Impact of Consequence Information on Insurance Choice

Abstract

Insurance choices are often hard to rationalize by standard theory and frequently appear suboptimal. A key reason may be that people are unable to map the cost-sharing features of plans to their distribution of financial consequences. We develop and experimentally test a decision aid that provides this mapping to simplify comparisons of plan options. In two experiments mirroring typical health insurance decisions, we find that when people choose plans using standard feature-based information, they violate dominance at high rates. Our distribution-based decision aid substantially reduces dominance violations, and also changes choice patterns in situations where there is no dominant option. Choice patterns under feature-based menus can be most easily rationalized by models of heuristic choices, such as minimizing premium or deductible. With the decision aid, though, significantly more people have choice patterns that are better explained by expected utility theory. We compare our distribution-based approach to an alternative of providing estimates of the expected value of costs, which is the most common decision-support available in most insurance markets. Providing expected values affects choices in a similar direction as our consequence-based approach, but in a more muted fashion, and is only about half as effective at reducing dominance violations.

The effect of information disclosure on demand for high‐load insurance

“…The rightmost columns of Table 5 provide some directional indication that disclosure might help in reducing the number of FOSD choices made by subjects. This would be consistent with other studies on the topic (e.g., Samek and Sydnor, 2017). A Fisher's exact test comparing the incidence of FOSD violators in each treatment to those in the control show that these differences are not statistically significant.…”

More options, more problems? Lost in the health insurance maze

“…Panel A of Figure 6 shows supporting evidence for H1, as the probability to choose one of the normatively meaningful deductible plans moves from 46.07% to 60%, which is statistically significant (p = 0.0416). A clear visualization of the relationship between incurred health costs and costs borne by the insured for each deductible plan thus seems to improve choices, as also found in Samek and Sydnor (2020).…”

“…Gruber et al (2021) find that machine learning‐based decision support tools can be effective in improving decisions and even be a substitute for skilled expert advice. In a recent contribution, Samek and Sydnor (2020) document the significant effects of presenting subjects with a visual representation of the financial consequences of health cost, conditional on plan choice, on health insurance choice optimality within an incentivized laboratory setting as well as a hypothetical‐choice experiment in the field. Samek and Sydnor (2020) rely on self‐reported health status to identify the utility‐maximizing option, focusing on poor decision‐making resulting from a lack of the ability to map incurred health costs to financial consequences, given particular health insurance plans.…”

“…In particular, relative to the benchmark setting of the extant system, we evaluate decisions under conditions with a reduced number of choices (i.e., only the two optimal options) as well as with an increased level of information available. The latter is motivated by the recent work of Samek and Sydnor (2020), who present participants with so-called "consequence graphs" that map individuals' total incurred health costs (i.e., costs caused by accessing medical services) to individual health spending (i.e., costs borne by the insured) for a set of contracts in a graphical representation. Please note that we use the term "consequence graph" generally to refer to the representation that maps incurred health costs into total costs borne by the insured.…”

“…Please note that we use the term "consequence graph" generally to refer to the representation that maps incurred health costs into total costs borne by the insured. It differs from the notion of the consequence graph, as used originally by Samek and Sydnor (2020), who incorporate in the mapping the distribution of outcomes using a quantile approach. Thus, our analysis is not a direct test of the same approach.…”