Download the full version of the ebook at ebookfinal.

com

Regression Models for Categorical Dependent

Variables Using Stata 3rd Edition J. Scott Long

https://ebookfinal.com/download/regression-models-for-
categorical-dependent-variables-using-stata-3rd-edition-j-
scott-long/

OR CLICK BUTTON

DOWNLOAD EBOOK

Download more ebook instantly today at https://ebookfinal.com

Instant digital products (PDF, ePub, MOBI) available
Download now and explore formats that suit you...

Regression Models for Categorical Dependent Variables

Using Stata 1st Edition J. Scott Long

https://ebookfinal.com/download/regression-models-for-categorical-
dependent-variables-using-stata-1st-edition-j-scott-long/

ebookfinal.com

Regression Using JMP 1st Edition Rudolf J. Freund

https://ebookfinal.com/download/regression-using-jmp-1st-edition-
rudolf-j-freund/

ebookfinal.com

Regression Analysis An Intuitive Guide for Using and

Interpreting Linear Models 1st Edition Jim Frost

https://ebookfinal.com/download/regression-analysis-an-intuitive-
guide-for-using-and-interpreting-linear-models-1st-edition-jim-frost/

ebookfinal.com

Data Analysis Using Regression and Multilevel Hierarchical

Models 1st Edition Andrew Gelman

https://ebookfinal.com/download/data-analysis-using-regression-and-
multilevel-hierarchical-models-1st-edition-andrew-gelman/

ebookfinal.com
Microeconometrics Using Stata Second Edition Volume II
Nonlinear Models and Casual Inference Methods A. Colin
Cameron
https://ebookfinal.com/download/microeconometrics-using-stata-second-
edition-volume-ii-nonlinear-models-and-casual-inference-methods-a-
colin-cameron/
ebookfinal.com

Bootstrap Tests for Regression Models 1st Edition Leslie

Godfrey (Auth.)

https://ebookfinal.com/download/bootstrap-tests-for-regression-
models-1st-edition-leslie-godfrey-auth/

ebookfinal.com

A handbook of statistical analyses using Stata 3rd ed

Edition Brian S. Everitt

https://ebookfinal.com/download/a-handbook-of-statistical-analyses-
using-stata-3rd-ed-edition-brian-s-everitt/

ebookfinal.com

Counselling for Post traumatic Stress Disorder 3rd Edition

Michael J Scott

https://ebookfinal.com/download/counselling-for-post-traumatic-stress-
disorder-3rd-edition-michael-j-scott/

ebookfinal.com

Bayesian Models for Categorical Data Wiley Series in

Probability and Statistics 1st Edition Peter Congdon

https://ebookfinal.com/download/bayesian-models-for-categorical-data-
wiley-series-in-probability-and-statistics-1st-edition-peter-congdon/

ebookfinal.com
Regression Models
For Categorical
Dependent Variables
Using Stata

T h i r d E d it i o n
Regression Models for Categorical
Dependent Variables Using Stata
Third Edition
Regression Models for Categorical
Dependent Variables Using Stata
Third Edition

J. SC O T T LONG
Departments o f Sociology and Statistics
Indiana University
Bloomington, Indiana

.JEREMY FREESE
Department o f Sociology and Institute for Policy Research
Northwestern University
Evanston, Illinois

Press
A S ta ta Press Publication
S tataC orp LP
College Station, Texas
 ® Copyright © 2001, 2003, 2006, 2014 by StataC orp LP
All rights reserved. F irst edition 2001
Revised ed ition 2003
Second ed ition 2006
Third edition 2014

Published by Stata Press, 4905 Lakeway Drive, C ollege S tation , Texas 77845
T ypeset in LM^X 2e
Printed in the United S ta tes of America
10 9 8 7 6 5 4 3 2 1

ISBN-10: 1-59718-111-0
ISBN-13: 978-1-59718-111-2

Library o f Congress C ontrol Number: 2014948009

N o part o f this book m ay be reproduced, stored in a retrieval system , or transcribed, in any

form or by any m eans— electronic, mechanical, ph otocop y, recording, or o th erw ise— without
th e prior written perm ission of StataCorp LP.

Stata, S r a ia , Stata P ress, M ata, m a ia . and N etC ourse are registered trademarks of
S tataC orp LP.

S ta ta and Stata Press are registered trademarks w ith the World Intellectual P roperty Organi­
zation o f the United N ations.

N etCourseNow is a tradem ark of StataCorp LP.

I£Te X 2£ is a trademark o f the American M athem atical Society.

O ther brand and product nam es are registered tradem arks or trademarks o f their respective
com panies.
To our parents
Contents
L is t o f fig u res x ix
P re f a c e xxi

I G eneral inform ation 1

1 I n t r o d u c ti o n 7
1.1 W hat is this book a b o u t ? ............................................................................ 7
1.2 Which models are considered?...................................................................... 8
1.3 Whom is this book for? ............................................................................... 9
1.4 How is the book o rg a n iz e d ? ......................................................................... 9
1.5 The SPost so ftw a re ......................................................................................... 11
1.5.1 Updating S t a t a ............................................................................... 12
1.5.2 Installing SPost 1 3 ............................................................................ 13
Uninstalling S P o s t9 ........................................................................ 14
Installing SPost 13 using search ................................................. 14
Installing S P ostl3 using net i n s t a l l ........................................... 16
1.5.3 Uninstalling S P o s t l 3 ...................................................................... 17
1.6 Sample do-files and d a ta s e ts ......................................................................... 17
1.6.1 Installing the spost 13_do p a c k a g e ............................................... 17
1.6.2 Using spex to load data and run e x a m p le s ................................ 17
1.7 Getting help with S P o s t ................................................................................ 18
1.7.1 W hat if an SPost command does not w o r k ? ............................ 18
1.7.2 Getting help from the a u th o r s ...................................................... 19
W hat we need to help y o u ........................................................... 20
1.8 Where can I learn more about the models? ............................................. 21
2 I n tr o d u c tio n to S ta t a 23
The S ta ta i n t e r f a c e .......................................................
23
Abbreviations.............................................................. 27
Getting h e l p ............................................................................... 27
2.3.1 O nline help ...................................................................... 27
2.3.2 P D F m a n u a l s ................................................................... 28
2.3.3 E rro r m e s s a g e s .............................................................. 28
2.3.4 A sking for h e l p .............................................................. 28
2.3.5 O th e r r e s o u r c e s .............................................................. 29
The working d i r e c t o r y ................................................................. 29
Stata file t y p o s ............................................................................... 30
Saving o u tp u t to log files............................................................. 30
Using and saving d a t a s e t s .......................................................... 32
2.7.1 D a ta in S ta ta f o r m a t ................................................... 32
2.7.2 D a ta in o th e r formats ................................................ 33
2.7.3 E n terin g d a ta by h a n d ................................................ 33

Size lim itations on d a t a s e t s ...................................................... 34

Do-files ............................................................................................. 34

2.9.1 A dding c o m m e n ts .......................................................... 35

2.9.2 Long l i n e s ....................................................................... 36

2.9.3 S to pping a do-file while it is r u n n i n g .................... 37

2.9.4 C rea tin g d o -files............................................................. 37

2.9.5 R ecom m ended stru ctu re for d o -file s ....................... 38

Using S ta ta for serious d ata a n a l y s i s ..................................... 40

Syntax of S ta ta co m m a n d s......................................................... 41

2.11.1 C o m m a n d s ...................................................................... 43

2.11.2 V ariable lists ................................................................ 43

2.11.3 if an d in q ualifiers.........................................................
2.11.4 O ptions ..........................................................................
Managing d a t a .............................................................................
2.12.1 Looking a t your d a t a ..................................................
Contents

2.12.2 Getting inform ation about variables.......................

2.12.3 Missing v a lu e s ..............................................................
2.12.4 Selecting o b s e r v a tio n s ..............................................
2.12.5 Selecting variables ....................................................
2.13 Creating new v a ria b le s ..............................................................
2.13.1 The generate co m m an d .............................................
2.13.2 The replace c o m m a n d .............................................
2.13.3 The recode c o m m a n d .................................................
2.14 Labeling variables and v a l u e s .................................................
2.14.1 Variable la b e ls ..............................................................
2.14.2 Value l a b e l s .................................................................
2.14.3 The notes c o m m a n d .................................................
2.15 Global and local m a c r o s ..........................................................
2.16 Loops using foreach and forvalues..........................................
2.17 G ra p h ics........................................................................................
2.17.1 The graph c o m m a n d .................................................
2.18 A brief t u t o r i a l ...........................................................................
2.19 A do-file te m p la te ........................................................................
2.20 Conclusion.....................................................................................
3 E s tim a tio n , te s tin g , a n d fit
3.1 E stim ation.....................................................................................
3.1.1 S tata’s o u tp u t for ML e s tim a tio n .......................
3.1.2 ML and sample s iz e .................................................
3.1.3 Problems in obtaining ML e s t i m a t e s ................
3.1.4 Syntax of estim ation c o m m an d s..........................
3.1.5 Variable lists ..........................................................
Using factor-variable notation in the variable list
Specifying interaction and p o ly n o m ia ls .............
More on factor-variable notation .......................
3.1.6 Specifying the estimation s a m p l e .......................
x mts

M issing d a t a ............................................................................. 93
Inform ation about missing v a l u e s ..................................... 95
P ostestim ation commands and th e estimation sam ple . 98
3.1.7 W eights and survey d a t a ...................................................... 99
C om plex survey designs ...................................................... 100
3.1.8 O ptions for regression m o d e l s ............................................ 102
3.1.9 R o b ust standard e r r o r s ......................................................... 103
3.1.10 R eading the estim ation o u tp u t ........................................ 105
3.1.11 S toring estim ation results .................................................. 107
(Advanced) Saving estim ates to a f ile .............................. 108
3.1.12 R eform atting output w ith estim ates t a b l e .................... 111
3.2 T estin g ......................................................................................................... 114
3.2.1 O ne-tailed and two-tailed tests ........................................ 115
3.2.2 W ald and likelihood-ratio tests ........................................ 115
3.2.3 W ald tests with test and testp arm ................................. 116

3.2.4 LR tests w ith l r t e s t ................................................................ 118

Avoiding invalid LR tests .................................................. 120

3.3 Measures of fit ....................................................................................... 120

3.3.1 S yntax of f i t s t a t ...................................................................... 120

3.3.2 M ethods and formulas used by f i t s t a t ........................... 123

3.3.3 Exam ple of f its ta t................................................................... 129

3.4 estat postestim ation c o m m a n d s......................................................... 130

3.5 C onclusion................................................................................................. 131
4 M e th o d s o f i n t e r p r e t a t io n 133
4.1 Comparing linear and nonlinear models ........................................ 133
4.2 Approaches to interpretation ............................................................ 136
4.2.1 M ethod of interpretation based on p red ictio n s.............. 137
4.2.2 M ethod of interpretation using p a ra m e te rs..................... 138
4.2.3 S ta ta and SPost commands for in te rp re ta tio n .............. 138
4.3 Predictions for each observation........................................................ 138
Contents X1

4.4 Predictions at specified v a lu e s ..................................................................... 139

4.4.1 Why use th e in* commands instead of m a rg in s ? ..................... 140
4.4.2 Using margins for p re d ic tio n s ...................................................... 141
Predictions using interaction and polynom ial t e r m s ............. 146
Making m ultiple p re d ic tio n s ........................................................ 146
Predictions for groups defined by levels of categorical variables 150
4.4.3 (Advanced) Nondefault predictions using m a rg in s .................. 153
The predictQ o p t i o n ..................................................................... 153
The expression() o p t i o n .............................................................. 154
4.4.4 Tables of predictions using m ta b le ............................................... 155
mtable with categorical and count o u tc o m e s .......................... 158
(Advanced) Combining and form atting tables using mtable . 160

4.5 Marginal effects: Changes in p re d ic tio n s................................................. 162

4.5.1 Marginal effects using m argins..................................................... 163
4.5.2 Marginal effects using m t a b l e ..................................................... 164
4.5.3 Posting predictions and using m lin c o m ...................................... 165
4.5.4 Marginal effects using m c h a n g e .................................................. 166
4.6 Plotting predictions......................................................................................... 171
4.6.1 Plotting predictions with m arg in sp lo t........................................ 171
4.6.2 Plotting predictions using m g e n .................................................. 173
4.7 Interpretation of p a ra m e te rs ......................................................................... 178
4.7.1 The listcoef c o m m a n d ................................................................... 179
4.7.2 Standardized coefficients................................................................ 180
4.7.3 Factor and percentage change coefficients.................................. 184

4.8 Next s te p s ........................................................................................................... 184

II M od els for specific kinds of outcom es 185

5 M o d e ls for b in a ry o u tc o m e s : E s tim a tio n , te s tin g , and fit 187
5.1 The statistical m o d e l....................................................................................... 187
5.1.1 A latent-variable m odel................................................................... 188
Xll Contents

5.1.2 A nonlinear probability m o d e l ..................................................... 192

5.2 E stim ation using logit and probit com m ands ......................................... 192
5.2.1 E xam ple of logit m o d e l................................................................... 194
5.2.2 C om paring logit and p r o b i t ......................................................... 196
5.2.3 (Advanced) Observationspredicted perfectly..............................197
5.3 Hypothesis t e s t i n g ..........................................................................................200
5.3.1 Testing individual c o e ffic ie n ts...................................................... 200
5.3.2 Testing multiple c o e ffic ie n ts......................................................... 203
5.3.3 Com paring LR and Wald t e s t s ...................................................... 205
5.4 Predicted probabilities, residuals, and influential observations . . . . 206
5.4.1 P redicted probabilities using p r e d i c t .........................................206
5.4.2 Residuals and influential observations using p r e d i c t .............. 209
5.4.3 L east likely o b se rv a tio n s................................................................ 216
5.5 Measures of fit ................................................................................................218
5.5.1 Inform ation c r ite ria ................................................................ ... 219
5.5.2 Pseudo-R 2’s .......................................................................................221
5.5.3 (Advanced) Hosmer Lemeshow s t a t i s t i c ...................................223
5.6 Other com m ands for binary outcomes .......................................................225
5.7 C onclusion........................................................................................................... 225
6 M o d els for b i n a r y o u tco m es: I n t e r p r e t a t i o n 227
6.1 Interpretation using regressionc o e ffic ie n ts ..................................................228
6.1.1 Interpretation using odds r a t i o s ...................................................228
6.1.2 (Advanced) Interpretation using y * ............................................ 235
6.2 Marginal effects: Changes in p r o b a b ilitie s ............................................... 239
6.2.1 Linked variables.................................................................................241
6.2.2 Summ ary measures of c h a n g e ...................................................... 242
MEMs and M E R s............................................................................ 243
A M E s ............................................................................................... 243
S tandard errors of marginal e f f e c t s ...........................................244
6.2.3 Should you use the AME, the MEM, or the M E R ? .............. 244
Contents xiii

6.2.4 Examples of m arginal e f f e c ts .............................................. 246

AMEs for continuous variables.............................................. 248

AMEs for factor v a r ia b le s .................................................... 251

Summary table of A M E s........................................................ 252

Marginal effects for sub grou ps.............................................. 254

MEMs and M E R s .................................................................... 255

Marginal effects w ith powers and in te r a c tio n s ................ 259

6.2.5 The distribution of marginal e f fe c ts ...................................... 261

6.2.6 (Advanced) Algorithm for computing the distribution of

effects ........................................................................................ 265

6.3 Ideal ty p e s.................................................................................................. 270

6.3.1 Using local means with ideal t y p e s .................................... 273

6.3.2 Comparing ideal types with statistical t e s t s ...................... 274

6.3.3 (Advanced) Using macros to test differences between ideal

ty p e s ............................................................................................ 275

6.3.4 Marginal effects for ideal types ........................................... 278

6.4 Tables of predicted probabilities........................................................... 280

6.5 Second differences comparing marginal e f f e c ts ................................. 285

6.6 Graphing predicted p r o b a b ilitie s ........................................................ 286

6.6.1 Using m arg in sp lo t.................................................................... 287

6.6.2 Using mgen with the graph com m and ................................. 290

6.6.3 Graphing m ultiple p rediction s.............................................. 293

6.6.4 Overlapping confidence intervals........................................... 297

6.6.5 Adding power term s and plotting p r e d ic tio n s ................ 301

6.6.6 (Advanced) G raphs with local m e a n s ................................. 303

6.7 Conclusion.................................................................................................. 308

M o d e ls fo r o rd in a l o u tc o m e s 309

7.1 T he statistical m o d e l................................................................................ 310

7.1.1 A latent-variable m odel........................................................... 310

7.1.2 A nonlinear probability m o d e l.............................................. 314

7.2 Estimation using ologit and opro b it..................................................... 314

xiv Contents

7.2.1 E xam ple of ordinal logit m o d e l ................................................... 315

7.2.2 P redicting p e rfe c tly .......................................................................... 319
7.3 Hypothesis t e s t i n g ............................................................................................ 320
7.3.1 T esting individual c o e ffic ie n ts ......................................................321
7.3.2 T esting multiple c o e ffic ie n ts ......................................................... 322
7.4 Measures of fit using f i t s t a t ........................................................................... 324
7.5 (Advanced) Converting to a different p a ra m e te riz a tio n .........................325
7.G The parallel regression a s s u m p tio n .............................................................. 326
7.6.1 Testing the parallel regression assumption using oparallel . . 329
7.6.2 Testing the parallel regression assumption using b ran t .. . 330
7.6.3 Caveat regarding the parallel regression a ssu m p tio n ............331
7.7 Overview of in te rp re ta tio n ............................................................................331
7.8 Interpreting transformed c o e ffic ie n ts ........................................................332
7.8.1 Marginal change in y* ..................................................................332
7.8.2 Odds r a t i o s ..................................................................................... 335
7.9 Interpretations based on predicted p ro b a b ilitie s .................................... 338
7.10 Predicted probabilities with p r e d ic t............................................................ 339
7.11 Marginal e ffe c ts ................................................................................................341
7.11.1 P lotting marginal e f f e c ts ............................................................... 344
7.11.2 M arginal effects for a quick o v e r v ie w ....................................... 350
7.12 Predicted probabilities for ideal t y p e s .......................................................351
7.12.1 (Advanced) Testing differences between ideal types
.............................................................................................................354

7.13 Tables of predicted probabilities...................................................................355

7.14 Plotting predicted p ro b a b ilitie s................................................................... 359
7.15 Probability plots and marginal effects ..................................................... 364
7.16 Less common models for ordinal o u tc o m e s ...............................................370
7.16.1 T he vStereotype logistic m o d e l ..................................................... 370
7.16.2 The generalized ordered logit m o d e l........................................... 371
7.16.3 (Advanced) Predictions w ithout using factor-variable notation 374
Contents xv

7.16.4 The sequential logit m o d e l............................................................ 378

7.17 C onclusion.......................................................................................................... 382
8 M o d e ls fo r n o m in al o u tc o m e s 385
8.1 The multinomial logit m o d e l......................................................................... 386
8.1.1 Formal statem ent of the m o d e l .................................................. 390
8.2 Estimation using the mlogit c o m m a n d ...................................................... 390
Weights and complex s a m p le s .................................................... 391
O p t i o n s ........................................................................................... 391
8.2.1 Example of M N L M ........................................................................ 392
8.2.2 Selecting different base o u tc o m e s ............................................... 395
8.2.3 Predicting p e rfe c tly ........................................................................ 397
8.3 Hypothesis t e s t i n g ......................................................................................... 398
8.3.1 mlogtest for tests of the M N L M .................................................. 398
8.3.2 Testing the effects of the independent v a r ia b le s ..................... 399
8.3.3 Tests for combining altern ativ es.................................................. 403
8.4 Independence of irrelevant a lte rn a tiv e s......................................................407
8.4.1 Hausm an-M cFadden test of I I A .................................................. 408
8.4.2 Small-Hsiao test of IIA ............................................................... 409
8.5 Measures of fit ................................................................................................ 411
8.6 Overview of in te rp re ta tio n ............................................................................ 411
8.7 Predicted probabilities with p re d ic t............................................................ 412
8.8 Marginal effects................................................................................................ 415
8.8.1 (Advanced) The distribution of marginal e ffe c ts ..................... 420
8.9 Tables of predicted probabilities................................................................... 423
8.9.1 (Advanced) Testing second d iffe re n c e s ..................................... 425
8.9.2 (Advanced) Predictions using local means and subsamples . 428
8.10 Graphing predicted p r o b a b ilitie s ................................................................ 432
8.11 Odds r a t i o s ....................................................................................................... 435
8.11.1 Listing odds ratios with lis tc o e f...................................................435
8.11.2 Plotting odds r a ti o s ......................................................................... 436
xvi Contents

8.12 (Advanced) A dditional models for nom inal o u tc o m e s ............................ 444

8.12.1 S tereotype logistic regression ......................................................... 445
8.12.2 C onditional logit model .................................................................454
8.12.3 M ultinom ial probit model w ith I I A ............................................465
8.12.4 Alternative-specific m ultinom ial p r o b i t ..................................... 469
8.12.5 R ank-ordered logit model .............................................................475
8.13 C onclusion............................................................................................................479
9 M o d e ls for c o u n t o u tc o m e s 481
9.1 The Poisson d is trib u tio n ................................................................................. 481
9.1.1 F ittin g the Poisson distribution w ith the poisson com m and 483
9.1.2 Com paring observed and predicted counts with mgen . . . . 484
9.2 The Poisson regression m o d e l ........................................................................ 487
9.2.1 E stim ation using p o isso n .................................................................488
Exam ple of the P R M ...................................................................... 489
9.2.2 F actor and percentage changes in E(y | x ) ................................490
Exam ple of factor and percentage ch ange................................. 492
9.2.3 M arginal effects on E(y | x ) ......................................................... 493
Examples of marginal e ffe c ts.........................................................495
9.2.4 Interpretation using predicted probabilities...............................496
Predicted probabilities using m table and m ch an g e.................496
Treating a count independent variable as a factor variable . 498
Predicted probabilities using m g e n ...........................................500
9.2.5 Comparing observed and predicted counts to evaluate model
sp e c ific a tio n ......................................................................................501
9.2.6 (Advanced) Exposure t i m e ............................................................. 504
9.3 The negative binomial regression m o d e l ..................................................507
9.3.1 E stim ation using n b re g .................................................................... 509
NB1 and NB2 variance f u n c t i o n s .............................................. 509
9.3.2 Exam ple of N B R M .......................................................................... 510
9.3.3 Testing for overdispersion ............................................................. 511
Contents XV11

9.3.4 Comparing th e PRM and NBRM vising estim ates table . . . 511
9.3.5 Robust standard e r r o r s .................................................................. 512
9.3.6 Interpretation using E(y | x ) ......................................................... 514
9.3.7 Interpretation using predicted pro b ab ilities.............................. 516
9.4 Models for truncated c o u n t s ........................................................................ 518
9.4.1 Estimation using tpoisson and t n b r e g ........................................ 521
Example of zero-truncated m o d e l .............................................. 521
9.4.2 Interpretation using E(y | x ) ......................................................... 523
9.4.3 Predictions in the estimation s a m p l e ......................................... 524
9.4.4 Interpretation using predicted rates and probabilities . . . . 525
9.5 (Advanced) The hurdle regression m o d e l................................................... 527
9.5.1 Fitting the hurdle m o d e l................................................................ 528
9.5.2 Predictions in the sample ............................................................. 531
9.5.3 Predictions a t user-specified v a l u e s ............................................ 533
9.5.4 Warning regarding sample specification...................................... 534
9.6 Zero-inflated count m o d e ls ............................................................................ 535
9.6.1 Estimation using zinb and z i p ...................................................... 538
9.6.2 Example of zero-inflated m o d els................................................... 539
9.6.3 Interpretation of coefficients......................................................... 540
9.6.4 Interpretation of predicted probabilities .................................. 541
Predicted probabilities with m t a b l e ........................................... 542
Plotting predicted probabilities w ith m g e n ............................. 543
9.7 Comparisons among count models ............................................................. 544
9.7.1 Comparing mean probabilities...................................................... 545
9.7.2 Tests to com pare count m o d e l s ................................................... 547
9.7.3 Using countfit to compare count m o d e l s .................................. 551
9.8 C onclusion........................................................................................................... ^58
R e fe re n c e s 561
A u t h o r in d e x 569
S u b je c t in d e x 573
Figures

2.1 T he S ta ta user in te rfa c e ............................................................................... 24

4.1 A simple linear m o d e l .................................................................................. 134

4.2 A simple nonlinear m o d e l ............................................................................ 136

5.1 Relationship between latent variable y* and P r(y = 1) for the BRM . 189
5.2 Relationship between the linear model y* = a + fix + e and the
nonlinear probability model P r (y = 1 | x) = F ( a + (3 x )...................... 191
5.3 The distinction between an outlier and an influential observation . . 210

G.l M arginal change and discrete change in the B R M ................................. 239

(i.2 Overlapping confidence intervals compared w ith discrete change . . 300

7.1 Relationship between observed y and latent y* in O R M with one

independent v a ria b le ..................................................................................... 311
7.2 Plot of predicted probabilities and cumulative probabilities for the
ordered logit model .................................................................................... 362

9.1 The Poisson probability density function (P D F ) for different rates . . 482
Preface
As with previous editions, our goal in writing this book is to make it routine to carry out
the complex calculations necessary to fully interpret regression models for categorical
outcomes. Interpreting these models is complex because the models are nonlinear.
Software packages th a t fit these models often do not provide options that make it simple
to compute th e quantities th at are useful for interpretation; when they do provide these
options, there is usually little guidance as to how to use them . In this book, we briefly
describe the statistical issues involved in interpretation and then show how you can use
Stata to make these computations.
While o u r purpose remains the same, this third edition is an almost complete rewrite
of the second edition almost every line of code in our SPost commands has been rew rit­
ten. Advances in computing and th e addition of new features to Stata has expanded
the possibilities for routinely applying more sophisticated m ethods of interpretation.
As a result, ideas we noted in previous editions as good in principle are now much
more straightforw ard to implement in practice. For example, while you could com pute
average m arginal effects using com m ands discussed in previous editions, it was difficult
and few people did so (ourselves included). Likewise, in previous editions, we relegated
methods for dealing with nonlinearities and interactions on the right-hand side of the
model to th e last chapter, and our impression was that few readers took advantage of
these ideas because they were com paratively difficult and error-prone to use.1
These lim itations changed w ith the addition of factor variables and the m argins
command in S tata 11. It took us personally quite a while to fully absorb the potential
of these powerful enhancements and decide how best to take advantage of them. Plus,
Stata 13 added several features th a t were essential for w hat we wanted to do.
This th ird edition considers the same models as the second edition of the book. We
still find these to be the most valuable models for categorical outcomes. And, as in
previous editions, our discussion is limited to models for cross-sectional data. W hile
we would like to consider models for panel data and other hierarchical data structures,
doing so would at least double the size of an already long book.

1. Those w ho have read previous editions will note that this last chapter has been dropped entirely. In
addition to covering linked variables o f the right-hand side, that chapter also discussed adapting our
com m ands to other estimation com m ands; however, this is now obsolete because m argins works
with m ost estim ation commands. We also dropped the section on working effectively in Stata
because Long’s (2009) Workflow of Data Analysis Using Stata covers these topics in detail.
XXII Preface

We note, however, t h a t m any of our SPost com m ands—such as m tab le. mgen, and
mchange (hereafter referred to as the m* com m ands) are based on m arg in s and can
b e used w ith any m odel th a t is supported by m arg in s. This is a substantial change
from our earlier p rc h a n g e . p rg en . p rta b , and p rv a lu e commands, which only worked
w ith th e models discussed in th e book. A second m ajor improvement is th a t our m*
com m ands work w ith w eights and survey estim ation, because these are supported by
m a rg in s.
SPost was originally developed using S tata 4 an d S ta ta 5. Since then, our commands
have often been enhanced to use new features in S tata. Sometimes these enhancements
have led to code th at was not as efficient, robust, or elegant as we would have liked. In
S P o stl3 . we rewrote m uch of the code, incorporated better returns, improved output,
a n d removed obscure or obsolete features.

How to cite

Our commands are n o t officially part of S ta ta. We have written them in an effort
to contribute to the com m unity of researchers whose work involves extensive use of the
models we cover. If you use our commands or other m aterials in published work, we
ask th a t you cite our work in th e same way th a t you cite other useful sources. We
ask th a t you simply cite th e book rather than providing different citations to different
commands:

Long, J. S., and J. Freese. 2014. Regression Models for Categorical Depen­
dent Variables in Stata. 3rd ed. College S tation, TX: Stata Press.

Thanks

Hundreds of people have contributed to our work since 2001. We cannot possibly
mention them all here, b u t we gratefully acknowledge them for taking the tim e to give
us their ideas. Thousands of students have taken classes using previous editions of the
books, and many have given us suggestions to make th e book or the com m ands more
effective.
In writing this third edition, several people deserve to be mentioned. Ian Anson
and Trent Mize tested com m ands and provided comments on draft chapters. Tom
VanHeuvelen ran labs in two classes that used early versions of the commands, helped
students work around bugs, m ade valuable suggestions on how to improve th e com­
mands, provided detailed comments on each chapter, was a sounding board for ideas,
and was an exceptional research assistant. Rich W illiam s gave us many suggestions that
improved the book and our commands. He has a (sometimes) valuable gift for finding
bugs as well. Scott Long gratefully acknowledges the support provided by the College
of Arts and Sciences at Indiana University.
People at StataCorp provided their expertise in many ways. More th a n this, though,
we are grateful for their engagem ent and support of our project. Jeff Pitblado was enor­
Preface XXlll

mously helpful as we incorporated factor variables and m argins into our commands. His
advice made our code far more com pact and reliable. Vince Wiggins provided valuable
advice on our graphing commands and helped us understand m argins better. Lisa
Gilmore, as always, did a great job moving the book from draft to print. Most im­
portantly, discussions with David Drukker stimulated our thinking about a new edition
and, as always, asked challenging questions th a t made our ideas better.

Illinois and Indiana Jerem y Freese

August 2014 Scott Long
Part I

General information
Our book is about using S tata to fit and interpret regression models with categorical
outcomes, w ith an emphasis 011 interpretation. The book is divided into two parts.
Part I contains general information th a t applies to all the regression models th a t are
then considered in detail in part II.

• C h a p te r 1 is a brief orienting discussion that also includes critical information

about installing a collection of S ta ta commands, known collectively as SPostl3,
th at we have w ritten to facilitate the interpretation of regression models. W ithout
these com m ands, you cannot do m any of the things we suggest in later chapters.
• C h a p te r 2 includes an introduction to Stata for those who have not used the
program or are just beginning to use it. I11 addition to this basic information on
the S ta ta interface, the chapter includes an introduction to using macros, returns,
and loops. Because these tools arc used extensively in later chapters, we encourage
more advanced Stata users to a t least skim these sections.
• C h a p te r 3 considers issues of estim ation, testing, and assessing fit th at are com­
mon to all the models considered in later chapters. We discuss both the statistical
issues involved and the S tata com m ands that carry out these operations. R eaders
already using S tata may be familiar with much of this material. Still, we think
the chapter is worth at least a quick read through, paying particular attention to
factor-variable notation, which is used extensively in later chapters.
• C h a p t e r 4 considers commands and approaches to interpretation that are used
with all the regression models in part II of the book. Most importantly, we discuss
the m arg in s command along with the m table. mgen. and mchange commands we
have w ritten th a t use m argins. You do not need to m aster this material 011 first
reading, but you should at least skim each section so th a t you can retu rn to
relevant sections as you read later chapters.

Part II encompasses chapters 5-9 and is organized by the type of outcome being m od­
eled. These chapters apply the m ethods introduced in chapters 3 and 4 using the package
of commands we show you how to install in chapter 1 .
 We m ust add some words of caution: First, we have not be able to test our commands
w ith every m argins-com patible estim ation command. We have been encouraged that
o u r com m ands have worked with other models we have used in our research. Nonethe­
less. if you use our m* com m ands with other models, you should include the d e t a i l s
op t ion so th a t you can com pare th e output from m arg in s with the sum m aries provided
by o u r com m ands.
Second, m argins is an extremely powerful com m and th at has features for applica­
tio n s th a t we have not considered, such as experim ental design. O ur philosophy in
designing our commands is to allow them to allow these options, which are passed along
to m a rg in s to do the com putation. Everything should work, but we have not been
ab le to te s t every option. While we could have designed our commands to intercept all
m a rg in s options that we have not tested, this seemed less useful than allowing you to
try them . If you use options that are not discussed in th e book, please let us know if
you encounter problems.
T h ird , just because m argins can compute a particular statistic does not mean that
it is reasonable to interpret that statistic. It can estim ate statistics th a t are valuable
an d appropriate for a given model, and it will also compute the sam e statistics for
another model for which those statistics are inappropriate. The burden of what to
com pute is left with the user. This is more broadly tru e of m argins: its great power
and flexibility also put an extra burden of responsibility on the user for making sure
results are interpreted correctly.
Another cost of the remarkable generality (if m argins is that very general routines
are used to make computations. These routines are slower sometimes much slower—
th an routines that are optimized for a specific model. As a consequence, our earlier
SPost commands (which we now refer to as SPost9), are actually much faster than the
corresponding commands in SPostl3. Interestingly, our m* commands take about as
long to run today as our earlier commands took a decade ago. While those moving from
SPost9 to SPostl3 might be put off by how much slower th e commands are, we think the
advantages are overwhelming. W ith each new release of Stata, m arg in s is faster, and
your com puter is likely to be more powerful. For now, however, in o u r own research,
we take advantage of S ta ta 13 where m argins is noticeably faster and use S tata/M P to
take advantage of multiple computing cores.
The new methods of interpretation th at are possible using m arg in s and our m*
commands sometimes require using loops and macros. They also require you to think
carefully about how you want to compute predictions to interpret your model. We
have marked some sections as Advanced to indicate th a t they involve more advanced
m ethods of interpretation or require more sophisticated Stata programming. T he box at
the beginning of each of these sections explains w hat is being done in th a t section, why it
is im portant, and when new users might want read the entire section. “Advanced” does
not m ean that the content of these sections is less valuable to some readers; indeed, we
believe these sections include some of the most im portant contributions of this edition.
Nor does it mean the commands are too difficult for substantive researchers. Rather, we
think some readers m ight benefit from finishing the other sections in a chapter before
reading the advanced sections.
 We strongly encourage you to be a t your computer so th a t you can experiment w ith
the commands as you read. Initially, we suggest you replicate w hat is done in the book
and then experim ent w ith the commands. T he spostl3_do packages (see section 1.6.1)
will download to your working directory the datasets we use and do-files to reproduce
most of the results in the book. In th e examples shown throughout th e book, we assume
the commands are being run in a working directory in which th e s p o s t 13_do package has
been installed. We have written the spex command (standing for ;‘S ta ta postestimation
examples” ), which makes it simple to use th e datasets and run the baseline estim ation
commands we use in the book. For example, the command spex l o g i t downloads th e
data and fits th e model we use as the baseline example for the binary logit model. After
you type it, you are immediately ready to explore postestim ation commands.
For each ty p e of outcome that we consider in later chapters, we rely primarily 011 a
single running example throughout. We have found this works best in teaching these
materials. However, it does make selecting examples very challenging. We wanted
examples to be interesting, accessible to diverse audiences, simple enough to follow
easily w ithout being trivial, representative of what you m ight find in other data, and
illustrative of key points. In trying to balance these sometimes conflicting goals, we
use examples th a t do not always make a compelling case for a particular method of
interpretation. For example, an effect might be small or a plot rather uninteresting.
We hope you will not decide 011 this basis that the m ethod being illustrated will be
ineffective with your data. A given m ethod might not be effective in your application,
but the nature of interpretation in nonlinear models is that you often need to try m ultiple
approaches to interpretation until you find the one that is most effective. Of course,
in some cases, the relationship you were expecting simply is not in the d ata you are
analyzing. W hile we have tried dozens of variations and approaches to interpretation
for each model, we cannot show them all. Just because we do not show a particular
approach to interpretation for a given model does not imply th a t you should not consider
that approach.

Conventions

We use several conventions throughout the book. S tata com m ands, variable names,
filenames, and o u tput are presented in a typewriter-style font l i k e th is . Italics are
used to indicate th a t something should be substituted for the word in italics. For
example, l o g i t variablelist indicates th a t the command l o g i t is to be followed by a
list of variables.
When o u tp u t from Stata is shown, the command you would type into S ta ta is
preceded by a period (which is the S ta ta prompt). For example,

. logit lfp age wc he k5, nolog

Logistic regression Number of obs 753
(output om itted )
To re p ro d u c e th e outp u t, do not type the period before the command. If following
along, only ty p e those com m ands with a dot prom pt. Commands w ithout the dot
p ro m p t are ju s t shown as examples. Also, as just illustrated, when we have deleted part
of th e o u tp u t, we indicate th is with output omitted. Keystrokes are set in this font. For
exam ple, Alt-f means th a t you are to hold down th e Alt key while pressing the f key.
T h e h ead in g s for sections th a t discuss advanced topics are indicated as Advanced.
W e refer to entries in th e S tata manuals by using the Stata convention in which
th e a b b rev iatio n for the m anual is presented in brackets and the topic in boldface (for
exam ple, [r ] lo g it). Typing h e lp m anuals in S tata will provide more inform ation about
th ese abbreviations.
T h e screenshots th at we present are from S tata 13 for Windows. If you are using
a different operating system or version, your screen might not look the sam e. See the
S ta ta C o rp publication G etting Started with Stata for your operating system for further
d etails.

Th e SPost commands

M any of the commands th a t we discuss are com m ands that we have w ritten, and
as such, they are not part of official Stata. To follow examples, you m ust install these
com m ands as described in section 1.5. Although we assume you are using S ta ta 12 or
later, most commands should work in Stata 11 (though we cannot support problems
you might encounter in S ta ta 11).
S ta ta 13 added two features that we find extrem ely valuable. First, o u tp u t for factor
variables is much clearer. Second, it is possible to use m argins to com pute average
discrete changes for variables th a t are not binary (full details are given in chapter 3).
T he s p o s t 13_ado package cannot be installed along w ith the older spost9_ado.
W hile the SPost9 commands a sp rv a lu e . c a s e 2 a lt, m isschk. m logplot, mlogview,
praccum. prchange. p rc o u n ts, prgen, p rta b , and p rv a lu e have been dropped from
SPostl3, th eir functionality remains in other commands. The pr* com m ands are re­
placed by th e m* commands, m logplot and mlogview. w ritten using S ta ta 7 graphics
and dialog boxes, have been replaced by m lo g itp lo t and th e related m changeplot com­
mands. m isschk is no longer needed because of the introduction of S ta ta ’s m is s ta b le
command. T he a sp rv alu e and c a s e 2 a lt commands have been mostly superseded by
changes S ta ta has made to fitting models with alternative-specific variables.
Still, if you have used SPost9, you might want to use some of the old commands.
W ith this in mind, we created the spost9_legacy package th a t includes com m ands that
have been dropped. The versions of the commands in this package are not exactly the
same as those in the spost9_ado package, but they have the same syntax as the earlier
commands. We cannot, however, provide technical support for this package. If you
need to run the SPost9 commands as described in the second edition—for example, to
continue work on a project using these commands you should uninstall s p o s t 13_ado
and then install spost9_ado.
Getting help

We are gratified th at many people have bought our book, b u t as a consequence, we

receive many emails with questions. W hile we try to respond to everyone who contacts
us, this is not always possible. Please read section 2.3 for information on the best way
to resolve questions or problems as quickly as possible. We appreciate it.
1 Introduction

1.1 What is this book about?

O ur book shows you effective and efficient ways to use regression models for categorical
and count outcomes. It is a book about d ata analysis and is not a formal treatment
of statistical models. To be effective in analyzing data, you w ant to spend your time
thinking about substantive issues and not laboring to get your software to generate the
results of interest. Accordingly, good d a ta analysis requires good software and good
technique.
Although we believe th a t these points apply to all data analyses, they are particularly
im portant for th e regression models th a t we examine. These models are nonlinear;
consequently, the simple interpretations th a t are possible in linear models are no longer
appropriate. In nonlinear models, the effect of each variable on the outcome depends
on the level of all variables in the model. Because of this nonlinearity, which we discuss
in detail in chapter 3, no method of interpretation can fully describe th e relationships
among the independent variables and th e outcome. Rather, a series of postestimation
explorations are needed to uncover the m ost im portant aspects of these relationships.
If you limit your interpretations to the standard output of estim ated slope coefficients,
your interpretation will usually be incomplete and sometimes even misleading.
In the linear regression model (LRM), most of the work of interpretation is com­
plete once the estim ates are obtained. You simply read off the coefficients, which can
1)0 interpreted as follows: “For a unit increase in x y is expected to increase by ¡3^
units, holding all other variables constant.” In nonlinear models, such as logit or neg­
ative binomial regression, additional com putations are necessary after the estimates
are obtained. Indeed, when interpreting nonlinear models, m ost of the work involves
sometimes complex postestim ation analyses, which are the focus of our book.
To make these computations, we use S ta ta ’s postestimation commands along with
commands that we have written. W ithout these commands, the computations are time
consuming and error-prone. All in all, it is not fun work, and it is tem pting to limit your
analyses to an uninform ative table of param eter estimates. Fortunately, the commands
we discuss in this book make sophisticated postestimation analysis routine and even
enjoyable. Although these analyses can take a lot of work, our commands reduce the
tedium so that you can focus on substantive issues.

7
 8 Chapter 1 Introduction

1.2 W hich models are considered?

R egression m odels analyze th e relationship between an explanatory variable and an
ou tco m e variable while controlling for the effects of other variables. The LRM is probably
the m o st commonly used regression model in the social sciences. A key advantage of
th e L R M is th e ease of interpreting results. Unfortunately, this model applies only
to cases in which the dependent variable is unbounded. Using the LRM when it is
not ap p ro p riate may produce coefficients that are biased and inconsistent, and there
is n o th in g advantageous ab o u t the simple interpretation of results th at are incorrect.
F ortunately, many appropriate models exist for categorical outcomes, and these models
are th e focus of our book. We cover models for four kinds of dependent variables: binary
outcom es, ordinal outcomes, nominal outcomes, and count variables.
B inary outcomes have two values, such as whether a citizen voted in the last election,
w hether a patient was cured after receiving some medical treatment, or whether a
respondent attended college. The regression models and commands we consider include
b in ary logit (lo g it) and binary probit (p ro b it).
O rdinal outcomes have m ore than two categories th a t are assumed to be ordered on
a single, underlying dimension. For example, a survey m ight ask if you would be “very
likely” , “somewhat likely”, or “not at all likely” to take a new subway to work, or if you
agree with the president on “all issues”, “most issues” , “some issues” , or “almost no
issues” . We focus on the ordered logit (o lo g it) and ordered probit (o p r o b it) models,
but we also consider the sequential logit model ( s e q lo g it) , stereotype logistic regression
( s l o g i t ) , and the generalized ordered logit (g o lo g it2 ), which is also appropriate for
nominal outcomes.
Nominal outcomes also have more than two categories, but the categories are not,
ordered. Examples include th e mode of transportation a person takes to work (for ex­
ample, bus, car, train) or an individual’s employment statu s (for example, employed,
unemployed, out of the labor force). The prim ary model we consider is the multino­
mial logit model (m logit) along with its counterpart, the multinomial probit model
with uncorrelated errors (m probit). We also review th e related conditional logit model
( c l o g i t and a s c lo g it), the alternative-specific m ultinom ial probit with correlated er­
rors (asm probit), and the rank-ordered logit model ( r o lo g it) .
Finally, count variables count the number of tim es something has happened, such as
the number of articles w ritten by a scientist or the num ber of patents a biotechnology
company has obtained. We begin with the Poisson regression model (p o isso n ), followed
by the negative binomial regression model (nbreg), the zero-truncated Poisson and
negative binomial models (tp o is s o n and tn b reg ), the hurdle regression model, and
lastly, the zero-inflated Poisson and negative binomial models (zip and z in b ).
Although this book covers many models for different types of outcomes, they are all
models for cross-sectional d ata. We do not consider models for survival or event-history
data, even though Stata has a powerful set of commands for dealing w ith these data.
We recommend Cleves et al. (2010) and the Stata Survival Analysis Reference Man-
1.4 How is the book organized? 9

ual for more inform ation on these types of models. Likewise, we do not consider any
models for panel or other multilevel data, even though S tata contains commands for fit­
ting these models. For additional information, see Rabe-Hesketh and Skrondal (2012),
Cameron and Trivedi (2010), and the Sta ta Longitudinal-Data/Panel-Data Reference
Manual.

3 Whom is this book for?

We expect th at readers of this book will vary considerably in their knowledge of both
statistics and S tata. W ith this in mind, we have tried to structure the book to accom­
m odate the diversity of our audience. Minimally, however, we assiune th at you have a
solid familiarity w ith the linear regression model and that you are comfortable using the
basic features of th e operating system of your computer. Although we have provided
sufficient information about each model so th a t you can read each chapter without prior
exposure to the models discussed, we strongly recommend that you do not use this book
as your sole source of information on the models (see section 1.8 for reading recommen­
dations). Our book will be most useful if you have already studied or are studying the
models considered herein in conjunction with reading our book.
Ideally, you are running Stata 13 or later. Most of our examples will, however, run
in Stata 11 and 12. If you are using a version of Stata earlier th a n S tata 11, we suggest
th a t you instead use the second edition of our book (Long and Freese 2006). However,
w ith the powerful new features in S tata 13 and the new m ethods of interpretation in
this third edition, we hope you decide instead to upgrade your software. To make the
m ost out of the book, you will need access to th e Internet to download our commands,
datasets, and sam ple programs (see section 1.5 for details). For information about
obtaining Stata, sec the StataCorp website at http://w w w .stata.com .

4 How is the book organized?

Chapters 2 4 introduce materials th at are essential for working with the models we
present in the later chapters:

C h a p te r 2 : I n tr o d u c tio n to S ta t a reviews the basic features of Stata that are

necessary to get new or inexperienced users up and running w ith the program.
New users should work through th e brief tutorial that we provide in section 2.18.
This introduction is by no means comprehensive, so we include information on
how to learn more about using S tata. Those who are familiar with Stata can ski])
this chapter, although even these readers might benefit from scanning it.

C h a p te r 3: E s tim a tio n , te stin g , a n d fit reviews Stata com m ands for fitting mod­
els. testing hypotheses, and com puting measures of model fit. Those who regularly
use Stata for regression modeling might be familiar with much of this material;
however, we suggest at least a quick review of the m aterial. Most importantly,
10 Chapter 1 Introduction

you should read our detailed discussion of factor-variable notation, which was in-
tro d u c e d in S ta ta 11. Understanding how to use factor variables is essential for
tin 1 m eth o d s of interpretation presented in the later chapters.

C h a p t e r 4: M e th o d s o f in te r p r e ta tio n is an overview of various approaches to

in te rp re tin g regression models. We introduce th e m argins command th a t is part
of official S tata and th e m table, mgen, and mchange commands th a t are part of
S P o stl3 . This chapter is essential background before proceeding to p a rt II. Study
th is ch a p te r carefully, even if you arc an advanced user. Readers new to S ta ta are
likely to find th at this chapter has more detail than initially needed; therefore,
th ro u g h o u t the chapter, we suggest which sections you may wish to only skim on
first reading.

P a r t II covers regression models for different types of outcomes.

C h a p t e r s 5 a n d 6: M o d e ls for b in a ry o u tc o m e s begin with an overview of how

th e binary logit and probit models are derived and how they can be fit. After
th e m odel has been fit, we show how to test hypotheses, compute residuals and
influence statistics, and calculate scalar m easures of model fit. C h ap ter 6 uses
postestim ation commands that assist in interpretation using predicted probabili­
ties, discrete and m arginal change in the predicted probabilities, and for the logit
model, odds ratios. Because binary models provide a foundation on which many
models for other kinds of outcomes are derived, and because these two chapters
provide more detailed explanations of common tasks than later chapters do, we
recommend reading these chapters carefully even if you are interested mainly in
another type of outcome.

C h a p t e r 7: M o d els for o rd in a l o u tc o m e s presents th e ordered logit and ordered

probit models. Wc show how these models are fit and how to test hypotheses
about coefficients. We also consider tests of a key assum ption of b o th models,
known as the parallel regression assumption. For interpreting results, we discuss
m ethods similar to those described in chapter G, and we also discuss interpretation
in term s of a latent dependent variable. M ethods of interpretation using predicted
probabilities apply directly to models for nominal outcomes, so it is useful to fa­
miliarize yourself with these methods before proceeding to chapter 8. T his chapter
also details the implications of assuming th at an ordinal model is appropriate for
your outcome and recommends th a t you use models for nominal outcom es as part
of your evaluation of ordinal models.

C h a p t e r 8: M odels for n o m in a l o u tc o m e s focuses oil th e multinomial logit model.

We show how to test hypotheses that involve m ultiple coefficients and discuss tests
of a key assumption known as the independence of irrelevant alternatives assump­
tion. M ethods of interpretation using predictions are identical to those for ordinal
models. Interpretation using odds ratios is a simple extension of th e methods
introduced in chapter 6, although the m ultinomial logit model’s m any parame­
ters make the process of interpretation much m ore complicated. To deal with
 1.5 The SPost, software 11

this complexity, we present a graphical method for summarizing the parameters.

The multinomial probit model w ithout correlated errors is discussed briefly, and
then the m ultinom ial logit model is used to explain the stereotype logit model.
This model, which is often used with ordinal outcomes, also has applications with
nominal outcomes. These models assum e case-specific independent variables (each
independent variable has one value for each observation). We end the chapter with
a short review of models th a t also include alternative-specific data, in which some
variables vary over the alternatives for each individual, such as an individual’s sim­
ilarity to each candidate in an election. We consider the conditional logit model
and the alternative-specific m ultinomial probit model, th e latter of which allows
correlations between alternative-specific error terms. Lastly, we present the rank-
ordered logistic regression model, which can be used when you have information
about the ranking of outcomes as opposed to information about only the selected
or most preferred outcome.

C h a p te r 9: M o d e ls fo r co u n t o u tc o m e s begins with the Poisson and negative

binomial regression models, including a test to determine which model is appro­
priate for your data. We also show how to incorporate differences in exposure
time into param eter estimation. Next, we consider interpretation for changes in
the predicted rate and changes in the predicted probability of observing a given
count. The rest of the chapter deals with models that address problems associated
with having too many zeros relative to w hat the model predicts or having no zeros
at all. We s ta rt with zero-truncated models for which zeros are missing from the
outcome variable, perhaps because of the way the d ata were collected. We then
merge a binary model and a zero-truncated model to create the hurdle model.
We also consider fitting anti interpreting zero-inflated count models, which are
designed to account for the many zero counts often found in count outcomes.

.5 The SPost software

From our point of view, one of the best things about Stata is how easy it is to add your
own commands. If S tata does not have a com m and you need or some command does not
work the way you like, you can program a new command yourself, and it will work as if it
were part of official S tata. We have created a suite of programs, referred to collectively
as SPostl3 (S tata postestimation commands for version 13), for the postestimation
interpretation of regression models. These commands must be installed before you can
try the examples in later chapters.

I f you have u s e d S P o s t b efo re, re a d th is ! For this book, we completely rewrote

our earlier SPost commands, which we will refer to as SPost9. If you have the
spost9_ado package installed on yoiu computer, you should uninstall it (details
below) before you install the s p o s t 13_ado package.
 12 Chapter 1 Introduction

T o got th e most out of th is book, yon need to try each method using both the
official S ta ta com m ands and o u r SPostl3 commands. Ideally, you are running S ta ta 13
or la te r. If you are running S ta ta 11 or Stata 12, m ost of our examples will work, but
a few valuable features new in S tata 13 will not be available. Before we discuss how
to in sta ll our commands and update your software, we have suggestions for new S tata
users, those w ith earlier versions of Stata, and those who used SPost9:

If you are new to Stata. If you have never used Stata, you m ight find the instructions in
th is section to be confusing. It might be easier if you skim the material now and return
to it afte r you have read the introduction to S tata in chapter 2.

If you are using Stata 10 or earlier. The SPostl3 comm ands used in this book will not
w ith ru n w ith Stata 10 and earlier. You can use SPost9 contained in the spost9_ado
package, which is described in the second edition of our book (Long and Freese 200G).
However, if you are investing the time to learn these m ethods, we think you are much
b e tte r off upgrading your software so that you can use spostl3_ado.

If you are using an earlier version of SPost. Before using the spostl3_ado package, you
m ust uninstall any earlier versions of SPost, such as the spostado package or the
spost9_ado package. SPostl3 replaces our earlier p rv a lu e , p rtab , prgen, prchange,
and praccum commands with the more powerful m tab le, mgen, rachange, an d m l i s t a t
com m ands based on S tata’s remarkable m argins com m and, which did not exist when
the previous edition of this book was written. If you want to use our new com m ands but
also want access to the SPost9 commands, you can install the sp o st9 _ leg acy package.
Details are given below.

1.5.1 Updating Stata

Before you install SPostl3. we strongly recommend th at you update your version of
S tata. This does not mean to upgrade to S tata 13, but rather to make sure you have
the latest updates for whatever version of S tata you are running. You should do this
even if you have just installed S tata because the D V D or download th a t you received
m ight not have the latest changes to the program. W hen you arc online and in S tata, you
can update S tata by selecting C h eck for U p d a te s from th e H elp menu; equivalently,
you can type the command u p d ate query, as we did here:
 1.5.2 Installing SP ostlS 13

Stata/MP111•(Results]
fie Edit Oki Graphic* Slalntics User Window

T his screen shows the update status and recommends an action. Our installation is up
to date. If it was not, S ta ta would recommend an update and would provide instructions
on how to do th a t.

.5.2 Installing S P o stl3

We begin with some background of w hat happens when you install user-written com­
mands in Stata. T he good news is th at once they arc installed, these commands behave
ju st like official S ta ta commands. Program s th a t add commands to S ta ta are contained
in files that end in the extension .ado, which stands for autom atic do-file. For exam­
ple, l i s t c o e f . ado is the file that contains the command l i s t c o e f . W hen you type the
command l i s t c o e f , S ta ta automatically runs l i s t c o e f .ado. T he ado-files, along with
supplementary files th at might have other suffixes, are included as p art of a package.
In Stata, a package comprises a list of included files along with instructions on how to
install them on your computer.
To install SPostl3, you install the sp o stl3 _ a d o package. We consider two methods
for installation, b u t first we explain how to uninstall the spost9_ado package.
 Chapter 1 Introduction

U n in stalling S P o s t 9

1<> d e te rm in e it you have th is package installed, type the command ado. If spost9_ado
is liste d , you ca n uninstall it, by typing

. ado uninstall spost9_ado

package spost9_ado from http://www.indiana.edu/~jslsoc/stata
Distribution-date: 25Jul2012
(package uninstalled)

Installing S P o s tl3 using search

'The s e a r c h word command searches an online database wherein StataC orp keeps track
ol u ser-w ritten additions to S ta ta. Typing search sp o stl3 _ a d o opens a Viewer window
th a t looks like this:

d Viewer - search spostl 3_ado

File Edit History Help

search spostl 3_ado /D,

search spostl3 *do X

Dialog * I Also ->ee~ Jump To ■*

.. . . 1 . _
search for spostl3_ado (manual: [R] search)

Search of official help files, FAQs, Examples, SJ3 , and STBs

Web resources from Stata and other users

(contacting http://www.stata.com)

2 packages found (Stata Journal and STB listed first)

spostl3_ado from http://www.indiana.edu/-jslsoc/stata

Distribution-date: 13May2014 / spostl3_ado | SPostl3 commands from Long
and Freese (under preparation) / Regression Models for Categorical
Outcomes using Stata, 3rd Edition. / Support
www.indiana.edu/~jslsoc/spost.htm / Scott Long ([email protected]) &

spost9_legacy from http://www.indiana.edu/-jslsoc/stata

Distribution-date: 18Feb2014 / spost9_legacy | SPost9 commands not
included in spostl3_ado. / From Long and Freese, 2014, Regression Models
for Categorical Outcomes / using Stata, 3rd Edition. / Support
www.indiana.edu/~jslsoc/spost.htm / Scott Long ([email protected]) &

(click here to return to the previous screen)

(end of search)
_ _ _ _ _ _ _ _ _ _ _ _ _____ v
Rea<ty CAP -N U M :OVRl
1.5.2 Installing SP ostl 3 15

W hen you click on s p o s tl3 _ a d o from h t t p : / / w w w .i n d i a n a . e d u / ~ j s l s o c / s t a t a ,

which is a blue link, a new Viewer window opens:

2l Viewer - net describe spost13_ado, from(http://www.indiana.edu/~jslsoc/stata) - n

File Edit History Help

J5*l fv net describe spost13_ado, from(http://www.indiana.edu/~jslsoc/s Q

net describe sp o stl 3_ado. from (— X

package spostl3_ado from http://www.indiana.edu/~jslsoc/scata

TITLE
Discribucion-dace: 13May2014

DESCRIPTI0N/ADTHOR (S)
sposcl3_ado | SPostl3 commands from Long and Freese (under preparation)
Regression Models for Categorical Outcomes using Stata, 3rd Edition.
Support w w w .indiana.edu/~j slsoc/apost.htm
Scott Long ([email protected]) & Jeremy Freese ([email protected])

INSTALLATION FILES (click here to install)

3postl3_ado/mlogitplot .ado
spostl3_ado/mlogitplot.sthlp
spostl3_ado/_orme_data.ado
spostl3_ado/_orme_data_getmatrix.ado
spostl3_ado/_orme_data_getme.ado
spostl3_ado/_orme_data_getor.ado
spo3tl3_ado/_orme_data_plotpairs.ado
spostl3_ado/_orme_data_plotvariables.ado
spost 13_ado/_orme_graph.ado
spo3tl3_ado/_orme_3yntax. ado
spostl3_ado/orplot.ado
3postl3_ado/orplot.8thlp
spostl3_ado/meplot.ado
spostl3 ado/meplot.sthlp
Ready [CAP ||NUM |QVR|

Click on the linked text th at says ( c l i c k h e r e to i n s t a l l ) . After a delay during

which files are downloaded, Stata will respond with one of the following messages:

i n s t a l l a t i o n c o m p le te means th at the package has been successfully installed and

that you can now use the commands. Ju s t above the i n s t a l l a t i o n co m p lete
message, S ta ta tells you the directory where the files were installed.

a l l f i l e s a lr e a d y e x i s t and a re up t o d a t e means th at your system already has

the latest version of the package. You can now use the commands.
 '*> Chapter 1 Introduction

t h e f o l l o w i n g f i l e s e x i s t and a r e d i f f e r e n t indicates that your system already

lias files w ith the sam e nam es as those in the package being installed and th a t the
ex istin g files differ from those in the package. T h e nam es of the differing files are
listed , a n d you are given several options. Most likely, the files listed are earlier
versions of our program s, so you should select the option Force i n s t a l l a t i o n
r e p l a c i n g a l r e a d y - i n s t a l l e d f i l e s . This is not as ominous as it sounds. Be­
cause th e files 011 our website are the latest versions, you want S ta ta to replace
y o u r current files with these new files.

c a n n o t w r i t e in d i r e c t o r y directoryname means th a t you do not have w rite privileges

to th e directory where S ta ta is trying to install the files. This usually occurs only
w hen you are using S ta ta 011 a network. We recom m end that you contact your
netw ork adm inistrator an d ask whether our com m ands can be installed using the
instructions given above. If you cannot wait for a network adm inistrator to install
th e commands or to give you the needed write access, you can install th e programs
to any directory where you do have write permission, including a flash drive or
your personal directory 011 a network. For example, suppose th at you want to
install SPostl3 to your directory called d :\u sern am e (which can be any directory
where you have write access). You should use th e following commands:

. cd d:\username
d :\username
. mkdir ado
. sysdir set PERSONAL "d:\username\ado"
. net set ado PERSONAL
. net search spost
(contacting http ://w w w .stata.com)

Then follow the installation instructions provided above for installing SPostl3.
If you get the error “could not create directory” after typing m kdir ado, you
probably do not have w rite privileges to the directory.

If you install ado-files to your own directory, th en each tim e you begin a new Stata
session you must tell S ta ta where these files are located. You do th is by typing
s y s d i r s e t PERSONAL directory?larne, where directoryname is the location o f the
ado-files. For example,

. sysdir set PERSONAL d:\username\ado

Installing SPo stl3 using net install

You can also install the s p o s t 13_ado package entirely from the Command window. If
the m ethod listed above does not work, the following steps might. While online, type

. net from http://www.indiana.edu/-jslsoc/stata/

Available packages will be listed in the Results window. You can click on s p o s t 13_ado
and follow the instructions, or you can type
 1.6.2 Using spex to load data and run examples 17

. net install spostl3_ado

n e t g e t can be used to download supplem entary files (for example, datasets and sample
do-files) from our website. For example, to download the package sp o stl3 _ d o (discussed
below), type

. net get spostl3_do

These files are downloaded to the current working directory (see chapter 2 for a full
discussion of the working directory).

1.5.3 Uninstalling S P o stl3

If you want to uninstall our commands, simply type ado u n i n s t a l l spostl3_ado. When
things do not seem to work right, our first suggestion is to uninstall spostl3_ado and
then reinstall it.

1.6 Sample do-files and datasets

Although we hope you will soon be using the methods in the book with your own
d ata, we think it is valuable to first reproduce our examples and then modify them
to try new things. To facilitate this, we have w ritten the spex com m and (included in
spostl3_ado), which makes it easy to load our d ata and run our baseline models. We
also created the sp o stl3 _ d o package, which lets you download th e d ata and do-files for
each chapter.

1.6.1 Installing the spostl3_do package

T he spostl3_do package contains the datasets used in the book along with do-files
th a t reproduce m ost of th e analyses. The do-files have names like rm 3ch9-count .do
and contain comments th a t link the com m ands to sections of the book. To down­
load these files to your working directory, while in Stata and online, type sea rch
spostl3_do. In th e Viewer window th at opens, click on the blue link spostl3_do from
h t t p : / / w w w .i n d i a n a .e d u / - js l s o c /s t a t a . A new Viewer window will open, and you
can follow the instructions to download th e files.
In the examples shown throughout th e book, we assume the com m ands are being run
from within a working directory in which the spostl3_do package has been installed.
These do-files assum e th a t you are using S ta ta 13.1. If you are using S ta ta 12, install
the sp o stl3 _ d o l3 package instead.

1.6.2 Using spex to load data and run examples

Experimenting w ith the postestimation commands that we discuss requires th at you

have first fit, the appropriate model. In our examples, we show you how to open a dataset
and fit models as you would if you were working with your own data. Accordingly, we
 18 Chapter 1 Introduction

begin w ith a u s e com m and to load the data and then use an estimation command, such
as l o g i t , to fit th e model.
To m ake it simpler for you to experiment with th e methods in later chapters, we
have w ritte n th e command sp ex (S tata postestimation examples). Typing spex com-
m a n d n a m e will produce our prim ary example for th a t estimation com m and. If you
type s p e x l o g i t , for example, S tata will automatically load the data and fit the model
th a t serves as our main logit example. Alternatively, you can specify the nam e of any
d a ta s e t th a t we use (spex datasetnam e) , and spex will load those d ata b u t not fit any
m odel. By default, spex looks for the dataset on our website. If it does not find th e
d a ta s e t there, it will look in th e current working directory and all the directories where
S ta ta searches for ado-files. For more information, type h e lp spex.
T h e running examples in this edition of the book are different from those used in
th e second edition. If you w ant to run the earlier examples, use the spex9 com m and.
For exam ple. spex9 l o g i t runs the logit example th a t was used with spost9_ado.

1.7 Getting help with SPost

Because things do not always work as intended and commands that we say will work
might not, we have some troubleshooting recommendations for you. We ask you to
please read th is section carefully to try to resolve any problems you may be experiencing
with SPostl3. If none of these suggestions fixes the issue, you can then contact us.

1.7.1 W hat if an SPost command does not work?

We assum e here that you have installed SPostl3 but th a t some of or all the com m ands
do not work. Here are some things to consider:

1. Make sure Stata is properly installed and up to date. Typing v e r i n s t will verify
th a t S ta ta has been properly installed. Typing u p d a te query will tell you whether
the version you are running is up to date and w hat you should do next. If you are
running S tata over a network, your network adm inistrator may need to do this
for you. See [u] 28 U s in g th e In te r n e t to k e e p u p to d a te and [Ft] u p d a te .
2. Make sure SPostl3 is up to date. Type ad o u p d a te , update to check, or uninstall
s p o s t 13_ado and then reinstall it.
3. If you get the error message unrecognized command, there are several possibilities.

a. If th e command used to work, consider whether you are working on a different

com puter or station in a computer lab. User-written programs must be
installed on each machine th at you use.
b. If you sent a do-file using SPostl3 commands to another person who cannot
get the commands to work, th at person should verify he or she has SPostl3
installed. Your do-file will not work with SPost9.
 1.7.2 Getting help from the authors 19

c. If you get the error message u n r e c o g n iz e d command: strangename after typ­

ing one of our commands, where strangename is not the name of the command
th at you typed, it means th at S ta ta cannot find an ancillary ado-file that the
command needs. We recommend th at you uninstall the s p o s tl3 _ a d o package
and then reinstall it.

4. If you get an error message th at you do not understand, click on the blue return
code beneath th e error message for more information about the error.
5. Often, what appears to be a problem w ith one of our commands is actually a
mistake the user has made. We know this because we make these mistakes, too.
For example, make sure that you are not using = when you should be using ==.
6 . Because our com m ands are for use after you have fit a model, they will not have
the information needed to operate properly if Stata was not successful in fitting
your model. O ur commands should tra p such errors but sometimes do not, so
make sure th ere were no problems w ith the last model fit.
7. Irregular value labels can cause com m ands to fail. Where possible, we recom­
mend using labels th a t have fewer th a n eight characters and contain no spaces
or special characters other than underscores (_). If you are having problems and
your variables do not meet this standard (especially the labels for your dependent
variable), then try changing your value labels with the l a b e l command (details
are given in section 2.15).
8 . Unusual values of the outcome categories can also cause problems. For ordinal
or nominal outcomes, some of our com m ands require th at all the outcome values
be integers between 0 and 99. The behavior of some official S tata commands can
also be confusing when unusual values are used. For these types of outcomes, we
strongly recommend using consecutive integers starting w ith 1 .

.7.2 Getting help from the authors

If you have tried everything we recommended in section 1.7.1 and you are still encoun­
tering an error, the next step is to contact us. We hear from hundreds of readers and
do our best to help. To make this easier for us, please carefully follow the suggestions
in this section.
We encourage you to sta rt by reviewing W illiam Gould’s blog entry “How to suc­
cessfully ask a question on Statalist” (2 0 1 0 ). His advice will increase your chances of
getting your question answered, either from the Statalist, from us, or elsewhere. In
addition, we have found th a t in the process of carefully preparing a question, we often
find the solution ourselves.
Here are other suggestions to make it easier for us to answer your question, which
will also increase your chances of getting a prom pt answer:
 20 Chapter 1 Introduction

1. C heck h ttp ://w w w .in d ian a.ed u /-jslso c/sp o st.h tra and
h ttp ://w w w .iiid ian a.ed u /-jslso c/sp o stJielp .litm for advice on what to try before
co n tac tin g us. There m ight be recent information th a t solves your problem.
2. M ake sure th at both S ta ta and spostl3_ado are up to date. We keep repeating
th is because it is the m ost common solution to problems our readers bring to us.
3. Look a t the sample hies in the spostl3_do package. It is sometimes easiest to
figure o u t how to use a command by seeing how others use it. Try w hat you are
doing on one of our d atasets and see if you can reproduce it.
4. If you still have a problem , send us the inform ation described in the next section.

W hat we need to help you

To solve your problem, we need to be able to reproduce it. Simply describing the
problem rarely is sufficient. Please send us the following:

1. A do-file that reproduces the problem (see a sample do-file at the end of this
section).

a. Include your name, email address, and a description of the problem.

b. Begin with the com m ands about and sp o stl3 w h ic h , which displays the ver­
sions of software you are using.
c. Include the results of summarize used in th e analysis (not all variables in the
dataset) and the results of ta b u la te for categorical variables.
d. Include only the com m ands needed to reproduce or explain th e problem.
Remove all unrelated commands.
e. Remove all references to specific directories, such as lo g u s in g
c :\d a ta \p r o je c t3 \p r o b le m , te x t or u se c :\d a t a \ p r o j e c t 3 \s a m p le .d t a .
Our computer will not. have your directory structure, so your do-file will not
run on our com puter. The do-file should read th e data from th e working
directory and save the log file to the working directory.

2. The dataset used by th e do-file. This should be a small dataset extracted from
the full dataset you arc using. Only send the variables used in your do-file and
create a dataset with only a subset of your observations (assuming, of course, that
the error is reproduced with the smaller sample).
3. A plain text log file showing the error. Do not send the log in SMCL format. To
avoid this, add the t e x t option to your log com m and, for example, lo g u sin g
m yproblem .log, t e x t re p la c e .

Send this information to [email protected] or [email protected].

 1.8 Where can I learn more about the models? 21

Here is an example of w hat a do-filc m ight look like:

capture log close

log using yourname.log, text replace
* mtable generates a variable not found error.
* scott long - jslong®indiana.edu - 2014-05-09
about
spostl3which
use jslong-error.dta
logit y xl x2
* the following command causes the error
mtable xl, at(x2=(l(l)3))
log close

.8 Where can I learn more about the models?

If you want to learn more about the regression models that are covered in this book, we
recommend the following valuable sources:

Cameron, A. C., and P. K. Trivedi. 2005. Microeconometrics: M ethods and Applica­

tions. New York: Cambridge University Press. This is an excellent introduction
to the m ethods and models discussed in this book, as well as models for panel
data.

Cameron, A. C., an d P. K. Trivedi. 2010. Microeconometrics Using Stata. Rev. ed.

College Station, TX: S tata Press. This companion to Microeconometrics: Methods
and Applications (Cameron and Trivedi 2005) shows how to use S ta ta for cross-
section and panel models.

Cameron, A. C., and P. K. Trivedi. 2013. Regression Analysis o f Count Data. 2nd
ed. Cambridge: Cambridge University Press. This is a definitive reference about
count models.

Hardin, .1. W.. and J. M. Hilbe. 2012. Generalized Linear Models and Extensions. 3rd
ed. College Station, TX: Stata Press. This is a thorough review of the generalized
linear model (or GLM) approach to modeling and includes detailed information
about using these models with Stata.

Hosmer, D. W., J r., S. Lemeshow, and R. X. Sturdivant. 2013. Applied Logistic

Regression. 3rd ed. New York: Wiley. This book, written prim arily for biostatis­
ticians and medical researchers, considers logit models for binary, ordinal, and
nominal outcomes. T he authors often discuss how their recommendations can be
executed using Stata.

Long, J. Scott. 1997. Regression M odels for Categorical and Lim ited Dependent
Variables. Thousand Oaks, CA: Sage. This book provides m ore details about the
models discussed in our book.
 Chapter 1 Introduction

l'ra in , K. 2009. Discrete Choice M ethods with Simulation. 2nd ed. New York: Cam­
b rid g e University Press. This is a thorough review of a wide range of models for
d isc re te choice and includes details on new m ethods of estimation using simulation.

W o o ld rid g e, J. M. 2010. Econom etric Analysis o f Cross Section and Panel Data. 2nd
<‘d. C am bridge, MA: M IT Press. This is a comprehensive review of e c o n o m e t r ic
m e th o d s for cross-section and panel data.
2 Introduction to Stata

This book is about fitting and interpreting regression models using Stata; to earn our
pay, we must get to these tasks quickly. W ith th a t in mind, this chapter is a relatively
concise introduction to S ta ta 13 for those with little or no familiarity with the software.
Experienced Stata users can skip this chapter, although a quick reading might be useful.
We focus on teaching the reader what is necessary to work through the examples later
in th e book and to develop good working techniques when using S ta ta for d ata analysis.
These discussions are not exhaustive; often, we show you either our favorite approach
or th e approach th a t we think is simplest. One of the great things about Stata is that
there are usually several ways to accomplish the sam e thing—so if you find a better way
than what we have shown you, use it!
You cannot learn how to use S tata simply by reading. We strongly encourage you to
try th e commands as we introduce them. We have also included a tutorial in section 2.18
th a t covers the basics of using Stata. Indeed, you might want to try the tutorial first
and then read our detailed discussions of th e commands.
T he screenshots in this chapter were created in Stata 13.1 running under Windows
using the default windowing preferences. If you have changed the defaults or are running
S ta ta under Unix or Mac OS, your screen m ight look slightly different. Those of you
new to Stata, regardless of the operating system you are using, should examine the
appropriate Getting Started manual, available in P D F format with your copy of Stata,
for further details: G etting Started with Stata for Mac, G etting Started with Stata
for Unix, or G etting Started with Stata for Windows. How to access this and other
documentation from within S tata is discussed in section 2.3.2.
For further instruction beyond what is provided in this chapter, look at the resources
listed in section 2.3. We assume th at you know how to load S tata on the computer you
are using and that you are familiar with your com puter’s operating system. By this,
we mean that you should be comfortable copying and renaming files, working with
subdirectories, closing and resizing windows, selecting options w ith menus and dialog
boxes, and so on.

2.1 The Stata interface

Figure 2.1 shows w hat S ta ta looks like after several commands have been entered and
d ata have been loaded into memory. The five m ain windows th a t you will use most
often are the Review, Results, Command. Variables, and Properties windows. Except

23
21 Chapter 2 Introduction to Stata

lor th e large Results window, each lias its name listed in its title bar. There are also
o th e r, m ore specialized windows, such as the Viewer, D ata Editor, Variables Manager,
Do-file E d ito r, Graph, and G raph Editor windows.

Past commands Results are Variable list Data properties

appears here typed here appears here appears here

F igure 2.1. The S tata user interface

T h e C o m m a n d w indow is where you type com mands th at are executed when you
press Enter. As you ty p e commands, you can edit them at any time before pressing
Enter. Pressing Page Up brings the most recently used command into the Com­
mand window, where you can edit it and then press Enter to run the modified
command.

T h e R e s u lts w indow echoes the command typed in the Command window (the com­
mands are preceded by a called the dot prom pt, as shown in figure 2.1) and
then displays the o u tp u t from th at command. W ithin the window, you can high­
light text and right-click on th a t text to see options for copying th e highlighted
text. H ie C opy T a b le option copies the selected lines to the Clipboard, and
C o p y Table as H T M L allows you to copy the selected text as an H T M L table
(see page 114 for m ore information). You also have the option to print the con­
tents of the window. Only the most recent ou tp u t is available this way; earlier
Exploring the Variety of Random
Documents with Different Content
 The man fidgeted in his chair, glanced out of the window, and
then took a long pull at his cigar.
“Bored you, didn’t it?” asked the girl. “I knew it would, but you
insisted on my telling it, and you’re the only one that knows it. I’m
really getting garrulous.”
“Do you think $5,000 would be enough to get the papers fixed
up?”
“Oh, yes, that would be quite enough, for I inquired about it. It
would take me there and back again and pay all expenses.”
“And you’d give me half?”
“Why, of course I would. Who wouldn’t?”
You know the old saying about a sucker being born every
minute. I could go on and make the usual hot finish to this story, but
what’s the use when two lines will suffice. She got the money, of
course, and he got what is known in the language of The Line as the
lemon. Very sour it was for this hard, wise fellow, and they say that
now every time he passes a manicure parlor he turns his head the
other way and says things which wouldn’t look well in print.
There were times when she did things that were
unconventional
 INVESTING IN A HUSBAND
Money makes the mare go.
Sure.
That is, sometimes, if it’s the right kind of a mare and there is
enough money.
Take out all the “ifs” and “buts” and it will be all right.
The world began with a man, Adam, and the woman came later,
but the finish will be different, for there will be a woman in the last
ditch giving or ready to give the avenging angel the stiffest kind of
an argument.
This story differs from the Creation in that it begins with a
woman, as all stories of to-day should. And why not? for take the
lady out of the case and there’s no story and never will be. The slim
finger of a woman, you know, is in every pie. Sometimes it improves
the flavor and sometimes it spoils it—that’s a matter of luck—and
there are men who have tried pies or many fingers, whichever simile
you prefer, and the result in their cases is always the same.
The girl in this story had birth, and blood, and breeding behind
her. She also had good looks and a little money, and that is about all
that anyone wants. Add to that a fairly nice disposition and you have
reached the limit.
Of course, she wasn’t perfect by any means. She was a bit
whimsical and peculiar, and her moods were as apparent as the
moving pictures thrown on a sheet in the theatre. She was unusual
in that her moods were reflected in her face with all the truthfulness
of a mirror. That was the reason that some said she was good-
looking, while others contended that she was most ordinary. Take
her as I’ve often seen her, when she was cheerful and happy-go-
lucky, and while there was nothing about her features that was
regular she was attractive enough for anyone, and she could make a
good many young fellows turn their heads to look after her as she
passed down the street.
Then again something would happen, and she would seem to
age ten years in as many hours, and a crop of deep lines and
wrinkles would spring out like magic. But she had magnetism, and
she was forever standing at the fork of two roads, one of which led
to good and the other to bad. To her it was the toss of a coin which
one she would take.
It was while she was in a thoughtful mood, debating with
herself, that the man came along. There’s an apology goes with that,
for he hadn’t a vote yet, and he was very youthful in his ways and of
that age where a youngster is apt to tell more than is good for him,
and to stray from the field of fact. Of course, it’s not a crime—it’s
only a period. With his red cheeks and baby complexion he looked
like a cross between a stick of peppermint candy and one of
Raphael’s cherubs. He was as pretty a piece of embroidery as ever
asked his mother for spending money, and when the girl saw him
she immediately threw out a line and took him in tow. Inside of
twenty-four hours she had her monogram indelibly stamped on him,
and he was hers. Hand in hand they went out to see the world and
become real sports, and it wasn’t long before wine was the limit and
it wasn’t half good enough at that. They left a lurid streak up and
down the line, but it soon faded out, for they weren’t financially
strong enough to make a splash that would attract any more
attention than a pair of tiny gold fish in a two-dollar aquarium.
After all, it amounts to nothing more or less than a question of
capacity—stomach as well as purse, and it is rarely that the two
harmonize. The man with the yard-wide thirst is often handicapped
by a purse with complete or partial paralysis.
 And then these two fell in with other company in the shape of a
man and woman whose nuptials had been attended by incidents of a
more or less exciting character, the star part of which was an
elopement which savored more of desire than genius in its
arrangements. They had succeeded so well in their new venture that
they owned the entire contents of a flat across the river in Jersey,
and being still in the throes of love themselves—or thinking they
were—they were headquarters for everything that seemed like an
affair of the heart. Some who were not their friends were unkind
enough to say that it was nothing more nor less than a case of
misery loving company, and that being on the coals themselves this
couple enjoyed leading others to the broiler. But that’s unkind and
really ought not to be believed.
However, many a racket came off in the flat, and they all went
as hot a pace as wind and weather permitted, until even a rank
outsider would have said it was time for a minister to get on the job
and do what he could to make things legal.
The cork popped from a bottle of wine and the juice of the grape
sizzled out.
“What do you say, Kid, let’s get married?”
“All right, I’m game if you are; you can’t phaze me,” she said.
“Well, how about to-night?”
“The sooner the better.”
Talk about quick action, it was here with a vengeance.
Four people on a ferryboat, then an elevated railroad and the
ringing of a minister’s door bell.
It’s all very simple.
The dinner afterward in a cafe, very informal, you know, to
harmonize with the ceremony, with a couple of quarts for luck
sandwiched in by cocktails and highballs; then a few brief telegrams:
“Married to-night; wish us luck;” you know the rest.
 It was all right, after all, apparently, and everybody did wish
them luck, even if there were a few bad spots in the job. But, you
see, they suited themselves and there was no one else to be taken
into consideration, not even the relatives. This going around and
holding consultations in advance is no good, and people who are in
love or who think they are in love don’t want advice of any kind,
except the kind that rings the door bell of a minister’s hut or buys a
wedding ring and sends it with the words:
“Get busy before it is too late.”
I’m no critic, and I don’t pretend to criticise here. I’m simply
telling a story which may or may not be true, but I’m not going to be
responsible for it any more than the man who rents a place and
plants flowers in the garden is responsible for the architecture of the
house on the premises.
It is said that the bride in this case was kind enough to supply
the funds for the honeymoon, while the nice boy supplied the beauty
and called it even. In the eyes of the lady it seems a fair enough
proposition, but harsh things are liable to be said of such a
combination, even though it is no one’s business.
When they returned from the fields of fruits and flowers the boy
had made up his mind, like the Count Boni de Castellane, that being
a husband was much better than holding down a job in an office,
and so they settled in New York like a pair of pigeons after a long
flight. He had no more idea of the responsibilities of married life
than a six-months’-old infant has of playing the races. With a place
to sleep and a feed bag always ready for his face he was satisfied,
but that was because of his youth. You see, marrying from the
cradle has both its advantages and its drawbacks, according to the
way you look at it.
For him every morning was Christmas, and the tree was always
fixed up with something nice with his name on it. Do you blame him
for looking pleasant? Press the button for a dollar, press it twice and
you get five. Just as easy as drawing money out of the bank when
you have a check book.
But with all going out and nothing coming in it doesn’t last long,
and when he had swept up all the spare change in sight he began to
cast his covetous eye upon the big bundle that was tied up with a
woolen string.
He knew something about the racing game—just enough to get
stung when the time came—and he knew a man who was good
enough to offer him a half interest in a racing mare that had been
kept under cover for a year or so, but who could, if she was let out,
beat anything that ever wore pigskin. To that infantile mind of his
this was the one great chance of a lifetime and the thousand-dollar
bill was the key which would unlock the door to wealth.
Money without working for it.
Why it was a pipe. Besides, it made a beautiful and alluring tale
for the bride, who had reached that stage where she didn’t want her
boy away from her, not even for a minute. With the thousand he
would make the initial investment, and with the rest of the bank roll
he would bet. With paper and pencils they sat at the table one night
and rolled up two thousand to the fortune of a Rockefeller.
How easy it is to make money that way. All you have to do is to
begin with any amount, even a penny, and if your pencil holds out
you’ll have a million in less than no time, but you can’t buy anything
with it—there’s the trouble. The man in the insane asylum who
imagined that every stone in the construction of the building was of
pure gold and that it belonged to him was just as rich in his own
mind as the wealthiest human being in the world—and happier, too,
I’ll bet you.
They planned it all out, even to the trip to Europe on the
winnings of the first big race, for she would carry odds of not less
than 20 to 1, because she was unknown.
A little trip down to the bank and out came the money in brand
new bills that were very good to look at.
 So the first step was taken, and the boy made up his mind that
he had turned his back forever upon such things as ten-dollar-a-
week jobs.
It doesn’t require any ingenuity or brains for a man to separate
himself from such things as thousand-dollar bills—in fact it’s quite
easy. Consequently it didn’t require any brain work on the part of the
boy to deplete the account by just that amount within a very short
time. For his new bill he received in return a slip of paper which
stated that he was the half owner of the racing mare known as Blue
Monday, and that in consideration of his paying one-half of the
training expenses of the said mare he was to be entitled to one-half
of the winnings, less jockey fees and other incidentals.
To him it sounded beautiful and it took not less than one quart
to celebrate this new business venture—paid for by the lady, of
course, but still, in view of the fact that they were one, it was all
right.
Then there began to come to him via the U. S. Mail, certain
sundry statements concerning the expenses of putting this fine bit of
horse flesh into the proper condition to bring home the money, and
the request for immediate remittance. There was variety enough
about these statements, too, to satisfy the most fastidious, and the
amounts ranged all the way from six dollars and fifty cents to an
even hundred. The clever mind of the bride took in the situation at a
glance, but the faith of the optimistic kid held as fast as a ship’s
anchor to a rock ledge, and he could see nothing but success in the
near future.
You know there is never a day so far away that it doesn’t come
at last. So it was that the day of the long expected race arrived and
down deep in the trousers pockets of the Pink Cheeked One was
$150, the last shot in the locker.
“It’s all right, Kid,” he said to her. “It’s just as I thought, she’s a
twenty-five to one shot, and I’m going to plank every cent down. At
those odds we’ll take home with us $3,750, and I guess that’ll hold
us for awhile. How about it?”
“But suppose she doesn’t win?”
“Doesn’t win? What’s the matter with you—are you getting cold
feet? How can she lose? Didn’t we clock her this morning on the try-
out and didn’t she beat the track time? Wait till you know more
about this game and you’ll see where I’m right.”
I don’t know much more about it than that, but the files of
papers of that date show me that Blue Monday, mare, 3-year-old,
was entered for the Seaside stakes of $1,500, at odds of 25 to 1;
there was a good start, with her in the lead. At the quarter she had
fallen back to fourth, at the half she had crept up until she lapped
the second horse.
She finished seventh.
I should say that blue-eyed boy was looking for a job the next
day, but I’m not fortune teller enough to know whether he
connected or not.
 TRAINING AN OLD SPORT
Come and listen to the siren song of the New York girl, and
perhaps it may interest you for awhile. There is no question about it
unless you are a bronze statue standing on a gray stone pedestal in
some park, or a cigar store Indian with an Hebraic nose and a
wooden tomahawk. In the first place the New York girl has been
conceded to be a wonder and about the best in the world in looks as
well as in figure. She has a fine complexion when she gives it a
chance to show itself, and, like the little girl in the story book, when
she’s good she’s very, very good, and when she’s bad she’s a peach.
The thing is to pick out the right one, and your chances for that are
just as good as drawing to a pair in poker. Some say it’s luck, while
others favor the science idea.
With that for an overture, let’s ring the bell for the curtain to go
up on the charming little two-act play, entitled “The Redemption of a
Sport.”
The Old Sport has been up against every proposition the sun
ever shone on, and there was nothing he wasn’t fly to. He had paid
board for blondes and brunettes as well as a few Leslie Carters, to
say nothing of an Albino he once took a fancy to. He was an early
and late bird, and he was known up and down the line by his first
name, which is a distinction that it usually takes a lot of money or a
number of years, and sometimes both, to acquire, and even then it’s
not a lead pipe cinch that you’ll land it right.
 A light flashed out on the landing and revealed the figure
of a beautiful woman

This fellow was good to the girls, and could be relied on for a
five-case note on a hurry touch at any time, for he had no buttons
on his pockets, and he knew that safe deposit vaults in heaven are
only used for the storing of golden crowns in hot weather.
 “If I can’t take my money with me,” he said once, “then I’ll
spend it here, for if there’s anything in the world that I hate it is to
think that there’s going to be a lot of hungry relatives picking over
the bones of my estate before I get comfortably settled in the six
feet of real estate that no one can beat me out of. The money’s got
to be spent some time, and I’m going to be the one to get the credit
for it because it’s mine.”
But there came a time in his life when he felt that he wanted to
get away from the mob. He had been stung by the bee of
domesticity and didn’t know it. What he did know was that he
wanted a place with a real woman in it, where he could hang his hat
and that he could call his own. If he had wanted to put his brains at
work he would have known that it was nothing more nor less than
the law of nature which had him fast—that same law which makes a
bird build a nest in a tree, or a wild animal pre-empt a bed of moss
under the roots of a certain tree.
It was the home instinct.
So he began to cast his eye around for a side partner whom he
could have and hold, even if he had to coax her up to the altar with
a marriage license printed in red and gold and lasso her with a
wedding ring. From that time on he was always on the alert for the
right one to come along, and every time he heard a sound like a
skirt he made an investigation. In about ten days he turned down all
the Dollies and Mauds of the Line, for he couldn’t see where they
would have a look-in if the cook happened to leave in a hurry and he
arrived home with a backwoods appetite. You see he wanted a gas-
stove performer who could in an emergency tell the difference
between a roast and a ragout in the raw state, and who could juggle
with a lot of cold grub in the ice box, and turn out a square meal
that was not only hot but nourishing. He was tired of restaurant
hash, anyhow, and he was longing for the kind of biscuits that
mother used to make.
He figured for awhile on a girl named Elsie, who could make a
cocktail to beat the band, and who could also drink more and get
away with it than any of the rest. She was a good looker, too, and
she had trotted in double harness before, but he found out that she
was a bit promiscuous in her tastes, and he didn’t care to feel that
he had to stay at home all the time in order to keep her from
entertaining any stranger in a pair of trousers who happened along.
So he put a red cross, which means “Danger, Keep Off,” opposite her
name, and began looking in another direction.
He changed his tactics completely.
“I’m on now,” he said to himself. “I’ll hunt up some nice little
innocent girl who doesn’t know anything of the world, and who has
taken a course in a cooking school. I want the kind whose ambition
in life is to be boss of a nice three-story house, and who doesn’t
care any more for Broadway than a hobo does for a hot bath. I’ll just
hunt up some mother’s girl who has her hair hanging down her back
in a big, thick braid, and I’ll sing her a song that’ll make her think
I’m the real thing on wheels.”
So with that very laudable and commendable idea he started
out. He didn’t figure that a tough old nut like he was had any right
to go up against a game like that, and that his play was to mix with
people of his own class. But you’ll find in nine cases out of ten that
the worse a man is or has been the more innocence and purity he
wants when he is figuring on giving a sky pilot a chance to make a
dollar or two.
But having made up his mind the kind of a field he was going to
hunt, the next question was how to break in. All the girls he knew
were, without exception, of the brand which are at their best when
the lights are turned on, who rent flats for business purposes, and
who change quarters when an intimation is made by the captain of a
police precinct that the change will do them good. To save his life he
couldn’t figure out this new proposition, and he was like the man
who bought a new double-barreled shotgun and then found out he
couldn’t get a permit to hunt the birds the old farmer owned.
 And now right here, at the critical moment, in steps fate, luck, or
destiny, it doesn’t matter which, for they are all the same, and
shuffles the cards for a new deal.
An automobile on Broadway bumped hard enough into the rear
end of a hansom cab to almost throw the driver from his seat and to
make him swear a blue streak of profane eloquence. The usual
crowd collected, and in the bunch caught there by the sudden rush
of curious and morbid humanity was the Old Sport. He pushed with
both elbows to free himself and then stepped back testily. A girl
behind him cried out with pain, and he turned suddenly around to
find himself face to face with as choice a little blonde as ever carried
books home from school, and, furthermore, she had a braid down
her back.
“I beg your pardon, did I hurt you?” he asked.
“I’m afraid you did; you stepped on my foot.”
“Well, just take my arm and let me help you out of this crowd.”
Easy if you only know how and the chance comes your way.
The Old Sport wasn’t really old—not over forty—and he was
there with the looks, and the little lady rather liked the way he
framed up, as anyone could see by the way she cuddled up to him
as she limped along. His heart was beating it like a yeggman coming
East on a brake beam, and already he was figuring on how to handle
this new proposition.
If it had been one of those other girls he would have said:
“You just send your trunk up to my place, and we’ll go around
and have a talk to a minister; how about it?”
But he couldn’t say that to this girl with the pink in her cheeks
and the fluffy hair that had never been up against the peroxide.
“Foot pretty bad, Kid?” was the way he broke the ice.
“Oh, no, thank you, it’s all right now, but it hurt me a lot at first.”
 “Live far from here?” he came back again.
“No, not very far; only Fifty-third street.”
There was only ten blocks to go, and when they got to the last
one he knew all about her. He knew that she was living with her
aunt, and that she was taking music lessons because some day she
hoped to be able to teach. As they paused for a moment on the
corner, he said:
“If you should happen along on Forty-second street to-morrow
about 2, I’ll be glad to see you.”
It was a bit crude, but it went all right and the date was made.
When she walked away he stood looking after her, and he noticed
that she had a nice trim figure, a dainty little foot and that she
stepped out like a thoroughbred.
“You for me,” he remarked, and then he hustled back to find
some one he could treat, so great was his joy.
So there’s the picture, to use a theatrical term, and the curtain
goes down on it for the end of the first act.
Now, you and I and some of the rest of the thirsty crowd will go
out and have a drink between acts, but it’s a warm night and instead
of one drink there’s half a dozen. Time flies when you’re in good
company and the Old Sport was taking no chances. Ten interviews
with the girl—ten good, square, honest talks at the rate of a talk a
day—and she consented to take a chance with him and tell the folks
afterward. He was on the level, though, and when she went home a
couple of days later she had the little certificate with her, and after a
few tears Auntie was invited around to visit her new nephew and
look over the new house.
As for the Sport, he settled down as comfortably as an old buff
Cochin-China hen on a dozen eggs, and he made up his mind that
he had been missing a good many years of real dyed-in-the-wool
happiness while he was traveling The Line with the bunch and
throwing all kinds of booze under his belt.
 But when the weeks began to add themselves into months he
grew a bit restless of nights and it came pretty hard when any of the
boys asked him to come along and help them crack a bottle. He took
the Mrs. to the show once in a while, but it was always a case of
hurry home as soon as the orchestra began to play “My Country, ’Tis
of Thee.” He didn’t want to take a chance of being caught by any of
the Merry-Merrys who were out for the rent and guyed for “marrying
decent.” Once or twice he thought he had made a mistake and that
the change was too great or too sudden for him, but an hour later
when he had his slippers on and was planted in the big armchair in
the corner, he knew he wouldn’t make any kind of a change for the
world, and he felt that he had lost a good many years out of his life
in not getting into this kind of a game sooner. Like an old fire horse,
he was all right as long as he didn’t smell fire. But the time was
coming, and it was as sure as rent, taxes or death.
It came when he went out one night to be gone not more than a
half hour, and when he tried his key in the lock it was 2 A. M., and
the girl, her eyes red from crying with the desertion and the
loneliness of it all, had fallen asleep, fully dressed, across the foot of
the bed. He was very sorry and penitent, but for all that he went out
the next night just the same, and after that he was never in. He was
back on the old trail, mixing once more, to the great delight of the
crowd. The novelty of home had worn off, and when his wife waited
up for him she usually found him too drunk to understand what she
was saying to him. From one step it is easy to take another, or, as
the Chinese say, the creeper always walks in the end. He took to
bringing friends home with him at all hours, especially between
three and six in the morning, and their arrival was always made
apparent by the wild time they had scrambling up the stairs.
Now, in this story—as in real life—always keep your eye on the
lady. It doesn’t make any difference where she comes from, whether
it’s New York City or Lower Squankum, New Jersey, she is either one
of two things, very clever or very dull. There is no medium, for what
may seem to you like a medium is only a counterfeit and not the real
article. For every ninety-nine dull women there is one clever woman;
for every ninety-nine clever women there is one ace who tops the
rest as easily as Mont Blanc tops an ant hill. The wife in this case
was not one of the dullards, that’s a cinch. If she had been she
would have made an idiot of herself and acted the way the rest of
them do—which is a great nuisance and annoying to any man. She
was a genius, and I ask you to take off your hat to her—as I do.
“I notice,” she remarked to Old Sport one morning, “that you
never bring more than one friend home with you when you arrive.
Why don’t you bring half a dozen, or three, anyhow? It would be
much more companionable.”
He was a bit on his guard at first, but she convinced him that
she was serious about it, and then he began to congratulate himself
that he had his wife well in hand.
Two nights later he arrived with half a dozen of the hottest
hooters that ever held an all-night session in a furnished flat. He let
them in with his key, and as they paused at the foot of the stairs, a
clock from somewhere chimed out a silvery “three.”
“Come on, boys; open house here; everything goes,” said Old
Sport. “My wife says my friends are good enough for her if they’re
good enough for me. Come on.”
He, with another, made the start up the stairs, but they hadn’t
gone more than a few steps when a brilliant light from the landing
somewhere fairly dazzled them.
Directly in front of them, apparently in the act of stepping out of
a huge picture frame, was the symmetrical figure of an almost nude
woman. The light struck her just right and brought out every detail.
“Great,” shouted someone from the foot of the stairs.
“Shut up, you fool, it’s my wife,” answered the Sport. “Put out
that light up there, do you hear? Put it out.”
But it blazed away as steadily as ever, and there was no
movement on the part of the figure, except that the full bosom rose
and fell with the regularity of her breathing.
 The Sport turned around on the stairs.
“Come out of here, you fellows; this is going too far. Come on,
skiddoo, all of you.”
And when the last one had gone out he slammed the door
behind them. What happened inside is none of your business, nor
mine, either, because I don’t believe in scandal, but any evening the
Old Sport is wanted he will be found at his home address with his
wife and a kid who looks like him.
As for the lady; she has a genius that she is just beginning to
appreciate.
 CONCERNING A SYRIAN BEAUTY
Transplant the Oriental to the Occident, or in plain words bring a
nice-looking girl from the East to New York, for instance, and nine
times out of ten there is sure to be something doing. Most of the
doings, to be sure, are under the rose, but every once in a while
some hint bobs to the surface and the news is wafted about by
every breeze of a whisper.
In his very handsomely appointed suite of apartments on the
upper West Side is a young fellow who has good enough blood in his
veins to be game and take his medicine, and with sense enough to
keep his mouth shut. Across the bridge of his nose are three knife
cuts made by a blade that was very keen, which was held by a hand
that knew its business. His doctor tells him that it is not at all
serious, even though inconvenient—you know how doctors talk when
there is a good fat fee at the other end of the line. He also says that
there is nothing in the world that will prevent and eradicate those
three disfiguring scars, even after the wound has been thoroughly
healed and every possible surgical precaution taken.
And there’s the rub.
Through all the rest of his life this man, upon whom the world
has been smiling since his birth, will be marked with the signs of his
folly.
So much for the present.
Now for the recent past.
 Put her in tights and she would have been an Oriental
sensation

The woman was a Syrian beauty with sloe eyes and an olive skin
that was like a piece of copper-hued satin, so soft and smooth and
free from blemish was it. There was a faint flush of red in her
cheeks, too, as if the hot blood was trying to break through the
tender skin. Her lips were red and full, and because of all that riot of
color her teeth showed whiter than they really were. She had,
besides, small feet and slim, trim ankles.
Any wise man will appreciate that and understand why they are
brought into this story. Up to the age of twenty-five the male animal
looks at the female face and is satisfied. After that no such casual
scrutiny satisfies him. First face, hair and general contour, then
ankles, and often it is the last view which does the work or turns the
trick, which is the same thing, only it is expressed differently. This is
with the assumption, of course, that the man has enough
discrimination to want quality, not quantity. Quantity is unwieldy and
unsatisfactory from every viewpoint except from that of the
gentleman who is in the butcher business, and who wants a
standing advertisement for his shop. Embonpoint is all right in
sausages but not in women, excepting—and that is understood—
those on dime museum platforms.
The first name of the lady was Dekka, the rest was
unpronounceable and we’ll let it go at that. She was a seller of
Oriental goods, not from a Tenderloin standpoint, but real
merchandise such as is recognized by the law—laces, draperies, bits
of cunningly embroidered silks, and even rugs, which she called
carpets, with the accent on the first syllable. Her stock was carried in
a dress suit case which was handled by her “brother,” who was also
a Syrian, and he only resembled her because he, too, had black
eyes, an olive skin and dark crispy hair, to say nothing of his small
feet.
Day after day they went in and out of houses, flats and
apartments, visiting none but the best, and calling an express wagon
into service when a rug display was necessary. She was the brains of
the combination and did all the selling. His job was done when he
put the satchel down by her side. Then he effaced himself and was
invisible until she was ready to exit, when he made a mysterious
reappearance from somewhere.
And that’s the soup of the story; the roast follows.
 The Jap valet to the young man of means and leisure announced
to him one afternoon that a dark lady—makes you think of the
queen of spades, doesn’t it?—wanted to see him and wouldn’t take
no for an answer.
“Bring her in,” said Jimmy, who was feeling in just the right kind
of a humor to see anyone, even a man to whom he owed money,
and in a moment she had slipped into the room as lightly as a cat
walking on wet grass. There was the sound of her French heels
hitting the bare spots on the polished floor that was music to him,
and he wondered what there was in the meeting of leather and
wood that was so attractive and just a bit different from anything he
had ever heard before.
She courtesied in a friendly, intimate sort of a way, and then
spoke:
“Good day; the lady? Can I show her some laces? Very fine.”
There was just the faintest touch of an accent in her voice, but it
was rather pleasant than otherwise, and it seemed to have a very
soothing effect on him.
“There is no lady here,” he laughed, “that is, not yet.”
“Ah, too bad, and such a nice place, too. It is so beautiful.”
She half turned as if to go, and he stepped toward her.
“What have you got to sell? I might buy something.”
“You are so kind; I have them here,” and she motioned to the
next room. “My brother bring them, then he go ’way. It is very heavy
to carry all the time.”
“Yama,” called he, “bring it in, whatever it is,” and in a moment
the Jap came lugging the leather case.
Jimmy noted how deftly the shapely brown fingers unfastened
the brass catches, and as she leaned over he found himself studying
her with the eye of a man who has seen and known a great many
women of all kinds and all nationalities with one or two exceptions,
and one of the exceptions was Syrian. A faint perfume, the odor of
which he failed to recognize, seemed to fill the room, and he knew it
came from her, and he became suddenly aware that he was taking
more interest in the saleswoman than he was in the goods she was
about to offer him.
When the bag had been opened and the contents tumbled out
promiscuously, without any attempt at order or display, she sat down
on the rug beside them. She picked out a lace scarf and carefully
smoothing out its folds held it before him.
“Very fine,” she said; “all made by hand, see?” and she pointed
to the heavy embroidery.
“It’s all right,” he answered, but he wasn’t looking at the silk, he
was looking straight in her eyes and wondering why it was he had
never met a woman with eyes as black as those before.
“You are not looking,” she said.
“I am,” he replied.
“At the scarf, I mean.”
“No, there is something better.”
“But I am only selling the scarf to you,” and she began to fold it
up while her cheeks became more red.
“What’s the price?” asked Jimmy.
“Only $6, and very cheap.”
“All right, I’ll take it; let me see what else you’ve got there.”
And presently they were both sitting on the rug, he on one side
of the bag and she on the other. In a half hour he had spent one
hundred dollars, but to save his life he couldn’t have told what it was
he had bought and, what was more, he didn’t care.
He laid the crisp new bill on her knee, and as she began to fold
up the remnant of her stock he asked questions.
 “You said your brother went around with you. Is he really your
brother or something else?”
“My own brother; why should I tell you a lie?”
“I don’t know except that there are a great many brothers and
cousins in this world who are not brothers or cousins at all, except
as a matter of convenience. You know, I think you are a nice little
girl and I fancy I’m getting just a bit gone on you. I don’t mind
buying things from you, but I should like it if you and I could be
friends.”
By this time they were standing up; the suit case had been
closed and it was still between them, as if it was a sort of a
guardian.
“Couldn’t you stay here and have a little lunch with me? We’ll
have it right away and you’ll be away in an hour. Where’s your
brother?”
“Oh, he always waits somewhere—outside, maybe.”
“In the other room?”
“Oh, no; sometimes in the hall and sometimes in the street;
sometimes he goes away and comes back again.”
“Well, this time he can wait a little longer. Yama,” calling to the
Jap, “get some lunch and hurry up.”
He picked up the barrier of a dress suit case and put it one side,
then he walked over to her and putting his arm around her waist,
pulled her toward him and kissed her squarely on the mouth.
“Oh,” she cried, “what are you doing?”
“Kissing you. I’ve bought your silks and now I’m ready to invest
in kisses, and I find,” he remarked, as he kissed her again, “that
your kisses are the best.”
The blood leaped to his brain, and he held her so tightly that it
seemed as if he would crush her.
 “You’ve made me fall in love with you,” he said, and that strange
Oriental perfume which came to him from her seemed to make him
mad. “I want you to go away with me; will you? We’ll go wherever
you like, and you will not have to sell those things any more. You
can have all the money to spend that you want and you will be a
lady.”
Here was a picture strong enough to turn the head of any
woman, much less a Syrian straight from peasant stock, brought into
the world by accident, with a face like a Madonna and with a supple,
pliant figure that made men turn around and look after her. A girl
who had known what privation and hardship was, and who came of
a race where women were born to be servants and made to wait on
men, the masters. Her beauty had brought her nothing and now it
had suddenly become an asset, a stock in trade of so great value
that for the rest of her life she would know neither work, nor care,
nor trouble. The blood rushing through her veins made her dizzy and
her head fell forward as her eyes half closed. One brown arm crept
up and around the neck of this strong, broad-shouldered American,
and it kept her from falling to the floor in the excess of her emotion.
He felt her going, and picking her up, carried her to the big armchair
over in the corner, where she cuddled up like a rabbit. She was
clasping and unclasping her fingers nervously as he stood looking at
her and her half-closed eyes never once met his.
“What’s the matter?” he asked, bending over. “Can I do anything
for you?”
“No,” she whispered; “I was only thinking of my brother.”
“You don’t want to mind him; he’s all right wherever he is.”
“Not that, but he might not want—he might not like you to—to
love me,” and she looked up at him.
“We’ll take care of your brother all right. Because he is your
brother I will do what I can for him. Why, I will——”
The voice of the Jap came from the other room just as Jimmy
was settling himself on the edge of the big chair, and had his arm
around the Syrian’s neck.
“No,” it said, “you wait; I see.”
There was an angry voice raised in expostulation, and then
before the man could move the brother came bounding through the
parted curtains. He paused for just one brief moment and then
shrieked:
“Dekka.” He said something else, too, but it was in his own
language and only the woman understood, but whatever it was it
made her shrink still lower in her seat and cover her face with her
hands. He was on Jimmy like a cat, and three times, even though
the frightened Jap was trying to pull him off, he cut, and each cut
was across the bridge of the nose, and the knife blade went as true
and sure to the mark as though it was in the hands of a surgeon on
a patient who was under ether. Then with one firm grip on the wrist
of the girl he dragged her to the door and out, while the faithful
Yama was using the silk scarfs—the ones which had just been
bought—trying to staunch the flow of blood.
And that’s the story.
And the moral of it is that every man should stick to his own
race and his own blood, Caucasian to Caucasian and Oriental to
Oriental, for there are some things in this world that don’t mix any
more than oil and water.
The first pair are in the ring, the talk ceases, and the
show is on
 THE REJUVENATION OF PATSY
We’ll just take in a fight to-night for a change. I’ve had you
Down the Line, over on the East Side, in the wine joints, behind the
scenes, and in half a dozen of the so-called swell restaurants, and all
the time there have been all kinds of punching matches going on in
a dozen different halls, “Clubs,” they are called, just to sidestep the
stern arm of the law, but what difference does it make to a good
sport so long as the men are well matched and they are willing to
mix it at all times?
Three rounds are the limit, but there is a lot doing between bell
and bell—enough to make even the most seasoned ringster sit up
and look around as if to say:
“Now here is some punching that does a man’s heart good—it
seems like old times, when——.” You know the rest about the days
of long ago, and if you listen to him he will hand you a line of talk
that will put you away for the count.
You may talk as you like about all the sports you know, but after
all there is nothing like a good go with the gloves between a pair
who know their business, and there are few men who have any red
blood in their veins who will not go a long ways to see a slugfest. Of
course you’ll always find up against some bar a bunch of dead ones
who will stretch their arms and say:
“Not for mine; I’ve seen all I want to see, and I wouldn’t go
around the corner to get a ringside seat at a go between Roosevelt
and Kaiser Wilhelm.”
 There’s a screw loose somewhere in these fellows, or else they
are drying of dry rot and don’t know it. Nine out of ten of them are
bigger around the waist than they are around the chest, and they
invariably talk loud.
There’s a little club that I know of where you can get a great run
for your money, and we will go there.
It’s a case of come early and avoid the rush, for when the gong
rings for the first bout there is only standing room left and that is at
a premium because the prices are low. The manager doesn’t have to
bother his head about making matches because the “talent” comes
to him, and it often happens that the men who furnish the
preliminaries are picked from out of the audience. These three-round
affairs have done a lot to bring out a bunch of new ones; any young
fellow who knows any part of the game can go on and get a try-out.
He earns a few dollars and if he proves to be good, he is boosted
along the line.
There is a mixed crowd on hand to-night, and you can expect a
good card. In one of the ringside seats is the district attorney, a man
who loves a fair fight in or out of the ring. Further up are a few
brokers who have thought it worth while to come down here for one
night, anyhow. It is safe to say that every class in life is represented,
the man who is worth a million rubs elbows with the ten-dollar-a-
week clerk and they fraternize as freely as though they were chums.
“This Abe Attell is a clever boy, but they say he hasn’t the
punch,” ventures the clerk.
“Yes, I saw him recently and he made that big fellow look like a
cart horse,” returns the man of money.
The fellow who paid one-tenth of his weekly stipend to join the
club for that one night, which, by the way, is the system employed
to evade the law on the subject, pulls out a cigarette, and asks:
“Can I trouble you for a light?”
 “No trouble at all,” comes the cheerful answer, and a glowing
perfecto, which cost not less than thirty-five cents, is handed over.
That miscellaneous crowd is welded into one solid mass by the
masonry of sport, even though individual opinions are retained, and
the opinion of a seasoned ring-goer is set hard and deep as the rock
of Gibraltar.
The smoke is wafted back and forth like the tidal currents of the
sea and the exertions of a hundred devotees of nicotine are adding
to it every moment. An interminable buzz of voices fills the big room,
and there is fight in the very air.
“I tell you the old man could lick O’Brien any day he wanted to;
he’s got the punch and he can stand the gaff, ain’t that enough?”
This in a strident voice from the cheaper seats, and it was answered
at once by an argument that was apparently deemed irrefutable:
“Why didn’t he do it?”
Near the door is a fight bug whom no one ever heard of, and
who is interesting simply because he is a freak. He is voluble,
emphatic and vainglorious.
“I kin beat Britt an’ he knows it, an’ dat’s the reason he won’t
give me a chanst. He’d be a pipe fer me, ‘cos I’d infight him, an’ he
couldn’t stand my body punchin’. Dere’s where I’m great—on dose
body blows. I challenged him three times an’ he never paid no
attention to me. He’s afraid uv me, dat’s what he is. I kin beat ’em
all if dey’ll only cum to me.”
“You couldn’t beat a carpet,” shouts a wit, and the bug is
temporarily squelched.
The noise of the voices is suddenly emphasized—the first pair
are coming and the show is on. Into the ring they climb from
opposite corners, principals and seconds, and then, more leisurely,
as befits the dignity of his exalted position, comes the announcer.
They all have the same speech, which has been doing duty for
generations, and this one is no different from the rest: