Specificity Versus Generality: A Meta-Analytic Review of the Association Between Trait Disgust Sensitivity and Moral Judgment

RHM Account

For RHM and colleagues (Haidt et al., 1994; Rozin & Fallon, 1987; Rozin et al., 1994, 2008), disgust is elicited in response to four broad classes of stimuli: core elicitors (such as food, body products, animals), animal nature (or reminder) elicitors (such as sex, death, hygiene, body envelope violations), interpersonal elicitors (such as contact with strangers or ‘undesirables’), and moral elicitors (such as certain moral offences) (Rozin et al., 2008). RHM propose that core disgust originated as an elaboration of distaste, which was later co-opted to a range of elicitor classes beyond those directly associated with disease avoidance. When considering the emotions that these different elicitor classes generate, RHM seem to suggest that the experiential, physiological, and behavioral responses are consistent across classes and all serve broadly ‘protective’ functions, ranging from protecting the body from disease or infection (core disgust) to more abstract meanings: protecting the body, soul, and social order (for animal nature, interpersonal and moral disgust).

In an attempt to validate these classes of elicitors, Haidt et al. (1994) developed the Disgust Scale (D-Scale), a 32-item measure of individual differences in sensitivity to each disgust class. The D-Scale contains 16 items that specifically ask respondents to rate on a 3-point scale how disgusting they perceive certain stimuli to be, as well as 16 aversive true-false statements that do not specifically probe disgust per se (e.g., “it would bother me tremendously to touch a dead body”). Although the D-Scale recognizes numerous classes of elicitors theorized to cover the four-domain structure of disgust noted above, the empirically extracted subscales yield inadequate reliabilities, leaving it a rather poor measure of the domain specificity of disgust sensitivity (Haidt et al., 1994). Resultingly, the authors recommended that the D-Scale should be interpreted as a measure of general disgust sensitivity because it yielded good internal reliability when the classes were aggregated.

Updated RHM Account

Olatunji and colleagues (2007) developed a revised version of RHM's scheme (D-Scale-R) to address some of the above measurement issues. After item and factor analyses, the D-Scale-R retained 25 items (of the original 32) loading on to three factors: core, contamination, and animal reminder. Core disgust sensitivity includes items relating to rotting foods, waste products and small animals, and is claimed to assess offense taken at pathogen threats; contamination disgust sensitivity, captures aversion to potential sources of contact contamination, such as touching toilet seats and drinking from another person's cup; and lastly, the animal reminder disgust sensitivity domain taps into body envelope violations and references to death, and is thought to capture aversion to reminders of humans’ animal nature.

Numerous studies have replicated the factor structure of the D-Scale-R (e.g., Olatunji et al., 2009; van Overveld et al., 2011) and other work has attested to the unique predictive utility of the distinct disgust types. For example, the subscales have seen extensive use in the clinical domain: contamination disgust sensitivity is specifically related to obsessive compulsive concerns about sexual orientation (Ching et al., 2018), and animal reminder disgust sensitivity is specifically related to a variety of anxiety concerns (Olatunji et al., 2014), as well as blood-injury fears (Olatunji et al., 2008). There is also evidence of subscale specificity outside of the clinical domain. For instance, core disgust sensitivity has a special relation with prejudice towards homosexuals (Olatunji, 2008), whereas contamination disgust sensitivity has a special relation with prejudice towards immigrants (Aarøe et al., 2017).

Taken together, these various lines of research point to the utility in distinguishing between at least three types of disgust sensitivity: core, contamination, and animal reminder.

Despite the sound psychometric properties and unique predictive utility of these subscales, the RHM and updated RHM accounts and measures have been criticized on a number of fronts. First, there are considerable theoretical issues with animal reminder disgust. As Tybur et al. (2009) point out, it is not clear why disgust is suited to perform the function of avoiding existential thoughts, such as fear of death and recognition of our phylogenetic connection with non-human animals. Confirming these validity issues is recent research showing that (a) reminders of death and mutilation elicit an emotional experience that is more akin to empathic concern than disgust (Kupfer, 2018) and (b) human-animal similarities fail to even elicit disgust (Kollareth & Russell, 2017). Second, neither the D-Scale (aggregates items across different elicitor classes) nor the revised version (excludes the sexual and moral elicitor classes) adequately captures the sexual and moral elicitors of disgust, and therefore do not allow for exploration of these two types of disgust. Third, the D-Scale-R potentially introduces some redundancy since both the core and contamination subscales might be interpreted as an index of pathogen avoidance (Tybur et al., 2018). Fourth, that both instruments ask for responses to a variety of negative states (e.g., “bothered”, “upset”) and not exclusively reports of “disgust”, has led some scholars to suggest the D-Scale and D-Scale-R might better be characterized as measures of general negative affect rather than disgust (Cusimano et al., 2018). In sum, these limitations raise important concerns about the construct validity of the D-Scale and D-Scale-R and potentially undermine the interpretability of their associations with other psychological constructs.

Tybur Account

To address some of the conceptual and methodological issues in RHM's accounts and measures, Tybur and colleagues (2009, 2013) developed an alternative account which posits three domains of disgust: pathogen, sexual, and moral. On this view, domains are distinguished not only by elicitors and ultimate functions but also by (somewhat) different sets of psychological, physiological, and behavioral responses (Tybur et al., 2009, 2013). Accordingly, different domains may indeed reflect different emotions (at least to some degree) rather than different manifestations of the same emotion.

Pathogen disgust evolved to foster avoidance of contact with infectious disease-causing agents. It is triggered by cues such as bodily products, animals, poor hygiene, decomposing organic matter, and in this way, pathogen disgust overlaps with at least two of the three updated RHM domains (core and contamination). The pathogen disgust response is akin to the canonical disgust response discussed above.

Sexual disgust evolved to foster avoidance of contact with people who jeopardize reproductive fitness. It is triggered by cues indicative of low mate value (e.g., promiscuous partners) and is somewhat characterized by the canonical disgust response, although the information processing architecture subserving this adaptation is attuned to cues of sexual value and motivates avoidance of sexual contact (rather than contact per se).

Moral disgust evolved to communicate and coordinate condemnation of rule violators and is triggered by rule violations (especially those violations that are likely to be condemned by others, such as lying, cheating, and stealing). The moral disgust response is not theorized to resemble canonical disgust, but rather consists of signaling behaviors designed to regulate and coordinate condemnation (e.g., facial and vocal expressions of disgust).

The Three Domain Disgust Scale (TDDS; Tybur et al., 2009) was developed to assess individual differences in sensitivity to these types of disgust. Factor analyses from the original and subsequent replication studies demonstrate support for a three-factor model (DeBruine et al., 2010; Tybur et al., 2009, 2011). Moreover, there is evidence of the unique predictive validity of the subscales, and thereby differences between disgust types. For instance, research examining contamination fears in obsessive-compulsive disorder patients has found that pathogen disgust sensitivity specifically predicts fear of contact contamination, whereas sexual disgust sensitivity specifically predicts fear of mental contamination (Poli et al., 2019). More evidence of subscale specificity comes from outside of the clinical domain: on the one hand sexual disgust sensitivity is uniquely related to moral condemnation of recreational drugs (Kurzban et al., 2010), and on the other hand pathogen disgust sensitivity is uniquely related to women's preferences for masculine men (DeBruine et al., 2010; Jones et al., 2013).

Tybur's account and measurement are also not without limitations. Primary among them is that ‘moral disgust’ may be better thought of as rhetorical concept that actually resembles the characteristic signatures of anger instead of disgust (for more elaboration, see Nabi, 2002), a criticism that has garnered a fair amount of empirical support in recent years (Gutierrez et al., 2012; Kollareth & Russell, 2019; Royzman et al., 2014). This has implications for interpreting correlations between moral disgust and other variables: if moral disgust is a (closer) proxy for anger, then it may be anger that is doing the explanatory work rather than disgust per se.

The criticisms of both approaches notwithstanding, updated RHM and Tybur models (and measurements) suggest that it may be worth distinguishing between kinds of disgust (at least in terms of elicitor classes). One of the key foci of this meta-analysis is to assess the extent to which the specific measures of disgust sensitivity discussed above (and the subscales thereof) are related to moral judgment (see Method for more detail), and hence whether the role of disgust (in particular disgust sensitivity) in moral judgment is best characterized as a general effect (manifest across disgust types) or one that is more specific (particular to one or few disgust types).

Moral Domain Specificity Versus Generality

Evidence of moral domain specificity is mixed. Some work suggests that disgust sensitivity (of one kind or another) is associated exclusively with violations of the moral foundation of sanctity or purity. In moral judgment research, measures of moral violations are commonly derived from Moral Foundations Theory (Graham et al., 2011, 2013), a theory which suggests the moral universe consists of a variety of moral foundations or domains which are aligned with fitness challenges across our evolutionary history and to which our moral intuitions are attuned. These domains are Care/Harm, Fairness/Cheating, Loyalty/Betrayal, Authority/Subversion, Sanctity/Degradation (aka Purity) (and Liberty/Oppression ² ). Each foundation is characterized by specific adaptive challenges, triggers, characteristic virtues, and emotions. It is worth noting that there are other theoretical accounts of morality within the psychological literature that carve-up the moral universe quite differently. The theory of Morality as Cooperation, for instance, suggests that our sense of morality is grounded in a collection of biological and cultural solutions to the problems of cooperation and proposes seven distinct domains of cooperation which are considered morally relevant across cultural contexts (Curry et al., 2019). However, because the majority of disgust-immorality effects in the literature are derived from studies that use the Moral Foundations Theory taxonomy, the present analysis also follows the Moral Foundations Theory taxonomy (and unless otherwise specified, we take moral domains to reflect the domains of Moral Foundations Theory). As mentioned, because of shared evolutionary pressures, one's sensitivity to condemning sanctity moral violations should be closely linked to one's sensitivity to disgust; and there is empirical work consistent with this. Indeed, some studies have demonstrated a unique disgust-sanctity association (e.g., Horberg et al., 2009), whereas other highly powered studies (e.g., Wagemans et al., 2018) are suggestive of a less strict ‘primarily purity’ effect, in which disgust sensitivity is associated with condemnation of a variety of moral violations, but is most pronounced for the condemnation of sanctity moral violations (for similar findings see, Garvey & Ford, 2014; Olatunji et al., 2016; van Dijke et al., 2018; van Leeuwen et al., 2017).

By contrast, other work indicates disgust-immorality effects are more general, such that disgust sensitivity is associated with condemnation of violations in domains other than sanctity. Landy and Piazza (2019), for example, found that disgust sensitivity was associated with harsher moral judgments of harm and fairness transgressions. Similarly, Chapman and Anderson (2014) found that disgust sensitivity was associated with condemnation of harm transgressions even after controlling for potential confounds such as trait anger and social conservatism. The studies by the aforementioned authors, however, do not include sanctity violations to directly compare their disgust-immorality associations with, rendering their results inconclusive as to whether there any meaningful differences in the size of associations between moral domains. Further evidence in favor of moral domain generality can be found in a recent cross-cultural study in which van Leeuwen and colleagues (2017) showed that pathogen disgust is more strongly related to authority transgressions than sanctity transgressions.

Disgust Specificity Versus Generality

Extant work is also ambiguous about disgust-specificity. This ambiguity derives in part from a variety of methodological and measurement issues. First, some studies looking at disgust sensitivity and moral judgment ignore questions of disgust specificity altogether and use either ad hoc or truncated measures that pay no heed to disgust type (e.g., Horberg et al., 2009; Inbar et al., 2009; Schein et al., 2016). Second, some studies that do use validated measures with the potential to explore disgust-specificity do not report disgust type effects, and instead aggregate across elicitor classes (e.g., Garvey & Ford, 2014; Landy & Piazza, 2019; Wagemans et al., 2018).

The limited work that has explicitly considered different dimensions of disgust sensitivity yield results suggestive of some degree of disgust type specificity. For example, across several studies, Karinen and Chapman (2019) found that core disgust sensitivity (subscale of D-Scale-R) was more strongly associated with moral condemnation of harm transgressions than the other subscales in the D-Scale-R. Furthermore, using the TDDS, van Leeuwen and colleagues (2017) found that sexual disgust sensitivity was more strongly associated with moral condemnation of sanctity transgressions than was the pathogen disgust sensitivity subscale (for an analogous finding, see Kurzban et al., 2010).

In sum, extant work on disgust sensitivity and moral judgment is ambiguous about the role of moral domain as a moderator of the disgust-immorality effect and is at least indicative of the possibility that multiple types of disgust sensitivity are associated with moral judgments. In the current meta-analysis, we use the moderators disgust type and moral domain to clarify questions of specificity and generality in the disgust-immorality association, and to also articulate the pattern of disgust-immorality associations that future theorizing must consider.

Literature Search

Articles were obtained by searching in PsycINFO, Google Scholar and Web of Science during March 2018 using the following search string: combination: (“disgust sensitivity” OR “trait disgust” OR “disgust scale*” OR “disgust propensity”) AND (“moral judgement*” OR “moral judgment*” [USA spelling] OR “moral violation*” OR “moral transgression*” OR “moral foundation*”). Theses and dissertations were retrieved via ProQuest using the same combinations of keywords.

Descendant (or forward) searches were conducted on key empirical articles (Chapman & Anderson, 2014; Horberg et al., 2009) using the PsycINFO “cited by” function. Ancestor (or backward) searches were performed by searching the reference lists of all the studies that met the inclusion criteria. In addition, we also searched the reference lists of several major narrative reviews on disgust and moral judgment (Cameron et al., 2015; Chapman & Anderson, 2013; Piazza et al., 2018; Russell & Giner-Sorolla, 2013). As depicted in Figure 1, these searches generated 1,072 unduplicated citations. Requests for unpublished literature were also sent to the electronic mailing lists of the Society of Social and Personality Psychology, and via direct email to authors of key published studies in the area.

OPEN IN VIEWER

Figure 1.

Meta-analysis literature search flow diagram.

Study Eligibility

After initial article-level screening was conducted by scanning titles and abstracts, the following eligibility criteria were applied to the remaining studies within articles. A study was included if:

The study reported zero-order correlation(s) between measures of trait disgust sensitivity (and/or subscales thereof) and moral judgments (more detail below). Studies that reported partial associations (e.g., effects in multiple regressions or mediation contexts) were included if bi-variate correlations were also reported. In cases in which zero-order correlations were unreported, we contacted the authors by email to request them.

Coding of Study Characteristics and Substantive Moderators

Primary coding for study characteristics and moderators was performed by the lead author. Double coding was not performed on the grey literature sent to us (approximately 30% of the total effects) because codes were completed by other researchers in the coding template provided to them (moderator codes closely resembled those found in Table 1). Out of the remaining eligible effects to be coded, 78% were independently double coded by Warren and Azaad. Interrater reliabilities were calculated using Cohen's kappa for categorical moderators and Pearson's r for continuous moderators. All reliabilities were in an acceptable range (0.83–1; M = .96). Disagreements were resolved through discussion. A summary of how these variables were coded is presented in Table 1. The coding details for the key substantive moderators, relating to disgust sensitivity type and moral domain, are discussed in detail below.

OPEN IN VIEWER

Table 1.

Coding scheme for moderators

Moderator	Definition	Coding details
Source type	Status of publication from which the effect was derived	Published, unpublished
Sample type	Sample from which the effect was derived	Student, general community*, Mturk (Amazon's mechanical turk: online database), YourMorals (online database)
Sample location	Location from which the effect was derived**	USA, EU, UK, India, Japan, Australia, other
Gender	Percentage of female participants for the effect	Continuous
Age	Mean age of participants in years for the effect	Continuous
Sample size	Number of participants that the analysis of the effect was conducted on	Continuous
Ethnicity	Percentage of participants identifying with specific ethnic group (Caucasian, Asian, African American, Latino) for the effect	Separate continuous moderator for each of the noted ethnicities
Nationality	Percentage of nationality make-up (USA, EU, UK, Asia, India) for the effect	Separate continuous moderator for each noted nationality
Education	Percentage of participants’ highest educational attainment (high school, completing university, completed university, and graduate school) for the effect	Separate continuous moderator for each noted education level
Politician orientation	Percentage of participants’ ideological self-placement (Liberal, Moderate, Conservative) for the effect	Separate continuous moderator for each political orientation
Disgust sensitivity measure length	Was the measure of disgust sensitivity from which the effect was derived based on a full-length scale or a shortened scale?	Full, short
Disgust sensitivity measure reliability	Cronbach's alpha for relevant disgust sensitivity scale/sub-scale for the effect	Continuous
Moral judgment type	What kind of judgment was the effect based on?	Normative evaluation, punishment
Moral judgment measure	What was the measure of moral judgment the effect was based on?	Moral Foundations Questionnaire, Moral Foundations Vignettes, Moralization of Everyday Life, ad hoc
Moral judgment measure length	How many items were used to assess the moral judgment?	Continuous
Moral judgment measure reliability	Cronbach's alpha for relevant moral judgment scale/sub-scale for the effect	Continuous
Disgust prime	Was a manipulation of disgust administered during the experiment?	Disgust manipulation; no disgust manipulation
Disgust sensitivity measure	How was disgust sensitivity measured?	D-Scale (aggregate); D-Scale-R (aggregate); D-Scale-R Core; D-Scale-R Animal reminder; D-Scale-R Contamination; TDDS-pathogen; TDDS-sexual; TDDS-moral; other***
Moral domain	What domain does the moral judgment(s) fall into?	Care, Fairness, Loyalty, Authority, Sanctity

Note. *General community refers to non-Mturk, non-student, and non-YourMorals based samples (see endnote iii for treatment of YourMorals data); **USA = United states of America, EU = European Union, UK = United Kingdom; ***other refers to non-validated measures of disgust sensitivity (e.g., ad hoc measures); coded moderators were determined by the studies available to us. Only moderator codes with four or more effects (stemming from different samples) were retained for analyses.

Effect Size Estimation, Transformation, and Combination

Pearson's correlation coefficient (r) was the focal statistic for meta-analytic procedures; all effects in the meta-analysis were scored such that positive correlations indicated that higher disgust sensitivity corresponded to higher ratings of moral wrongness. The majority of effects were reported in this way, except for two studies that reported the effect in terms of group mean differences in moral wrongness as a function of a categorical disgust sensitivity variable (Chapman & Anderson, 2014; Nichols, 2002). For these studies, the mean difference was converted into a Pearson correlation by first transforming effect sizes into d's (standardized mean differences) from t statistics, and then from d to r using the appropriate formulas outlined in Lipsey and Wilson (2001). For unpublished reports, authors either sent (a) raw data necessary to compute the correlations or (b) information and correlations in a coding template provided to them.

Following the formulas outlined in Lipsey and Wilson (2001), all coded effects were transformed into a Fisher's z coefficient and weighted by N–3 (inverse variance). This procedure addresses problems with standard error formulation for r (Lipsey & Wilson, 2001). All meta-analytic computations were performed on the transformed z values, and then converted back to r for presentation.

Meta-Analytic Procedure

Sample-level dependency in our dataset was substantial; we obtained 512 effects from 74 samples. This is because it is common for different types of disgust and different domains of moral judgment to be measured in the same samples. Given that our dataset contained substantial sample-based dependence among effects, we conducted a multilevel meta-analysis (with three-levels) to model this dependency, so that we could minimize data loss typical of the more common methods of ensuring independence among effects (e.g., aggregation, selection, shifting units).

In our meta-analytic dataset, each sample (indexed by subscript _j) may yield multiple effects (indexed by subscript _i) (we use notation from, Cheung, 2014). To deal with this dependency, we adopted a multilevel analysis strategy that takes into account (a) sampling variance at Level 1, (b) within-sample variance at Level 2, and (c) between-sample variance at Level 3. More specifically, the model, without moderators is:

E S i j = β 0 + u ( 3 ) j + u ( 2 ) i j + e i j

Each error term, u ( 3 ) j , u ( 2 ) i j , and e i j was assumed to be independently and normally distributed with mean zero and variances, Var ( u ( 3 ) j ), Var ( u ( 2 ) i j ), and Var ( e i j ). The last of these, the sampling variance (in this case for Fisher’s z), is the known sampling variance for each sample, calculated as 1 / ( N − 3 ) , where N is the sample size; thus only β 0 , u ( 3 ) j , and u ( 2 ) i j were estimated (Cheung, 2014).

Using this approach, we began by estimating the overall effect size, in the context of the above-specified model, and also the within and between sample variances as indicators of heterogeneity. We tested the significance of these heterogeneity components and the proportion of the total variance between observed effects accounted for by variance at each level (I²). Given that variability in observed effects was not due merely to sampling variance (see below), we followed these initial analyses with moderator analyses in which categorical or continuous moderators were added to the above model. We also considered publication bias using three strategies: inspection of funnel plots, Egger's test (Egger et al., 1997), and moderation by publication status. All analyses used Restricted Maximum Likelihood estimation and were implemented using the metafor package for R (Viechtbauer, 2010).

Overall Effect and Heterogeneity

The first goal of the current meta-analysis was to obtain an overall weighted-mean correlation for the disgust-immorality association. Two effects were more than three standard deviations from the unweighted mean effect size. However, neither of these were influential cases (Cook's d's = .08, .11). As such, all 512 effects (from a total of 72,443 participants ³ ) were used in the analyses. To estimate the overall mean effect, we first fit a three-level model to our data, which yielded a mean effect size of r = .23, SE = .01, t = 19.1, p < .001, 95% CI = [.20, .25] ⁴ . We conducted likelihood ratio tests comparing models with and without respective variance components and found significant variance between effects within samples, Var ( u ( 2 ) i j ) = .012, χ²(1) = 14,285.89, p < .001, as well as between samples, Var ( u ( 3 ) j ) = .01, χ²(1) = 63.93, p < .001. Further, we computed I² values - using formulas from Higgins and Thompson (2002) – to estimate heterogeneity in effect sizes attributable to (a) sampling variance, (b) within-sample, between-effect variance, and (c) between-sample variance. Heterogeneity due to sampling variance was small (I² = 2.36%), and well below the 75% threshold proposed by Hunter and Schmidt (2004), thus justifying moderator analyses. Both within-sample variance (I² = 63.15%) and between-sample variance (I² = 34.49%) contributed substantially to overall heterogeneity.

For comparison purposes, we fitted a two-level model (a traditional random effects model) based on sample-level aggregated effects, which yielded highly similar results, r = .22, SE = .01, z = 18.77, p < .001, 95% CI = [.20, .24]. Further, the two-level model indicated that there was significant heterogeneity in the set of aggregated effects, Q(74) = 204.12, p < .001, I² = 71.55%.

Publication Bias

We used multiple methods to assess publication bias. Visual inspection of a funnel plot did not reveal the obvious asymmetry indicative of publication bias (Figure 2). Consistent with this, the Egger regression method (Egger et al., 1997), did not indicate publication bias, F(1, 134) = .07, p = .79. Additionally, source type did not moderate disgust-immorality associations; that is, effect sizes did not differ between published and unpublished effects, F(1, 510) = 2.44, p = .12 (see Table 2 for further detail). Thus, the results detailed here are unlikely to be affected by publication bias.

OPEN IN VIEWER

Figure 2.

Funnel plot of standard error and effect size (Fisher's Z) of published effects.

OPEN IN VIEWER

Table 2.

Source and sample moderators

Moderator	k	r/β	C	C+	95% CI	F(df)	Model R2	F(df)+
Source type	512					2.44(1,510)	0.48
Published	136	0.25***	a		[0.21, 0.28]
Unpublished	376	0.21***	a		[0.18, 0.24]
Sample type	509					3.05(3,505)*	1.78	2.06(2,414)
Student	195	0.22***	a,b	a,b	[0.18, 0.25]
General community	64	0.16***	a	a	[0.10, 0.21]
Mturk	230	0.25***	b	b	[0.22, 0.28]
YourMorals	20	0.18*	a,b		[0.03, 0.32]
Sample location	479					1.29(6,472)	1.62
USA	301	0.24***	a		[0.21, 0.27]
EU	76	0.17***	b		[0.11, 0.23]
UK	17	0.17*	a,b		[0.04, 0.31]
India	15	0.18*	a,b		[0.02, 0.34]
Japan	15	0.25**	a,b		[0.08, 0.40]
Australia	35	0.20***	a,b		[0.11, 0.30]
Other	20	0.15**	a,b		[0.06, 0.24]
Age	503	0			[−0.00, 0.00]	0.01(1,501)	0.00
Gender	508	−0.09			[−0.27, 0.09]	0.98(1,506)	0.19
Caucasian	229	0.31			[−0.09, 0.62]	2.38(1,227)	1.04
Asian	161	0.06			[−0.44, 0.53]	0.05(1,159)	0.03
African American	164	0.57			[−0.94, 1.00]	0.29(1,162)	0.18
Latino	157	0.37			[−0.94, 0.99]	0.13(1,155)	0.09
Nationality USA	212	0			[−0.12, 0.13]	0.00(1,210)	0.00
Nationality UK	225	−0.05			[−0.24, 0.15]	0.24(1,223)	0.11
Nationality EU	210	−0.42			[−1.00, 0.97]	0.12(1,208)	0.06
Nationality Australia	210	−0.06			[−0.38, 0.27]	0.13(1,208)	0.06
Nationality Asia	225	−0.12			[−0.73, 0.61]	0.08(1,223)	0.04
Nationality India	210	0.01			[−0.16, 0.18]	0.02(1,208)	0.01
Completing University	252	−0.06			[−0.15, 0.03]	1.77(1,250)	0.70
Completed University	242	0.16^			[−0.01, 0.31]	3.62(1,240)^	1.49	0.08(1,175)
Graduate School	242	0.10			[−0.22, 0.40]	0.36(1,240)	0.15
High School	242	0.29			[−0.13, 0.62]	1.89(1,240)	0.78
Liberal	141	−0.14			[−0.35, 0.09]	1.37(1,139)	0.98
Conservative	138	−0.06			[−0.45, 0.36]	0.07(1,136)	0.05
Moderate	123	0.27*			[0.06, 0.45]	6.59(1,121)*	5.16	0.30(1,21)

Note. ***p < .001., **p < .01., *p < .05, ^p < .10; k = number of effects; r = mean effect size for each moderator; β = unstandardized regression coefficient; C = contrast indices, moderator levels that share a letter index do not differ significantly at p = .05; C⁺ = multivariate contrast indices controlling for correlated moderators, moderator levels that share a letter index do not differ significantly at p = .05; CI = 95% confidence intervals; F(df) = test statistic for moderator heterogeneity; Model R² indicates the percentage of the total variance explained across the three levels; F(df)⁺ = multivariate test statistic for moderator heterogeneity (controlling for correlated moderators). No values in the F(df)⁺ column indicate that multivariate analysis was not performed because certain criteria were not met (for our strategy, see online Supplemental Materials).

Source and Sample Moderators

As can be seen in Table 2, all sample types yielded significant effects. Despite this, pairwise comparisons showed that Mturk samples (from which the majority of effects came) produced larger effects than did general community samples. However, after accounting for correlated moderators, multivariate analyses indicated that sample type was not a significant moderator of the disgust-immorality association.

The proportion of conservatives, moderates, and liberals in a sample did not significantly moderate the effect, although the effect of the latter trended towards attenuation (as evidenced by an asymmetric confidence interval).

None of gender, age, education level or ethnicity significantly moderated disgust-immorality associations. Neither, as noted earlier, did source type. Although nationality did not significantly moderate the effect (and disgust-immorality associations were significant in all locations), pairwise comparisons did suggest larger effects from US than European samples.

Method and Design Moderators

As highlighted in Table 3, more reliable measures of both disgust sensitivity and moral judgment increased the magnitude of disgust-immorality associations. Moderating effects of disgust sensitivity measure reliability and moral judgment measure reliability remained marginally significant and significant, respectively, after controlling for correlated moderators. And although disgust sensitivity measure length (full versus short) did not moderate effects, moral judgment measure length did, with moral judgment measures containing a higher number of items producing significantly larger disgust-immorality associations. We also note that the scale used to measure moral judgment marginally moderated the effect in univariate (effects were larger for Moralization of Everyday Life Scale ⁵ than Moral Foundations Questionnaire and ad hoc) but not in multivariate analyses. Furthermore, effects were unmoderated by disgust prime and moral judgment type (normative evaluation versus punishment).

Substantive Moderators

Moral Domain

Univariate analyses revealed significant moderation by moral domain, explaining most of the coded heterogeneity in effect sizes (see Table 4). No multivariate analysis was performed because no moderators were substantially correlated with moral domain. Although all five domains of moral judgment produced significant disgust-immorality associations, these effects were significantly strongest for the sanctity domain (compared to all other domains) and significantly weakest for the fairness domain (compared to all other domains). The effects of care, authority and loyalty did not differ from each other, but were each significantly weaker than the effects of sanctity and stronger than effects of fairness. Figure 3 summarizes the results of the moderator analyses for the moral domain.

OPEN IN VIEWER

Figure 3.

Caterpillar plot of effect sizes (Fisher's Z) by moral domain.

Disgust Sensitivity Measure and Moral Domain

To further test specificity in disgust-immorality associations, we investigated whether the effects of disgust sensitivity measure varied as a function of moral domain (disgust type × moral domain). These analyses were run to (a) test the specific a priori predictions of Moral Foundations Theory that canonical disgust is uniquely (or primarily) associated with moral judgments of sanctity transgressions (Graham et al., 2013) and (b) explore other patterns of associations between types of disgust and domains of moral judgment.

To this end, we tested whether or not there was an interaction between moderator's disgust sensitivity measure and moral domain. Results revealed that the interaction between disgust sensitivity measure and moral domain was statistically significant, F(32, 434) = 4.23, p < .001, indicating that disgust sensitivity measure-moral domain combinations explain significant heterogeneity in disgust-immorality associations. The results from meta-analytic multilevel regression testing moderation by disgust sensitivity measure within each of the moral domains are documented in Table 5. Specifically, within each moral domain, we (a) estimated the average correlation for each disgust sensitivity measure and (b) tested whether correlations significantly differed across disgust sensitivity measures. To ensure sufficient data was available in each moderator analysis, we required at least four effects (derived from independent samples) to be coded for the disgust sensitivity measure – moral domain combinations in Table 5. This meant that particular moderator level combinations were excluded from our analyses (e.g., D-Scale × fairness).

OPEN IN VIEWER

Table 5.

Disgust sensitivity measure effects as a function of moral domain

Moderator	k	r/β	C	95% CI	F(df)	Model R2
Care	121				3.25(8,112)**	18.84
D-Scale (aggregate)	5	0.26***	a,b,c,d	[0.14, 0.37]
D-Scale-R (aggregate)	27	0.22***	b,d	[0.17, 0.26]
Core	6	0.21***	a,b,c,d	[0.13, 0.29]
Animal Reminder	6	0.16***	a,b,c,d	[0.07, 0.24]
Contamination	6	0.13**	c	[0.05, 0.22]
TDDS-pathogen	23	0.15***	c	[0.10, 0.20]
TDDS-sexual	21	0.23***	b,d	[0.18, 0.28]
TDDS-moral	20	0.24***	d	[0.19, 0.29]
Other	7	0.13**	c	[0.05, 0.22]
Fairness	90				11.52(4,81)***	36.25
D-Scale-R (aggregate)	13	0.15***	a	[0.09, 0.22]
TDDS-pathogen	21	0.08**	a	[0.03, 0.13]
TDDS-sexual	22	0.09***	a	[0.04, 0.13]
TDDS-moral	21	0.24***	b	[0.19, 0.29]
Other	9	0.07	a	[−0.02, 0.16]
Sanctity	122				16.51(5,113)***	42.22
D-Scale (aggregate)	4	0.32***	a	[0.17, 0.45]
D-Scale-R (aggregate)	25	0.37***	a	[0.31, 0.42]
TDDS-pathogen	26	0.23***	b,c	[0.18, 0.29]
TDDS-sexual	32	0.41***	a	[0.36, 0.45]
TDDS-moral	26	0.22***	b,c	[0.16, 0.28]
Other	6	0.12^	c	[−0.00, 0.24]
Authority	78				5.71(4,69)***	24.88
D-Scale-R (aggregate)	15	0.22***	a	[0.17, 0.28]
TDDS-pathogen	18	0.24***	a	[0.19, 0.28]
TDDS-sexual	19	0.26***	a	[0.21, 0.30]
TDDS-moral	18	0.23***	a	[0.18, 0.28]
Other	4	0.00	b	[−0.11, 0.10]
Loyalty	68				4.16(4,59)**	21.99
D-Scale-R (aggregate)	8	0.24***	a	[0.18, 0.30]
TDDS-pathogen	17	0.18***	a	[0.13, 0.23]
TDDS-sexual	18	0.21***	a	[0.17, 0.26]
TDDS-moral	17	0.20***	a	[0.15, 0.25]
Other	4	0.03	b	[−0.07, 0.14]

Notes. ***p < .001., **p < .01., *p < .05, ^p < .10; k = number of effects; r = mean effect size for each moderator; β = unstandardized regression coefficient; C = contrast indices, moderator levels that share a letter index do not differ significantly at p = .05; CI = 95% confidence intervals; F(df) = test statistic for moderator heterogeneity; Model R² indicates the percentage of the total variance explained across the three levels.

For authority and loyalty, although disgust sensitivity measure was a significant moderator, pairwise comparisons revealed that this was driven by ‘other’ disgust sensitivity measures (of which there were very few, k = 4) producing significantly weaker effects than the TDDS subscales and D-Scale-R (aggregate), all of which produced significant effects that did not differ from each other (Table 5).

For fairness, the effect for moral disgust was significantly stronger than for each of the other disgust sensitivity measure types available for that analysis (i.e., TDDS subscales and D-Scale-R [aggregate]).

For care, significant moderation appears to be accounted for primarily by the TDDS-pathogen, D-Scale-R contamination and ‘other’ measures yielding smaller effects than the rest of the coded measures (although the ‘other’ and contamination effects were based on rather small k's). Among the TDDS subscales, the pathogen effect is significantly smaller than both moral and sexual disgust sensitivity, indicating that to the extent disgust sensitivity plays a role in the condemnation of care violations, it is disgust sensitivity types other than pathogen that have stronger effects. D-Scale-R (aggregate) also has significantly stronger effects than both its contamination subscale (although not its core and animal reminder subscales) and TDDS-pathogen. Although the D-Scale also produced effects for care comparable to the D-Scale-R, the small set of effects upon which its results are based is reason for pause in over-interpretation.

Finally, for the sanctity domain, significant moderation by disgust sensitivity measure indicated that among the TDDS subscales, it is sexual disgust sensitivity (rather than the pathogen or moral disgust sensitivity) that produces the strongest effects (Table 5). Counter to claims of Moral Foundations Theory, which would predict pathogen disgust to be the strongest predictor here, sexual disgust appears to account for significantly more variation in judgments of sanctity violations. The aggregate D-Scale and D-Scale-R effects are of comparable size to that of sexual disgust, although again, the former effect is based on too few effects to warrant strong inference.

Unfortunately, as with most of the other moral domains except for care, too few effects were estimated from the D-Scale-R subscales to assess their roles in predicting judgments of sanctity violations.

Moral Domain

The finding that disgust sensitivity (averaged across measures) was related to all moral domains (care, fairness, sanctity, authority, and loyalty), but was most strongly related to the sanctity domain, is inconsistent with hypotheses that propose either strict moral domain specificity (e.g., sanctity domain; Moral Foundations Theory; Graham et al., 2013) or strict moral domain generality (e.g., constructivism; Cameron et al., 2015). On balance, our moral domain moderator results are consistent with the ‘primarily purity’ (or primarily sanctity) account of disgust-immorality effects, recently supported in Wagemans et al.'s (2018) large-scale, systematic exploration of disgust sensitivity effects across moral foundations.

It comes as no surprise that disgust sensitivity is closely linked to sanctity moral concerns, as disgust-based motives (e.g., disease avoidance) are theorized to ground sanctity moral concerns (Graham et al., 2013); however, it is less clear why disgust sensitivity is also linked to non-sanctity sanctity moral concerns. One possibility is that the association between disgust and non-sanctity moral judgments is artefactual. For instance, some of the effects might be overestimated because of the shared variance between disgust and other specific emotions, such as anger, that also correspond to socio-moral violations. Recent work, though, has largely put to rest this concern by showing that disgust sensitivity predicts judgments of non-sanctity violations when controlling for trait anger (Chapman & Anderson, 2014; Karinen & Chapman, 2019; c.f. Horberg et al., 2009). Alternatively, some of the effects might be overestimated because of the shared variance between disgust and a general tendency to respond with negativity (Landy & Piazza, 2019). Some comfort over this concern can be found in recent regression results indicating that the relationship between disgust sensitivity and non-sanctity moral judgments survive covariation of negative affect (an imperfect proxy for negativity in judgments) (Karinen & Chapman, 2019).

Assuming that there is a robust relationship between disgust and non-sanctity moral judgments, there might be several psychological mechanisms that explain this relation. First, one potential mechanism is interpersonal value. In line with the notion that disgust involves trade-offs between pathogen avoidance (or sexual avoidance) and the benefits of social contact (or sexual contact), recent work has shown that people are less comfortable with engaging in acts that involve infection risk with those who have lower interpersonal value (e.g., disked individual, dishonest stranger) (Tybur et al., 2020). Following this logic, more disgust sensitive people might be more likely to interpret certain individuals, such as norm violators, as non-valued conspecifics and thus avoid them. Moreover, moral condemnation might be used as a way to justify inclinations to avoid norm violators, or anyone else perceived to be of low interpersonal value. Thus, interpersonal value may help explain the relationship between disgust and non-sanctity judgments in the current meta-analysis.

Second, another potential mechanism is self-other similarity. Recently, Mentser and Nussinson (2020) found that more disgust sensitivity people are more likely to perceive unfamiliar others as less similar to oneself, presumably as a form of protection against the infection (or sexual) risks involved in social (or sexual) contact. Low similarity perceptions are also implicated in a range of social phenomena including basic first impressions, prejudice towards outgroups, and moral character judgments (e.g., Bocian et al., 2021; Lydon et al., 1988; McDonald et al., 2017). In a similar manner, people might perceive the actions of norm violators to be maximally distinct from the self and recruit moral condemnation as a way to express this. Thus, disgust sensitivity might be associated with moral judgments because more disgust sensitive people may perceive a greater discrepancy between themselves and the norm violators than lower disgust sensitivity individuals.

Third, a further possibility is that disgust-based mechanisms (e.g., disease voidance, promiscuity avoidance) might (partly) explain the moralization of non-sanctity domains. On the one hand, the non-sanctity effects might be due to disease avoidance mechanisms (i.e., canonical disgust). For instance, the authority domain might be linked to disgust because adhering to tradition and established societal practices might provide some protection against the risk of disease transmission (Murray et al., 2011). Similarly, the loyalty domain might be linked to disgust because moralizing in-group/out-group boundaries may have decreased the risk of contact with people from foreign ecologies who may carry novel diseases (Fincher & Thornhill, 2008). On the other hand, the non-sanctity effects might be due to sexual avoidance mechanisms (i.e., sexual disgust), such as promiscuity avoidance. For instance, moralizing authority (e.g., religion, family) and loyalty (e.g., partner trustworthiness, faithfulness) might reduce the risk of mate poaching and other related promiscuous behaviors. The notion that sexual restrictiveness is driving disgust-immorality effects is explored in more detail below.

Disgust Type

Interpretations of the moderator disgust sensitivity measure (our proxy for disgust type) is less straightforward than that of the moral domain because many of the effects come from composite scales, D-Scale (k = 14) and D-Scale-R (k = 107). One issue with the D-Scale and D-Scale-R composites involves their inclusion of animal reminder elicitors. Theoretically, the existence of an anti-animal reminder mechanism is doubtful (Tybur et al., 2009, 2013); and empirically, aversive reactions to animal reminders (including injuries, gore, mutilation, body-envelope violations) are based on empathy rather than disgust (Kupfer, 2018). This latter point raises the possibility that empathy might partly explain disgust-immorality effects derived from D-Scale and D-Scale-R composite scales. One can circumvent the issues with the animal reminder construct by considering the core and contamination subscale effects of the D-Scale-R (the poor subscale reliabilities of the D-Scale do not allow for this). However, we urge caution in interpreting these subscale results because they were estimated on a very small number of effects (k = 10) and were predominantly based on only care transgressions. Another issue is the fact that the D-Scale composite conflates different elicitor classes (e.g., canonical, sexual). The implication of this is that there is no way of knowing which elicitor class (or classes) accounts for the effects of the D-Scale on moral judgment. By contrast, the D-Scale-R omits the sexual elicitors and hence inhibits inquiry of sexual disgust motives in moral judgment altogether.

Unlike the D-Scale and D-Scale-R, the TDDS permits for more straightforward interpretations of the results of disgust type on moral judgment because (a) animal reminder elicitors are not present in the TDDS (thus the effects of the TDDS on moral judgments are not confounded by empathy) and (b) the measure is consistently used by researchers at the subscale level (thus facilitating investigations of disgust type specificity [e.g., canonical disgust, sexual disgust] on moral judgment). However, in the same way animal reminder disgust raises validity problems for D-Scale and D-Scale-R, the moral disgust construct raises validity concerns for the TDDS. Specifically, relations between TDDS-moral and moral judgments might be obscured due to the moral disgust construct representing an emotion that is closer to anger than disgust (Nabi, 2002). Given the ambiguities just described in interpreting certain measures of disgust (i.e., D-Scale composite, D-Scale-R composite, TDDS-moral), we focus our discussion specifically on how TDDS-pathogen disgust and TDDS-sexual disgust are related to moral judgment and the implications of these patterns for current and future theorizing.

Our results revealed that despite both pathogen and sexual disgust yielding significant effects, TDDS-pathogen was a significantly weaker predictor of moral judgment (aggregated across moral domains) than TDDS-sexual. Interactions between different disgust types and different moral domains indicated that this difference is driven by stronger effects of sexual disgust within the care and sanctity domains. Thus, neither strict disgust type generality (e.g., constructivism; Cameron et al., 2015) nor specificity (e.g., canonical disgust; Moral Foundations Theory; Graham et al., 2013) adequately describe these patterns of associations.

The finding that sexual disgust is more strongly predictive of condemnation of sanctity transgressions than is canonical disgust ⁶ is consistent with recent evolutionary accounts that view reproductive strategies as a determinant of individual political and moral values (Petersen, 2018; Tybur et al., 2015; van Leeuwen et al., 2017). A growing body of research suggests that various measures of sexual restrictiveness (including sexual disgust sensitivity) are related to values and attitudes broadly supportive of restricting promiscuity and promoting monogamy, including conservative political ideology (Billingsley et al., 2018; Kubinski et al., 2018; Tybur et al., 2010, 2015), religiosity (Weeden & Kurzban, 2013), opposition to gay marriage (Pinsof & Haselton, 2016), and recreational drug use (Kurzban et al., 2010; Quintelier et al., 2013). Functionally, sexual disgust is claimed to be part of an adaptation designed to process the costs and benefits of risky sexual encounters (Tybur et al., 2013). Endorsing a suite of traditional normative practices and attitudes may be one way of minimizing general exposure to fitness compromising sexual encounters. The results of the current meta-analysis complement the above findings by extending the scope to moral values and judgments, emphasizing the importance of sexual disgust (i.e., sexual strategies account; Tybur et al., 2015) – as opposed to canonical disgust (i.e., disease avoidance account; Graham et al., 2013) – in explaining variance in moral values and judgments (especially in the sanctity domain). Future work should heed this pattern of results and at the very least simultaneously examine both pathogen and sexual disgust to distinguish the respective roles of each disgust type (and thereby distinguish between the disease avoidance account and sexual strategies account) in the moral domain. Additionally, a logical next step is to examine which sexual disgust motives underlie these sexual disgust-immortality effects. Indeed, it has recently been suggested that there are multiple domains of sexual disgust (Taboo, Oral sex, Promiscuity, Hygiene, BDSM, and Same-sex attraction), each of which represents a different kind of mating problem that sexual disgust supposedly evolved to solve (Crosby et al., 2020). Future work should make use of the Sexual Disgust Inventory to elucidate the different sexual disgust motives involved in moral judgment, as TDDS-sexual disgust measure doesn't allow for these domain-level distinctions to be made.

Revisiting the Sanctity (Purity) Foundation

A sensible question to consider next is whether the conceptualization of sanctity (purity) as a moral domain that responds to violations that involve disease is accurate (Graham et al., 2013). Our finding that sexual disgust is more strongly related to the sanctity domain than is canonical disgust is inconsistent with this conceptualization. Rather than performing a disease mitigating function, our results suggest that the sanctity domain might function to regulate sexual behavior, by facilitating moral condemnation of actions that undermine monogamy and conventional sexuality (e.g., promiscuity, infidelity, casual sex). Future theorizing on the topic should consider that the moralization of sanctity norms might not result from just disease avoidance mechanisms but also sexual avoidance mechanisms. Recognizing this complexity might involve dividing the sanctity domain into two sub-domains, distinguishing norms enforcing disease mitigation from norms enforcing conventional sexuality.

Even if one grants that the conceptualization of the sanctity construct is accurate (i.e., rooted in disease avoidance; Graham et al., 2013; Haidt & Graham, 2007), various measures of sanctity have questionable construct validity, thus posing a problem for interpretating the sanctity domain (Gray et al., 2021). For instance, the most widely used measure of sanctity concerns – the Moral Foundations Questionnaire sanctity subscale - contains an item that has nothing to do with disease avoidance (“Whether or not someone acted in a way that God would approve of”; Graham et al., 2011). Hence, using the Moral Foundations Questionnaire sanctity subscale (65% of the effect in this meta-analysis were based on the Moral Foundations Questionnaire) might underestimate (canonical) disgust-sanctity effects because some of the sanctity items do not represent the sanctity construct. However, we suspect that this concern is unlikely to dramatically alter our results, since the other five items that form part of the sanctity subscale are nominally about disease (or could at least be interpreted in this way). Nonetheless, future research should develop more construct valid measures of the sanctity domain.

Implications for Causal Theories of Disgust in Moral Judgment

Our results are also consistent with the idea that disgust has a causal effect on moral judgment (although not sufficient to justify the claim). There are two key points to be made here. First, as discussed earlier, research on disgust sensitivity is agnostic as to whether the causal role of disgust in moral judgment is incidental or integral. Taken with Landy and Goodwin (2015) (which suggests that incidental canonical disgust does not play a clear causal role in moral judgment) our finding that disgust sensitivity is robustly associated moral judgment suggests that integral disgust (of various kinds) might be a better candidate for studying the causal role of disgust in moral judgment. Surprisingly, there is a dearth of moral judgment research incorporating the manipulation of integral disgust (for an exception, see Wisneski & Skitka, 2017). We encourage future work studying the causal role of disgust in moral judgment to explore this possibility.

Second, our finding that sexual disgust explained more of the variance in some moral judgments than canonical disgust raises the possibility of sexual disgust as a candidate for playing a causal role in moral judgment (especially of the sanctity domain). To date, experimental approaches to disgust – incidental and integral – have only examined inductions of canonical disgust on moral judgment, and thus it is an open question as to whether sexual disgust causally influences moral judgment. Future work should attempt to manipulate the different sexual disgust motives and study the effects on the different domains of moral judgment (while controlling for pathogen disgust).