A Modern Approach to Regression with R » (1st Edition)

A Modern Approach to Regression with R focuses on tools and techniques for building regression models using real-world data and assessing their validity. A key theme throughout the book is that it makes sense to base inferences or conclusions only on valid models.

The regression output and plots that appear throughout the book have been generated using R. On the book website you will find the R code used in each example in the text. You will also find SAS-code and STATA-code to produce the equivalent output on the book website. Primers containing expanded explanations of R, SAS and STATA and their use in this book are also available on the book website.

The book contains a number of new real data sets from applications ranging from rating restaurants, rating wines, predicting newspaper circulation and magazine revenue, comparing the performance of NFL kickers, and comparing finalists in the Miss America pageant across states.

One of the aspects of the book that sets it apart from many other regression books is that complete details are provided for each example. The book is aimed at first year graduate students in statistics and could also be used for a senior undergraduate class.

1 Introduction 1

1.1 Building Valid Models 1

1.2 Motivating Examples 1

1.2.1 Assessing the Ability of NFL Kickers 1

1.2.2 Newspaper Circulation 1

1.2.3 Menu Pricing in a New Italian Restaurant in New York City 5

1.2.4 Effect of Wine Critics' Ratings on Prices of Bordeaux Wines 8

1.3 Level of Mathematics 13

2 Simple Linear Regression 15

2.1 Introduction and Least Squares Estimates 15

2.1.1 Simple Linear Regression Models 15

2.2 Inferences About the Slope and the Intercept 20

2.2.1 Assumptions Necessary in Order to Make Inferences About the Regression Model 21

2.2.2 Inferences About the Slope of the Regression Line 21

2.2.3 Inferences About the Intercept of the Regression Line 23

2.3 Confidence Intervals for the Population Regression Line 24

2.4 Prediction Intervals for the Actual Value of Y 25

2.5 Analysis of Variance 27

2.6 Dummy Variable Regression 30

2.7 Derivations of Results 33

2.7.1 Inferences about the Slope of the Regression Line 34

2.7.2 Inferences about the Intercept of the Regression Line 35

2.7.3 Confidence Intervals for the Population Regression Line 36

2.7.4 Prediction Intervals for the Actual Value of Y 37

2.8 Exercises 38

3 Diagnostics and Transformations for Simple Linear Regression 45

3.1 Valid and Invalid Regression Models: Anscombe's Four Data Sets 45

3.1.1 Residuals 48

3.1.2 Using Plots of Residuals to Determine Whether the Proposed Regression Model Is a Valid Model 49

3.1.3 Example of a Quadratic Model 50

3.2 Regression Diagnostics: Tools for Checking the Validity of a Model 50

3.2.1 Leverage Points 51

3.2.2 Standardized Residuals 59

3.2.3 Recommendations for Handling Outliers and LeveragePoints 66

3.2.4 Assessing the Influence of Certain Cases 67

3.2.5 Normality of the Errors 69

3.2.6 Constant Variance 71

3.3 Transformations 76

3.3.1 Using Transformations to Stabilize Variance 76

3.3.2 Using Logarithms to Estimate Percentage Effects 79

3.3.3 Using Transformations to Overcome Problems due to Nonlinearity 83

3.4 Exercises 103

4 Weighted Least Squares 115

4.1 Straight-Line Regression Based on Weighted Least Squares 115

4.1.1 Prediction Intervals for Weighted Least Squares 118

4.1.2 Leverage for Weighted Least Squares 118

4.1.3 Using Least Squares to Calculate Weighted Least Squares 119

4.1.4 Defining Residuals for Weighted Least Squares 121

4.1.5 The Use of Weighted Least Squares 121

4.2 Exercises 122

5 Multiple Linear Regression 125

5.1 Polynomial Regression 125

5.2 Estimation and Inference in Multiple Linear Regression 130

5.3 Analysis of Covariance 140

5.4 Exercises 146

6 Diagnostics and Transformations for Multiple Linear Regression 151

6.1 Regression Diagnostics for Multiple Regression 151

6.1.1 Leverage Points in Multiple Regression 152

6.1.2 Properties of Residuals in Multiple Regression 154

6.1.3 Added Variable Plots 162

6.2 Transformations 167

6.2.1 Using Transformations to Overcome Nonlinearity 167

6.2.2 Using Logarithms to Estimate Percentage Effects: Real Valued Predictor Variables 184

6.3 Graphical Assessment of the Mean Function Using Marginal Model Plots 189

6.4 Multicollinearity 195

6.4.1 Multicollinearity and Variance Inflation Factors 203

6.5 Case Study: Effect of Wine Critics' Ratings on Prices of Bordeaux Wines 203

6.6 Pitfalls of Observational Studies Due to Omitted Variables 210

6.6.1 Spurious Correlation Due to Omitted Variables 210

6.6.2 The Mathematics of Omitted Variables 213

6.6.3 Omitted Variables in Observational Studies 214

6.7 Exercises 215

7 Variable Selection 227

7.1 Evaluating Potential Subsets of Predictor Variables 228

7.1.1 Criterion 1: R2-Adjusted 228

7.1.2 Criterion 2: AICc, Akaike's Information Criterion 230

7.1.3 Criterion 3: AICc, Corrected AIC 231

7.1.4 Criterion 4: BIC, Bayesian Information Criterion 232

7.1.5 Comparison of AIC, AICc and BIC 232

7.2 Deciding on the Collection of Potential Subsets of Predictor Variables 233

7.2.1 All Possible Subsets 233

7.2.2 Stepwise Subsets 236

7.2.3 Inference After Variable Selection 238

7.3 Assessing the Predictive Ability of Regression Models 239

7.3.1 Stage 1: Model Building Using the Training Data Set 239

7.3.2 Stage 2: Model Comparison Using the Test Data Set 247

7.4 Recent Developments in Variable Selection-LASSO 250

7.5 Exercises 252

8 Logistic Regression 263

8.1 Logistic Regression Based on a Single Predictor 263

8.1.1 The Logistic Function and Odds 265

8.1.2 Likelihood for Logistic Regression with a Single Predictor 268

8.1.3 Explanation of Deviance 271

8.1.4 Using Differences in Deviance Values to Compare Models 272

8.1.5 R2 for Logistic Regression 273

8.1.6 Residuals for Logistic Regression 274

8.2 Binary Logistic Regression 277

8.2.1 Deviance for the Case of Binary Data 280

8.2.2 Residuals for Binary Data 281

8.2.3 Transforming Predictors in Logistic Regression for Binary Data 282

8.2.4 Marginal Model Plots for Binary Data 286

8.3 Exercises 294

9 Serially Correlated Errors 305

9.1 Autocorrelation 305

9.2 Using Generalized Least Squares When the Errors Are AR(1) 310

9.2.1 Generalized Least Squares Estimation 311

9.2.2 Transforming a Model with AR(1) Errors into a Model with iid Errors 315

9.2.3 A General Approach to Transforming GLS into LS 316

9.3 Case Study 319

9.4 Exercises 325

10 Mixed Models 331

10.1 Random Effects 331

10.1.1 Maximum Likelihood and Restricted Maximum Likelihood 334

10.1.2 Residuals in Mixed Models 345

10.2 Models with Covariance Structures Which Vary Over Time 353

10.2.1 Modeling the Conditional Mean 354

10.3 Exercises 368

Appendix: Nonparametric Smoothing 371

References 383

Index 387