Wednesday, June 18, 2014

Republican presidents are bad for our health?

Infant mortality in the US fluctuates with the political party of the President.  A paper published in the  June 4 issue of the widely respected International Journal of Epidemiology ("Us Infant Morality and the President's Party," Rodriguez et al.) reports that between 1965-2010, infant mortality rates were 3% higher when a Republican was president than when the president was a Democrat.

Rodriguez et al. write that previous "political epidemiology" has been cross-national, and has attempted to determine the effect of policy on public health; welfare states, national health systems vs not, higher social expenditures or medical expenditures per capita vs lower, etc. Income inequality was found to be correlated with public health until the data were re-analyzed and additional variables controlled (Avendano, 2012), leading Avendano to question whether income inequality was in fact causal, as opposed to either spurious or real but only correlation with some unmeasured variable(s).  Further studies have attempted to determine which social factors are actually causal; some have suggested social expenditure and the generosity of family policies may be. In this way, political policies may affect public health, but actual causality, rather than just correlation, is difficult to determine.

Rodriguez et al. posed the question, ‘Is the political party of the president of the USA associated with an important, objective and sensitive measure of population health, infant mortality?’ The idea is that the party in power drives macroeconomic policy, and macroeconomic policy influences the socioeconomic milieu, affecting variables that affect health and mortality.

Infant mortality has fallen dramatically since 1965, from a total of 24.7 per 1000 births to 6.1 in 2010.  In the graph below, the authors have removed the trend, and show total infant mortality, neonatal and post-neonatal mortality, by president, for blacks and whites.  During Democratic administrations, all rates are lower, across the board, on average 3% lower.



Logged IMR, NMR, and PMR residual trends and presidential partisan regimes, 1965–2010; Source, Rodriguez et al., 2014
The statistical effects are essentially the same for Blacks and Whites, but infant mortality among Black infants is about two times higher than it is among Whites, so the absolute effect is larger.  The percentages may be small -- very small, in fact -- but their consistency does lend them credibility.

Several things stand out about these results.  First, as the authors point out, the implementation of policies that might have had a direct effect on infant mortality -- Johnson's Great Society and Medicaid in the 1960's, or expansion of Medicaid eligibility between 1979 and 1992 don't correlate with these periodic dips in IMRs.  That would be the easy explanation.  But this means that the correlation with political party may have little or nothing to do with policy differences.

Or, Rodriguez et al. suggest, the correlation could reflect real, cyclical changes in socioeconomic conditions for mothers and infants, depending on national policy.  Or, differential availability of abortion, since high risk fetuses may be more likely to be aborted than fetuses at lower risk.  Or, it might reflect differing attitudes toward health disparities, with Democrats more likely than Republicans to use government to address them -- but what actual governmental policy is implemented, or eliminated, and thus responsible for the fluctuations is anybody's guess.

But there's something curious about these findings.  Neonatal mortality, death before 28 days of life, is generally considered to be due to conditions of pregnancy or congenital abnormalities, while post-neonatal infant mortality, death between 29 days and 1 year, includes sudden infant death syndrome, which isn't correlated with socioeconomic status, but PMR is also considered to be a reflection of socioeconomic conditions.  If that's so, then neonatal mortality should look quite different from post-neonatal mortality in this study, but it doesn't.  It shouldn't be fluctuating with policy differences or income inequality or whatever political or economic factors, if any, might be responsible for the trend reported here.  And, one would expect there to be a more marked difference between Black neonatal and post-neonatal mortality, since health disparities are most reflected in Black infant deaths.

Equally problematic is that one might expect that presidential terms are short relative to the lag time between implementing a new policy and its effects.  The study did allow for a one year lag time, but still, most health policies don't have immediate impact. So the incumbent's party may be irrelevant to what happens during his term, or it would at least be the successor's (sometimes the same sometimes different) party.  Do people's expectations, based on the current President's outlook, change their behavior in subtle ways? Sounds plausible, and would have nothing to do with the policy change itself, but so many people are uninvolved, uninterested in, or skeptical of the political system that this might not be much of an explanation. And the pattern goes back before CNN and FOX imitation news organizations had much intentionally motivating influence on what people thought or were aware of.

Still, social and political epidemiology are interesting approaches to understanding the underlying causes of ill health and mortality.  The fields look at risk factors several steps removed from those generally considered as causes of disease, so that AIDS, or malaria, e.g., might be attributed to poverty rather than HIV infection or being bitten by a parsite-carrying mosquito, and legitimately so.  That is, the idea is that poverty increases one's risk of exposure to diseases, and if you eliminate poverty you eliminate risk.  The difficulty, of course, is that enacting public health policy that calls for eliminating poverty is a lot more difficult than distributing bed nets or clean needles.

And, the problem of identifying cause from correlation is huge with such metadata.  It can be pretty much pure guess work to pull causal factors from the social or political hat, as this paper suggests -- in fact, if something like, to make something up, differences in completeness of registration of vital statistics in Republican and Democratic years were responsible for this cyclical dip, it would look just the same as if the cause were changing policy. The point is, that we just don't know.  In addition, it's hard to avoid interpreting results from one's particular political point of view -- maybe there's something interesting in this paper, maybe there's not, but it's very hard to know.

24 comments:

Holly Dunsworth said...

What if whatever infant mortality reflects is also what causes us to choose a president?

Holly Dunsworth said...

Which is complicated by the fact that it's not always a landslide and we have the electoral college. Phleh.

Ken Weiss said...

Nice try, but if that were the case, then wouldn't the _next_ president's party reflect that? Saddened parents would vote differently the next election, or something. Or maybe if your baby dies, then if you're a Dem you react one way and if you're an elephant the other? This gets all tangled up! It's easier just to blame the Republicans if you are not worried about overpopulation, and the Dems if you think the earth is too crowded already.

Holly Dunsworth said...

I get that we're not making serious claims here, but that is what I tried to write: The condition, described here by infant mortality, is also the condition voters are living in when they pick the next president.

Anne Buchanan said...

I was just collecting quotes for a slide: "Poverty causes HIV/AIDS," "AIDS causes poverty," "poverty causes malaria," "poverty 'ages' genes of young children".... There's the issue of avoiding assuming causation from correlation, but there's perhaps the deeper issue of what we even mean by 'cause'.

Ken Weiss said...

It's all jumbled up as to whether the incumbent's party is responsible, or his predecessor's. So the upshot is, if you want to have a healthy baby, either vote Democrat, or wait to see if Hillary runs and wins. Tell your man to go have a beer and watch some sports or something.....for the next couple of years!

Holly Dunsworth said...

If I were asked to provide hypotheses for these data, they'd be quite different from what Anne reported. I'd say (1) that conditions reflected by infant mortality play a role in how presidential elections go, or (2) infant mortality reporting (by mothers) and data collection and reporting are both slightly different during different administrations. Both of these make slightly more sense than a current president's political party having any effect on the infant mortality during his administration.

Ken Weiss said...

You can argue it in almost any way, I think. The lag time issue is to me the most problematic one. The effect of 3% change is too small I'd say to affect an election result, since most states vote is far more than 3% edge, I think. Attitudes about reporting might be more likely for something less clear-cut than a death, unless prenatals were not considered.

But now I can't tell if anything serious can be said about this, or even whether any explanation is less likely than some artifact in the data.

In tomorrow's post we deal with correlations and how to interpret them (or not)....

Holly Dunsworth said...

That's not what I meant. Conditions as measured by a 3% difference in infant mortality, here, like as a proxy measure for some larger gestalt, are the gestalt conditions that influence voting, not infant mortality itself.

Holly Dunsworth said...

Also, I could be wrong, but abortions can go into neonatal death stats, depending.

Anne Buchanan said...

Holly, I didn't think of it the other way around -- conditions that affect infant mortality rates affecting presidential elections. I like that hypothesis.

I actually think the results are a statistical artifact of some sort, though other than reporting differences, I don't know what that would be. As I noted in the post, to me it's odd that neo and post neonatal mortality show the same pattern.

But this all illustrates the problem with this kind of study -- you've got a correlation, and not only is it hard to pick apart what or whether it tells you about causation, it's impossible to even know the direction of causation. Is this study anything more than a political statement in disguise? I really don't know, but that's because it's impossible to know.

Holly Dunsworth said...

If it was a political statement in disguise, it's not very strong given how a president's party isn't the main determinant of what happens to the USA while he's in office.

Holly Dunsworth said...

Or maybe it is, what do I know about cause!?! :)

Ken Weiss said...

Yes, I think I understood. What I was saying was that I didn't see how the higher-level gestalt factor would have such small but replicable effects, even as a proxy. I guess I'm mixed as to whether this paper is to be taken seriously or not, and if so, then how could one ever even pose a clear analysis to identify the factor.

For one thing, the patter isn't just high vs low rates, but a sine-wave like pattern of increase and decline. Maybe some such thing affects elections, but maybe there is some other kind of explanation. For example--just to illustrate--what if there were relevant generational factors in some behavioral trait such that kids want to be different from their parents. It just appears too regular to me. But of course I'm worse than just guessing, with absolutely no extra knowledge from what's in the paper.

Anne Buchanan said...

The political statement would be simple, and go something like this, I think: political epidemiology often looks at the effects of economic inequality as the cause of ill health. Inequality is higher in Republican administrations, thus Republican presidents are bad for public health, of which infant mortality is an indicator.

Ken Weiss said...

That's plausible but very generic. Again I am not persuaded because of the lag-time issues, the sinusoidal regularity....and the fact that the pattern seems to decline as an administration goes on, reaching a low point before the next admin takes over.

Anonymous said...

An interesting discussion here. http://economistsview.typepad.com/economistsview/2013/11/infant-mortality-and-the-presidents-party.html

Anne Buchanan said...

Thanks. It looks like no one knows what to make of this.

Manoj Samanta said...

Are they raw data or based on rounds of massaging? The sine-wave looks too strange to be real data.

Anne Buchanan said...

No, massaged. They corrected for the steep 50 year declining trend.

Ken Weiss said...

The problem is that all data need some 'massaging', and it's a judgment call. But I agree that the pattern seems far too regular not to raise questions about the data, or about something else. However, the usual kinds of massaging (if done right) doesn't induce that sort of pattern.

Anonymous said...

The author of this commentary on the published paper states that Rodriguez et al did not find differences between neonatal and postneonatal, and between blacks and whites. Well, reading the paper (and just looking at the patterns in the graph) there ARE differences: Effects for postneonatal are much bigger than for neonatal, and the authors say that even though proportional effects for blacks and whites are similar, there are huge differences between absolute effects for blacks and whites.

Anonymous said...

I don't understand why people here are saying 3% is a small effect? Have you read the literature in infant mortality? 3% is perhaps amongst the biggest effects out there, if not the biggest reported!

Anne Buchanan said...

Anonymous, I'm not sure what you mean by 3% being "among the biggest effects out there." My point was just that because infant mortality rates in the US are already low, even if not the lowest in the world, 3% of that is pretty small. Not nothing, and if it's your infant, it doesn't matter how insignificant the statistics look.

The effects of no prenatal care, or of poverty, or of no or poor post-natal care can be much larger than a 3% rise. IMRs in populations without such care are higher by orders of magnitude. See this table, e.g.

And yes, I think neonatal mortality should basically not reflect an effect at all. It should be seen in post neonatal deaths. I don't see the large differences you're seeing. But perhaps we're talking past each other here.