Moral dumbfounding occurs when people defend a moral judgement even though they cannot provide a reason in support of this judgement. It manifests as an admission of not having reasons, or the use of unsupported declarations (\it's just wrong") or tautological reasons (\because it's incest") as justi cations for a judgment. It typically occurs for harmless taboos. It is cited as evidence for intuitionist or dual-process theories of moral judgement over rationalist approaches, however the phenomenon remains poorly understood. We test a con ict in dualprocesses explanation of moral dumbfounding, where moral dumbfounding is an example of con ict between a habitual response (making a judgement) and a response that results from deliberation (providing a reason for the judgement). One prediction of this explanation is that under speci c manipulations (e.g., cognitive load), responses should vary in predictable ways. The dumbfounding paradigm involves three possible responses: (a) providing reasons for a judgement (System 2/deliberative); (b) accepting the counter-arguments and rating the behaviour as \not wrong" (System 1/habitual); (c) a dumbfounded response (System 1/habitual). Cognitive load manipulations have been shown to inhibit deliberative responding. We present 4 studies in which dumbfounded responding was investigated under cognitive load manipulations. We hypothesised that rates of providing reasons would be reduced under cognitive load. The identi cation of reasons was inhibited in Studies 1 and 3, but not in Studies 2 and 4. The results do not provide strong evidence that moral dumbfounding be explained as con ict in dual-processes. Future research should investigate this further, addressing methodological limitations identi ed.
Moral dumbfounding occurs when people defend a moral judgement even though
they cannot provide a reason in support of this judgement [
Despite the in uence of moral dumbfounding on the morality literature, the
phenomenon is not well understood. Until recently [
Drawing on dual-process explanations of moral judgement [
One prediction of explaining moral dumbfounding as con ict and adopting
a dual-process model of moral judgement, is, that under speci c manipulations,
moral judgements should vary in predictable ways. In the dumbfounding paradigm,
there are generally three possible responses: (a) providing reasons for a judgement;
(b) accepting the counter-arguments and rating the behaviour as \not wrong";
(c) a dumbfounded response. These responses can be positioned in terms of
dualprocesses, whereby option (a) requires deliberation (traditionally System 2), and
options (b) and (c) are grounded largely in habitual responding (traditionally
System 1) [
Above we suggest that identifying reasons for a judgement requires more deliberation
than accepting the counter-arguments, (accepting the \logical" argument, and
adopting the judgement that is grounded in reasons is System 1/habitual response).
We also hypothesised that identifying reasons requires more deliberation than
a dumbfounded response (the initial judgement of the behaviour was habitual).
Manipulations of cognitive load are known to inhibit deliberative responding,
[
We conducted four studies in which we tested the hypothesised relationship between cognitive load and dumbfounded responding. We hypothesised that a cognitive load manipulation would inhibit the identi cation of reasons, leading to higher rates of dumbfounding or changing judgements (or both). 4
The aim of Study 1 was to investigate if a cognitive load manipulation in uenced participants' ability to justify their judgement. 4.1
Study 1: Method
Participants and Design Study 1 was a between subjects design. The dependent
variable was response to the critical slide (see below). The independent variable
was cognitive load with two levels: present and absent [
A total sample of 66 participants (55 female, 11 male; M age = 22.42, min = 18, max = 57, SD = 6.86) took part. Participants were undergraduate students, postgraduate students, and alumni from Mary Immaculate College (MIC), and University of Limerick (UL). Participation was voluntary and participants were not reimbursed for their participation.
Procedure and Materials Data were collected using an online questionnaire. Data collection took place in a designated computer laboratory in MIC. The experimenter remained in the laboratory for the duration of the study. Participants were rst presented with an information sheet and consent form.
Participants in the experimental condition were presented with an eight digit number/letter string and asked to memorise the sequence. After 30 seconds, the experiment progressed to the next slide. Participants had the option to click \ok" and progress to the next slide after 15 seconds.
Participants were then presented with the \Julie and Mark" (Incest ) vignette
[
Dumbfounding was measured using the critical slide [
Study 1: Results Threre was no di erence in initial judgement depending on cognitive load, t (63.9) = 1.256, p = 0.214. The responses to the critical slide for the experimental group (N = 33) and the control group (N = 33) are displayed in Fig. 1. A chi-squared test for independence revealed a signi cant association between experimental condition and response to the critical slide, 2(2, N = 66) = 7.53, p = .023, V = .27: under cognitive load more participants (15) selected \There is nothing wrong" than in the control group (5), the observed power power was .69. The aim of the Study 1 was to investigate if dumbfounded responding was in uenced by cognitive load. Speci cally, adopting a dual-systems model of moral judgement, it was hypothesised that cognitive load would lead to reduced levels of deliberative responding, leading to a reduction in successfully identifying reasons for judgements. This may lead to (a) increased levels of dumbfounding, or (b) increased selecting of the \nothing wrong" response. As predicted we found higher rates of selecting of the \nothing wrong" response in the cognitive load group compared to the control group. 5
Study 1 demonstrated interesting variability in responses to the critical slide
depending on cognitive load. The aim of Study 2 was to assess the replicability
of the results of Study 1, using an online sample. In Study 1, the experimenter
was in the room with the participants. This made it more di cult for participants
to cheat on the memory task. This is not possible with an online sample. An
alternative cognitive load manipulation was taken from De Neys and Schaeken
[
Study 2: Method Participants and Design Study 2 was a between subjects design. The dependent variable was response to the critical slide. The independent variable was cognitive load with two levels: high and low.
A total sample of 100 participant (56 female, 44 male; M age = 38.38, min = 19, max = 72, SD = 12.41) took part. Participants in this sample were recruited using Amazon's MTurk. Participants were paid $0.50 for their participation. Participants were recruited from English speaking countries or from countries where residents generally have a high level of English (e.g., The Netherlands, Denmark, Sweden).
Procedure and Materials Data were collected using an online questionnaire.
Materials were largely the same as in Study 1, with a change to the cognitive
load manipulation. Cognitive load was manipulated using a dot-pattern memory
task [
Participants were presented with a 3 x 3 grid containing a dot pattern.
This image disappeared after one second. Participants then answered a question
relating to the moral judgement task. Following this, participants were asked
to reproduce the dot-pattern. All participants took part in the memory task,
and cognitive load was manipulated by varying the complexity of the patterns
presented [
Study 2 proceeded in much the same way as Study 1. There were four target questions during which participants were engaged in the memory task. A di erent pattern was presented before each of the following: the initial judgement, the initial opportunity to provide reasons, the critical slide, and the revised judgement. After each of these questions participant were required to reproduce the pattern. As in Study 1, dumbfounding was measured using the critical slide.
a
b
Threre was no di erence in initial judgement depending on cognitive load, t (97.77)
= 0.962, p = 0.339. The responses to the critical slide for the experimental group
(N = 51) and the control group (N = 49) group are displayed in Fig. 3. A
chi-squared test for independence revealed no association between experimental
condition and response to the critical slide, 2(2, N = 100) = 1.07, p = .585, V
= .10. The observed power was .14.
Engagement with the Memory Task It is possible that the di erence
in results observed between Study 1 and Study 2 is due to the alternative
manipulation of cognitive load employed. In Study 1, the control group did not
engage in any task, however, adopting De Neys and Shaeken's procedure [
Rates of successful reproduction of the dot patterns in Study 2 were much
lower than reported by De Neys and Shaeken [
The memory task involved correctly placing dots in a 3 x 3 grid. For the scoring of this task, each of the nine places in the grid could be marked/not marked correctly or incorrectly, making 9 the total possible number of correct responses. If a person misplaced one dot in the pattern this would count for 2 incorrect places in the grid: the mark in the incorrect place, and the absence of a mark in the place it should have been. A participant who received a score of 7, could reasonably be taken to have engaged with the task, and simply made a slip. As such, this was taken as the cut-o point for identifying engagement. This resulted in 56 participants being identi ed as engaging with the memory task, and 44 being identi ed as not engaging with the task.
Responses to critical slide for participants who engaged with the memory task and participants who did not engage with the memory task were analysed separately (see Fig. 4). A chi-squared test for independence revealed an association between engagement in the memory task and response to the critical slide, 2(2, N = 100) = 7.68, p = .021, V = .28. The observed power was .70. 5.3
Study 2: Discussion The aim of Study 2 was to replicate Study 1. As a replication it failed. However, interesting variability was observed when engagement with the memory task was accounted for. Two problems with the cognitive load manipulation were identi ed. Firstly, the control did not serve as an appropriate control. Secondly, (and unsurprisingly) the manipulation is only e ective if participants engage with the memory task. A follow-up study addresses each of these. 6
In Study 2 the role of engagement with the memory task emerged as an important moderator of the e ectiveness of the cognitive load manipulation. Study 3 was conducted in order to test if cognitive load a ects participants' ability to identify reasons for their judgements, when accounting for engagement with the memory task.
100% e s n o p s e r cah75% e g n it c e l e tss 50% n a p tiir c a fop25% e g a t n e c r e P 0% Participants and Design Study 3 was a between subjects design. The dependent variable was response to the critical slide. The independent variable was cognitive load with two levels: present and absent.
Following the elimination of 34 participants who scored less than 7 on the memory task we were left with a nal sample of 129 participants (74 female, 55 male; M age = 40.26, min = 20, max = 72, SD = 13.04). Participants in this sample were recruited through MTurk (under the same conditions as Study 2). Procedures and Materials Study 3 was the same as Study 2 with two changes. The control group did not take part in a memory task, and to avoid task fatigue, the dot patterns presented alternated between the easy 3-dot patterns and the complex 4-dot patterns.
A score of 7 or higher on the memory task that accompanied the critical slide was selected as the measure of engagement with the memory task. Only participants who engaged with the task were eligible for analysis. Other than the two changes described above, Study 3 was the same as Study 2. 6.2
Study 3: Results Threre was no di erence in initial judgement depending on cognitive load, t(123.06) = 0.165, p = 0.870. The responses to the critical slide for the experimental group (N = 68) and the control group (N = 61) are displayed in Fig. 5. A chi-squared test for independence revealed a signi cant association between experimental condition and response to the critical slide, 2(2, N = 129) = 6.51, p = .039, V = .223. The observed power was .63.
100% e s n o p s e r cah75% e g n it c e l e tss 50% n a p itir c a fpo25% e g a t n e c r e P 0% Study 3 demonstrated the predicted relationship between engagement with a cognitive load task and providing reasons for a judgement. As expected, engagement with a cognitive load task reduced the rates of providing reasons. However, Study 3 did not include an objective manipulation check. As such a follow up study was conducted that included an objective manipulation check. 7
Participants and Design Study 4 was a between subjects design. The dependent variable was response to the critical slide. The independent variable was cognitive load with two levels: present and absent.
Following the removal of 29 participants for non-engagement with the memory task, we were left with total sample of 127 participants (84 female, 43 male; M age = 41.19, min = 21, max = 74, SD = 13.91). Participants in this sample were recruited through MTurk (under the same conditions as Studies 2 and 3).
Procedures and Materials Study 4 was the same as Study 3 with one
change, the inclusion of a manipulation check. A prose paragraph was included
after participants made their revised judgements. Participants were then asked
three comprehension questions relating to the prose paragraph. It was expected
that participants in the control group would perform better at this task than
participants under cognitive load [
Condition
Cognitive load Control 29.7% 19.0% Threre was no di erence in initial judgement depending on cognitive load, t (139.46) = 0.795, p = 0.428. The responses to the critical slide for the experimental group (N = 64) and the control group (N = 61) are displayed in Fig. 6. Though the pattern of responses is in the predicted direction, a chi-squared test for independence revealed no signi cant association between experimental condition and response to the critical slide, 2(2, N = 127) = 2.047, p = .359, V = .13. The observed power was .23. The predicted relationship between cognitive load and dumbfounded responding was not observed in Study 4.
The responses to the manipulation check questions were investigated. There was no di erence in the number of correct answers to these questions between the cognitive load group and the control group F (1, 124) = .33, p = .569, partial 2 = .003. There was also no di erence in time taken to read the vignette between the groups F (1, 125) = 2.57, p = .112, partial 2 = .020.
Study 4: Discussion The relationship between cognitive load and providing reasons observed in previous studies was not observed in Study 4. However, the manipulation check did not reveal any di erences between the control group and the experimental group, suggesting that the cognitive load manipulation was ine ective in Study 4. 8
Combined Results, Discussion, and Conclusion
100%
The combined results for Studies 1-4 are displayed in Fig. 7. Two measures of engagement with the memory task were taken in the combined analysis that follows, (a) a score of 7 or higher (as in previous analyses) and (b) a score of 8 or higher. When engagement with the memory task was accounted for (using the primary measure of engagement (N = 422, 221 experimental, 201 control), a signi cant association between cognitive load and responses to the critical slide was found, 2(2, N = 422) = 12.675, p = .002, V = .17. The observed power was .90. When the stricter measure of engagement was employed (N = 413, 212 experimental, 201 control), a signi cant association between cognitive load and response to the critical slide was found, 2(2, N = 413) = 14.847, p <.001, V = .19. The observed power was .94. Furthermore, a mini meta-analysis was conducted and found that cognitive load signi cantly in uenced responding across all studies in 2(8) = 23.81, p = .002 (Fisher's method); or when weighting for sample size, z = 2.95, p = .002 (Stou er's Z-score method).
The studies presented here provide some weak evidence for a con ict in dual-process explanation of moral dumbfounding (cognitive load inhibits reason giving). The dependent variable in these studies is nominal/categorical. This is not well suited for identifying small or subtle e ects. Responses in the dumbfounding paradigm are not easily in uenced by experimental manipulation, leading to inconsistent results, and the ndings of the studies presented here are inconclusive. Future research should employ more robust manipulations and controls for the engagement of participants, to resolve the inconsistencies in the results presented here.