The Mystery of the Health Insurance Experiment: An Extended Attrition Analysis

Presenter: Jeremy C. Green, Yale University

Abstract

Rationale: The Health Insurance Experiment was the most expensive social experiment ever conducted in the United States. The experiment found that health insurance plans can save money through the use of cost sharing in their plan designs without causing deleterious impacts on the health of their enrollees. Following the experiment a quarter century of public and private health policy has aimed to impose cost sharing on individuals through copayments and deductibles in order to avoid the use of potentially unnecessary health services that plan enrollees may seek if all of their health care services were paid for by their health insurance. Several recent articles in the health policy literature have suggested that the experimental results may in truth be a spurious artifact of an attrition differential between the experimental cost sharing treatment groups and the free fee-for-service plan control group.

Methodology: In this paper I first discuss the recent arguments surrounding the potential problem of attrition and then proceed to re-analyze the original experimental data. I examine some reasons that participants gave for their attrition. I then use multiple imputation (MI) to fill-in the missing data and re-estimate treatment effects using multiple imputation estimation.

Results: Of the 176 attrition cases in the treatment group, 25 participants left the experiment voluntarily primarily because they were returning or changing to other insurance because it provides better or less expensive coverage, or is more convenient than the experimental plan, and 23 participants left the experiment voluntarily primarily because they were returning or changing to other health insurance because of difficulty paying for their health care under the experimental plan. Diagnostic analysis of the multiple imputation procedure finds that hospital admissions are very difficult to predict which may be because illness is a largely random event.

Conclusions: The best information we have regarding the reasons for attrition, are in those reasons for attrition provided by the individuals leaving the experiment. Behavioral economic theories may suggest that people do not like the very idea of trading of health and dollars; individuals may have left the experiment because they were worried about being able to pay for their health care, even if they would have been able to. It is not possible to know whether or not this benign explanation for attrition is true.