0.1 Question 3 — What Explains the Relationship Between Religiosity and the Frequency of Seeking Moral Advice from AI Chatbots? (Exploratory Indirect Associations)

To identify plausible psychological pathways and statistical mediators of the relationship between religiosity and AI moral advice seeking, we examined several candidate mediators (Table 2). As our data are cross-sectional, we present these mediation models as exploratory, first-pass statistical tests of plausible indirect associations rather than confirmatory test of causal mechanisms. We first conducted regression analyses (simple regression and multiple regression models adjusted for demographic covariates) to establish which candidate mediators were significantly associated with the Religious Behavior Score. Even though we pre-registered the analysis, because of the large number of candidate mediators tested (9 in Study 1, 10 in Study 2), we applied a highly conservative Bonferroni correction to avoid false positives in our first-pass selection. We then restricted mediation analyses to those reaching significance under this corrected threshold (\(p < .05 / 9 = .0056\) in Study 1; \(p < .05 / 10 = .0050\) in Study 2) in at least one path-a model. Finally, we estimated parallel mediation models for each study using structural equation modeling. The parallel mediation models were pre-registered as exploratory (without specifying a priori how to structure the parallel models). To minimize criterion contamination and multicollinearity among multiple mediators within the same cluster, we selected the mediator that accounted for the largest proportion of variance from each significantly associated cluster.

Table 1: Study 1: Religious Behavior Score Predicting Each Candidate Mediator

Model 1

Model 2

Mediator

β

SE

95% CI LL

95% CI UL

p

Sig (Corrected)

β

SE

95% CI LL

95% CI UL

p

Sig (Corrected)

Overall Frequency of Seeking Moral Advice

0.37

0.05

0.27

0.47

< .001

Yes

0.33

0.06

0.22

0.44

< .001

Yes

Overall Interest in Seeking Moral Advice

0.31

0.05

0.21

0.41

< .001

Yes

0.27

0.06

0.16

0.39

< .001

Yes

General Open-Mindedness

0.23

0.05

0.13

0.34

< .001

Yes

0.27

0.06

0.15

0.38

< .001

Yes

Open-Mindedness on Moral Issues

0.15

0.05

0.04

0.25

0.006

No

0.19

0.06

0.07

0.31

0.002

Yes

Intellectual Humility (CIHS)

-0.04

0.05

-0.15

0.06

0.439

No

0.06

0.06

-0.06

0.19

0.337

No

Intellectual Humility (MMIH)

-0.06

0.05

-0.16

0.05

0.297

No

0.01

0.06

-0.12

0.13

0.885

No

Belief in Moral Objectivity

0.26

0.05

0.15

0.36

< .001

Yes

0.19

0.06

0.06

0.31

0.003

Yes

Perceived Valence of AI Chatbots as Moral Advisors

0.26

0.05

0.16

0.36

< .001

Yes

0.23

0.06

0.10

0.35

< .001

Yes

Perceived Authority of AI Chatbots as Moral Advisors

0.26

0.05

0.16

0.36

< .001

Yes

0.25

0.06

0.13

0.37

< .001

Yes

Table 2: Study 2: Religious Behavior Score Predicting Each Candidate Mediator

Model 1

Model 2

Mediator

β

SE

95% CI LL

95% CI UL

p

Sig (Corrected)

β

SE

95% CI LL

95% CI UL

p

Sig (Corrected)

Overall Frequency of Seeking Moral Advice

0.37

0.05

0.27

0.47

< .001

Yes

0.34

0.06

0.23

0.44

< .001

Yes

Overall Interest in Seeking Moral Advice

0.30

0.05

0.20

0.40

< .001

Yes

0.30

0.06

0.18

0.41

< .001

Yes

Open-Mindedness on Moral Issues

0.12

0.05

0.01

0.22

0.030

No

0.11

0.06

-0.01

0.23

0.074

No

Belief in Moral Objectivity

0.31

0.05

0.21

0.41

< .001

Yes

0.24

0.06

0.13

0.36

< .001

Yes

Perceived Authority of AI Chatbots as Moral Advisors

0.17

0.05

0.07

0.28

0.001

Yes

0.08

0.06

-0.04

0.20

0.178

No

Tendency to Anthropomorphize AI Chatbots

0.22

0.05

0.12

0.33

< .001

Yes

0.18

0.06

0.06

0.29

0.004

Yes

Fear of Negative Judgement

-0.06

0.05

-0.16

0.05

0.290

No

-0.01

0.06

-0.12

0.11

0.913

No

Self-Reflective Tendencies

0.07

0.05

-0.04

0.17

0.213

No

0.13

0.06

0.01

0.25

0.040

No

Deference to Authority (MFQ-2)

0.53

0.05

0.45

0.62

< .001

Yes

0.30

0.05

0.21

0.39

< .001

Yes

Deference to Authority (MAC-Q)

0.32

0.05

0.22

0.42

< .001

Yes

0.15

0.06

0.04

0.26

0.006

No

In Study 1, Religious Behavior Score was significantly associated with most candidate mediators. In both simple and multiple regression models (where continuous variables were standardized to report \(\beta\) coefficients to facilitate comparison), Religious Behavior Score predicted overall frequency of and interest in seeking moral advice from non-AI, non-religious sources (cluster a), general open-mindedness, open-mindedness on moral issues, and intellectual humility measured by CIHS and MMIH (cluster b), belief in moral objectivity (cluster c), and perceived valence and authority of AI chatbots as moral advisors (cluster d). Intellectual humility (CIHS and MMIH; cluster b) did not reach significance under the corrected threshold and was therefore excluded from subsequent analyses. In Study 2, Religious Behavior Score similarly predicted overall frequency of and interest in seeking moral advice (cluster a), belief in moral objectivity (cluster c), and perceived authority of AI chatbots (cluster d). Among the newly added Study 2 measures, tendency to anthropomorphize AI chatbots (cluster e) and both measures of deference to authority (cluster f) were significantly associated with Religious Behavior Score. Open-mindedness on moral issues (cluster b), fear of negative judgment, and self-reflective tendencies (cluster f) did not reach significance under the corrected threshold and were excluded from subsequent statistical mediation analyses.

Table 3: Study 1: Individual Statistical Mediation (Indirect Effect) Analyses

Path a

Path b

Indirect (ab)

Direct c'

Full Model

Mediators

X -> M

M -> Y
(given X)

Effect

95% CI LL

95% CI UL

X -> Y
(given M)

Proportion
Mediated

Selected for SEM

Overall Frequency of Seeking Moral Advice

0.26***

1.00***

0.26***

0.18

0.33

0.06

81.23%

Yes

Overall Interest in Seeking Moral Advice

0.21***

0.91***

0.19***

0.12

0.26

0.13**

58.85%

No

General Open-Mindedness

1.02***

0.08***

0.08***

0.04

0.12

0.24***

24.73%

Yes

Open-Mindedness on Moral Issues

1.14**

0.04***

0.04**

0.01

0.08

0.27***

13.79%

No

Belief in Moral Objectivity

3.82***

0.00

0.00

-0.03

0.03

0.31***

1.03%

No

Perceived Valence of AI Chatbots as Moral Advisors

0.20***

0.57***

0.11***

0.06

0.16

0.20***

35.22%

No

Perceived Authority of AI Chatbots as Moral Advisors

0.21***

0.55***

0.12***

0.07

0.17

0.20***

37.55%

Yes

Table 4: Study 2: Individual Statistical Mediation (Indirect Effect) Analyses

Path a

Path b

Indirect (ab)

Direct c'

Full Model

Mediators

X -> M

M -> Y
(given X)

Effect

95% CI LL

95% CI UL

X -> Y
(given M)

Proportion
Mediated

Selected for SEM

Overall Frequency of Seeking Moral Advice

0.24***

0.99***

0.24***

0.16

0.31

0.12*

66.04%

Yes

Overall Interest in Seeking Moral Advice

0.20***

0.82***

0.16***

0.11

0.23

0.19***

46.05%

No

Belief in Moral Objectivity

0.30***

0.04

0.01

-0.02

0.05

0.34***

3.72%

No

Perceived Authority of AI Chatbots as Moral Advisors

0.16**

0.71***

0.11**

0.05

0.18

0.24***

31.55%

Yes

Tendency to Anthropomorphize AI Chatbots

0.12***

0.78***

0.09***

0.05

0.14

0.26***

26.03%

Yes

Deference to Authority (MFQ-2)

1.77***

0.06**

0.11***

0.05

0.18

0.24***

31.76%

Yes

Deference to Authority (MAC-Q)

13.40***

0.01***

0.08***

0.04

0.13

0.28***

22.48%

No

Across both studies, the association between Religious Behavior Score and frequency of seeking moral advice from AI chatbots was most strongly accounted statistically for by an indirect effect via individuals’ overall tendency to seek moral advice from diverse non-AI, non-religious sources (cluster a). Perceived authority of AI chatbots as moral advisors (cluster d) served as a consistent secondary indirect pathway in both studies. Additionally, in Study 1, open-mindedness on moral issues (cluster b) contributed to a smaller but reliable indirect association (whereas in Study 2, it did not survive the initial Bonferroni correction and was not tested). Among the new candidate mediators introduced in Study 2, deference to authority (cluster f; both MFQ-2 and MAC-Q) and tendency to anthropomorphize AI chatbots (cluster e) also emerged as significant individual indirect pathways. Belief in moral objectivity (cluster c), fear of negative judgment and self-reflective tendencies (cluster f; Study 2 only) did not significantly mediate the relationship between Religious Behavior Score and frequency of seeking moral advice from AI chatbots and were excluded from parallel mediation analyses.

To examine the unique statistically contribution of each candidate mediator while accounting for their intercorrelations, we estimated a parallel statistical mediation model for each study using the lavaan package in R (Rosseel 2012). For each study, we selected the mediator with the highest proportion of variance explained from each cluster significantly predicted by Religious Behavior Score under our conservative Bonferroni correction. For Study 1, we included three mediators: overall frequency of seeking moral advice from non-AI sources (cluster a), open-mindedness on moral issues (cluster b), and perceived authority of AI chatbots as moral advisors (cluster d). For Study 2, we included four mediators: overall frequency of seeking moral advice from non-AI sources (cluster a), perceived authority of AI chatbots as moral advisors (cluster d), tendency to anthropomorphize AI chatbots (cluster e), and deference to authority — MFQ-2 (cluster f) (open-mindedness on moral issues [cluster b] was excluded as it did not survive the Bonferroni correction). All mediators included were allowed to intercorrelate, making both models saturated. Specific indirect effects were estimated using 5,000 bootstrap resamples.

Figure 1: Study 1: Parallel Mediation Model
Figure 2: Study 2: Parallel Mediation Model

Overall, in Study 1, the parallel statistical model explained 54.8% of the variance in the frequency of seeking AI moral advice (R²=0.548). The total indirect effect was b=0.293 (95% CI [0.212, 0.376]), with a residual direct association of b=0.023 (p=0.567). In Study 2, the model explained 59.2% of the variance (R²=0.592), with a total indirect effect of b=0.254 (95% CI [0.154, 0.357]) and a direct association of b=0.103 (p=0.037).

The parallel models revealed a consistent two-pathway pattern of indirect associations across both studies. The indirect association via the disposition to moral consultation from non-AI, non-religious sources (cluster a) was the dominant pathway in both studies, followed by a significant indirect association via the perceived authority of AI chatbots as moral advisors (cluster d). In contrast, the indirect association via open-mindedness on moral issues (cluster b) was tested only in Study 1 and was not statistically significant in the parallel model. In Study 2, neither the indirect association via the tendency to anthropomorphize AI chatbots (cluster e) nor via deference to authority (cluster f) reached significance.

References

Rosseel, Yves. 2012. “Lavaan: An R Package for Structural Equation Modeling.” Journal of Statistical Software 48 (2): 1–36. https://doi.org/10.18637/jss.v048.i02.