Health insurance policy and behavior in a virtual environment

We introduce a new experimental approach to measuring the effects of health insurance policy alternatives on behavior and health outcomes over the life course. In a virtual environment with multi-period lives, subjects earn virtual income and allocate spending, to maximize utility, which is converted into cash payment. We compare behavior across age, income and insurance plans—one priced according to an individual’s expected cost and the other uniformly priced through employer-implemented cost sharing. We find that 1) subjects in the employer-implemented plan purchased insurance at higher rates; 2) the employer-based plan reduced differences due to income and age; 3) subjects in the actuarial plan engaged in more health-promoting behaviors, but still below optimal levels, and did save at the level required, so did realize the full benefits of the plan. Subjects had more difficulty optimizing choices in the Actuarial treatment, because it required more long term planning and evaluating benefits that compounded over time. Contrary, to model predictions, the actuarial priced insurance plan did not increase utility relative to the employer-based plan.


Citation: Tracy JD, James KA, Kaplan H, Rassenti S (2021) An investigation of health insurance policy and behavior in a virtual environment. PLoS ONE 16(4): e0248784.

Editor: James H. Cardon, Brigham Young University, UNITED STATES

Received: September 26, 2020; Accepted: March 4, 2021; Published: April 6, 2021

Copyright: © 2021 Tracy et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: Data are available in the University’s digital commons at

Funding: The authors received no specific funding for this work.

Competing interests: The authors have declared that no competing interests exist.


US health care costs consume 17.5% of GDP [1], or nearly 10,000 per person, close to twice that of other developed nations [23]. At the household level, these costs can result in bankruptcies and other financial hardships [4]. Yet, even with this vast spending, there are still unmet medical needs. 64% of Americans say that have avoided or delayed medical care in the last year due to expected costs [5]. This combination of high costs and unmet need have led to on-going debate about healthcare and insurance reform.

Much of the debate regarding health insurance policy concerns individual freedom and responsibility versus universal coverage and reducing health disparities [6]. Proponents of universal coverage argue that it improves the average health of the population, decreases disparities in health outcomes within the population, and generates efficiencies by eliminating the transactions costs associated with private insurance billing systems. Proponents of systems emphasizing individual responsibility argue two points: first, that such plans allow people to exert their own priorities in health decisions regarding how much and what kind of coverage to purchase; and second, if insurance premiums and co-pays depend on health behavior, such as smoking, exercise and diet, people will have incentives to improve their health, reducing the total costs of health care.

There is a growing body of research designed to measure the effects of policy alternatives with respect to individual responsibility and social cost sharing. One approach is to take advantage of cross-national data to measure the impacts of health insurance alternatives on patient behavior, population health and health disparities [78]. That research shows that health outcomes in the U.S. have been decreasing in recent decades relative to other developed countries, who all practice universal cost sharing. However, given that many other nation-level variables co-vary with insurance policy [78], it has been difficult to determine whether differences in health insurance policies are responsible for the trend and to pinpoint the causes of the decreasing relative status of the US.

Another approach is to take advantage of ‘natural experiments’ associated with policy changes, such the Oregon and other Medicaid Expansions [911], the SSDI Accelerated Benefits Demonstration Project [12], RomneyCare [13] and the Affordable Care Act (ACA)–ObamaCare [14] These studies confirm that health care usage increases in response to more generous insurance benefits. With the exception of three more recent studies they have been less conclusive, however, about the health benefits from the additional spending and services. There is evidence that self-reported health increases with increased health insurance coverage in Oregon [15] and with RomneyCare in Massachusetts [13], but effects on objective measures of health were not detected (see also [12]) with respect to mortality following the ACA. The assessment of long term impacts is limited by the fact that all are relatively recent reforms, in comparison to the decades over which chronic diseases develop. More recently, Miller et al. [16], Borgschulte and Vogler [17], and Goldin et al. [18] all find that the ACA reduced mortality.

There is also a rich literature of field experiments offering financial incentives for health promoting behavior. Giné et al. [19] show that financial incentives can promote smoking cessation. Volpp et al [20] show that incentives can induce subjects to lose weight in the short term, but in the long-term, post-incentive period the weight is regained. Incentives can also be used to increase gym attendance and exercise. Again, it is not clear that healthy behaviors are sustained in the post-incentive period. See [2122] for systematic reviews. In general, this research shows financial incentives influence people to adapt healthier behaviors. However, given their short time frames, such experiments are limited in their ability to measure the effects of interventions on health outcomes and cost.

These bodies of research illustrate the methodological challenges in assessing the effects of policy alternatives with respect to population health and total health care costs. The vast majority of healthcare costs and disease burdens are associated with important risk factors directly or indirectly behavioral, such as an obesogenic diet, lack of physical activity, smoking and drug use. In addition, the effects of behavior accumulate over many decades, suggesting that full policy impacts may accumulate slowly over time. Another challenge is that people can prepare for, and prevent, disease events along several different fronts. Examples are saving money, purchasing health insurance, changing habitual behavior, and engaging in preventative surveillance and treatment. Since different policies may affect the value of investing in those fronts, assessing total impacts will require an understanding of potential substitutions and complements between policies and individual behavior. Because of these challenges, there is very little causal evidence regarding the impacts of healthcare policy alternatives, despite the very precise statistics on costs and morbidity and the vigor of the policy debate.

Laboratory experiments can be important guides and complements to field experiments which are orders of magnitude costlier, riskier, and more time consuming. The laboratory offers many distinct advantages: the possibility to explore outside the realm of current feasibility, and the possibility to examine changes at the societal level, avoiding potential biases caused by ignoring counterfactuals due to selective participation and defection. Most importantly, the laboratory allows clear identification of moral hazard and adverse selection because the parameters of the environment are perfectly known to the experimenter, and measurements of all outcomes are precise and never obscured by difficult-to-discover costs or self-reported values. Finally, if a particular experimental treatment has negative consequences, they are directly manifested by diminished cash payoffs to participants rather than physical consequence in the real world. Salient results are guaranteed if participants simply prefer more reward to less [23]. Laboratory controlled decision-making experiments have proven to be a valid way to reduce complicated systems to essential operational features that can be used to test the impacts of rule-making policies before much costlier real-world implementation is attempted. For example, experimental electricity markets [24] have examined and improved the operational principles of the annual $14.5 trillion dollar energy industry (8.3% GDP), in which a tiny 1% improvement in efficiency is worth billions! Energy generation and distribution is fraught with complex non-linear dynamics and stochastic uncertainties, and it was highly regulated and long thought to be unamenable to incentive-based management. The fact that results from laboratory experiments with student subjects had external validity and were informative about about potential policy reforms in these complex markets, gives promise that controlled lab experiments can also provide insight into how policy alternatives might influence health, healthcare and insurance decisions.

This paper introduces a new, complementary experimental economic approach to address those challenges in examining the effects of health insurance policy alternatives on behavior and health outcomes over the life course. This experiment compares two types of premium systems for health insurance, one based on individual responsibility and the other on group-level cost and risk sharing.

Our experimental design tests predictions as to how cost sharing in health insurance premiums impact the following outcomes: 1) decisions about whether to purchase insurance; 2) behaviors that prevent or increase the risk of adverse health events; 3) overall health levels; 4) longevity; 5) savings; 6) overall welfare and 7) disparities in health and welfare due to income and age.

The paper proceeds as follows. The next section presents detailed descriptions of the the experimental environment and research design, the theoretical model motivating the experiment, and theoretical results which show how the optimal time paths of consumption and investments in health should vary as a function of the insurance and income treatments, from which we make predictions of subject behavior. Section three presents the empirical results and analyzes the treatment effects on subject behavior. The discussion section presents a summary of our main findings and discusses some weaknesses in the current design and appropriate directions for future research. The conclusion addresses how our approach complements existing approaches.


Model intuition

Our test environment, based on the Grossman [25] lifecycle model of health investment, is designed to model the multiple health-related decisions people make in their daily lives and the different ways people can prepare for and ameliorate disease events. Subjects live multi-period lives and have a stock of health that naturally deteriorates each period, and declines further if the subject suffers a stochastic illness event [26], which we call a health ‘shock.’ Each period, subjects receive an income (an allocation of virtual currency) that depends positively on their income class and their individual health [27] Subjects must then decide on how to allocate their income to current consumption versus health-related investments and savings. Health-related investments include the purchase of health insurance, the purchase of health recovery from deterioration and shocks, and expenditures on physical condition, which we call resilience, that lowers the size of potential future shocks. The ability to invest in resilience is designed to capture preventative behaviors such as exercise, diet and abstention from smoking, drugs, etc. whereas health recovery investment is meant to model services from medical providers, pharmaceuticals, etc. The probability of health ‘shocks’ increases during the second half of life to capture the effects of aging. Insurance pays for health recovery from shocks but does not compensate for natural deterioration.


The first treatment is actuarially-based insurance, in which the cost of each individual subject’s premium depends on that subject’s current expected health care costs to the insurer, which, in turn, depends on that subject’s age, health state, and health-related behavior. The actuarial plan embodies the concept of individual responsibility in a competitive insurance marketplace. Each subject chooses whether or not to purchase insurance given her personal premium. To alleviate the costs of premiums and continuing health degradation, subjects can invest in resilience, save to buffer against health shocks, and directly invest in improving current health. The actuarial plan smooths the lumpy costs of health shocks and provides incentives for individuals to improve their health. However, it may generate significant inequality in health outcomes, due to variable premium costs associated with individual conditions.

The second treatment is employer-based insurance, resembling the plans most Americans have. Premiums are largely subsidized by the employer (funded through a reduction in wages), and there is social cost sharing: all employees have no choice but to contribute to the employer’s costs through universal proportionate wage reductions. The premium the employer must pay the insurer is determined by the average expected costs of all employees who choose the plan. The proportion of the premium paid directly by the employee is fixed, regardless of age, income, health state or behavior. The employer treatment and its incentives differ significantly from the actuarial plan. First, since premiums do not depend on resilience, there is less incentive to invest in prevention (often referred to as, ex-ante ‘moral hazard’). (ex ante moral hazard concerns behavior that proceeds the insured event, and distinct from ex post moral hazard which concern behavior post event, such as seeking greater treatment than one would had they been uninsured.) Second, people with greater risk (the ‘old’) would have a greater incentive to purchase the insurance since their expected health costs would be greater (often referred to as, ‘adverse selection’). However, this plan is expected to generate less inequality in health outcomes due to social cost sharing.

Subjects were assigned to one of four experimental treatments groups in a 2×2 experimental design. One treatment dimension was the actuarial versus employer insurance plan, and the other was high versus low income. In addition, all subjects face a young and an old period during each life with the latter characterized by higher health risks (probability of shocks), so our statistical analysis is necessarily 2x2x2.


This subsection presents the general model employed in our experimental design. We build upon Grossman [25] where resources are limited and can be allocated between enjoying life and ensuring that it continues. Our subjects live for a maximum of T periods. They each have two state variables: health Ht and savings St. Each period, health degrades by a fixed amount δ, and is improved through an investment of  into health. Increasing health allows a subject to become more productive, earn greater income and spend less time attending to ailments; thus, income available to be spent by the subject during each period of the experiment is dependent on health. Any income not invested on health or current consumption is deposited into a savings account available for future consumption or investments in health. Subjects derive utility (which we refer to as ‘joy’ in the subject interface) from current investments in joy . Joy is a concave function of dollars invested in consumption, moderated by health, as being healthy allows one to enjoy life more. Joy is our primary outcome and cumulative joy earned determines subjects payouts.

In addition to the standard elements of the Grossman model, we add the prospect of stochastic negative health shocks, and two additional investments, resilience  and insurance . Should a shock occur that period, resilience investment reduces the magnitude of the shock, while insurance, if purchased, will pay to recover any health lost due to the shock.

Listed below is the event timeline faced by subjects in our model. Each period, t, is subdivided into the following four stages:

  1. Investments  withdrawn from income and savings.
  2. Intermediate health Ht is determined for the current period considering health at the end of last period, health degradation δ, health improvement , and whether a stochastic health shock  is realized.
  3. Joy  and period income M(Ht) are computed based on investments and intermediate health.
  4.  health is recovered if insurance was purchased and a health shock occurred, yielding a final health Ht+1.