HMO plans, self-selection and utilization of health care services

Abstract

This study examines the effect of health maintenance organizations (HMOs) on the use of health care services among the privately insured, nonelderly population. To consider jointly the possible self-selection bias and high frequency of zero observations in the applied utilization measures, we accommodate the endogeneity of health plan choice decisions in the censored regression model. Using data from the 2000 Medical Expenditure Panel Survey, we find strong evidence for favourable self-selection into HMO plans. Health maintenance organization enrollment is found to encourage greater use of office-based and hospital outpatient services. Overall satisfaction with the quality of care among HMO members is relatively low compared to that among nonHMO members. These findings suggest that more effort is needed to develop management strategies in HMOs in order to contain the moral hazard in utilization and assure the quality of service provided within the network of HMO providers.

Acknowledgements

We thank Qi Li, Donald Deere and Badi Baltagi for their superb research support and seminar participants at Texas A&M University, KDI School of Public Policy and Management, Seoul National University (Graduate School of Public Health) and the 2004 American Public Health Association Annual Conference for helpful comments. We are also indebted to the anonymous referees for suggestions. We, however, retain sole responsibility for all errors.

Notes

¹ We exclude individuals who have multiple coverage as the combination of private and public insurance because their utilization of health care services is reported as a total annual amount which renders it difficult to determine what portion of the total annual number of visits to specific health services is associated with their coverage and choice of a private insurance plan.

² Being ‘privately insured’ is defined as being insured by any private insurance for all 12 months of 2000. Then, changes in insurance status and insurance plan at any point within the year are not a concern that may complicate the results.

³ Medical expediture panel survey 2000 provides the enrollment status at three different points during a year. Being ‘enrolled in HMOs’ means that a person is reported as being covered by an HMO plan for all three rounds of the survey. There are individuals who were enrolled only for some part of the year. However, among those enrolled in an HMO at any time during 2000, over 80% were enrolled all year.

⁴ Health, United States (2003) reports the percentage of the population enrolled in HMOs by geographical region in 2000: Northeast 36.5%, Midwest 23.2%, South 22.6% and West 41.7%. However, the number of HMO plans available in a region shows the opposite pattern: Northeast 98, Midwest 161, South 203, West 106. These data show that the extent of popularity of HMO plans varies across regions which reflects the variations in adopting new administrative health care systems and in the need to manage quality and utilization.

⁵ To avoid potential bias in including individuals who do not have a choice as to whether to join an HMO, we replicate the health plan choice analysis using a more limited sample, restricting the population to those who have choices in selecting their health care plan. Although the sample size is reduced from 7474 to 2428, it does not change the earlier findings presented in Table 5. Furthermore, we conduct the health plan choice estimation without including health risk/attitudinal measures to assure that the possible endogeneity of these variables does not alter the main findings in Table 5. The full results are available upon request to the authors.

⁶ Estimates of the HMO effect on the utilization of both types of service are all insignificant and change signs across specifications in uncorrected models. In corrected models, a significantly negative, self-selection bias due to endogenous insurance plan choice and a highly significant positive effect of HMO enrollment on the utilization level are found for both service types regardless of the estimation specification. In the log-linear model in which individuals with zero visits to service sites are excluded from the estimation, the magnitude of coefficients is smaller than that in the linear-model. It implies that exclusion of zero observations leads to loss of relevant information and, as a result, provides diluted evidence on the utilization performance of HMO plans. Changes in the sum of squared residuals (SSR) illustrate the importance of considering the statistical feature of dependent variables and selection bias; removing zero observations to avoid the problem of the high frequency of zero observation reduces the SSR, which is further reduced with endogeneity correction. The full set of results is available upon request to the authors.