Alan Acock’s book, A Practical Guide to Logistic Regression Using Stata, is written for students and researchers who are new to logistic regression and who want to focus on applications rather than theory. This guide teaches when and why logistic regression is appropriate, how to easily fit these models by using Stata, and how to interpret and present the results.
The book begins with a review of OLS regression and an introduction to the concepts of logistic regression. It compares and contrasts these two methods and explains why logistic regression is usually the better approach to modeling binary outcome data. Along the way, readers will learn about parameter estimation for logistic regression models.
The author then turns his attention to interpreting the models and assessing model fit. The book demonstrates how to transform the coefficients into more interpretable odds ratios and how to estimate relative risks when appropriate. Acock next explains tools such as the pseudo-R², likelihood-ratio tests, Akaike’s information criterion (AIC), and Schwarz’s Bayesian information criterion (BIC) and shows how to use these tools to assess the fit of the model to the data.
Subsequent chapters focus on assessing a model’s predictive utility using sensitivity, specificity, and receiver operating characteristic (ROC) curves. These concepts are explained clearly and demonstrated with practical examples.
The book concludes with a detailed discussion of how to build models with different kinds of predictor variables, how to use Stata‘s margins command to transform the model coefficients to predicted probabilities, and how to use marginsplot to create easily interpretable visualizations of the results. The author includes many examples using continuous and categorical predictors, illustrates various interactions between different predictor variables, and explains complications that may arise, such as multicollinearity.
A Practical Guide to Logistic Regression Using Stata provides a comprehensive, applications-oriented introduction to modeling binary outcomes using logistic regression. Readers at all levels will learn the skills to confidently fit, assess, interpret, and visualize these models using their own data.
1.2 Ways to report results
1.3 Using Stata
2.2 Exploring the data
2.3 Labeling values for categorical variables
2.4 Saving the edited dataset
3.2 What logistic regression can tell us
3.2.1 Robust and cluster–robust estimation
Clustered sample design
4.2 Interpreting ORs as a percentage difference for binary predictors
5.2 How does logistic regression fit better than ordinary least-squares linear regression?
6.2 Fitting logistic regression models with multiple predictors
6.3 Interpreting ORs for quantitative predictors
6.4 Selecting the right base level for categorical predictors
7.2 Information criteria
7.3 Identifying cases that the model fits poorly
8.2 Estimation of sensitivity and specificity
9.2 Comparing tests
10.2 Data preparation
10.3 Estimating the ORs
10.4 The margins command
10.4.2 Estimating the risk for quantitative predictors
10.4.3 Estimating for a combination of categorical and quantitative variables
11.2 Graphs of categorical predictors that include three or more categories
11.3 Graphs with one quantitative predictor
11.4 Graphs with one quantitative and one categorical predictor
11.5 Graphs of a pair of categorical predictors
11.6 Graphs of a pair of categorical predictors
12.2 Estimating the curve (uncentered predictor)
12.3 Centering, collinearity, and nonessential collinearity
12.4 Estimating the curve (centered x)
12.5 Compare centered and uncentered models
12.6 Use of a quadratic with logistic regression
13.2 Interaction of a categorical and a quantitative variable using logistic regression
13.3 Estimating and interpreting probabilities (uncentered)
13.4 Interaction of categorical variables
13.5 Interaction of quantitative variables
14.2 Selected postestimation commands
15.1 Collinearity and multicollinearity
15.2 Sample size
15.3 Small-sample bias
15.4 Relative risk
