The computational approach has become an invaluable tool in many fields that are directly relevant to research in religious phenomena. Yet the use of computational tools is almost absent in the study of religion. Given that religion is a cluster of interrelated phenomena and that research concerning these phenomena should strive for multilevel analysis, this article argues that the computational approach offers new methodological and theoretical opportunities to the study of religion. We argue that the computational approach offers 1.) An intermediary step between any theoretical construct and its targeted empirical space and 2.) a new kind of data which allows the researcher to observe abstract constructs, estimate likely outcomes, and optimize empirical designs. Because sophisticated multilevel research is a collaborative project we also seek to introduce to scholars of religion some general computational issues, and finally applications that model behavior in religious contexts.

Much experimental evidence shows that minimally counterintuitive concepts, which violate one intuitive ontological expectation of domain-specifi c natural kinds, are remembered as well as or better than intuitive concepts with no violations of ontological expectations, and much better than maximally counterintuitive concepts with more than one violation of ontological violations (Barrett and Nyhoff , 2001;Boyer and Ramble, 2001;Atran and Norenzayan, 2004;Gonce et al., 2006). It is also well established that concepts rated as high in imagery, (e.g., apple) are recalled better than concepts that are low in imagery (e.g., justice; see Paivio, 1990). We conducted three studies to test whether imagery levels aff ected recall rates of intuitive, minimally counterintuitive, and maximally counterintuitive concepts. In study 1, we obtained imagery level ratings for 48 three-word items. In study 2, we used the ratings obtained in study 1 in a 2 × 3 recall task in which imagery (high vs. low) was manipulated along with counterintutiveness (intuitive vs. minimally counterintuitive vs. maximally counterintuitive). High imagery items were recalled signifi cantly better than low imagery items for intuitive and maximally counterintuitive items but not for minimally counterintuitive items. Study 3, replicated the fi ndings from study 2 in a 2 × 2 study using a larger number of intuitive and minimally counterintuitive items. In both studies, High imagery items were recalled signifi cantly better than low imagery items for intuitive but not for minimally counterintuitive items. Th us, minimally counterintuitive concepts appear insulated from imagery eff ects on recall.

Owning up to negative group traits is facilitated by autonomy and demonstrates benefits for ingroup and intergroup processes.

Detection of fine-grained opinions and beliefs holds promise for improved social media analysis for social science research, business intelligence, and government decision-makers. While commercial applications focus on mapping landscapes of opinions towards brands and products, our goal is to map "sociostructural" landscapes of perceptions of social groups. In this work, we focus on the detection of views of social group status differences. We report an analysis of methods for detecting views of the legitimacy of income inequality in the U.S. from online discussions, and demonstrate detection rates competitive with results from similar tasks such as debate stance classification.