People Who Believe In Heaven Commit More Crimes

This picture decreases rape rates
Heaven and hell

Some people who really ought to know better—but don’t—reported on the peer-reviewed paper “Divergent Effects of Beliefs in Heaven and Hell on National Crime Rates” by Azim F. Shariff and Mijke Rhemtulla and have concluded two contradictory findings.

  1. Belief in Heaven raises the crime rate
  2. Belief in Hell lowers the crime rate

The authors put it thusly: “Supernatural benevolence…may actually be associated with less prosocial behavior.” They say their findings “raise important questions about the potential impact of religious beliefs on global crime.”

Even so once an august news agency as CBS (which gave us today’s headline) reports “Believing if you are on a ‘highway to hell’ could impact whether or not if you commit a crime.” This matches the breathless coverage in many other places, proving once more that statistics should not be practiced without a license. Here’s what happened.

Our stalwart statistician want-to-bes took data from 1981-1984, a time long ago, a time when the Berlin Wall still held back the throngs of Westerners seeking to join the East German socialist paradise, and then they took more data from 1990-1993, and then again from 1994-199, and then still more from 1994-1999, and, would you know it?, still more from 1999-20004, and finally, not satisfied with all that, more data from 2005-2007.

The “data” consisted of answers from the World Values Surveys and European Value Surveys (different surveys, as in different). Citizens in 67 countries were asked questions on religious beliefs and other matters, such as their very own personal “Big Five Inventory of personality differences.” Only it wasn’t 67 countries. Sometimes it was 56 countries, others times 48, and yet other times 46, and then 51, 48, 47, 43, and even a lowly 39. Obviously—and the authors must agree, since they remain mute on this subject—nothing could have changed in these beliefs, not in nature or proportion, from 1981 to 2007.

They asked citizens Do you believe, yes or no, in “Heaven,” “Hell,” and “God.” They asked how many times do you attend religious services. They computed murder rates, robbery rates, and rape rates (in France in honor of film artist Roman Polanski, this was presumably modified to rape-rape rates). They even looked at car theft rates and human trafficking rates. No word on whether these measured rates matched the same years as the demographic data (I’d guess not).

And then they, yes, computed a “weighted average” for each of all these variables—even the variables “conscientiousness, agreeableness and neuroticism”—for each country.

And then? Then the real magic. All of these strange “variables were entered into a series of linear regression equations, with each crime regressed on beliefs in heaven and hell as well as all covariates.”

You know what’s coming next, dear reader. You know the thrills that await. For you, loyal visitor, know better than others the excitement that comes with tables filled with numbers with asterisks. For those asterisks lead to footnotes, and in those footnotes are found the joy of the small p-value, and in wee p-values are found promotions, glory, and honor.

To test the theories attested to above, regression models were used. Now the regression coefficient (see below if you don’t understand these metaphysical creations) between belief in Heaven and assault rate is 1.7, a number given not just one, not a mere two, but three whole asterisks! But don’t claim triumph yet, because the regression coefficient between belief in Hell and assault rate is -1.7, which also has three asterisks.

Steady on, dear one. Because if my math is right, this means the model coefficient adjustment for somebody who believes in both Heaven and Hell (which are most people who believe in either) is 1.7 – 1.7 = 0. Never mind, never mind. It’s the wee p-values that count. At least as far as publishing articles goes.

Yet there’s something screwy in my analysis, because let’s recall. This isn’t an individual’s belief in Heaven or Hell: it’s a nation’s (sort-of) weighted average of belief versus a nation’s (sort-of) weighted average crime rate. Yet this doesn’t stop our authors from writing: “Belief in hell predicted lower [overall] crime rates [coefficient -1.9, p < .001], whereas belief in heaven predicted higher crime rates [coefficient 1.9, p < .001].”

Which is still a wash if, as if very likely, the weighted (sort-of) average of Heaven believers equals the weighted (sort-of) average of Hell believers inside a country. The reason I’m probably right about this is found in their Table 1: all the positive coefficients of Heaven belief are matched by more or less equal negatives coefficients for Hell belief. Even worse for our authors’ theory, nearly all of these coefficients are similarly sized, and this is so even after “adjusting” for covariates like “Urbanicity” and “Life expectancy.”

Am I right? Gallup in 2004 reports that about 80% of Americans believe in Heaven and 70% in Hell. (The USA wasn’t in this data, of course.) And this site shows that belief in Heaven and Hell is, as we thought, nearly the same in a wide variety of countries. Significantly, it is nowhere wildly disparate.

Now a regression is a model of the central parameter of a normal distribution (ND), a distribution which is used to quantify uncertainty in some thing, like crime rate. We can write their unadjusted model like this (stick with me):

     central parameter ND for crime rate = b0 + b1Heaven + b2Hell

where we substitute in a nation’s (sort-of) weighted average of Heaven and Hell belief. If b1 = b2, as the estimates do in this paper, and if Heaven and Hell percentages are equal, as they are in most places, then the equations simplifies to

     central parameter ND for crime rate = b0

where b0 is just some number which is of no interest to anybody.

You cannot just examine b1 alone or b2 alone: you must look at both simultaneously. The entire study is thus a wash, nearly certainly a figment of (unconscious) data manipulation. Ignoring all the other ways the study goes wrong (mixing years wantonly, etc.), what would prove me mistaken is if the (sort-of) weighted averages beliefs in Heaven and Hell were not about the same, but were everywhere (or in most places) different. That is not likely, as we have seen.

And that’s just the misinterpreted statistics. Don’t get me started on discussions like this one given by the authors in defense of their theory:

Divine punishment, on the other hand, has emerged as a cultural tool to overcome a number of those limitations. Unlike humans, divine punishers can be omniscient, omnipotent, infallible, and untouchable-and therefore able to effectively deter transgressors who may for whatever reason be undeterred by earthly policing systems.

Good grief!


Thanks to Mike Flynn who suggested this (depressing) topic.


  1. Noblesse Oblige

    Yes, Good grief!

    Did the authors worry about Pascal’s Bet, which might suggest an opposite conclusion: You have nothing to lose and something to gain if you don’t commit crimes.

  2. intrepid_wanders

    It is a shame to be so close, but to sin and miss the mark. The authors were so close to tying their coefficients to the golden ratio((1 + sqrt(5))/2). If that were to occur, the connection to the golden rule could not be far behind. Sad day, indeed.

  3. JH

    According to Figure 1 in the paper, in general, countries whose dominant religions are Musilin (blue dots) and other (red dots) have smaller differences between the proportion of heaven believers and the proportion of hell believers. (There is a yellow dot labeled as “US” with an approximate difference of 10%.) They seem to have a lower crime rate. Interesting! Is it because people don’t report crime as much in those countries?

    We know that a regression model does NOT imply causation. So, which one of the following describes a cause-and-effect relationship?
    A) “Belief in Heaven raises the crime rate. Belief in Hell lowers the crime rate.”
    B) “Here, we investigate these effects at the societal level, showing that the proportion of people who believe in hell negatively predicts national crime rates whereas belief in heaven predicts higher crime rates.”

    Note the unit of analysis is country, not individual. BTW, although we’d never know if b1=b2, if indeed b1 = b2, then the equations simplifies to Mean (crime rate| Heaven, Hell) = b0 + b (Heaven + Hell).

  4. JH

    If indeed b1 = b2 = b, then the equations simplifies to Mean (crime rate| Heaven, Hell) = b0 + b *(Heaven + Hell).

    I want to define b clearly.

  5. JH

    I don’t usually read your tweets.

    @SusanFelicity @deBeauxOs1 @ChesterScoville Still awaiting your *relevant* criticisms. Have a go. If you dare.

    What is this all about?

  6. Luis Dias

    IDK JH, seems like someone didn’t like the debunkin’ taking place here. Gosh how I hate that “debunking” word. Internet is so filled with it “I’m gonna debunk his arse like there’s no tomorrow”

  7. Briggs


    Click on the “View conversation” and you’ll see the string of tweets. It’s not especially fascinating, however.

    The figure you mention is not the data, but “residuals” (of a sort) and of no interest given the other flaws.


    I agree and will from now seek a better word.

  8. JH

    Mr. Briggs,

    No, they are not residuals of any sort. The vertical axis is the standardized crime rate (data).

    Figure 1. Crime rate z-scores as a function of how much higher the proportion of a nation that believes in heaven is compared to the proportion that believes in hell. R2 = .54.

  9. Briggs


    Oh geez. Let’s don’t argue over trivia. Is my main criticism of this paper right or wrong?

  10. JH

    Mr. Briggs,

    Trivia? I disagree because whether my conclusion is correct hinges on it. A different vertical axis tells a totally different story in this case.

    What is your main criticism? I see no real debunking of anything because you have misinterpreted their results (I must have not been clear on this point), and you have not analyzed the data. Yes, there might be multi-collinearity in the data, but speculations don’t prove anything.

  11. Briggs


    So you agree that my analysis is right. (If you cannot see, look harder.)

    I offered lots of evidence (and more is within an easy search) to show that “multi-collinearity” is right (the near equal belief rates of Heaven and Hell). Have you any to offer it is wrong? Because speculation that it might be wrong?

  12. JH

    Mr. Briggs,

    Where is your analysis? Are you saying that you have the data? That there might or might not be multi-collinearity in the data is not a claim or a speculation. It’s just a simple concept of data analysis.

    The differences in percentage of people who believe in heaven compared to hell are NOT nearly the same (nor all positive) in a wide variety of countries… unless “nearly the same” =”a difference within 30%;” see the graph in the link you referred to.

    Here is their main conclusion.

    “Here, we investigate these effects at the societal level, showing that the proportion of people who believe in hell negatively predicts national crime rates whereas belief in heaven predicts higher crime rates.”

    Show me why this conclusion is incorrect. How about just giving me two scatter plots of (Crime Rate versus Heaven) and (Crime Rate versus Hell)?

    I have no interest in debunking anything. If you don’t get my point, so be it.

  13. Briggs


    You win, JH. I think you ought to use this example to teach your students “How To Do Regression Properly.”

  14. JH

    Mr. Briggs,

    “How to use Regression Properly?”

    Yes, this might be a good example for both classic and Bayesian regression analysis. Even just for basic demonstrations and interpretations of descriptive statistics. I’d have to try it out. On second though, I need to ask for the data though…, no, too much trouble to ask people for data.

    When you grade a report or referee a paper, do you think about “winning?” I don’t.

  15. Briggs


    The good news is that your students will find success publishing!

  16. Ye Olde Statistician

    Shoot, I once found a near-perfect correlation between the %imported passenger cars sold in the US (Y) and the %women in the labor force (X). I would be more interested in how one does a correlation in which X and Y are not actually measured on the same units but on aggregates. The “crime rate” and “belief in heaven/hell” are compiled on countries, which is a little like correlating the height of a tree (Y) with the girth of a tree (X) where Y and X are measured on different samples of different trees at different times, and not on the self-same tree. IOW, did the people who committed the crimes hold those beliefs, or were the beliefs held by other people, not committing the crimes, who happened to live in the same country?

    The second question is definitional. Different jurisdictions define crimes differently, set thresholds differently, etc. Grand theft auto may be low in countries where there are few autos. There is also the issue of reported crime vs. actual crime. In some countries, crimes may be under-reported more than than in others. Some countries in the past have been known to jigger their published crime rates for political or ideological reasons.

    Then one has the possibility that a crook may eschew a belief in hell out of self-interest. He does not want to believe that he will be punished and so withholds belief. The more crooks there are in a country, the greater the gap between heaven-belief and hell-belief.
    + + +

    The Z-score is a standardized residual for the model Y=Y-bar, nicht wahr?

  17. cb

    If people think there are serious negative consequences to some or other action, they WILL be MUCH less likely to do it. If anyone thinks this has to be ‘proved’ via statistics, then please go shoot yourself. In the head, at least thrice. With a shotgun. And then put a stick of dynamite in your mouth and light it.

    This ‘study’ should have started by considering the above, and then seeing if new-agers (heaven-4-all; there is no spoon, er, hell) committed more crimes than Christians (heaven-4-few; yes there is a hell). Another test would HAVE to be done, then, is considered whether atheists (both schools: the self-identified, AND those who ‘believe nothing supernatural exists’) commit more crimes than Christians (repeat with atheists vs new-agers.)

    I say Christians in the above, since the behaviors of other groups (example those who love Pieces), are often legally criminal in Western countries. This aspect cannot be ignored: the Piecers, for example, have an actual Holy Duty to be criminal towards non-Piecers – which makes a study unnecessary.

    For fun, one should also consider, oh lets say, they hippies: who have allowed and are allowing a very great many deaths due to the complete ban on DDT. Those are DEAD people – yet the hippies smile and wave, smile and wave. Socialism-communism… more mega-deaths, more smiling and waving.
    Mmmm, well, I don’t think a ‘study’ is necessary to realize that hippies are super-criminals; as are politicians, the PC, etc.

    In fact, I would imagine that the ultimate purpose of such studies are to disallow the peasants to draw common-sense conclusions: only the Intellectual Elites are allowed to have REAL opinions, after all.

Leave a Reply

Your email address will not be published. Required fields are marked *