Politics is a Matter of Life and Death (Times 23,000)

by Larry Bartels on January 30, 2012 · 27 comments

in Other social science,Policy,Political Parties

That’s the message I take from a recent book by James Gilligan, a psychiatrist at New York University. In Why Some Politicians Are More Dangerous than Others, Gilligan documents a striking statistical connection between changing rates of violent death in the United States over the past century and the party of the president. He concludes that Republican administrations are “risk factors for lethal violence,” and that the only reason they have not produced “disastrously high epidemic levels” of suicides and homicides is that Democrats have repeatedly undone their damage. (I’ve added handsome hand-coloring to Gilligan’s key  figure in order to highlight the partisan pattern.)

Gilligan found that, over the 108 years covered by his analysis (1900-2007), the age-adjusted suicide rate increased by an average of 9.7 per million over each Republican four-year term but decreased by an average of 11.1 per million over each Democratic term. The age-adjusted homicide rate increased by an average of 3.6 per million over each Republican term but decreased by an average of 4.2 per million over each Democratic term.

These differences may sound small, but they are not. According to the CDC’s latest National Vital Statistics Report, there were 37,793 suicides in the U.S. in 2010 (122 per million population) and 16,065 homicides (52 per million population). Applying Gilligan’s figures to those baseline levels (and assuming 1% population growth per year) implies a projected total of 213,000 violent deaths over the next four years under a Democratic president, but 236,000 under a Republican—a difference of 23,000 lives. (These estimates reflect my calculations based on Gilligan’s data; even they greatly understate the long-run implications of the partisan differences he identifies, since a higher or lower death rate at the end of one term becomes a higher or lower baseline for subsequent fluctuations.)

Gilligan attributes these very different trends in violent death rates to consistent differences in the parties’ policies and performance with respect to unemployment, economic inequality, and recessions, which in turn have substantial effects on “individual emotional and psychological health and welfare.” (The first of these links draws, in part, upon my work.) He argues that intense shame or humiliation is usually “the immediate psychological motive, or cause,  of violent behavior,” and that “these feelings can be stimulated and exacerbated by many stressors in the social environment, one of the most powerful and common of which is the experience of being fired from one’s job, or for any other reason suffering a severe loss of socio-economic status; this experience has been more frequent and prolonged under Republican than under Democratic administrations throughout the twentieth century, and compensatory measures to reduce the intensity of the humiliation (e.g. the WPA under Roosevelt) have been more extensive and effective under Democratic than under Republican presidents.”

Gilligan argues that analyses like his “provide an empirical basis for evaluating different political parties and their social and economic policies and achievements—an assessment grounded in the human sciences, such as public health, preventive medicine, and economic and epidemiological statistics. These can serve as fact-based alternatives to the much more frequent opinion-based assertions and predictions concerning political parties, candidates, and  policies. At a time when every knowledgeable person is calling for evidence-based medicine, isn’t it time we also had evidence-based politics?”

This piece is cross-posted on the VU on Politics blog.

{ 27 comments }

Chris January 30, 2012 at 8:34 am

Correlation is not causation. Typically, when the sun sets, the moon rises. Does not mean the sun causes the moon to rise. To propose that one party’s election or avowed policies, as opposed to enacted policies, can account for the complex factors leading to murder & suicide seems to me to fall into the trap of seeking evidence that supports one’s beliefs rather than explanatory science.

greenweaver January 30, 2012 at 9:04 am

The moon only rises when the sun sets once per sidereal month. On this day it is “full”. The sun’s light is reflected by the entire visible surface of the moon. We know there is a causal relationship. The early empirical research that proved it was inspired by the initial observation of correlation between the relative positions of the moon and sun and the moon’s phase. You decry continued political science but most of us look forward to the follow-up studies and “evidence based politics”.

Keeping Track January 31, 2012 at 12:09 am

To expand on Greenweaver, the fact that the Earth steadily revolves means that about half the time the sun is somewhere in the sky and independently half the time the moon is somewhere within view. Ergo about 1/4 th of the time (or 1/2 of all daytime) sun and moon are both in the sky, in defiance of many children’s stories.

matt w January 31, 2012 at 11:03 am

I think the issue is partly that the moon tends to be hard to see when the sun is up — thus meaning that the correlation between sunset and seeing the moon is in fact causal.

(Though Chris’s example is inapt, I think it is important to note the difference between correlation and causation here.)

Nor December 23, 2012 at 9:25 am

Hm. First off, you are totally wrong about the damn moon. Try looking up sometime. Secondly, suicide is very closely tied to poverty, lack of social services like health care, food stamps, and housing, and economic recession. I am sure you might be able to imagine why. Since Republicans tend to cut social services and increase the gap between the rich and the poor, often leading to increased stresses on poor people, more suicide. Not off the wall. Numbers to back it up. Totally what you would expect. Not shocking at all. Everyone knows a (modern) Republican in the White House hurts the poor. Especially the poor.
And correlation is a perfectly legitimate way to study social phenomenon. Mostly because people tend to object when they are used in experiments. Something about ethics?

Fed funk January 30, 2012 at 9:00 am

You may also want to consider what effect the Great Depression and two world wars had on crime.

David Kiger January 30, 2012 at 10:05 am

The President is not an all powerful dictator. For example, the massive decline in the Clinton years (note the stagnation in the first couple!) may be attributed to the Republican congress stopping his “crazy left-wing agenda” and forcing him to come more towards the middle.

As a basically liberal person, I’d love to be able to point to this as a validation of my political beliefs. However, I’m more than a little skeptical.

John Jay January 30, 2012 at 10:50 am

“He argues that intense shame or humiliation is usually ‘the immediate psychological motive, or cause, of violent behavior,’”

Anyone aware of a scholarly source for that claim?

Steve Tesser January 30, 2012 at 4:39 pm

“Anyone aware of a scholarly source for that claim?” – That kind of supportive documentation would most likely be found in Gilligan’s book, rather than a 300-word review.

John Jay January 30, 2012 at 8:49 pm

One would hope. As I’m not familiar with the claim and don’t have the book, I hoped a Monkey Cage reader might be able to direct me to a source.

Lucas January 31, 2012 at 1:19 pm

Gilligan, being a leading expert on the psychology of violence, sites himself as well as one external source for the initial claim on pg. 97 (the first page of the chapter The Shame of it All). Full citation here:

1 Gilligan, Violence; Gilligan, “Shame, Guilt and Violence”; James Gilligan, “Exploring Shame in Special Settings: A Psychotherapeutic Study,” pp.475 in Christopher Cordess and Murray Cox, eds., Forensic Psychotherapy: Crime, Psychodynamics and the Offender Patient, Vol. II, London: Jessica Kingsley, 1995.

I would imagine that more detailed support for this claim could be found in these texts.

John Jay January 31, 2012 at 11:07 pm

Yeah. I try to be evidenced-based, so I’ll let the data guide me, but seeing Gilligan’s association with psychodynamics should give any person of science pause.

Rob Robinson January 30, 2012 at 12:02 pm

Aside from the really, really questionable mechanism proposed between the President and these changes, doesn’t this raise the same problem of timing that we find in Bartels’ own (much more defensible) research? That is, what sort of lag would we expect in policy-making that would lead to increased suicide?

Josh R. January 30, 2012 at 12:03 pm

Regarding the homicide part, too little attention seems to be given to important historical causes of these patterns, including demographic patterns.

You have the following red periods:
a.) 1900-1914 or so: not sure what is going on there.
b.) 1918-1932 or so: the era of Prohibition! And, not coincidentally, the rise of national crime syndicates and murder over smuggling. A Democratic President residing over this era would likely have seen a similar trend. You then have a precipitous decline that begins with the end of Prohibition and also covers WWII (less murders likely because you have so many young men fighting overseas).
c.) 1952-1960: looks to be rather flat.
d.) Periods of 1968 to 1976 and 1980 to 1988: problem here is that, for the former, crime is already going up and that the causes have a strong demographic factor. Murder (and crime more generally) is primarily a (young) male enterprise – not coincidentally you have population booms during these periods (Baby Boom and a smaller boom) that is likely pushing those numbers up regardless of the party of the Presidency.
e.) 2000-2008 – look rather level (and probably declined since crime declined in general during this era).

The only era that doesn’t have a clear explanation is the first one, but otherwise broader demographic/historical factors seem to be in play beyond party differences.

Lucas January 31, 2012 at 1:20 pm

Have you read the book?

Bob Loblaw January 30, 2012 at 12:20 pm

It could also be the opposite: that a certain zeitgeist in the country leads to election of a certain party. Maybe when the country is feeling like hurting itself or others it votes Republican.

RobC January 30, 2012 at 1:28 pm

Reading Gilligan’s partisan idiocy masquerading as science makes me want to kill myself.

Jack January 30, 2012 at 1:31 pm

I would think Jim Campbell would say that this is really a lag effect, so in fact it’s Democratic administrations causing higher murder/suicide rates and Republicans causing lower rates. We can’t expect to see policy effects immediately, correct?

See the Bartels/Campbell divide over Democratic and Republican regimes’ effects on economic indicators to realize this was at least partially said with tongue firmly planted in cheek.

Adam January 30, 2012 at 2:28 pm

Although there are sources of continuity within each party during this time period (and the authors’ mechanism does entail one of these continuities), the Republican (Democratic) Party from 1900 – 1928 (1900-1948) != the modern Republican Party (Party) in several important aspects.

I’m interested to see if the results hold up in the post World War II period. They don’t seem to based on the graph above.

I second most of the other criticisms concerning the mechanism and other events that correlate with the partisan control of the presidency in the first half of the 20th century.

Larry Bartels January 30, 2012 at 6:44 pm

The slogan that “Correlation is not causation” can be very fruitful when it is offered as a spur to further scientific thinking and analysis—as in many of these responses—but very unhelpful when it is (mis)understood as providing, in itself, good grounds for dismissing any controversial causal assertion.

In the concluding chapter of his book, Gilligan lists seven criteria proposed by the International Agency for Research on Cancer to “establish beyond reasonable doubt whether a given agent (e.g. cigarettes) could be regarded as causing a given outcome (e.g. lung cancer).” The criteria will sound familiar to anyone who has taken a basic course in scientific inference: there must be a plausible mechanism connecting the proposed cause and effect, a monotonic relationship between the dose and response, appropriate timing of cause and effect, consistency with results of other studies, and so on. He then summarizes his evidence regarding each of these criteria in turn, arguing that he has succeeded in “identifying a chain of evidence that can explain what is otherwise an inexplicable association.” Obviously, readers may disagree; but the fact that “Correlation is not causation” does not constitute evidence against his argument.

Gilligan addresses most of the alternative hypotheses suggested here at some point in his book, though often in much less detail than I would wish. Two more general points seem worth bearing in mind. First, since the partisan differences appear in levels of both suicide and homicide, a compelling alternative hypothesis will have to encompass both those domains. Most of the specific suggestions offered here focus on homicides, but the partisan difference is actually larger for suicides; so, even if the counter-arguments for homicides were entirely compelling, they would leave most of the overall relationship between partisanship and lethal death rates unaddressed. Second, the plausibility of any hypothesis—including any hypothesis offered as an alternative to Gilligan’s—declines with the number of moving parts needed to make it work. So the fact that we can come up with six different ad hoc “explanations” for the six major turns in Gilligan’s trend-line should not be very impressive to anyone who approaches the problem with an open mind—especially if those “explanations” cannot themselves be shown to work consistently across a range of cases.

The connection between shame and violent behavior seems to be grounded primarily in Gilligan’s own previous research (e.g., http://www.amazon.com/Violence-Reflections-National-James-Gilligan/dp/0679779124). More concretely, there are substantial scientific literatures connecting unemployment and suicides and unemployment and homicides. So, if unemployment is consistently more prevalent in Republican administrations than in Democratic administrations—it is—why would we _not_ expect the lethal death rate to be higher, as well?

The strong (I am tempted to say “violent”) resistance to Gilligan’s hypothesis among people exposed to a 500-word summary of the argument seems to me to reveal a fascinating squeamishness about the real implications of political choices. Gilligan himself does not seem to be entirely immune; he writes at one point of “a sense of crossing a boundary in moving from the consideration of violent deaths, whether self-inflicted or inflicted by others, to issues of political parties and social justice. And I wonder if others share this sense of transgression.” The answer seems to be “yes.” Nevertheless, if we are ever to develop a political science worthy of the name—that is, the sort of “evidence-based politics” he calls for in the last passage quoted in my post—we will have to transgress some boundaries, entertain some controversial hypotheses, and take seriously the hard work of distinguishing between “partisan idiocy masquerading as science” and real, inevitably uncertain science with real, sometimes uncomfortable, political implications.

David Hulan January 30, 2012 at 7:34 pm

Closer looking at the graph seems to indicate that the violent death rate only rose significantly under some Republicans and only fell significantly under some Democrats. Specifically, there were sharp increases under T. Roosevelt, Coolidge, Hoover, and Nixon, and sharp decreases under Wilson, FDR, and Clinton. But not much change under Taft, Harding, Eisenhower, Reagan, or the Bushes on the Republican side, or under Truman, Kennedy, Johnson, or Carter for the Democrats. I make no attempt to assign significance to any of this, though it’s true that there were no sharp rises under Democrats nor declines under Republicans.

Lasse Aaskoven January 31, 2012 at 6:27 am

It is interesting but again I think structural and economic variables is the real cause of both the in- and decreases in violent death and the shift in Democratic and Republican presidential leadership. Even if this i controlled for, the argument as already stated assumes that the US President has dictatorial powers or at least is the crucial factor in shaping public policy.

While I fully support the idea put forward by dr. Bartels regarding “evidence based politics”, I think using this kind of data to develop a “Vote Republican if you want more violents deaths” argument is both flawfull and at best distorting more generalized political science argument.

To contrast this kind of data evaluation let me give you and example from my own country (Denmark). In the post WWII-period the tax revenue as a percent of GDP rose more rapidly under the bourgouis governments in Denmark (normally a coaliton between a liberal (European meaning) and conservative party) than under the leadership for the Social-Democratic party. But would a “evidence based politics approach” then lead us to conclude: “Vote conservative if you want higher taxes” and “Vote Social-democratic if you want lower taxes”?

Larry Bartels January 31, 2012 at 12:55 pm

I don’t know the Danish case. If “In the post WWII-period” means “pretty consistently over the past 60 years,” then, yes, I think the voters should do some updating. But the U.S. case is not that complicated; the parties’ policies are, on average, consistent with their professed ideologies and with the (economic) preferences of their core constituencies, and they have been for a very long time.

Incidentally, while Gilligan’s analysis is limited to the U.S., he cites independent studies finding that suicide rates are also higher under conservative parties in Australia and Britain. If his argument is correct, that should also be true in other places where conservative parties have produced more (and more stressful) unemployment and economic inequality–but _not_ in places where conservative parties have run more generous welfare states.

Lasse Aaskoven April 11, 2012 at 4:08 pm

Some more on the partisan effect of mortality (health based).

José A. Tapia Granados. (2010). Politics and health in eight European countries: A comparative study of mortality decline under social democracies and right-wing governments. Social Science & Medicine 71 (2010) 841-850

Karl Pearson December 31, 2012 at 3:33 pm

Idiotic bivariate analysis. One could just as well posit that people elect Republicans to deal with especially violent and stressful times, and that the times would be even more violent had Republicans not been elected. The data would support that conclusion as much as it supports Gilligan’s. Gilligan needs remedial education in research design and causal inferences, and should work on his multivariate modeling skills.

Larry Bartels May 18, 2013 at 12:54 pm

“How Austerity Kills”: http://www.nytimes.com/2013/05/13/opinion/how-austerity-kills.html?pagewanted=all&_r=1&.

“The Public Health Effect of Economic Crises and Alternative Policy Responses in Europe: An Empirical Analysis”: http://www.ncbi.nlm.nih.gov/pubmed/19589588.

_The Body Economic: Why Austerity Kills_: http://www.amazon.com/The-Body-Economic-Austerity-Kills/dp/0465063985.

Lawrence Zigerell May 18, 2013 at 2:09 pm

The Stuckler et al. article linked to above claims that, in its sample, each 1% increase in unemployment associated with an estimated mean of 310 additional suicide deaths and an estimated mean of 40 additional homicide deaths…but with an estimated mean of 630 *fewer* road traffic deaths. So a 1% increase in unemployment is associated with *saving* lives, if the analysis is restricted to these three types of death.

Of course, there is no reason to prioritize these three types of death over other types of death, but concern for only three types of death trumps the concern of the original post with only suicide and homicide death.

The influence of public policy on public health is an important area of study for political science. But before we can use this research to inform policy or vote choice, we need a fuller picture of the influence of particular policies on everything that these policies affect in both the short term and the long term.

Comments on this entry are closed.

Previous post:

Next post: