Measuring supernatural belief implicitly using the Affect Misattribution Procedure

ABSTRACT

Asking about religious beliefs, or lack thereof, is a sensitive and complex issue. Due to cultural norms, people may be motivated to respond in a socially desirable way. In addition, deliberating about beliefs may yield different responses than intuition-based responses. To develop a better understanding of the relationship between intuition and self-reported belief, we developed a new implicit measure of supernatural belief. Specifically, we adapted the Affective Misattribution Procedure (AMP) to measure supernatural belief. In a preregistered online study of 404 American participants, we found that the strength of associations between supernatural entities (e.g., god, devil, heaven) and the concept “real” (as opposed to the concept “imaginary”) predicted self-reported supernatural belief and self-reported religious behavior, and these associations were of comparable magnitude to those found in studies where supernatural belief was measured implicitly using the Implicit Association Test (IAT). These results provide provisional evidence that the AMP can be used as an implicit measure of supernatural belief.

KEYWORDS:

• Affect Misattribution Procedure
• belief
• implicit
• prime
• religiosity
• Semantic Misattribution Procedure
• supernatural

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 Here we list all published studies that we could find that used the IAT (either the original or “single target” variant) as an implicit measure of supernatural belief. The IAT has also been used to study other religion-related topics, such as, implicit religiosity/spirituality (Bachmann, 2014; Crescentini, Urgesi, Campanella, Eleopra, & Fabbro, 2014; Klein, Hood, Silver, Keller, & Streib, 2016; LaBouff, Rowatt, Johnson, Thedford, & Tsang, 2010; Wenger & Yarbrough, 2005), categorization of concepts as religious versus paranormal (Weeks, Weeks, & Daniel, 2008; Weeks & Gilmore, 2017), and paranormal beliefs not typically promoted by established religions but more associated with “New Age” beliefs (e.g., witchcraft, telepathy, divination; Stieger & Hergovich, 2013). Given our specific focus on using implicit measures to study supernatural belief, we do not further discuss these other (related) uses of implicit measures in the literature.

2 r(59) = .31, 95% CI [.06, .52], p = .01 (Shariff et al., 2008); r(101) = .22, 95% CI [.03, .40], p < .05 (Irwin, 2014); r(31) = .23, 95% CI [−.12, .53], p = .20 (Lindeman et al., 2016); and r(140) = .22, 95% CI [.06, .37], p < .05 (Dentale et al., 2018).

3 Conducted used the R package Metafor (Viechtbauer, 2010). Forest plot of the meta-analysis is available in supplement at https://osf.io/a4wnv/.

4 There is considerable debate about what implicit measures actually measure (Brownstein, Madva, & Gawronski, 2019). Some scholars have argued that they do not measure fully fledged “beliefs,” but measure some other species of cognitive state, such as “aliefs” (Gendler, 2008), “patchy endorsements” (Levy, 2015), or “unconscious imaginings” (Sullivan-Bissett, 2018). While we do not deny the importance of this debate, for ease of exposition we refer to the measures as “implicit measures of belief” while remaining agonistic about whether the associations being measured really are best described as “beliefs.”

5 For an accessible primer on implicit measures and their strengths and limitations, see Payne and Gawronski (2010).

6 In the study that introduced the AMP (Payne et al., 2005), participants were asked to make affective judgments (e.g., judgments about the pleasantness of the target). Subsequently, the AMP has been adapted to study semantic judgments (e.g., judgments about the meaning of the target; Deutsch & Gawronski, 2009). Some scholars (e.g., Sava et al., 2012) refer to the semantic variant of the AMP as the Semantic Misattribution Procedure (SMP). However, the overall structure of the task is essentially unchanged. Consequently, for ease of exposition we refer to both variants of the task as the AMP.

7 We did not use a formal power analysis to determine sample size as we are employing a novel paradigm and have no strong reason to predict particular effect sizes for the associations between variables. Instead, we selected a minimum sample size of 400 because this fit with our available resources and is a considerably larger sample than the four earlier studies that examined correlations between the IAT and an explicit measure of supernatural belief: n = 33, 61, 103, and 142.

8 14 participants picked “other” as their affiliation. They were asked to “please specify” in an open response box. We read these responses and coded those that entered specific religious groups as “affiliated” (e.g., “Jehovah’s Witness”) and those that did not enter specific religious groups as “unaffiliated” (e.g., “spiritual but not religious”).

9 Participants are presented with an instructions screen and a cover story about written Chinese having many different pictographs that mean “real” and “imaginary” (A “Survey Flow” document which reports the cover story is available in supplement at https://osf.io/a4wnv/.).

10 More precisely, the 21 primes were sampled without replacement. Once all 21 primes had been sampled a further 21 primes were sampled without replacement, and so on until all trials had been completed. In total, each participant is presented with 80 different Chinese pictographs.

11 We sought to present the prime supraliminally because previous research suggests that a quick, yet clear presentation of the prime may enhance the misattribution of emotions/attitudes onto the ambiguous target item (Cameron et al., 2012). To this end, we settled on 175 ms.

12 Our aim was to present each participant with 84 trials; that is, four complete “blocks” of 21 trials without replacement. However, due to a coding error each participant was presented with only 80 trials. This meant that participants do not see all 21 words in the fourth “block” of trials. Nonetheless, because analyses are based on mean scores for three categories of prime this minor discrepancy in number of presentations is very unlikely to bias results.

13 The Greenhouse-Geisser sphericity statistic was ε = .797 and the Huynh-Feldt statistic was ε = .800. Because the Greenhouse-Geisser sphericity statistic is ε > .750 we followed a recommended practice of applying the Huynh-Feldt correction instead of the Greenhouse-Geisser correction (Field, 2017).

Additional information

Funding

This research was generously supported by the Australian Research Council [grant number DP180102384] and the John Templeton Foundation [grant numbert 60624]. The opinions expressed are those of the authors and do not necessarily reflect the views of these funders.