Blog

You know how it goes:  “there are three kinds of lies:  lies, damned lies, and statistics.”

One thing I’ve been following over time is the debate about the minimum wage; most recently, the CBO produced a report projecting the effects of a substantially elevated minimum wage:  it would significantly raise income for many low-wage workers, while putting other worker out of a job.  Supporters of wage boosts saw this as good news (eh, those newly unemployed would probably cycle into and out of jobs and working half the year at a doubled salary, and enjoying free time the other half) and opponents, understandably enough, saw otherwise.  And everyone’s got a study that proves what they want to believe and analysis of their opponents’ studies demonstrating why those studies are wrong.  Heck, the other day there was a claim on twitter that when the minimum wage had been hiked in San Francisco or some such place, “only” the low-rated restaurants closed up shop, so it was no big deal.

It’s all a mess.  And, to be honest, it’s not simply people intentionally deceiving others by manipulating their statistics, or even unintentionally doing so, but that there are so many factors feeding into how a local or regional or national economy is doing at any given time, and so much difficulty untangling the impact of a change and all its knock-on effects, that there is no simple answer.  It stands to reason that an employer faced with boosting employees’ pay will raise prices or try to do with fewer work hours, and I have to say that over the past few days I’ve spent more time in more McDonalds’ (OK, two of them) than has been the case in a while, and they’ve installed automated ordering stations in each of them.  And it’s not a simple as saying, “if an employee has work hours cut in half but a doubled pay, then it’s a win because they have more personal time,” because the employer has to keep the business going with half as much production from the employee, because, one presumes, they had already been aiming at keeping that employee busy and productive during their shift.

So look, what follows isn’t “statisticians/economists are liars,” because these instances don’t prove that anyone acted in bad faith and they really just illustrate more the degree to which, to quote a famous doll, “math is hard.”

The question at hand is this:

Did expansion of Medicaid save lives, and, by extension, the decision by some states to reject Medicaid expansion cost lives?

A working paper out today says “yes.”  It’s titled “Medicaid and Mortality: New Evidence from Linked Survey and Administrative Data” by Sarah Miller, Sean Altekruse, Norman Johnson, and Laura R. Wherry.  (Side note:  the National Bureau of Economic Research working papers are paywalled but I put on my journalist hat and requested access some time ago.  Yay, me!)

They took a look at mortality data by state for the 6 years prior to and the 4 years after Medicaid expansion, and sliced and diced it to look only at those folks who would have benefitted from expansion, that is, poor pre-65 adults.  (I admit that I don’t quite follow exactly they integrated this all together.)  They conclude that there are real, statistically significant decreases in mortality for those states which expanded Medicaid.

Here’s their bottom line:

Our estimated change in mortality for our sample translates into sizeable gains in terms of the number of lives saved under Medicaid expansion. Since there are about 3.7 million individuals who meet our sample criteria living in expansion states, our results indicate that approximately 4,800 fewer deaths occurred per year among this population, or roughly 19,200 fewer deaths over the first four years alone. Or, put differently, as there are approximately 3 million individuals meeting this sample criteria in non-expansion states, failure to expand in these states likely resulted in 15,600 additional deaths over this four year period that could have been avoided if the states had opted to expand coverage.

(Now, I would dispute the framing of “additional deaths” — even if Medicaid expansion is unambiguously a tool to reduce deaths, failing to adopt it is not an action that increases deaths above a baseline.  It’s also a bit misleading to use the 15,600 as a “headline” figure when it’s a four-year calculation rather than over a single year.)

Expressed differently, they found a percentage point reduction of 0.13.  In year one, the probability of dying in an expansion state relative to a non-expansion state decreased by 0.09 percentage points.  In years 2 and 3, by 0.1 percentage points, and in year 4, 0.2 percentage points — in all such cases meeting the usual tests for statistical significance.

So, look, my instinct here is to be skeptical, since up to now, there had not been such proof.  In fact, the best possible way of evaluating the effect of expanded government provision of health services, a randomized trial, occurred in Oregon, when they gave some poor folks, but not others, Medicaid via a lottery in 2008, then measured the outcomes.   The effects?

This randomized, controlled study showed that Medicaid coverage generated no significant improvements in measured physical health outcomes in the first 2 years, but it did increase use of health care services, raise rates of diabetes detection and management, lower rates of depression, and reduce financial strain.

Which would suggest that any impacts would be more long-term, if prevention measures take a while to take effect.  The new paper even references this as a part of the relevant background (and notes a calculated 16% reduction in mortality had such a large confidence interval as to be useless).

But the new analysis finds an unambiguous, immediate drop in mortality for Medicaid-expanding states relative to non-Medicaid expanding states.  This seems too-good-to-be-true, especially since this is occurring at exactly the time when the opioid crisis is causing such a dramatic rise in death rates; Medicaid expansion begin in 2014, and that’s pretty much when death rates exploded per the charts this report.

(What about claims that the expansion of Medicaid ironically increased opioid deaths because people newly-able to receive painkillers through Medicaid were newly prone to opioid addiction and newly able to fill prescriptions and resell the pills?  I’m not going to dig into this too much here; so far as I can tell, there’s just as much “proof” for one side or the other of the debate as with the minimum wage.  An article I was pointed to via twitter, “Causality, Stories, Medicaid, and Opioids” by Andrew Goodman-Bacon at Vanderbilt and doctoral candidate Emma Sandoe at Harvard, presents what they think is a slam-dunk case that the opioid crisis was well under way before Medicaid expansion and thus unconnected to it, but a graph that they present as key evidence, showing that there was already a higher rate of drug overdoses in the expansion states, beginning earlier, in 2010 (before then, there was no difference between these two groups), so that Medicaid could not be the cause, shows a significant (and I mean visually, not statistically) jump in overdose rates in 2014, and further still in 2015 and 2016.  But in any case, if the hardest-hit states are expansion states, and this study’s data shows deaths decreased on an overall basis, then one could make the case that the data supports Medicaid expansion being so fantastic that it even balanced out opioid death increases with even greater decreases in other types of deaths.)

And, for further context, we’re talking about a calculated reduction of 4,800 deaths annually in expansion states and 3,900 potentially fewer deaths in the non-expansion states, for a total of 8,700 fewer actual/hypothetical deaths per year.  (Is this math right?  The expansion states have, in total, much greater population, so it doesn’t entirely make a lot of sense for there to have been nearly as many hypothetical-lives-saved in the smaller number of non-expansion states than actual/calculated lives saved in the expansion states.)  Yes, those are 8,700 people per year whose loved ones per the calculations didn’t/wouldn’t have to say good bye to them too soon.  But there are all manner of government programs touting the lives they would save, directly or indirectly, and there’s always a cost-benefit analysis.

For comparison, in the relevant adult, non-elderly population (actually ages 20 – 64), there were 703,298 deaths reported according to the CDC’s website out of a total population of 3,436,501,449.  This compares to 697,132 deaths in 2016 and 677,192 in 2015 — and death rates that have been increasing nearly every year since this online tool‘s data starts in 1999.  The crude rate was 328.3 deaths per 100,000 in this age group in 1999, and is now 365 — but at the same time, this is called a “crude” rate for a reason, and year-over-year comparisons really need further adjustments to reflect that our population is, in general, aging, so an increasing death rate is, in part, simply because more of the 20 – 64s are older and at greater risk of dying, rather than necessarily saying anything about life expectancy, though that’s a piece of things, too, as the rise in “deaths of despair” is causing drops in life expectancy for middle-aged folk.

A few other thoughts:

It disturbs me that the very regional nature of Medicaid expansion — the non-expanders pretty nearly coincide with the South and Great Plains states (see this map) — means that it would be very difficult to truly differentiate between other influences on death rates impacting different regions of the country differently, and the Medicaid expansion.  The study authors attempt to test for this by looking a relative differences in mortality between these two groupings of states as if the “event” started in 2010 rather than 2014; they find “no effect on . . . mortality in expansion states during this pre-ACA period” but visually, where they find an immediate and sudden drop in relative mortality coinciding with the Medicaid expansion in 2014, as soon as they look back four further years, the data actually seems to suggest that there’s a lot of variation in relative mortality (at least at the scale that they show) from one year to the next, which says to me that the changes in relative mortality in 2014 and subsequent years may be statistically significant but that the association with Medicaid expansion might not be the correct identified cause.  I would be interested in a different sort of test — if they took 2014 data and sliced and diced it differently – say, separating out different regions of the country, such as East vs. West, rural vs. urban, etc., — would they see similar (apparent) impacts?

It is also, again, striking that the change should be so immediate when the conventional wisdom has always been that, prior to Medicaid providing benefits for the childless adult poor population, their immediate medical crises were taken care of via a combination of county hospitals and statutory requirements that any hospital treat any patient in the ER, but that what they missed was important long-term care like chronic disease management.  And this leads me to wonder whether there is a placebo effect going on here — by which I don’t mean the self-perception aspect of the placebo effect (people rating pain levels lower when they’ve been given a sham treatment) but the fact that the experience of being treated has a real, meaningful effect on patient well-being and, it seems to me, could be working to reduce mortality at these very small levels even as the Oregon study couldn’t identify specific measures of health that were impacted.

Now, again, so far as I can tell, the math is all fine, and I don’t have any reason to call out the authors on the basis of their methodology, and (except for a small bit of phrasing) they’re not writing in an ideological manner.  And, as a reminder, I’ve said from my earliest days of blogging that, while I reject the notion of “positive rights” or that people have a “right to health care” it is still an appropriate action for government to take, to make provision for the medical treatment of those members of society as cannot do so on their own, though my preference has been some variation of VoucherCare or a Staff Model HMO, the latter because it is clear to me that we simply cannot have a health system in which anyone can truly go to any doctor at any time, but that individuals need coordinated care and, yes, that includes, in part, saying “no.”

But none of this is as simple as “doing the math.”