Advanced Statistics in Criminology and Criminal Justice 5th Edition David Weisburd David B Wilson Alese Wooditch Chester Britt
Advanced Statistics in Criminology and Criminal Justice 5th Edition David Weisburd David B Wilson Alese Wooditch Chester Britt
com
https://ebookmeta.com/product/advanced-statistics-in-
criminology-and-criminal-justice-5th-edition-david-weisburd-
david-b-wilson-alese-wooditch-chester-britt-2/
OR CLICK BUTTON
DOWNLOAD NOW
https://ebookmeta.com/product/industrial-microbiology-1st-edition-
david-b-wilson-editor/
ebookmeta.com
Sylver And Gold Book I The Sylver And Gold Series Michelle
Larkin
https://ebookmeta.com/product/sylver-and-gold-book-i-the-sylver-and-
gold-series-michelle-larkin/
ebookmeta.com
This Will Not Pass: Trump, Biden, and the Battle for
America's Future First Edition Jonathan Martin
https://ebookmeta.com/product/this-will-not-pass-trump-biden-and-the-
battle-for-americas-future-first-edition-jonathan-martin/
ebookmeta.com
https://ebookmeta.com/product/soviet-motor-gunboats-of-world-war-ii-
the-red-army-s-river-tanks-from-stalingrad-to-berlin-przemyslaw-
budzbon/
ebookmeta.com
https://ebookmeta.com/product/functional-polymers-for-controlled-drug-
release-umile-gianfranco-spizzirri-editor/
ebookmeta.com
https://ebookmeta.com/product/computer-graphics-programming-in-opengl-
with-c-2nd-edition-v-scott-gordon/
ebookmeta.com
Death on the Water 1st Edition Cj Birch
https://ebookmeta.com/product/death-on-the-water-1st-edition-cj-birch/
ebookmeta.com
David Weisburd
David B. Wilson
Alese Wooditch
Chester Britt
Advanced
Statistics in
Criminology and
Criminal Justice
Fifth Edition
Advanced Statistics in Criminology
and Criminal Justice
Advanced Statistics
in Criminology
and Criminal Justice
Fifth Edition
David Weisburd
Department of Criminology, Law and Society, George Mason
University, Fairfax, VA, USA and Institute of Criminology,
Faculty of Law, Hebrew University of Jerusalem, Jerusalem, Israel
David B. Wilson
Department of Criminology, Law and Society, George Mason
University, Fairfax, VA, USA
Alese Wooditch
Department of Criminal Justice, Temple University, Philadelphia,
PA, USA
and
Chester Britt
Department of Sociology, Iowa State University, Ames, IA, USA
David Weisburd David B. Wilson
Department of Criminology, Law and Society Department of Criminology, Law and Society
George Mason University George Mason University
Fairfax, VA, USA Fairfax, VA, USA
Institute of Criminology Chester Britt (deceased)
Faculty of Law Department of Sociology
Hebrew University of Jerusalem Iowa State University
Jerusalem, Israel Ames, IA, USA
Alese Wooditch
Department of Criminal Justice
Temple University
Philadelphia, PA, USA
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Contents
Chapter one
Introduction 1
Proportionality Review and the Supreme Court of New Jersey: A Cautionary Tale 3
Generalized Linear Models 7
Special Topics 13
References 14
Chapter two
Multiple Regression 15
Overview of Simple Regression 17
Extending Simple Regression to Multiple Regression 23
Assumptions of Multiple Regression 27
Measurement Error in the Independent Variables 32
Regression Diagnostics 33
Dealing with Outliers and Influential Cases 38
Testing the Significance of Individual Regression Coefficients 40
Assessing Overall Model Fit and Comparing Nested Models 41
Comparing Regression Coefficients Within a Single Model: The Standardized Regression Coefficient 46
Correctly Specifying the Regression Model 48
Model Specification and Building 50
An Example of a Multiple Regression Model 53
Chapter Summary 59
Key Terms 60
Symbols and Formulas 61
Exercises 63
Computer Exercises 66
References 72
v
vi CONTENTS
Chapter three
Multiple Regression: Additional Topics 73
Nominal Variables with Three or More Categories in Multiple Regression 76
Nonlinear Relationships 80
Interaction Effects 92
An Example: Race and Punishment Severity 96
An Example: Punishment Severity 105
The Problem of Multicollinearity 109
Chapter Summary 112
Key Terms 113
Symbols and Formulas 113
Exercises 114
Computer Exercises 118
References 126
Chapter four
Logistic Regression 127
Why Is It Inappropriate to Use OLS Regression for a Dichotomous Dependent Variable? 130
Logistic Regression 136
A Substantive Example: Adoption of Compstat in U.S. Police Agencies 146
Interpreting Logistic Regression Coefficients 151
Comparing Logistic Regression Coefficients 158
Evaluating the Logistic Regression Model 166
Statistical Significance in Logistic Regression 169
Chapter Summary 173
Key Terms 175
Symbols and Formulas 176
Exercises 178
Computer Exercises 181
References 185
Chapter five
Multiple Regression with Multiple Category Nominal or Ordinal
Measures 187
Multinomial Logistic Regression 190
Ordinal Logistic Regression 205
Chapter Summary 219
Key Terms 220
Formulas 221
Exercises 222
Computer Exercises 225
References 231
CO NTENTS vii
Chapter six
Count-Based Regression Models 233
The Poisson Distribution 236
Poisson Regression 239
Over-Dispersion in Count Data 249
Quasi-Poisson and Negative Binomial Regression 251
Zero-Inflated Poisson and Negative Binomial Regression 255
Chapter Summary 259
Key Terms 260
Symbols and Formulas 261
Exercises 262
Computer Exercises 263
References 271
Chapter seven
Multilevel Regression Models 273
A Simple Multilevel Model 277
Random Intercept Model with Fixed Slopes 287
Random Coefficient Model 295
Adding Cluster (Level 2) Characteristics 300
Chapter Summary 309
Key Terms 310
Symbols and Formulas 311
Exercises 312
Computer Exercises 315
References 319
Chapter eight
Statistical Power 321
Statistical Power 323
Components of Statistical Power 326
Estimating Statistical Power and Sample Size for a Statistically Powerful Study 335
Summing Up: Avoiding Studies Designed for Failure 346
Chapter Summary 347
Key Terms 348
Symbols and Formulas 348
Computer Exercises 349
References 365
Chapter nine
Randomized Experiments 367
The Structure of a Randomized Experiment 368
The Main Advantage of Experiments: Isolating Causal Effects 371
viii CONTENTS
Chapter ten
Propensity Score Matching 417
The Underlying Logic Behind Propensity Score Matching 419
Selection of Model for Predicting Propensity for Treatment 421
Matching Methods 422
Assessing the Quality of the Matches 427
Sensitivity Analysis for Average Treatment Effects 431
Limitations of Propensity Score Matching 433
Chapter Summary 435
Key Terms 436
Symbols and Formulas 437
Exercises 437
Computer Exercises 438
References 448
Chapter eleven
Meta-analysis 451
A Historical Note 454
The Logic of Meta-analysis 455
The Effect Size 456
Meta-analysis of Effect Sizes 467
Forest Plots 474
Moderator Analysis 475
Handling Statistically Dependent Effect Sizes: Robust Standard Errors 480
Publication Selection Bias 482
Chapter Summary 485
Key Terms 486
Symbols and Formulas 486
Exercises 490
Computer Exercises 491
References 496
CO NTENTS ix
Chapter twelve
Spatial Regression 499
Why Can’t We Use OLS Regression with Spatial Data? 501
How Do We Define Spatial Relationships? 502
What Is Spatial Regression? 510
Which Type of Spatial Regression Should I Use? 514
Spatial Regression Example 518
Chapter Summary 523
Key Terms 524
Symbols and Formulas 525
Exercises 526
Computer Exercises 528
References 535
Glossary 537
Index 543
C h a p t e r o n e
Introduction
Special Topics
2
PR O PO R TIO N ALITY R EVIEW AND THE SUPREME COURT . . . 3
1
We draw heavily in this section from Weisburd (2001).
2
See Bienen (1996).
4 CHAPTER ONE: INTRO DUCTION
the question. Logistic analyses run in SAS would not converge. (Baldus and New
Jersey Administrative Office of the Courts 1991b, p. 11)
Put in lay terms, there were too few cases and too many variables for a
logistic regression to be estimated. In statistical terms, as we note in Chap. 4,
the models estimated could not converge, or reach a single statistical
solution. As we describe in Chap. 4, lack of convergence is an indicator of
serious problems in the specification of a logistic regression model. Profes-
sor Baldus goes on to explain what was done in order to overcome this
problem:
To deal with this problem we used discriminant analysis, which is capable of
estimating regression coefficients with the same properties as logistic regression
coefficients. Most importantly, discriminant analysis can handle a much larger
number of independent variables. We tested the comparability of the results
from the two procedures with small models that both methods could handle.
The results were comparable, and the discriminant analysis showed no signs of
bias or tendency toward misspecifications. (Baldus and New Jersey Administrative
Office of the Courts 1991b, p. 11)
Faced with the reality that the preferred logistic regression technique
could not provide a statistical solution, Baldus looked to an alternative
approach. He and his colleagues used discriminant functions, which are
not based on maximum likelihood estimates and therefore do not face
problems of convergence (see Chap. 4), as a first step in estimating logistic
regression procedures. Importantly, the use of an alternative estimating
technique did not purge the models of the problems that caused lack of
convergence in the first place. It merely allowed estimation of coefficients
even though such problems were present. Baldus, however, notes that a
series of diagnostic techniques were used that suggested that significant
biases did not develop from taking this alternative approach.
Despite Baldus’ assurances, there are elements of these models that
make them highly suspect, especially for a decision as important as death
penalty sentencing. Many of the coefficients in these models were very
large, sometimes referred to as jumbo coefficients. One of the measures
suggested that the odds of being sentenced to death are 400,000 times
higher for those cases in which a public official is killed. This based on a
sample of only 132 cases.
An additional suggestion of the problems inherent in this approach, and
perhaps a more easily understood indication that the ground was not very
solid, is gained when we examine the 95% confidence intervals around
probability estimates for individual defendants given by Baldus in his report
(Fig. 1.1). A confidence interval in this case provides a basic method for
assessing how stable the estimates gained are for each specific defendant.
The tighter the interval, the more stable the estimate. The larger the interval,
the less confidence we can put in the specific result.
6 CHAPTER ONE: INTRO DUCTION
Predicted probabilities of death sentence and associated confidence interval for 113 penalty
Figure 1.1
trial cases, 1983-1991