Department of Economics

Econometric Analysis of Panel Data

Professor William Greene Phone: 212.998.0876
Office: KMC 7-90 Home page:people.stern.nyu.edu/wgreene
Email: [email protected]
URL for course web page:
people.stern.nyu.edu/wgreene/Econometrics/PanelDataEconometrics.htm

Final Examination: Spring 2017

This is a ‘take home’ examination. Today is Tuesday, May 2, 2017. Your answers are due on
Friday, May 12, 2017. You may use any resources you wish – textbooks, computer, the web, etc. – but
please work alone and submit only your own answers to the questions.

The five parts of the exam are weighted as follows:

Part I. Literature 20
Part II. The Mundlak Estimator 20
Part III. Panel Data Regressions 50
Part IV. Binary Choice Models 50
Part V. A Loglinear Model 60

Note, in parts of the exam in which you are asked to report the results of computation, please filter your
response so that you present the numerical results as part of an organized discussion of the question. Do not
submit long, unannotated pages of computer output. Some of the parts require you to do some
computations. Use Stata, R, NLOGIT, MatLab or any other software you wish to use.

Part I. Literature

Locate a published study in a field that interests you that uses a panel data based methodology.
Describe in no more than one page the study, the estimation method(s) used and the conclusion(s) reached
by the author(s).

Part II. The Mundlak Approach in Estimation

Many recent studies have revived Mundlak’s approach to modeling common effects in linear
regression and nonlinear models. Describe in detail the standard common effects models. How is the
Mundlak estimator motivated? How is it employed? Show how the estimator provides a constructive test
for fixed vs. random effects.
Part III. Panel Data Regressions

The course website contains an abbreviated version of the WHO health outcomes data set,

http://people.stern.nyu.edu/wgreene/Econometrics/WHO-balanced-panel.csv

and as an nlogit project,

http://people.stern.nyu.edu/wgreene/Econometrics/WHO-balanced-panel.lpj

The csv file is a text, comma delimited file that should be directly readable by other programs such as Stata
and R. The original data set contained 840 observations as an unbalanced panel for 191 countries. It also
contained data for some internal political districts such as the 24 states of Mexico and the provinces of
Canada and Australia. This panel retains the data for the 140 countries that contain all 5 years of data. The
variables in the file are

COUNTRY = Country name (text)

ID, STRATUM = Country ID. Ignore STRATUM
YEAR = 1993, …, 1997
COMP and LOGCOMP = WHO health outcome measure and its log
DALE and LOGDALE = WHO life expectance and its log
EDUC, LOGEDUC, LOGEDUC2 = Education, log and square of log
HLTHEXP, LOGHEXP, LOGHEXP2 = Health expenditure, log and square of log
PUBTHE = Share of health expenditure paid by government
LOGED_EX = LOGHEXP * LOGEDUC
GINI = Gini coefficient income distribution
TROPICS = Dummy variable for tropical country
POPDEN, LOGPOPDN = Population density, people per square kilometer and log
GDPC, LOGGDPC = Per capita GDP and log
T93,…,T97 = Year dummy variables
GEFF = World bank measure of government effectiveness
VOICE = World Bank measure of political efficacy
OECD = OECD member dummy variable
MEANLCMP = Country mean of log COMP
MEANLHC = Country mean of log EDUC
MEANLHC2 = Country mean of log EDUC squared
MEANLEXP = Country mean of log HEXP

Note that COMP, DALE, EDUC and HLTHEXP are time varying, but all other measured variables are
time invariant.

The WHO model originally specified was

yit =  + 1x1,it + 2x2,it + 11x1,it2 + 22x2,it2 + 12x1,itx2,it + it

where
y = logCOMP, x1 = logEDUC, x2 = logHEXP.

Call this Model A. This is a translog production function. The authors found that the values of 12 implied
a nonconcave production function, and fixed 22 and 12 both to zero in their final presentation. Call this
restricted model Model B.
a. Fit the “pooled” model and report your results.

b. Using the pooled model, test the null hypothesis of Model B against the alternative Model A.

c. Using the formulation of Model B, fit a random effects model and a fixed effects model. Use your
estimation results to decide which is the preferable model. If you find that neither panel data model is
preferred to the pooled model, show how you reached that conclusion. As part of the analysis, test the
hypothesis that there are no “country effects.”

d. Using the Mundlak approach, determine which model, fixed or random effects is preferred.

e. Assuming that there are “latent individual (county) effects,” the asymptotic covariance matrix that is
computed for the pooled estimator, s2(X′X)-1, is inappropriate. What estimator can be computed for the
covariance matrix of the pooled estimator that will give appropriate standard errors?

f. The hypothesis of constant returns to scale in the translog model (Model A) would be

H0: 1 + 2 = 1 and 11 + 22 + 212 = 0

Test this hypothesis in the context of Model A.

g. The 2004 Health Economics paper by Greene argued that WHO did not handle the obvious heterogeneity
across countries appropriately. Variables GINI, TROPICS, logPOPDN, logGDPC, GEFF, VOICE, OECD
all capture dimensions of this heterogeneity. Extend the random effects model to include some (or all) of
these variables and test the hypothesis that they significantly add to the explanatory power of the model.

h. Are there “time effects” in the data. One approach find out would be to add the time variables (less one
of them) to the preferred regression model and test for their joint significance. A second approach would
be to use a CHOW test to test for homogeneity of the regression model over the 5 years. Test the
homogeneity assumption using your preferred pooled model.
Part IV. Binary Choice Models

The course website describes the “German Manufacturing Innovation Data.” The actual data are not
published on the course website. We will use them for purposes of this exercise, however. You can obtain
them by downloading either a csv file,

http://people.stern.nyu.edu/wgreene/Econometrics/probit-panel.csv

or an nlogit project file,

http://people.stern.nyu.edu/wgreene/Econometrics/probit-panel.lpj

This data set contains 1,270 firms and 5 years of data for 6,350 observations in total – a balanced panel. The
variables that you need for this exercise are described in the data sets area of the course home page,

http://people.stern.nyu.edu/wgreene/Econometrics/PanelDataSets.htm

(The csv file can easily be ported to other software such as R, SAS and Stata.) I am interested in a binary
choice model for the innovation variable, IP. You will fit your model using at least three of the independent
variables in the data set. With respect to the model you specify,

A. THEORY
(a) If you fit a pooled logit model, there is the possibility that you might be ignoring unobserved heterogeneity
(effects). Wooldridge argues that when one fits a probit model while ignoring unobserved heterogeneity, the
raw coefficient estimator (MLE) is inconsistent, but the quantity of interest, the “Average Partial Effects” might
well be estimated appropriately. Explain in detail what he has in mind here.
(b) Suppose we were to estimate a “fixed effects” probit model by “brute force,” just by including the 1,270
dummy variables needed to create the empirical model. What would the properties of the resulting estimator
likely be? What is “the incidental parameters problem?”
(c) How would I proceed to use Chamberlain’s estimator to obtain a consistent slope estimator for the fixed
effects logit model.
(d) Describe in detail how to fit a random effects logit model using quadrature and using simulation for the part
of the computations where they would be necessary, under the assumption that the effects are uncorrelated with
the other included exogenous variables.
(e) Using the random effects logit model that you described in part (d), describe how you would test the
hypothesis that the same logit model applies to the four different sectors in the data set
(CONSGOOD,FOOD,RAWMTL,INVGOOD).

B. PRACTICE
(a) Fit a pooled probit model using your specification. Provide all relevant estimation results. (Please condense
and organize the results in a readable form.)
(b) Fit a random effects probit model.
(c) Use the Mundlak (correlated random effects) approach to approximate a fixed effects model. Recall this
means adding the group means of the time varying variables to the model, then using a random effects model.
(d) Note the difference between the estimates in (b) and (c). Which do you think is appropriate? Explain.

Tip for nlogit users: You can use

CREATE ; new variable = GroupMean(variable,pds=5)$
To obtain the group means you need for a variable.
Part V. A Loglinear Model

This semester, we have examined several ‘loglinear models,’ including the logit model for binary
choice, Poisson and negative binomial models for counts and the exponential model for a continuous
nonnegative random variable. We will now examine one more loglinear model. The nonnegative,
continuous random variable y|x has a Weibull distribution:

f ( y | x )   i P yiP 1 exp   i yi P  , y  0, P  0,
i  exp(x i ). The first element of x i is a constant term, 1.
(We examined a version of this model in Assignment 5.) Estimation and analysis is based on a
sample of N observations on yi,xi. The conditional mean function is

1  P 1   P 1 
E[yi|xi] =    exp(xi )   (Note the minus sign.)
i  P   P 

The variables used in the regressions are described below.

--------+---------------------------------------------------------------------
Variable| Mean Std.Dev. Minimum Maximum Cases Missing
--------+---------------------------------------------------------------------
INCOME| .352135 .176857 .001500 3.067100 27326 0
logINCOM| -1.157442 .491452 -6.502290 1.120732 27326 0
AGE| 43.52569 11.33025 25.0 64.0 27326 0
EDUC| 11.32063 2.324885 7.0 18.0 27326 0
HSAT| 6.785662 2.293725 0.0 10.0 27326 0
MARRIED| .758618 .427929 0.0 1.0 27326 0
HHKIDS| .402730 .490456 0.0 1.0 27326 0
--------+---------------------------------------------------------------------

The data set is a panel. There are 7,293 groups with group sizes ranging from 1 to 7. This exercise
will examine a variety of regression formulations. I have done the estimation for you; the results
appear below. Some of the questions will involve a small amount of ancillary computation.

A. I propose to estimate the parameters (P,α,) by maximum likelihood. The results are shown in
regression 1 below. Derive the log likelihood function, likelihood equations and Hessian. Show
precisely how to use Newton’s method to estimate the parameters. How will you obtain asymptotic
standard errors for your estimator? Test the hypothesis of ‘the model.’ That is, test the hypothesis
that all of the coefficients are equal to zero (except the constant term) using the likelihood ratio test.

B. There are several interesting special cases of the Weibull model. If P = 1, the model reduces to
the exponential model discussed in class. We considered three different ways to test a parametric
restriction such as this, Wald, Likelihood ratio and LM tests. Using the results of regressions 1, 2
and 3 below, carry out the three tests. Do the results of the three tests agree?
C. The conditional mean function shown above suggests a nonlinear least squares approach. Note
that the conditional mean function can be written

  P 1 1
E[ y | x]  exp log       x   exp(x)
  P  
where  is the constant term and  is the remaining parameters, and x1 is all variables not including
the constant term. Thus,  and  have different constant terms, and are otherwise the same. The
nonlinear least squares results are shown in regression 4. How do the two results compare? We now
have two possible estimators of . In theoretical terms, which is better, MLE or NLS? Why? Do the
empirical results support your argument?

D. The likelihood equations for estimation of (P,) imply that E[yP|x] = 1/. Prove this result.

E. Derive the partial effects for the Weibull conditional mean function, E[y|x]/x. Compute the
partial effects at the means of the data. Hint: ((P+1)/P) for the P in regression 1 equals .88562.
How would you obtain standard errors for your estimated partial effects? Explain in detail.

F. Regression 5 presents linear least squares results for the regression of –y on x. (The minus sign on
y changes the sign of the coefficients so they will be comparable to the earlier results.) How do these
results compare to the MLEs in part A? How do they compare to the results in part E? Why would
they resemble the results in part E?

G. The log of a Weibull distributed variable has a type 1 extreme value distribution. The expected
value of logy is -x + , where  is the Euler-Mascheroni constant, 0.57721566…. Regression 6
presents the results of linear regression of –logy on x. Which other result should these resemble? Do
they?

H. Since these are panel data, it is appropriate to rebuild the model to accommodate the unobserved
heterogeneity. Explain the difference between fixed and random effects models. How would they
appear in the loglinear model formulated here?

I. Regressions 7 and 8 show FEM and REM.

(1) What is the incidental parameters problem? Would the result apply to the model shown in (7)?
(2) Show how the parameters of the random effects model in regression 8 are computed. I.e.,
describe how the maximum simulated likelihood estimator is computed.
(3) Regression 9 presents estimates or a random effects model that also contains the group means of
the regressors. As noted earlier, this Mundlak style treatment helps to distinguish the FE and RE
specifications. Based on the results given, which appears to be the preferable model, FE or RE?

J. Some have argued that marital status might be endogenous in an income equation when there are
households that have two working people. (You probably thought people married for love.) To
investigate in the present model, I will use a control function approach. Regression 10 presents a
probit eqution for marital status based on age, education, gender and whether the household head has
a white collar job. The variable GENRES is the generalized residual from this model,
GENRES = q(x)/(qx) where q = 2Married – 1. The expected value of GENRES is zero, and
since it is the derivative of logL with respect to the constant term, it will sum to zero in the sample. I
am going to use GENRES as a control function? What is a control function, and why will I use it in
the INCOME model?

K. Regression 11 presents estimates of the Weibull INCOME model that includes the control
function. Regression 12 is similar to 11, but regression 12 includes normal heterogeneity in the
model in the form of what appears to be a random effect – a random constant. But, this is not a panel
data model look closely at the results and note that the ‘panel’ has one period. The implied two
equation model underlying 12 is

MARRIEDi* = z + ui, MARRIEDi = 1[MARRIEDi* > 0], ui ~ N[0,1].

INCOMEi* ~ Weibull(i,P) where i = exp(xi + i)

where (i,ui) have a bivariate normal distribution with means (0,0), standard deviations (,1) and
correlation . The endogeneity issue turns on . The coefficient on GENRES in the model in
regression 12 will approximate . So, based on the estimated model, marry for money
(endogenous,  not equal to zero) or marry for love (exogenous,  equal to zero)?

L. In this model, the argument in parts J and K about MARRIED could also be made about health
satisfaction, HSAT. But, HSAT is an ordered outcome, coded 0,1,2 (bad, middling, good) in our
data. How would you proceed to deal with endogneity of HSAT in this model?
1. Weibull, MLE
-----------------------------------------------------------------------------
Weibull (Loglinear) Regression Model
Dependent variable INCOME
Log likelihood function 12133.14495
Restricted log likelihood 1195.24508 (Log likelihood when  = 0)
Chi squared [ 7](P= .000) 21875.79975
Significance level .00000
McFadden Pseudo R-squared -9.1511775
Estimation based on N = 27326, K = 8
Inf.Cr.AIC = -24250.3 AIC/N = -.887
--------+--------------------------------------------------------------------
| Standard Prob. 95% Confidence
INCOME| Coefficient Error z |z|>Z* Interval
--------+--------------------------------------------------------------------
|Parameters in conditional mean function
Constant| 1.67075*** .01433 116.62 .0000 1.64267 1.69883
AGE| .00086*** .00022 3.91 .0001 .00043 .00130
EDUC| -.05084*** .00073 -69.23 .0000 -.05228 -.04940
HSAT| -.01233*** .00077 -15.96 .0000 -.01385 -.01082
MARRIED| -.16990*** .00371 -45.79 .0000 -.17717 -.16262
FEMALE| -.02041*** .00334 -6.11 .0000 -.02696 -.01386
HHKIDS| .06403*** .00375 17.07 .0000 .05668 .07139
|Scale parameter for Weibull model
P_scale| 2.13722*** .00495 431.40 .0000 2.12751 2.14693
--------+--------------------------------------------------------------------
***, **, * ==> Significance at 1%, 5%, 10% level.
-----------------------------------------------------------------------------
2. Exponential, MLE
-----------------------------------------------------------------------------
Exponential (Loglinear) Regression Model
Dependent variable INCOME
Log likelihood function 1558.04494
Restricted log likelihood 1195.24508
Chi squared [ 5](P= .000) 725.59973
Significance level .00000
McFadden Pseudo R-squared -.3035360
Estimation based on N = 27326, K = 6
Inf.Cr.AIC = -3104.1 AIC/N = -.114
--------+--------------------------------------------------------------------
| Standard Prob. 95% Confidence
INCOME| Coefficient Error z |z|>Z* Interval
--------+--------------------------------------------------------------------
|Parameters in conditional mean function
Constant| 1.85106*** .04834 38.29 .0000 1.75632 1.94580
AGE| .00158** .00064 2.48 .0133 .00033 .00283
EDUC| -.05438*** .00268 -20.27 .0000 -.05963 -.04912
HSAT| -.01101*** .00275 -4.00 .0001 -.01641 -.00561
MARRIED| -.26249*** .01568 -16.75 .0000 -.29322 -.23177
HHKIDS| .06619*** .01399 4.73 .0000 .03877 .09360
--------+--------------------------------------------------------------------

3. Constrained Weibull, MLE

-----------------------------------------------------------------------------
Weibull (Loglinear) Regression Model
Dependent variable INCOME
LM Stat. at start values 21526.22099
LM statistic kept as scalar LMSTAT
Log likelihood function 1558.04494
Restricted log likelihood 1195.24508
Chi squared [ 6](P= .000) 725.59973
Significance level .00000
McFadden Pseudo R-squared -.3035360
Estimation based on N = 27326, K = 7
Inf.Cr.AIC = -3102.1 AIC/N = -.114
--------+--------------------------------------------------------------------
| Standard Prob. 95% Confidence
INCOME| Coefficient Error z |z|>Z* Interval
--------+--------------------------------------------------------------------
|Parameters in conditional mean function
Constant| 1.85106*** .09976 18.56 .0000 1.65553 2.04658
AGE| .00158 .00130 1.22 .2233 -.00096 .00412
EDUC| -.05438*** .00574 -9.47 .0000 -.06563 -.04312
HSAT| -.01101** .00556 -1.98 .0477 -.02190 -.00011
MARRIED| -.26249*** .02881 -9.11 .0000 -.31896 -.20603
HHKIDS| .06619** .02827 2.34 .0192 .01078 .12160
|Scale parameter for Weibull model
P_scale| 1.0*** .00672 148.81 .0000 .98683 .10132D+01
--------+--------------------------------------------------------------------
***, **, * ==> Significance at 1%, 5%, 10% level.
Model was estimated on May 04, 2014 at 10:57:45 AM
-----------------------------------------------------------------------------
4. Nonlinear Least Squares, y on exp(-b’x)
-----------------------------------------------------------------------------
Nonlinear least squares regression ............
LHS=INCOME Mean = .35214
Standard deviation = .17686
Fit R-squared = .11070
Adjusted R-squared = .11073
Model test F[ 5, 27320] (prob) = 680.2(.0000)
--------+--------------------------------------------------------------------
| Standard Prob. 95% Confidence
UserFunc| Coefficient Error z |z|>Z* Interval
--------+--------------------------------------------------------------------
B_ONE| 1.92270*** .02202 87.33 .0000 1.87955 1.96585
B_AGE| -.00022 .00030 -.75 .4535 -.00081 .00036
B_EDUC| -.05378*** .00103 -52.09 .0000 -.05580 -.05175
B_HSAT| -.01072*** .00132 -8.12 .0000 -.01330 -.00813
B_MARR| -.25986*** .00821 -31.66 .0000 -.27594 -.24377
B_KIDS| .05581*** .00664 8.40 .0000 .04279 .06883
--------+--------------------------------------------------------------------

5. Linear Least Squares, -y on b’x

-----------------------------------------------------------------------------
Ordinary least squares regression ............
LHS=MINCOME Mean = -.35214
Standard deviation = .17686
---------- No. of observations = 27326 DegFreedom Mean square
Regression Sum of Squares = 93.8115 5 18.76231
Residual Sum of Squares = 760.870 27320 .02785
Total Sum of Squares = 854.682 27325 .03128
---------- Standard error of e = .16688 Root MSE .16687
Fit R-squared = .10976 R-bar squared .10960
Model test F[ 5, 27320] = 673.68410 Prob F > F* .00000
--------+--------------------------------------------------------------------
| Standard Prob. 95% Confidence
MINCOME| Coefficient Error z |z|>Z* Interval
--------+--------------------------------------------------------------------
Constant| -.03873*** .00815 -4.75 .0000 -.05470 -.02275
AGE| .00012 .00010 1.14 .2535 -.00008 .00032
EDUC| -.02088*** .00044 -47.07 .0000 -.02175 -.02001
HSAT| -.00366*** .00046 -8.03 .0000 -.00455 -.00277
MARRIED| -.08630*** .00260 -33.21 .0000 -.09140 -.08121
HHKIDS| .02024*** .00238 8.51 .0000 .01558 .02489
--------+--------------------------------------------------------------------

6. Linear Least Squares, -logy on b’x

-----------------------------------------------------------------------------
Ordinary least squares regression ............
LHS=MLINCOME Mean = 1.15744
Standard deviation = .49145
---------- No. of observations = 27326 DegFreedom Mean square
Regression Sum of Squares = 968.991 5 193.79827
Residual Sum of Squares = 5630.67 27320 .20610
Total Sum of Squares = 6599.66 27325 .24152
---------- Standard error of e = .45398 Root MSE .45393
Fit R-squared = .14682 R-bar squared .14667
Model test F[ 5, 27320] = 940.30851 Prob F > F* .00000
--------+--------------------------------------------------------------------
| Standard Prob. 95% Confidence
MLINCOME| Coefficient Error z |z|>Z* Interval
--------+--------------------------------------------------------------------
Constant| 2.03085*** .02217 91.60 .0000 1.98740 2.07430
AGE| .00190*** .00028 6.73 .0000 .00135 .00246
EDUC| -.05651*** .00121 -46.82 .0000 -.05887 -.05414
HSAT| -.01175*** .00124 -9.47 .0000 -.01418 -.00932
MARRIED| -.34733*** .00707 -49.14 .0000 -.36118 -.33348
HHKIDS| .06628*** .00647 10.25 .0000 .05361 .07896
--------+--------------------------------------------------------------------

7. Fixed Effects Weibull, MLE

-----------------------------------------------------------------------------
FIXED EFFECTS Weibul Model
Dependent variable INCOME
Log likelihood function 34910.40335
Estimation based on N = 27326, K =7299
Inf.Cr.AIC = -55222.8 AIC/N = -2.021
Unbalanced panel has 7293 individuals
Skipped 0 groups with inestimable ai
Weibull loglinear regression model
--------+--------------------------------------------------------------------
| Standard Prob. 95% Confidence
INCOME| Coefficient Error z |z|>Z* Interval
--------+--------------------------------------------------------------------
|Index function for probability
AGE| -.04322*** .00055 -78.85 .0000 -.04429 -.04214
EDUC| -.07959*** .00616 -12.91 .0000 -.09167 -.06750
HSAT| -.00339*** .00088 -3.85 .0001 -.00511 -.00166
MARRIED| -.18215*** .00836 -21.80 .0000 -.19853 -.16578
HHKIDS| .07732*** .00550 14.06 .0000 .06654 .08810
|Scale parameter for Weibull distribution
P_scale| 5.77115*** .02935 196.61 .0000 5.71362 5.82868
--------+--------------------------------------------------------------------
***, **, * ==> Significance at 1%, 5%, 10% level.
Model was estimated on May 04, 2014 at 10:25:21 AM
-----------------------------------------------------------------------------

8. Random Effects Weibull, Maximum Simulated

Likelihood
-----------------------------------------------------------------------------
Random Coefficients WeiblReg Model
Dependent variable INCOME
Log likelihood function 19489.51857
Restricted log likelihood 1558.04494
Chi squared [ 1](P= .000) 35862.94726
Significance level .00000
McFadden Pseudo R-squared -11.5089579
Estimation based on N = 27326, K = 8
Inf.Cr.AIC = -38963.0 AIC/N = -1.426
Unbalanced panel has 7293 individuals
Simulation based on 100 Halton draws
Weibull loglinear regression model
--------+--------------------------------------------------------------------
| Standard Prob. 95% Confidence
INCOME| Coefficient Error z |z|>Z* Interval
--------+--------------------------------------------------------------------
|Nonrandom parameters
AGE| -.01369*** .00015 -91.51 .0000 -.01398 -.01339
EDUC| -.06413*** .00057 -111.59 .0000 -.06525 -.06300
HSAT| -.00478*** .00059 -8.05 .0000 -.00594 -.00361
MARRIED| -.19181*** .00307 -62.53 .0000 -.19782 -.18580
HHKIDS| .08751*** .00271 32.26 .0000 .08219 .09282
|Means for random parameters
Constant| 2.43436*** .01030 236.40 .0000 2.41418 2.45455
|Scale parameters for dists. of random parameters
Constant| .50166*** .00138 364.84 .0000 .49896 .50435
|Scale parameter for Weibull distribution
P_scale| 4.15999*** .01130 368.03 .0000 4.13783 4.18214
--------+--------------------------------------------------------------------
***, **, * ==> Significance at 1%, 5%, 10% level.
Model was estimated on May 04, 2014 at 10:28:37 AM
-----------------------------------------------------------------------------
9. Random Effects Weibull, Maximum Simulated
Likelihood with Group Means
-----------------------------------------------------------------------------
Random Coefficients WeiblReg Model
Dependent variable INCOME
Log likelihood function 21443.98658
Restricted log likelihood 1735.72267
Chi squared [ 1](P= .000) 39416.52782
Significance level .00000
McFadden Pseudo R-squared -11.3545005
Estimation based on N = 27326, K = 13
Inf.Cr.AIC = -42862.0 AIC/N = -1.569
Unbalanced panel has 7293 individuals
Simulation based on 100 Halton draws
Weibull loglinear regression model
--------+--------------------------------------------------------------------
| Standard Prob. 95% Confidence
INCOME| Coefficient Error z |z|>Z* Interval
--------+--------------------------------------------------------------------
|Nonrandom parameters
AGE| -.04140*** .00041 -100.59 .0000 -.04221 -.04060
EDUC| -.07436*** .00382 -19.47 .0000 -.08185 -.06687
HSAT| -.00775*** .00076 -10.19 .0000 -.00924 -.00626
MARRIED| -.20973*** .00473 -44.37 .0000 -.21899 -.20046
HHKIDS| .08242*** .00454 18.17 .0000 .07353 .09132
gmnAGE| .04656*** .00043 107.94 .0000 .04571 .04741
gmnEDUC| .03803*** .00384 9.90 .0000 .03050 .04556
gmnHSAT| -.01136*** .00083 -13.69 .0000 -.01299 -.00974
gmnMARRI| .01497** .00587 2.55 .0107 .00347 .02646
gmnHHKID| -.01393** .00603 -2.31 .0210 -.02575 -.00210
|Means for random parameters
Constant| 1.41275*** .00930 151.96 .0000 1.39453 1.43097
|Scale parameters for dists. of random parameters
Constant| .46708*** .00118 395.64 .0000 .46477 .46940
|Scale parameter for Weibull distribution
P_scale| 4.38591*** .01205 363.95 .0000 4.36229 4.40953
--------+--------------------------------------------------------------------
10. Probit Model for Marital Status, MLE
-----------------------------------------------------------------------------
Binomial Probit Model
Dependent variable MARRIED
--------+--------------------------------------------------------------------
| Standard Prob. 95% Confidence
MARRIED| Coefficient Error z |z|>Z* Interval
--------+--------------------------------------------------------------------
|Index function for probability
Constant| .20370*** .05761 3.54 .0004 .09079 .31662
AGE| .02234*** .00076 29.28 .0000 .02084 .02383
EDUC| -.03308*** .00367 -9.02 .0000 -.04027 -.02589
FEMALE| -.12946*** .01727 -7.50 .0000 -.16330 -.09562
WHITEC| -.03858** .01861 -2.07 .0382 -.07506 -.00210
--------+--------------------------------------------------------------------

11. Weibull with Control Function, MLE

-----------------------------------------------------------------------------
Weibull (Loglinear) Regression Model
Dependent variable INCOME
Log likelihood function 12160.11190
Restricted log likelihood 1195.24508
Chi squared [ 7](P= .000) 21929.73366
Significance level .00000
--------+--------------------------------------------------------------------
| Standard Prob. 95% Confidence
INCOME| Coefficient Error z |z|>Z* Interval
--------+--------------------------------------------------------------------
|Parameters in conditional mean function
Constant| 1.19668*** .03721 32.16 .0000 1.12374 1.26962
AGE| -.00528*** .00051 -10.39 .0000 -.00627 -.00428
EDUC| -.04215*** .00107 -39.51 .0000 -.04425 -.04006
HSAT| -.01251*** .00077 -16.34 .0000 -.01401 -.01101
MARRIED| .67313*** .06091 11.05 .0000 .55375 .79251
HHKIDS| .05201*** .00371 14.03 .0000 .04475 .05927
GENRES| -.49197*** .03508 -14.02 .0000 -.56073 -.42321
|Scale parameter for Weibull model
P_scale| 2.13826*** .00492 434.41 .0000 2.12862 2.14791
--------+--------------------------------------------------------------------

12. Weibull with Normal Heterogeneity and Control Function,

Maximum Simulated Likelihood
-----------------------------------------------------------------------------
Random Coefficients WeiblReg Model
Dependent variable INCOME
Log likelihood function 13158.01422
Restricted log likelihood 1563.62291
Sample is 1 pds and 27326 individuals
Simulation based on 10 Halton draws
--------+--------------------------------------------------------------------
| Standard Prob. 95% Confidence
INCOME| Coefficient Error z |z|>Z* Interval
--------+--------------------------------------------------------------------
|Nonrandom parameters
AGE| -.00331*** .00046 -7.24 .0000 -.00421 -.00241
EDUC| -.04512*** .00097 -46.49 .0000 -.04702 -.04321
HSAT| -.01211*** .00072 -16.84 .0000 -.01352 -.01070
MARRIED| .45855*** .05526 8.30 .0000 .35024 .56686
HHKIDS| .06500*** .00369 17.60 .0000 .05776 .07224
GENRES| -.39467*** .03169 -12.45 .0000 -.45678 -.33256
|Means for random parameters
Constant| 1.31805*** .03473 37.95 .0000 1.24999 1.38612
|Scale parameters for dists. of random parameters
Constant| .25368*** .00194 130.76 .0000 .24988 .25748
|Scale parameter for Weibull distribution
P_scale| 2.64003*** .00733 359.95 .0000 2.62566 2.65441
--------+--------------------------------------------------------------------