USING PROPENSITY SCORES WITH QUASI-EXPERIMENTAL DESIGNS

By William Holmes

TABLE OF CONTENTS

            PREFACE

Approach of the Book

            Example Data

                        The General Social Survey Panel

                        The Health and Retirement Study

            Improving Knowledge

            Acknowledgements

1. INTRODUCTION:QUASI-EXPERIMENTS AND NON-EQUIVALENT GROUPS

Quasi-Experiments and Inference

            Threats to valid inference

            Propensity scores

Quasi-Experiments and Observational Studies

Cross-sectional designs

Pre-post comparison groups

Dose response designs

Panel studies

Longitudinal studies

Broken experiments

Adequacy and Sufficiency of Causal Inference

2. CAUSAL INFERENCE USING CONTROL VARIABLES

Controlling Confoundedness

            Matching as controlling

            Stratifying as controlling

            Weighting as controlling

            Adjusting as controlling

            Multivariate models for controlling

Selecting Control Variables

            Theory selected controls

            Research selected controls

            Ad hoc controls

            Pre-test controls

            Misspecification in causal models

            Consistency in using controls

Getting Consistent Estimates

            Instrumental variable controls

            Estimating with instrumental variables

            Two-Stage Least Squares

            Detecting selection bias

            Removing selection bias

            Checking for misspecification

Summary

3. CAUSAL INFERENCE USING COUNTERFACTUAL DESIGNS

Controlled Experiments

Challenges to Counter-Factual Designs

Natural Experiments

Matching Samples

Propensity Matching

Key variable Matching

Distance Matching

Assessing Matching Results

Sample Weighting

Adequacy and Sufficiency of Matching

Causal Inference with Matching

4. PROPENSITY APPROACHES OF QUASI-EXPERIMENTS

Estimating Propensity Scores

            Regression estimation of propensities

Logistic estimation of propensities

Discriminant function estimation of propensities

Estimation Complications

Checking Imbalance Reduction

            Standardized deviation criteria

            Percent reduction

            Clinical/Substantive criteria

            Graphical criteria

Matching

            Choosing calipers

Using distance criteria

Dealing with dropped cases

Stratifying

Regressing

Adequacy and sufficiency of propensity estimates

5. PROPENSITY MATCHING

One-One Matching

Matching similar propensities

Using calipers

Using distance criteria

Dealing with dropped cases

One-Many Matching

Using calipers

Using distance criteria

Greedy Matching

Nearest-Neighbor Matching

Unequal Cases in Groups

Assessing Adequacy and Sufficiency of Matching

6. PROPENSITY MATCHING: OPTIMIZED SOLUTION

Full Matching

Optimizing Criteria

Optimizing Procedures

Optimization and Network Flow

Genetic Optimized Matching

Adequacy and Sufficiency of Optimized Solutions

7. PROPENSITIES AND WEIGHTED LEAST SQUARES REGRESSION

Propensities as Weights

Weighting Options

Weighted Regression

Assessing Regression Results

Assessing Adequacy and Sufficiency of Weighting

8. PROPENSITIES AND COVARIATE CONTROLS

Controlling Options

Adjustment Options

Propensities versus Time 1 controls

Propensities and Time 1 controls

Assessing Covariate Results

Assessing Adequacy and Sufficiency of Covariates

9. USE WITH GENERALIZED LINEAR MODELS

Generalized Linear Models

Logistic Regression

Matched data with GZLM

Weighted data with GZLM

Covariate data with GZLM

Adequacy and Sufficiency of GZLM

10. PROPENSITY WITH CORRELATED SAMPLES

Paired Samples

Repeated Measures

Repeated Variable ANOVA

Geographically Correlated Samples

Cox Regression

11. HANDLING MISSING DATA

Identifying Missing Data

Imputation of Missing Data

Propensity Estimation with Missing data

Matching with Missing Data

Stratifying with Missing Data

Covariance Control with Missing Data

Sensitivity Analysis

12. REPAIRING BROKEN EXPERIMENTS

When things go wrong

Assessing the damage

Developing a Strategy

Matching Repairs

Weighting Repairs

Control Variable Repairs

REFERENCES

APPENDICES

A. STATA COMMANDS FOR PROPENSITY USE

B. R COMMANDS FOR PROPENSITY USE

C. SPSS COMMANDS FOR PROPENSITY USE

D. SAS COMMANDS FOR PROPENSITY USE

INDEX

return to Home

Revised 1/31/2013