Medigap Selection Effects: A Study

The dissertation begins by highlighting the significant and sustained increase in healthcare expenditures globally, and specifically in the United States. US per capita healthcare costs increased dramatically from $4,878 in 2000 to $9,523 in 2014, representing a rise from 13.8% to 17.5% of the GDP, reaching a total of $3 trillion in 2014. This trend is not unique to the US; developed nations like Germany and Switzerland show a National Health Expenditure-GDP ratio around 11%. Medicare spending alone reached $618.7 billion in 2014, comprising 20% of total National Health Expenditure and growing by 5.5% from 2013. This escalating cost of healthcare establishes the context for investigating the efficiency and potential problems within the Medigap insurance market.

The study focuses on the Medigap insurance market, heavily regulated by the Federal government. This regulation restricts insurance companies' use of pre-existing conditions for risk classification during open enrollment, resulting in limited information for the underwriting process. This limitation creates significant information asymmetry—a key focus of the study. The dissertation argues that this information asymmetry within the Medigap market presents an ideal setting to analyze economic problems stemming from adverse selection and moral hazard. Insurance contracts based on asymmetric information risk attracting high-risk individuals while deterring low-risk individuals, potentially leading to market inefficiencies or even collapse, as per Akerlof's (1970) work. The presence of moral hazard exacerbates these issues further, suggesting the need for a comprehensive economic analysis of this critical healthcare insurance segment.

The high cost of Medicare spending associated with Medigap coverage is a major concern for policymakers. Some argue that Medigap insurance, particularly plans with first-dollar coverage, incentivize beneficiaries to overutilize healthcare services. This concern has led policymakers to consider strategies to reduce Medigap coverage to curb Medicare spending growth. The dissertation addresses this concern and explores crucial research questions. First, is there a link between the type of individual (high-risk versus low-risk) and their likelihood of purchasing Medigap insurance? This question is analyzed by examining the relationship between Medigap purchase and self-rated health. Secondly, does moral hazard exist in the Medigap market, and if so, what is its impact? The dissertation attempts to answer this by investigating the above questions and drawing conclusions about moral hazard's presence and influence.

The introduction cites the work of Fang, Keane, and Silverman (2008), who investigated the Medigap market using data from the Medicare Current Beneficiaries Survey (MCBS) and the Health and Retirement Survey (HRS). Their study found that those with Medigap coverage spent $4,000 less on medical expenditure than those without, conditional on Medigap price. However, controlling for health status, the study showed that those with Medigap spent $2,000 more. This discrepancy suggests advantageous selection in the market. This prior research highlights the complexity of the Medigap market and the potential for seemingly contradictory findings, emphasizing the need for further investigation into the interplay of factors like cognitive ability and risk preference. The current dissertation builds upon this foundation, aiming to provide a deeper understanding of the Medigap market dynamics and the underlying economic forces at play.

The literature review begins by examining studies on the health insurance market, focusing on Medigap insurance. Wolfe and Goddeeris (1991) used the Retirement History Survey (RHS) from 1977-1979 (2059 observations) to analyze moral hazard and selection effects in the Medigap market. Their findings indicated the presence of adverse selection, although the effect's magnitude was small enough not to cause significant efficiency issues. They also observed a strong positive correlation between wealth and insurance demand. In contrast, other research, such as that by Avitabile (2009) using data from the Survey of Health, Ageing and Retirement in Europe, found no significant positive correlation between the probability of having private insurance and the probability of ex-post risk (hospital treatment). This highlights the varied findings concerning selection effects in different contexts and datasets.

Yao, Schmit, and Sydnor (2012) studied a micro-health insurance market in Pakistan and analyzed the relationship between claim history and renewal decisions. They found that households with larger previous claims were slightly more likely to renew their insurance. However, renewed households were significantly less risky than newly enrolled households, indicating that risk types evolve over time and advantageous selection might occur. This dynamic aspect of risk further complicates the analysis of selection effects in insurance markets. The finding emphasizes the importance of considering the temporal dimension of risk when studying insurance markets.

Cutler and Zeckhauser (1997) explored the impact of selection effects in two distinct settings: the Group Insurance Commission of Massachusetts and Harvard University. At Harvard, where contributions were equal across plans, adverse selection led to a ‘death spiral’ and the collapse of the PPO plan within three years. In contrast, the Group Insurance Commission of Massachusetts, which subsidized 85% of premiums regardless of plan cost, avoided this death spiral. This comparison highlights how different subsidy mechanisms can significantly influence selection effects and market stability in health insurance. This section further underscores the importance of the design and implementation of healthcare insurance plans and its profound effect on the market's health.

The review extends to life insurance and long-term care insurance. Mahdavi (2005) analyzed the correlation between life insurance and mortality rates, considering risk type and risk aversion. They proposed that sufficient risk aversion to reduce mortality rates and high processing costs can lead to advantageous selection, while insufficient risk aversion and low processing costs might result in adverse selection. Finkelstein and McGarry (2006) studied the long-term care insurance market. They noted a lack of significant correlation between insurance purchase and long-term care use, but they argue this doesn't negate asymmetric information. They provided evidence of both adverse selection on risk type and advantageous selection on risk preference, suggesting the importance of considering multiple dimensions of private information in this setting. Their work also introduces a framework for testing asymmetric information in insurance markets that directly informs the methodology used in this dissertation.

Chiappori and Salanié (2000) investigated asymmetric information in the French automobile insurance market, focusing on less experienced drivers. Using various econometric models (including Probit and nonparametric models), they found no evidence of adverse selection. The analysis touches upon the theoretical foundation of the “positive correlation property” which posits a positive correlation between risk type and insurance coverage. The review notes that while empirical evidence in other insurance markets sometimes supports this correlation, contradictions exist. This raises questions about the assumption of only uni-dimensional private information on risk types and sets the stage for considering more complex scenarios in the dissertation's analysis of the Medigap market.

The core dataset for this dissertation is the Medical Expenditure Panel Survey (MEPS), specifically the Household Component (MEPS-HC). Sponsored by the Agency for Healthcare Research and Quality, MEPS has been running since 1996 and collects data through a multi-stage sampling process. The MEPS-HC is a subsample of households that previously participated in the National Health Interview Survey (NHIS). Each year, a new panel of households is selected and followed for two consecutive years, involving five rounds of interviews with the non-institutionalized US population. The study utilizes data from 2009, 2010, and 2011. Although MEPS is an overlapping panel design, this study employs a cross-sectional analysis focusing on each year separately rather than exploiting the panel structure to analyze changes over time.

The sampling method employed by MEPS-HC is not simple random sampling; therefore, the data cannot be treated as a straightforward representation of the general US population. The researchers acknowledge the complexities of the survey design and the need to account for this in their analysis. The survey design incorporates complex features including stratification and clustering, and sample weights are crucial to adjust for potential biases in the sampling process. Some groups may be under-sampled (due to non-response), and others may be over-sampled. Sample weights are applied to correct for these discrepancies to obtain estimates that more accurately reflect the population. The initial number of respondents were 2,344 (2009), 2,274 (2010), and 2,314 (2011). After removing those with zero sample weights, the adjusted sample sizes were 2,295, 2,213, and 2,256 respectively. This careful consideration of the data's structure is essential for ensuring accurate and reliable results.

The analysis centers on several key variables. Total health expenditure is a primary variable, used as a measure of healthcare utilization. Medigap coverage is treated as a binary variable (0 or 1), indicating whether an individual holds Medigap insurance. Crucially, self-rated health, obtained directly from participants, serves as a proxy variable for self-assessed risk type. This variable, an ordinal categorical variable, allows the researchers to indirectly analyze the relationship between risk perception and Medigap insurance purchasing behavior. The inclusion of self-rated health allows researchers to move beyond simple correlations and address the complexities of multi-dimensional private information in the Medigap market.

The study employs several econometric models to analyze the data. Initially, a conventional positive correlation test is conducted to examine the relationship between risk type (proxied by self-rated health) and Medigap coverage. However, this method is acknowledged as a starting point and is supplemented by more robust techniques. Multiple linear regression models examine the relationship between total health expenditure and Medigap coverage, controlling for premium variables (age, gender, smoking status). Probit models are used to analyze the binary Medigap coverage decision. Seemingly Unrelated Regressions (SUR) are applied to account for the potential correlation between error terms in the equations for health expenditure and Medigap coverage. The choice of econometric techniques is justified in the context of the data's characteristics and the research questions being addressed.

The analysis begins with a conventional positive correlation test, examining the relationship between Medigap coverage and total health expenditure. This initial test controls for premium variables (age, gender, smoking status and interactions). The results from 2009, 2010, and 2011 data are then interpreted. For 2009, a significant positive correlation was found, suggesting that those with Medigap coverage had 47% higher total health expenditure. This could be attributed to moral hazard (increased healthcare utilization due to insurance). For 2010 and 2011, this effect was smaller and insignificant, indicating a smaller magnitude of moral hazard in those years. Alternatively, individuals might possess private information about their risk type, impacting their Medigap decision; however, the impact is only statistically significant in 2009.

To delve deeper, the study leverages the ‘self-rated health’ variable from MEPS as a proxy for self-perceived risk. This variable is used to study whether individuals accurately assess their risk level and how their self-assessment affects their health expenditure and Medigap choice. By employing a Probit model (with ‘excellent’ health as the omitted category), the researchers analyze how individuals' self-rated health relates to their Medigap coverage. The findings show that those who self-report as having less than excellent health status are less likely to have Medigap insurance after controlling for premium variables. The analysis then considers the broader context by including additional explanatory variables to account for multi-dimensional private information potentially influencing coverage decisions.

Recognizing potential correlation between error terms in the regression models for total health expenditure and Medigap coverage, the study employs Seemingly Unrelated Regressions (SUR). The SUR approach increases efficiency by considering the correlation between the errors in the health expenditure equation and the Medigap coverage equation, compared to independent models. Including further controls (risk aversion, cognitive limitations, education, and poverty categories), strong and significant effects of self-selected health status on total health expenditure remain. A much stronger positive correlation emerges after controlling for the additional variables, demonstrating the influence of private information not fully captured by standard risk classification. This strengthened correlation is observed across different socio-economic groups. These findings hold consistently across the 2010 and 2011 datasets.

The combined findings suggest both advantageous selection and moral hazard are present. Using independent models, the researchers initially fail to find a positive correlation between self-rated health and Medigap coverage. However, applying the SUR model reveals a significant negative correlation, consistent with advantageous selection. This suggests that healthier individuals are less likely to purchase Medigap. The finding of moral hazard is supported by observing that the correlation between total health expenditure and Medigap coverage becomes significantly positive only after controlling for error term correlation. This signifies that moral hazard mitigates the negative correlation expected from pure advantageous selection. The findings demonstrate a complex interaction between individual risk assessment, insurance choice, and healthcare utilization in the Medigap market.

The discussion section begins by addressing the limitations of using age, gender, and smoking status as proxies for the premium variables used by insurance companies to classify risk. The study acknowledges that these factors do not fully represent the complexities of premium setting in the Medigap market. In some states, community rating is used (same premium regardless of age or health), while in others, age rating is used (premiums vary with age). Even with smoking status and gender considered, age remains the most significant factor influencing premium calculations. The researchers acknowledge that this simplification could introduce potential biases into the analysis.

Another limitation acknowledged is the possibility of cream skimming by insurance companies. Some individuals may have attempted to enroll in Medigap coverage outside the open enrollment period and been rejected. This possibility suggests that the observed data might not fully capture the preferences of all individuals interested in Medigap coverage, potentially leading to a biased representation of the market dynamics. This would skew the results related to adverse or advantageous selection in the Medigap market, especially since only individuals who successfully enrolled are included in the analysis.

A significant limitation is the treatment of Medigap coverage as a binary variable (holding coverage or not). This simplification ignores the variation in Medigap plan types and their different coverage levels. Medigap plans have evolved since their standardization in 1992, with newer plans like M and N introduced in 2010 offering varied premium and out-of-pocket cost structures. By treating Medigap as binary, the analysis loses important details on the heterogeneity of plans and their potential impact on individual choices and health expenditure. Future research could benefit from considering Medigap plan types as a multi-level categorical variable to incorporate this valuable level of detail, which could lead to more precise and comprehensive findings related to adverse and advantageous selection.

The concluding section summarizes the dissertation's findings. The study initially uses data on self-rated health, total health expenditure, and Medigap purchase, controlling for risk factors used by insurance companies and including other variables such as risk aversion, cognitive limitations, education, and poverty levels. Using independent Probit and log-linear models, the researchers failed to find a positive correlation between risk type and Medigap coverage. However, with the SUR model which considers the correlation between error terms for health expenditure and Medigap coverage, a negative correlation was found, supporting the existence of advantageous selection. The study's limitations highlight areas for future research, specifically incorporating more nuanced measures of Medigap plan types and more complete representation of insurance company practices in determining premiums.