Download the full version of the textbook now at textbookfull.

com

Econometrics of panel data : methods and

applications First Edition Biørn

https://textbookfull.com/product/econometrics-of-
panel-data-methods-and-applications-first-edition-
biorn/

Explore and download more textbook at https://textbookfull.com

 Recommended digital products (PDF, EPUB, MOBI) that
you can download immediately if you are interested.

Time Series and Panel Data Econometrics First Edition M.

Hashem Pesaran

https://textbookfull.com/product/time-series-and-panel-data-
econometrics-first-edition-m-hashem-pesaran/

textbookfull.com

Principles of Econometrics: Theory and Applications 1st

Edition Mignon

https://textbookfull.com/product/principles-of-econometrics-theory-
and-applications-1st-edition-mignon/

textbookfull.com

Relative fidelity processing of seismic data : methods and

applications 1 edition Edition Wang

https://textbookfull.com/product/relative-fidelity-processing-of-
seismic-data-methods-and-applications-1-edition-edition-wang/

textbookfull.com

Touch of Wrath 1st Edition Autumn Reed

https://textbookfull.com/product/touch-of-wrath-1st-edition-autumn-
reed/

textbookfull.com
The Complete Textbook of Phlebotomy 5th Edition Lynn B
Hoeltke

https://textbookfull.com/product/the-complete-textbook-of-
phlebotomy-5th-edition-lynn-b-hoeltke/

textbookfull.com

Modern C Efficient and Scalable Application Development

Leverage the modern features of C to overcome difficulties
in various stages of application development First Edition
Richard Grimes
https://textbookfull.com/product/modern-c-efficient-and-scalable-
application-development-leverage-the-modern-features-of-c-to-overcome-
difficulties-in-various-stages-of-application-development-first-
edition-richard-grimes/
textbookfull.com

EAI International Conference on Technology, Innovation,

Entrepreneurship and Education: TIE'2018 Ping Zheng

https://textbookfull.com/product/eai-international-conference-on-
technology-innovation-entrepreneurship-and-education-tie2018-ping-
zheng/
textbookfull.com

Lonely Planet Cancún Cozumel the Yucatán Lonely Planet

https://textbookfull.com/product/lonely-planet-cancun-cozumel-the-
yucatan-lonely-planet/

textbookfull.com

The Nurse Practitioner in Urology: A Manual for Nurse

Practitioners, Physician Assistants and Allied Healthcare
Providers 2nd Edition Michelle Lajiness
https://textbookfull.com/product/the-nurse-practitioner-in-urology-a-
manual-for-nurse-practitioners-physician-assistants-and-allied-
healthcare-providers-2nd-edition-michelle-lajiness/
textbookfull.com
Data Lake Architecture Designing the Data Lake and
Avoiding the Garbage Dump First Edition Bill Inmon

https://textbookfull.com/product/data-lake-architecture-designing-the-
data-lake-and-avoiding-the-garbage-dump-first-edition-bill-inmon/

textbookfull.com
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

Econometrics of Panel Data

OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

Econometrics
of Panel Data
Methods and Applications

Erik Biørn

1
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

3
Great Clarendon Street, Oxford, ox2 6dp,
United Kingdom
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide. Oxford is a registered trade mark of
Oxford University Press in the UK and in certain other countries
© Erik Biørn 2017
The findings, interpretations, and conclusions expressed in this work are entirely
those of the authors and should not be attributed in any manner to the World Bank,
its Board of Executive Directors, or the governments they represent.
The moral rights of the author have been asserted
First Edition published in 2017
Impression: 1
All rights reserved. No part of this publication may be reproduced, stored in
a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by licence, or under terms agreed with the appropriate reprographics
rights organization. Enquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above
You must not circulate this work in any other form
and you must impose this same condition on any acquirer
Published in the United States of America by Oxford University Press
198 Madison Avenue, New York, NY 10016, United States of America
British Library Cataloguing in Publication Data
Data available
Library of Congress Control Number: 2016937737
ISBN 978–0–19–875344–5
Printed in Great Britain by
Clays Ltd, St Ives plc
Links to third party websites are provided by Oxford in good faith and
for information only. Oxford disclaims any responsibility for the materials
contained in any third party website referenced in this work.
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

P R E FACE

Panel data is a data type used with increasing frequency in empirical research in eco-
nomics, social sciences, and medicine. Panel data analysis is a core field in modern
econometrics and multivariate statistics, and studies based on such data occupy a growing
part of the field in all the mentioned disciplines. A substantial literature on methods
and applications has accumulated, also synthesized in survey articles and a number of
textbooks. Why then write another book on panel data analysis? I hope that some of my
motivation and part of the answer will be given in the following paragraphs.
The text collected in this book has originated, and been expanded through several years,
partly in parallel with courses I have given at the University of Oslo for master’s and
doctoral students of economics. I have been interested in the field, both its models and
methods and applications to social sciences, for more than forty years. The first drafts were
lecture notes, later expanded and synthesized into course compendia, first in Norwegian,
later in English. In compiling the text, I have had no ambition of giving an account of
the history of panel data analysis and its various subfields, some of which have a longer
history than others. Within some 350 pages it is impossible to give a complete coverage,
and the choice of topics, and the depth of discussion of each of them, to some extent reflect
my preferences. Some readers may miss topics like cross-section-time-series analysis in
continuous time, duration analysis, and analysis of non-linear panel data models (outside
the limited dependent variables field). Topics I give specific attention to, more than many
comparable texts, I think, are coefficient identification of models mixing two-dimensional
and individual-specific variables, regression models with two-way random effects, models
and methods for handling random coefficients and measurement errors, unbalanced
panel data, and panel data in relation to aggregation. Problems at the interface between
unbalance and truncation, and between micro-econometrics and panel data analysis, are
also discussed.
It goes without saying that in a book dealing with data showing temporal–spatial vari-
ation, matrix algebra is unavoidable. Although panel data have a matrix structure, such
algebra should not be a core matter. I have experienced that many students starting on the
topic feel parts of the matrix algebra, especially when written in a dense, compact style,
to be an obstacle. I therefore set out to explain many of the models and methods in some
detail to readers coming to panel data analysis for the first time, including students famil-
iar with basic statistics and classical multiple regression analysis, and applied researchers.
Some technical material is placed in appendices. Yet some advanced and ‘modern’ topics
are discussed. Since one of my intentions is that students should be given the chance to
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

vi PREFACE

get an acceptable grasp of basic ideas without having to be involved in extensive matrix-
algebraic exercises, most chapters start with a simple example in scalar (or elementary
vector) notation, and then attempt to take the reader gradually to more complex cases in
fuller matrix notation, even if this necessitates some repetition.
The first chapters survey rather elementary materials. I believe that the initial sections
of the first eight chapters may be useful for bachelor students, provided their knowledge
of multiple regression analysis and basic mathematical statistics is sufficient. Chapters 1–3
and 5 contain mostly basic topics, Chapters 4 and 6–10 contain intermediate level topics,
and Chapters 11 and 12 contain somewhat advanced topics. I have, as far as possible, tried
to keep each chapter self-contained. Yet some connections exist. It may also be helpful to
note that Chapter 6 builds on Chapters 2, 3, and 5, that Chapters 6 and 7 expand topics in
Chapters 2–4, that Chapters 7 and 8 are methodologically related, that Chapter 11 builds
on Chapters 9 and 10, and that Chapter 12 builds on Chapters 3 and 6. Each chapter has
an initial summary, some also have a concluding section.
Some readers may miss exercises with solutions, which would have expanded the size
of the book. On the other hand, examples of applications, some from my own research,
or experiments, and illustrations utilizing publicly available data are included in several
chapters. My view of the panel data field is that the core topics have, to some extent, the
character of being building blocks which may be combined, and the number of potential
combinations is large. Not many combinations are discussed explicitly. Certain potential
combinations—for example, dynamic equations with random coefficients for unbalanced
panel data and random coefficients interacting with limited dependent variables with
measurement errors—have (to my knowledge) hardly been discussed in any existing text.
Although primarily aiming at students and practitioners of econometrics, I believe that
my book, or parts of it, may be useful also for students and researchers in social sciences
outside economics and for students and research workers in psychology, political science,
and medicine, provided they have a sufficient background in statistics. My intention, and
hope, is that the book may serve as a main text for lecturing and seminar education at
universities. I believe that parts may be useful for students, researchers, and other readers
working on their own, inter alia, with computer programming of modules for panel data
analysis.
During the work, I have received valuable feedback from many colleagues and students,
not least PhD students writing their theses, partly under my supervision, on applied panel
data topics. Questions frequently posed during such discussions have undoubtedly made
their mark on the final text. I want to express my gratitude to the many students who
have read, commented on, and in others ways been ‘exposed to’ my notes, sketches, and
chapter drafts. Their efforts have certainly contributed to eliminate errors. I also thank
good colleagues for many long and interesting discussions or for having willingly spent
their time in reading and commenting on drafts of preliminary versions of the various
sections and chapters, sometimes more than once. I regret that I have been unable to take
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

PREFACE vii

all their advice into account. I specifically want to mention (in alphabetical order) Jørgen
Aasness, Anne Line Bretteville-Jensen, John K. Dagsvik, Xuehui Han, Terje Skjerpen,
Thor Olav Thoresen, Knut R. Wangen, and Yngve Willassen. Needless to say, none of
them are to be held responsible for remaining errors or shortcomings. The text has been
prepared by the author in the Latex document preparation software, and I feel obliged
to the constructors of this excellent scientific text processor. Last, but not least, I express
my sincere gratitude to Oxford University Press, in particular Adam Swallow and Aimee
Wright, for their belief in the project and for support, encouragement, and patience.
Erik Biørn
Oslo, March 2016
OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

C O NTEN TS

LIST OF TABLES xvii

1 Introduction 1
1.1 Types of panel variables and data 1
1.2 Virtues of panel data: Transformations 3
1.3 Panel data versus experimental data 8
1.4 Other virtues of panel data and some limitations 9
1.5 Overview 11

2 Regression analysis: Fixed effects models 14

2.1 Simple regression model: One-way heterogeneity 15
2.1.1 Individual-specific intercepts and coefficients 15
2.1.2 Individual-specific intercepts, common coefficient 16
2.1.3 Homogeneous benchmark model 19
2.1.4 How are the estimators related? 21
2.2 Multiple regression model: One-way heterogeneity 23
2.2.1 Individual-specific intercepts and coefficients 23
2.2.2 Individual-specific intercepts, common coefficients 25
2.2.3 Homogeneous benchmark model 27
2.2.4 How are the estimators related? 30
2.3 Simple regression model: Two-way heterogeneity 33
2.3.1 Individual- and period-specific intercepts 33
2.3.2 Homogeneous benchmark model 36
2.3.3 How are the estimators related? 39
2.4 Multiple regression model: Two-way heterogeneity 41
2.4.1 An excursion into Kronecker-products: Definition 41
2.4.2 Matrix formulae with Kronecker-products: Examples 42
2.4.3 Panel data ‘operators’: Bilinear and quadratic forms 43
2.4.4 Individual- and period-specific intercepts 46
2.4.5 Homogeneous benchmark model 49
2.4.6 How are the estimators related? 50
2.5 Testing for fixed heterogeneity 53
2.5.1 One-way intercept and coefficient heterogeneity 54
2.5.2 Two-way intercept heterogeneity 55
Visit https://textbookfull.com
now to explore a rich
collection of eBooks, textbook
and enjoy exciting offers!
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

x CONTENTS

Appendix 2A. Properties of GLS 56

Appendix 2B. Kronecker-product operations: Examples 61

3 Regression analysis: Random effects models 65

3.1 One-way variance component model 66
3.1.1 Basic model 66
3.1.2 Some implications 67
3.2 GLS with known variance components 68
3.2.1 The GLS problem and its reformulation 68
3.2.2 GLS as OLS on transformed data 70
3.2.3 Four estimators and a synthesis 71
3.3 GLS with unknown variance components 74
3.3.1 The problem 74
3.3.2 ‘Estimation’ of variance components from disturbances 75
3.3.3 Synthesis: Stepwise estimation 76
3.3.4 Testing for random one-way heterogeneity 78
3.3.5 Disturbance serial correlation and heteroskedasticity 79
3.4 Maximum Likelihood estimation 81
3.4.1 The problem 81
3.4.2 Simplifying the log-likelihood 82
3.4.3 Stepwise solution 83
3.5 Two-way variance component model 84
3.5.1 Basic assumptions 84
3.5.2 Some implications 85
3.6 GLS with known variance components 87
3.6.1 The problem and its reformulation 87
3.6.2 GLS as OLS on transformed data 88
3.6.3 Five estimators and a synthesis 89
3.7 GLS with unknown variance components 92
3.7.1 The problem 92
3.7.2 ‘Estimation’ of variance components from disturbances 93
3.7.3 Synthesis: Stepwise estimation 95
3.7.4 Testing for random two-way heterogeneity 95
3.8 Maximum Likelihood estimation 96
3.8.1 The problem 96
3.8.2 Simplifying the log-likelihood 96
3.8.3 Stepwise solution 97
Appendix 3A. Two theorems related to GLS Estimation 99
Appendix 3B. Efficiency in the one-regressor case 100
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

CONTENTS xi

4 Regression analysis with heterogeneous coefficients 107

4.1 Introduction 107
4.2 Fixed coefficient models 108
4.2.1 Single-regressor models without intercept 108
4.2.2 Synthesis 113
4.2.3 Including intercepts 114
4.2.4 Multi-regressor models 115
4.3 Random coefficient models 116
4.3.1 Single-regressor models and their estimators 117
4.3.2 Multi-regressor model and its GLS estimator 123
4.3.3 Coefficient prediction 127
Appendix 4A. Matrix inversion and matrix products: Useful results 132
Appendix 4B. A reinterpretation of the GLS estimator 134

5 Regression analysis with unidimensional variables 136

5.1 Introduction 136
5.2 Properties of moment matrices in unidimensional variables 137
5.2.1 Basics 137
5.2.2 Implications for regression analysis 138
5.3 One-way fixed effects models 139
5.3.1 Simple model with two regressors 139
5.3.2 The general case 143
5.4 One-way random effects models 145
5.4.1 Simple model with two regressors 145
5.4.2 The general case 147
5.5 Two-way models 150
5.5.1 Fixed effects two-regressor model 150
5.5.2 Random effects two-regressor model 154

6 Latent heterogeneity correlated with regressors 157

6.1 Introduction 157
6.2 Models with only two-dimensional regressors 158
6.2.1 A simple case 158
6.2.2 Generalization 161
6.3 Models with time-invariant regressors 164
6.3.1 A simple case 164
6.3.2 Generalization 166
6.3.3 The Wu–Hausman test 168
6.4 Exploiting instruments: Initial attempts 171
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

xii CONTENTS

6.4.1 A simple example 171

6.4.2 Generalization 172
6.4.3 Reinterpretation using a recursive system 173
6.5 Exploiting instruments: Further steps 175
6.5.1 Choosing instruments: General remarks 176
6.5.2 Estimation in case of exact identification 177
6.5.3 Estimation in case of overidentification 181
Appendix 6A. Reinterpretation: Block-recursive system 185
Appendix 6B. Proof of consistency in case of exact identification 187

7 Measurement errors 189

7.1 Introduction 189
7.2 A homogeneous model 190
7.2.1 Model setup: Basic probability limits 191
7.2.2 Basic (disaggregate) estimators 192
7.2.3 Aggregate estimators 194
7.2.4 Combining inconsistent estimators 198
7.2.5 Difference estimators 199
7.3 Extension I: Heterogeneity in the equation 202
7.3.1 Model 202
7.3.2 Between and within estimation 203
7.3.3 Combining inconsistent estimators 204
7.4 Extension II: Heterogeneity in the error 206
7.4.1 Model 206
7.4.2 Estimation 206
7.4.3 Combining inconsistent estimators 209
7.5 Extension III: Memory in the error 209
7.5.1 One-component error with memory 210
7.5.2 Three-component specification 212
7.6 Generalized Method of Moments estimators 215
7.6.1 Generalities on the GMM 215
7.6.2 GMM-estimation of the equation in differences 217
7.6.3 Extensions and modifications 222
7.7 Concluding remarks 223
Appendix 7A. Asymptotics for aggregate estimators 224

8 Dynamic models 227

8.1 Introduction 227
8.2 Fixed effects AR-models 229
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

CONTENTS xiii

8.2.1 A simple model 229

8.2.2 Within estimation: Other difference transformations 230
8.2.3 Equation in differences: Simple estimation by level IVs 234
8.2.4 Equation in differences: System GMM with level IVs 236
8.2.5 Equation in levels: Simple estimation by difference IVs 238
8.2.6 Equation in levels: System GMM with difference IVs 238
8.2.7 Panel unit roots and panel co-integration 240
8.2.8 Bias correction 242
8.3 Fixed effects AR-models with exogenous variables 242
8.3.1 Model 242
8.3.2 Equation in differences: System GMM with level IVs 244
8.3.3 Equation in levels: System GMM with difference IVs 246
8.3.4 GMM estimation: AR-model versus EIV-model 249
8.3.5 Extensions and modifications 250
8.4 Random effects AR(1) models 250
8.4.1 A simple AR(1) model 250
8.4.2 Including exogenous variables 252
8.5 Conclusion 254
Appendix 8A. Within estimation of the AR coefficient: Asymptotics 255
Appendix 8B. Autocovariances and correlograms of yit and yit 257

9 Analysis of discrete response 261

9.1 Introduction 261
9.2 Binomial models: Fixed heterogeneity 262
9.2.1 Simple binomial model: Logit and probit parameterizations 263
9.2.2 Binomial model with full fixed heterogeneity 269
9.2.3 Binomial model with fixed intercept heterogeneity 270
9.2.4 Fixed intercept heterogeneity, small T and conditional ML 270
9.3 Binomial model: Random heterogeneity 275
9.3.1 Model 275
9.3.2 ML estimation 277
9.4 Multinomial model: Other extensions 278
9.4.1 Model 279
9.4.2 Likelihood function and ML estimation 281
9.4.3 ML estimation conditional on the individual effects 282
Appendix 9A. The general binomial model: ML Estimation 283
Appendix 9B. The multinomial logit model: Conditional ML estimation 285
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

xiv CONTENTS

10 Unbalanced panel data 287

10.1 Introduction 287
10.2 Basic model and notation 289
10.2.1 Formalization of the non-balance 289
10.2.2 Individual-specific, period-specific, and global means 291
10.3 The fixed effects case 292
10.3.1 The one-regressor case 292
10.3.2 Between- and OLS-estimation in the one-regressor case 294
10.3.3 The multi-regressor case 296
10.3.4 Between- and OLS-estimation in the general case 297
10.4 The random effects case 298
10.4.1 The one-regressor case 299
10.4.2 The multi-regressor case 302
10.4.3 Estimating the variance components: FGLS 303
10.5 Maximum Likelihood estimation 306
10.5.1 A convenient notation and reordering 306
10.5.2 Reformulating the model 307
10.5.3 ‘Estimating’ the variance components from disturbances 308
10.5.4 The log-likelihood function 309
Appendix 10A. Between estimation: Proofs 312
Appendix 10B. GLS estimation: Proofs 313
Appendix 10C. Estimation of variance components: Details 315

11 Panel data with systematic unbalance 317

11.1 Introduction: Observation rules 317
11.2 Truncation and censoring: Baseline models 320
11.2.1 Basic concepts and point of departure 320
11.2.2 Truncation and truncation bias 321
11.2.3 Censoring and censoring bias 324
11.2.4 Maximum Likelihood in case of truncation 326
11.2.5 Maximum Likelihood in case of censoring 327
11.3 Truncation and censoring: Heterogeneity 328
11.3.1 Truncation: Maximum Likelihood estimation 329
11.3.2 Censoring: Maximum Likelihood estimation 330
11.4 Truncation and censoring: Cross-effects 332
11.4.1 Motivation and common model elements 333
11.4.2 Implied theoretical regressions for the observable variables 334
11.4.3 Stepwise estimation: Examples 335
11.4.4 Extensions: ML estimation—Heterogeneity 338
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

CONTENTS xv

Appendix 11A. On truncated normal distributions 340

Appendix 11B. Partial effects in censoring models 345

12 Multi-equation models 347

12.1 Introduction 347
12.2 Regression system with one-way random effects 348
12.2.1 Model description 349
12.2.2 Matrix version: Two formulations 349
12.2.3 GLS estimation 353
12.2.4 Maximum Likelihood estimation 355
12.3 Regression system with two-way random effects 358
12.3.1 Model description 358
12.3.2 Matrix version: Two formulations 359
12.3.3 GLS estimation 362
12.4 Interdependent model: One-equation estimation 362
12.4.1 Within and between two-stage least squares 364
12.4.2 Variance components two-stage least squares 366
12.5 Interdependent model: Joint estimation 371
12.5.1 Within and between three-stage least squares 372
12.5.2 Variance components three-stage least squares 374
Appendix 12A. Estimation of error component covariance matrices 376
Appendix 12B. Matrix differentiation: Useful results 378
Appendix 12C. Estimator covariance matrices in interdependent models 379

REFERENCES 383
INDEX 394
OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

L I S T O F TA B L E S

2.1 Marginal budget share, promille 32

3.1 Transforming the GLS problem into an OLS problem 71
3.2 Marginal budget share, promille 77
3.3 Relative efficiency of estimators 78
4.1 Hospital-specific coefficient of ABF dummy 130
4.2 Estimated coefficient mean of ABF dummy 131
4.3 ABF coefficient: distribution of estimates 131
5.1 Equation with individual specific regressors 149
6.1 Wage equation: impact of work experience on wage rate 161
6.2 Wage equation example: including education 166
6.3 Fixed vs. Random effects: petrol consumption 170
6.4 Wage equation example: estimates from Hausman–Taylor procedure 185
7.1 Probability limits of aggregate estimators in Model (7.1)–(7.2) 196
7.2 Probability limits of aggregate estimators in Model (7.55)–(7.56) 207
7.3 Probability limits in models without and with error memory 211
7.4 Input–output elasticities and their inverses 221
8.1 Labour demand equation: OLS estimates 248
8.2 Labour demand equation: Arellano–Bond and Blundell–Bond GMM estimates 249
8.3 Asymptotics for ML estimators of β and δ in Model (8.63) 253
9.1 Binomial logit analysis of drug use: latent heterogeneity not accounted 268
9.2 Binomial logit analysis of drug use: heterogeneity, as random effects, accounted for 278
10.1 Unbalanced panel data: health satisfaction example 305
10.2 Correlation, regressor vs. regressand hsat 306
11.1 Censoring and truncation: equation for female log(wage rate) 332
12.1 Wage equation, simple: OLS and panel data GLS-2SLS estimates 370
12.2 Wage equation, extended: OLS and panel data GLS-2SLS estimates 370
OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

1 Introduction

Panel data, longitudinal data, or combined time-series/cross-section data are terms used
in econometrics and statistics to denote data sets which contain repeated observations
on a selection of variables from a set of observation units. The observations cover simul-
taneously the temporal and the spatial dimension. Examples are: (1) time-series for
production, factor inputs, and profits in a sample of firms over a succession of years;
(2) time-series for consumption, income, wealth, and education in a sample of persons
or households over several years; and (3) time-series for manufacturing production, sales
of medical drugs, or traffic accidents for all, or a sample of, municipalities or counties,
or time-series for variables for countries in the OECD, the EU, etc. Examples (1) and (2)
relate to micro-data, (3) exemplifies macro-data.
‘Panel data econometrics is one of the most exciting fields of inquiry in econometrics
today’ (Nerlove, Sevestre, and Balestra (2008, p. 22)), with a history going back to at least
1950. We will not attempt to survey this interesting history. Elements of a survey can be
found in Nerlove (2002, Chapter 1); see also Nerlove (2014) as well as Griliches (1986,
Sections 5 and 6) and Nerlove, Sevestre and Balestra (2008). A background for the study
of panel data, placing it in a wider context, may also be given by a quotation from a text
on ‘modern philosophy’:
Space and time, or the related concepts of extension and duration, attained special prominence
in early modern philosophy because of their importance in the new science . . . Metaphysical
questions surrounding the new science pertained to the nature of space and time and their relation
to matter. Epistemological questions pertained to the cognition of space itself or extension in
general . . . and also to the operation of the senses in perceiving the actual spatial order of things.
(Hatfield (2006, p. 62))

In this introductory chapter, we first, in Section 1.1, briefly define the main types of
panel data. In Section 1.2 we illustrate, by examples, virtues of panel data and explain in
which sense they may ‘contain more information’ than the traditional data types cross-
section data and time-series data. Section 1.3 briefly contrasts panel data with experimen-
tal data, while some other virtues of panel data, as well as some limitations, are specified
in Section 1.4. An overview of the content of the book follows, in Section 1.5.

1.1 Types of panel variables and data

Several types of panel data exist. The one which perhaps first comes to mind is balanced
panel data, in which the same individuals are observed in all periods under consideration.
Visit https://textbookfull.com
now to explore a rich
collection of eBooks, textbook
and enjoy exciting offers!
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

2 ECONOMETRICS OF PANEL DATA

Using i as subscript for the unit of observation and t as subscript for the time-period, and
letting the data set contain N units and T time-periods, the coverage of a balanced panel
data set can be denoted as i = 1, . . . , N; t = 1, . . . , T.
Balanced panel data have a matrix structure. We need, in principle, three subscripts to
represent the observations: one for the variable number; one for the individual (unit)
number; and one for the period (year, quarter, month, etc.). A balanced panel data set can
therefore be arranged as three-dimensional matrices. Quite often N is much larger than T,
but for panels of geographic units, the opposite may well be the case. Regarding asymp-
totics, the distinction between N → ∞ and T → ∞, often denoted as ‘N-asymptotics’
(cross-sectional asymptotics) and ‘T-asymptotics’ (time-serial asymptotics), will often
be important. In ‘micro’ contexts, ‘short’ panels from ‘many’ individuals is a frequently
occurring constellation.
Quite often, however, some variables do not vary both across observation unit and
over time-periods. Examples of variables which do not vary over time, time-invariant
variables, are for individuals: birth year; gender; length of the education period (if for
all individuals the education has been finished before the sample period starts); and to
some extent attitudes, norms, and preferences. For firms they are: year of establishment;
sector; location; technical strength; and management ability. Such variables are denoted
as individual-specific or firm-specific. Examples of variables that do not vary across
individuals, individual-invariant variables, may be prices, interest rates, tax parameters,
and variables representing the macro-economic situation. Such variables are denoted
as time-specific or period-specific. As a common term for individual-specific and time-
specific variable we will use unidimensional variables. Variables showing variation across
both individuals and time-periods are denoted as two-dimensional variables. Examples
are (usually) income and consumption (for individuals) and production and labour input
(for firms).
The second, also very important, category is unbalanced panel data. Its characteristic is
that not the same units are observed in all periods, but some are observed more than once.
There are several reasons why a panel data set may become unbalanced. Entry and exit
of units in a data base (e.g., establishment and close-down of firms and marriages and
dissolution of households) is one reason, another is randomly missing observations in
time series. A particular type of unbalance is created by rotating panel data, which emerges
in sample surveys when the sample changes systematically in a way intended by the data
collector. Another major reason why unbalanced panel data may occur is endogenous
selection, meaning, loosely, that the selection of units observed is partly determined by
variables our model is intended to explain. This may complicate coefficient estimation
and interpretation if the selection mechanism is neglected or improperly accounted for
in the modelling and design of inference method. Asserting that selection problems are
potentially inherent in any micro-data set, panel data as well as cross-section data, is hardly
an exaggeration.
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

INTRODUCTION 3

1.2 Virtues of panel data: Transformations

Several advantages can be obtained by utilizing panel data instead of time-series data or
cross-section data in an empirical investigation. Panel data are in several respects ‘richer’.
As a general characteristic we may say: pure time-series data contain no information about
individual differences, and pure cross-section data contain no information about period-
specific differences. We therefore are unable from time-series data to explore effects of
individual-specific variables and from cross-section data to examine effects of time-
specific variables. Panel data do not have, or have to a far smaller degree, these limitations,
not least because such data admit many useful transformations. Nerlove, Sevestre, and
Balestra (2008, pp. 4–5) give the following remarks on the analyst’s need to give due
attention to the way the actual data type is generated:
In many applications in the social sciences, especially in economics, the mechanism by which the
data are generated is opaque . . . Understanding the process by which the observations at hand
are generated is of equal importance. Were the data for example obtained from a sample of firms
selected by stratified random sampling . . .? In the case of time series, the data are almost always
“fabricated” in one way or another, by aggregation, interpolation, or extrapolation, or by all three.
The nature of the sampling frame or the way in which the data are fabricated must be part of the
model specification on which parametric inference or hypothesis testing is based.

We will, with reference to an example, illustrate differences in the ‘information’ con-

tained in pure time series data and pure cross-section data on the one hand, and in
balanced panel data on the other, and the transformations made possible by the latter.
Consider an equation explaining the conditional expectation of y linearly by x, z, and q,
where y and x are two-dimensional variables, z is individual-specific, and q is period-
specific. We assume

E(yit |x, z, q) = k + xit β + zi α + qt γ ,

where k is an intercept; β, α, and γ are coefficients and x, z, q are (row) vectors containing
all values of (x, z, q) in the data set; and i and t are index individuals and time periods,
respectively. We assume that xit , zi , qt , β, α, and γ are scalars, but the following argument
easily carries over to the case where the variables are row-vectors and the coefficients are
column-vectors. This expression is assumed to describe the relationship between E(y|x, z, q)
and x, z, q for any values of i and t. Let uit = yit −E(yit |x, z, q), which can be interpreted
as a disturbance, giving the equivalent formulation

yit = k + xit β + zi α + qt γ + uit , E(uit |x, z, q) = 0, for all i, t.

First, assume that the data set is balanced panel data from N individuals and T periods,
so that we can specify
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

4 ECONOMETRICS OF PANEL DATA

yit = k + xit β + zi α + qt γ + uit , E(uit |x, z, q) = 0,

(1.1)
i = 1, . . . , N; t = 1, . . . , T,

where now (x, z, q) denote vectors containing the values of (xit , zi , qt ) for i =
1, . . . , N; t = 1, . . . , T. A researcher may sometimes give primary attention to β and
wants to estimate it without bias, but α and γ may be of interest as well. Anyway,
we include zi and qt as explanatory variables because our theory implies that they
are relevant in explaining yit , and we do not have experimental data which allow
us to ‘control for’ zi and qt by keeping their values constant in repeated samples. If
uit has not only zero conditional expectation, but also is homoskedastic and serially
uncorrelated, we can from the NT observations on (yit , xit , zi , qt ) estimate (k, β, α, γ )
by ordinary least squares (OLS), giving Minimum Variance Linear Unbiased Estima-
tors (MVLUE), the ‘best possible’ linear estimators, or Gauss–Markov estimators of the
coefficients.
What would have been the situation if we only had had access to either time-series or
cross-section data? Assume first that pure time-series, for individual i = 1 (i.e., N = 1) in
periods t = 1, . . . , T, exist.1 Then Model (1.1) should be specialized to this data situation
by (conditioning on z1 is irrelevant)

y1t = (k+z1 α) + x1t β + qt γ + u1t , E(u1t |x1· , q) = 0, t = 1, . . . , T, (1.2)

where x1· = (x11 , . . . , x1T ). From time-series for y1t , x1t , and qt , we could estimate β, γ ,
and the composite intercept k + z1 α. This confirms: (i) pure time-series data contain no
information on individual differences or on effects of individual-specific variables; (ii) the
intercept is specific to the individual (unit); (iii) the coefficient α cannot be identified, as it
belongs to a variable with no variation over the data set (having observed z1 is of no help);
and (iv) the coefficients β and γ can be identified as long as x1t and qt are observable and
vary over periods. If u1t is homoskedastic (over t) and shows no serial correlation (over t),
then OLS applied on (1.2) will give estimators which are MVLUE for these coefficients in
the pure time-series data case.
Next, assume that a cross-section, for period t = 1 (i.e., T = 1), for individuals
i = 1, . . . , N, exists. Then (1.1) should be specialized to (conditioning on q1 is irrelevant):

yi1 = (k+q1 γ ) + xi1 β + zi α + ui1 , E(ui1 |x·1 , z) = 0, i = 1, . . . , N, (1.3)

1 We here consider the case with time-series from only one individual (unit) to retain symmetry with the
pure cross-section case below. Most of our following conclusions, however, carry without essential modifica-
tions over to situations with aggregate time-series for a sector or for the entire economy, since the equation
is linear. But individual-specific time-series are far from absent. We could, for example, possess annual time-
series of sales, stock prices, or employment for a specific company.
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

INTRODUCTION 5

where x·1 = (x11 , . . . , xN1 ). From cross-section data for yi1 , xi1 , and zi , we could estimate
β, α, and the composite intercept k+q1 γ . This confirms: (i) pure cross-section data contain
no information on period-specific differences or on the effects of period-specific variables;
(ii) the intercept is specific to the data period; (iii) the coefficient γ cannot be identified,
as it belongs to a variable with no variation over the data set (having observed q1 is of no
help); (iv) the coefficients β and α can be identified as long as xi1 and zi are observable and
vary across individuals. If ui1 is homoskedastic (over i) and serially uncorrelated (over i),
then OLS applied on (1.3) will give MVLUE for these coefficients in the pure cross-section
data case.
By panel data we may circumvent the problem of lack of identification of α from
times series data, when using the ‘time-series equation’ (1.2) and of γ from cross-
section data, when using the ‘cross-section equation’ (1.3). Moreover, we may control
for unobserved individual-specific or time-specific heterogeneity. We illustrate this from
(1.1), by first taking the difference between the equations for observations (i, t) and (i, s),
giving

yit −yis = (xit −xis )β +(qt −qs )γ +(uit −uis ), E(uit −uis |x, q) = 0,
(1.4)
i = 1, . . . , N; t, s = 1, . . . , T (t  = s),

from which zi vanishes and, in contrast to (1.1), E(uit |z) = 0 is not required for consistency
of OLS. We consequently ‘control for’ the effect on y of z and ‘retain’ only the variation
in y, x, and q. Having the opportunity to do so is crucial if zi is unobservable and reflects
unspecified heterogeneity, but still is believed to affect yit . To see this, assume that zi is
unobservable and correlated with xit (across i) and consider using OLS on (1.1) with zi
excluded, or on

yi1 = constant + xi1 β + ui1 , i = 1, . . . , N,

where ui1 is a disturbance in which we (tacitly) include the effect of zi . This gives a biased
estimator for β, since ui1 , via the correlation and the non-zero value of α, captures the
effect of zi on yi1 : we violate E(ui1 |z) = 0. This will not be the case if we instead use OLS
on (1.4).
By next in (1.1) taking the difference between the equations for observations (i, t) and
(j, t), it likewise follows that

yit −yjt = (xit −xjt )β +(zi −zj )α+(uit −ujt ), E(uit −ujt |x, z) = 0,
(1.5)
i, j = 1, . . . , N (i  = j); t = 1, . . . , T,

from which qt vanishes and, in contrast to (1.1), E(uit |q) = 0 is not required for con-
sistency of OLS. We consequently ‘control for’ the effect on y of q and ‘retain’ only the
 OUP CORRECTED PROOF – FINAL, 19/9/2016, SPi

6 ECONOMETRICS OF PANEL DATA

variation in y, x and z. Having the opportunity to do so is crucial if qt is unobservable and

reflects unspecified heterogeneity, but still is believed to affect yit . To see this, assume that
qt is unobservable and correlated with xit (over t) and consider using OLS on (1.1) with
qt excluded, or on

y1t = constant + x1t β + u1t , t = 1, . . . , T,

where ui1 is a disturbance in which we (tacitly) include the effect of qt . This gives a biased
OLS estimator for β, since u1t , via the correlation and the non-zero value of γ , captures
the effect of qt on y1t : we violate E(u1t |q) = 0. This not will be the case if we instead use
OLS on (1.5).
If only individual time-series (N = 1) are available, we may perform the transformation
leading to (1.4), but not the one leading to (1.5). Likewise, if one cross-section (T = 1) is
the only data available, we may perform the transformation leading to (1.5), but not the
one leading to (1.4). Transformations which may give both (1.4) and (1.5) are infeasible
unless we have panel data.
We also have the opportunity to make other, more complex, transformations of (1.1).
Deducting from (1.4) the corresponding equation when i is replaced by j (or deducting
from (1.5) the corresponding equation with t replaced by s) we obtain

(yit −yis )−(yjt −yjs )

= [(xit −xis )−(xjt −xjs )]β +(uit −uis )−(ujt −ujs ),
(1.6)
E[(uit −uis )−(ujt −ujs )|x] = 0,
i, j = 1, . . . , N (i  = j), t, s = 1, . . . , T (t  = s).

By this double differencing, zi and qt disappear, and neither E(uit |z) = 0 nor E(uit |q) = 0
is needed for consistency of the OLS estimators. We thus control for the effect on y of
both z and q and retain only the variation in x. To see this, assume that both zi and qt are
unobservable and correlated with xit (over i and t, respectively), and consider using OLS
on either (1.1) with both zi and qt omitted, on

y1t = constant + x1t β + u1t , t = 1, . . . , T,

or on

yi1 = constant + xi1 β + ui1 , i = 1, . . . , N,

while (tacitly) including the effect of, respectively, (qt , zi ), qt , and zi in the equation’s
disturbance, which will give biased (inconsistent) estimators for β. This will not be the
case when using OLS on (1.6).
Random documents with unrelated
content Scribd suggests to you:
two hours’ traffic of the stage often compel a playwright to telescope
time; but no other dramatist has ever dared so violent a
compression as this.
And this is how Verrall solves the difficulty “with lucidity, skill and
brilliance,” so Professor Norwood tells us. The story of the series of
beacons is a lie concocted by the wife and her lover. There is only
one beacon, which Ægisthus lights when he discovers the landing of
Agamemnon; it is to warn his accomplice that she may make ready
to murder her husband. And as Agamemnon is actually on shore
when this single beacon flames up, he is able to arrive in the middle
of the play. If we accept this solution of the difficulty we are
compelled to believe that Æschylus wrote a play, instantly accepted
as a masterpiece, which had to wait for more than two thousand
years for a British scholar to explain away an impossibility. This
explanation is undoubtedly lucid and skilful and brilliant; but none
the less is it a specimen of undramatic criticism. It could never have
been put forward by anyone who had an elementary knowledge of
the principles of playmaking.
A dramatist never tells lies to his audience; and the audience
always accepts the statements of his characters as true—unless he
himself takes care to suggest that a given statement is false. The
play has to be taken at its face value. The characters talk on purpose
to convey all needful information to the spectators. Æschylus may
make the queen lie to the king, but when she does this the audience
is aware of the truth or surmises it. The dramatist never hesitates to
let his characters deceive one another; but if he knows his business
he does not deceive the spectators. In real life Agamemnon could
not arrive for a fortnight after Troy had fallen; but the Athenian
audience could not wait in their seats two weeks, so Æschylus
frankly brings on Agamemnon; and the spectators were glad to
behold him, asking no inconvenient questions, because they were
eager to see what would happen to him. It might be a contradiction
of the fact, but it was not a departure from the truth, since the king
would assuredly come home sooner or later. Everyone familiar with
Sarcey’s discussion of the conventions of the drama is aware that
the spectators in the theater are never sticklers for fact; they are
willing to accept a contradiction of fact, if that contradiction is for
their own profit, as it was in this case. And they accept it
unthinkingly; and it is only when they hold the play in their hands to
pick it to pieces that they discover any “difficulty.”

III

To say this is to say that Verrall, however lucid and skilful and
brilliant, was a discoverer of mares’ nests. And a host of undramatic
critics have skilfully exercised their lucid brilliance in discovering
mares’ nests in Shakspere’s plays. Most of them are stolid Teutons,
with Gervinus and Ulrici in the forefront of the procession. They
analyzed the tragedies of Shakspere with the sincere conviction that
he was a philosopher with a system as elaborate as those of Kant
and Hegel; and they did not seem to suspect that even if a dramatist
is a philosopher he is—and must be—first of all a playwright, whose
invention and construction are conditioned by the theater for which
he is working. Even in the greatest plays philosophy is a by-product;
and the main object of the great dramatist is always to arouse and
retain and reward the interest of his immediate audience.
He must make his story plain to the comprehension of the average
playgoer; and he must therefore provide his characters with motives
which are immediately apparent and instantly plausible. Shakspere is
ever anxious that his spectators shall not be misled, and he goes so
far as to have his villains, Richard III and Iago, frankly inform the
audience that they are villains, a confession which in real life neither
of these astute scoundrels would ever have made to anybody. The
playwright knows that if he loses his case before the jury, he can
never move for a retrial; the verdict is without appeal. It may be
doubted whether any dramatist has ever cared greatly for the
opinion of posterity. Assuredly no popular playwright—and in their
own day every great dramatist was a popular playwright—would
have found any compensation for the failure of his play in the hope
and expectation that two hundred or two thousand years later its
difficulties might be explained by a Verrall, however lucid and skilful
and brilliant this belated expounder might be.
There are two Shaksperian mares’ nests which may be taken as
typical, altho the eggs in them are not more obviously addled than in
a host of others. One was discovered in ‘Macbeth,’ in the scene of
Banquo’s murder. Macbeth incites two men to make way with
Banquo; but when the deed is done, three murderers take part in it.
Two of them are the pair we have seen receiving instructions from
Macbeth. Who is the third? An undramatic critic once suggested that
this third murderer is no less a person than Macbeth himself, joining
his hired assassins to make sure that they do the job in workmanlike
fashion. The suggester supported his suggestion by an argument in
eight points, no one of which carries any weight, because we may
be sure that if Shakspere had meant Macbeth to appear in person,
he would have taken care to let the audience know it. He would not
have left it hidden to be uncovered two and a half centuries after his
death by the skilful lucidity of a brilliant undramatic critic.
It is reasonably certain that Burbage, who acted Richard III and
Hamlet, also acted Macbeth; and Shakspere would never have sent
this renowned performer on the stage to take part in a scene
without justifying his share in it and without informing the spectators
that their favorite was before them. Shakspere was an actor himself;
he knew what actors wanted and what they liked; he took good care
of their interests; and we may rest assured that he never asked
Burbage to disguise his identity. If he had meant the third murderer
to be Macbeth, we should have had the stage direction, “Enter two
murderers with Macbeth disguised.” As it is, the stage direction reads
“Enter three murderers.”
The other mare’s nest has been found in ‘King Lear.’ It has often
been pointed out that Cordelia is absent from a large portion of the
action of the tragedy, altho her presence might have aided its
effectiveness. It has been noted also that Cordelia and the Fool are
never seen on the stage together. And this has prompted the
suggestion that the Fool is Cordelia in disguise. Here again we see
the undramatic critic at his worst. If Shakspere had meant this, he
would have made it plain to the spectators the first time Cordelia
appeared as the Fool,—otherwise her assumption of this part would
have been purposeless, confusing, futile. Whatever poignancy there
might be in the companioning of the mad king by his cast-off
daughter all unknown to him, would be unfelt if her assumption of
the Fool’s livery was not at once recognized. The suggestion is not
only inacceptable, it is unthinkable by anyone who has even an
elementary perception of the playmaking art. It could have
emanated only from an undramatic critic who was familiar with ‘King
Lear’ in the study and not on the stage, who regarded the sublimest
of Shakspere’s tragedies as a dramatic poem and not as a poetic
drama planned for the playhouse.
Yet this inept suggestion can be utilized to explain the fact that
Cordelia and the Fool never meet before the eyes of the spectators.
The cast of characters in ‘King Lear’ is very long; and quite possibly
it called for more actors than there were in the limited company at
the Globe. We know that in the Tudor theater a performer was often
called upon to sustain two parts. It is possible that the shaven lad
who impersonated Cordelia was the only available actor for the Fool,
and that therefore Cordelia—at whatever loss to the effectiveness of
the play—could not appear in the scenes in which the Fool had to
appear. Cordelia did not don the disguise of the Fool; but the same
performer may have doubled the two parts. That much of
supposition can be ventured for whatever it may be worth.

IV

It is in England and in Germany that the undramatic critics have

been permitted to disport themselves most freely and most
frequently. In France they have never been encouraged to pernicious
activity. That the French have not suffered from this pest may be
due to the honorable existence of the Théâtre Français, where the
masterpieces of French tragedy and of French comedy have been
kept alive on the stage for which they had been written; or it may be
due to the fact that in the literature of France the drama has been
continuously more important than it has been in the literature of any
other country.
In England and in Germany the drama has had its seasons of
abundance and its seasons of famine, whereas in France, altho there
might be poor harvests for a succession of years, harvests of some
sort there always have been. No period in French literature is as
devoid of valid drama as that in English literature during the first
three-quarters of the nineteenth century. From 1800 to 1870 the
plays of our language which were actable were unreadable and the
plays which were readable were unactable. It is in the periods of
penury, when there is a divorce between literature and the drama,
that the undramatic critic is inspired to chase rainbows. As there is
then no vital drama in the theater, and as the pieces then exhibited
on the stage have little validity, the undramatic critic is led to the
conclusion that since the theater can get along without literature, so
the drama can get along without the theater. And that way madness
lies.
There is this excuse for the supersubtle critics of the Italian
Renascence that they lived not long removed from the middle ages,
in which all memory of the acted drama had been lost and in which
the belief was general that the comedies of Plautus and Terence had
been composed, not for performance by actors in a theater and
before an audience, but for a single reciter who should deal with
them as a modern elocutionist might stand and deliver ‘Pippa Passes’
or the ‘Cenci.’ But there is no excuse for the English-speaking
expounders of Sophocles and Shakspere, because they cannot help
knowing that the plays of the Athenian were written to be performed
in the Theater of Dionysus and that the plays of the Elizabethan
were written to be performed in the Globe theater.
A friend of mine, not yet forty, told me that as an undergraduate
he had read half-a-dozen Greek plays with a professor, who was an
enthusiastic admirer of Greek literature, who had spent a winter in
Athens, and who had acquired modern Greek. This professor spared
no pains to make his students appreciate the poetic beauty of
Athenian tragedy; but never once did he call their attention to the
circumstances of original performance or arouse their interest by
pointing out the theatrical effectiveness of the successive situations.
To this ardent lover of Greece and of Greek literature, the
‘Agamemnon,’ the ‘Œdipus,’ and the ‘Medea’ were only poems in
dialog; they were not plays composed to be acted, adjusted to the
conditions of the Athenian theater, and conforming to the
conventions tacitly accepted by the Attic audience.
But worse remains behind. The writer of the chapters on
Shakspere in the composite ‘Cambridge History of English Literature,’
deals skilfully and cautiously with the dates of composition and
performance of each of the plays; but he criticizes them with no
examination of their theatrical effectiveness. It is scarcely too much
to say that he considers them as dramatic poems intended to be
read rather than as poetic dramas intended to be acted. Nothing in
either of his chapters is evidence that he ever saw a comedy or a
tragedy of Shakspere’s on the stage. He reveals no knowledge of the
principles of playmaking; and it may be doubted whether he
suspects the existence of these principles. And in one passage of his
commentary he has given us the absolute masterpiece of undramatic
criticism:

It is, of course, quite true that all of Shakspere’s plays were written to be acted;
but it may be questioned whether this is much more than an accident arising from
the fact that the drama was the dominant form of literature. It was a happy
accident, because of the unique opportunity this form gives of employing both the
vehicles of poetry and prose.

(1921)
 III
OLD PLAYS AND NEW PLAYGOERS
 III
OLD PLAYS AND NEW PLAYGOERS

Every dramatist is of necessity subdued to what he works for—the

playgoers of his own generation in his own country. Their approval it
is that he has to win first of all; and if they render a verdict against
him he has no appeal to posterity. It is a matter of record that a play
which failed to please the public in its author’s lifetime never
succeeded later in establishing itself on the stage. Partizans may
prate about the dramatic power of the ‘Blot in the ’Scutcheon,’ but
when it is—as it has been half-a-dozen times—galvanized into a
semblance of life for a night or a fortnight, it falls prone in the
playhouse as dead as it was when Macready first officiated at its
funeral. Even the ‘Misanthrope,’ mightiest of Molière’s comedies and
worthy of all the acclaim it has received, was not an outstanding
triumph when its author impersonated Alceste, and it has rarely
rewarded the efforts of the succession of accomplished actors who
have tried to follow the footsteps of the master; it is praised, it is
admired; but it does not attract the many to the theater, because it
does not give them abundantly the special pleasure that only the
theater can bestow. ‘Tartuffe’ and the ‘Femmes Savantes’ do this and
also half-a-score of Molière’s lighter and less ambitious pieces,
supported by stories more theatrically effective than that of the
‘Misanthrope.’
The playwright who is merely a clever craftsman of the stage has
no higher aim than to win the suffrages of his contemporaries. He
knows what they want—for he is one of them—and he gives them
what they want, no more and no less. He does not put himself into
his plays; and perhaps his plays would be little better if he did. He is
strenuously and insistently “up to date,” as the phrase is; and as a
result he is soon “out of date.” He writes to be in the fashion; and
the more completely he portrays the fleeting modes of the moment,
the more swiftly must he fall out of fashion. The taste of the day is
never the taste of after days; and the journalist-dramatist buys his
evanescent popularity at a price. Who now is so poor as to pay
reverence to Kotzebue and to Scribe, who once had all the managers
at their feet? No maker of plays, not Lope de Vega or Dumas—
Alexander the Great—was more fertile than Scribe in the invention of
effective situations, none was ever more dextrous in the knotting
and unknotting of plots, grave and gay. But his fertility and his
dexterity have availed him little. He wrote for his own time, not for
all time. What sprang up in the morning of his career and bloomed
brightly in the sunshine, was by night-fall drooping and withered and
desiccated.
The comic dramatists of the Restoration had perforce to gratify the
lewd likings of vicious spectators who wanted to see themselves on
the stage even more vicious than they were. Congreve and Wycherly
put into their comedies what their contemporaries relished, a game
flavor that stank in the nostrils of all decent folk. The Puritan shrank
with horror from the picture in which the Impuritan recognized his
own image. So it was that a scant hundred years after they had
insulted the moral sense (which, like Truth, tho “crushed to earth will
rise again; the eternal years of God are hers,”) they were swept from
the stage. What had delighted under Charles II disgusted under
George IV.
Even the frequent attempt to deodorize them failed, for, as
Sheridan said—and he knew by experience since he had made his
‘Trip to Scarborough’ out of the ‘Relapse’—the Restoration comedies
were “like horses; you rob them of their vice and you rob them of
their vigor.” Charles Lamb, who had a whimsical predilection for
them, admitted that they were “quite extinct on our stage.”
Congreve’s pistol no longer discharged its steel bullets; and Wycherly
no longer knocked his victims down with the butt of his gun. Yet
they died hard; I am old enough to have seen Daly’s company in the
‘Trip to Scarborough’ and the ‘Recruiting Officer,’ in the ‘Inconstant,’
in ‘She Would and She Would Not’ and the ‘Country Girl’ (Garrick’s
skilful cleansing of Wycherly’s unspeakable ‘Country Wife’)—all of
which reappeared because they had appealing plots, amusing
situations and lively characters and because they did not portray the
immorals of the days of Nell Gwyn.
Yet when an adroit playwright who seeks to please the public of
his own time by the representation of its manners, happens to be
also a creative artist, enamored of life, he is sometimes able so to
vitalize his satire of a passing vogue that it has abiding vigor. This is
what Molière did when he made fun of the ‘Précieuses Ridicules.’
Even when he was writing this cleverest of skits, the cotery which
had clustered around Madame de Rambouillet was disintegrating and
would have disappeared without his bold blows. But affectation is
undying; it assumes new shapes; it is always a tempting target; and
Molière, by the magic of his genius, transcended his immediate
purpose. He composed a satire of one special manifestation of
pretence which survives after two centuries and a half as an
adequate satire of all later manifestations. The Précieuses in Paris
have long since been gathered to their mothers; so have the
Esthetes across the channel in London; and soon they will be
followed to the grave by the Little Groups of Serious Thinkers who
are to-day settling the problems of the cosmos by the aid of empty
phrases. No one sees the ‘Précieuses Ridicules’ to-day without
recognizing that it is almost as fresh as it was when Madame de
Rambouillet enjoyed it.
The man of genius is able to please his own generation by his
depiction of its foibles and yet to put into his work the permanent
qualities which make it pleasing to the generations that come after
him. The trick may not be easy, but it can be turned. How it shall be
done,—well, that is one of the secrets of genius. In the case of the
‘Précieuses Ridicules’ we can see that Molière framed a plot for his
lively little piece that is perennially pleasing, a plot which only a little
modified was to support two popular successes nearly two centuries
later,—the ‘Ruy Blas’ of Victor Hugo and the ‘Lady of Lyons’ of
Bulwer-Lytton. He tinged his dialog with just enough timeliness to hit
the taste of the town in 1658; and he did not so surcharge it as to
fatigue the playgoers of Paris two centuries and a half later.

II

The likings of the groundlings who stood in the yard of the Globe
theater when Shakspere began to write plays were coarser and
grosser than those of the burghers whom Molière had to attract to
the Petit-Bourbon; and unfortunately Shakspere in his earlier efforts
was not as cautious as Molière. In the Falstaff plays, for example,
the fat knight is as alive to-day as when Elizabeth is fabled to have
expressed the wish to have him shown in love. But the talk of his
companions, Nym and Pistol, is too thickly bespangled with the tricks
of speech of Elizabethan London to interest American and British
theater-goers three hundred years later. There is but a faded appeal
in topical allusions which need to be explained before they are
appreciated and even before they are understood; and in the
playhouse itself footnotes are impossible.
In his earliest pieces, written during his arduous apprenticeship to
the craft of playmaking, when he was not yet sure of his footing in
the theater, Shakspere had to provide parts for a pair of popular fun-
makers,—Will Kempe and another as yet unidentified. They were
lusty and robust comedians accustomed to set the house in a roar as
soon as they showed their cheerful faces. They created the two
Dromios, the two Gobbos, Launce and Speed, Costard and Dull; and
it is idle to deny that not a little of the talk that Shakspere put in
their mouths is no longer laughter-provoking; it is not only too
topical, too deliberately Tudor, it is also too mechanical in its effort at
humor to move us to mirth to-day. Their merry jests,—Heaven save
the mark!—are not lifted above the level of the patter of the
“sidewalk comedians” of our variety-shows. They are frankly
“clowns”; and Shakspere has set down for them what the
groundlings expected them to utter, only little better than the rough
repartee and vigorous innuendo and obvious pun which they would
have provided for themselves if they had been free to do as they
were wont to do. What he gives them to say is rarely the utterance
of the characters they were supposed to be interpreting; and this is
because the two Dromios are parts only, are not true characters, and
are scarcely to be accepted even as types.
A difference of taste in jests, so George Eliot declared, is “a great
strain on the affections”; and it would be insulting to the creator of
Bottom and Falstaff to pretend that we have any affectionate regard
for Costard and Dull, for Launce and Speed. It is only when
Shakspere was coming to the end of his apprenticeship that he
found out how to utilize the talents of Kempe and of Kempe’s
unknown comrade in comedy, in parts which without ceasing to be
adjusted to their personalities were also accusable characters,
Dogberry and Touchstone. But when we come to Touchstone we are
forced to perceive that Shakspere was the child of his own age even
when he refrained from echoing its catchwords. He was cleaner than
the majority of his rivals, but he was near enough to Rabelais to be
frank of speech. On occasion he can be of the earth, earthy. He
bestows upon Touchstone a humor which is at times Rabelaisian in
its breadth, in its outspoken plainness of speech, assured of the
guffaws of the riffraff and rabble of a Tudor seaport, but a little too
coarse for the descendants of the Puritans on either side of the
Atlantic to-day. Nearly fifty years ago when Harry Beckett was
rehearsing in ‘As You Like It’ for one of the infrequent Shaksperian
revivals that Lester Wallack ventured to make, he told me
sorrowfully that his part had been sadly shorn, some of Touchstone’s
best lines having been sacrificed in deference to the increasing
squeamishness of American audiences.
These accessory comic parts are not alone in their readjustment to
the modifying moods of a later age. The point of view changes with
every generation, and with every change a character is likely to be
seen from a different angle. No dramatist, whatever his genius, can
foresee the future and forecast the fate of his creatures. The
centuries follow one another in orderly procession, and they are
increasingly unlike. Moreover, the dramatist of genius, by the very
fact that he is a genius, is forever building better than he knew. He
may put a character into a play for a special purpose; and after a
century or two that character will loom larger than its creator dreamt
and will stand forward, refusing to keep the subordinate place for
which it was deliberately designed. We listen to the lines he utters
and we read into them meanings which the author could not have
intended, but which, none the less, are there to be read by us.
We may even accept as tragic a figure whom the playwright
expected to be received as comic and who was so received by the
audience for which the playwright wrote. Sometimes this is a
betrayal of his purpose, as it is when aspiring French actors have
seen fit to represent the Figaro of Beaumarchais (in the ‘Marriage of
Figaro,’ not in the ‘Barber of Seville’) as a violent and virulent
precursor of the French Revolution; or as it is when the same French
actors insist on making the Georges Dandin of Molière a subject for
pity, tear-compelling rather than laughter-provoking.
It is not a betrayal, however, rather is it a transfiguration when the
Shylock of Shakspere is made to arouse our sympathy. I make no
doubt that Shakspere projected Shylock as a comic villain, at whom
he intended the spectators to laugh, even if they also shuddered
because of his bloodthirstiness. Yet by sheer stress of genius this
sinister creature, grotesque as he may be, is drawn with such
compelling veracity that we cannot but feel for him. We are shocked
by the insulting jeers of Gratiano at the moment of his discomfiture.
We are glad that his plot against Antonio has failed; none the less do
we feel that he has been miserably tricked; we are almost ready to
resent the way in which the cards have been stacked against him.
To anyone who has familiarized himself with the attitude of
Elizabethan playgoers toward usurers and toward the Jews, it is
evident that Shakspere intended the ‘Merchant of Venice’ to be a
Portia play; its action begins with talk about Belmont and it ends at
Belmont itself; and Shylock is absent from the final act. In spite of
this intent of the author, the ‘Merchant of Venice’ has become in our
eyes a Shylock play. In fact, Macready four-score years ago used to
appear in a three-act version which ended with the trial scene,—a
most inartistic perversion of the comedy. After all, the ‘Merchant of
Venice’ is a comedy, even if its love-story is sustained and stiffened
by a terrible underplot. Shakspere created the abhorrent Shylock
that the lovely Portia could cleverly circumvent him and score off him
and put him to shame. His hardness of heart was to make more
refulgent her brightness of soul. Shylock was set up to be scorned
and hated and derided; he is a vindictive moneylender, insisting on a
horrible penalty; no one in the play has a good word for him or a
kindly thought; his servant detests him and his daughter has no
natural affection for him.
When all is said, we cannot but feel that Shakspere in his
treatment of Shylock displays a callousness not uncommon in
Elizabethan England. And yet—and yet Shakspere is true to his
genius; he endows Shylock with life. The Jew stands before us and
speaks for himself; and we feel that we understand him better than
the genius who made him. Our sympathy goes out to him; and altho
we do not wish the play to end otherwise than it does, we are
almost ready to regard him as the victim of a miscarriage of justice,
guilty though he is. Ellen Terry has quoted from a letter of Henry
Irving’s a significant confession: “Shylock is a ferocity, I know—but I
cannot play him that way!” Why couldn’t he? It was because the
nineteenth century was not the sixteenth, because Victorian
audiences were not Elizabethan, because the peoples who have
English for their mother-tongue are less callous and more civilized
than their forebears of three hundred years ago.

III

While it is more than three hundred years since Shakspere wrote

the ‘Merchant of Venice,’ it is less than a hundred and fifty since
Sheridan wrote the ‘School for Scandal.’ The gap that yawns
between us and Sheridan is not so wide or so deep as the gulf that
divides us from Shakspere; but it is obvious enough. Even a hundred
years ago Charles Lamb declared that the audiences of his time were
becoming more and more unlike those of Sheridan’s day, and that
this increasing unlikeness was forcing the actors to modify their
methods, a little against their wills. Sheridan’s two brilliant comedies
continue to delight us by their solidity of structure, their vigor of
characterization and their insistent sparkle of dialog. In the ‘Rivals’
Sheridan is following in the footsteps of his fellow Irishman,
Farquhar, and in the ‘School for Scandal’ he is matching himself
against Congreve. In both he was carrying on the tradition of
Restoration comedy, with its coldheartedness, its hard glitter, its
delineation of modes rather than morals. It is perhaps too much to
assert that most of his characters are unfeeling; but it is not too
much to say that they are regardless of the feelings of others—
perhaps because their own emotions are only skin-deep.
It is true that in the ‘Rivals’ Sheridan threw a sop to the admirers
of Sentimental Comedy and introduced a couple of high-strung and
weepful lovers, Falkland and Julia, who are forever sentimentalizing.
But this precious pair have been found so uninteresting that in most
of the later performances of the ‘Rivals’—all too infrequent, alas!—
they have been omitted altogether or disgraced by relegation to the
background.
The vogue of Sentimental Comedy was waning when Sheridan
wrote, and it disappeared before he died, yet the playgoers of
London and of New York were becoming more tender-hearted than
their ancestors who had delighted in the metallic harshness of
character-delineation customary in Restoration comedy. They were
beginning to look for characters with whom they could sympathize
and to desire the villains to remain consistent in their villainy. They
were unwilling to remain in what Lamb termed “the regions of pure
comedy, where no cold moral reigns.” Lamb called the ‘School for
Scandal’ incongruous in that it is “a mixture of sentimental
incompatibilities,” Charles Surface being “a pleasant reality” while
Joseph Surface was “a no less pleasant poetical foil to it.”
 The original performer of Joseph was John Palmer; and Lamb
asserted that it required his consummate art “to reconcile the
discordant elements.” Then the critic suggested, and this was a
century ago, that

a player with Jack’s talents, if we had one now, would not dare do the part in the
same manner. He would instinctively avoid every turn which might tend to
unrealize, and so to make the character fascinating. He must take his cue from the
spectators, who would expect a bad man and a good man as rigidly opposed to
each other as the death-beds of those geniuses are contrasted in the prints.

A little later in the same essay—the incomparable analysis of

‘Artificial Comedy’—Lamb pointed out that “Charles must be loved
and Joseph hated,” adding that

to balance one disagreeable reality with another, Sir Peter Teazle must be no
longer the comic idea of a fretful old bachelor bridegroom, whose teasings (while
King played it) were evidently as much played off at you as they were meant to
concern anybody on the stage,—he must be a real person, capable in law of
sustaining an injury,—a person towards whom duties are to be acknowledged,—
the genuine crim. con. antagonist of the villainous seducer Joseph. To realize him
more, his sufferings under his unfortunate match must have the downright
pungency of life,—must (or should) make you not mirthful but uncomfortable, just
as the same predicament would move you in a neighbor or old friend.

I cannot count the number of occasions on which I have enjoyed

the performance of the ‘School for Scandal,’—but they must amount
to a score at the least. I recall most clearly John Gilbert’s Sir Peter;
and I can testify that he had preserved the tradition of King. He was
the fretful old bachelor bridegroom, who, when the screen fell and
discovered Lady Teazle in the library of Joseph Surface, was
wounded not in his heart but in his vanity. He preserved the comic
idea, as Sheridan had designed. But John Gilbert was the only Sir
Peter I can recall who was able to achieve this histrionic feat.
Of all the many Lady Teazles it has been my good fortune to see,
Fanny Davenport stands out most sharply in my memory,—perhaps
because she was the first I had ever beheld and perhaps because
she was then in the springtime of her buoyant beauty. Certainly
when the screen fell she was a lovely picture, like Niobe all tears.
Her repentance was sincere beyond all question. She renounced the
comic idea, which is that Lady Teazle has been caught in a
compromising situation by the elderly husband with whom she is in
the habit of quarrelling. Fanny Davenport saw only the pathos of the
situation; and she made us see it and feel it and feel for her and
hope that her impossible husband would accept her honest
explanation,—the explanation which indeed he would have to accept
since we as eye-witnesses are ready to testify that it is the truth, the
whole truth and nothing but the truth.
But this rendering of the part is discomposing to the comic idea;
and it forces a modification of method upon the actor of Charles
Surface. It is in deference to the comic idea that when the screen
falls Sheridan made Charles see the humor of the situation and only
the humor of it. He is called upon to chaff Sir Peter and Lady Teazle
and Joseph, one after the other. If the actor speaks these lines with
due regard to the comic idea which created Sir Peter as a peevish
old bachelor bridegroom and Lady Teazle as a frivolous woman of
fashion, and if the actor of Sir Peter and the actress of Lady Teazle
take the situation not only seriously but pathetically as they would in
a twentieth century problem-play, then Charles’s speech is heartless
and almost brutal. Now Charles is a character as sympathetic to the
audience in his way as Lady Teazle is in hers. Charles is to be loved
as Joseph is to be hated. And so the impersonator of Charles is
compelled to modify his method, to transpose his lines and to
recognize that the robust raillery natural to him and appropriate to
the predicament must be toned down in deference to our more
delicate susceptibilities.
He laughs at Sir Peter first; and then he turns to Joseph, who is
fair game and whom the spectators are glad to see held up to scorn.
He says “you seem all to have been diverting yourselves here at hide
and seek and I don’t see who is out of the secret.” With this he turns
to Lady Teazle and asks, “Shall I beg your ladyship to inform me?”
So saying he looks at her and perceiving that she is standing silent
and ashamed, with downcast eyes, he makes her a bow of apology
for his levity. Finally with another thrust at his brother, the unmasked
hypocrite, he takes his departure airily, leaving them face to face. If
the comic idea suffers from this contradiction of the intent of the
comic dramatist, it must find what consolation it can in its sense of
humor.

IV

A large share of the success of even the masterpieces of the

drama, comic and tragic, is due to the coincidence of its theme and
its treatment with the desires, the opinions and the prejudices of the
contemporary audiences for whose pleasure it was originally
planned. But the play, comic or tragic, as the case may be, can
survive through the ages (as the ‘Merchant of Venice’ and the
‘School for Scandal’ have survived) only if this compliance has not
been subservient, if the play has the solidity of structure and the
universality of topic which will win it a welcome after its author is
dead and gone. What is contemporary is three parts temporary, and
what is up-to-date is certain soon to be out-of-date. Nevertheless it
is always the audience of his own time and of his own place that the
playwright has to please, first of all; and if their verdict is against
him he has lost his case. Plays have their fates no less than books;
and the dispensers of these fates are the spectators assembled in
the playhouse. The dramatist who ignores this fact, or who is
ignorant of it, does so at his peril. As Lowell once put it with his
wonted pungency, “the pressure of public opinion is like the pressure
of the atmosphere; you cannot see it, but it is sixteen pounds to the
square inch all the same.”
(1921)
 IV
TRAGEDIES WITH HAPPY ENDINGS
 IV
TRAGEDIES WITH HAPPY ENDINGS

In Mrs. Wharton’s acute and often penetrating analysis of ‘French

Ways and Their Meaning,’ she dwelt upon the innate intellectual
honesty of the French, “the special distinction of the race, which
makes it the torch-bearer of the world”; and she asserted that
Bishop Butler’s celebrated declaration, “Things are what they are and
will be as they will be,” might have been “the motto of the French
intellect.” She called it “an axiom that makes dull minds droop, but
exalts the brain imaginative enough to be amazed before the marvel
of things as they are.”
She pointed out that in Paris the people who go to the moving-
pictures to gaze at an empty and external panorama are also the
people who flock to the state-subventioned theaters, the Français
and the Odéon, to behold the searching tragedies of Corneille and
Racine, immitigably veracious in the portrayal of life as it is on the
lofty plane of poetry:

The people who assist at these grand tragic performances have a strong enough
sense of reality to understand the part that grief and calamity play in life and in
art; they feel instinctively that no real art can be based on a humbugging attitude
toward life, and it is their intellectual honesty which makes them exact and enjoy
its fearless representation.

This intellectual honesty Mrs. Wharton failed to find in the

audiences of our American theaters, because it is not a habitual
possession of Americans generally. And she ventured to quote a
remark which she once heard Howells make on our theatrical taste.
They had been talking about the pressure exerted upon the
American playwright by the American playgoing public, compelling
him to wind up his play, whatever its point of departure, with the
suggestion that his hero and heroine lived happily ever after, like the
prince and princess who are married off at the end of the fairy-tale.
Mrs. Wharton declared that this predilection of our playgoers did not
imply a preference for comedy, but that, on the contrary, “our
audience wanted to be harrowed (and even slightly shocked) from
eight till ten-thirty, and then consoled and reassured before eleven.”
“Yes,” said Howells, “what the American public wants is a tragedy
—with a happy ending.”
And Mrs. Wharton added her own comment that what Howells
said of the American attitude in the theater “is true of the whole
American attitude toward life.” In other words we Americans both in
the playhouse and out of it are lacking in the intellectual honesty
which the French possess. We are not convinced, and we are not
willing to let our plays, and even our novels, convince us, that
“things are what they are and will be as they will be.”
With the praise that Mrs. Wharton bestowed upon the French, no
one who has profited by the masterpieces of French literature could
cavil for a moment. The French are intellectually honest, more so
than any other modern nation, and perhaps as much so as the
Greeks. There is abundant insincerity in our drama and in our fiction;
and no one long familiar with either is justified in denying this. But,
none the less, Howells’s characteristically witty remark has not
perhaps all the weight which Mrs. Wharton attached to it. And it
instantly evokes the desire to ask questions. Is it really true that we
Americans like tragedies with happy endings? And, supposing this to
be true, are we the only people who have ever revealed this
aberration? Finally, if we have revealed it, are there any special
reasons for this manifestation of our deficiency in intellectual
honesty?
 Having propounded these three queries, I propose to answer them
myself as best I can, and as the farseeing reader probably expected
me to do; and it appears to me prudent to commence by considering
the second of them, leaving the first to be taken up immediately
thereafter. Are we Americans the only people who like tragedies with
happy endings? Here we have a starting point for a discursive
inquiry into the tastes of the playgoing public in other countries and
in other centuries. Nor need we begin this leisurely loitering by too
long a voyage, for we have only to go back a hundred years, more
or less, and to tarry a little while in France itself.

II

It was in the minor theatres of Paris at the end of the eighteenth

century and at the beginning of the nineteenth that there was slowly
developed a new type of play, the melodrama. Its first masters were
Ducange and Pixérécourt, who had profited by the experience of
their ruder forerunners and who taught the secrets of their special
craft to their more expert followers, the fertile Bouchardy, for one,
and, for another, the only lately departed Dennery, the most adroit
and the most inventive of them all.
A melodrama may be described briefly as a play with a plot and
nothing but a plot; it abounds in situations enthralling, intricately
combined and adroitly presented; and it contains characters
simplified to types, drawn in profile and violently stencilled with the
primary colors. It has a Hero, who struggles against his fate and
struggles in vain until the final episode, when the Villain, as black as
he is painted, is cast into outer darkness, the entirely white Hero
being then rewarded for all his sufferings and for all his struggles
with the hand of the equally pale Heroine, truly the female of his
species. The melodrama may be devoid of veracity, but it is
compelling in its progressive interest. It is dextrously devised to
delight audiences which want “to be harrowed (and even slightly
shocked) from eight to ten-thirty, and then consoled and reassured
before eleven.” In short, it is “a tragedy with a happy ending.”
 What could be more tragic than the tale of the ‘Two Orphans’? In
that ultimate masterpiece of melodrama, two lovely sisters, one of
them blind, are severally lost in Paris in the wickedest days of the
Regency. We are made to follow their appalling misadventures; and
we behold them again and again in danger of death and worse than
death. The sword of Damocles was suspended over their fair heads
from the first rising of the curtain until within five minutes of its final
fall. The odds are a hundred to one, nay, a thousand to one, against
their escaping unscathed from their manifold perils. And yet,
nevertheless, at the very end, the clouds lift, sunshine floods the
stage; and the two heroines are left at last to live happy like two
princesses with their two princes in the most entrancing of fairy
tales. And many thousand Parisian audiences, laying aside their
intellectual honesty for the occasion, dilated with the right emotion,
sobbed at the sorrows of the sisters, cheered the rescuers and
venomously hissed the villains who had pursued them.
So completely were the playgoers of Paris subdued to what they
worked in, that the makers of melodrama were emboldened to
strange tricks. Théophile Gautier once described a long-forgotten
melodrama by Bouchardy, himself long forgotten, in which an
important character was killed off in the third act. Then in the fifth
act when the unfortunate but immaculate Hero was absolutely at the
mercy of his vicious enemies, and when he could extricate himself
from the toils only if he had the talisman he had been seeking in
vain,—the needed password, the necessary key, the missing will, the
incriminatory documents, or whatever you prefer—when all is lost,
even honor, then in the very nick of time, the character who was
killed off in the third act, and dead beyond all question, reappears
and gives the Hero the talisman (whatever it was). The Hero
receives this with joy, commingled with surprize. “I thought you
were dead!” he cries; “how is it that you are here now?” “Ah,”
answers the traveller from beyond, “that—that is a secret that I
must carry back with me into the tomb!”
It is only fair to record that Parisian melodrama was not often as
rude and as crude as this in its subterfuges and its expedients.
Indeed, it sometimes rose to a far higher level, as in the ‘Don César
de Bazan’ of Dennery and in the ‘Lyons Mail’ of Moreau, Giraudin and
Delacour. It even served as the model for Victor Hugo’s superb and
sonorous ‘Hernani’ and ‘Ruy Blas,’ in which he flung the rich
embroidered mantle of his ample lyricism over an arbitrary skeleton
of deftly articulated intrigue, as artificial as it was ingenious.
In its earlier manifestations it was imitated in Great Britain,
notably by Edward Fitzball, the first playmaker who perceived the
theatrical possibilities of the legend of the ‘Flying Dutchman.’ Fitzball
did not disdain to intimate that he considered himself the “Victor
Hugo of England,”—which tempted Douglas Jerrold to remark that
Fitzball was really only the “Victor No Go” of England. In its later
manifestations the melodrama of the French supplied a pattern for
the ‘Silver King’ of Henry Arthur Jones, one of the most satisfactory
specimens of this type of play. The ‘Silver King’ won the high
approval of Matthew Arnold, who called it an honest melodrama,
relying necessarily “for its main effect on an outer drama of
sensational incidents” and none the less attaining the level of
literature because the dialog and the sentiments were natural.
By the side of the British ‘Silver King’ of Henry Arthur Jones may
be set the American ‘Secret Service’ of William Gillette, which also
relies for its main effect on an outer drama of sensational incidents;
and yet the sensational incidents are so fitly chosen and so artfully
interwoven that they serve to set off the very human hero, an
accusable character, a Union spy, with a divided duty before him.
Toward the end of the play it becomes evident that this brave and
resourceful man is doomed to death; and to this fatality he is himself
resigned, wilfully throwing away a chance to escape and welcoming
a speedy exit from his impossible position. Yet, once more, just
before the curtain falls, the dramatist intervenes, like a god from the
machine, sparing his hero’s life, and even permitting the spectators
to foresee that hero and heroine will live happily ever after, thus
consoling and reassuring the audience before eleven o’clock.
 I make bold to say that this happy ending is not inartistic and that
it does not outrage our intellectual honesty, for the obvious reason
that ‘Secret Service’ is not essentially a tragedy; it is a serio-comic
story which never uplifts us to the serene atmosphere of the
irresistible and the inevitable in which tragedy lives. It is too brisk in
its humor, too lively in its representation of the externalities of life, to
justify a fatal conclusion. A true tragedy must not only end sadly, it
has also to begin sadly; it has to impress us subtly with a sense of
impending disaster, inherent in its theme. What Stevenson said of
the short-story, when that is as dramatic as it can be, is applicable to
the drama itself. “Make another end to it?” he wrote in answer to a
suggestion to that effect. “Ah, yes, but that is not the way I write;
the whole tale is unified. To make another end, that is to make the
beginning all wrong.... The body and end of a short-story is bone of
the bone and blood of the blood of the beginning.” In other words
the beginning of a melodrama never demands a tragic ending, and
rarely even permits it.

III

Altho modern melodrama was developed in the totally unliterary

minor playhouses of Paris more than a hundred years ago, the
playgoers of France had not had to wait until the early nineteenth
century or even until the early eighteenth to be consoled and
reassured by a tragedy with a happy ending. It was in the first half
of the seventeenth century that Corneille took over from a Spanish
original the first and fieriest of his tragedies, the ‘Cid,’—the story of
which leads up to one of the strongest situations in all dramatic
literature. The duty is suddenly laid upon a high-strung warrior to
fight a duel to the death with the father of the woman whom he
loves and who loves him. Seemingly the deadly stroke of his sword
has severed the lovers forever, for how could a woman wed the red-
handed slayer of her father? Yet it is with this prospective wedding,
abruptly brought about, that Corneille ends his play; and he was so
dextrous a dramatist, so abundant in emotion and so persuasive in
eloquence, that he was able to carry his audience with him, even at
the cost of their intellectual honesty.
Nor did the playgoers of England have to await the importation of
French melodrama in the original package before they could enjoy
reassurance and consolation after being harrowed and even slightly
shocked. Indeed, the Londoners had this pleasure provided for them
even earlier than it had been vouchsafed to the Parisians. All
students of the history of our stage are familiar with the type of play
known as tragi-comedy. Its name sufficiently describes it, a name
apparently first used in the prolog to a play by Plautus and revived
by the Italian theorists of the theater. Dramas of this species sprang
up spontaneously in Italy, in Spain and in France; and we find the
form flourishing in England in the second half of the sixteenth
century, altho it cannot be said to have been more popular among
the English than it was among the French. Shakspere’s somber
‘Measure for Measure’ is the most immediately obvious example; and
at the performance of this play the spectators were harrowed, and
even more than slightly shocked, by a succession of powerful
situations, only to be at last reassured and consoled by a happy
ending, mechanically and unconvincingly brought about.
In the course of time tragi-comedy modified its methods and
became the dramatic-romance, of which Beaumont and Fletcher’s
‘Philaster’ may be taken as one characteristic specimen and
Shakspere’s ‘Cymbeline’ as another. Perhaps it would be more exact
to say that the dramatic-romance is only an insular sub-species of
sentimental tragi-comedy. Most of the best known of the dramatic-
romances of Beaumont and Fletcher (or of Fletcher and Massinger)
conform to the definition of tragi-comedy, as Professor Ristine has
skilfully condensed this from a defence of the type made by Guarini,
author of the ‘Pastor Fido’:

Tragi-comedy, far from being a discordant mixture of tragedy and comedy, is a

thorough blend of such parts of each as can stand together with verisimilitude,
with the result that the deaths of tragedy are reduced to the danger of death, and
the whole in every respect a graduated mean between the austerity and the
dignity of the one and the pleasantness and ease of the other.
 This Italian definition of Renascence tragi-comedy can be
transferred to modern melodrama of the more literary kind,—the
‘Silver King,’ for example, and ‘Secret Service,’ in which we find the
graduated mean between austere dignity and easy pleasantness.
After quoting from Guarini, Professor Ristine gave his own analysis
of the elements combined in English tragi-comedy:

Love of some sort is the motive force; intrigue is rife; the darkest villainy is
contrasted with the noblest and most exalted virtue. In the course of an action ...
in which the characters are enmeshed in a web of disastrous complications,
reverse and surprise succeed each other with lightning rapidity.... But final disaster
is ingeniously averted.... Wrongs are righted, reconciliation sets in, penitent villainy
is forgiven, and the happy ending made complete.

In its turn this American description of English tragi-comedy is

applicable also to French melodrama of the less literary kind,—the
‘Lyons Mail’ and the ‘Two Orphans.’
It is possible to find at least one tragedy with a happy ending
amid the two score plays which alone have come down to us from all
the hundreds acted in the Theater of Dionysus before the assembled
citizens of Athens,—probably the most intelligent body of playgoers
to which any dramatist has ever been privileged to appeal. The
‘Alcestis’ of Euripides is a beautiful play, grave, inspiring and moving;
yet it has been a constant puzzle to the historians of Greek
literature, who have never been quite able to declare what manner
of tragic drama it is, since it has one character who is frankly
humorous and since it has a happy ending,—the revivification of the
pathetic heroine who had given her life to save her husband and
who is brought back by Hercules, after a combat with Death.

IV

After this desultory ramble through the history of the drama in

other centuries and in other countries, we are in better case to
consider the first of the three questions suggested by Mrs. Wharton’s
assertion that we Americans are deficient in the intellectual honesty
which is a recognized characteristic of the French. Is it really true
that we like tragedies with happy endings? If it is true, we are no
worse off than the English in the time of Shakspere, the French in
the time of Corneille and in the time of Hugo, the Greeks in the time
of Euripides. But is it true?
It might be urged in our defence that we do not in the least object
to the death of the hero and the heroine (or of both together) in the
music-drama; and it must be admitted that in serious opera a tragic
ending is not only acceptable but is actually expected. It might be
pointed out that the final death of the heroine has never in any way
interfered with the immense popularity of a host of star-plays,
‘Adrienne Lecouvreur,’ the ‘Dame aux Camélias,’ ‘Froufrou,’ ‘Théodora’
and ‘La Tosca.’ It might be permissible to record that the death of
‘Cyrano de Bergerac,’ (a fatal termination not inherent in the theme
of that heroic comedy and in fact almost inconsistent,) did not
dampen the pleasure of the American playgoer.
These things must be taken for what they are worth; and perhaps
they are not really pertinent to our immediate inquiry, since opera is
a very special form of the dramatic art, making an appeal of its own
within arbitrary limits, and since a star-play is relished by the
majority largely as a vehicle for the exhibition of the histrionic
versatility of the star herself or himself, a last dying speech and
confession affording the performer an excellent opportunity for the
display of his or her virtuosity.
We must go behind Mrs. Wharton’s rather too sweeping
accusation and center attention on a single point. American
playgoers of to-day enjoy and hugely enjoy seeing on the stage
stories which are harrowing, which deal liberally with life and death,
and which after all end happily, sending us home consoled and
reassured. So have the playgoers of other lands in other times; and
the real question is whether we refuse to accept the tragic end when
this is ordained by all that has gone before, when it is a fate not to
be escaped. In other words, have we the intellectual honesty which
shall compel us to accept George Eliot’s stern declaration that
“consequences are unpitying”?
Thus put, the question is not easy to answer.
For myself I am inclined to think that when we are at liberty to
choose between the happy and the unhappy ending, when one or
the other is not imposed upon us by the action or by the atmosphere
of the story set before us, we tend to prefer a conclusion which
dismisses the hero and the heroine to a vague future felicity. But I
am inclined also to believe that we do not shrink from the bitterest
end if this has been foreordained from the beginning of time, if the
author has been skilful enough and sincere enough to make us feel
that his tragedy could not possibly have any other than a tragic
termination.
In the ‘Second Mrs. Tanqueray’ the fatal ending is obligatory; it
grows out of the nature of things; and the play has established itself.
In ‘Mid-Channel’ there is no way out of the difficulty in which the
heroine has entangled herself, except through the door of death. On
the other hand, the plot of the ‘Notorious Mrs. Ebbsmith’ cried aloud
for a tragic ending, which the author refused to grant; and perhaps
this is one reason why the piece has never taken hold on our
playgoing public despite its indisputable qualities.
As it happens there have been seen on our stage in the first and
second decades of the twentieth century four plays, unequal in
sincerity and different in texture, but all of them variants of the
same theme. Two are British, ‘Iris’ by Sir Arthur Pinero, and the
‘Fugitive’ by John Galsworthy; and two are American, the ‘Easiest
Way’ by Eugene Walter and ‘Déclassée’ by Miss Zoe Akins. In each of
them we are invited to follow the career of a young woman who
loves luxury and who moves through life along the line of least
resistance, until at last the ground gives way beneath her feet. ‘Iris’
was the first of the four; it is the most delicately artistic and the
most veracious. The ‘Easiest Way’ is perhaps the most vigorous. The
‘Fugitive’ is pallid and futile. ‘Déclassée’ is the least important of
them all, as it is the least original. The two last-named pieces are
unsatisfactory when we bring them to the bar of our intellectual
honesty; and yet they both end with the death of the heroine, an
arbitrary exit out of the moral entanglements in which she has
involved herself. The two earlier plays have a more truly tragic
ending, since they leave the heroine alive yet bereft of all that
makes life worth living. No one of the four sent the spectators home
reassured and consoled.

There might seem to be no necessity to put the third question

now that the second has been discussed. And yet there may be
profit in asking ourselves whether there are any special reasons why
the American playgoing public might be expected to lapse from
intellectual honesty and to compel our playwrights to violate the
logic of their stories and to stultify themselves to achieve a puerile
fairy-tale conclusion. Mrs. Wharton put forward one such reason,
when she asserted that our attitude in the theater is characteristic
“of the whole American attitude toward life.” Here she is drawing an
indictment against the American people and not merely against
American playgoers.
To enter upon that broad problem would take me too far afield,
too far, that is, from the theater itself, within the walls of which this
inquiry must be confined. Are there any conditions in the American
theater which make against the sincere and searching portrayal of
life? I must confess that I think there is at least one such condition,
the possible consequences of which are disquieting. This is the
change in the composition of the audiences in our American theaters
from what they were half-a-century ago—which is as far back as my
own memories as a playgoer extend. I think that the average age of
the spectators is now considerably less than it was when I was a
play-struck boy; and I think also that the proportion of women is
distinctly larger than it was in those distant days. If I am right in
believing that this change has taken place, and also in anticipating
that it is likely to be even more evident in the years that are to