Health Econometrics Using Stata by Partha Deb, Edward C. Norton, and Willard G. Manning provides an excellent overview of the methods used to analyze data on healthcare expenditure and use. Aimed at researchers, graduate students, and practitioners, this book introduces readers to widely used methods, shows them how to perform these methods in Stata, and illustrates how to interpret the results. Each method is discussed in the context of an example using an extract from the Medical Expenditure Panel Survey.
After the overview chapters, the book provides excellent introductions to a series of topics aimed specifically at those analyzing healthcare expenditure and use data. The basic topics of linear regression, the generalized linear model, and log and Box-Cox models are covered with a tight focus on the problems presented by these data. Using this foundation, the authors cover the more advanced topics of models for continuous outcome with mass points, count models, and models for heterogeneous effects. Finally, they discuss endogeneity and how to address inference questions using data from complex surveys.
The authors use their formidable experience to guide readers toward useful methods and away from less recommended ones. Their discussion of “health econometric myths” and the chapter presenting a framework for approaching health econometric estimation problems are especially useful for this aspect.
List of tables
List of figures
Preface
Notation and typography
1. INTRODUCTION
Outline
Themes
Health econometric myths
Stata friendly
A useful way forward
2. FRAMEWORK
Introduction
Potential outcomes and treatment effects
Estimating ATEs
A laboratory experiment
Randomization
Covariate adjustment
Regression estimates of treatment effects
Linear regression
Nonlinear regression
Incremental and marginal effects
Model selection
In-sample model selection
Cross-validation
Other issues
3. MEPS DATA
Introduction
Overview of all variables
Expenditure and use variables
Explanatory variables
Sample dataset
Stata resources
4. THE LINEAR REGRESSION MODEL: SPECIFICATION AND CHECKS
Introduction
The linear regression model
Marginal, incremental, and treatment effects
Marginal and incremental effects
Graphical representation of marginal and incremental effects
Treatment effects
Consequences of misspecification
Example: A quadratic specification
Example: An exponential specification
Visual checks
Artificial-data example of visual checks
MEPS example of visual checks
Statistical tests
Pregibon’s link test
Ramsey’s RESET test
Modified Hosmer–Lemeshow test
Examples
Model selection using AIC and BIC
Stata resources
5. GENERALIZED LINEAR MODELS
Introduction
GLM framework
GLM assumptions
Parameter estimation
GLM examples
GLM predictions
GLM example with interaction term
Marginal and incremental effects
Example of marginal and incremental effects
Choice of link function and distribution family
AIC and BIC
Test for the link function
Modified Park test for the distribution family
Extended GLM
Conclusions
Stata resources
6. LOG AND BOX – COX MODELS
Introduction
Log models
Log model estimation and interpretation
Retransformation from ln(y) to raw scale
Error retransformation and model predictions
Marginal and incremental effects
Comparison of log models to GLM
Box–Cox models
Box–Cox example
Stata resources
7. MODELS FOR CONTINUOUS OUTCOMES WITH MASS AT ZERO
Introduction
Two-part models
Expected values and marginal and incremental effects
Generalized tobit
Full-information maximum likelihood and limited-information maximum likelihood
Comparison of two-part and generalized tobit models
Examples that show similarity of marginal effects
Interpretation and marginal effects
Two-part model example
Two-part model marginal effects
Two-part model marginal effects example
Generalized tobit interpretation
Generalized tobit example
Single-index models that accommodate zeros
The tobit model
Why tobit is used sparingly
One-part models
Statistical tests
Stata resources
8. COUNT MODELS
Introduction
Poisson regression
Poisson MLE
Robustness of the Poisson regression
Interpretation
Is Poisson too restrictive?
Negative binomial models
Examples of negative binomial models
Hurdle and zero-inflated count models
Hurdle count models
Zero-inflated models
Truncation and censoring
Truncation
Censoring
Model comparisons
Model selection
Cross-validation
Conclusion
Stata resources
9. MODELS FOR HETEROGENEOUS EFFECTS
Introduction
Quantile regression
MEPS examples
Extensions
Finite mixture models
MEPS example of healthcare expenditures
MEPS example of healthcare use
Nonparametric regression
MEPS examples
Conditional density estimator
Stata resources
10. ENDOGENEITY
Introduction
Endogeneity in linear models
OLS is inconsistent
2SLS
Specification tests
2SRI
Modeling endogeneity with ERM
Endogeneity with a binary endogenous variable
Additional considerations
GMM
Stata resources
11. DESIGN EFFECTS
Introduction
Features of sampling designs
Weights
Clusters and stratification
Weights and clustering in natural experiments
Methods for point estimation and inference
Point estimation
Standard errors
Empirical examples
Survey design setup
Weighted sample means
Weighted least-squares regression
Weighted Poisson count model
Conclusion
Stata resources
References
Author index
Subject index
