Religion and Economic Growth across Countries

Let's start with the puzzle that wouldn't quite fit into the usual growth stories. If richer countries tend to be less religious, is that because prosperity erodes faith, or because certain religious traits actually help economies grow? Barro and McCleary went after that knot directly. They didn't settle for simple correlations. They asked a sharper question: when you separate "believing" from "belonging," do doctrinal beliefs like heaven and hell push growth one way, while churchgoing pulls it another? And then they built the machinery to make that answer credible. Their data come from an unusual synthesis: six international survey waves between the early nineteen eighties and nineteen ninety-nine, stitched together from the World Values Survey, the International Social Survey Programme, and the Gallup Millennium Survey. That gives them country-level fractions for two things that matter here—monthly church attendance and beliefs in heaven or hell—across as many as fifty-nine countries. Those religiosity measures are matched to three decade-long growth spells of real income per person—from nineteen sixty-five to nineteen seventy-five, from nineteen seventy-five to nineteen eighty-five, and from nineteen eighty-five to nineteen ninety-five—along with the usual growth controls. Everything is averaged at the country level because the question is macro: how does a nation's religious environment line up with its economic trajectory? There's a finance-style twist to the way they handle the religion variables. Each is transformed using a logit—take the log of x divided by one minus x, where x is the fraction attending or believing. That keeps predicted values on a sensible zero-to-one scale and makes the slope easy to read. In this setup, the marginal effect on the original fraction equals the coefficient times x times one minus x. You don't need to love the algebra to get the point. It's a way to talk about percentage-of-people variables in a linear world without breaking the math. Now, the elephant in the room is causality. Religiosity might respond to growth, and not the other way around. So they instrument. The instruments are deliberately institutional and slow-moving: whether a country had an official state religion around nineteen seventy, whether the state appointed or approved religious leaders in the nineteen seventies, and how religiously plural the country was around nineteen eighty, measured as one minus the Herfindahl index across major religions. They also use the composition of adherence—shares of Protestant, Muslim, Hindu, and so on, with Catholic as the baseline—as predictors of religiosity. These variables are strong at explaining attendance and beliefs—joint significance is about as decisive as it gets—and, crucially, they don't look like direct levers on short-run growth. When they check the over-identifying restrictions—the test that asks, "Do these instruments sneak into the growth equation directly?"—the answer is no; the p-value sits around 0.25, which is consistent with the exclusion story they want to tell. If you're wondering why these features make good instruments, this is the logic: they're powerful predictors of religiosity, they move slowly, and they're unlikely to be driven by a decade's worth of gross domestic product growth. Before we get to growth, it helps to know what religiosity looks like in the cross-section. The raw correlations with income are negative: richer places tend to have lower churchgoing and lower reported beliefs. But the partial patterns are more nuanced. Education actually raises participation once you hold income, urbanization, life expectancy, and age structure constant. The coefficient on schooling in the attendance equation is about 0.19 in the logit scale; walking that into plain terms, an extra couple of years of average schooling—roughly the spread you see across countries—lines up with something like a ten-point increase in monthly attendance from a mean of a third of adults. Urbanization pulls the other way. A move from a more rural to a more urban country—say, an increase by 0.15 on the urbanization share—goes with about a five-point drop in attendance. Beliefs track with modernization too: more education is associated with stronger beliefs in heaven and hell, while more urbanization and longer lives correlate with weaker ones. Younger populations lean more religious than older ones in these data, especially on attendance. The religious marketplace itself matters. More plural countries—places where no single church dominates—have higher attendance and stronger beliefs, even after you net out development and demographics. The pluralism index comes in positively for all three religiosity dimensions. State arrangements bite hard as well. Where there's an official church, attendance and beliefs are higher. Where the government regulates religion—appointing or approving religious leaders—they're lower. And the Cold War imprint is unmistakable: Communist regimes show large negative associations with both participation and belief, followed by a modest rebound in ex-Communist countries late in the nineteen nineties. If you're wondering why these features make good instruments, this is the logic: they're powerful predictors of religiosity, they move slowly, and they're unlikely to be driven by a decade's worth of gross domestic product growth. Okay, the payoff. In the growth regressions—with all the baseline macro controls you'd expect: starting income, schooling, urbanization, life expectancy, age shares, investment, fertility, openness, terms of trade, inflation, governance—the contrast between believing and belonging is stark. When they instrument for religiosity and include monthly church attendance and belief in hell together, the coefficient on attendance is negative, and the coefficient on hell is positive. The headline numbers are tidy mirror images: about minus 0.0095 on attendance and plus 0.0094 on hell, each statistically tight. Translate with their standard deviations in the logit world—roughly 1.20 for attendance and 1.06 for hell—and you get this: a one-standard-deviation increase in monthly churchgoing is associated with about a 1.1 percentage-point lower annual growth rate, while a one-standard-deviation increase in belief in hell maps to about a 1.0 point higher rate. If you worry that religious composition is doing the work—maybe countries with more Protestants just happen to grow faster—they try that too. Include the religion shares directly in the growth equation, and the pattern holds. Attendance gets more negative, about minus 0.016, and hell gets more positive, about plus 0.014; the pair is jointly significant at the one-in-a-thousand level. Switch from hell to heaven as the belief measure, and you see a similar but gentler version of the same thing: attendance is still negative—around minus 0.010 without the shares—and belief in heaven is positive but smaller, roughly 0.0069. Add the composition shares, and the signs stay put; magnitudes nudge a bit. What happens when you throw both beliefs into the mix at once? Hell still carries the signal. In the joint specification with attendance, hell's coefficient is about 0.010 and heaven hovers near zero. In the version that also includes the religion shares, hell rises to roughly 0.017, heaven to a small positive, and attendance becomes more negative. Tests that ask if the heaven and hell coefficients are equal don't quite clear the statistical bar. P-values sit around 0.12 to 0.13, so the safer reading is directional: hell tends to show the stronger link to growth, heaven the weaker one, and churchgoing tugs downward throughout. They also rotate the belief variable. Use belief in an afterlife instead of the specific doctrines, and the pattern stays intact: attendance negative, afterlife belief positive. Bring in belief in God, and the wheels come off—belief in God is small and indistinguishable from zero once you control for attendance. And if you swap in a self-described "religiousness" measure, it behaves like attendance: negative, not positive. One clever check asks whether the power of hell beliefs differs by religious tradition. You might imagine that hell in a majority-Protestant country, or a majority-Muslim one, would mean different things for behavior. They allow interactions between hell belief and religious composition, and they don't get significant differences. The p-values—0.76 in one extended run, 0.98 in another—say no. That's a useful non-result: the growth link runs through believing, not through a specific denominational mix. Under the hood, a couple of design choices keep the estimates from drifting. On the religiosity side, they estimate three equations—attendance, belief in heaven, belief in hell—together using seemingly unrelated regressions, so they can handle correlated errors across survey waves and sources. They include intercepts by data source so you don't confuse a Gallup quirk for a cultural shift. On the growth side, they use three-stage least squares with beginning-of-period instruments, plus those institutional religion variables and the composition shares. And they line up timing so that the variables meant to be "causes" come before the growth spells. None of this makes the problem trivial, but it does keep reverse causation and omitted-variable bias from swallowing the conclusions. What does all this mean in human terms? Barro and McCleary's read is that beliefs function like outputs of a religion sector. Think norms about honesty, thrift, and diligence—traits we know from other work can raise productivity. Attendance, by contrast, looks more like an input cost. Time, attention, and resources devoted to communal practice may help sustain the belief system, but conditional on holding those beliefs fixed, more attendance shows up as a drag on growth. You don't have to buy a particular theology to see the economic distinction: ideas can be productivity-enhancing; resource use can be costly. There are, as always, caveats. Instruments that look exogenous over a decade might have their own long-run feedback loops. Religious shares could, in principle, affect growth beyond their role in shaping beliefs and attendance, despite that 0.25 over-identification p-value that says the exclusion story isn't rejected. Measurement isn't perfect; some waves are richer for Christian-majority countries and thinner elsewhere, and beliefs in heaven and hell aren't available in every survey round. And the headline numbers are average effects across very different places. They're not destiny for any given country. But the cross-checks are reassuring. Instruments are jointly powerful in predicting religiosity—those first-stage tests light up. Include the composition shares directly, and the believing-versus-belonging pattern persists. Swap belief measures, and the qualitative story stays put. Even the crude correlations that once seemed to show "growth kills religion" turn out, in the multivariate view, to hide a more interesting split: development reshapes how people practice and what they affirm, and those two dimensions point growth in opposite directions. If you're listening for policy, tread lightly. This is not an argument to engineer beliefs, and the authors don't go there. It does say: when thinking about the cultural underpinnings of growth, pay attention to the content of beliefs, not just the intensity of participation. It also says: the institutional setting—state churches, regulation, pluralism—shapes religiosity in predictable ways, and those ways can be used to parse cause from effect. Where could we learn more? Two directions feel especially promising. One is history: richer, longer-run data on church-state arrangements and religious competition would tighten the instrument story and let us see whether the growth links hold over centuries, not decades. The other is micro-to-macro. If beliefs about the afterlife change saving, effort, or trust at the individual level—as many lab and field studies suggest—then building up from those mechanisms to the aggregate growth effect would give this story even firmer footing. For now, the cleanest takeaway is simple to say and subtle to act on: when it comes to growth, what people believe seems to matter more than how often they show up. Believing and belonging are not the same economic thing. And once you can tell them apart, the data start to make sense.