The opiate of the elites

In case you didn’t see our graph-laden Vox EU article, here it is. The Obama reference is already a bit stale but the content is still fresh, I hope . . .

Barack Obama attracted attention recently by describing small-town Americans who were “bitter” at economic prospects who “cling to guns or religion’’ in frustration. This statement, made during the height of the Democratic nomination battle, has received a lot of attention, but it represents a common view. For example, Senator Jim Webb of Virginia wrote, “Working Americans have been repeatedly seduced at the polls by emotional issues such as the predictable mantra of ‘God, guns, gays, abortion and the flag’ while their way of life shifted ineluctably beneath their feet.’’ And this perspective is not limited to Democrats. For example, conservative columnist David Brooks associates political preference with cultural values that are modern and upscale (“sun-dried tomato concoctions”) or more traditional (“meatloaf platters”).

All these claims fit generally into the idea of religion as the opiate of the masses, the idea that social issues distract lower-income voters from their natural economic interests. But there is an opposite view, associated with political scientist Ronald Ingelhart, of post-materialism—the idea that, as people and societies get richer, their concerns shift from mundane bread-and-butter issues to cultural and spiritual concerns.

Which story better describes how Americans vote? Who are the values voters? Are they the poor (as implied by the “opiate of the masses’’ storyline) or the rich (as would be predicted by “post-materialism”)?

Case studies are interesting but do not resolve the question. Thomas Frank described how Kansas is full of socially conservative Republicans at all income levels. But then there is south Texas, whose low-income Latinos are socially conservative on many issues but vote for Democrats. Manhattan’s upper west side remains strongly Democratic despite its steadily increasing income level, but the suburbs of Dallas are full of high-income conservative Republicans.

There are many ways of looking at social class, attitudes, and voting. We’ll take a demographic approach and compare religious to secular voters.

Regular churchgoers are about 15% more likely than non-attendees to vote Republican. Perhaps surprisingly, this big religion gap did not show up until 1992, when Bill Clinton ran against George H. W. Bush, as we show in Figure 1.


Figure 1. Difference in probability of voting for the Republican candidate for president, comparing people who went to church at least once per week to nonattenders. Nothing much was happening until 1992, when all of a sudden George H. W. Bush received 20% more of the vote among religious than among the nonreligious.

Back in 1980, Jerry Falwell’s Moral Majority and other Religious Right organizations played a prominent role in rallying support for Ronald Reagan and other Republican candidates. But the gap between religious and non-religious in voting was actually less for Ronald Reagan—in both 1980 and 1984—than for Gerald Ford in 1976. As Glaeser and Ward (2006) point out, the recent political divisions associated with religious belief coincide with the geographic pattern of richer states supporting the Democrats and poorer states going Republican.

So religion matters. For whom does it matter? Does it matter for the frustrated masses, seduced by emotional issues, or for the less economically-pressed elites? We can answer the question by measuring the religious/secular gap among voters at different income levels.


Figure 2. Support for George W. Bush, as a function of income, plotted separately for frequent church attenders, moderate attenders, and non-attenders. The difference between rich and poor is large for religious people but disappears among the non-religious.

The difference in Republican support, comparing regular religious attendees to non-attendees, is huge for rich voters but low among the poor; see Figure 2. This result—that church attendance predicts voting more for the rich than the poor—is consistent with the finding of Ansolabehere, Rodden, and Snyder (2007) that “low-income Americans are significantly less inclined to vote based on moral values than are high-income groups.” They find the impact of economic issues on voting is larger for regular churchgoers, residents of Republican-leaning states, and rural voters than for non-churchgoers, residents of Democratic states, and urban or suburban voters.

To connect to Figure 2: the line for regular churchgoers is steep, while the line for non-attenders is flat. Thus, income predicts how you vote—if you are religious. We had earlier found that income predicted voting more in poor states than in rich states (Gelman, Shor, Bafumi, and Park, 2007). This again fits the story of post-materialism, that economic concerns are more important in poorer areas, with social and religious issues mattering more among the rich.

The United States is far from unique in having religion as a political division. Religious and secular voters differ no more in America than in France, Germany, Sweden, and many other European countries, consistent with the post-materialist notion that people in richer countries have the luxury of voting on social issues. Figure 3 tells the story.


Figure 3. For each of thirty countries (ordered by per-capita GDP), estimated vote for conservative parties by income and religious attendance. In each plot, the solid, light, and dotted lines show the estimates for frequent religious attenders, occasional attenders, and nonattenders. With only a few exceptions, churchgoers and higher-income people are much more likely to vote for conservative parties. The curve show logistic regression fits, which can be misleading because the actual pattern is not always smooth; for example, in some countries, middle-income people vote more conservatively than the rich or the poor. The purpose of this figure is to quickly show overall patterns of richer or more religious people voting conservatively in different countries.

Huber and Stanig (2007) and Huber (2007) noted that, within each country, the differences between rich and poor voting tend to be larger among religious voters; however, the differences between rich and poor—both in their voting patterns and in the size of the religion gap—are larger here than in most other countries, a finding also consistent with post-materialism. Incomes are more unequal in the United States than in other rich countries today, and so it makes sense that rich and poor vote more differently. And, indeed, the Democratic and Republican parties are farther apart on issues of economic redistribution than are left and right parties in most European countries.

Religious Americans are more Republican than secular Americans, but the difference between them is mostly among the middle class and rich—the “post-materialist” values voters. The evidence does not support the idea that lower-income Americans are voting based on “God, guns, and gays.”


This article is based on material in some of the material in chapters 6 and 7 of the forthcoming book, Red State, Blue State, Rich State, Poor State: Why Americans Vote the Way They Do, by Andrew Gelman, David Park, Boris Shor, Joseph Bafumi, and Jeronimo Cortina. Related material appears at

Our analyses used U.S. poll data from the National Election Study and the Annenberg Election Survey and international data from the Comparative Study of Electoral Systems. Self-reports of religious attendance can not necessarily be trusted (see Haraway, Marler, and Chaves, 1993, along with the follow-ups cited here. Nonetheless, we follow the usual practice in social science and work with the survey responses, assuming that people who say they attend church weekly are more religious than those who say they do not attend, whatever their actual practices.

Here are the other references cited, in order of appearance:

Webb, Jim (2006). Class struggle: American workers have a chance to be heard. Wall Street Journal, 15 November.

Brooks, David (2001). One nation, slightly divisible. Atlantic Monthly, December.

Inglehart, Ronald (1971). The silent revolution in post-industrial societies. American Political Science Review 65, 991-1017.

Frank, Thomas (2005). What’s the Matter With Kansas? How Conservatives Won the Heart of America. New York: Macmillan.

Shapiro, Walter (2005). What’s the matter with Central Park West? Atlantic Monthly, March.

Glaeser, Edward, and Ward, Bryce (2006). Myths and realities of American political geography. Unpublished paper.

Ansolabehere, Stephen, Rodden, Jonathan, and Snyder, James (2006). Purple America. Journal of Economic Perspectives 20, 97-118.

Gelman, Andrew, Shor, Boris, Bafumi, Joseph, and Park, David (2007). Rich state, poor state, red state, blue state: what’s the matter with Connecticut? Quarterly Journal of Political Science 2, 345-367.

Huber, John, and Stanig, Piero (2007). Why do the poor support right-wing parties? A cross-national analysis. Unpublished paper.

Huber, John (2007). Religious belief, religious participation, and social policy attitudes across countries. Unpublished paper.

8 thoughts on “The opiate of the elites

  1. Being a Swede, these graphs surpised me quite a lot. Especially the differences between Sweden, Denmark and Norway. The Scandinavian coutries are among the most secular countries in the world, so I would assume that religions would play a very minor role in ones political preferences. That appear to be the case in Norway and to a somewhat lesser degree in Denmark. In Sweden, however there appear to be a huge effect. We do have an openly religous party in Sweden (the Christian Democrats, as opposed to either Norway and Denmark), but their current support is howering around 4%. So do sample size play a role here or do religion actually play a much greater role in peoples lives in Sweden than I ever imagined? Interesting…

  2. PKI,

    Jeronimo should be able to answer this one; he's the one who analyzed the international data. A lot depends on which parties are coded as left and right within each country.

  3. PKI,

    I think you're reading a little too much from the graph. It shows only that more religious people vote differently than less religious people–it doesn't say anything about how important religion is in the country. Say only 4% of Swedes are regular church attenders, and that all of them vote for the Christian Democrats, but no one else does. Then there's a big difference in vote choice by religiosity, but religion is still marginal to the lives of the vast majority of Swedes.

  4. Hi Andrew,

    In figure 2 here, Mississippians are more likely to be found on the "poor" side, and New Yorkers are more likely to be found on the "rich" side. Put another way, P(state=MS|income=poor) > P(state=NY|income=poor).

    But the difference between these conditional probabilities probably varies depending on reported frequency of church attendance.

    Elsewhere, you've observed that in the big rich states (CA, NY, CT…), income is negatively correlated with reported frequency of church attendance, whereas in small poor states (MS, AR, AL…), the correlation is reversed.
    So it's reasonable to imagine that

    P(state=MS|income=poor & church=weekly) >>> P(state=NY|income=poor & church=weekly)
    P(state=MS|income=poor & church=monthly) >> P(state=NY|income=poor & church=monthly)
    P(state=MS|income=poor & church=never) > P(state=NY|income=poor & church=never)

    If this is true, then there is a bigger geographical shift in who is represented as one moves across the church=weekly line than there is in who is represented as one moves across the church=never line.

    The church=weekly line could then be characterized as "a subset of Americans for whom income predicts state of residence somewhat", and the church=never line as "a subset of Americans for whom income does not predict state of residence".

    Since geography is an important predictor of the probability of voting for Bush, one expects the slope of the church=weekly line in Figure 2 to be greater than the slope of the church=never line.

    I'm not questioning the claim that nationally for 2004 voters, the effect of income on vote was greater for reported church-goers. But is religiosity what mattered?

    Many variables could produce a three-way division of Americans with the same "predicts geography or not" categories as you have here. Some likely ones are "number of coats owned," "body mass index," and race. This makes me wonder whether it's wise to focus on the reported frequency of attendance as an issue.

    I would be interested to see how things look if you could measure the effect of reported church attendence more specifically. (As some would describe it, by "controlling for geography" and other variables that correlate with reported church attendance and would explain the increasing income-vote slopes you see on their own). Simulations might be able to show whether this is an issue at all.

    The other issue I think is relevant in these discussions has been mentioned nicely here:… I'll just add that it's somewhat related to the observation I recently made about the "increasing life-expectancy gap". On the five-point income scale you frequently use, a -2 is probably poor no matter where you are in the US. But a +2 in MS is likely a "truly weathly Mississipian" who has a very high standing in the community, whereas a +2 in New Jersey might be a typical suburban worker. As a result, -2 and +2 are far apart in Mississippi, but relatively close in New Jersey. This is more or less the case for different church-going categories, even within a state. (Visual: Picture a rich Mississippian who goes to church every week among the congregation; then picture a rich New Jerseyan who goes to church every week among the congregation. Is the picture the same?) Adding state or county income averages into the model doesn't address this problem, I don't think. You need not only varying-slope, varying-intercept, but also varying-horizontal-scale and varying-horizontal-zero.

    It's not impossible to imagine that the "true" horizontal income scale is not the same for the three lines in Figure 2, and the right segment of the plot's box should really be titled to the left (and the lines squished accordingly). The slopes wouldn't portray the same story if income=-2 is the same as true_income=2, but "rich" means true_income = 0 for church=monthly, while it means true_income=2 for church=weekly.


  5. The difficulty I have with Figure 3 is how it fits into the rest of the text: It starts with the interesting finding that in the US religion appears to matter less for poor people then for rich people, that is, there is a difference in slopes. Figure 3 seems to suggest that this pattern is most apperent in the US, and that the lines are approximately paralel in many countries. The US being an exceptional case would be an interesting conclusion, but I don't see that in the text. Am I misreading figure 3?

  6. "Religious Americans are more Republican than secular Americans, but the difference between them is mostly among the middle class and rich—the “post-materialist” values voters. The evidence does not support the idea that lower-income Americans are voting based on “God, guns, and gays.”"

    "The evidence" or "that evidence"? It seems to me that people like Webb and Obama are positting that a subset of poor voters, non-urban voters who vote republican, are doing so because at the time that they vote they're prioritizing religious and gun-owning (and a couple others that may vary between Obama, Webb and others) values over their perceived or actual economic self-interest.

    It seems that this could still be the case even if an equal number of poor religious people vote democrat, and even if religious rich people vote disproportionately Republican.

    The argument seems to be that poor non-urban people would vote Democrat in much larger numbers if they valued their perceived or actual economic interest over their religious and gun-owning values.

    So, it seems to me that a reasonable place to start is polls of poor non-urban people who voted Republican, asking them to rank what influenced their vote, and which party is better in their opinion on particular issues (their economic interest, their religious values, gun owning values, etc.).

  7. Steve,

    Yes, we have some three-way plots (income, religious attendance, and state) in the book. Regarding your comment about relative income: the variation among incomes is much greater within than between stats, and in any case the relative-income argument doesn't really explain the different slopes for different groups. But I'm sure that the ideas you're mentioning are part of the story. When income predicts voting, we're not saying income is causing people to vote in a certain way. Ultimately, it's correlated with variables such as issue attitudes and ideology that are associated with voting for Democrats or Republicans.


    Our main point with Figure 3 was that, contrary to what is sometimes believed, religion is not uniquely divisive in American politics. Religious attendance is correlated with conservative voting in many places. (Also, I thought it was appropriate to put some European data in my Vox EU column.)


    I'm sure that a lot of people at all income levels vote for reasons of God, guns, and gays. But the data I've seen don't support the idea that it's lower-income people who are doing it. I agree completely that what these polls are telling us is relative proportions.

    Regarding your specific question, we have looked at the "What is the most important issue to you" question in the National Election Study, but we couldn't get much out of it.

  8. I think figure 3 is interesting for the reasons I have given before, but I don't think it gives very compelling evidence that religion is also divisive in other countries. This figure shows that people who frequently visit church are more likely to vote conservative, but this distinction is not going to matter if there are very few people who frequently attent church, as is the case in the Netherlands and apperently also in Sweden. So the fact that the lines are further appart for both the Netherlands and Sweden than in the US does not mean that religion is more divisive in Sweden and the Netherlands than in the US.

Comments are closed.