This entry-level text offers clear and concise guidelines on how to select, construct, interpret, and evaluate count data. Written for researchers with little or no background in advanced statistics, the book presents treatments of all major models using numerous tables, insets, and detailed modeling suggestions. It begins by demonstrating the fundamentals of modeling count data, including a thorough presentation of the Poisson model. It then works up to an analysis of the problem of overdispersion and of the negative binomial model, and finally to the many variations that can be made to the base count models. Examples in Stata, R, and SAS code enable readers to adapt models for their own purposes, making the text an ideal resource for researchers working in health, ecology, econometrics, transportation, and other fields.
Author is a leading scholar in the field
Written for researchers in a variety of disciplines who have little to no background in modeling count data or advanced statistics
Stata, R, and SAS code provided for examples used throughout; complete with guidelines on how best to use models and software
Joseph M. Hilbe, Arizona State University
Joseph Hilbe is a solar system ambassador with NASA's Jet Propulsion Laboratory, California Institute of Technology; an Adjunct Professor of Statistics at Arizona State University; an Emeritus Professor at the University of Hawaii; and a statistical modeling instructor for Statistics.com, a web-based continuing-education program in statistics. He is the author of several books on statistical modeling and serves as the coordinating editor for the Cambridge University Press series Predictive Analytics in Action.
Author(s): Joseph M. Hilbe
Edition: 1
Publisher: Cambridge University Press
Year: 2014
Language: English
Pages: C, xvi, 283, B
Tags: Математика;Теория вероятностей и математическая статистика;Математическая статистика;Прикладная математическая статистика;
Preface
CHAPTER 1 Varieties of Count Data
SOME POINTS OF DISCUSSION
1.1 WHAT ARE COUNTS?
1.2 UNDERSTANDING A STATISTICAL COUNT MODEL
1.2.1 Basic Structure of a Linear Statistical Model
1.2.2 Models and Probability
1.2.3 Count Models
1.2.4 Structure of a Count Model
1.3 VARIETIES OF COUNT MODELS
1.4 ESTIMATION – THE MODELING PROCESS
1.4.1 Software for Modeling
1.4.2 Maximum Likelihood Estimation
1.4.3 Generalized Linear Models and IRLS Estimation
1.5 SUMMARY
CHAPTER 2 Poisson Regression
SOME POINTS OF DISCUSSION
2.1 POISSON MODEL ASSUMPTIONS
2.2 Apparent Overdispersion
2.3 CONSTRUCTING A “TRUE” POISSON MODEL
2.4 POISSON REGRESSION: MODELING REAL DATA
2.5 INTERPRETING COEFFICIENTS AND RATE RATIOS
2.5.1 How to Interpret a Poisson Coefficient and Associated Statistics
2.5.2 Rate Ratios and Probability
2.6 EXPOSURE: MODELING OVER TIME, AREA, AND SPACE
2.7 PREDICTION
2.8 POISSON MARGINAL EFFECTS
2.8.1 Marginal Effect at the Mean
2.8.2 Average Marginal Effects
2.8.3 Discrete Change or Partial Effects
2.9 SUMMARY
CHAPTER 3 Testing Overdispersion
SOME POINTS OF DISCUSSION
3.1 BASICS OF COUNT MODEL FIT STATISTICS
3.2 OVERDISPERSION: WHAT, WHY, AND HOW
3.3 TESTING OVERDISPERSION
3.3.1 Score Test
3.3.2 Lagrange Multiplier Test
3.3.3 Chi2 Test: Predicted versus Observed Counts
3.4 METHODS OF HANDLING OVERDISPERSION
3.4.1 Scaling Standard Errors: Quasi-count Models
3.4.2 Quasi-likelihood Models
3.4.3 Sandwich or Robust Variance Estimators
3.4.4 Bootstrapped Standard Errors
3.5 SUMMARY
CHAPTER 4 Assessment of Fit
SOME POINTS OF DISCUSSION
4.1 ANALYSIS OF RESIDUAL STATISTICS
4.2 LIKELIHOOD RATIO TEST
4.2.1 Standard Likelihood Ratio Test
4.2.2 Boundary Likelihood Ratio Test
4.3 MODEL SELECTION CRITERIA
4.3.1 Akaike Information Criterion
4.3.2 Bayesian Information Criterion
4.4 SETTING UP AND USING A VALIDATION SAMPLE
4.5 SUMMARY AND AN OVERVIEW OF THE MODELING PROCESS
4.5.1 Summary of What We Have Thus Far Discussed
CHAPTER 5 Negative Binomial Regression
SOME POINTS OF DISCUSSION
5.1 VARIETIES OF NEGATIVE BINOMIAL MODELS
5.2 NEGATIVE BINOMIAL MODEL ASSUMPTIONS
5.2.1 A Word Regarding Parameterization of the Negative Binomial
5.3 TWO MODELING EXAMPLES
5.3.1 Example: rwm1984
5.3.2 Example: medpar
5.4 ADDITIONAL TESTS
5.4.1 General Negative Binomial Fit Tests
5.4.2 Adding a Parameter – NB-P Negative Binomial
5.4.3 Modeling the Dispersion – Heterogeneous Negative Binomial
5.5 SUMMARY
CHAPTER 6 Poisson Inverse Gaussian Regression
SOME POINTS OF DISCUSSION
6.1 POISSON INVERSE GAUSSIAN MODEL ASSUMPTIONS
6.2 CONSTRUCTING AND INTERPRETING THE PIG MODEL
6.2.1 Software Considerations
6.2.2 Examples
6.3 SUMMARY – COMPARING POISSON, NB, AND PIG MODELS
CHAPTER 7 Problems with Zeros
SOME POINTS OF DISCUSSION
7.1 COUNTS WITHOUT ZEROS – ZERO-TRUNCATED MODELS
7.1.1 Zero-Truncated Poisson (ZTP)
7.1.2 Zero-Truncated Negative Binomial (ZTNB)
7.1.3 Zero-Truncated Poisson Inverse Gaussian (ZTPIG)
7.1.4 Zero-Truncated NB-P (ZTNBP)
7.1.5 Zero-Truncated Poisson Log-Normal (ZTPLN)
7.1.6 Zero-Truncated Model Summary
7.2 TWO-PART HURDLE MODELS
7.2.1 Poisson and Negative Binomial Logit Hurdle Models
7.2.2 PIG-Logit and Poisson Log-Normal Hurdle Models
7.2.3 PIG-Poisson Hurdle Model
7.3 ZERO-INFLATED MIXTURE MODELS
7.3.1 Overview and Guidelines
7.3.2 Fit Tests for Zero-Inflated Models
7.3.3 Fitting Zero-Inflated Models
7.3.4 Good and Bad Zeros
7.3.5 Zero-Inflated Poisson (ZIP)
7.3.6 Zero-Inflated Negative Binomial (ZINB)
7.3.7 Zero-Inflated Poisson Inverse Gaussian (ZIPIG)
7.4 SUMMARY – FINDING THE OPTIMAL MODEL
CHAPTER 8 Modeling Underdispersed Count Data – Generalized Poisson
SOME POINTS OF DISCUSSION
SUMMARY
CHAPTER 9 Complex Data: More Advanced Models
TYPES OF DATA AND PROBLEMS DEALT WITH IN THIS CHAPTER
9.1 SMALL AND UNBALANCED DATA – EXACT POISSON REGRESSION
9.2 MODELING TRUNCATED AND CENSORED COUNTS
9.2.1 Truncated Count Models
9.2.2 Censored Count Models
9.2.3 Poisson-Logit Hurdle at 3 Model
9.3 COUNTS WITH MULTIPLE COMPONENTS – FINITE MIXTURE MODELS
9.4 ADDING SMOOTHING TERMS TO A MODEL – GAM
9.5 WHEN ALL ELSE FAILS: QUANTILE COUNT MODELS
9.6 A WORD ABOUT LONGITUDINAL AND CLUSTERED COUNT MODELS
9.6.1 Generalized Estimating Equations (GEEs)
9.6.2 Mixed-Effects and Multilevel Models
9.7 THREE-PARAMETER COUNT MODELS
9.8 BAYESIAN COUNT MODELS – FUTURE DIRECTIONS OF MODELING?
9.9 SUMMARY
APPENDIXS AS Code
POISSON
POISSON WITH PEARSON DISPERSION SCALED SES
POISSON WITH ROBUST VARIANCE ESTIMATOR
POISSON WITH EXPOSURE (OFFSETS)
POISSON WITH MARGINAL EFFECTS (BOTH AVERAGE ME AND ME AT THE MEAN)
NB2 – NEGATIVE BINOMIAL (TRADITIONAL)
ZERO-INFLATED POISSON
ZERO-INFLATED NEGATIVE BINOMIAL
ZERO-TRUNCATED POISSON
ZERO-TRUNCATED NEGATIVE BINOMIAL
FINITE MIXTURE MODEL
OBSERVED VERSUS PREDICTED COUNTS CHI2 GOODNESS-OF-FIT (0 20 VISITS)
CENSORED POISSON MODEL
Bibliography
Index
Back Cover
