Today Vox had an article about a new study of health insurance in America that annoyed me in a number of ways, and highlighted both Vox’s patronizing know-it-all style, and the simplistic economics-worship of some of its writers (in this case Ezra Klein). It’s one of those Economics-101 “Yes, you think A but really this [insert shallow confounder] means not A! Wow!” arguments that are beloved of pop economists, and it’s really frustrating to see it being trotted out now by Vox at a time when the USA is going through a major ideological battle over universal health coverage.
This post may turn out to be a bit long and kind of technical, depending on how frustrated I get reading further on the topic as I write …
The basic argument
Klein has taken a new working paper by Finkelstein et al and used a few of its apparently central findings to build up a story around a question. Finkelstein et al analyzed the Orgeon Health Insurance Experiment to find out how much money medicaid recipients were willing to give up in exchange for medicaid, and how much utility they get from their health insurance. As part of this they found that the uninsured actually don’t pay for much of their treatments: only 20% of their out-of-pocket expenses are paid by them, the rest being shouldered by someone else. This is a central part of Klein’s discussion and, in my opinion, a terribly uninformative finding. Klein has a whole section of the article about how the uninsured are actually “kinda-insured,” which is kinda-true but also kinda completely misses the point, in a very important way that, in my opinion, says a lot about the reasons Americans are having so much difficulty with this whole universal health coverage (UHC) thing. He then moves on to a discussion of the findings of the original Oregon Health Insurance Experiment paper in the New England Journal of Medicine, which found that medicaid wasn’t actually that good for a lot of its recipients; there is a lot wrong with this paper and a lot of reasons its findings need to not be over-stated, but Klein doesn’t really consider them, and gives the study findings more weight than (in my opinion) they deserve. He then goes on to one of those discussions that only economists have, which I guess they expect the rest of us to take seriously, that are deeply poisonous in their basic assumptions, and often wrong: “is health insurance worth it?” This is like the classic economics paper on why voting is a waste of time: superficially appealing but absolutely and completely wrong. He finishes with an important statement, that health insurance should be assessed in terms of the value it offers people, but then juxtaposes value with cost-control as if the two things are mutually contradictory. Pretty much everything in this Vox article is superficially right but deeply wrong, and I want to talk about why it’s wrong and what this means for the health insurance debate.
The Oregon Health Insurance Experiment
The Oregon Health Insurance Experiment (OHIE) is at the heart of the findings in this Vox article, but it’s probably not something we should put too much faith in. Basically, the Oregon state government expanded medicaid places a few years ago, but it could only expand to 30,000 so it ran a lottery for the 70,000 potentially eligible people. The 30,000 potentially eligible people then applied for medicaid, with many getting rejected, and Finkelstein cunningly convinced the government to let her study the results. This is a joyous opportunity for health insurance research because it offers a randomized controlled trial (RCT) of access to health insurance: the gold standard of medical research, enabling us to eliminate a whole bunch of confounders and explore only the effect of health insurance.
Unfortunately there are many problems with the Oregon Health Insurance Experiment and the original paper which launched it to fame. First and foremost, although 30,000 people won the lottery, winning the lottery only increased the probability of accessing medicaid by “25 percentage points” because many didn’t apply or were ineligible, and many non-winners somehow finnagled their way into medicaid. Thus the “Experiment” suffers from massive contamination of the kind that usually renders an RCT ineligible for publication, because most of the intervention group ended up as controls and some of the control group ended up as interventions. While the process of assignment to these two groups was random, the process of transition between groups and final allocation was not, and in fact is decided by a very clear set of factors with a high risk of confounding, such as age, unemployment, etc. The second big problem with the OHIE is that the follow-up period was only 2 years, but lottery winners went on a waiting list, so the actual follow-up time from starting medicaid to study end was less than 2 years, but many of the outcomes they studied (blood pressure awareness, treatment and control, for example) require long follow-up, and key outcomes such as financial catastrophe (see below) are dependent on much longer follow-up times and/or retrospective analysis. Note that the non-medicaiders received a full 2 years follow-up, another minor source of bias. The third problem is that many lifestyle and consumption variables that are crucial to understanding the home-financing impacts of health insurance were obtained from a mailed survey with 15,500 respondents (out of 70,000 in the original study!), one of the most infamous ways of introducing bias into studies (respondents to mail surveys are even less normal than you, dear reader(s)). In contrast, surveys of health financing issues in developing nations (which in my opinion are the gold standard of health financing surveys) routinely get 95-98% follow-up in detailed, complex door-to-door interviews. I have said before on this blog that I think American health finance researchers could learn a lot from what the developing world is doing, and this is another example. The fourth problem is the choice of outcomes: even in systems that are completely free (such as the NHS), health outcomes that can be analyzed over just two years of follow-up are heavily dependent on health-seeking behaviors and non-financial access barriers (e.g. work and time off), and the best measure of health success in a health insurance plan is in serious but often rare outcomes – all-cause mortality, hospitalization, that sort of thing. Also, the OHIE didn’t do much analysis of financial outcomes, which are the main point of health insurance programs. Finally, the study is only ethical if you squint and tilt your head: randomizing people to receive health care is not ethical, and the only reason this study gets grace on that count is that America’s system is insane, but the general ethical view of the medical establishment is that just because the state does something convenient, that doesn’t mean it’s ethical to participate in studies of that thing (see e.g. debate in the British Medical Journal for the Godwin-level examples). Regardless, most people accept the validity of the OHIE, so let’s run with it for now, bearing in mind its flaws: flawed papers often still have a lot to tell us.
The uninsured are “kinda-insured”
In my view the central flaw of the Vox opinion piece lies with its uncritical acceptance of the working paper’s finding that only 20% of expenses were paid for by people without insurance, and the implications of this. The Vox article states:
It’s perhaps easiest to explain this through example. Imagine John breaks his leg. If John is uninsured, his brother, Mike, pays for his medical care. But if John has Medicaid, then the government pays for his care. John got medical care either way. So in this case, Medicaid’s money actually didn’t go to John so much as it went to his brother, because it was his brother who actually would have ended up paying the tab.
This is the kind of superficial gotcha that economists like Ezra Klein love, and it’s annoying and … superficial. There is a large body of research on the health financing aspects of health insurance, and a key concept used in that literature is distress financing. In developing nations, distress financing is defined variously as using any of the following strategies to pay for medical care: selling assets from the home or family business, using savings, calling on family members for financial support, or withdrawing children from school to work [yes, you read that right: this is what lack of health insurance does]. What John did was distress financing, and one of the goals of universal health coverage is to reduce or eliminate the incidence of distress financing. Sure, Mike is better off if John gets medicaid, but in health financing we don’t care about Mike, Tom, Dick or Harry: we are designing a system that protects John from financial catastrophe and distress financing. This is because it is of no interest to us if Mike spends his money on a plasma-screen tv or his brother’s appendix or indeed his own, the purpose of health insurance is to pool risk, that is to ensure that no person – whether directly afflicted or not – has to spend unexpected amounts of money on health care. No doubt there are people out there whose monthly premiums are paid for by friends, sugar daddies or family. We don’t care. The important point is that we have established a universal risk pool into which everyone pays, and everyone draws. It’s no concern to us whether Mike pays for John or John pays for John or John’s sugar daddy pays for John, and typically health insurance research doesn’t ask about how premiums are paid, so why should we care how out-of-pocket expenses are paid? So Klein’s example completely misrepresents the moral purpose of health insurance, by assuming the wrong things about why we have health insurance, and misunderstanding the tools that are available to understand how health insurance works.
I also think Klein has misunderstood the working paper on this issue, because I don’t think the working paper makes as big an issue of this distribution of costs as he does. Finally, if John and Mike are sharing the cost of their health care, then really what’s happening there is that they are establishing a very inefficient, unregulated risk-pooling mechanism – a private version of medicaid. When John gets medicaid we aren’t seeing a situation where suddenly Mike is better off because John can pay for his own care, we’re seeing a situation where Mike is better off because John has been drawn into a larger, better-managed, better-regulated risk pool.
Estimating the utility of health insurance
The working paper is largely aimed at estimating the utility of health insurance, and it uses techniques from economics that I’m definitely not qualified to critique. I know nothing about utility functions or their optimization, so a lot of the language and techniques are a mystery to me. However, there seem to be a couple of aspects of their analysis that insert strong biases. For starters, their assumption 3 on page 8:
Individuals choose m and c optimally, subject to their budget constraint
which is explained as:
The assumption that the choices [of some functions] are individually optimal is a nontrivial assumption in the context of health care where decisions are often taken jointly with other agents (e.g., doctors) who may have ddifferent objectives and where the complex nature of the decision problem may generate individually sub optimal decisions
That isn’t to say health insurance is useless, or that medical care doesn’t help. But we’re probably paying too much and getting too little, and now that we’re a lot closer to a world where every American who wants health insurance can afford it, we should be focusing on making sure that all that health insurance we’re buying is actually delivering the health we’re expecting.