[Attention Conservation Notice: About 3,500 words on the factual, scientific, and philosophical problems of a paper which was surely not intended to be taken seriously as science or philosophy. Nick Matzke comes at it from a different angle at The Panda’s Thumb, and more briefly.]
Evolutionary geneticist Jerry Coyne has an article coming out in the journal Evolution, in which he demonstrates yet again why excellence in one field does not guarantee competence in any other field. The paper aims to do several things: to advance an argument about why evolution is so controversial in US political culture, to settle scores with people who take different approaches to stopping creationist attacks on evolution, to use pro-evolution activism to advance his own antireligious agenda, to sketch out a model of why religion persists in some societies and not in others, and to propose a new approach for science advocacy: to protect science, we should focus on ending religion, and to end religion, we should focus on ending social inequality.
I applaud Coyne for recognizing that social inequality, especially as measured by income inequality, is an important factor in science literacy and in general. If only there were some sort of mass movement already dedicated to that issue, but readers of Coyne’s blog would find only one passing reference to the inequality-driven Occupy Wall Street protests of last fall. Coyne’s paper demonstrates no particular awareness, indeed, that there’s a large literature on the factors driving income inequality, and the consequences of income inequality for societies, let alone of existing efforts by activists to alleviate inequality and its harmful consequences.
Set that aside, though: the paper wasn’t published in a political science or sociology journal, and while one can argue that the audience of evolutionary biologists would benefit from such context in a paper so far from their training, one could also argue that such context would have made the paper too long, though I think it might have simply forced him to focus on his actual argument.
But one would think that the statistically-trained reviewers at a journal like Evolution would have recognized the dangers of the analysis Coyne offers to justify his claims, an analysis which relies heavily on correlations between the traits of countries (e.g., percent supporting evolution, income inequality, aggregate measures of religiosity). Such correlations between group properties are known as ecological correlations, and are prone to various biases, including the ecological fallacy.
Making the leap from correlation to causation is always hard. But it gets especially bad with the sort of data Coyne is working with. The correlations you find between aggregate measurements of populations are not necessarily the same as the correlations you’d find at the level of individuals, which are the relevant correlations for assessing causation.
For a fascinating example of how ecological correlation can lead us astray, consider Gelman, et al.’s paper “Rich state, poor state, red state, blue state: What’s the matter with Connecticut?,” which explores why the Democratic party does best in the wealthiest states, even though Republicans are seen as the party of the wealthy -Â why is there a positive correlation between a state’s median income and its Democratic vote while there’s a negative correlation between individual wealth and an individual’s probability of voting Democratic? It turns out that the individual correlation between income and voting weakens as a state’s income rises, so that wealthy people in poorer states are more Republican than wealthy people in wealthy states, thus inverting the correlation when you aggregate the individual data.
Similar confounding factors are likely to complicate the data Coyne relies upon. Far better for Coyne’s case would be polls which ask each respondent a question about perceived economic safety, about personal religiosity, and about personal views of evolution and science. Another benefit of such a study is that it would allow us to control for other fairly important social factors which correlate with income inequality.
In a well-designed lab experiment, this is less of an issue. You can control the relevant variables directly and plan around potential confounding factors. In using observational data, especially for comparisons between nations or nations where the populations self-select and change their behaviors in response to the different social structures, it’s a lot dodgier to lean too heavily on correlations.
Coyne’s causal model holds that income inequality (and economic insecurity in general) -> religiosity -> creationism. To back this causal model, he works from aggregate levels of support for evolution in 34 industrialized nations, correlating those with national measures of religiosity, then correlates national religiosity with national income inequality. As we know, correlation is not causation, and this pattern of correlations could imply many other causal pathways: e.g., religiosity -> creationism -> income inequality or income inequality -> creationism -> religiosity.
Certainly, the inequality -> religion -> creationism causal chain has an intuitive logic, but there’s an intuitive logic to other causal chains of these three items. Correlation alone doesn’t let us decide what order makes the most sense for a causal chain. For instance, Coyne and other gnu atheists have long sought to attribute all manner of social ills to the persistence of religion, so might high levels of religiosity cause both social inequality and creationism? Or might income inequality create little incentive to learn science, and then low science literacy leaves the population without a viable alternative to religious explanations? There are, naturally, ways to statistically test these different causal chains, but this would require Coyne to have tested his model by seeking to falsify alternative explanations, or even in considering potential confounding factors, and he shows no interest in rigorously testing his model.
As with all such correlational studies, it’s also possible that the three variables at issue are connected by some fourth factor not present in Coyne’s model. The most obvious factor which Coyne ignores is education. Studies consistently show that, even after controlling for religiosity, how educated someone is has a tremendous impact on his or her views on evolution. Countries with better educational systems also tend to be more accepting of evolution, and countries with high income inequality tend to have worse education systems. The correlations Coyne uses to justify his causal model may, then, be spurious. Oddly, his paper never even mentions differences between national educational systems in discussing different nation’s attitudes towards science. Nor does the paper discuss differences in economic and social development, as measured by GDP or GNI or broader measures like the Human Development Index (which includes economic factors, life expectancy, educational statistics, and other factors to generate a more comprehensive picture of a nation’s development).
This omission is surprising because controlling for GNI per capita is common in such international comparisons. Lots of things correlate with economic development, and factoring that out of a comparison is usually an important first step. Indeed, among the 34 countries Coyne examines, income inequality has a strong negative correlation with GNI per capita, and the median years of schooling among adults has a strong positive correlation with GNI per capita. Many measures of religiosity also correlate with GNI per capita and income inequality. These correlations make causality especially tricky to ascertain, though he could have controlled for them if he wanted to.
My own statistical noodling with those data suggests that religious measures are consistently less important than the combined effects of economic and social development in predicting national levels of support for evolution, at least within the 34 countries from Miller, et al. which Coyne uses as the basis for his paper. Here are the correlations between percent backing evolution and several economic, religious, and societal measures (the economic variables, including income inequality measurement GINI, come from the World Bank and the UN, the data on life expectancy and schooling come from the UN as well, and the religion variables come from surveys by Gallup and the World Values Survey):
|% pro-evo.||GINI||GNI per cap.||HDI (UN)||Rel’n imp’t||God imp’t||Believe in God||Belong to rel’n||Comforted by rel’n||Life expect.||Years of school|
|GNI per cap.||0.46||-0.38||–||0.83||-0.40||-0.35||-0.23||0.02||-0.40||0.74||0.36|
|Believe in God||-0.50||0.38||-0.23||-0.37||0.83||0.92||–||-0.82||0.93||-0.15||-0.56|
|Belong to rel’n||0.20||-0.09||0.02||0.18||-0.61||-0.68||-0.82||–||-0.70||-0.04||0.51|
|Comforted by rel’n||-0.61||0.37||-0.40||-0.50||0.92||0.97||0.93||-0.70||–||-0.28||-0.55|
|Years of school||0.36||-0.44||0.36||0.65||-0.62||-0.57||-0.56||0.51||-0.55||0.12||–|
Note that the smallest correlations involve the fraction of the public who say they belong to a religion (according to data from the World Values Survey). A nation’s income and education and general development doesn’t much change how many people belong to a religion, nor is religious membership influenced by how people conceive of their god(s) or of religion itself (not as strongly as other religious variables are, at least).
This suggests that being religious means something different in more developed nations, an important point to bear in mind given Coyne’s odd definitions. Coyne introduces his own definition of “religion” (focusing on supernatural belief), and then adopts a non-specialist dictionary definition of “religiosity” for much of the paper: “religiousness; religious feeling or belief.” Later, he switches to defining religiosity by “the frequency of prayer among citizens,” following the method of an author he cites favorably. But one can be religious without praying, and one can presumably pray without having firm religious beliefs or feelings. Coyne’s nonstandard definition of religion and his dictionary definition of religiosity are, of course, offered without any citation to the extensive literature on these topics generated by anthropologists, sociologists, and comparative religion scholars.
This is why I pulled together other sorts of measures of religiosity in the table above. Religiosity isn’t just about supernatural belief, and it isn’t just about how often one prays. It can also mean that one feels like a member of a group, or performs certain rituals for cultural or social reasons, or it can manifest in how someone views one’s religion (or religion in general). In regression models trying to parse out how these variables influence acceptance of evolution, I quickly found that even when I pare down the (multicollinear, thus hard to model) religious variables, strong interaction effects between those variables and the others dominate the models. The effects of the economic and social variables consistently match or exceed the effects of religious variables. These strong interaction effects get us back to the inherent dangers of working with ecological correlations, and would seem to invalidate the analysis Coyne reports. The data at hand are simply not the right ones to work with in sorting out causality, yet Coyne builds the substantive component of his paper on exactly this weak foundation.
In any event, the substantive component is fairly small, and is preceded by several sections that don’t really connect with the concluding discussion of inequality. He opens, for instance, by trying to argue that evolution denial is driven by religion per se, and not simply a certain subset of religions. This doesn’t bear up under scrutiny, since the Pew report he cites only compares Protestants, Catholics, and the religiously unaffiliated, not the many other religious groups in this country. A different Pew report from the same year shows that Jews, Buddhists, and Hindus are more accepting of evolution than the religiously unaffiliated, and so are Unitarians (not shown on that graph, but the data are available from Pew). In other words, the difference isn’t just religion vs. nonreligion: the kind of religion involved matters a lot, and we know that based on the same data Coyne mis-cites to reach the opposite conclusion.
With equal disregard for the evidence, he also insists there is “no evidence that resistance to evolution reflects a lack of outreach on the part of teachers and scientists.” But the very same surveys which Coyne wrongly cites to support his claims about religion also show that more educated Americans are much more likely to accept evolution. In addition, the surveys show that respondents who regularly watch science TV shows, who regularly visit science websites (NOAA.gov and ScienceDaily.com are the examples offered), or who regularly read science magazines are more accepting of evolution, and that the effects of these forms of science outreach are cumulative (the more of them a respondent partook, the more accepting of evolution). Education and outreach by scientists matters a great deal.
Coyne doesn’t actually mention those data, and may well be unfamiliar with their existence. He dismisses the possibility that education and outreach might matter by first mischaracterizing the argument of Unscientific America by Chris Mooney and Sheril Kirshenbaum, and then arguing:
there has been plenty of outreach. Now, more than at any time in my life, I see Americans awash in popular science–evolution in particular. Bookstores teem with volumes by Stephen Gould, Steven Pinker, Richard Dawkins, Edward O. Wilson, and Jared Diamond; evolutionary psychology is all the rage; natural history museums have become user friendly; there are dozens of blogs about evolution; and entire television channels are devoted to science and nature. Science education for laypeople isn’t hard to come by, and we have more popular commentary on evolution than ever before.
The point of Unscientific America was that there are multiple discrete cultures in America. In one of those, people read NOAA.gov, subscribe to Scientific American, and watch NOVA. These people respect science and agree with it. They tend to be better educated and wealthier than other Americans, and whiter and less rural, too. In addition to this small Scientific America, there’s a large chunk of the public that doesn’t seek out science, and in the modern media environment, can easily avoid it. This Unscientific America is not reached by many of the extant tools of outreach used by scientists precisely because they decided at some point that science simply wasn’t their thing. Mooney and Kirshenbaum’s book was not simply aimed, as Coyne claims, at chastising scientists for “cloistering ourselves in the lab rather than reaching out to the public.” Their point was that there are indeed scientists doing good work, but that there needs to be more and better outreach. Nothing Coyne presents does anything to undercut that point. Science education for laypeople isn’t hard to find, but it is easy to avoid if it’s dull or poorly targeted.
Coyne doesn’t so much refute Mooney and Kirshenbaum as restate their premise, with less data and less appreciation for the research that already exists on the topic.
Coyne also squeezes in a splash of amateur philosophy amidst his amateur sociology. He once more tries to build a case that science and religion are incompatible. Since he is merely restating the same old story, my objections are the same. But since we’re harping on his abuse of statistics, it’s worth pointing out that he tries to claim that the high rates of atheism among members of the elite National Academy of Sciences membership (93%) demonstrates that religion and science are incompatible.
But consider: until recently, the membership of the NAS was about 90% male. Does this mean that there’s some incompatibility between a pair of X chromosomes and ability to do science? I can’t find data on the racial background of NAS members, but I’d bet we’d find a similarly unrepresentative pattern there, and be loathe to assert that such a correlation indicated incompatibility. It seems far more likely that the disparity in religious membership in NAS relative to the general public represents an extreme case of the same system of self-selection that drives the differing interest in science among various parts of American culture. Differences, it should again be noted, that are accentuated by the income disparities Coyne so recently discovered.
Need I mention that Coyne’s paper includes a gratuitous reference to the Templeton Foundation? That he attacks “accommodationism” (adopting a new definition for the protean term)? That he claims, wrongly (wrongly!), that accommodationism has no evidence in its favor? That he insists it is “not science or philosophy but … theology” to assert that science and religion might be compatible, but does not chide himself for engaging in theology when he invests several pages in arguing the opposite view? Compared to the sloppiness of the rest of the paper, these jarringly incongruous sections hardly merit mention. As polemic, the paper is too dull, and its scientific failings are so obvious and fatal that I simply could not fathom how it was published in a prestigious journal like Evolution if I didn’t know Coyne to be the past president of the society which publishes the journal. It seems power has its privileges.
At the end of the day, I agree with Coyne that so long as the dominant form of American religion is anti-evolution, we’ll have problems with creationism in schools. Which suggests two possible solutions. One, which Coyne advocates exclusively, involves eradicating religion. He likes to toss that idea around, and it works OK as a slogan, but doesn’t suggest any obvious platform of actions that would actually eradicate religion (“Europe did it!” is not a platform). The other solution, which Coyne rejects for reasons that have less to do with evidence than personal aversion, involves changing the dominant form of religion. Doing that would involve outreach by scientists to religious leaders and religious communities, encouraging those who are already pro-evolution to speak out more, those who are on the fence to come out for evolution, and those that are anti-evolution to at least more fully confront the current state of evolutionary science, as well as the full range of theological approaches to evolution.
I think that latter strategy has a lot of potential. Scientific studies show that telling audiences that it is possible to be religious and to accept evolution is one of the most effective way to change their mind about evolution, and those studies are backed by years of experience by activists on the ground. A growing number of evangelical scientists are voicing their support for evolution, and opening up internal discussions within evangelical churches that will at least soften opposition to evolution, and may well be turning people around. Mainline Protestant churches are issuing more and stronger statements in support of evolution and evolution education, and leaders in many religious traditions are taking the opportunity of Evolution Weekend to urge churchgoers not to reject evolution.
The second strategy doesn’t require a complete revolution in our social system. We should, of course, work towards a more equitable economy, and my record on that point is, I dare say, stronger than Coyne’s. But doing so will not happen quickly, nor will any consequent change in society’s religious makeup. I don’t want science education to wait on a back burner for the conclusion of these social revolutions. I think there’s a deep need to uproot the social legacy of slavery and Jim Crow, of gender discrimination, of union-busting, of kleptocratic traditions and rules in Washington and our state capitols, of legacy college admissions, and a host of other tools of oppression and economic division. We don’t, however, need to treat those big, complicated fights as a necessary prerequisite of fixing science literacy. Fixing those inequities in American society could take centuries more, and I don’t think science literacy can wait.
Furthermore, education is a big part of the solution to those economic problems. People with college degrees earn thousands of dollars more per year than those without. A high school diploma is worth a lot of money, too. And a good high school education is one of the best ways to get access to the best colleges. But tuition has grown so rapidly at private universities (like Coyne’s University of Chicago) and even at public schools (which have been gutted to fund tax cuts for the rich), that a college degree is outside the reach of many people. And many others leave college with crushing debt, debts which cripple their ability to pursue graduate training, to establish a family, or simply to give back to their communities.
Education is a silver bullet, and science education is a .357 Magnum. That’s one big reason the President keeps putting scientific training at the forefront of his agenda. And if universities and their senior professors committed themselves to that mission, it could go a long way towards closing income gaps. It would mean lowering tuition, forgiving more student loans, spending down their enormous endowments, and building more connections to high schools. Fortunately, the University of Chicago has long operated its own school, and now runs a system of charter schools throughout Chicago. A U of C professor truly interested in fighting income inequality has lots of opportunities to contribute.
And, for what it’s worth, many religious groups have been at the vanguard of efforts to alleviate social inequities. The US Conference of Catholic Bishops came out against Rep. Paul Ryan’s budget plan on these grounds, arguing that it fails to protect the common good, that it does not protect “the least of these,” and calling for “shared sacrifice by all, including raising adequate revenues, eliminating unnecessary military and other spending, and fairly addressing the long-term costs of health insurance and retirement programs.” The Catholic Worker Movement has long been among the most vigorous opponents of income inequality in society. Many of the Occupy camps last fall hosted tents from religious groups seeking to support the 99% Movement’s efforts to right social inequities, ministering to and protecting the camps through police raids and internal divisions. Religious communities have done a lot to advance the cause of social justice, and as Coyne catches up on the challenges of social inequity, he would do well to reach out to those religious communities and learn from their experiences.