LIMDEP
LIMDEP
Version 7.0
User's Manual
Revised Edition
by
William H. Greene
Econometric Software, Inc.
© 1998 Econometric Software, Inc. All rights reserved.
This software product, including both the program code and the accompanying documentation, is
copyrighted by, and all rights are reserved by Econometric Software, Inc. No part of this product, either the
software or the manual, may be reproduced, stored in a retrieval system, or transmitted in any form or by any
means without prior written permission of Econometric Software, Inc.
LIMDEPTM is a trademark of Econometric Software, Inc. All other brand and product names are trademarks
or registered trademarks of their respective companies.
Econometric Software, Inc.
15 Gloria Place
Plainview, NY 11803
Voice 516-938-5254
Fax
516-938-2441
E-mail [email protected]
Econometric Software, Australia
41B Excelsior Avenue
Castle Hill NSW 2154
Australia
Fax
61-2-899-6674
i
LIMDEP

LICENSE AGREEMENT
You have only the non-exclusive right to use this computer program. You may use the program on
a single computer at any one time. If necessary, you may make one copy of this program for backup in
support of your own use on the single machine. This program and manual may not be distributed to any
other party. This program may not be electronically transferred from one computer to another or over a
network without specific authorization by Econometric Software, Inc. You may not modify, adapt, translate,
or change the program or the manual without prior written permission of Econometric Software, Inc.
DISCLAIMER OF WARRANTY
This program is provided ‘as is’ without warranty of any kind. You are responsible for the choice of
this program and for all results obtained with it. Econometric Software shall have no liability or
responsibility to the purchaser or any other entity for loss or damage alleged to be caused by the use of this
program or its documentation. This disclaimer of any liability shall include, but is not limited to, any loss of
business, savings, or profits, or any other incidental or consequential loss or physical damage resulting from
use of this program or its documentation.
DEFECTIVE MEDIA REPLACEMENT
The media on which LIMDEP is provided are warranted to be free of physical defects in material or
workmanship in normal usage and will be replaced without charge if such physical defects are reported to
Econometric Software, Inc. within thirty (30) days of the date of purchase.
ii
iii
Preface

LIMDEP is a flexible computer program for estimating the sorts of models most frequently analyzed
with cross section data. Its range of capabilities include basic linear regression and descriptive statistics, the
full set of techniques normally taught in the first year of a graduate econometrics sequence, and many
advanced techniques such as nested logit models, parametric duration models, Poisson regressions with right
censoring and nonlinear regressions estimated by instrumental variables and the GMM. LIMDEP's menu of
options is as wide as that of any other general purpose program available, though, as might be expected,
longer in some dimensions and shorter in others (notably time series analysis). This is not an all purpose
program which will solve every imaginable problem. In spite of claims to the contrary, that program does
not exist. If it did, it would be impossibly cumbersome. Rather, our guiding principle in the construction of
LIMDEP is a desire to make as many of the widely used techniques and nonlinear models as simple to apply
as possible. LIMDEP is best suited to the analysis of cross sections and relatively standard problems of time
series analysis, such as ARMAX models, distributed lags, and low order autocorrelation.
This program has developed over many years (since 1980), initially to provide an easy to use tobit
estimator - hence the name, ‘LIMited DEPendent variable models.’ The accumulated suggestions of many
colleagues, students, and users too numerous to thank individually have led to the current package. In
particular, the help of Aline Quester, Nathaniel Beck, Andrea Long, David James, Bill Spitz, Charles
Hallahan, Terry Seaks, Rich Goldstein and Steffen Kuehnel has been considerable. Version 7.0 continues
our periodic cycle of collecting, then incorporating the many suggestions we receive from our users. We
also update LIMDEP every few years to incorporate new developments in econometrics and to meet the
changing demands of our users.
Version 7.0 also reflects the major participation of David Hensher and Michael Lowe at
Econometric Software, Inc., Australia. The Windows 95/NT version of LIMDEP was developed in
collaboration with Michael Lowe.
To the best of our knowledge, the code of this program is correct as described. However, no
warranty is expressed or implied. Users assume responsibility for the selection of this program to achieve
their desired results and for the results obtained.
William H. Greene
Econometric Software, Inc.
15 Gloria Place
Plainview, New York 11803
January, 1998
iv
v
Preface to Version 7.0

Version 7.0 of LIMDEP continues our efforts to produce major upgrades to the program while
maintaining compatibility with earlier versions. Version 7.0 features a new set of estimation programs
including NLOGIT, an optional FIML estimator for nested logit models, enhanced capabilities, a
redesigned interface and, with this revision, full conversion to the Windows 95/NT operating system.
Revised Interface
Current LIMDEP users will see vast improvements in LIMDEP’s interface, such as:
•
•
•
•
•
•
•
Conversion of the user interface to the Windows 95/NT operating systems
Redesigned screens for input and output
Greater use of menus and function keys
Expanded HELP and a slightly abbreviated manual which is now in the help file.
Greater use of algebraic syntax in mathematical commands (e.g., b'x instead of Dot(X,b).)
Full use of algebra in all logical expressions
More informative diagnostics and a complete trace of execution, with diagnostics and
indexation of commands to the output file
Improved
formats in all program output
•
A
program
for constructing output tables of several sets of estimates
•
New Models
Version 7.0 features the following new estimation programs:
•
•
•
•
•
•
•
•
•
•
•
•
•
•
Nested tobit models
Grouped data models with sample selection
Count data models with altered probability of the zero outcome
Count data models with sample selection
Bivariate probit models with heteroscedasticity
Multivariate and multinomial probit models
Random coefficients models
Partial observability models
Nonlinear multiple equations models and GMM estimators
Bivariate tobit model
Random effects tobit, probit, and logit models
Fixed and random effects models for count data
Random parameters logit model
Models of heterogeneity and underreporting for count data
In addition to the new estimators, many new features have been added for manipulating panel data. With
Version 7.0, all data panel models may be based on unbalanced data sets.
vi
Nested Logit Models
NLOGIT is a full information maximum likelihood estimator for two to four level nested logit
models. This full-featured, extremely flexible estimator is, to our knowledge, the only FIML estimator of its
kind available as part of an integrated econometrics package. Users of NLOGIT will have all of LIMDEP’s
functions available to them. NLOGIT is an optional feature of LIMDEP 7.0 and is sold separately.
New Programming Features and Tools
Version 7.0 offers added power and enhanced capabilities to existing commands as well as new
programming features such as:
•
•
•
•
•
•
•
•
Completely revised and simplified matrix algebra program
Subroutines with parameter lists
Loops using DO FOR, DO WHILE, and DO UNTIL
Indexation of variables in lists for looping over a set of variables
Revisions of the editor
Major enhancements to MINIMIZE
Bootstrapping and subsampling
New commands for analyzing model results such as restrictions and hypothesis tests.
Marginal Effects
The computation of marginal effects has been automated for all of the limited and qualitative
dependent variable models. This will eliminate the large amount of effort that was needed to obtain partial
effects in models such as the multinomial logit, sample selection, and various probit models. With the
addition of ;Marginal Effects to the model command, you will now obtain a full set of statistical output
(identifier, estimate, P-value, and t-ratio) for all estimated marginal effects computed at the overall sample
means and at the means of additional strata that you may specify.
Constraints and Tests
All models estimated with LIMDEP may now be specified with equality and fixed value restrictions.
The ;Rst=specification feature that was introduced in Version 6.0 to allow cross equation equality
restrictions in the multinomial logit model has now been extended to all of the other nonlinear models in
LIMDEP’s menu. This option allows you to force coefficients in a model to equal other coefficients or to
take fixed values rather than be estimated.
LIMDEP contains many tools for carrying out hypothesis tests, including Wald, LM, and LR tests.
The programming features now make it easier to carry out tests of nonnested hypotheses. Version 7.0
provides a new feature that will allow Wald tests of any number of sets of linear or nonlinear restrictions of
the parameters of any model. Every model command can be followed immediately by as many WALD;...
restrictions$ commands as desired. A standard set of labels is provided which makes them simple to use
and invariant to the model specification. (Version 6.0’s ;Test option requires that parameters be identified as
B(j) where j is the position of the parameter in the coefficient vector. Thus, if the specification changes, so
must the restriction. In Version 7.0, coefficients are denoted B_name, where ‘name’ is the name of the
variable multiplied by the coefficient. This does not change from one model to another.)
vii
Incompatibilities Between the Windows and DOS Versions

There are a number of major differences between the Windows and DOS versions of LIMDEP 7.0,
including, of course, the obvious change to a screen, menu oriented interface. But, we emphasize that the
differences that do exist concern only how instructions are given to the program. There are no differences in
the statistical capabilities of the two programs or in their other data manipulation routines. In general, many
options exist in the Windows program that are not available in the DOS program. In addition, some of the
commands in the DOS program are absent from the Windows program, or are handled in a different fashion.
The following lists the major differences in the basic operation of LIMDEP:
•
•
•
•
•
•
•
•
•
•
•
•
•
DEMO: The Windows program does not contain the demonstration.
DISK: Features of the file system are handled using Windows’ standard features
EDIT: The screen text editor in the Windows program replaces the DOS command prompt.
EXTRACT: Large data sets are no longer handled with temp files. (See Chapter 5.)
LOAD/SAVE: These work as before, but operate more easily in the File menu.
PAUSE: Output is accumulated in the output window, and can be reviewed as desired.
PRINT: Data listings should be sent to a file, then printed later if necessary.
RESET: This operation is handled with the Project menu.
REVIEW: This command is now available in the Tools menu.
SHOW: You can bring a file into the editor, but better to switch to another window.
To SHOW a file, just use File:Open in the main screen menu.
STATUS: Operation of the project window replaces this function. (See Chapter 3.)
STOP/EXIT/QUIT: Select File:Exit to leave LIMDEP.
SYSTEM: You can just open a DOS window for this function.
This manual contains separate instructions on operation of the two programs. Other differences between the
Windows and DOS programs are described in Chapters 2 and 3. A full summary of the Windows menus is
given in Section 3.10.
viii
ix
Incompatibilities Between Versions 6.0 and 7.0

Version 7.0 is nearly fully backward compatible with Version 6.0. The only substantive difference
is in the matrix commands listed below, though users of the bivariate probit model should make note of the
change in the way that fitted values are computed for this model. Users of Version 6.0 may note that SAVE
files written by Version 7.0 are far smaller than their Version 6.0 counterparts. This is due to a much more
efficient way of storing the information. Nonetheless, SAVE files written with Version 6.0 can be read by
Version 7.0. We do suggest that Version 6.0 SAVE files be recreated with 7.0, as some additional
information is being placed in the newer files. (Of course, SAVE files written by Version 7.0 cannot be read
by Version 6.0.)
Matrices
In matrix commands and programs, a matrix that is specified by enumerating its elements must now
be enclosed in square brackets, i.e., ‘[...].’. This applies to both sets of numbers which comprise matrices
and to sets of matrices which comprise partitioned matrices. For examples,
and
MATRIX ; A = 1,2 / 3,4 $ must now be MATRIX ; A = [1,2 / 3,4 ] $
MATRIX ; A = M11/M21,M22 $ must now be MATRIX ; A = [M11/M21,M22] $.
Because of the variety of different arrangements that may appear, the effect of omitting the brackets is
unpredictable. In most cases, a diagnostic will appear, but in others, a matrix different from the intended one
will be created without apparent error. For instance, the first example above creates a 1H1 matrix containing
1.0000.
Panel Data Models
Frontier models with panel data previously used
; Pds = T for fixed number of periods,
; Lhs = Y,Ti for variable number of periods.
They now use ; Pds = T where T is the fixed number of periods,
or
; Pds = Ti where Ti is the count variable for variable group sizes.
This same syntax is also used for LOGIT, TOBIT, PROBIT, and POISSON. Since none of these four
commands with variable number of periods (or fixed for TOBIT and POISSON) were available in Version
6.0, FRONTIER is the only command actually changed by this update.
Loglinear survival models (SURV ; Model = ...) with time varying covariates that were set up like
panels were previously specified with
SURV ; Lhs = Time, Censor, Ti ; ...
where Ti gives the number of records, which could vary by observation, or could be fixed. This syntax is
still supported, but users may use, instead,
SURV ; Lhs = Time, Censor ; Pds = T or Ti,
where Ti may be a number if every observation has the same number of periods, or it may be a variable,
used exactly the same as before.
x
Bivariate Probit Predictions
Predicted values for the bivariate probit model are computed as follows:
1. Compute the four cells for the bivariate distribution based on b1′x1, b2′x2,D. We then compute
P[y1=1,y2=1] = BVN(b1′x1,b2′x2,D)
P[y1=1,y2=0] = M(b1′x1) - P[y1=1,y2=1] = P[y1=1] - P[y1=1,y2=1]
P[y1=0,y2=1] = P[y2=1] - P[y1=1,y2=1]
P[y1=0,y2=0] = 1 - P[y1=1,y2=0] - P[y1=0,y2=1] - P[y1=1,y2=1]
2. The prediction is then the (y1,y2) associated with the cell with the largest probability.
In Version 6.0, the bivariate frequency distribution listed after the coefficient output was computed using the
univariate probabilities. I.e., If b1′x1 > 0, predict y1=1, and likewise for y2. In Version 7.0, the bivariate
distribution is used instead.
SAVE Files
Version 6.0 provided a method of using a ‘SAVE’ file to store some intermediate results. (See the
Version 6.0 manual, page 60.) In particular, this file allowed you to put in a file
1. The listing of variables produced with the ;List option on model commands,
2. The parameters and covariance matrices after estimation, and
3. The raw second moments, means, standard deviations, covariance matrix, and
correlation matrix from DSTAT.
This has been discontinued. All of the results above can be placed directly into the output file. For 1, it is
automatic. For 2 and 3, simple MATRIX commands, with ;List will produce the same results. The use of
this type of file has been discontinued. Note this not the file associated with the SAVE command discussed
above.
OR in Logical Expressions
In Version 6.0, the ‘+’ character was used to indicate ‘OR’ in logical expressions, such as
CREATE;If(AGE=0 + INCOME=0)BADDATA=1$ in which BADDATA equals 1 if AGE equals 0 or if
INCOME equals 0. In Version 7.0, the operator ‘|’ is used instead. Thus, the counterpart to the command
above is CREATE;If(AGE=0 | INCOME=0)BADDATA=1$
BEEP and NOBEEP
The BEEP and NOBEEP commands have been dropped. In addition, all beeping at the end of file
entry, procedure entry, etc. has been discontinued. (LIMDEP is quieter.)
xi
Some Common Questions

The following are some of the questions most commonly asked about using LIMDEP:
In reading a data file, LIMDEP claims to have reached the end of the file after reading exactly one half
of the observations. Why?
1. Your format statement is inadequate. For example:
READ ; File = ... ; Nvar = 3 ; Format = (2F5.3) ...
This command reads three variables but gives only two format codes. This requires
two lines of data per observation. The second line is used to give the third
variable, and one number is lost.
2. (Very unlikely) You do not have a format statement. But, the data file that you are
using has lines that are over 300 characters wide. LIMDEP reads 300 characters.
If it does not find all of the values it expects, it goes to the next line.
The iterations failed to converge. A diagnostic stated that a minimum of the function could not be
found.
This is common with models with correlation coefficients, such as the selection models and the
bivariate probit models. It should not happen with univariate probit or any kind of logit models. It should be
quite rare with LDVs such as the tobit model. Check:
1. Scaling the data. Are the variables of very different magnitudes? Try to avoid this. This is
especially problematic in routines that involve quadrature (random effects probit/ordered probit)
and in routines that fit correlation coefficients (bivariate probit).
2. Collinearity is also a problem in nonlinear models.
A strategy: When you are having trouble estimating a model, try estimating it with a very small
subset of the variables. Choose one or two independent variables (for each equation if necessary) and
estimate with just them (even if you do not believe the specification). Build up the model from a small base.
When problems emerge, you will know where to look in the data.
Why isn’t LIMDEP able to read my Excel spreadsheet file?
Current Windows versions of these programs write an enormous amount of superfluous (for our
purposes) information into these files, and LIMDEP gets confused by it. But, if you use either of these
programs to create your data, at the time you wish to write the spreadsheet file, go to the menu of file types
and choose the .WK1 format. LIMDEP will be able to read the .WK1 file.
The log-likelihood function for my model is positive. Is this possible?
Log-likelihood functions are only required to be negative for models with discrete dependent
variables. Tobit models, regression models, duration models, etc. are based on continuous distributions.
Thus, the log-likelihoods for these models can be positive.
xii
During the iterations, I get a diagnostic that a variance (or standard deviation) is not positive or that a
correlation coefficient is outside the admissable range (-1,+1). Does this mean my estimates are not
useable?
No. LIMDEP searches for the maximizer of a log-likelihood by taking a step from a valid parameter
estimate toward a new estimate. If the new estimate is invalid, as in the query, LIMDEP issues a warning,
then tries a shorter step. If the iterations ultimately did converge, then the warnings can be ignored. If
convergence is not achieved, the warnings might be helpful in finding out why.
The Cox proportional hazards model seems to take a very long time to iterate.
It does. This model is extremely computation intensive. Every iteration requires essentially N2
passes through the data, not just one.
Can I expand the data array?
For the DOS program, use a temp file. You can in the mainframe version. See the Appendix. In
the Windows program, you need only adjust the workspace to whatever amount of space you need. The
program handles everything else. See Section 3.10.2.
My computations require a matrix larger than 10000 cells.
Unfortunately, this limit is hard coded into LIMDEP. On the other hand, you should never have to
create a matrix whose length is the number of observations in the sample. Chapter 10 contains extensive
details on how to handle data matrices. Chances are there is a way to do your computations without having
to create a matrix this large.
If you believe you have found a bug in LIMDEP, please do the following:
If it is something obvious, just write or call and let us know. We’ll fix the program and send a
replacement with alacrity. But, it is much more likely that it will be something subtle. For example, suppose
LIMDEP crashed in the middle of an iteration because it tried to take the square root of a negative number.
This sort of thing usually requires us to execute the program with your data and your commands. Try to
produce the error with as small a data set and command set as possible. Send us enough material, including
your trace file and the data, so that we can reproduce the problem. When we find the problem, we'll fix it
and replace your program. Please try to avoid sending huge data sets. Also, in almost all cases, it is
extremely helpful if you can send us the TRACE.LIM file that is associated with your problem run.
xiii
Abbreviated Table of Contents

Part I Setting Up and Getting Started 1

Chapter 1 Introduction to LIMDEP 3
1.1. The LIMDEP Program 3
1.2. Econometric Techniques 4
1.3. Getting Started 5
Chapter 2 Installing and Executing LIMDEP 7
2.1.
2.2.
2.3.
2.4.
2.5.
2.6.
2.7.
2.8.
Introduction 7
Equipment 7
Installation 7
Execution - Starting LIMDEP 10
Retaining Results Between Sessions - SAVE and LOAD 14
Exiting LIMDEP 17
Restarting During a Session - The RESET Command 18
The Trace File 19
Chapter 3 Entering Commands - Basics of Operation 21
3.1. Introduction 21
3.2. Basic Command Entry 21
3.3. Commands 30
3.4. Input Files - Entering Commands from a File 32
3.5. Work Areas, Projects, and the STATUS Command 36
3.6. Program Output and the Output Window 43
3.7. HELP 44
3.8. Features of the DOS Program 45
3.9. A Summary of Commands 50
3.10. Summary of the Windows Desktop 54
Part II Data Input, Export, and Transformation 63

Chapter 4 Entering Data and Creating Data Files 65
4.1.
4.2.
4.3.
4.4.
4.5.
4.6.
4.7.
Introduction 65
The Data Area and the ROWS Command 65
Using the Data Editor 66
Reading Data Files 70
Formatted Data 73
Large Data Sets 77
Spreadsheet Files and Binary Files 77
xiv
Abbreviated Table of Contents

4.8. Adding Observations - The APPEND Command 79
4.9. Writing Data Files - the WRITE Command 80
4.10. Listing Data 80
Chapter 5 Data Transformations 81
5.1.
5.2.
5.3.
5.4.
5.5.
Introduction 81
The CREATE Command 81
Recoding Variables - The RECODE Command 94
Sorting Variables - The SORT Command 95
The DELETE and RENAME Commands 96
Chapter 6 Panel Data and the Discrete Choice Model 97
6.1.
6.2.
6.3.
6.4.
6.5.
6.6.
6.7.
6.8.
Introduction 97
Programs Which Use Panel Data 97
Conventions 97
Stratification Indicators 98
Indicators and Group Size Variables for Panel Data 99
Matrix Commands for Panel Data 101
Converting Data for the Discrete Choice Models 101
Merging Invariant Variables into a Panel 102
Chapter 7 Variable Lists, the Current Sample, and Missing Data 105
7.1.
7.2.
7.3.
7.4.
7.5.
Introduction 105
Lists of Variables 105
The NAMELIST Command 106
The Current Sample 108
Missing Data 113
Part III Model Estimation 117

Chapter 8 Commands for Estimating Models 119
8.1.
8.2.
8.3.
8.4.
8.5.
8.6.
8.7.
Introduction 119
Model Commands 119
Output from Estimation Programs 120
Output Files 123
The Review Window - Tables of Model Results 124
Numerical Optimization 125
Multivariate Normal Probabilities 134
Chapter 9 Analyzing Models and Testing Hypotheses 137
9.1.
9.2.
9.3.
9.4.
9.5.
9.6.
Introduction 137
Model Components 138
Standard Output 141
Marginal Effects 144
Retrievable Results 147
The Last Model and Functions of Parameters 149
Abbreviated Table of Contents
xv

9.7. Creating and Displaying Predictions and Residuals 151
9.8. Imposing Equality and Fixed Value Restrictions 155
9.9. Linear Functions of Model Parameters 157
9.10. Nonlinear Restrictions - The WALD Command 162
9.11. Tests Based on Likelihood Functions - LR and LM Tests 165
9.12. Hausman Tests 171
9.13. Moment Based Specification Tests 172
Chapter 10 Matrix Algebra 177
10.1.
10.2.
10.3.
10.4.
10.5.
10.6.
10.7.
10.8.
10.9.
Introduction 177
Entering MATRIX Commands 180
Using MATRIX Commands with Data 185
Manipulating Matrices 188
Entering, Moving, and Rearranging Matrices 196
Matrix Functions 200
Sums of Observations 204
Examples 211
Changes from Earlier Versions of LIMDEP 217
Chapter 11 Manipulating Scalars with the CALCULATE Command 219
11.1.
11.2.
11.3.
11.4.
11.5.
11.6.
Introduction 219
Command Input in CALCULATE 220
Output from CALCULATE 222
Forms of CALCULATE Commands - Conditional Commands 224
Scalar Expressions 226
Calculator Functions 227
Part IV Data Description 235

Chapter 12 Describing Data 237
12.1.
12.2.
12.3.
12.4.
12.5.
Introduction 237
Summary Statistics 237
Histograms 242
Cross Tabulations 245
Box-Jenkins Time Series Identification 248
Chapter 13 Scatter Diagrams and Plotting 251
13.1.
13.2.
13.3.
13.4.
13.5.
13.6.
Introduction 251
Printing and Exporting Figures 251
Scatter Plots 252
Multiple Scatter Plots - The SPLOT Command 258
Plotting Matrices - The MPLOT Command 259
Plotting Functions - The FPLOT Command 261
xvi
Abbreviated Table of Contents

Part V Single and Multiple Equation Regression Models 263

Chapter 14 The Linear Regression Model 265
14.1.
14.2.
14.3.
14.4.
14.5.
Introduction 265
Least Squares Regression 265
Predictions and Residuals 268
Hypothesis Tests 273
Stepwise Regression 289
Chapter 15 Heteroscedasticity in the Linear Model 291
15.1.
15.2.
15.3.
15.4.
Introduction 291
Correcting the OLS Covariance Matrix 291
Estimating Models with Heteroscedasticity 294
Testing for Heteroscedasticity 304
Chapter 16 Autocorrelation in the Linear Model 307
16.1.
16.2.
16.3.
16.4.
16.5.
16.6.
Introduction 307
Correcting the OLS Covariance Matrix 307
Correcting for First Order Autocorrelation 308
Autocorrelation with a Lagged Dependent Variable 311
Differencing and Higher Order Autocorrelation 312
Testing for Autocorrelation 313
Chapter 17 Linear Models with Panel Data 315
17.1.
17.2.
17.3.
17.4.
17.5.
17.6.
Introduction 315
Data Arrangement and Setup 316
One Way Fixed and Random Effects Models 318
Two Factor Fixed and Random Effects Models 338
Random Coefficients Model 344
Groupwise Heteroscedasticity, Correlation, and Autocorrelation 347
Chapter 18 ARIMA, ARMAX and Distributed Lag Models 357
18.1.
18.2.
18.3.
18.4.
18.5.
Introduction 357
Box-Jenkins ARIMA and ARMAX Models 357
Polynomial Distributed Lag Models 365
The Geometric Lag Model 370
Roots of Dynamic Equations 375
Chapter 19 Nonlinear Least Squares, 2SLS, IV, and GMM Estimation 377
19.1.
19.2.
19.3.
19.4.
19.5.
19.6.
Introduction 377
The Box-Cox Regression Model 377
Nonlinear Least Squares for Nonlinear Regression 389
Two Stage Least Squares Estimation of Linear Models 401
Nonlinear Two Stage Least Squares Estimation 406
GMM Estimation of Nonlinear Models 407
Abbreviated Table of Contents
xvii

Chapter 20 Systems of Regression Equations 409
20.1.
20.2.
20.3.
20.4.
20.5.
Introduction 409
Linear SURE Models Estimated by GLS 409
Maximum Likelihood Estimation of Constrained Linear Systems 420
Instrumental Variables (3SLS) Estimation of a Set of Equations 427
Nonlinear Systems of Regression Equations 429
Part VI Qualitative, Discrete, and Count Dependent Variables 437

Chapter 21 Probit, Logit, and Other Models for Binary Choice 439
21.1.
21.2.
21.3.
21.4.
21.5.
21.6.
21.7.
21.8.
21.9.
Introduction 439
The Probit Model for Binary Choice 443
Heteroscedasticity in the Probit Model 450
Panel Data and Random Effects in the Probit Model 454
The Binomial Logit Model 459
Fixed and Random Effects in the Binomial Logit Model 464
User Defined Index Models for Binary Choice 467
Semiparametric Analysis of Binary Choice - MSCORE 469
Nonparametric Regression Based on Binary Choice - NPREG 476
Chapter 22 Bivariate and Multivariate Probit and Partial Observability Models 483
22.1.
22.2.
22.3.
22.4.
22.5.
Introduction 483
Estimating the Bivariate Probit Model 483
Bivariate Probit Models with Sample Selection 492
Bivariate Probit Models with Partial Observability 493
The Multivariate Probit Model 494
Chapter 23 Ordered Probability Models 497
23.1.
23.2.
23.3.
23.4.
23.5.
23.6.
23.7.
Introduction 497
Commands for Basic Ordered Probit and Logit Models 498
A Random Effects Model for Panel Data 505
Stratification 507
Heteroscedasticity 507
Sample Selection 509
Technical Details 510
Chapter 24 Multinomial Logit and Discrete Choice Models - LOGIT and NLOGIT 513
24.1.
24.2.
24.3.
24.4.
24.5.
24.6.
24.7.
24.8.
24.9.
Introduction 513
The Multinomial Logit Model 514
NLOGIT - Discrete Choice Models 519
Models for Discrete Choice 520
Data for the Discrete Choice Models 521
Model Commands for the Discrete Choice Model 527
Options for the Discrete Choice Model 529
Alternatives to the Multinomial Logit Model 537
Applications 547
xviii
Abbreviated Table of Contents

Chapter 25 Nested Logit Models - NLOGIT 557
25.1.
25.2.
25.3.
25.4.
25.5.
25.6.
25.7.
Introduction 557
Mathematical Specification of the Model 558
Commands for FIML Estimation 559
Sequential (Two Step) Estimation of Nested Logit Models 573
Applications 577
Combining Data Sets and Scaling in Discrete Choice Models 587
Technical Details 591
Chapter 26 Poisson and Negative Binomial Regression Models for Count Data 595
26.1.
26.2.
26.3.
26.4.
26.5.
Introduction 595
Poisson and Negative Binomial Regression Models 598
Count Data Models for Panel Data 629
Zero Inflated Poisson (ZIP) Models 640
Models for Sample Selection 649
Part VII Limited Dependent Variables and Frontier Regression Models 659

Chapter 27 Tobit, Censoring, and Truncation Models 661
27.1.
27.2.
27.3.
27.4.
27.5.
27.6.
27.7.
Censored Regression Models 661
The Tobit Model for Censored Data 662
Variants of the Tobit Model 674
Technical Notes for the Tobit Models 695
The Truncated Regression Model 700
The Grouped Data Regression Model 703
The Lognormal Regression Model 708
Chapter 28 Sample Selection and Switching Regression Models 711
28.1.
28.2.
28.3.
28.4.
28.5.
28.6.
Introduction 711
Regression Models with Sample Selection 712
Different Specifications of the Selection Equation 721
Mixtures of Discrete, Censored, and Continuous Variables 735
Technical Details 739
Switching Regression Models 741
Chapter 29 Stochastic Frontier Regression Models 753
29.1.
29.2.
29.3.
29.4.
29.5.
Introduction 753
Model Command 754
Results for the Stochastic Frontier Estimators 755
Applications 757
Log-Likelihood Functions for the Frontier Models 765
Abbreviated Table of Contents
xix

Part VIII Duration Models 767

Chapter 30 Nonparametric Analysis of Duration Data 769
30.1.
30.2.
30.3.
30.4.
30.5.
Introduction 769
Life Tables 769
Commands for Life Tables 771
Applications 772
Mathematical Details for the Homogeneity Tests 782
Chapter 31 Semiparametric and Proportional Hazard Models 783
31.1.
31.2.
31.3.
31.4.
31.5.
31.6.
Introduction 783
The Proportional Hazards Model 783
Commands for the Proportional Hazards Model 784
Output for the Proportional Hazards Model 789
The Ordered Logit Model 790
Applications 791
Chapter 32 Parametric Models of Duration 803
32.1.
32.2.
32.3.
32.4.
32.5.
32.6.
32.7.
Introduction 803
Loglinear Models for Survival Data 804
Commands for Estimating Loglinear Models 805
Output from Loglinear Model Commands 807
Extensions of the Loglinear Models 808
Applications 817
Technical Details 825
Part IX Nonlinear Optimization and LIMDEP Programming 827

Chapter 33 Nonlinear Optimization 829
33.1.
33.2.
33.3.
33.4.
33.5.
Introduction 829
The MINIMIZE/MAXIMIZE Commands 829
Optimization Problems 834
Output from MINIMIZE/MAXIMIZE 835
Applications 836
Chapter 34 Analyzing Nonlinear Functions 839
34.1.
34.2.
34.3.
34.4.
34.5.
Introduction 839
Plotting a Function 839
Variances for Nonlinear Functions 842
Differentiation 844
Integration 845
xx
Abbreviated Table of Contents

Chapter 35 Programming Tools 849
35.1.
35.2.
35.3.
35.4.
35.5.
35.6.
Introduction 849
Estimation Programs and Post-Processing 849
Numeric Computation 850
Batching Commands 851
Flow Control 853
Looping with DO Statements 855
Chapter 36 Using the Command Editor 857
36.1.
36.2.
36.3.
36.4.
36.5.
Introduction 857
The Windows Text Editor Window 857
Using the Editor in the DOS Program 858
The SILENT Command 860
Files 861
Chapter 37 Procedures and the EXECUTE Command 863
37.1.
37.2.
37.3.
37.4.
37.5.
37.6.
37.7.
Introduction 863
Storing a Procedure 863
The Procedure Library 865
Executing a Stored Procedure 865
Looping with the EXECUTE Command 871
Editing Procedures 874
Line Editor for the Mainframe Version 876
Appendix - The Mainframe Version 877
A.1.
A.2.
A.3.
A.4.
A.5.
The Mainframe Version of LIMDEP 877
The Fortran Source Code 877
Installation 878
Installation on Specific Systems 881
Using the Mainframe Version 882
References 885
Author Index 893
Subject Index 895
Command Index 903
Table of Contents

Part I Setting Up and Getting Started 1

Chapter 1 Introduction to LIMDEP 3
1.1. The LIMDEP Program 3
1.2. Econometric Techniques 4
1.3. Getting Started 5
Chapter 2 Installing and Executing LIMDEP 7
2.1. Introduction 7
2.2. Equipment 7
2.3. Installation 7
2.3.1. Windows 95 and Windows NT 7
2.3.2. DOS and Windows 3.1 8
2.4. Execution - Starting LIMDEP 10
2.4.1. Beginning a Session on a PC Under Windows 95/NT 10
2.4.2. Beginning an Interactive Session on a PC Using DOS or Windows 3.x 11
2.4.3. Starting Up on a Mainframe 13
2.5 Retaining Results Between Sessions - SAVE and LOAD 14
2.5.1. Saving a Windows 95/NT Session 15
2.5.2. Saving a DOS or Mainframe Session 15
2.5.3. Fast Input of a Data Set with OPEN/LOAD 17
2.6. Exiting LIMDEP 17
2.7. Restarting During a Session - The RESET Command 18
2.8. The Trace File 19
Chapter 3 Entering Commands - Basics of Operation 21
3.1. Introduction 21
3.2. Basic Command Entry 21
3.2.1. Loading a Project and Opening the Editor in the Windows Program 22
3.2.2. Entering Commands at the Main Prompt and Using the Editor in the DOS
Program 25
3.2.3. Interactive Command Entry in the Mainframe Program 30
3.3. Commands 30
3.3.1. Syntax 30
3.3.2. Naming Conventions and Reserved Names 32
3.4. Input Files - Entering Commands from a File 32
3.4.1. Submitting an Input File to the Windows Program 34
3.4.2. Submitting an Input File to the DOS or Mainframe Program 35
xxii
Table of Contents

3.5. Work Areas, Projects, and the STATUS Command 36
3.5.1. Work Areas 36
3.5.2. The Project Window in the Windows Program 37
3.5.3. The STATUS Command in the DOS and Mainframe Programs 41
3.6. Program Output and the Output Window 43
3.6.1. The Windows Output Window 43
3.6.2. Opening an Output File 43
3.6.3. Editing Your Output - Edit, Cut, Paste, and Copy 44
3.7. HELP 44
3.7.1. HELP in the Windows Program 44
3.7.2. The HELP Command in the DOS Program 44
3.8 Features of the DOS Program 45
3.8.1. Using the File System 45
3.8.2. Examining Files with the SHOW Command 45
3.8.3. The DISK Command and the Disk Manager 46
3.8.4. Using DOS - The SYSTEM Command 47
3.8.5. Demonstration and a Tutorial 47
3.9. A Summary of Commands 50
3.10. Summary of the LIMDEP Desktop 54
3.10.1. The LIMDEP Windows 55
3.10.2. The Main Menus 56
3.10.3. The Toolbar 61
3.10.4. The Command Window 61
3.10.5. Commands and Menu Items 62
Part II Data Input, Export, and Transformation 63

Chapter 4 Entering Data and Creating Data Files 65
4.1. Introduction 65
4.2. The Data Area and the ROWS Command 65
4.3. Using the Data Editor 66
4.3.1. Data Editor in the Windows Program 66
4.3.2. Data Editor in the DOS Program 68
4.4. Reading Data Files 70
4.4.1. Variable Names 70
4.4.2. Missing Values 72
4.4.3. Transposed Data Files - Reading by Variables 73
4.5. Formatted Data 73
4.5.1. Converting Blanks to Missing Values 75
4.5.2. Format Codes - Some Pointers 75
4.6. Large Data Sets 77
4.7. Spreadsheet Files and Binary Files 77
4.8. Adding Observations - The APPEND Command 79
4.9. Writing Data Files - the WRITE Command 80
4.10. Listing Data 80
Table of Contents
xxiii

Chapter 5 Data Transformations 81
5.1. Introduction 81
5.2. The CREATE Command 81
5.2.1. Algebraic Transformations 82
5.2.2. Conditional Transformations 84
5.2.3. CREATE Functions 86
5.2.4. Random Number Generators 92
5.2.5. Editing Data 94
5.3. Recoding Variables - The RECODE Command 94
5.4. Sorting Variables - The SORT Command 95
5.5. The DELETE and RENAME Commands 96
Chapter 6 Panel Data and the Discrete Choice Model 97
6.1.
6.2.
6.3.
6.4.
6.5.
6.6.
6.7.
6.8.
Introduction 97
Programs Which Use Panel Data 97
Conventions 97
Stratification Indicators 98
Indicators and Group Size Variables for Panel Data 99
Matrix Commands for Panel Data 101
Converting Data for the Discrete Choice Models 101
Merging Invariant Variables into a Panel 102
Chapter 7 Variable Lists, the Current Sample, and Missing Data 105
7.1. Introduction 105
7.2. Lists of Variables 105
7.3. The NAMELIST Command 106
7.3.1. Uses of Namelists 107
7.3.2. Deleting Namelists 107
7.3.3. Indexing Variables in Namelists 108
7.4. The Current Sample 108
7.4.1. Cross Section Data 109
7.4.2. Time Series Data 111
7.4.3. Random Sampling from the Current Sample - The DRAW Command 112
7.5. Missing Data 113
7.5.1. Reading Missing Data 113
7.5.2. Missing Data in Transformations 113
7.5.3. Missing Data in Scalar and Matrix Algebra 114
7.5.4. Missing Data in Estimation Routines 114
7.5.5. Automatically Bypassing Missing Data - The SKIP Command 115
xxiv Table of Contents

Part III Model Estimation 117

Chapter 8 Commands for Estimating Models 119
8.1. Introduction 119
8.2. Model Commands 119
8.3. Output from Estimation Programs 120
8.3.1. Displaying Covariance Matrices 122
8.3.2. Controlling Output to the Screen - FAST and SILENT Execution 122
8.4. Output Files 123
8.4.1. Displaying the Contents of the Output File and the Execution Trace 123
8.4.2. Messages in the Trace File - The TYPE and TIMER Commands 124
8.5. The Review Window - Tables of Model Results 124
8.6. Numerical Optimization 125
8.6.1. Technical Display During Optimization 125
8.6.2. Interrupting the Iterations 126
8.6.3. Exit Codes 127
8.6.4. Iteration Controls 127
8.6.5. Starting Values 130
8.6.6. Hints for Iterative Estimation 131
8.6.7. Technical Details on Optimization 132
8.7 Multivariate Normal Probabilities 134
Chapter 9 Analyzing Models and Testing Hypotheses 137
9.1. Introduction 137
9.2. Model Components 138
9.2.1. Constant Terms 138
9.2.2. Using Weights 138
9.2.3. Lags and Logarithms 139
9.2.4. Partial Differences 140
9.3. Standard Output 141
9.4. Marginal Effects 144
9.5. Retrievable Results 147
9.6. The Last Model and Functions of Parameters 149
9.7. Creating and Displaying Predictions and Residuals 151
9.8. Imposing Equality and Fixed Value Restrictions 155
9.9. Linear Functions of Model Parameters 157
9.9.1. Computing Linear Functions of Parameters 158
9.9.2. F Tests in Linear Models 159
9.9.3. Wald Tests in Nonlinear Models 161
9.10. Nonlinear Restrictions - The WALD Command 162
9.11. Tests Based on Likelihood Functions - LR and LM Tests 165
9.11.1. Likelihood Ratio Tests 165
9.11.2. Lagrange Multiplier Tests 166
9.12. Hausman Tests 171
9.13. Moment Based Specification Tests 172
Table of Contents
xxv

Chapter 10 Matrix Algebra 177
10.1. Introduction 177
10.2. Entering MATRIX Commands 180
10.2.1. Dialog Mode in the Windows Program 180
10.2.2. Dialog Mode in the DOS Program 181
10.2.3. Dialog Mode in the Mainframe Program 181
10.2.4. Command Mode MATRIX Commands 181
10.2.5. Forms of MATRIX Commands - Conditional Commands 182
10.2.6. Output 182
10.2.7. Matrix Results in Output Files 183
10.2.8. Unformatted Output 184
10.2.9. Statistical Output 185
10.2.10. Descriptive Statistics for the Elements in a Matrix 185
10.2.11. Plotting Matrices 185
10.3. Using MATRIX Commands with Data 185
10.3.1. Data Matrices 186
10.3.2. Computations Involving Data Matrices 187
10.4. Manipulating Matrices 188
10.4.1. Matrix Work Areas 189
10.4.2. Naming and Notational Conventions 189
10.4.3. Matrix Dimensions 190
10.4.4. Placing Matrix Results in Scalars 190
10.4.5. Matrix Expressions 191
10.4.6. Scalar Multiplication of a Result - Using CALCULATE 194
10.4.7. Adding the Same Scalar to Every Element of a Matrix 195
10.4.8. Raising a Matrix to a Power 195
10.5. Entering, Moving, and Rearranging Matrices 196
10.6. Matrix Functions 200
10.6.1. Functions of One Matrix 200
10.6.2. Functions of Two or More Matrices 204
10.7. Sums of Observations 204
10.8. Examples 211
10.9. Changes from Earlier Versions of LIMDEP 217
Chapter 11 Manipulating Scalars with the CALCULATE Command 219
11.1.
11.2.
11.3.
11.4.
Introduction 219
Command Input in CALCULATE 220
Output from CALCULATE 222
Forms of CALCULATE Commands - Conditional Commands 224
11.4.1. Documenting Calculations in the Output and Trace Files 225
11.4.2. Work Space for the Calculator 225
11.5. Scalar Expressions 226
11.6. Calculator Functions 227
xxvi Table of Contents

Part IV Data Description 235

Chapter 12 Describing Data 237
12.1. Introduction 237
12.2. Summary Statistics 237
12.2.1. Weights and Missing Observations 238
12.2.2. Stratification 239
12.2.3. Matrix Functions for Describing Panel Data 239
12.2.4. Sample Quantiles 241
12.3. Histograms 242
12.4. Cross Tabulations 245
12.4.1. Output 245
12.4.2. Testing the Independence Assumption 246
12.4.3. Analyzing Frequency Data 247
12.4.4. An Application 247
12.5. Box-Jenkins Time Series Identification 248
Chapter 13 Scatter Diagrams and Plotting 251
13.1. Introduction 251
13.2. Printing and Exporting Figures 251
13.2.1. Printing and Saving Graphs in the Windows Program 251
13.2.2. Printing and Saving Graphs in the DOS Program 251
13.3. Scatter Plots 252
13.3.1. Plotting One Variable Against Another 252
13.3.2. Time Series Plots 254
13.3.3. Plotting Several Variables Against One Variable 255
13.3.4. Options for Scaling the Figure 256
13.3.5. Grids and Lines in the Plotting Field 256
13.3.6. A Program for Plotting Confidence Regions 257
13.4. Multiple Scatter Plots - The SPLOT Command 258
13.5. Plotting Matrices - The MPLOT Command 259
13.5.1. Plotting Autocorrelation and Partial Autocorrelation Functions 259
13.5.2. Examining an Estimation Criterion (Log-likelihood) Function 260
13.6 Plotting Functions - The FPLOT Command 261
Part V Single and Multiple Equation Regression Models 263

Chapter 14 The Linear Regression Model 265
14.1. Introduction 265
14.2. Least Squares Regression 265
14.2.1. Omitted Variables 266
14.2.2. Retrievable Results 267
14.2.3. Application 267
Table of Contents
xxvii

14.3. Predictions and Residuals 268
14.3.1. Plotting Residuals 269
14.3.2. Standardized Residuals and Regression Diagnostics 271
14.4. Hypothesis Tests 273
14.4.1. Restricted Least Squares and Linear Restrictions 274
14.4.2. Testing Nonlinear Restrictions 277
14.4.3. Tests of Structural Change 277
14.4.4. Tests of Nonnested Hypotheses 280
14.4.5. Testing for Linearity vs. Loglinearity 286
14.4.6. CUSUM Test of Model Stability 286
14.5. Stepwise Regression 289
Chapter 15 Heteroscedasticity in the Linear Model 291
15.1. Introduction 291
15.2. Correcting the OLS Covariance Matrix 291
15.3. Estimating Models with Heteroscedasticity 294
15.3.1. Weighted Least Squares 294
15.3.2. A Model of Multiplicative Heteroscedasticity 294
15.3.3. Variance Proportional to the Square of the Mean 299
15.3.4. Autoregressive Conditional Heteroscedasticity (ARCH) Model 300
15.4. Testing for Heteroscedasticity 304
Chapter 16 Autocorrelation in the Linear Model 307
16.1.
16.2.
16.3.
16.4.
16.5.
16.6.
Introduction 307
Correcting the OLS Covariance Matrix 307
Correcting for First Order Autocorrelation 308
Autocorrelation with a Lagged Dependent Variable 311
Differencing and Higher Order Autocorrelation 312
Testing for Autocorrelation 313
Chapter 17 Linear Models with Panel Data 315
17.1. Introduction 315
17.2. Data Arrangement and Setup 316
17.2.1. Contiguous Data 317
17.2.2. Groupwise Data Summary Statistics 317
17.3. One Way Fixed and Random Effects Models 318
17.3.1. Commands for One Factor Models 319
17.3.2. Program Output for One Factor Models 320
17.3.3. Saved Results 321
17.3.4. Application 322
17.3.5. Restricted Least Squares 326
17.3.6. Robust Estimation of the OLS Covariance Matrix 326
17.3.7. The Group Means Estimator 327
17.3.8. Autocorrelation 328
17.3.9. Two Stage Least Squares for the Fixed Effects Model 331
17.3.10. Technical Details on Estimation of One Factor Models 333
xxviii Table of Contents

17.4. Two Factor Fixed and Random Effects Models 338
17.4.1. Program Output for Two Factor Models 339
17.4.2. Application 340
17.4.3. Technical Details 342
17.5. Random Coefficients Model 344
17.5.1. Application 345
17.5.2. Technical Details 346
17.5.3. Predicting Group Specific Coefficient Vectors 346
17.6. Groupwise Heteroscedasticity, Correlation, and Autocorrelation 347
17.6.1. Command and Options 347
17.6.2. Technical Details 349
17.6.3. Application 350
Chapter 18 ARIMA, ARMAX and Distributed Lag Models 357
18.1. Introduction 357
18.2. Box-Jenkins ARIMA and ARMAX Models 357
18.2.1. Model Command 358
18.2.2. Model Output 359
18.2.3. Example 360
18.2.4. Technical Details 364
18.3. Polynomial Distributed Lag Models 365
18.3.1. Example 368
18.4. The Geometric Lag Model 370
18.4.1. Application to the GNP/Money Data 373
18.5. Roots of Dynamic Equations 375
Chapter 19 Nonlinear Least Squares, 2SLS, IV, and GMM Estimation 377
19.1. Introduction 377
19.2. The Box-Cox Regression Model 377
19.2.1. Model Commands 378
19.2.2. Output and Saved Results 381
19.2.3. Application 382
19.2.4. Technical Details 387
19.3. Nonlinear Least Squares for Nonlinear Regression 389
19.3.1. Command for Nonlinear Regression 390
19.3.2. Specification of the Regression Function 390
19.3.3. Recursive Functions and Providing Derivatives 393
19.3.4. Options for Nonlinear Least Squares 394
19.3.5. Model Output and Retrievable Results 395
19.3.6. Imposing Restrictions and Testing Hypotheses 396
19.3.7. Application 397
19.3.8. Technical Details 400
19.4. Two Stage Least Squares Estimation of Linear Models 401
19.4.1. Autocorrelation with a Lagged Dependent Variable 401
19.4.2. Robust Estimation of the 2SLS Covariance Matrix 402
19.4.3. Two Stage Least Squares Estimation with Fixed Effects 403
19.4.4. Model Output for the 2SLS Command 403
19.4.5. Application 403
Table of Contents
xxix

19.5. Nonlinear Two Stage Least Squares Estimation 406
19.6. GMM Estimation of Nonlinear Models 407
Chapter 20 Systems of Regression Equations 409
20.1. Introduction 409
20.2. Linear SURE Models Estimated by GLS 409
20.2.1 Command for SURE Estimation 410
20.2.2. Options for the Generalized Least Squares Procedure 410
20.2.3. Model Output for the GLS Estimator 412
20.2.4. Application - The Translog System 412
20.2.5. Application - Generalized Least Squares 414
20.2.6. Technical Details on Generalized Least Squares 419
20.3. Maximum Likelihood Estimation of Constrained Linear Systems 420
20.3.1. Command for ML Estimation of Constrained SURE Systems 421
20.3.2. Model Output for the Maximum Likelihood Estimator 422
20.3.3. Application 422
20.3.4. Technical Details 426
20.4. Instrumental Variables (3SLS) Estimation of a Set of Equations 427
20.5. Nonlinear Systems of Regression Equations 429
20.5.1. Commands for Nonlinear Systems - The NLSUR Command 429
20.5.2. Output and Saved Results from NLSUR 432
20.5.3. Application 433
20.5.4. Technical Details 435
Part VI Qualitative, Discrete, and Count Dependent Variables 437

Chapter 21 Probit, Logit, and Other Models for Binary Choice 439
21.1. Introduction 439
21.1.1. Formulations of Discrete Choice Models 439
21.1.2. The Index Function Approach to Binary Choice 440
21.1.3. Grouped and Individual Data for Discrete Choice Models 441
21.1.4. Weights 441
21.1.5. Choice Based Sampling 441
21.1.6. Methods of Estimation and Options 442
21.1.7. Restrictions 443
21.2. The Probit Model for Binary Choice 443
21.2.1. Command for the Univariate Probit Model 444
21.2.2. Output and Retrievable Results from the Probit Model 445
21.2.3. Applications 447
21.3. Heteroscedasticity in the Probit Model 450
21.3.1. Application 452
21.3.2. Technical Details 453
21.4. Panel Data and Random Effects in the Probit Model 454
21.4.1. Output from the Random Effects Estimator 454
21.4.2. Retrievable Results 455
21.4.3. Application 455
21.4.4. Technical Details 458
xxx
Table of Contents

21.5. The Binomial Logit Model 459
21.5.1. Output from the LOGIT Command 460
21.5.2. Applications 460
21.5.3. Heteroscedasticity in the Binomial Logit Model 462
21.6. Fixed and Random Effects in the Binomial Logit Model 464
21.7. User Defined Index Models for Binary Choice 467
21.8. Semiparametric Analysis of Binary Choice - MSCORE 469
21.8.1. Command for MSCORE 470
21.8.2. Output from MSCORE 473
21.8.3. Technical Details 474
21.8.4. Extensions 475
21.9. Nonparametric Regression Based on Binary Choice - NPREG 476
21.9.1. Command for NPREG 476
21.9.2. Output from NPREG 478
21.9.3. Application 479
Chapter 22 Bivariate and Multivariate Probit and Partial Observability Models 483
22.1. Introduction 483
22.2. Estimating the Bivariate Probit Model 483
22.2.1. Options for the Bivariate Probit Model 484
22.2.2. Heteroscedasticity 485
22.2.3. Model Results for the Bivariate Probit Model 486
22.2.4. Marginal Effects 487
22.2.5. Application 489
22.2.6. Technical Details 491
22.3. Bivariate Probit Models with Sample Selection 492
22.4. Bivariate Probit Models with Partial Observability 493
22.5. The Multivariate Probit Model 494
22.5.1. Marginal Effects 495
22.5.2. Retrievable Results 496
22.5.3. Technical Considerations 496
Chapter 23 Ordered Probability Models 497
23.1. Introduction 497
23.2. Commands for Basic Ordered Probit and Logit Models 498
23.2.1. Censoring in the Ordered Probability Models 500
23.2.2. Output from the Ordered Probability Estimators 500
23.2.3. Applications 501
23.3. A Random Effects Model for Panel Data 505
23.4. Stratification 507
23.5. Heteroscedasticity 507
23.6. Sample Selection 509
23.7. Technical Details 510
Table of Contents
xxxi

Chapter 24 Multinomial Logit and Discrete Choice Models - LOGIT and NLOGIT 513
24.1. Introduction 513
24.2. The Multinomial Logit Model 514
24.2.1. Model Command for the Logit Models 514
24.2.2. Output for the Logit Models 515
24.2.3. Application 517
24.2.4. Computing Predicted Probabilities for the Multinomial Logit Model 518
24.3. NLOGIT - Discrete Choice Models 519
24.4. Models for Discrete Choice 520
24.5. Data for the Discrete Choice Models 521
24.5.1. Setting Up the Data 521
24.5.2. Fixed and Variable Numbers of Choices 522
24.5.3. Types of Data on the Choice Variable 524
24.5.4. Entering Data on a Single Line 525
24.6. Model Commands for the Discrete Choice Model 527
24.7. Options for the Discrete Choice Model 529
24.7.1. Weighting and Choice Based Sampling 529
24.7.2. Predicted Probabilities and Inclusive Values 530
24.7.3. Marginal Effects and Elasticities 531
24.7.4. Descriptive Statistics 532
24.7.5. Simulating the Model 533
24.7.6. Lagrange Multiplier, Wald, and Likelihood Ratio Tests 534
24.7.7. Testing the Independence of Irrelevant Alternatives 534
24.7.8. Scaling the Data 536
24.7.9. Adding Descriptive Output 537
24.8. Alternatives to the Multinomial Logit Model 537
24.8.1. Heteroscedastic Extreme Value Model 537
24.8.2. Random Parameters (Mixed) Logit Model 540
24.8.3. Multinomial Probit Model 543
24.9. Applications 547
Chapter 25 Nested Logit Models - NLOGIT 557
25.1. Introduction 557
25.2. Mathematical Specification of the Model 558
25.3. Commands for FIML Estimation 559
25.3.1. Data Setup 559
25.3.2. Model Command 560
25.3.3. Utility Functions 561
25.3.4. Inclusive Value Parameters 565
25.3.5. Starting Values for Parameters and Fixed Parameters 567
25.3.6. Marginal Effects and Elasticities 569
25.3.7. Inclusive Values, Utilities, and Probabilities 570
25.3.8. Testing the Independence of Irrelevant Alternatives 571
25.3.9. A Model of Covariance Heterogeneity 572
25.4. Sequential (Two Step) Estimation of Nested Logit Models 573
25.5. Applications 577
25.6. Combining Data Sets and Scaling in Discrete Choice Models 587
25.6.1. Joint Estimation 588
xxxii Table of Contents

25.6.2. Sequential Estimation 590
25.7. Technical Details 591
25.7.1. Log-Likelihood and Derivatives for the Three Level Nested Logit Model 591
25.7.2. Computations for the Covariance Matrix for Two Step Estimates 594
Chapter 26 Poisson and Negative Binomial Regression Models for Count Data 595
26.1. Introduction 595
26.2. Poisson and Negative Binomial Regression Models 598
26.2.1. Censoring and Truncation 599
26.2.2. Output for the Count Data Models 601
26.2.3. Applications of the Poisson Model 602
26.2.4. The Negative Binomial Model 613
26.2.5. Technical Details 615
26.2.6. Poisson and Negative Binomial Models with Unobserved Heterogeneity 616
26.2.7. Poisson Models with Underreporting 623
26.3. Count Data Models for Panel Data 629
26.3.1. Output 630
26.3.2. Applications 631
26.3.3. Technical Details 633
26.3.4. A Semiparametric Poisson Regression with Heterogeneity and Population
Splitting (Intermittency) for Panel Data 635
26.4. Zero Inflated Poisson (ZIP) Models 640
26.4.1. Commands for the ZIP Models 642
26.4.2. Output for the ZIP Models 644
26.4.3. Applications 645
26.4.4. Technical Details 648
26.5. Models for Sample Selection 649
26.5.1. Commands for the Sample Selection Model 649
26.5.2. Technical Details 651
26.5.3. A Limited Information, Nonlinear Least Squares Approach 651
26.5.4. Full Information Maximum Likelihood Estimation 653
26.5.5. Commands for Estimating Selection Models 655
26.5.6. Application 656
Part VII Limited Dependent Variables and Frontier Regression Models 659

Chapter 27 Tobit, Censoring, and Truncation Models 661
27.1. Censored Regression Models 661
27.2. The Tobit Model for Censored Data 662
27.2.1. Output for the Tobit Model 664
27.2.2. Applications 665
27.3. Variants of the Tobit Model 674
27.3.1. Heteroscedasticity 674
27.3.2. Random Effects in Panel Data 676
27.3.3. Tobit Model with Sample Selection 678
27.3.4. Simultaneous Equations Tobit Model 683
Table of Contents
xxxiii

27.4.
27.5.
27.6.
27.7.
27.3.5. Bivariate and Sequential, Nested Tobit Models 686
27.3.6. Nonnormal Distributions for the Tobit Model 687
Technical Notes for the Tobit Models 695
The Truncated Regression Model 700
The Grouped Data Regression Model 703
27.6.1. Grouped Data with Sample Selection 704
27.6.2. Applications 706
The Lognormal Regression Model 708
Chapter 28 Sample Selection and Switching Regression Models 711
28.1. Introduction 711
28.2. Regression Models with Sample Selection 712
28.2.1. Estimation of the Standard Model 713
28.2.2. A Model For Treatment Effects - Using All Observations 716
28.2.3. Simultaneous Equations Models with Selectivity 717
28.2.4. Tobit Model with Selectivity 718
28.2.5. A Model of Incidental Truncation 718
28.2.6. Qualitative Dependent Variables 720
28.3. Different Specifications of the Selection Equation 721
28.3.1. The Univariate Probit Model 721
28.3.2. A Binary Logit Selection Model 721
28.3.3. Multinomial Logit Selection Model 722
28.3.4. Discrete Choice Selection Rule 725
28.3.5. Tobit Selection Rule 727
28.3.6. Ordered Probit Selection Rule 730
28.3.7. Bivariate Probit Selection Rule 733
28.4. Mixtures of Discrete, Censored, and Continuous Variables 735
28.4.1. One Variable Censored 735
28.4.2. Both Variables Binary 737
28.4.3. Sample Selection with Two Treatments 738
28.5. Technical Details 739
28.6. Switching Regression Models 741
28.6.1. Model Commands 742
28.6.2. Results for Switching Regression Models 744
28.6.3. Application 746
28.6.4. Technical Details 751
Chapter 29 Stochastic Frontier Regression Models 753
29.1.
29.2.
29.3.
29.4.
Introduction 753
Model Command 754
Results for the Stochastic Frontier Estimators 755
Applications 757
29.4.1. Cross Section Models 760
29.4.2. Panel Data Model 762
29.4.3. A Method of Moments Estimator for a Normal-Gamma Model 763
29.5. Log-Likelihood Functions for the Frontier Models 765
xxxiv Table of Contents

Part VIII Duration Models 767

Chapter 30 Nonparametric Analysis of Duration Data 769
30.1. Introduction 769
30.2. Life Tables 769
30.3. Commands for Life Tables 771
30.3.1. Tables for Individuals and Specific Exit Times 771
30.3.2. Stratification 772
30.4. Applications 772
30.4.1. Strike Duration Data 772
30.4.2. An Example With Stratification 778
30.5. Mathematical Details for the Homogeneity Tests 782
Chapter 31 Semiparametric and Proportional Hazard Models 783
31.1. Introduction 783
31.2. The Proportional Hazards Model 783
31.3. Commands for the Proportional Hazards Model 784
31.3.1. Plotting the Survival and Integrated Hazard Functions 785
31.3.2. Keeping the Survival and Integrated Hazard Functions 785
31.3.3. Time Dependent Covariates 785
31.3.4. Stratification 789
31.4. Output for the Proportional Hazards Model 789
31.5. The Ordered Logit Model 790
31.6. Applications 791
31.6.1. Cox’s Proportional Hazards Model 791
31.6.2. The Ordered Logit Model 800
Chapter 32 Parametric Models of Duration 803
32.1.
32.2.
32.3.
32.4.
32.5.
Introduction 803
Loglinear Models for Survival Data 804
Commands for Estimating Loglinear Models 805
Output from Loglinear Model Commands 807
Extensions of the Loglinear Models 808
32.5.1. Time Varying Covariates 808
32.5.2. Heterogeneity 810
32.5.3. Split Population Survival Models 811
32.5.4. Truncation 813
32.5.5. The Gompertz Model 814
32.5.6. The Generalized F Model 815
32.5.7. Heterogeneity in the Scale Parameter for Loglinear Models 816
32.6. Applications 817
32.6.1. A Comparison of Loglinear Models 817
32.6.2. The Generalized F Distribution 824
32.7. Technical Details 825
32.7.1. The Scoring Method for the Loglinear Models 825
32.7.2. Estimation with Time Varying Covariates 825
Table of Contents
xxxv

Part IX Nonlinear Optimization and LIMDEP Programming 827

Chapter 33 Nonlinear Optimization 829
33.1. Introduction 829
33.2. The MINIMIZE/MAXIMIZE Commands 829
33.2.1. Options 830
33.2.2. Function Definition 830
33.2.3. Subfunctions 832
33.2.4. Supplying Derivatives 833
33.3. Optimization Problems 834
33.3.1. Minimizing a Simple Function 834
33.3.2. Solutions to Equations 834
33.3.3. Minimizing a Sum of Terms 835
33.4. Output from MINIMIZE/MAXIMIZE 835
33.5. Applications 836
Chapter 34 Analyzing Nonlinear Functions 839
34.1.
34.2.
34.3.
34.4.
34.5.
Introduction 839
Plotting a Function 839
Variances for Nonlinear Functions 842
Differentiation 844
Integration 845
Chapter 35 Programming Tools 849
35.1.
35.2.
35.3.
35.4.
35.5.
Introduction 849
Estimation Programs and Post-Processing 849
Numeric Computation 850
Batching Commands 851
Flow Control 853
35.5.1. Logical Expressions 853
35.5.2. Examples 854
35.6. Looping with DO Statements 855
Chapter 36 Using the Command Editor 857
36.1.
36.2.
36.3.
36.4.
36.5.
Introduction 857
The Windows Text Editor Window 857
Using the Editor in the DOS Program 858
The SILENT Command 860
Files 861
36.5.1. Reading a File into the Editor 861
36.5.2. Writing an ASCII File from the Editor 861
Chapter 37 Procedures and the EXECUTE Command 863
37.1. Introduction 863
xxxvi Table of Contents

37.2. Storing a Procedure 863
37.3. The Procedure Library 865
37.4. Executing a Stored Procedure 865
37.4.1. Silent Execution 865
37.4.2. Parameters and Character Strings in Procedures 866
37.4.3. Macros - The STRING Command 867
37.4.4. Repeated Execution of a Procedure 868
37.4.5. Execution with One Scalar Parameter 868
37.4.6. Pausing During Execution 869
37.4.7. Conditional Execution 869
37.4.8. Defining Exit (Convergence) Criteria 869
37.4.9. Query for Exit from Iterations 870
37.4.10. Query for a Parameter to Use in the Procedure 871
37.5. Looping with the EXECUTE Command 871
37.5.1. Looping Over a Set of Variables 872
37.5.2. Indexing in Namelists 873
37.5.3. Loops Within Procedures 873
37.6. Editing Procedures 874
37.6.1. Editing in the Windows Program 874
37.6.2. Editing Procedures in the DOS Program 875
37.7. Line Editor for the Mainframe Version 876
Appendix - The Mainframe Version 877
A.1. The Mainframe Version of LIMDEP 877
A.2. The Fortran Source Code 877
A.3. Installation 878
A.3.1. Program LIMDEP - The Main Program 878
A.3.2. Subroutine SETNPT - The Default Logical Units 880
A.3.3. Subroutine PROMPT - Output to a Terminal 880
A.3.4. Subroutine GETLN - Input from a Terminal 880
A.3.5. Subroutine FILES - Opening Files 880
A.3.6. Subroutine FCLOSE - Closing Files 880
A.4. Installation on Specific Systems 881
A.4.1. Initiation 881
A.4.2. Installation on a Sun 881
A.4.3. Installation on a VAX 881
A.4.4. Installation on Other Non-IBM Systems 882
A.4.5. Installation on IBM and Amdahl Systems 882
A.5. Using the Mainframe Version 882
A.5.1. Commands that Are Not Supported 883
A.5.2. Input and Output Files 883
A.5.3. Differences in the READ Command 883
A.5.4. SAVE, LOAD, and SHOW 884
References 885
Author Index 893
Subject Index 895
Command Index 903
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement