UBC undergrads think atheists are less trustworthy than rapists

In this recently published paper, UBC psychologist Will Gervais and his coauthors consider the causes of prejudice against atheists. In one of their studies, the authors tell UBC undergraduate subjects about a jerk called Richard:

Richard is 31 years old. On his way to work one day, he accidentally backed his car into a parked van. Because pedestrians were watching, he got out of his car. He pretended to write down his insurance information. He then tucked the blank note into the van’s window before getting back into his car and driving away.

Later the same day, Richard found a wallet on the sidewalk. Nobody was looking, so he took all of the money out of the wallet. He then threw the wallet in a trash can.

Respondents are then asked which is more likely:

    (a) Richard is a teacher.
    (b) Richard is a teacher and XXXXX.

XXXX is randomized across respondents and is either “a Christian,” “a Muslim,” “a rapist,” or “an atheist.”

The idea here is that simply asking people about prejudices can be misleading, as no one likes to think that they are prejudiced. The correct answer to the question is always response (a): the set of people who are teachers and XXXXX can never be larger than, and will generally be smaller than, the set of people who are teachers. People responding (b) are committing the conjunctive fallacy. According to the paper, people commit this fallacy when the group membership (XXXXX) is “deemed representative” of the description (of, here, Richard).

The results:

As the authors note, Vancouver is one of the most liberal and secular areas of North America. Yet even in Vancouver, and at UBC, the graph shows that respondents are more prejudiced against atheists than against rapists.

Robin Hanson wonders whether atheists actually are less trustworthy, guessing probably so, but not as much as people think. A cursory scan of the relevant research suggests otherwise, although causation is very difficult to uncover in this case. For example, Heaton (2006):

Considerable research in sociology, criminology, and economics aims to understand the effect of religiosity on crime. Many sociological theories positing a deterrent effect of religion on crime are empirically examined using ordinary least-squares (OLS) regressions of crime measures on measures of religiosity. Most previous studies have found a negative effect of religion on crime using OLS, a result I am able to replicate using county-level data on religious membership and crime rates. If crime affects religious participation, however, OLS coefficients in this context suffer from endogeneity bias. Using historic religiosity as an instrument for current religious participation, I find a negligible effect of religion on crime and a negative effect of crime on religion. To further explore the relationship between religion and crime, I examine variation in crime incidence before and after Easter. Consistent with the IV results, I find no evidence of a decrease in crime following Easter.

10 Responses to “UBC undergrads think atheists are less trustworthy than rapists”

  1. Chris, am I missing something really obvious here? Since a tiny fraction of the population (I hope) are rapists, but atheists are – what – 10, 20, more – percent of the population, the person is of course much more likely to be a teacher and an atheist than a teacher and a rapist. Even the numbers given in your paper suggest that people believe that teacher/rapists are, relative to their representation in the population, much more likely to commit hit and runs than teacher/atheists. Is this really taken as evidence of discrimination against atheists?

  2. “Yet even in Vancouver, and at UBC, the graph shows that respondents are more prejudiced against atheists than against rapists.”

    Actually the graph shows that there are no statistically significant differences between atheists and rapists. They didn’t really talk about these differences as if they were significant, did they?

  3. Frances, good point, I don’t think that population frequencies (among 31 year old males named Richard) of the various religious groups are discussed in the paper. But, since answering (b) contradicts the axioms of probability, I don’t think it’s meaningful to try to interpret the pattern of incorrect answers assuming some coherent probabilistic model for the students’ reasoning.

    This type of question is apparently a standard preference revelation, as we might call it, tool in experimental psychology, although I didn’t bother reading background papers to see how controversial it is.

    Suppose we ignore the “teacher &” part, and wave our hands a little, and pretend the responses are estimates of Pr ( group | criminal ). That would help us explain why the estimates suggest Pr ( atheist | criminal ) > Pr ( rapist | criminal ), but notice from the graph that we would then conclude that the students believe Pr ( atheist | criminal ) is about an order of magnitude greater than Pr ( Christian | criminal ). Since the proportion of Christians in the population is roughly an order of magnitude greater than the proportion of atheists, we conclude, using Bayes, that students believe an atheist is two orders of magnitude more likely to be criminally selfish than a Christian. Also, if students believe the proportion of atheists in the 31 year old male population is similar to the proportion of rapists in that population, we would also infer that students believe atheists and rapists are roughly equally likely to be criminals. However, again, all of this is a big stretch, as we’re using probability theory to interpret responses which cannot be squared with probabilistic reasoning.

    On your second post: I was referring to point estimates; the paper does not suggest the difference between atheists and rapists in that graph is statistically significant.

  4. I think Frances’ point is right. It’s surely unwise to assume that because subjects suffer from a probabilistic fallacy in one context, they never use probabilistic reasoning. The argument that rapists are few is intuitive.

  5. Bizarre, would never have thought of Canada as so irrational (the US, of course, are barmy. I just always assumed that Canadians were saner). Would be interested to know the results of such a question in countries where most of the population is atheist, or functionally so in spite of going to church, such as Britain or other countries in Europe.

  6. Does the ordering of the options impact the ratios?

    Do society norms interfere with a valid interpretation of the data?

    For example, we all have received rigorous training about the ills of discrimination based on race or religion — so those two are out. That leaves rapists and atheists. Theoretically, societal norms concerning violence and sexual crimes could push some to choose atheists as the safest or most comfortable remaining option available.

    Obviously, I’m just throwing around some wild speculation. It would be interesting to figure out “where” the result is coming from.

  7. > the graph shows that respondents are more prejudiced against atheists than against rapists.

    … ? really?

    Do you guys know the meaning of the error bars in the graph?

    The graph explicitly states that atheist, rapist and muslims have the same probability. Only christians have a somewhat lower probability but still are indistinguishable from muslims.

  8. The graph does not show that those groups “have the same probability.” That conclusion hinges on committing the logical error of accepting the null hypothesis. Rather, the point estimates suggest that atheists are least trustworthy, followed by rapists (odds ratio 1.27). As I note above, the difference between rapists and Muslims is not statistically significant, but the difference between atheists and the other groups is: contrary to another common logical error in hypothesis testing, the overlapping confidence intervals do not imply that we fail to reject the null. By the same token, contrary to your claim, we cannot tell from the results presented in the paper whether the difference between Christians and Muslims is statistically significant. Finally, the point estimate for Christians is less than 5% whereas it is close to 50% for atheists—referring to that gap as “somewhat lower probability” does not seem very accurate.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s