In Senate testimony last week, one of those testifying offered the observation, “One million men and women each year are turning to bankruptcy in the aftermath of a serious medical problem—and three-quarters of them have health insurance.” In a column in the Washington Post last week, Professor Elizabeth Warren (who also gave the just quoted testimony) stated, “[H]alf [of bankruptcy filers] said that illness or medical bills drove them to bankruptcy,” an assertion that was repeated at the Hearings last week on the bankruptcy reform legislation. Most of the Democratic Senators in attendance accepted the assertion that 50% of bankruptcies are caused by health problems without question (even going so far as to silence me when I raised doubts about the credibility of that figure). It has also been widely reported in the media.
Professor Warren writes, “With the dramatic rise in medical bankruptcies now documented, this tired approach would be no different than a congressional demand to close hospitals in response to a flu epidemic.” This figure was used to throw cold water on the bankruptcy reform legislation, and it is expected that Sen. Feinstein at least will propose an amendment.
But is it true that it is “now documented” that 50% of bankruptcies are caused by health problems?
The conclusion is based on a study in Health Affairs. Reviewing the study, it appears that the estimate that 50% of bankruptcy filings are precipitated by a “serious medical problem” cannot be supported based on what that study actually examined.
First, the study comes on the heels of many studies over many decades that find mixed evidence for the belief that a substantial number of consumer bankruptcies are caused by health problems. Those that did find some relationship often found a very small relationship, which explains why Professor Warren has described it as “a dramatic rise” in medical bankruptcies. For instance, in an earlier book, Professor Warren and co-authors wrote, “The central finding is that medical debt is not an especiallly important burden for most debtors.” In a more recent article, it was observed that “until the 1990s . . . most empirical studies of bankruptcy did not find illness, injury, or medical debt to be a major cause of bankruptcy.” Indeed, in the Health Affairs article, it is stated that medical bankruptcies increased 23-fold over the past two decades. No previous credible study has ever found anything approximating the conclusion that 50% of bankruptcies are caused by medical problems. The appearance of such a huge anomaly usually augurs caution in interpreting the results in light of the massive contrary results on the other side. Such caution is warranted here.
In fact, the “finding” in this article of a massive rise in medical bankruptcies appears to actually be a result in the way in which medical bankruptcies are counted, rather than an actual change in the numbers. They draw their data from two sources. First, self-identified bankruptcy filers who say that some medical event “caused” their bankruptcy. Second, analysis of “objective” facts on filers bankruptcy papers that find either (1) debtor or spouse lost at least 2 weeks of work-related income because of illness or injury or (2) uncovered medical bills exceeding $1,000 in 2 years before bankruptcy, or (3) debtors who say they had to mortgage their home to pay medical bills (which for some reason they list as an “objective” factor rather than a self-identified factor.
Do these findings support the claim that 50% of bankruptcy filings were caused by a “serious medical problem”?
First, consider the self-identified filers. Among the self-identified factors that are listed as “medical” causes of bankruptcy in Exhibit 2 of the article are the following: illness or injury, birth/addition of new family member, death in family, alcohol or drug addiction, uncontrolled gambling. First, it is surely open to question whether uncontrolled gambling or a death in the family really should count as a “medical” problem. More generally, the category “illness or injury” is very broadly defined in the study, and there is no apparent limit on the time frame over which the illness or injury occurred, or the severity. So classifying all of these factors as medical problems that have “caused” bankruptcy certainly seems open to question.
Second, the “objective” measures from the debtors bankruptcy petitions are, if anything, even more questionable. First, the authors count anything above 2 weeks of lost work income as a “serious medical problem.” There appears to be no time frame over which this is measured, nor does it apparently even need to be consecutive lost work. So, for instance, if a restaurant waiter called in sick for 2 weeks or more in some indeterminate period of time prior to filing bankruptcy, this would presumably count as a serious medical problem.
Nor does the requirement of $1,000 in unpaid medical bills within 2 years of bankruptcy seem like a very plausible measure of serious financial problems. Again, it is pretty easy to rack up $1,000 in unpaid medical bills over a 2 year period, especially if elective procedures not covered by insurance are added in. Moreover, it is well-understood that debtors who are falling into bankruptcy pick and choose which debts they pay, paying down their mortgage or nondichargeable debts for instance, while not paying their unsecured debts, such as medical and credit card debt. So the fact that the medical debts were unpaid says little, because it may reflect strategic payment of debts prior to bankruptcy.
So the categorization of what counts as a “serious medical problem” is quite questionable in this study. But there is a more fundamental problem that this concern hints at–there is no control group in this study. It is usually Statistics 101 that in order to infer causation from a data observation, it is necessary to have a control group. Absent a control group, it is not clear how the authors can make their claims.
So, for instance, one would want to know how many Americans missed 2 weeks of work or had a $1,000 in medical bills and didn’t file bankruptcy. This is precisely why other previous studies have failed to find much of a correlation between health problems and bankruptcy–almost every family in America has a health problem, death in the family, or gives birth every year. Most of them do not file bankruptcy. In short, I suspect a lot of people had medical problems comparable to those who filed bankruptcy, but did not file bankruptcy. Of course, we will never know, because the authors have no control group to determine whether those in bankruptcy were more prone to illness or injury than the population at large.
Moreover, the authors do not compare the amount of medical debt they found to other debt or obligations that bankrupt debtors had. So, for instance, they would count as a medical bankruptcy a debtor who had $1,001 in medical bills, even if that debtor had say $50,000 in student loans, car loans, and other debt. It would be absurd, it seems to me, to say that the $1,001 in medical expenses “caused” that bankruptcy. Nonetheless, it would counted in this study, because the authors do not control for medical debt as a percentage or in relation to the debtors overall debt.
But the problems do not end there. For instance, the authors claim (page W5-71) that from 1981-2001, medical bankruptcies increased 23-fold, citing a study from 1981 published in “As We Forgive Our Debtors.” I have read and reread the relevant chapter of that book, and have been unable to determine exactly what criteria were used to classify medical bankruptcies there, and how they compare to here. It appears that the measure used in the earlier work was the pure narrowest form of self-identified filers, those who stated that they filed bankruptcy because of a health problem. In the current study, it appears the authors ask the self-identified filers if health problems were “a reason” for bankruptcy. I can find no evidence that the authors there counted as medical bankruptcies any bankruptcy where the debtor had above a specified amount of medical expenses. Even if it were the case, there is no evidence that the $1,000 figure chosen in the current study was adjusted for inflation over the prior study. Nor is there any indication that the authors attempted to adjust the medical expenses that are found in the current study for increases in debtor’s income. So again, it seems like they have just changed their method of counting, not the actual substance.
The authors also do not provide any causal explanation for what could have changed in the medical system to produce a 23-fold increase in health=related bankruptcies in 20 years, and specifically note that the percentage of those in bankruptcy who have health insurance has changed little over that time.
In fact consider the following passage from “As We Forgive Our Debtors”:
Our central finding is that crushing medical debt is not the widespread bankruptcy phenomenon that many have supposed. To the extent that the typical debtors in bankrutpcy are painted as sympathetic characters because they are struggling with insurmountable medical debts, these data show that ‘typical’ is the wrong adjective. Only a few debtors find thmselves in such extreme circumstances…. About half of all debtors carry some medical debt, and many carry substantial medical debt. Althought these medical debts are not the obvious cause of the debtors’ bankruptcies they are part of their financial troubles.” (p. 173).
Again, what seems to have changed is not the frequency of the underlying problem, but simply the way the data is counted and classified. In the earlier study, the authors noted that half of filers had some amount of medical debt, but recognized that relatively small amounts of unpaid unsecured medical debt or minor injuries were likely not the cause of bankruptcy, because this is a part of the financial life for almost every American family. For the debtors in the earlier study, the medical debt that was found was relatively small in comparison to the bankrupts’ other debts. In the more recent study, the authors have simultaneously increased what counts as a “medical problem” and classified even relatively small and trivial medical expenses and problems as bankruptcies “caused” by medical problems. Changing the way you count and classify the same data is not the same thing as finding a 23-fold increase in the underlying problem itself.
I close with an illustration that tries to put the major flaws of this study in perspective and the policy recommendations that have been drawn from it. Suppose that I wanted to find out how many Americans filed bankruptcy because of tax problems. I then interviewed bankruptcy filers and checked their financial records, and counted as a “tax-caused bankruptcy” anyone who either (1) paid $1,000 or more in taxes during the past two years, or (2) anyone who said that if he didn’t have to pay taxes he wouldn’t have had to file bankruptcy because he would have had more money for his other bills. I suspect that under that criteria I would find a pretty substantial number of “tax-caused bankruptcies.” I then conclude that, as a result, we shouldn’t make people pay taxes if they believe it might make them file bankruptcy, and that any unpaid tax obligations should get a blanket discharge in bankruptcy (unlike current law, which makes them largely nondischargeable).
Obviously, my hypothetical study of “tax-caused bankruptcies” would be sheer nonsense. I would have no control group (how many other people paid taxes and didn’t file bankruptcy), I would have no information about how large my tax payments were relative to other obligations (mortgage, student loans, etc.), and my data would be subject to high rates of self-reporting bias. You would object–“almost everyone pays taxes, what is so unique about this group?” My policy proposal would be ridiculous. In short, my hypothetical study would be properly dismissed as junk science because it fails to use even the most basic statistical controls and techniques.
Let me emphasize–I do not deny that many bankruptcies are caused by health problems. This is why the bankrutpcy reform bill carves out several specific exceptions for treatment of health expenses and health insurance. In theory, the number may be as high as some now say, although as noted, the overwhelming number of studies fail to find anything approximating such a high number. But if it is true, that conclusion cannot be based on this article that is published in Health Affairs that got so much press last week and so much interest in the United States Senate. The statistical classification and methods are just too questionable to support that conclusion.
Comments are closed.