------
0 | 6,792 1,446 | 8,238
| 82.45 17.55 | 100.00
-----------+----------------------+----------
1 | 813 10,954 | 11,767
| 6.91 93.09 | 100.00
-----------+----------------------+----------
Total | 7,605 12,400 | 20,005
| 38.02 61.98 | 100.00

xtline draws line plots for panel data.

2 Estimation using xtreg

The basic linear unobserved effects panel data model may is

yit = Xit β + ci + uit (1)

(For a full explanation of the symbols see Wooldridge page 251, etc.). In equation 1 ui
is the unit specific residual and differs between units but not across time within units.
Averaging equation 1 over time we get
ȳi = X̄i β + ci + ūi (2)
Subtracting equation 2 from equation 1 gives equation 3
(yit − ȳi ) = (Xit − X̄i )β + (uit − ūi ) (3)
These three equations form the basis for the various ways of estimating β.
xtreg ...,fe gives the fixed effects or within estimator of β and is derived from
equation 3. It is equivalent to performing OLS on equation pd3.
xtreg ...,be gives the between effects and corresponds to OLS estimation of
equation 2.
xtreg ...,re gives the random effects estimator and is a weighted average of
the within and between effects estimator. The random effects estimator is equivalent to
estimating

(yit − θȳi ) = (Xit − θ X̄i )β + (1 − θ)ci + (uit − θūi ) (4)

where θ is a function of σ2c and σ2u .
xtreg ...,mle produces maximum liklihood estimates of the random effects es-
timator.
For other options available with the xtreg command see the on-line help files or
the STATA manuals.

9
. tsset idcode year
panel variable: idcode, 1 to 5159
time variable: year, 68 to 88, but with gaps

.
. qui gen age2 = ageˆ2

. qui gen ttl_exp2 = ttl_expˆ2

. qui gen tenure2 = tenureˆ2

. gen byte black = race==2

.
. * OLS
. regress ln_w grade age* ttl_exp* tenure* black not_smsa south

Source | SS df MS Number of obs = 28091

-------------+------------------------------ F( 10, 28080) = 1681.47
Model | 2402.22796 10 240.222796 Prob > F = 0.0000
Residual | 4011.63592 28080 .142864527 R-squared = 0.3745
-------------+------------------------------ Adj R-squared = 0.3743
Total | 6413.86388 28090 .228332641 Root MSE = .37797

------------------------------------------------------------------------------
ln_wage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
grade | .0629238 .0010313 61.01 0.000 .0609024 .0649452
age | .038598 .003467 11.13 0.000 .0318025 .0453935
age2 | -.0007082 .0000563 -12.57 0.000 -.0008186 -.0005978
ttl_exp | .0211279 .002335 9.05 0.000 .0165511 .0257046
ttl_exp2 | .0004473 .0001246 3.59 0.000 .0002031 .0006916
tenure | .0473687 .0019626 24.14 0.000 .0435219 .0512156
tenure2 | -.002027 .0001338 -15.15 0.000 -.0022893 -.0017648
black | -.0699386 .0053207 -13.14 0.000 -.0803673 -.0595098
not_smsa | -.1720455 .0051675 -33.29 0.000 -.182174 -.161917
south | -.1003387 .0048938 -20.50 0.000 -.1099308 -.0907467
_cons | .2472833 .0493319 5.01 0.000 .1505903 .3439762
------------------------------------------------------------------------------

.
. * Fixed-effects model (within-group estimator)
. xtreg ln_w grade age* ttl_exp* tenure* black not_smsa south, fe

Fixed-effects (within) regression Number of obs = 28091

Group variable (i): idcode Number of groups = 4697

R-sq: within = 0.1727 Obs per group: min = 1

between = 0.3505 avg = 6.0

10
 overall = 0.2625 max = 15

F(8,23386) = 610.12
corr(u_i, Xb) = 0.1936 Prob > F = 0.0000

------------------------------------------------------------------------------
ln_wage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
grade | (dropped)
age | .0359987 .0033864 10.63 0.000 .0293611 .0426362
age2 | -.000723 .0000533 -13.58 0.000 -.0008274 -.0006186
ttl_exp | .0334668 .0029653 11.29 0.000 .0276545 .039279
ttl_exp2 | .0002163 .0001277 1.69 0.090 -.0000341 .0004666
tenure | .0357539 .0018487 19.34 0.000 .0321303 .0393775
tenure2 | -.0019701 .000125 -15.76 0.000 -.0022151 -.0017251
black | (dropped)
not_smsa | -.0890108 .0095316 -9.34 0.000 -.1076933 -.0703282
south | -.0606309 .0109319 -5.55 0.000 -.0820582 -.0392036
_cons | 1.03732 .0485546 21.36 0.000 .9421497 1.13249
-------------+----------------------------------------------------------------
sigma_u | .35562203
sigma_e | .29068923
rho | .59946283 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(4696, 23386) = 5.13 Prob > F = 0.0000

.
. * Between-group estimator
. xtreg ln_w grade age* ttl_exp* tenure* black not_smsa south, be

Between regression (regression on group means) Number of obs = 28091

Group variable (i): idcode Number of groups = 4697

R-sq: within = 0.1591 Obs per group: min = 1

between = 0.4900 avg = 6.0
overall = 0.3695 max = 15

F(10,4686) = 450.23
sd(u_i + avg(e_i.))= .3036114 Prob > F = 0.0000

------------------------------------------------------------------------------
ln_wage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
grade | .0607602 .0020006 30.37 0.000 .0568382 .0646822
age | .0323158 .0087251 3.70 0.000 .0152105 .0494211
age2 | -.0005997 .0001429 -4.20 0.000 -.0008799 -.0003194
ttl_exp | .0138853 .0056749 2.45 0.014 .0027598 .0250108
ttl_exp2 | .0007342 .0003267 2.25 0.025 .0000936 .0013747
tenure | .0698419 .0060729 11.50 0.000 .0579361 .0817476
tenure2 | -.0028756 .0004098 -7.02 0.000 -.0036789 -.0020722

11
 black | -.0564167 .0105131 -5.37 0.000 -.0770272 -.0358061
not_smsa | -.1860406 .0112495 -16.54 0.000 -.2080949 -.1639862
south | -.0993378 .010136 -9.80 0.000 -.1192091 -.0794665
_cons | .3339113 .1210434 2.76 0.006 .0966093 .5712133
------------------------------------------------------------------------------

.
. * Random-effects model (GLS estimator)
. xtreg ln_w grade age* ttl_exp* tenure* black not_smsa south, re

Random-effects GLS regression Number of obs = 28091

Group variable (i): idcode Number of groups = 4697

R-sq: within = 0.1715 Obs per group: min = 1

between = 0.4784 avg = 6.0
overall = 0.3708 max = 15

Random effects u_i ˜ Gaussian Wald chi2(10) = 9244.87

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

------------------------------------------------------------------------------
ln_wage | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
grade | .0646499 .0017811 36.30 0.000 .0611589 .0681408
age | .036806 .0031195 11.80 0.000 .0306918 .0429201
age2 | -.0007133 .00005 -14.27 0.000 -.0008113 -.0006153
ttl_exp | .0290207 .0024219 11.98 0.000 .0242737 .0337676
ttl_exp2 | .0003049 .0001162 2.62 0.009 .000077 .0005327
tenure | .039252 .0017555 22.36 0.000 .0358114 .0426927
tenure2 | -.0020035 .0001193 -16.80 0.000 -.0022373 -.0017697
black | -.0530532 .0099924 -5.31 0.000 -.0726379 -.0334685
not_smsa | -.1308263 .0071751 -18.23 0.000 -.1448891 -.1167634
south | -.0868927 .0073031 -11.90 0.000 -.1012066 -.0725788
_cons | .2387209 .0494688 4.83 0.000 .1417639 .335678
-------------+----------------------------------------------------------------
sigma_u | .25790313
sigma_e | .29069544
rho | .44043812 (fraction of variance due to u_i)
------------------------------------------------------------------------------

. estimates hold re

.
. * Random-effects model (Gaussian ML or fully iterated GLS estimator)
. xtreg ln_w grade age* ttl_exp* tenure* black not_smsa south, mle

Fitting constant-only model:

Iteration 0: log likelihood = -13690.161
Iteration 1: log likelihood = -12819.317
Iteration 2: log likelihood = -12662.039

12
Iteration 3: log likelihood = -12649.744
Iteration 4: log likelihood = -12649.614

Fitting full model:

Iteration 0: log likelihood = -8922.145
Iteration 1: log likelihood = -8853.6409
Iteration 2: log likelihood = -8853.4255
Iteration 3: log likelihood = -8853.4254

Random-effects ML regression Number of obs = 28091

Group variable (i): idcode Number of groups = 4697

Random effects u_i ˜ Gaussian Obs per group: min = 1

avg = 6.0
max = 15

LR chi2(10) = 7592.38
Log likelihood = -8853.4254 Prob > chi2 = 0.0000

------------------------------------------------------------------------------
ln_wage | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
grade | .0646093 .0017372 37.19 0.000 .0612044 .0680142
age | .0368531 .0031226 11.80 0.000 .030733 .0429732
age2 | -.0007132 .0000501 -14.24 0.000 -.0008113 -.000615
ttl_exp | .0288196 .0024143 11.94 0.000 .0240877 .0335515
ttl_exp2 | .000309 .0001163 2.66 0.008 .0000811 .0005369
tenure | .0394371 .0017604 22.40 0.000 .0359868 .0428875
tenure2 | -.0020052 .0001195 -16.77 0.000 -.0022395 -.0017709
black | -.0533394 .0097338 -5.48 0.000 -.0724172 -.0342615
not_smsa | -.1323433 .0071322 -18.56 0.000 -.1463221 -.1183644
south | -.0875599 .0072143 -12.14 0.000 -.1016998 -.0734201
_cons | .2390837 .0491902 4.86 0.000 .1426727 .3354947
-------------+----------------------------------------------------------------
/sigma_u | .2485556 .0035017 70.98 0.000 .2416925 .2554187
/sigma_e | .2918458 .001352 215.87 0.000 .289196 .2944956
-------------+----------------------------------------------------------------
rho | .4204033 .0074828 .4057959 .4351212
------------------------------------------------------------------------------
Likelihood-ratio test of sigma_u=0: chibar2(01)= 7339.84 Prob>=chibar2 = 0.000

3 Testing after xtreg

*

. /* After xtreg, re */
.
. estimates unhold re

13
.
. * Breusch & Pagan score test for random effects
. xttest0

Breusch and Pagan Lagrangian multiplier test for random effects:

ln_wage[idcode,t] = Xb + u[idcode] + e[idcode,t]

Estimated results:
| Var sd = sqrt(Var)
---------+-----------------------------
ln_wage | .2283326 .4778416
e | .0845038 .2906954
u | .066514 .2579031

Test: Var(u) = 0
chi2(1) = 14779.98
Prob > chi2 = 0.0000

.
. * Hausman specification test (compares fe and re)

.qui xtreg ln_wage grade age age2 ttl_exp ttl_exp2 tenure tenure2 not_smsa south,fe

F(4696, 23386) = 5.19 Prob > F = 0.0000

. estimates store fe

. qui xtreg ln_wage grade age age2 ttl_exp ttl_exp2 tenure tenure2 not_smsa south, re

. estimates store re

. hausman fe re

---- Coefficients ----

| (b) (B) (b-B) sqrt(diag(V_b-V_B))
| fe re Difference S.E.
-------------+----------------------------------------------------------------
age | .0359987 .0363062 -.0003075 .0013183
age2 | -.000723 -.000705 -.000018 .0000184
ttl_exp | .0334668 .0292321 .0042347 .0017085
ttl_exp2 | .0002163 .0002946 -.0000783 .0000529
tenure | .0357539 .0390983 -.0033444 .0005789
tenure2 | -.0019701 -.0020014 .0000313 .0000372
not_smsa | -.0890108 -.1268961 .0378853 .0063038
south | -.0606309 -.094716 .0340851 .008259
------------------------------------------------------------------------------
b = consistent under Ho and Ha; obtained from xtreg
B = inconsistent under Ha, efficient under Ho; obtained from xtreg

14
 Test: Ho: difference in coefficients not systematic

chi2(8) = (b-B)’[(V_b-V_B)ˆ(-1)](b-B)
= 142.53
Prob>chi2 = 0.0000

4 Prediction after xtreg

.
. * Syntax:
. * predict [type] newvarname [if exp] [in range] [, statistic nooffset]
. * where statistic is:
. * xb fitted values (the default)
. * stdp standard error of the fitted values
. * ue the combined residuals
. * xbu prediction, including effect
. * u the fixed effect component
. * e the random error component
.
. predict xb /* computes the linear predictor (the default) */
(option xb assumed; fitted values)
(443 missing values generated)

. predict stdp, stdp

(443 missing values generated)

5 Faster estimation of alternative models using xtdata

xtdata varlist ... produces a converted dataset of the variables specified or, if
varlist is not specified, all the variables in the data. Once converted, Stata’s ordinary
regress command may be used to perform various panel data regressions more quickly
than use xtreg. Before using xdata you must eliminate any variables that you do not
intend to use and that have missing values. After converting the data, with xdata you
may form linear transformations of the regressors. All nonlinear transformations of the
data must be done before conversion. .

. xtdata ln_w grade age* ttl_exp* tenure* black not_smsa south, fe clear

. regress ln_w grade age ttl_exp tenure black not_smsa south

Source | SS df MS Number of obs = 28091

15
-------------+------------------------------ F( 6, 28084) = 820.44
Model | 356.233455 6 59.3722424 Prob > F = 0.0000
Residual | 2032.33275 28084 .072366214 R-squared = 0.1491
-------------+------------------------------ Adj R-squared = 0.1490
Total | 2388.5662 28090 .085032617 Root MSE = .26901

------------------------------------------------------------------------------
ln_wage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
grade | .0375399 5.04e+08 0.00 1.000 -9.88e+08 9.88e+08
age | -.0026787 .0007876 -3.40 0.001 -.0042224 -.0011349
ttl_exp | .0287709 .0013209 21.78 0.000 .0261819 .0313599
tenure | .0114355 .0008422 13.58 0.000 .0097847 .0130863
black | (dropped)
not_smsa | -.0921689 .0088194 -10.45 0.000 -.1094553 -.0748825
south | -.0633396 .0101132 -6.26 0.000 -.083162 -.0435172
_cons | 1.121064 6.32e+09 0.00 1.000 -1.24e+10 1.24e+10
------------------------------------------------------------------------------

. regress ln_w grade age* ttl_exp* tenure* black not_smsa south

Source | SS df MS Number of obs = 28091

-------------+------------------------------ F( 9, 28081) = 651.21
Model | 412.443881 9 45.8270979 Prob > F = 0.0000
Residual | 1976.12232 28081 .07037222 R-squared = 0.1727
-------------+------------------------------ Adj R-squared = 0.1724
Total | 2388.5662 28090 .085032617 Root MSE = .26528

------------------------------------------------------------------------------
ln_wage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
grade | -.0147384 4.97e+08 -0.00 1.000 -9.75e+08 9.75e+08
age | .0359987 .0030904 11.65 0.000 .0299414 .0420559
age2 | -.000723 .0000486 -14.88 0.000 -.0008183 -.0006277
ttl_exp | .0334668 .0027061 12.37 0.000 .0281626 .0387709
ttl_exp2 | .0002163 .0001166 1.86 0.064 -.0000122 .0004448
tenure | .0357539 .0016871 21.19 0.000 .0324471 .0390607
tenure2 | -.0019701 .0001141 -17.27 0.000 -.0021937 -.0017465
black | (dropped)
not_smsa | -.0890108 .0086984 -10.23 0.000 -.10606 -.0719616
south | -.0606309 .0099763 -6.08 0.000 -.0801849 -.0410769
_cons | 1.222086 6.23e+09 0.00 1.000 -1.22e+10 1.22e+10
------------------------------------------------------------------------------

.
. xtdata ln_w grade age* ttl_exp* tenure* black not_smsa south, re ratio(.95) clear

------------------- theta --------------------

min 5% median 95% max
0.2750 0.2750 0.5741 0.7198 0.7377

16
. * (ratio is the ratio of the std. dev. of the individual effect and the
. * random error)
. regress ln_w constant grade age ttl_exp tenure black not_smsa south, noconstant

Source | SS df MS Number of obs = 28091

-------------+------------------------------ F( 6, 28085) =27121.45
Model | 11775.6413 6 1962.60688 Prob > F = 0.0000
Residual | 2032.33275 28085 .072363637 R-squared = 0.8528
-------------+------------------------------ Adj R-squared = 0.8528
Total | 13807.974 28091 .491544411 Root MSE = .269

------------------------------------------------------------------------------
ln_wage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
constant | (dropped)
grade | .1269649 .0013954 90.99 0.000 .1242299 .1296999
age | -.0026787 .0007876 -3.40 0.001 -.0042224 -.001135
ttl_exp | .0287709 .0013209 21.78 0.000 .026182 .0313599
tenure | .0114355 .0008422 13.58 0.000 .0097847 .0130863
black | (dropped)
not_smsa | -.0921689 .0088192 -10.45 0.000 -.109455 -.0748828
south | -.0633396 .010113 -6.26 0.000 -.0831616 -.0435175
------------------------------------------------------------------------------

. regress ln_w constant grade age* ttl_exp* tenure* black not_smsa south, noconstant

Source | SS df MS Number of obs = 28091

-------------+------------------------------ F( 9, 28082) =18682.05
Model | 11831.8517 9 1314.65019 Prob > F = 0.0000
Residual | 1976.12232 28082 .070369714 R-squared = 0.8569
-------------+------------------------------ Adj R-squared = 0.8568
Total | 13807.974 28091 .491544411 Root MSE = .26527

------------------------------------------------------------------------------
ln_wage | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
constant | (dropped)
grade | .0827449 .0035478 23.32 0.000 .0757911 .0896987
age | .0359987 .0030903 11.65 0.000 .0299415 .0420558
age2 | -.000723 .0000486 -14.88 0.000 -.0008183 -.0006277
ttl_exp | .0334668 .0027061 12.37 0.000 .0281627 .0387708
ttl_exp2 | .0002163 .0001166 1.86 0.064 -.0000122 .0004447
tenure | .0357539 .0016871 21.19 0.000 .0324472 .0390606
tenure2 | -.0019701 .0001141 -17.27 0.000 -.0021937 -.0017465
black | (dropped)
not_smsa | -.0890108 .0086982 -10.23 0.000 -.1060597 -.0719619
south | -.0606309 .0099761 -6.08 0.000 -.0801845 -.0410772
------------------------------------------------------------------------------

17
6 More general error structures
xtpcse and xtgls estimate linear panel data models using feasible GLS. xtpcse
computes OLS estimates with panel-corrected standard errrors, while xtgls computes
feasible GLS estimates. These commands allow estimation in the presence of AR(1)
autocorrelation within panels, as well as heteroskedasticity or cross-sectional correla-
tion across panels. In the case of cross-sectional correlation, xtgls requires T > n.

. use invest2, clear

. tsset company time

panel variable: company, 1 to 5
time variable: time, 1 to 20

.
. * OLS with panel-corrected standard errors
. xtpcse invest market stock /* Heterosk. and contemp. correlation (the default) */

Linear regression, correlated panels corrected standard errors (PCSEs)

Group variable: company Number of obs = 100

Time variable: time Number of groups = 5
Panels: correlated (balanced) Obs per group: min = 20
Autocorrelation: no autocorrelation avg = 20
max = 20
Estimated covariances = 15 R-squared = 0.7789
Estimated autocorrelations = 0 Wald chi2(2) = 755.43
Estimated coefficients = 3 Prob > chi2 = 0.0000

------------------------------------------------------------------------------
| Panel-corrected
| Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
market | .1050854 .0083183 12.63 0.000 .0887818 .1213891
stock | .3053655 .0330427 9.24 0.000 .240603 .3701281
_cons | -48.02974 10.81437 -4.44 0.000 -69.2255 -26.83397
------------------------------------------------------------------------------

. xtpcse invest market stock, corr(ar1) /* Heterosk., contemp. correlation and AR(1) autoc
(note: estimates of rho outside [-1,1] bounded to be in the range [-1,1])

Prais-Winsten regression, correlated panels corrected standard errors (PCSEs)

Group variable: company Number of obs = 100

Time variable: time Number of groups = 5
Panels: correlated (balanced) Obs per group: min = 20
Autocorrelation: common AR(1) avg = 20
max = 20
Estimated covariances = 15 R-squared = 0.5909

18
Estimated autocorrelations = 1 Wald chi2(2) = 124.32
Estimated coefficients = 3 Prob > chi2 = 0.0000

------------------------------------------------------------------------------
| Panel-corrected
| Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
market | .093367 .0125705 7.43 0.000 .0687294 .1180046
stock | .354706 .0571221 6.21 0.000 .2427486 .4666633
_cons | -39.39866 40.22722 -0.98 0.327 -118.2426 39.44524
-------------+----------------------------------------------------------------
rho | .8530976
------------------------------------------------------------------------------

. xtpcse invest market stock, corr(psar1) rhotype(tscorr)

Prais-Winsten regression, correlated panels corrected standard errors (PCSEs)

Group variable: company Number of obs = 100

Time variable: time Number of groups = 5
Panels: correlated (balanced) Obs per group: min = 20
Autocorrelation: panel-specific AR(1) avg = 20
max = 20
Estimated covariances = 15 R-squared = 0.8734
Estimated autocorrelations = 5 Wald chi2(2) = 483.87
Estimated coefficients = 3 Prob > chi2 = 0.0000

------------------------------------------------------------------------------
| Panel-corrected
| Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
market | .0976686 .009442 10.34 0.000 .0791627 .1161746
stock | .3726526 .0384121 9.70 0.000 .2973662 .447939
_cons | -46.95183 16.78803 -2.80 0.005 -79.85576 -14.0479
------------------------------------------------------------------------------
rhos = .4735903 .704354 .8977688 .5249498 .8558518
------------------------------------------------------------------------------

. xtpcse invest market stock, hetonly /* Heterosk., no contemp. correlation */

Linear regression, heteroskedastic panels corrected standard errors

Group variable: company Number of obs = 100

Time variable: time Number of groups = 5
Panels: heteroskedastic (balanced) Obs per group: min = 20
Autocorrelation: no autocorrelation avg = 20
max = 20
Estimated covariances = 5 R-squared = 0.7789
Estimated autocorrelations = 0 Wald chi2(2) = 720.01
Estimated coefficients = 3 Prob > chi2 = 0.0000

19
------------------------------------------------------------------------------
| Het-corrected
| Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
market | .1050854 .0090625 11.60 0.000 .0873232 .1228476
stock | .3053655 .0409468 7.46 0.000 .2251113 .3856198
_cons | -48.02974 14.20367 -3.38 0.001 -75.86841 -20.19106
------------------------------------------------------------------------------

. xtpcse invest market stock, hetonly corr(ar1) /* Heterosk. and AR(1) autocorr., no cont
(note: estimates of rho outside [-1,1] bounded to be in the range [-1,1])

Prais-Winsten regression, heteroskedastic panels corrected standard errors

Group variable: company Number of obs = 100

Time variable: time Number of groups = 5
Panels: heteroskedastic (balanced) Obs per group: min = 20
Autocorrelation: common AR(1) avg = 20
max = 20
Estimated covariances = 5 R-squared = 0.5909
Estimated autocorrelations = 1 Wald chi2(2) = 120.57
Estimated coefficients = 3 Prob > chi2 = 0.0000

------------------------------------------------------------------------------
| Het-corrected
| Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
market | .093367 .0128727 7.25 0.000 .0681369 .1185971
stock | .354706 .0587917 6.03 0.000 .2394763 .4699357
_cons | -39.39866 37.19875 -1.06 0.290 -112.3069 33.50954
-------------+----------------------------------------------------------------
rho | .8530976
------------------------------------------------------------------------------

.
. * Feasible GLS
. xtgls invest market stock, panel(iid) corr(indep) nmk

Cross-sectional time-series FGLS regression

Coefficients: generalized least squares

Panels: homoskedastic
Correlation: no autocorrelation

Estimated covariances = 1 Number of obs = 100

Estimated autocorrelations = 0 Number of groups = 5
Estimated coefficients = 3 Time periods = 20
Wald chi2(2) = 341.63
Log likelihood = -624.9928 Prob > chi2 = 0.0000

20
------------------------------------------------------------------------------
invest | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
market | .1050854 .0113778 9.24 0.000 .0827853 .1273855
stock | .3053655 .0435078 7.02 0.000 .2200918 .3906393
_cons | -48.02974 21.48016 -2.24 0.025 -90.13009 -5.929387
------------------------------------------------------------------------------

. * (same as regress ..., nmk uses n - k to normalize the RSS)

.
. xtgls invest market stock, i(company) panel(hetero) /* Heterosk., no contemp. correlatio

Cross-sectional time-series FGLS regression

Coefficients: generalized least squares

Panels: heteroskedastic
Correlation: no autocorrelation

Estimated covariances = 5 Number of obs = 100

Estimated autocorrelations = 0 Number of groups = 5
Estimated coefficients = 3 Time periods = 20
Wald chi2(2) = 865.38
Log likelihood = -570.1305 Prob > chi2 = 0.0000

------------------------------------------------------------------------------
invest | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
market | .0949905 .007409 12.82 0.000 .0804692 .1095118
stock | .3378129 .0302254 11.18 0.000 .2785722 .3970535
_cons | -36.2537 6.124363 -5.92 0.000 -48.25723 -24.25017
------------------------------------------------------------------------------

. xtgls invest market stock, panel(corr) corr(ar1) /* Heterosk., contemp. correlation and

Cross-sectional time-series FGLS regression

Coefficients: generalized least squares

Panels: heteroskedastic with cross-sectional correlation
Correlation: common AR(1) coefficient for all panels (0.8651)

Estimated covariances = 15 Number of obs = 100

Estimated autocorrelations = 1 Number of groups = 5
Estimated coefficients = 3 Time periods = 20
Wald chi2(2) = 153.66
Log likelihood = -491.3974 Prob > chi2 = 0.0000

------------------------------------------------------------------------------
invest | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------

21
 market | .0745101 .0091391 8.15 0.000 .0565978 .0924225
stock | .3150971 .0447361 7.04 0.000 .2274158 .4027783
_cons | -2.770019 13.78308 -0.20 0.841 -29.78435 24.24431
------------------------------------------------------------------------------

. matrix list e(Sigma)

symmetric e(Sigma)[5,5]
_ee _ee2 _ee3 _ee4 _ee5
_ee 5223.2164
_ee2 -101.18031 302.56293
_ee3 37.474924 146.75692 2578.9016
_ee4 -173.62446 57.848228 619.37254 262.40269
_ee5 -1093.8519 111.5931 537.76577 704.40596 8835.32

. xtgls invest market stock, panel(corr) corr(ar1) rhotype(dw)

Cross-sectional time-series FGLS regression

Coefficients: generalized least squares

Panels: heteroskedastic with cross-sectional correlation
Correlation: common AR(1) coefficient for all panels (0.8179)

Estimated covariances = 15 Number of obs = 100

Estimated autocorrelations = 1 Number of groups = 5
Estimated coefficients = 3 Time periods = 20
Wald chi2(2) = 203.26
Log likelihood = -495.6259 Prob > chi2 = 0.0000

------------------------------------------------------------------------------
invest | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
market | .0758752 .0090285 8.40 0.000 .0581796 .0935708
stock | .3289528 .0409971 8.02 0.000 .2485999 .4093056
_cons | -10.08235 11.9502 -0.84 0.399 -33.50432 13.33961
------------------------------------------------------------------------------

. xtgls invest market stock, panel(corr) corr(psar1)

Cross-sectional time-series FGLS regression

Coefficients: generalized least squares

Panels: heteroskedastic with cross-sectional correlation
Correlation: panel-specific AR(1)

Estimated covariances = 15 Number of obs = 100

Estimated autocorrelations = 5 Number of groups = 5
Estimated coefficients = 3 Time periods = 20
Wald chi2(2) = 331.55
Log likelihood = -484.6178 Prob > chi2 = 0.0000

22
------------------------------------------------------------------------------
invest | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
market | .0820264 .0081381 10.08 0.000 .066076 .0979767
stock | .3800689 .0313874 12.11 0.000 .3185508 .441587
_cons | -11.51848 12.69055 -0.91 0.364 -36.39151 13.35455
------------------------------------------------------------------------------

.
.
end of do-file

. exit, clear

23