black/white mortality differentials and american politics

There’s a new paper from Social Science and Medicine making the rounds with the provocative title “Black lives matter: Differential mortality and the racial composition of the U.S. electorate, 1970–2004.” The Monkey Cage has a write-up with a blunt (clickbait-y?) title that emphasizes the paper’s main question, Blacks die sooner than whites. How many votes has this cost Democrats? Something about this framing bothered me.

In the paper, Rodriguez et al* first document the persistent black/white mortality differential in the United States. One of the more depressing facts they report is how this differential has basically remained constant since 1960. It’s a nice data point to highlight how gains in civil rights are dramatically insufficient to secure real material progress when dominant actors have other tools at their disposal to maintain and even increase segregation and inequality.

From there, Rodriguez et al go on to ask their central question:

We evaluate a counterfactual: What would have been the effect on the 2004 general election if blacks had survived at the same rates as whites between the years 1970 and 2004?

The results show that the increase in the black electorate would not have been quite enough to swing the election to Kerry from Bush, but that “between 1970 and 2004 the outcomes of 7 close senate elections, and of 11 close gubernatorial elections would have been reversed from Republican to Democratic victors with the addition of black hypothetical survivors.”

On some level, all of this is sensible enough. But on another level, it’s deeply weird. The analysis requires the assumption that black and white votes remain distributed the same across Republican and Democratic candidates even as we imagine a massive shift in one of the fundamental features of the contemporary American political system: systemic racism.

Put differently, what would have had to be different in order for black/white mortality differentials to disappear? I can’t really imagine a world in which our political system is what it is (or anything very similar), and in which black/white mortality differentials were reduced to zero. Equalizing mortality would almost necessarily require reducing wealth and income differentials, residential segregation, occupational segregation, racialized mass incarceration, and all of the other sorts of systemic inequalities that produced and sustained that differential. It would require an America that was almost unrecognizable in terms of race relations, but somehow otherwise the same in terms of political parties and partisan affiliations.

Is there a better way to make and report on estimates of the “troubling feedback effect whereby premature deaths among blacks affect the balance of political power among blacks and whites in the United States” (as Robinson puts it at The Monkey Cage), one that recognizes the centrality of racial inequality to the contemporary political system?

* The authors have an interesting mix of affiliations across economics, public health, and geography. Several of them are affiliate with the University of Michigan, but we haven’t crossed paths.

Author: Dan Hirschman

I am a sociologist interested in the use of numbers in organizations, markets, and policy. For more info, see here.

10 thoughts on “black/white mortality differentials and american politics”

  1. “Put differently, what would have had to be different in order for black/white mortality differentials to disappear?…It would require an America that was almost unrecognizable in terms of race relations, but somehow otherwise the same in terms of political parties and partisan affiliations. ”

    I’ll again note that focusing on two races omits important racial dynamics, which in this case includes the “Hispanic paradox” in which Hispanic-Americans have longer life expectancies than non-Hispanic white Americans (PDF page 2 here).

    Like

    1. I guess it’s a matter of degree. I think it’s slightly more reasonable to imagine changing a single policy (permitting felons to vote). We could imagine a social movement emerging around the cause, pressuring Congress, a new constitutional interpretation, etc. It’s a specific policy, and even if it’s one that is widely supported right now*, it’s at least plausible to map out paths whereby that policy could change without making massive changes to other structures. For the mortality gap, it’s much harder for me to see what that could look like. You can’t simply pass a law saying “black and white voters will now die at equal rates.”

      I’m not against all counterfactuals, or this kind of counterfactual modeling in general. It just seemed off in this particular case, because the counterfactual wasn’t modeling a plausible policy option already under consideration [which could presumably occur without a major reshuffling of the political system] but a massive alteration of a deep social inequality [which presumably could not].

      * And Manza et al’s work suggests that there is widespread support for allowing those with felony convictions to vote.

      Like

  2. Although I like what the authors did, I tend to feel that this application relies too much on point estimates and ignores the variance. One could assume error in the demographic models of population growth/decline or excess mortality, or one could introduce other influences that would entail their own variance. To your point, one could examine the influence that a reduction in black excess mortality would have on White voting patterns — and could likely get decent point estimates from the social psych and political psych literatures on group threat — and then introduce those into the models. Adding these components would provide bounds around plausible effects. It makes for a worse headline, but better science, I think.

    Like

    1. To some degree, aren’t we already accepting these types of counterfactuals all the time?

      Imagine we run a regression. We have a variable in the right side of the equation: Race (1=white; 0=black). The coefficient B for that variable is the difference in Y between whites and blacks – ALL ELSE EQUAL.

      The assumption behind coefficient B is that cases included in the regression are EQUAL IN EVERYTHING, and that the difference in Y is only due to the circumstance that some of them are white and the rest black, right?

      How often do we think any coefficient in any regression is believable given the nature of these counterfactuals?

      That said, looking at the tables reported by the authors (both in the paper and in their methodological appendix), one can see there is a lot of stability in political behavior across states even though we know there is a lot of variation across states with regard to racial dynamics, income inequality, and history – to name a few.

      I also wonder the degree to which this paper helps us understand how “inequality here” means “inequality there”? The “out of sample extrapolation” becomes an useful way to illustrate theoretical and practical implications of a phenomenon that is very difficult to measure/assess: People who are not there.

      Liked by 1 person

      1. To some degree, aren’t we already accepting these types of counterfactuals all the time?

        You hit the nail on the head.

        I also find the value of extrapolating reasonable scenarios as a helpful theoretical exercise, provided that a) people don’t claim that the counterfactual scenario reflects truth and b) they assess the confidence with which they believe alternative scenarios exist.

        This topic deserves much more thorough treatment than I can give it here, but I agree with your analysis.

        Liked by 1 person

  3. This is a very good example of the difficulty in extrapolating from marginal effects to the overall shape of a function or its slope over a given range. If black mortality improved a little bit, this would probably not change voting patterns or party platforms, and we could forecast how that would affect political outcomes.

    The out-of-sample exercise in the paper is still a useful thought experiment, since it is a likely upper-bound on how much the mortality gap matters for the political process. If you want an alternate framing, maybe “here are some ways that systemic racism perpetuates itself through the political system” would work? And you could discuss the mortality gap but also the incarceration/felony gap and the voting rate gap. While it’s not literally true that 7 senate elections have been tipped by these racial inequalities, it is true that the current political order is predicated on those inequalities existing (just as the racial differences in political affiliations are predicated on the system that generates those inequalities).

    I am also not sure it is inconceivable that we could see big shifts in these gaps that change elections. Consider the incarceration gap: this could be shrunk dramatically by the reversal of current drug laws paired with ex post pardons for people convicted under the current regime. (I’m not clear on how big of a shift this would be – the numbers available here are internally inconsistent: http://en.wikipedia.org/wiki/Incarceration_in_the_United_States#Violent_and_nonviolent_crime but this data from the Bureau of Prisons suggests that half of people in prison are there on drug convictions: http://www.bop.gov/about/statistics/statistics_inmate_offenses.jsp). That would be a major shift, but it’s one that I optimistically see as conceivable during our lifetime, whereas improving mortality is much more complex and more difficult.

    Like

  4. To some degree, aren’t we already accepting these types of counterfactuals all the time?

    Imagine we run a regression. We have a variable in the right side of the equation: Race (1=white; 0=black). The coefficient B for that variable is the difference in Y between whites and blacks – ALL ELSE EQUAL.

    The assumption behind coefficient B is that cases included in the regression are EQUAL IN EVERYTHING, and that the difference in Y is only due to the circumstance that some of them are white and the rest black, right?

    How often do we think any coefficient in any regression is believable given the nature of these counterfactuals?

    That said, looking at the tables reported by the authors (both in the paper and in their methodological appendix), one can see there is a lot of stability in political behavior across states even though we know there is a lot of variation across states with regard to racial dynamics, income inequality, and history – to name a few.

    I also wonder the degree to which this paper helps us understand how “inequality here” means “inequality there”? The “out of sample extrapolation” becomes an useful way to illustrate theoretical and practical implications of a phenomenon that is very difficult to measure/assess: People who are not there.

    Like

Leave a Reply

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s