In dataset 1, we examined the difference in moral foundations between autistic individuals (n = 307) and typical individuals (n = 415). Moral foundations were measured with the MFQ. Participants were recruited through the Cambridge Autism Research Database (CARD). Between-subject effects from MANOVAs showed no significant differences between autistic and typical people for Care in either sex (p = 0.32 for females and 0.39 for males). This is an important and unexpected finding that supports hypothesis 1. The benefit of dataset 1 is that it allowed us to observe differences in moral foundations in autistic people and typical people.

Autistic females scored higher on the Fairness foundation than typical females (F(1, 447) = 13.00, p < 0.001). This supports hypothesis 2. Autistic females scored lower on Loyalty (F(1, 447) = 11.93, p < 0.01) and Authority (F(1, 447) = 8.27, p < 0.01) than typical females, supporting hypothesis 3. There was no significant difference in any of the five moral foundations between autistic males and typical males. The difference between autistic males and typical males on Fairness was not significance (p = 0.08). Overall, there was no support for hypotheses 2 and 3 for males.

Figure 1 displays mean differences between autistic people and typical people across each of the five moral foundations (means, SDs, and Cohen’s d is reported in Additional file 1: Table S1). As can be seen, this set of analyses show that overall, autistic people did not substantially differ from typical people on any of the five foundations.

Means, separated by sex, on the five moral foundations for autistic and typical people in dataset 1. Error bars are based on 95% CIs. This figure displays mean differences between autistic people (dark green) and typical people (light green) for scores on each of the five moral foundations. Panel A displays mean differences for females and panel B displays mean differences for males. P values are presented for significant and notable differences. Mean scores on the MFQ range from 0 to 5

Given that the analysis showed little to no general differences between autistic and typical people on the five moral foundations, we decided to make more nuanced comparisons related to moral foundations. In the next step of the analysis, we took this opportunity to examine the relationship between the Care and Fairness foundations by observing a Care/Fairness “ratio” for autistic and typical participants. The Care/Fairness “ratio” is the difference in endorsement on Care compared to Fairness, and allows us to observe if one foundation is valued more than another. Paired sample t-tests showed that autistic females endorsed Fairness over Care (t(130) = 2.98, p < 0.01). Autistic males also endorsed Fairness over Care (t(134) = 6.26, p < 0.001). This provided further support for hypothesis 2. For comparison purposes, typical females showed the opposite profile by endorsing Care over Fairness. (t(160) = 1.99, p < 0.05). However, typical males endorsed Fairness over Care t(112) = 2.61, p < 0.05). All means, SDs, and Cohen’s ds are presented in Additional file 1: Table S2. These results showed that the differences between autistic and typical people on their endorsement of the Care and Fairness foundations are only small, if any.

We next sought to better understand the different cognitive processes (i.e., empathizing and systemizing) that people use in their moral judgements. We leveraged additional data available for a subsample of autistic and typical participants who completed the EQ and SQ. Because of the smaller sample size (ns = 85 autistic people and 173 typical people), instead of conducting analyses separately for females and males, we conducted linear regressions while holding sex constant. Table 1 reports Pearson correlations and Additional file 1: Table S3 reports results from linear regressions. There are several notable results. First, EQ scores were positively associated with Care for both autistic (r = 0.25, p < 0.05) and typical people (r = 0.28, p < 0.01). SQ scores were positively associated with Fairness for autistic people (r = 0.24, p < 0.05), but not typical people (r = − 0.04, p = ns). This suggests that systemizing correlates with moral foundations for autistic people, but not for typical people. Furthermore, in general, this set of analyses shows that both empathizing and systemizing are indeed involved as cognitive processes in moral foundations and that people may differ in their moral foundations based on their empathizing and systemizing traits.

We previously discussed the important distinction between empathy profiles in autistic and typical people, namely, that autistic people tend to have lower cognitive empathy with intact or elevated affective empathy. This distinction might influence their moral foundations. Thus, in an exploratory analysis, we next examined the two major components of empathy (affective and cognitive empathy) and their relationship to moral foundations. Autistic people did not significantly differ from typical people on affective empathy (p = 0.174), however, autistic people did score significantly lower on cognitive empathy (p < 0.001) (Additional file 1: Table S4). This confirms that indeed in our sample, that autistic people have lower cognitive empathy with intact affective empathy, just as previous research suggests.

Given that our participants differed in terms of their cognitive and affective empathy, we next analyzed how these two different components of empathy are associated with moral foundations. As seen in Table 1, affective empathy was positively associated with Care for autistic (r = 0.29, p < 0.01) and typical people (r = 0.39, p < 0.01). Cognitive empathy was not significantly correlated with Care for autistic people (r = 0.12, p = ns), but was significantly correlated for typical people (r = 0.28, p < 0.01). This suggests that for autistic people, moral foundations are influenced more by their affective empathy levels than cognitive empathy levels. For typical people, their moral foundations seem to be influenced by both empathy components.

We next investigated if the number of autistic traits a person has is associated with their moral foundations. Results from Pearson correlations for the Autism Spectrum Quotient (AQ) and the social skills component of the EQ are reported in Additional file 1: Tables S3 and S5). The results were found to be largely non-significant which further provides evidence for our finding that autistic people and typical people do not differ much in terms of their moral foundations (i.e., if there were substantial differences, we would expect large associations with AQ scores).

To summarize, there were five important findings from dataset 1. First, autistic people overall had similar moral foundation scores to typical people. Second, autistic people scored the same on Care and this was true for both females and males. Third, the significant differences that were found were relatively small and showed that autistic females score higher on Fairness than typical females. In terms of the Care/Fairness ratio, both autistic females and males endorsed Fairness over Care. This suggests that Fairness may play a larger role in the moral judgements in autistic people than in typical people. Fourth, empathy is correlated with Care in both autistic and typical people. Fifth, systemizing correlated with Fairness in autistic people but not in typical people.

These correlations between moral foundations and both empathizing and systemizing in dataset 1 confirm our hypothesis that empathy and systemizing are two different constructs that influence the way in which people cognitively process moral judgments. However, the subsample size of people who completed the EQ and SQ in dataset 1 was relatively small. The sample size did not allow us to make observations about E–S types, which considers the standardized difference between scores on the EQ and SQ (Introduction and Methods). Therefore, we sought to leverage a larger dataset that included EQ and SQ scores with moral foundation scores. Furthermore, dataset 1 had no data on politics—how a person identifies politically is an extension of their moral judgments, and as discussed in the Introduction, indeed political identification and moral foundation scores.

To address the gap from dataset 1, dataset 2 was collected from www.YourMorals.org and consisted of more than 7000 typical participants who completed the MFQ and shortened versions of the EQ and SQ (Methods). This enabled us to observe how the five E–S cognitive types in the general population are associated with the five moral foundations. We calculated D-scores for each participant which is the basis of E–S classifications. D-scores are the standardized difference between a person’s scores on the EQ and SQ (see Methods). High D-scores indicate systemizing and low D-scores indicate empathizing. E–S cognitive type classifications are based on D-scores. The distribution of E–S types in dataset 2 is presented in Additional file 1: Table S6. The dataset did not ask participants about their clinical diagnoses so we were unable to identify participants who may have an autism diagnosis.

In the first stage of the analysis in dataset 2, we aimed to see if there was an association between D-scores and moral foundation scores. Thus, we calculated Pearson correlations between D-scores and scores on each of the five moral foundations (Table 2). As can be seen, D-scores were negatively correlated with Care for both females (r = − 0.26, p < 0.001) and males (r = − 0.24, p < 0.001). However, D-scores were also negatively correlated with Fairness for both females (r = − 0.08, p < 0.001) and males (r = − 0.10, p < 0.001), albeit to a lesser degree than Care. This was contrary to hypothesis 2, which predicted D-scores would be positively corelated with Fairness.

Since low D-scores indicate a drive to empathize while high D-scores indicate a drive to systemizing, the correlational results thus far from dataset 2 seem to suggest that Fairness scores are accounted for more by empathy than systemizing (logic and reasoning). To further examine if this is the case, we correlated each of the five moral foundations with EQ and SQ scores. The results showed the same pattern. Care was positively correlated with EQ and negatively correlated with SQ, for both females and males. Fairness was positively correlated with EQ for both sexes, but had an almost zero correlation with SQ for both females and males (see Table 2). This further contrasts hypothesis 2 and suggests that responses to Fairness items on the MFQ may not be driven by systemizing or reasoning more generally, in the general population, and is consistent with the results from the typical sample in dataset 1. Therefore, it appears that systemizing is only correlated with Fairness scores in autistic people, but not in typical people. Therefore, autistic people may rely more on their systemizing when making moral judgments about Fairness (as seen in dataset 1), but typical people do not to the same extent (as seen in datasets 1 and 2). This provides more evidence that even though autistic people and typical people may end up making similar moral judgments, the cognitive processes used to make those judgments may be different.

We next made more nuanced observations of moral foundations and empathizing and systemizing. Rather than relying on just D-scores alone, we observed how five different classifications of empathizing and systemizing are associated with moral foundations. Toward that end, in the second stage of analysis in dataset 2, we divided the sample into the five E–S types based on their D-scores. We then conducted MANOVAs separately for females and males to examine differences in moral foundation scores between the E–S types (see Additional file 1: Tables S7–S9 for the results, including all Ms, SDs, effect sizes, and p values, and results from post-hoc tests). Results from the MANOVAs showed that empathizing types (Extreme Type E and Type E) scored higher on Care than systemizing types (Type S and Extreme Type S), for both females and males. This provided support for hypothesis 1. However, empathizing types also scored higher on Fairness than systemizing types for both females and males. This contradicts hypothesis 2. Figure 2 displays mean scores on each of the five moral foundations by each of the five cognitive types.

Means, separated by cognitive type and sex, on the five moral foundations in dataset 2. Error bars are based on 95% CIs. Panel A displays results for females and panel B displays results for males

In dataset 1, we made observations about the nuanced relationships between Care and Fairness by observing the Care/Fairness ratio. Thus, in the third stage of the analysis of dataset 2, we examined the Care/Fairness ratio for each E–S type. Paired-sample t-tests showed that females with empathizing types (Type E and Extreme Type E) endorsed Care over Fairness (ps < 0.001). Males who were Extreme Type E endorsed Care and Fairness equally (p = 1.00), but males with a Type E endorsed Fairness over Care (p < 0.001). Both females and males who were Type S (extreme Type S) endorsed Fairness over Care (p < 0.001). Females who were Type S scored descriptively higher on Fairness than Care to a degree that approached, but was not significant (p = 0.08). Males who were Extreme Type S endorsed Fairness significantly more than Care (p < 0.001) (Ms and SDs are reported in Additional file 1: Table S9).

We were surprised that systemizing (as measured by the SQ) and systemizing types, were not more strongly associated to Fairness in the general population, as our second hypothesis had predicted. Rather, it appeared that lower empathizing may be contributing to a person’s Care and Fairness scores. Accordingly, to gain a better understanding of the role of empathizing and moral foundation scores, we conducted exploratory linear regressions, using SQ scores, cognitive empathy (measured via the perspective taking facet of the IRI) and affective empathy (measured via the empathic concern facet of the IRI) as predictors of Care (r2 = 0.27 for females and 0.31 for males) and Fairness (r2 = 0.09 for females and 0.19 for males) (Additional file 1: Table S10). For Care, affective empathy was the most significant predictor for females (β = 0.52, p < 0.001) and males (β = 0.56, p < 0.001). For males, SQ scores were a negative predictor of Care (β = − 0.14, p < 0.001). For Fairness, affective empathy was the only significant positive predictor for both females (β = 0.31, p < 0.001) and males (β = 0.45, p < 0.001). (For supplementary analysis, see Additional File 1: Fig S1). These analyses further showed that the affective component of empathy is contributing most to Care and Fairness in typical people, which is consistent with findings from dataset 1, and also that systemizing is not correlated to Fairness scores for typical people, which is also consistent with dataset 1. In other words, affective empathy levels appear to be more influential in their moral foundations of Care and Fairness than cognitive empathy, and systemizing levels.

In the fourth stage of the analysis in dataset 2, we extended the results by leveraging data about an additional 6th moral foundation proposed by Haidt (2012): Liberty. We leveraged new data from the YourMorals.org database and included participants who completed a 37-item survey that included 11 items that had been developed to explore Liberty, divided into two facets: Economic Liberty (e.g., “People who are successful in business have a right to enjoy their wealth as they see fit”) and Lifestyle Liberty (e.g., “People should be free to decide what group norms or traditions they want to follow”. See Methods and Additional File 1). We included only the 805 participants who had completed that survey and who had also completed the EQ and SQ questionnaires (see Methods). We correlated D-scores, EQ scores and SQ scores with the two liberty facets. We focused on D-scores rather than distinctions between the five cognitive types since the second part of the dataset had a smaller sample size (which would yield very small ns for the extreme cognitive types), compared to the first part of the dataset. As can be seen in Table 3, D-scores (high scores indicate systemizing and low scores indicate empathizing cognitive types) were positively correlated with both subsets of Liberty items. EQ scores were negatively correlated with Lifestyle Liberty and Economic Liberty. Surprisingly, SQ was not correlated with either of those two facets. The pattern of correlations was largely consistent for both female and male participants.

In the next stage of analysis, we addressed the following question: Do associations found for empathizing and systemizing and moral foundations also manifest in political identification? To address this question, we analyzed political self-identification in the dataset 2. D-scores were positively correlated with a 7-point scale for the liberalism-conservatism spectrum for females (r = 0.07, p < 0.001) and males (r = 0.09, p < 0.0001), meaning that conservatives had slightly higher D scores. This relationship is more clearly illustrated in Fig. 3, where we have plotted the percentage of men and women who self-identified as liberal or as conservative, as a function of E–S type. While most of the sample in dataset 2 identifies as liberal, the percentage of liberals decreases as we move from the drive to empathize to the drive to systemize. The opposite trend occurs for conservatives.

The percentage of self-identified liberals and conservatives for each cognitive type in Study 2. Blue solid lines indicate female liberals, blue dotted lines indicate male liberals, red solid lines indicate female conservatives, and red dotted lines indicate female conservatives

A unique aspect of dataset 2 is that it does not force participants to place themselves on the left–right axis (from liberal to conservative), unlike most surveys. Rather, it provides the option to identify as “libertarian.” We measured how D-scores differ across political self-identification for four categories: liberal, moderative, conservative, and libertarian. ANOVAs showed that libertarians had significantly higher D-scores than the three other political categories for females (F(3, 3323) = 22.17, p < 0.0001) and males (F(3, 3862) = 36.24, p 0.0001) (Ms, SDs and results from post hoc tests are in Additional file 1: Tables S11-S12). This relationship is most clearly illustrated in Fig. 4, which plots the percentage of participants with each cognitive type who self-identified as libertarian. Libertarians are almost non-existent among those who are extreme Type E, but are common among those who are extreme Type S: 10% of women and 17% of men. More specifically, the percentage of libertarians differed significantly across the five cognitive types, for both females (χ2 (4, 3458) = 46.39, p = 2.04 × 10–9) and males (χ2 (4, 4137) = 75.98, p = 1.24 × 10–15). The majority of people in each E–S type identified as liberal (68% for Extreme Type S to 90% for Extreme Type E in females, and 64% in Extreme Type S to 80% for Type E in males). However, libertarians within each cognitive type increased across the cognitive types reaching its highest proportion in Extreme Type S (10% for females and 17% for males) (Fig. 4) (percentages for political identification and E–S types are in Additional file 1: Table S13).

Percentage of self-identified libertarians for E–S cognitive type in dataset 2. The red line indicates females and the blue line indicates males

In the last stage of the analysis, we made observations about the contributions of sex differences compared to E–S cognitive types in predicting moral foundations across both datasets. Toward that end, we conducted stepwise linear regressions in both datasets. In dataset 1, we entered group type (autism and typical) in the first step as a control variable. Then we entered sex (female and male) in the second step, and D-score (the basis of E–S types) in the third step, and observed the amount of variance that D-scores explained over and above sex in predicting scores on the Care foundation. We then conducted a separate regression, this time with D-scores entered in the second step and sex entered in the third step. We repeated this set of analyses separately for all five moral foundations in dataset 1. Results showed that D-score accounted for a 7% increase in the variance explained (r2 = 0.10, change in r2 = 0.07, change in F change = 17.94, p < 0.001), which was an improvement from r2 = 0.03 using sex without D-scores. This suggests that D-scores account for at least three times more variance in the Care foundation than does sex. Sex did not significantly improve the model above and beyond D-scores, suggesting that the differences we observe between the sexes on the Care foundation may really be about differences in E–S types. There were no significant differences observed between D-scores and sex in predicting Fairness, Authority, or Loyalty. Sex did predict Sanctity above and beyond D-scores (r2 = 0.03, change in r2 = 0.02, change in F = 5.74, p < 0.05), but D-scores did not improve the model above sex (change in r2 = 0.001, change in F = 0.23, p = 0.63).

We conducted the same analyses in dataset 2, however, without included group type as a controlling variable since there was no diagnostic data in dataset 2. The results largely replicated. D-scores predicted Care above and beyond sex (r2 = 0.14, change in r2 = 0.05, change in F = 432.82, p < 0.001). In this dataset, sex also contributed variance above and beyond D-scores (r2 = 0.14, change in r2 = 0.03, change in F = 292.59, p < 0.001). As can be seen in comparing the changes in r2, D-scores accounted for almost two times more variance than did sex. Though D-scores and sex contributed significantly to the models for Fairness, Authority, Loyalty, and Sanctity foundations, they were not differences greater than 1% in the increased proportions of variance between sex and D-scores, suggesting that they had relatively similar contributions to those foundations. Overall, these sets of regression analyses from both datasets show that D-scores and sex contribute relatively equally to moral foundations, except for the case of Care, where D-scores account for two to three times more variance than does sex.