• “Health insurance for ‘humans'”

    Homo economicus does not buy health insurance. But ordinary humans do! And, unlike homo economicus, their shopping experience is distorted and constrained by all manner of frictions. In a recent NBER working paper, Benjamin Handel and Jonathan Kolstad consider the effects on health insurance choices of humans’ lack of information about plan attributes and the time and hassle costs they experience from and expect of plan use.

    In this paper we leveraged novel, individually-linked, administrative and survey data to show that both information frictions and perceived hassle costs are important factors for consumer health insurance choices at the large employer we study. We quantified the monetary implications for a variety of specific frictions, and revealed that including these friction measures in an expected utility framework typical of the structural insurance literature has potentially important implications for risk preference estimates. In our setting, omitting the typically unobserved friction measures leads to higher estimates of consumer risk aversion, which in turn directly impacts welfare analysis. In a simple menu design counterfactual analysis designed to highlight the welfare implications of our results, we find that, when we omit our additional friction measures from the model, the consumer welfare loss from risk exposure is approximately double that when these measures are included. While the direction and magnitude of this welfare result are specific to our setting, the analysis illustrates that accounting for these typically unobserved choice frictions can have potentially important implications for both choice and welfare analyses in insurance markets.

    Translation: Analyzing health plan choices without controlling for the frictions humans encounter, like incomplete information and transaction costs, yields biased results on the degree to which consumers seek protection from risk. How big a deal is this? Pretty big:

    Our estimates reveal the important role of the additional frictions. The baseline model, based on the administrative data alone [i.e., not controlling for frictions], predicts substantial risk aversion, with a mean constant absolute risk aversion (CARA) coefficient of 1.6 x 10-3. Framed in terms of a simple hypothetical gamble of similar scale, a consumer with this level of risk aversion would only be indifferent between not taking any action and taking on a gamble in which he gains $1000 with a 50% chance and loses $367 with a 50% chance. In other words, he would have to be paid a risk premium of roughly $633 in expectation to take on this risky bet. Incorporating measures of inertia, consumers are estimated to be less risk averse: the average one would be indifferent between no gamble and the same gamble that loses $812 with a 50% chance rather than $367. Our primary model — incorporating information frictions — leads to lower estimates of risk aversion relative to both baseline models: in the full model with all frictions the consumer would be indifferent if the gamble included a 50% loss of $913.

    Translation: Controlling for frictions is huge, leading to the conclusion that consumers (at least those in the sample studied) are far less risk averse than implied by a model that does not do so. Put another way, humans may not be choosing the plans they do for risk protection purposes. They may, instead, be making mistakes that homo economicus would not. I’m not even sure it’s fair to call them “mistakes.” Let’s just say, “we’re human.”


    • Translating to real world: a 75 yo F has a choice of two MA plans. One has front loaded costs but offers better LTC and rehab benefits, the other the reverse.

      She will likely require rehab at some point, but defers to the latter plan because of the temptation of lower initial premiums. She succombs financially in the long run, ie, experiences welfare loss.

      The 75 yo F does not understand the benefit, and thus, overlooks the risk–the risk she would averted–if she had the health literacy (a friction?) to select the optimal option on choosing?



    • As much as I applaud the sentiment in this post, I can’t help but wearily chime in to point out that real humans routinely minimize these frictions in this and every other arena down to their practical limits by hiring agents, or using other informal social mechanisms that have existed as long as humans have had the propensity to “truck, barter, and exchange.”

      Far more often than not, these emergent mechanisms generate outcomes which are superior to those generated by centralized bureaucratic mechanisms borne out of formal analysis of theoretical models and imposed from on high.

      If you disagree, I invite you to consider the methods that you employ to find a reliable babysitter, ponder the results you get with these methods, and contrast them with what you’d get if an incomplete list of babysitter attributes were parameterized, formalized into a model, optimized relative to a set of generic “welfare maximizing” preferences, and centrally administered by a remote bureaucracy. If you aren’t satisfied, repeat for criminal defense attorney, and continue right on down to pet grooming service, then tabulate the final score.

      Time to sigh very heavily, and drink to the memory of the recently departed Ronald Coase.

      • Maybe I should have added a footnote: Estimates in paper based on actual humans. Clicking through to the abstract would have revealed this as well. Sigh.

        • Is the central finding – that real consumers don’t evaluate health insurance coverage in the same way that a financial risk-minimizing algorithm would – actually surprising to anyone?

          That *can’t* be the case, since it’s been patently obvious for millenia that consumers evaluate purchases/transactions in multiple dimensions beyond price, so I can only hope that it’s the determination of the magnitudes associated with a sub-set of these choice frictions that anyone finds interesting or novel.

          @Mr. Mr.:

          Can you identify a single domain in which humans are given choices that one couldn’t document deviations from the supposed optimum? It’s quite a long leap from documenting such deviations exist, to demonstrating that the policies imposed to limit theoretical welfare losses actually do so, and leave consumers better off than they would have been otherwise.

      • Jay, if you read the paper, particularly the survey results, you’d notice that huge swaths of the respondents could not correctly identify facets of one of the plan options available, _even when they were enrolled in that plan_! (It’s on page 17,) It’s hard for me to see any outcome from such information deficiency as optimal.

        By the way, do agents even exist right now to help people pick health insurance?