Running head: MANIPULATING MORAL DUMBFOUNDING 1
1 Manipulating Moral Dumbfounding: Inhibiting the Identification of Reasons for a Moral
2 Judgement
1 2 1 1 3 Cillian McHugh , Marek McGann , Eric R. Igou , & Elaine L Kinsella
1 4 University of Limerick
2 5 Mary Immaculate College ~ University of Limerick
6 Author Note
7 All procedures performed in studies involving human participants were approved by
8 institutional research ethics committee and conducted in accordance with the Code of
9 Professional Ethics of the Psychological Society of Ireland, and with the 1964 Helsinki
10 declaration and its later amendments or comparable ethical standards. Informed consent was
11 obtained from all individual participants included in the study. The authors declare that
12 there are no potential conflicts of interest with respect to the research, authorship, and/or
13 publication of this article. All authors consented to the submission of this manuscript. This
14 work was part funded by Mary Immaculate College seed funding.
15 Correspondence concerning this article should be addressed to Cillian McHugh,
16 University of Limerick, Limerick, Ireland. E-mail: [email protected] MANIPULATING MORAL DUMBFOUNDING 2
17 Abstract
18 Moral dumbfounding occurs when people defend a moral judgement even though they
19 cannot provide a reason in support of this judgement. It manifests as an admission of not
20 having reasons, or the use of unsupported declarations (“it’s just wrong”) or tautological
21 reasons (“because it’s incest”) as justifications for a judgment. We test a dual-processes
22 explanation of moral dumbfounding, where moral dumbfounding is an example of conflict
23 between a habitual response (making a judgement) and a response that results from
24 deliberation (providing a reason for the judgement). The dumbfounding paradigm involves
25 three possible responses: (a) providing reasons for a judgement (deliberative/controlled
26 process); (b) accepting the counter-arguments and rating the behaviour as “not wrong”
27 (habitual/automatic process); (c) a dumbfounded response (habitual/automatic process).
28 Cognitive load manipulations have been shown to inhibit deliberative responding. We
29 present 5 studies in which dumbfounded responding was investigated under cognitive load
30 manipulations. We hypothesised that rates of providing reasons would be reduced under
31 cognitive load. The identification of reasons was inhibited in Studies 1 and 3, but not in
32 Studies 2, 4 or 5. The results provide weak evidence for a dual-process explanation of moral
33 dumbfounding. We found some evidence that dumbfounded responding may be linked with
34 Need for Cognition.
35 Keywords: moral dumbfounding, dual-processes, reasons, intuitions, need for cognition
36 Word count: 11,025 MANIPULATING MORAL DUMBFOUNDING 3
37 Manipulating Moral Dumbfounding: Inhibiting the Identification of Reasons for a Moral
38 Judgement
39 Moral dumbfounding occurs when people defend a moral judgement even though they
40 cannot provide a reason in support of this judgement (Haidt, 2001; Haidt, Björklund, &
41 Murphy, 2000; see also McHugh et al., 2017). It is seen as evidence for the intuitive nature of
42 moral judgements and has been cited as support for both intuitionist and dual-process
43 theories of moral judgement (e.g., Cameron, Payne, & Doris, 2013; Crockett, 2013; Cushman,
44 2013; Cushman, Young, & Greene, 2010; Greene, 2008; Haidt, 2001; Prinz, 2005). Despite
45 the influence of moral dumbfounding on the morality literature, the phenomenon is not well
46 understood. Until recently (e.g., Royzman, Kim, & Leeman, 2015; McHugh et al., 2017),
47 empirical evidence testing the phenomenon was limited to a single study with a final sample
48 of N = 30, unpublished in peer-reviewed form. The dumbfounding paradigm involves
49 presenting participants with a moral scenario, and asking them to judge the behaviour in the
50 scenario. Participants are then asked to provide reasons for their judgement. Typically the
51 scenarios in the dumbfounding paradigm depict “harmless taboos” (Haidt & Björklund, 2008,
52 p. 196). McHugh et al. (2017), building on the original work by Haidt et al. (2000),
53 identified two measurable responses that may be taken as indicators of moral dumbfounding.
54 Firstly, people may explicitly admit to not having reasons for their judgement. Secondly,
55 people may use unsupported declarations (“it’s just wrong”) or tautological reasons
56 (“because it’s incest”) as justifications for a judgement. The current research uses the
57 methods developed by McHugh et al. (2017) to test one potential explanation of moral
58 dumbfounding.
59 Moral Dumbfounding: A Dual-Process Perspective
60 Drawing on dual-process theories of moral judgement (e.g., Greene, 2008; Brand, 2016;
61 Cushman, 2013), the studies presented here aim to test the hypothesis that moral
62 dumbfounding can be explained as a result of a dual-process conflict (Bonner & Newell, 2010; MANIPULATING MORAL DUMBFOUNDING 4
63 De Neys, 2012; De Neys & Glumicic, 2008; Evans, 2007). Conflicts occur when a
64 habitual/intuitive response is different from a response that results from deliberation.
65 Examples of such conflicts include, base rate neglect problems (Bonner & Newell, 2010; De
66 Neys, 2012; De Neys & Glumicic, 2008; Evans, 2007), the conjunction fallacy (De Neys, 2012;
67 Tversky & Kahneman, 1983), and the persistence of apparently “irrational” behaviours
68 (Evans, 2007), and compulsive behaviours (e.g. overeating, smoking, or gambling, Evans,
69 2008).
70 Another example of this kind of dual-process conflict is identified by Lerner and
71 Goldberg (1999), discussing a study by Rozin, Markwith, and McCauley (1994). In this
72 study, Rozin et al. (1994) found that people reported reduced willingness to contact various
73 items that they believed had prior contact with (i) an AIDS victim, (ii) someone who had
74 been in a car accident, or (iii) a murderer. This unwillingness persisted despite assurances
75 that these items are sanitary (Rozin et al., 1994). The original unpublished moral
76 dumbfounding study (Haidt et al., 2000) included a similar task; in addition to three moral
77 judgement tasks, Haidt et al. (2000) included two non-moral tasks. As part of one of the
78 non-moral tasks, the experimenter dipped a sterilised cockroach in a glass of juice.
79 Participants were then asked to drink from the glass. Much like the study by Rozin et al.
80 (1994), participants were unwilling to drink from the glass (Haidt et al., 2000).
81 Haidt et al. (2000) present this response as equivalent to dumbfounded responding in a
82 moral task; in both tasks participants reject a given action (either performing it in the case
83 of the drink, or accepting it in the case of the moral vignettes) despite the context removing
84 the typical reasons given for doing so. This parity suggests that the cognitive processes that
85 lead to this unwillingness to contact items seen as contaminated (identified as dual-process
86 conflict by Lerner & Goldberg, 1999) may similarly lead to moral dumbfounding. It is
87 possible that dumbfounded responding may also be explained as the result of dual-process
88 conflict, where the making of a moral judgement is a habitual/intuitive response, and MANIPULATING MORAL DUMBFOUNDING 5
89 identifying reasons involves deliberation. According to this interpretation, conflict occurs
90 when deliberation fails to identify reasons for an intuitive judgement, and this results in a
91 dumbfounded response. We aim to test this explanation here.
92 Dual-Processes in the Dumbfounding Paradigm
93 Normal cases of conflict can be resolved either by (a) the over-riding of the habitual
94 response or (b) the ignoring of the inconsistent information from deliberation. Consider the
95 following problem (taken from De Neys, 2012, p. 29):
96 A psychologist wrote thumbnail descriptions of a sample of 1000 participants
97 consisting of 995 females and 5 males. The description below was chosen at
98 random from the 1,000 available descriptions.
99 Jo is 23 years old and is finishing a degree in engineering. On Friday nights, Jo
100 likes to go out cruising with friends while listening to loud music and drinking
101 beer. Which one of the following two statements is most likely?
102 (i) Jo is a man
103 (ii) Jo is a woman
104 Jo appears to present as stereotypically male, and so the intuitive/habitual response is
105 (i): Jo is a man. Deliberation (and attending to the make up of the sample) reveals that Jo
106 is 199 times more likely to be a woman (99.5%) than to be a man (0.5%). In this case there
107 is a clear conflict between the intuitive/habitual response and the deliberative response. This
108 conflict can resolved if a person accepts that their intuition was incorrect and adopts the
109 deliberative response. Alternatively a person may choose to ignore the inconsistent
110 information that resulted from deliberation and maintain their initial judgement.
111 In the dumbfounding paradigm, conflict cannot be as easily resolved using these MANIPULATING MORAL DUMBFOUNDING 6
112 strategies. Firstly, regarding an attempt to over-ride the intuitive/habitual moral judgement,
113 there is considerable evidence suggesting that people are unwilling to over-ride habitual
114 responses, or revise their intuitive judgements, on certain issues, and this is particularly true
115 for moral issues (Abelson, 1988; Kruglanski & Webster, 1996; Kruglanski, Webster, & Klem,
116 1993; McGregor, 2006a, 2006b; McGregor, Zanna, Holmes, & Spencer, 2001). This
117 robustness of the habitual/intuitive response means that resolving conflict requires either the
118 identification of reasons that are consistent with the initial judgement, or ignoring the
119 inconsistent information. If further reasons cannot be identified, and inconsistencies cannot
120 be resolved (e.g., because of direct questioning) a person may maintain their judgement
121 while also acknowledging that they cannot provide supporting reasons for it (i.e., presenting
122 as dumbfounded).
123 Two types of dumbfounded responses were identified by McHugh et al. (2017): (a)
124 unsupported declarations and (b) admissions of not having reasons. The salience of
125 inconsistent information is much greater when a person admits to not having reasons than
126 when they simply provide an unsupported declaration. As such, an unsupported declaration
127 is likely a much more attractive response than admitting to not having reasons for a
128 judgement. Indeed, McHugh et al. (2017) provide some suggestive evidence for this: when
129 dumbfounding was measured as the selecting of an unsupported declaration, rates of
130 dumbfounding were much higher than in other studies. In the absence of explicit reminders
131 that deliberation did not provide reasons to support their judgement, participants appeared
132 to readily dismiss inconsistencies between their initial judgement and the outcome of
133 deliberation.
134 Intuitive and Deliberative Responding
135 In traditional dual-process conflict studies there are generally two responses: a correct
136 response and an incorrect response; and two types of responses: logical (deliberative) and
137 intuitive (habitual). This binary correct/incorrect, logical/intuitive classification means that MANIPULATING MORAL DUMBFOUNDING 7
138 the dependent variables in these studies can be operationalised relatively easily, a response is
139 either correct or incorrect. The correct response always maps onto the logical (deliberative)
140 and various manipulations can frame the intuitive response as either correct or incorrect.
141 In contrast, there is no clear correct or incorrect answer in moral judgement tasks. In
142 addition to this, the study of moral dumbfounding (in the common case of an initial
143 judgement that the target behaviour is wrong) involves at least three different responses (1)
144 the providing of reasons (reason); (2) accepting the counter-arguments and rating the
145 behaviour as “not wrong” (nothing wrong); or (3) a dumbfounded response (dumbfounding).
146 The positioning of these responses options in terms of habitual or deliberative responses may
147 vary from person to person or from situation to situation.
148 The nature of the dumbfounding paradigm, which involves explicit arguments against
149 the most common judgements, requires that participants engage in some deliberation. This
150 means that all responses are likely the result of some level of deliberation in order to support
151 (or attempt to support) an initial intuition. A simplistic view of the possible responses
152 identified above would identify (1) reason as the successful alignment of deliberation and
153 intuition. (2) Nothing wrong is the over-riding of intuition by deliberation, while (3)
154 dumbfounding is the failure of deliberation to rationalise an intuition. This view however
155 presumes that there is only one intuition at play, and participants’ deliberations are an
156 attempt to provide reasons for this single intuition.
157 At least one other intuition that may become salient as part of the dumbfounding
158 paradigm has been identified. Participants may have intuitions relating to the nature of
159 moral knowledge, for example that moral judgements should be justifiable by reasons. During
160 the course of the study, this intuition (or alternative intuitions) may become salient. The
161 emergence of this intuition appears consistent with what has been observed in studies of
162 moral dumbfounding, whereby people appear to be motivated to identify reasons for their
163 judgement (McHugh et al., 2017; Royzman et al., 2015). This means that people experience MANIPULATING MORAL DUMBFOUNDING 8
164 more than a conflict between habitual and deliberative responding, they may also experience
165 competing intuitions. The emergence of these competing intuitions may be attributed to
166 deliberation, however, it is also possible that their emergence occurs as a result of the nature
167 of the dumbfounding paradigm independently of level of deliberation; participants are asked
168 to provide reasons for their judgements. The need for judgements to be justifiable by reasons
169 may be made salient by simply asking participants to provide justifications for their
170 judgements.
171 Three possible types of responses in the dumbfounding paradigm are listed above, (1)
172 reason; (2) nothing wrong; and (3) dumbfounding. In addition, two classes of (3)
173 dumbfounding have been identified: admissions of not having reasons, and unsupported
174 declaration (e.g., “It’s just wrong!” or “Incest is wrong.”). To reflect this, response (3)
175 dumbfounding may be described as (3a) admission and (3b) unsupported declaration . The
176 discussion below attempts to describe each of the responses 1-3 in terms of the varying roles
177 of deliberation and intuition, where at least two relevant (competing) intuitions have been
178 identified.
179 It is likely that (1) a reason response is the most desirable, it involves successfully
180 resolving the conflicting intuitions and providing reasons for a judgement. However, a reason
181 response also requires that deliberation is “successful”, that is, deliberation results in
182 successfully identifying a reason for a judgement. In the dumbfounding paradigm this is
183 made particularly difficult because commonly identified reasons for judgements are refuted
184 during the course of the study. This means that successfully identifying a reason requires
185 identifying a reason beyond the refuted reasons, or identifying shortcomings in the
186 refutations.
187 It is also likely that (3b) unsupported declaration requires the least deliberation.
188 Defending a judgement with a restatement of the judgement can be done without
189 deliberation. Furthermore, in restating a judgement and affirming a position, the salience of MANIPULATING MORAL DUMBFOUNDING 9
190 inconsistent intuitions may be reduced. This response may be likened to the “seizing and
191 freezing” behaviours described by Kruglanski and Webster (1996).
192 This leaves (2) nothing wrong and (3a) admission. It is probable that changing a
193 judgement requires more deliberation than admitting to not having a reason for it; in
194 changing a judgement a person would likely deliberate the strength of the counter-arguments.
195 It has also been shown that both types of dumbfounded response are more similar to each
1 196 other than to either (1) a reason response or (2) nothing wrong. It is likely that the
197 observed similarity between both dumbfounded responses would extend to levels of
198 deliberation. As such, dumbfounded responding is hypothesised to involve the least amount
199 of deliberation, and providing reasons requires the most amount of deliberation, with
200 changing judgement involving more deliberation than dumbfounded responding, but not as
201 much deliberation as successfully providing reasons.
202 Competing Intuitions. The only response that can be positioned as habitual or
203 deliberative with any confidence is (1) a reason response, which is certainly deliberative, as it
204 illustrates successful deliberation such that two competing intuitions may be aligned. The
205 relative roles of intuition and deliberation in each of the other responses are less clear. There
206 are at least two competing intuitions that may give rise to the remaining responses: (i) that
207 the behaviour is wrong, and (ii) that moral judgements should be grounded in reasons.
208 Furthermore, (3b) unsupported declaration is a clear endorsing of the first intuition (i)
209 that the behaviour is wrong over the second intuition (ii) that moral judgements should be
210 grounded in reasons. Arriving at this response is possible without deliberation. If a person
211 does not see a need for judgements to be justified by reasons, they are unlikely to engage in
212 a deliberative search for reasons.
213 The remaining responses (2) nothing wrong and (3a) admission, may be viewed as
1 This is based on the analysis of non-verbal behaviours such as laughing, smiling, silence; see McHugh et al. (2017) MANIPULATING MORAL DUMBFOUNDING 10
214 instances of selecting one of the competing intuitions following deliberation. Providing
215 response (2) nothing wrong, may also be viewed as selecting one of the competing intuitions
216 based on, and informed by deliberation, acknowledging the value of this deliberation process.
217 For this reason it is possible that response (2) nothing wrong, may, in some cases, involve
218 slightly more deliberation than (3a) an admission. The relative level of deliberation involved
219 in (2) nothing wrong and (3a) admission, may vary depending on the individual. Some
220 people may readily change their judgement based on new information, while others may not
221 see the need to justify their judgement by reasons.
222 In view of the above discussion, the responses may be (tentatively) ranked in order of
223 the relative role of deliberation. Beginning with the highest level of deliberation and ending
224 with the lowest, the responses may be ranked as follows: (i) providing reasons (successful
225 deliberation), (ii) accepting the counter-arguments and rating the behaviour as “not wrong”
226 (failed deliberation/deliberation over-riding an initial intuition), (iii) an admission of not
227 having reasons (failed deliberation/rejection of value of deliberation) and, (iv) an
228 unsupported declaration (failed deliberation/rejection of value of deliberation/deliberation
229 absent). As noted previously, providing reasons is the only response for which claims
230 regarding the relative role of deliberation and intuition can be made with any degree of
231 certainty.
232 Influences on Moral Dumbfounding
233 One prediction of explaining dumbfounding as conflict in dual-processes is that under
234 specific manipulations, responses in the moral dumbfounding paradigm should vary in
235 predictable ways. In addition to external manipulation, a second prediction of this conflict in
236 dual-processes explanation is that responses in the moral dumbfounding paradigm may
237 display variability that can be linked to specific individual difference variables. The studies
238 described here aim to investigate both of these possibilities. MANIPULATING MORAL DUMBFOUNDING 11
239 Influencing Moral Dumbfounding through Experimental Manipulation
240 Cognitive load has been shown to inhibit deliberative responding (e.g., De Neys, 2006;
241 Evans & Curtis-Holmes, 2005; Evans & Stanovich, 2013; Schmidt, 2016). Above, we
242 identified providing reasons as involving more deliberation than alternative responses in the
243 dumbfounding paradigm. This implies that cognitive load should inhibit the identification of
244 reasons for a judgement, leading to an increase in dumbfounded responding or an increase in
245 accepting the counter-arguments and revising the judgement made.
246 According to Greene et al. (2008), cognitive load reduces the likelihood of utilitarian
247 judgements but not deontological judgements. The judgements in dumbfounding paradigms
248 are typically deontological; the moral violations described in the scenarios in studies of moral
249 dumbfounding are violations of widely accepted deontic propositions. Changing a judgement
250 in the dumbfounding paradigm means rejecting a deontological judgement in favour of a
251 judgement informed by a utilitarian position (the counter-arguments highlight the lack of
252 harm in the scenarios). Drawing on Greene et al. (2008), who showed that utilitarian
253 judgements are negatively influenced by cognitive load, it is expected that cognitive load
254 should led to more dumbfounded responding, rather than changing of judgements. This
255 prediction is also purported to be supported by Haidt et al. (2000). In the opening note of
256 the original Haidt et al. (2000) report, they report that they conducted a second study in
257 which they manipulated cognitive load. They report that they found that cognitive load led
258 to increased levels of dumbfounding but did not influence judgements made. Beyond a brief
259 mention in the opening note, this cognitive load and moral dumbfounding study is not
260 reported in full in Haidt et al. (2000) or elsewhere.
261 An investigation of the dumbfounding under cognitive load can test two predictions of
262 dual-process models of moral judgement. Firstly, it is hypothesised that deliberative
263 responding generally will be inhibited by cognitive load, leading to less identification of
264 reasons for judgements. This inhibition may result in higher rates of selecting “there is MANIPULATING MORAL DUMBFOUNDING 12
265 nothing wrong” or higher rates of dumbfounded responding (or both). However, adopting
266 the work by Greene et al. (2008) suggests that the inhibition of deliberation should result in
267 higher rates of dumbfounded responding only.
268 Again, there are three possible responses in the dumbfounding paradigm: (1) providing
269 reasons; (2) a change in judgement; and (3) a dumbfounded response. Providing reasons has
270 been identified as requiring more deliberation than the other responses, and as such the
271 introduction of a cognitive load manipulation should reduce the providing of reasons in
272 favour of one of the other responses. It is not clear whether participants would be more
273 likely to revise their judgement or provide a dumbfounded response.
274 Individual Differences in Moral Dumbfounding
275 As noted above, it is likely that responses in the dumbfounding paradigm will vary
276 depending on individual differences. One individual difference variable linked to dual-process
277 approaches to cognition, therefore may be related to susceptibility to dumbfounding is Need
278 for Cognition (Cacioppo & Petty, 1982; Forsterlee & Ho, 1999; Petty, Cacioppo, & Kao,
279 1984; Petty, Feinstein, Blair, & Jarvis, 1996). The Need for Cognition Scale (NFC) is a
280 measure of an individual’s tendency “to engage in and enjoy effortful analytic activity”
281 (Forsterlee & Ho, 1999, p. 471; see also Cacioppo & Petty, 1982). In other words, it measures
282 a tendency to engage in deliberation (Evans & Stanovich, 2013). It is also related to a
283 person’s need to understand and make sense of the world (Forsterlee & Ho, 1999). It is
284 hypothesised that people who score highly on the NFC will be more likely to provide reasons
285 for their judgement. Related to this, people who score low on the NFC are likely to fail to
286 identify reasons for their judgement. That NFC is related to a need to understand and make
287 sense of the world suggests that of the people who fail to identify reasons for their
288 judgement, the people who revise their judgement will likely score higher on the NFC than
289 people who provide a dumbfounded response. MANIPULATING MORAL DUMBFOUNDING 13
290 The Current Studies
291 The studies described here aim to investigate test three predictions of a conflict in
292 dual-process explanation of moral dumbfounding. To test the first two predictions we
293 experimentally manipulated cognitive load. Our primary prediction is that a cognitive load
294 manipulation will inhibit people’s ability to provide reasons for their judgement, leading to
295 greater habitual responses (either nothing wrong or dumbfounding or both). A second
296 prediction, informed by Greene et al. (2008), that a cognitive load manipulation will inhibit
297 people’s ability to provide reasons for their judgement, leading to greater dumbfounding
298 specifically will also be tested. The third prediction that will be tested is that a person’s
299 tendency to provide reasons will be related to their score on the Need for Cognition scale
300 (Cacioppo & Petty, 1982; Petty et al., 1984). We hypothesise that as NFC increases people
301 will be less likely to present as dumbfounded. We conducted five studies to assess these
2,3 302 predictions. In all five studies, we report all measures, manipulations and exclusions.
303 Study 1 - College Sample
304 The aim of Study 1 was to investigate if a cognitive load manipulation influenced
305 participants’ ability to justify their judgement. We also measured Need for Cognition
2 A priori power analysis indicated that, for the primary research question (the influence of cognitive load on dumbfounded responding), in order to detect a large effect size (V = .35) with 80% power, a sample of N = 79 participants was required; in order to detect a medium effect size (V = .21) with 80% power a sample of N = 218 participants was required; in order to detect a small effect size (V = .07) with 80% power a sample of N = 1966 was required. This means that the studies described here are likely underpowered. To account for this we conclude with a combined analysis of all studies which includes mini meta-analyses.
3 A priori power analysis indicated that, for the secondary research question (the relationship between dumbfounded responding and Need for Cognition), in order to detect a large effect size (OR = 9) with 80% power, a sample of N = 11 participants was required; in order to detect a medium effect size (OR = 3.45) with 80% power a sample of N = 32 participants was required; in order to detect a small effect size (OR = 1.49) with 80% power a sample of N = 309 was required. Again, the studies described here are likely underpowered and we conclude with a combined analysis of all studies. MANIPULATING MORAL DUMBFOUNDING 14
306 (Cacioppo & Petty, 1982; Petty et al., 1984) as a potential moderator variable.
307 Study 1: Method
308 Participants and design. Study 1 was a between subjects design. The dependent
309 variable was rates of providing reasons/dumbfounding (measured using to the critical slide
310 with 3 response options: 1: providing reasons; 2: there is nothing wrong; 3: dumbfounded
311 response - admission). The independent variable was cognitive load with two levels: present
312 and absent. Cognitive load was manipulated by presenting participants with an eight digit
313 number letter string to be memorised. Need for Cognition (Cacioppo & Petty, 1982; Petty et
314 al., 1984) was included as a potential correlate and moderator variable.
315 A total sample of 66 participants (55 female, 11 male; M age = 22.42, min = 18, max =
316 57, SD = 6.86) took part. Participants in this sample were undergraduate students,
317 postgraduate students, and alumni from Mary Immaculate College (MIC), and University of
318 Limerick (UL). Participation was voluntary and participants were not reimbursed for their
319 participation.
320 Procedure and materials. Data were collected using an online questionnaire.
321 Data collection took place in a designated computer laboratory in MIC. The experimenter
322 remained in the laboratory for the duration of the study. Participants were first presented
323 with an information sheet and consent form. The main study proceeded when participants
324 had signed the consent form.
325 Participants in the experimental condition were presented with an eight digit
326 number/letter string and asked to memorise the sequence. After 30 seconds, the experiment
327 progressed to the next slide. Participants had the option to click “ok” and progress to the
328 next slide after 15 seconds.
329 Participants were then presented with the “Julie and Mark” (Incest) vignette (Haidt et
330 al., 2000). Participants rated on a 7-point Likert scale how right or wrong the behaviour of MANIPULATING MORAL DUMBFOUNDING 15
331 Julie and Mark was (where, 1 = Morally wrong; 4 = neutral; 7 = Morally right), and were
332 given an opportunity to provide reasons for their judgement. Following this, participants
333 were presented with a series of counter-arguments, which refuted commonly used
334 justifications for rating the behaviour as “wrong”.
335 Dumbfounding was measured using the critical slide (McHugh et al., 2017). This
336 contained a statement defending the behaviour and a question as to how the behaviour could
337 be wrong (“Julie and Mark’s behaviour did not harm anyone, how can there be anything
338 wrong with what they did?”). There were three possible answer options: (a) “There is
339 nothing wrong”; (b) an admission of not having reasons (“It’s wrong but I can’t think of a
340 reason”); and finally a judgement with accompanying justification (c) “It’s wrong and I can
341 provide a valid reason”. The order of these response options was randomised. Participants
342 who selected (c) were prompted to type a reason. The selecting of option (b), the admission
343 of not having reasons, was taken to be a dumbfounded response. Following the critical slide,
344 participants in the experimental condition were required to reproduce the eight digit
345 number-letter string sequence presented previously. Following this a post-discussion
346 questionnaire in which participants rated their response to the scenario across various
347 dimensions (Haidt et al., 2000).
348 Need for Cognition was measured using the short form of the Need for Cognition scale
349 (Cacioppo & Petty, 1982; Petty et al., 1984). This is an 18 item scale containing questions
350 relating to motivation to engage in thinking (e.g., “I would prefer complex to simple
351 problems”). Responses were recorded on a -4 to +4 Likert-type scale, where -4 = very strong
352 disagreement and +4 = very strong agreement.
353 Study 1: Results
354 Forty six participants (69.70%) rated the behaviour of Julie and Mark as wrong initially.
355 The mean initial rating of the behaviour was, M = 2.38, SD = 1.87. There was no difference MANIPULATING MORAL DUMBFOUNDING 16
356 in initial judgement depending on cognitive load, t(63.90) = 1.26, p = .214; d = 0.31.
100%
75% 63.6%
Condition 50% 45.5% Cognitive load Control 36.4%
25% 21.2% 18.2% 15.2%
Percentage of participants selecting each response 0% 15 5 6 7 12 21
Nothing Wrong Dumbfounded Reasons Response to Critical Slide
Figure 1 . Study 1: Responses to critical slide and for the experimental group (N = 33) and the control group (N = 33)
357 Participants who selected the admission of not having reasons on the critical slide were
358 identified as dumbfounded. Thirteen participants (19.70%) selected “It’s wrong but I can’t
359 think of a reason”. Thirty three participants (50%) selected “It’s wrong and I can provide a
360 valid reason”; and twenty participants (30.30%) selected “There is nothing wrong”.
361 Cognitive load manipulation check. Analysis of the responses to the cognitive
362 load manipulation revealed that nine participants (27.27%) successfully remembered the
363 sequence of numbers and letters in full. Responses to the manipulation check question
364 revealed that five participants (15.15%) found the memory task easy. Of these, four
365 participants both found the task easy and got the answer right. All participants correctly
366 remembered at least two digits, indicating at least some level of engagement with the MANIPULATING MORAL DUMBFOUNDING 17
Table 1 Study 1 – Observed counts, expected counts, and standardised residuals for each response to the critical slide depending on cognitive load
Cognitive Load Control
Observed count Nothing Wrong 15 5 Dumbfounded 6 7 Reasons 12 21 Expected count Nothing Wrong 10 10 Dumbfounded 6.5 6.5 Reasons 16.5 16.5 Standardised residuals Nothing Wrong 2.68* -2.68* Dumbfounded -0.31 0.31 Reasons -2.22* 2.22*
Note. * = sig. at p < .05; ** = sig. at p < .001
367 cognitive load manipulation.
368 We tested for differences between the cognitive load group and the control group for
369 Initial Judgement and Initial Confidence. An independent samples t-test revealed no
370 significant difference in initial rating in the cognitive load group, (M = 2.67, SD = 1.90),
371 and the control group, (M = 2.55, SD = 1.92), t(63.90) = 1.26 , p = 0.21; d = 0.31. An
372 independent samples t-test revealed no significant difference in initial confidence in the
373 cognitive load group, (M = 5.42, SD = 1.30), and the control group, (M = 5.09, SD = 1.83),
374 t(57.80) = 0.85 , p = 0.40; d = 0.21. In view of this, we assume that both groups engaged
375 equally with the task.
376 Cognitive load and dumbfounded responding. A chi-squared test for
377 independence revealed a significant association between experimental condition and response MANIPULATING MORAL DUMBFOUNDING 18
2 378 to the critical slide, χ (2, N = 66) = 7.53, p = .023, V = 0.34, the observed power was 0.69.
379 Under cognitive load fewer participants (12; 63.64%) provided reasons than in the control
380 condition (21; 36.36%). Similarly, under cognitive load more participants (15; 45.45%)
381 selected “There is nothing wrong” than in the control group (5; 15.15%). The responses to
382 the critical slide for the experimental group (N = 33) and the control group (N = 33) are
383 displayed in Figure 1. The observed counts, expected counts and standardised residuals are
384 displayed in Table 1.
385 Need for Cognition. A multinomial logistic regression revealed no significant
2 386 association between Need for Cognition and response to the critical slide, χ (2, N = 66) =
4 387 4.86, p = .088, the observed power was 0.49.
388 Study 1: Discussion
389 The aim of the Study 1 was to investigate if dumbfounded responding was influenced
390 by cognitive load. Specifically, adopting a dual-systems model of moral judgement, it was
391 hypothesised that cognitive load would lead to reduced levels of deliberative responding, and
392 that a reduction in rates of successfully identifying reasons for judgements would be
393 observed. This may lead to (a) increased levels of dumbfounding, or (b) increased selecting
394 of the “nothing wrong” response. As predicted we found lower rates of providing reasons in
395 the cognitive load group than in the control group. This led to higher rates of selecting
396 “there is nothing wrong” in the cognitive load group.
397 As discussed above, we hypothesised that moral dumbfounding occurs when people fail
398 to resolve a conflict between an initial intuition and an intuition that emerges following
399 deliberation. Under cognitive load a higher proportion of participants failed to provide
400 reasons for their judgements, and a higher proportion of participants selected the “There’s
4 As noted previously the studies presented here are likely underpowered. We conclude with a combined analysis. The relative probabilities of selecting each response depending on Need for Cognition for each study are presented in Appendix D. See Figure 11 for Study 1. MANIPULATING MORAL DUMBFOUNDING 19
401 nothing wrong” response. There was no difference in rates of dumbfounding between the
402 cognitive load group and the control group. It appears that when faced with an inability to
403 justify a judgement with reasons, it is preferable to revise the judgement than to
404 acknowledge the inconsistencies in the form of a dumbfounded response.
405 Study 2 - Online Replication
406 Study 1 demonstrated interesting variability in responses to the critical slide depending
407 on cognitive load. The aim of Study 2 was to assess the replicability of the results of Study 1,
408 using an online sample. In Study 1, the experimenter was in the room with the participants.
409 This made it more difficult for participants to cheat on the memory task. This is not possible
410 with an online sample. An alternative cognitive load manipulation was taken from De Neys
411 and Schaeken (De Neys & Schaeken, 2007), whereby a dot pattern is briefly presented to
412 participants, and participants are required to reproduce the dot pattern at a later stage.
413 Study 2: Method
414 Participants and design. Study 2 was a between subjects design. The dependent
415 variable was response to the critical slide. The independent variable was cognitive load with
416 two levels: high and low. Need for Cognition (Cacioppo & Petty, 1982; Petty et al., 1984)
417 was included as a potential correlate and moderator variable.
418 A total sample of 100 participants (56 female, 44 male; M age = 38.38, min = 19, max =
419 72, SD = 12.41) took part. Participants in this sample were recruited using Amazon’s MTurk
420 (Amazon Web Services Inc., 2016). Participants were paid $0.50 for their participation.
421 Participants were recruited from English speaking countries or from countries where
422 residents generally have a high level of English (e.g., The Netherlands, Denmark, Sweden).
423 Procedure and materials. Data were collected using an online questionnaire.
424 Materials were largely the same as in Study 1, with a change to the cognitive load
425 manipulation. Cognitive load was manipulated using a dot-pattern memory task (De Neys & MANIPULATING MORAL DUMBFOUNDING 20
426 Schaeken, 2007).
427 Participants were presented with a 3 x 3 grid containing a dot pattern. This image
428 disappeared after one second. Participants then answered a question relating to the moral
429 judgement task. Following this, participants were asked to reproduce the dot-pattern. All
430 participants took part in the memory task, and cognitive load was manipulated by varying
431 the complexity of the patterns presented (De Neys & Schaeken, 2007). The control group
432 were presented with simple patterns, containing three dots in a line, while the experimental
433 group were presented with more complex dot patterns containing 4 dots, see Figure 2.
a b
Figure 2 . Sample dot patterns - more simple for the control group (a) and higher complexity for the experimental condition (b)
434 Study 2 proceeded in much the same way as Study 1. There were four target questions
435 during which participants were engaged in the memory task. A different pattern was
436 presented before each of the following: the initial judgement, the initial opportunity to
437 provide reasons, the critical slide, and the revised judgement. After each of these questions
438 participants were required to reproduce the pattern. As in Study 1, dumbfounding was
439 measured using the critical slide.
440 Study 2: Results
441 Seventy seven participants (77%) rated the behaviour of Julie and Mark as wrong
442 initially. The mean initial rating of the behaviour was, M = 2.13, SD = 1.54. There was no MANIPULATING MORAL DUMBFOUNDING 21
443 difference in initial judgement depending on cognitive load, t(97.77) = 0.96, p = .339; d =
444 0.19.
445 Participants who selected the admission of not having reasons on the critical slide were
446 identified as dumbfounded. Twenty six participants (26%) selected “It’s wrong but I can’t
447 think of a reason”. Fifty participants (50%) selected “It’s wrong and I can provide a valid
448 reason”; and twenty four participants (24%) selected “There is nothing wrong”.
449 Cognitive load and dumbfounded responding. A chi-squared test for
450 independence revealed no association between experimental condition and response to the
2 451 critical slide, χ (2, N = 100) = 0.74, p = .690, V = 0.09, the observed power was 0.11. The
452 responses to the critical slide for the experimental group (N = 51) and the control group (N
453 = 49) are displayed in Figure 3.
454 Engagement with the Memory Task. It is possible that the difference in results
455 observed between Study 1 and Study 2 is due to the alternative manipulation of cognitive
456 load employed. In Study 1, the control group did not engage in any task, however, adopting
457 De Neys and Shaeken’s procedure (De Neys & Schaeken, 2007), participants in the control
458 group of Study 2 engaged in a memory task. It is possible that simply engaging in a memory
459 task led to differences in responses, and that level of difficulty (the manipulation that was
460 employed) was irrelevant. Indeed, the responding to the critical slide in the control group in
461 Study 2 is more similar to the responding in the experimental group in 1 than to the control
462 group in Study 1.
463 Rates of successful reproduction of the dot patterns in Study 2 were much lower than
464 reported by De Neys and Shaeken (De Neys & Schaeken, 2007). It appears that participants
465 in Study 2 did not engage with the memory task in the same way as participants in De Neys
466 and Shaeken’s studies (De Neys & Schaeken, 2007). We hypothesised that the effectiveness
467 of the cognitive load manipulation may be moderated by the degree to which people engage
468 with the manipulation. MANIPULATING MORAL DUMBFOUNDING 22
100%
75%
Condition 49.0% 51.0% 50% Cognitive load
Control
29.4% 26.5% 25% 21.6% 22.4%
Percentage of participants selecting each response 0% 11 13 15 11 25 25
Nothing Wrong Dumbfounded Reasons Response to Critical Slide
Figure 3 . Study 2: Responses to critical slide for the experimental group (N = 51) and the control group (N = 49)
469 The memory task involved correctly placing dots in a 3 x 3 grid. For the scoring of this
470 task, each of the nine places in the grid could be marked/not marked correctly or incorrectly,
471 making 9 the total possible number of correct responses. If a person misplaced one dot in
472 the pattern this would count for 2 incorrect places in the grid: the mark in the incorrect
473 place, and the absence of a mark in the place it should have been. A participant who
474 received a score of 7, could reasonably be taken to have engaged with the task, and simply
475 made a slip. As such, this was taken as the cut-off point for identifying engagement. This
476 resulted in 56 participants being identified as engaging with the memory task, and 44 being
477 identified as not engaging with the task.
478 Responses to critical slide for participants who engaged with the memory task and MANIPULATING MORAL DUMBFOUNDING 23
100%
75%
61.4% Condition
50% Engaged
41.1% Not Engaged 35.7%
25.0% 25% 23.2% 13.6%
Percentage of participants selecting each response 0% 13 11 20 6 23 27
Nothing Wrong Dumbfounded Reasons Response to Critical Slide
Figure 4 . Study 2: Responses to critical slide depending on engagement (N = 56) or non-engagement (N = 44) with the memory task
479 participants who did not engage with the memory task were analysed separately (see
480 Figure 4). A chi-squared test for independence revealed an association between engagement
2 481 in the memory task and response to the critical slide, χ (2, N = 100) = 6.68, p = .035, V =
482 0.26, the observed power was 0.11. The responses to the critical slide for the experimental
483 group (N = 51) and the control group (N = 49) are displayed in Figure 4. The observed
484 counts, expected counts and standardised residuals are displayed in Table 2.
485 Need for Cognition. A multinomial logistic regression revealed no significant
2 486 association between Need for Cognition and response to the critical slide, χ (2, N = 100) =
5 487 2.19, p = .334, the observed power was 0.24.
5 See Appendix D Figure 12 for relative probabilities of selecting each response depending on Need for Cognition. MANIPULATING MORAL DUMBFOUNDING 24
Table 2 Study 2 – Observed counts, expected counts, and standardised residuals for each response to the critical slide depending on cognitive load
Engaged Not Engaged
Observed count Nothing Wrong 13 11 Dumbfounded 20 6 Reasons 23 27 Expected count Nothing Wrong 13.44 10.56 Dumbfounded 14.56 11.44 Reasons 28 22 Standardised residuals Nothing Wrong -0.21 0.21 Dumbfounded 2.5* -2.5* Reasons -2.01* 2.01*
Note. * = sig. at p < .05; ** = sig. at p < .001
488 Study 2: Discussion
489 The aim of Study 2 was to replicate Study 1. As a replication it failed. However,
490 interesting variability was observed when engagement with the memory task was accounted
491 for. As predicted, under cognitive load (engagement with the memory task) rates of
492 providing reasons were reduced. In contrast to Study 1, the reduced rates of providing
493 reasons led to higher rates of dumbfounded responding (rather than increased rates of
494 selecting “There is nothing wrong”).
495 Two problems with the cognitive load manipulation were identified. Firstly, the control
496 did not serve as an appropriate control. Secondly, (and unsurprisingly) the manipulation is
497 only effective if participants engage with the memory task. A follow-up study addresses each
498 of these, using no task as the control for the cognitive load manipulation (e.g., Drolet & MANIPULATING MORAL DUMBFOUNDING 25
499 Frances Luce, 2004), and assessing the responses to the memory task to ensure participants
500 adequately engaged with the load manipulation.
501 Study 3 - Revised Online Replication
502 In Study 2 the role of engagement with the memory task emerged as an important
503 moderator of the effectiveness of the cognitive load manipulation. Study 3 was conducted in
504 order to test if cognitive load affects participants’ ability to identify reasons for their
505 judgements, when accounting for engagement with the memory task.
506 Study 3: Methods
507 Participants and Design. Study 3 was a between subjects design. The dependent
508 variable was response to the critical slide. The independent variable was cognitive load with
509 two levels: present and absent. Need for Cognition (Cacioppo & Petty, 1982; Petty et al.,
510 1984) was included as a potential correlate and moderator variable.
511 Following the elimination of 34 participants who scored less than 7 on the memory task
512 we were left with a final sample of 129 participants (74 female, 55 male; M age = 40.26, min
513 = 20, max = 72, SD = 13.04). Participants in this sample were recruited through MTurk
514 (under the same conditions as Study 2).
515 Procedure and Materials. Study 3 was the same as Study 2 with two changes.
516 The control group did not take part in a memory task, and to avoid task fatigue, in the dot
517 patterns presented to the experimental group, the dot patterns presented alternated between
518 the easy 3-dot patterns and the complex 4-dot patterns.
519 A score of 7 or higher on the memory task that accompanied the critical slide was
520 selected as the measure of engagement with the memory task. Only participants who
521 engaged with the task were eligible for analysis. Other than the two changes described above,
522 Study 3 was the same as Study 2. MANIPULATING MORAL DUMBFOUNDING 26
523 Study 3: Results
100%
75% 67.2%
52.9% Condition 50% Cognitive load
Control
24.6% 25.0% 25% 22.1%
8.2%
Percentage of participants selecting each response 0% 15 15 17 5 36 41
Nothing Wrong Dumbfounded Reasons Response to Critical Slide
Figure 5 . Study 3: Responses to critical slide for the cognitive load group (N = 68) and the control group (N = 61)
524 Ninety five participants (73.64%) rated the behaviour of Julie and Mark as wrong
525 initially. The mean initial rating of the behaviour was, M = 2.27, SD = 1.75. There was no
526 difference in initial judgement depending on cognitive load, t(122.69) = 0.15, p = .877; d =
527 0.03.
528 Participants who selected the admission of not having reasons on the critical slide were
529 identified as dumbfounded. Twenty two participants (17.05%) selected “It’s wrong but I
530 can’t think of a reason”. Seventy seven participants (59.69%) selected “It’s wrong and I can
531 provide a valid reason”; and thirty participants (23.26%) selected “There is nothing wrong”. MANIPULATING MORAL DUMBFOUNDING 27
Table 3 Study 3 – Observed counts, expected counts, and standardised residuals for each response to the critical slide depending on cognitive load
Cognitive Load Control
Observed count Nothing Wrong 15 15 Dumbfounded 17 5 Reasons 36 41 Expected count Nothing Wrong 15.81 14.19 Dumbfounded 11.6 10.4 Reasons 40.59 36.41 Standardised residuals Nothing Wrong -0.34 0.34 Dumbfounded 2.53* -2.53* Reasons -1.65 1.65
Note. * = sig. at p < .05; ** = sig. at p < .001
532 Cognitive load and dumbfounded responding. A chi-squared test for
533 independence revealed a significant association between experimental condition and response
2 534 to the critical slide, χ (2, N = 129) = 6.51, p = .039, V = 0.22, the observed power was
535 0.62. The responses to the critical slide for the experimental group (N = 68) and the control
536 group (N = 61) are displayed in Figure 5. The observed counts, expected counts and
537 standardised residuals are displayed in Table 3.
538 Need for Cognition. A multinomial logistic regression revealed no significant
2 539 association between Need for Cognition and response to the critical slide, χ (2, N = 129) =
540 6.43, p = .040, the observed power was 0.62. Need for Cognition explained between 4.64%
541 (Cox and Snell R square) and 7.70% (Nadelkerke R squared) of the variance in responses to
542 the critical slide. As Need for Cognition increased, participants were significantly more likely
543 to provide reasons than to present as dumbfounded, Wald = 6.08, p = .014, odds ratio = MANIPULATING MORAL DUMBFOUNDING 28
6 544 1.46, 95% CI [1.08, 1.97].
545 Study 3: Discussion
546 Study 3 partially demonstrated the predicted relationship between engagement with a
547 cognitive load task and response to the critical slide. Rates of dumbfounded responding were
548 significantly higher under cognitive load. Surprisingly, rates of providing reasons were not
549 significantly reduced under cognitive load. Study 3 did not include an objective
550 manipulation check. As such, we conducted a follow up study that included an objective
551 manipulation check.
552 Study 3 also provided some evidence that responses to the critical slide may be linked
553 with Need for Cognition. As predicted, participants who scored higher in Need for Cognition
554 were more likely to provide reasons than present as dumbfounded.
555 Study 4 - Online Replication with Manipulation Check
556 Study 3 found a significant relationship between cognitive load and response to the
557 critical slide and a significant relationship between Need for Cognition and response to the
558 critical slide. The aim of Study 4 was to replicate these findings. In addition Study 4
559 included a manipulation check to assess the effectiveness of the cognitive load manipulation
560 employed.
561 Study 4: Method
562 Participants and Design. Study 4 was a between subjects design. The dependent
563 variable was response to the critical slide. The independent variable was cognitive load with
564 two levels: present and absent. Need for Cognition (Cacioppo & Petty, 1982; Petty et al.,
565 1984) was included as a potential correlate and moderator variable.
6 See Appendix D Figure 13 for relative probabilities of selecting each response depending on Need for Cognition. MANIPULATING MORAL DUMBFOUNDING 29
566 Following the elimination of 29 participants who scored less than 7 on the memory task
567 we were left with a final sample of 127 participants (84 female, 43 male; M age = 41.19, min
568 = 21, max = 74, SD = 13.91). Participants in this sample were recruited through MTurk
569 (under the same conditions as Studies 2 and 3).
570 Procedure and Materials. Study 4 was the same as Study 3 with one change, the
571 inclusion of a manipulation check. A prose paragraph was included after participants made
572 their revised judgements. Participants were then asked three comprehension questions
573 relating to the prose paragraph. It was expected that participants in the control group would
574 perform better at this task than participants under cognitive load (Just & Carpenter, 1992).
575 Study 4: Results
100%
75% 65.1%
54.7% Condition
50% Cognitive load
Control
29.7%
25% 19.0% 15.6% 15.9%
Percentage of participants selecting each response 0% 19 12 10 10 35 41
Nothing Wrong Dumbfounded Reasons Response to Critical Slide
Figure 6 . Study 4: Responses to critical slide for the cognitive load group (N = 64) and the control group (N = 61) MANIPULATING MORAL DUMBFOUNDING 30
576 Ninety eight participants (77.17%) rated the behaviour of Julie and Mark as wrong
577 initially. The mean initial rating of the behaviour was, M = 2.09, SD = 1.62. There was no
578 difference in initial judgement depending on cognitive load, t(124.70) = 0.54, p = .589; d =
579 0.10.
580 Participants who selected the admission of not having reasons on the critical slide were
581 identified as dumbfounded. Twenty participants (15.75%) selected “It’s wrong but I can’t
582 think of a reason”. Seventy six participants (59.84%) selected “It’s wrong and I can provide a
583 valid reason”; and thirty one participants (24.41%) selected “There is nothing wrong”.
584 The responses to the manipulation check questions were investigated. There was no
585 difference in the number of correct answers to these questions between the cognitive load
586 group and the control group t(123.91) = 0.57, p = .569; d = 0.10. There was also no
587 difference in time taken to read the vignette between the groups t(63.40) = 1.62, p = .111; d
588 = 0.28.
589 Cognitive load and dumbfounded responding. A chi-squared test for
590 independence revealed no significant association between experimental condition and
2 591 response to the critical slide, χ (2, N = 127) = 2.05, p = .359, V = 0.13, the observed power
592 was 0.23. The responses to the critical slide for the experimental group (N = 64) and the
593 control group (N = 63) are displayed in Figure 6. The observed counts, expected counts and
594 standardised residuals are displayed in Table 4.
595 Need for Cognition. A multinomial logistic regression revealed no statistically
2 596 significant association between Need for Cognition and response to the critical slide, χ (2, N
7 597 = 127) = 1.5, p = .472, the observed power was 0.18.
7 See Appendix D Figure 14 for relative probabilities of selecting each response depending on Need for Cognition. MANIPULATING MORAL DUMBFOUNDING 31
Table 4 Study 4 – Observed counts, expected counts, and standardised residuals for each response to the critical slide depending on cognitive load
Cognitive Load Control
Observed count Nothing Wrong 19.00 12.00 Dumbfounded 10.00 10.00 Reasons 35.00 41.00 Expected count Nothing Wrong 15.62 15.38 Dumbfounded 10.08 9.92 Reasons 38.30 37.70 Standardised residuals Nothing Wrong 1.40 -1.40 Dumbfounded -0.04 0.04 Reasons -1.19 1.19
Note. * = sig. at p < .05; ** = sig. at p < .001
598 Study 4: Discussion
599 The significant relationship between cognitive load and response to the critical slide
600 observed in previous studies was not observed in Study 4. However, the pattern of responses
601 resembled the predicted relationship and resembled the pattern of responses observed in
602 previous studies. In addition, the manipulation check did not reveal any differences between
603 the control group and the experimental group, suggesting that the cognitive load
604 manipulation was ineffective in Study 4. Studies 2, 3, and 4 were conducted using online
605 samples recruited using Amazon’s MTurk (where experimental manipulations can be
606 challenging e.g., Crump, McDonnell, & Gureckis, 2013; Goodman, Cryder, & Cheema, 2013).
607 It is possible that, in addition to the manipulation being less effective than hoped, the online
608 sample did not engage with the study as effectively as the participants in Study 1. We MANIPULATING MORAL DUMBFOUNDING 32
609 conducted a final study to assess if the effect found in Study 1 could be replicated.
610 Study 5 - College Sample revisited
611 Given the inconclusive results across studies 1-4, Study 5 was to an attempt at a direct
612 replication of Study 1, using a larger sample.
613 Study 5: Method
614 Participants and design. Study 5 was a between subjects design. The dependent
615 variable was response to the critical slide. The independent variable was cognitive load with
616 two levels: present and absent. Need for Cognition (Cacioppo & Petty, 1982; Petty et al.,
617 1984) was included as a potential correlate and moderator variable.
618 A total sample of 204 participants (144 female, 59 male; M age = 20.56, min = 18, max
619 = 48, SD = 3.86) took part. Participants in this sample were undergraduate students,
620 postgraduate students, and alumni from Mary Immaculate College (MIC), and University of
621 Limerick (UL). Participation was voluntary and participants were not reimbursed for their
622 participation.
623 Procedure and materials. Data were collected using an online questionnaire.
624 Data collection took place in a designated computer laboratory in MIC. The experimenter
625 remained in the laboratory for the duration of the study. Participants were first presented
626 with an information sheet and consent form.
627 There were two sections of the online survey: the moral judgement task, and the Need
628 for Cognition scale. The order of presentation of these was randomised. At the beginning of
629 the moral judgement task, participants in the experimental condition were presented with an
630 eight digit number/letter string and asked to memorise the sequence. After 30 seconds, the
631 experiment progressed to the next slide. Participants had the option to click “ok” and
632 progress to the next slide after 15 seconds. MANIPULATING MORAL DUMBFOUNDING 33
633 Participants were then presented with the “Julie and Mark” (Incest) vignette (Haidt et
634 al., 2000). Participants rated how right or wrong the behaviour of Julie and Mark was, and
635 were given an opportunity to provide reasons for their judgement. Following this,
636 participants were presented with a series of counter-arguments, which refuted commonly
637 used justifications for rating the behaviour as “wrong”. Dumbfounding was measured using
638 the critical slide. Following the revised judgement participants were required to reproduce
639 the number/letter string.
640 Study 5: Results
641 A total of 165 participants (80.88%) rated the behaviour of Julie and Mark as wrong
642 initially. The mean initial rating of the behaviour was, M = 2.01, SD = 1.82. There was no
643 difference in initial judgement depending on cognitive load, t(191.10) = 1.04, p = .302; d =
644 0.15.
645 Participants who selected the admission of not having reasons on the critical slide were
646 identified as dumbfounded. Forty five participants (22.06%) selected “It’s wrong but I can’t
647 think of a reason”. one hundred Nineteen participants (58.33%) selected “It’s wrong and I
648 can provide a valid reason”; and forty participants (19.61%) selected “There is nothing
649 wrong”.
650 Cognitive load manipulation check. Initial check of responses to the memory
651 task revealed that 42 participants (42.86%) successfully remembered the sequence of
652 numbers and letters. Responses to the manipulation check question revealed that 22
653 participants (22.45%) found the memory task easy. Of these, 20 participants both found the
654 task easy and got the answer right. All participants correctly remembered at least two digits
655 suggesting engagement with the manipulation.
656 The cognitive load manipulation took place before the presenting of the vignette
657 describing the behaviour to be judged. This allowed for the possibility that participants MANIPULATING MORAL DUMBFOUNDING 34
100%
75% 64.2%
52.0% Condition 50% Cognitive load
Control
25.5% 25% 22.4% 17.0% 18.9%
Percentage of participants selecting each response 0% 22 18 25 20 51 68
Nothing Wrong Dumbfounded Reasons Response to Critical Slide
Figure 7 . Study 5: Responses to critical slide and for the experimental group (N = 98) and the control group (N = 106)
658 under cognitive load may not have engaged fully with the vignette when compared to the
659 control group. An independent samples t-test revealed no significant difference in initial
660 rating in the cognitive load group, (M = 2.15, SD = 1.97), and the control group, (M =
661 2.05, SD = 1.66), t(191.10) = 1.04 , p = 0.30; d = 0.15. An independent samples t-test
662 revealed no significant difference in initial confidence in the cognitive load group, (M = 5.51,
663 SD = 1.55), and the control group, (M = 5.83, SD = 1.49), t(199.05) = -1.50 , p = 0.14; d
664 = 0.21. In view of this, it was concluded that both groups engaged equally with the task.
665 Cognitive load and dumbfounded responding. A chi-squared test for
666 independence revealed no significant association between experimental condition and
2 667 response to the critical slide, χ (2, N = 204) = 3.08, p = .215, V = 0.12, the observed power
668 was 0.33. The predicted relationship between cognitive load and dumbfounded responding MANIPULATING MORAL DUMBFOUNDING 35
Table 5 Study 5 – Observed counts, expected counts, and standardised residuals for each response to the critical slide depending on cognitive load
Cognitive Load Control
Observed count Nothing Wrong 22.00 18.00 Dumbfounded 25.00 20.00 Reasons 51.00 68.00 Expected count Nothing Wrong 19.22 20.78 Dumbfounded 21.62 23.38 Reasons 57.17 61.83 Standardised residuals Nothing Wrong 0.98 -0.98 Dumbfounded 1.14 -1.14 Reasons -1.75 1.75
Note. * = sig. at p < .05; ** = sig. at p < .001
669 was not observed in Study 5. The responses to the critical slide for the experimental group
670 (N = 98) and the control group (N = 106) are displayed in Figure 7. The observed counts,
671 expected counts and standardised residuals are displayed in Table 5.
672 Need for Cognition. A multinomial logistic regression revealed no significant
2 673 association between Need for Cognition and response to the critical slide, χ (2, N = 204) =
674 5.83, p = .054, The observed power was 0.57. Need for Cognition explained between 2.73%
675 (Cox and Snell R square) and 4.19% (Nadelkerke R squared) of the variance in responses to
676 the critical slide. As Need for Cognition increased, participants seemed to be more likely to
677 provide reasons than to present as dumbfounded, Wald = 5.48, p = .019, odds ratio = 1.54,
678 95% CI [1.07, 2.21]. (As Need for Cognition increased, participants were also possibly more
679 likely to select there is nothing wrong than present as dumbfounded Wald = 2.47, p = .116, MANIPULATING MORAL DUMBFOUNDING 36
8 680 odds ratio = 1.42, 95% CI [0.92, 2.21].)
681 Study 5: Discussion
682 Study 5 failed to demonstrate the predicted relationship between cognitive load and
683 response to the critical slide. This provides some evidence to suggest that the failure of
684 Study 4 to demonstrate this effect is likely not due to the use of an online sample. As with
685 Study 4, the pattern of responses in Study 5 resembled the predicted relationship. The
686 overall relationship between Need for Cognition and response to the critical slide was
687 non-significant, however there were trends in the predicted direction. As noted previously,
688 the studies reported here lack the statistical power to detect anything other than a large
689 effect. As such, in the next section we aggregate the data from all the studies and conduct a
690 combined analysis. We also conduct mini-meta-analyses for the primary research question.
691 Combined Results
692 The total raw responses (without accounting for engagement with the memory task)
693 for each option on the critical slide depending on experimental condition for each study and
694 all studies combined are shown in Table 6. Taking the combined raw data for each condition
695 across all 5 studies, without accounting for engagement with the memory task, a significant
2 696 association between cognitive load and responses to the critical slide was found, χ (2, N =
697 660) = 9.79, p = .007, V = 0.12, the observed power was 0.81.
698 The total responses for each option on the critical slide for each study depending on
699 experimental condition, taking engagement with the memory task into account, are displayed
700 in Table 7. The combined results for Studies 1-5 are displayed in Figures 8 and 9. Given
701 this larger combined sample, two measures of engagement with the memory task were
702 assessed: (a) a score of 7 or higher (as in previous analyses) and (b) a score of 8 or higher.
8 See Appendix D Figure 15 for relative probabilities of selecting each response depending on Need for Cognition. MANIPULATING MORAL DUMBFOUNDING 37
Table 6 Responses to critical slide for all cognitive load studies (raw)
Cognitive Load Control
Study Response N percent N percent
Study 1* Nothing wrong 15 45% 5 15%
Dumbfounded 6 18% 7 21%
Reasons 12 36% 21 64%
Study 2 Nothing wrong 11 22% 13 27%
Dumbfounded 15 29% 11 22%
Reasons 25 49% 25 51%
Study 3* Nothing wrong 21 21% 15 25%
Dumbfounded 27 26% 5 8%
Reasons 54 53% 41 67%
Study 4 Nothing wrong 19 30% 12 19%
Dumbfounded 10 16% 10 16%
Reasons 35 55% 41 65%
Study 5 Nothing wrong 22 22% 18 17%
Dumbfounded 25 26% 20 19%
Reasons 51 52% 68 64%
All Studies* Nothing wrong 88 25% 63 20%
Dumbfounded 83 24% 53 17%
Reasons 177 51% 196 63%
Note. *significant variation at p < .05; **significant variation at p < .001 MANIPULATING MORAL DUMBFOUNDING 38
Table 7 Responses to critical slide for all cognitive load studies (controlling for engagement)
Cognitive Load Control
Study Response N percent N percent
Study 1* Nothing wrong 15 45% 5 15%
Dumbfounded 6 18% 7 21%
Reasons 12 36% 21 64%
Study 2 (engaged a)* Nothing wrong 13 23% 11 25%
Dumbfounded 20 36% 6 14%
Reasons 23 41% 27 61%
Study 3 (engaged a)** Nothing wrong 15 22% 15 25%
Dumbfounded 17 25% 5 8%
Reasons 36 53% 41 67%
Study 4 (engaged a) Nothing wrong 19 30% 12 19%
Dumbfounded 10 16% 10 16%
Reasons 35 55% 41 65%
Study 5 Nothing wrong 22 22% 18 17%
Dumbfounded 25 26% 20 19%
Reasons 51 52% 68 64%
All studies (engaged a)** Nothing wrong 84 26% 61 20%
Dumbfounded 78 24% 48 16%
Reasons 157 49% 198 64%
All studies (engaged b)** Nothing wrong 84 27% 61 20%
Dumbfounded 77 25% 48 16%
Reasons 149 48% 198 64%
Note. *significant variation at p < .05; **significant variation at p < .001 MANIPULATING MORAL DUMBFOUNDING 39
Table 8 Studies 1-5 – Observed counts, expected counts, and standardised residuals for each response to the critical slide depending on cognitive load
Cognitive Load Control
Observed count Nothing Wrong 84 61 Dumbfounded 78 48 Reasons 157 198 Expected count Nothing Wrong 73.89 71.11 Dumbfounded 64.21 61.79 Reasons 180.9 174.1 Standardised residuals Nothing Wrong 1.92 -1.92 Dumbfounded 2.75* -2.75* Reasons -3.86** 3.86**
Note. * = sig. at p < .05; ** = sig. at p < .001
703 When engagement with the memory task was accounted for using (a) the primary measure of
704 engagement (N = 626, 319 experimental, 307 control), a significant association between
2 705 cognitive load and responses to the critical slide was found, χ (2, N = 626) = 15.30, p <
706 .001, V = 0.16, the observed power was 0.95. The observed counts, expected counts, and
707 standardised residuals are displayed in Table 8.
708 When the stricter measure of engagement was employed (N = 617, 310 experimental,
709 307 control), a significant association between cognitive load and response to the critical
710 slide was found, (N = 617, 310 experimental, 307 control), a significant association between
2 711 cognitive load and responses to the critical slide was found, χ (2, N = 617) = 17.28, p <
712 .001, V = 0.17, the observed power was 0.97. MANIPULATING MORAL DUMBFOUNDING 40
713 Furthermore, a mini meta-analysis was conducted and found that cognitive load
2 714 significantly influenced responding across all studies χ (10) = 26.88, p = .003 (Fisher’s
715 method); or when weighting for sample size, z = 2.874, p = .002 (Stouffer’s Z-score method). MANIPULATING MORAL DUMBFOUNDING 41
100%
75% 64.5%
Condition 49.2% 50% Cognitive load
Control
26.3% 24.5% 25% 19.9% 15.6%
Percentage of participants selecting each response 0% 84 61 78 48 157 198
Nothing Wrong Dumbfounded Reasons Response to Critical Slide
Figure 8 . Studies 1-5: Responses to critical slide for the cognitive load groups (N = 309) and the control groups (N = 317) controlling for engagement (a) at > 6 correct MANIPULATING MORAL DUMBFOUNDING 42
100%
75%
67.2% 65.1% 63.6% 64.2% 61.4%
54.7% condition 52.9% 52.0% 50% Load/Engaged
41.1% Control 36.4%
25% % of participants providing reasons for their judgements
0% 12 21 23 27 36 41 35 41 51 68
Study 1 Study 2 a Study 3 a Study 4 a Study 5 N=66 N=100 N=129 N=127 N=204 Study
Figure 9 . Rates of declaring reasons and cognitive load across each study MANIPULATING MORAL DUMBFOUNDING 43
716 Need for Cognition
717 A multinomial logistic regression revealed a statistically significant association between
2 718 Need for Cognition and response to the critical slide, χ (2, N = 654) = 12.15, p = .002, The
719 observed power was 0.89. Need for Cognition explained between 2.52% (Cox and Snell R
720 square) and 3.94% (Nadelkerke R squared) of the variance in responses to the critical slide.
721 As Need for Cognition increased, participants were significantly more likely to provide
722 reasons than to present as dumbfounded, Wald = 11.01, p < .001, odds ratio = 1.27, 95% CI
723 [1.10, 1.46]. As Need for Cognition increased, participants were also significantly more likely
724 to select there is nothing wrong than present as dumbfounded Wald = 7.86, p = .005, odds
725 ratio = 1.27, 95% CI [1.07, 1.50]. The relative probabilities of selecting each response to the
726 critical slide depending on Need for Cognition are displayed in Figure 10.
1.00
0.75
Response to Critical Slide
Dumbfounded 0.50 Nothing Wrong
Reasons Predicted Probability
0.25
0.00
−4 −2 0 2 4 Need for Cognition
Figure 10 . Studies 1-5: Probability of selecting each response to the critical slide depending on Need for Cognition MANIPULATING MORAL DUMBFOUNDING 44
727 General Discussion
728 We conducted five studies to test a conflict in dual-processes explanation of moral
729 dumbfounding. We hypothesised that moral dumbfounding emerges as a result of conflict
730 between a habitual/intuitive response and deliberation. We found some evidence in support
731 of this explanation of moral dumbfounding. This evidence was not as strong as expected.
732 This may be due to limitations in the methods employed, or it may be evidence that the
733 explanation tested here does not capture the full complexity of the phenomenon.
734 Interpreting the Results
735 The relative roles of intuition and deliberation in producing each of the possible
736 responses in the dumbfounding paradigm were identified, and these responses were ranked
737 according to level of deliberation involved in each. The responses are as follows: (1) reason;
738 (2) nothing wrong; and (3) dumbfounding. We hypothesised that, (1) a reason response
739 involves the most deliberation. The relative roles of intuition and deliberation in the
740 remaining responses were less clear, particularly because of the possibility of that competing
741 intuitions may emerge. If deliberation fails to identify justifications for an initial intuition, a
742 person may find that they must choose between two competing intuitions (initial judgement;
743 that moral judgements should be justifiable). Despite this limitation, a clear candidate for
744 relying on deliberation the most was identified: (1) reason.
745 Cognitive load and Providing Reasons. Having identified a response that
746 clearly relied on deliberation more than the other responses, an experimental manipulation
747 that is known to influence deliberative responding was selected for study: cognitive load (De
748 Neys, 2006; Evans & Curtis-Holmes, 2005; Evans & Stanovich, 2013; Schmidt, 2016). In
749 addition to this, an individual difference variable that has to been linked to a tendency to
750 engage in deliberation was identified and selected for investigation (Need for Cognition,
751 Cacioppo & Petty, 1982; Evans & Stanovich, 2013; Petty et al., 1984). Five studies were
752 conducted investigating cognitive load and moral dumbfounding. The target response was MANIPULATING MORAL DUMBFOUNDING 45
753 “It’s wrong and I can provide a valid reason” on the critical slide. We hypothesised that
754 cognitive load would inhibit participants’ ability to provide reasons for their judgements, and
755 that under cognitive load reduced incidences of participants stating they could provide a
756 reason would be observed. We also hypothesised that providing reasons would be related to
757 Need for Cognition scores, such that participants who score higher on Need for Cognition
758 would be more likely to provide reasons than select any other response on the critical slide.
759 Study 1 showed a clear reduction in providing reasons for a judgement depending on
760 cognitive load (resulting in an increase in selecting “There is nothing wrong”). The results of
761 Study 2 indicated that any effect cognitive load had on ability to generate reasons for a
762 judgement was related to level of engagement with the cognitive load task. Study 3
763 demonstrated that when controlling for engagement with the cognitive load task, cognitive
764 load led to higher rates of dumbfounding. However, Studies 4 and 5 failed to replicate these
765 findings. Three studies described here therefore showed that participants actively engaging
766 with cognitive load task were consistently less effective at identifying reasons for their
767 judgements. Studies 4 and 5 showed a similar pattern of responses, however this was not
768 statistically significant.
769 Need for Cognition in the Dumbfounding Paradigm. There was no
770 significant relationship between Need for Cognition and the critical slide for Studies 1, 2, 4,
771 and 5. In Study 3, higher Need for Cognition predicted higher rates of providing reasons
772 than dumbfounded responding. In the combined analysis, higher Need for Cognition scores
773 predicted higher rates of both providing reasons and selecting “there is nothing wrong” and
774 reduced rates of dumbfounded responding.
775 Key Findings. The combined results of Studies 1-5 appear to indicate that engaging
776 in a cognitive load task leads to reduced incidences of providing reasons (as hypothesised).
777 Studies 4 and 5 to failed replicate this effect, so caution is advised when interpreting this
778 result, however, given the consistency in the pattern of responses, the clear result in the
779 combined analysis, and in the meta-analysis, it is likely that the failure to replicate the effect MANIPULATING MORAL DUMBFOUNDING 46
780 was due to a lack of statistical power. Furthermore, dumbfounded responding appeared to be
781 linked to lower Need for Cognition scores. Again, caution is advised in interpreting this
782 result, particularly given that only 1 (Study 3) of the Studies 1-5 individually found any
783 relationship between Need for Cognition and response to the critical slide.
784 Implications
785 Overall, the results of the studies described here provide some evidence for the conflict
786 in dual-processes explanation of moral dumbfounding identified here, however the
787 inconsistencies in the results poses a challenge to this finding. Despite the inconsistencies,
788 the studies described here do not provide convincing evidence that this explanation of moral
789 dumbfounding is wrong. From these somewhat mixed results, what seems clear is that moral
790 dumbfounding is not a simple matter of habitual or deliberative responses. The complexity
791 of moral dumbfounding becomes apparent when attempting to describe the various possible
792 responses in the paradigm in terms of the relative roles of deliberation and intuition. The
793 possibility that a failure of deliberation to provide reasons for an intuition may lead to
794 alternative and competing intuitions to become salient means that in some cases, moral
795 dumbfounding may involve conflict between competing intuitions.
796 The apparent conflict between competing intuitions is more complicated than conflict
797 between intuition and deliberation. Recall that the examples of conflict identified above
798 clearly pitted an intuitive response against a deliberative response, and resolution of this
799 conflict was relatively straight forward. For example, in base-rate neglect problems there is
800 only one correct answer. The intuitive response is incorrect and deliberation leads to an
801 alternative response. This conflict is easily resolved because deliberation has clearly identified
802 the intuitive response as incorrect. In contrast moral judgements are not clearly correct or
803 incorrect and the failure of deliberation to identify reasons for an intuition is not necessarily
804 evidence that the intuition is incorrect. That moral dumbfounding is more complicated than
805 classic cases of conflict in dual-processes does not necessarily mean that attempting to MANIPULATING MORAL DUMBFOUNDING 47
806 explain moral dumbfounding as the result of dual-process conflict is without merit. It is
807 possible that conflict in dual-processes is still implicated in moral dumbfounding. However,
808 where conflict is normally resolved through suppression of erroneous intuitive responses, in
809 the dumbfounding paradigm this conflict is not easily resolved leading participants are faced
810 with a conflict between competing intuitions. Given that this explanation identifies
811 dumbfounding as arising from conflict between competing intuitions (same underlying
812 mechanisms) as opposed to conflict between intuition and deliberation (different underlying
813 mechanisms) means that testing this explanation is particularly problematic.
814 Limitations and Future Directions
815 There are a number of related limitations of the studies conducted here that may have
816 contributed to the inconclusive results in different ways. Firstly, it is possible that the
817 inconclusive result is due to limitations in the method of data collection employed across
818 Studies 2, 3, and 4. Online data collection through MTurk is useful for ease of access to
819 willing participants resulting in highly efficient data collection. However, the absence of
820 oversight or supervision of participants taking part in the studies means that complex
821 manipulations such as cognitive load manipulations may not be as successful as in a
822 controlled laboratory setting (Crump et al., 2013; Goodman et al., 2013). In response to the
823 difficulties in manipulating cognitive load in an online study, we conducted an additional
824 study (Study 5) returning to a college sample where the study took place in a controlled
825 environment, under the supervision of the experimenter.
826 The failure of Study 5 to replicate the findings of Studies 1, 2, and 3 may be due to the
827 nature of the variables used. The dependent variable in these studies is nominal/categorical.
828 This variable type is not well suited for identifying small or subtle effects; consider the
829 extensive research identifying disgust effects on moral judgements (e.g., Cameron et al., 2013;
830 David & Olatunji, 2011; Landy & Goodwin, 2015; Rozin, Haidt, & MacCauley, 2009; Russell
831 & Giner-Sorolla, 2011; Sabo & Giner-Sorolla, 2017; Schnall, Haidt, Clore, & Jordan, 2008; MANIPULATING MORAL DUMBFOUNDING 48
832 Wheatley & Haidt, 2005). May (2014) noted that the effect of incidental disgust is largely on
833 the severity of a judgement, as opposed to altering the valence of the judgement. The use of
834 a binary right/wrong measure of judgements therefore would fail to identify any effect for
835 disgust on moral judgements. As such, the measures used in the dumbfounding paradigm
836 mean that identifying influences on dumbfounding is particularly challenging, because minor
837 influences on people’s ability to provide reasons may not yield measurable effects.
838 Conclusion
839 The studies described here aimed to investigate if cognitive load influenced the degree
840 to which people successfully identified reasons for their judgements. The individual
841 difference variable Need for Cognition was also investigated.
842 For 4 of the 5 studies conducted, there did not appear to be a relationship between
843 Need for Cognition and response to the critical slide, however in Study 3, and the final
844 analysis of all the studies together, lower Need for Cognition scores were associated with
845 dumbfounded responding. The effect size was quite small, and this effect only emerged in 1
846 study, and when all the data were combined. However, this may prove useful in furthering
847 our understanding of moral dumbfounding and the making of moral judgements more
848 generally. That dumbfounding may be weakly related to scoring lower on Need for Cognition
849 is interesting and may provide a useful area for follow-up work.
850 The primary aim of this paper was to test a dual-process conflict explanation of moral
851 dumbfounding. According to this view, providing reasons for a judgement is grounded in
852 deliberation, and dumbfounded responding is grounded in habitual responding. The extent
853 to which revising a judgement is grounded in deliberation or habitual responding is unclear,
854 and may vary depending on the individual. The key prediction of this explanation is that
855 inhibiting deliberative responding, through a cognitive load manipulation, would inhibit the
856 identification of reasons. MANIPULATING MORAL DUMBFOUNDING 49
857 The studies presented here offer suggestive evidence that cognitive load inhibits the
858 identification of reasons in the dumbfounding paradigm. Engagement with the cognitive load
859 manipulation on the part of the participants is essential for it to successfully inhibit the
860 identification of reasons. The pattern of responses across the studies is consistent with the
861 possibility that revising a judgement requires more deliberation than providing a
862 dumbfounded response, however this claim was not tested directly, and there is no clear
863 evidence in support of it. Some evidence in support of the primary prediction, that
864 inhibiting deliberative responding should inhibit the identification of reasons was found.
865 This suggests that a dumbfounded response is a more habitual/intuitive response whereas
866 providing reasons relies more on deliberative responding. But that the processes that give
867 rise to judgements and reason-giving in complex situations are not simply the operation
868 dual-systems or a conflict between them, but a complex of interacting processes in which
869 these aspects play a role.
870 Data Accessibility Statement
871 All participant data, and analysis scripts can be found on this paper’s project page on
872 the Open Science Framework at https://osf.io/wge5t/.
873 All statistical analysis was conducted using R (Version 3.6.1; R Core Team, 2017) and
874 the R-packages afex (Version 0.25.1; Singmann, Bolker, & Westfall, 2015), car (Version 3.0.3;
875 Fox & Weisberg, 2011; Fox, Weisberg, & Price, 2018), carData (Version 3.0.2; Fox et al.,
876 2018), citr (Version 0.3.0; Aust, 2016), DescTools (Version 0.99.28; et mult. al., 2019),
877 desnum (Version 0.1.1; McHugh, 2017), dplyr (Version 0.8.3; Wickham, Francois, Henry, &
878 Müller, 2017), emmeans (Version 1.4.1; Lenth, 2019), extrafont (Version 0.17; Chang, 2014),
879 foreign (Version 0.8.72; R Core Team, 2018), Formula (Version 1.2.3; Zeileis & Croissant,
880 2010), ggplot2 (Version 3.2.1; Wickham, 2009), knitr (Version 1.23; Xie, 2015), koRpus
881 (Version 0.11.5; Michalke, 2018a, 2019), koRpus.lang.en (Version 0.1.3; Michalke, 2019), lme4
882 (Version 1.1.21; Bates, Mächler, Bolker, & Walker, 2015), lmtest (Version 0.9.36; Zeileis & MANIPULATING MORAL DUMBFOUNDING 50
883 Hothorn, 2002), lsmeans (Version 2.30.0; Lenth, 2016), lsr (Version 0.5; Navarro, 2015),
884 Matrix (Version 1.2.17; Bates & Maechler, 2017), metap (Version 1.1; Dewey, 2017), mlogit
885 (Version 0.4.1; Croissant, 2013), nnet (Version 7.3.12; Venables & Ripley, 2002), papaja
886 (Version 0.1.0.9842; Aust & Barth, 2018), plyr (Version 1.8.4; Wickham et al., 2017;
887 Wickham, 2011), powerMediation (Version 0.2.9; Qiu, 2018), pwr (Version 1.2.2; Champely,
888 2018), reshape2 (Version 1.4.3; Wickham, 2007), scales (Version 1.0.0; Wickham, 2016),
889 sjstats (Version 0.17.4; Lüdecke, 2018), sylly (Version 0.1.5; Michalke, 2018b), tibble (Version
890 2.1.3; Müller & Wickham, 2017), VGAM (Version 1.1.1; Yee & Wild, 1996; Yee, 2010, 2013;
891 Yee & Hadi, 2014; Yee, Stoklosa, & Huggins, 2015), wordcountaddin (Version 0.3.0.9000;
892 Marwick, 2019), and zoo (Version 1.8.6; Zeileis & Grothendieck, 2005). MANIPULATING MORAL DUMBFOUNDING 51
893 References
894 Abelson, R. P. (1988). Conviction. American Psychologist, 43 (4), 267–275.
895 doi:10.1037/0003-066X.43.4.267
896 Amazon Web Services Inc. (2016). Amazon Mechanical Turk.
897 Aust, F. (2016). Citr: ’RStudio’ Add-in to Insert Markdown Citations. Retrieved from
898 https://CRAN.R-project.org/package=citr
899 Aust, F., & Barth, M. (2018). Papaja: Create APA manuscripts with R Markdown.
900 Retrieved from https://github.com/crsh/papaja
901 Bates, D., & Maechler, M. (2017). Matrix: Sparse and Dense Matrix Classes and Methods.
902 Retrieved from https://CRAN.R-project.org/package=Matrix
903 Bates, D., Mächler, M., Bolker, B., & Walker, S. (2015). Fitting Linear Mixed-Effects Models
904 Using lme4. Journal of Statistical Software, 67 (1), 1–48. doi:10.18637/jss.v067.i01
905 Bonner, C., & Newell, B. R. (2010). In conflict with ourselves? An investigation of heuristic
906 and analytic processes in decision making. Memory & Cognition, 38 (2), 186–196.
907 doi:10.3758/MC.38.2.186
908 Brand, C. (2016). Dual-Process Theories in Moral Psychology: Interdisciplinary Approaches
909 to Theoretical, Empirical and Practical Considerations. Springer.
910 Cacioppo, J. T., & Petty, R. E. (1982). The need for cognition. Journal of Personality and
911 Social Psychology, 42 (1), 116–131. doi:10.1037/0022-3514.42.1.116
912 Cameron, C. D., Payne, B. K., & Doris, J. M. (2013). Morality in high definition: Emotion
913 differentiation calibrates the influence of incidental disgust on moral judgments.
914 Journal of Experimental Social Psychology, 49 (4), 719–725. MANIPULATING MORAL DUMBFOUNDING 52
915 doi:10.1016/j.jesp.2013.02.014
916 Champely, S. (2018). Pwr: Basic Functions for Power Analysis. Retrieved from
917 https://CRAN.R-project.org/package=pwr
918 Chang, W. (2014). Extrafont: Tools for using fonts. Retrieved from
919 https://CRAN.R-project.org/package=extrafont
920 Crockett, M. J. (2013). Models of morality. Trends in Cognitive Sciences, 17 (8), 363–366.
921 doi:10.1016/j.tics.2013.06.005
922 Croissant, Y. (2013). Mlogit: Multinomial logit model. Retrieved from
923 https://CRAN.R-project.org/package=mlogit
924 Crump, M. J. C., McDonnell, J. V., & Gureckis, T. M. (2013). Evaluating Amazon’s
925 Mechanical Turk as a Tool for Experimental Behavioral Research. PLOS ONE, 8 (3),
926 e57410. doi:10.1371/journal.pone.0057410
927 Cushman, F. A. (2013). Action, Outcome, and Value A Dual-System Framework for
928 Morality. Personality and Social Psychology Review, 17 (3), 273–292.
929 doi:10.1177/1088868313495594
930 Cushman, F. A., Young, L., & Greene, J. D. (2010). Multi-system Moral Psychology. In J.
931 M. Doris (Ed.), The Moral Psychology Handbook (pp. 47–71). Oxford; New York:
932 Oxford University Press.
933 David, B., & Olatunji, B. O. (2011). The effect of disgust conditioning and disgust
934 sensitivity on appraisals of moral transgressions. Personality and Individual
935 Differences, 50 (7), 1142–1146. doi:10.1016/j.paid.2011.02.004
936 De Neys, W. (2006). Dual Processing in Reasoning: Two Systems but One Reasoner.
937 Psychological Science, 17 (5), 428–433. doi:10.1111/j.1467-9280.2006.01723.x MANIPULATING MORAL DUMBFOUNDING 53
938 De Neys, W. (2012). Bias and Conflict: A Case for Logical Intuitions. Perspectives on
939 Psychological Science, 7 (1), 28–38. doi:10.1177/1745691611429354
940 De Neys, W., & Glumicic, T. (2008). Conflict monitoring in dual process theories of
941 thinking. Cognition, 106 (3), 1248–1299. doi:10.1016/j.cognition.2007.06.002
942 De Neys, W., & Schaeken, W. (2007). When people are more logical under cognitive load:
943 Dual task impact on scalar implicature. Experimental Psychology, 54 (2), 128–133.
944 doi:10.1027/1618-3169.54.2.128
945 Dewey, M. (2017). Metap: Meta-analysis of significance values.
946 Drolet, A., & Frances Luce, M. (2004). The Rationalizing Effects of Cognitive Load on
947 Emotion-Based Trade-off Avoidance. Journal of Consumer Research, 31 (1), 63–77.
948 doi:10.1086/383424
949 et mult. al., A. S. (2019). DescTools: Tools for Descriptive Statistics. Retrieved from
950 https://cran.r-project.org/package=DescTools
951 Evans, J. S. B. T. (2007). On the resolution of conflict in dual process theories of reasoning.
952 Thinking & Reasoning, 13 (4), 321–339. doi:10.1080/13546780601008825
953 Evans, J. S. B. T. (2008). Dual-Processing Accounts of Reasoning, Judgment, and Social
954 Cognition. Annual Review of Psychology, 59 (1), 255–278.
955 doi:10.1146/annurev.psych.59.103006.093629
956 Evans, J. S. B. T., & Curtis-Holmes, J. (2005). Rapid responding increases belief bias:
957 Evidence for the dual-process theory of reasoning. Thinking & Reasoning, 11 (4),
958 382–389. doi:10.1080/13546780542000005
959 Evans, J. S. B. T., & Stanovich, K. E. (2013). Dual-Process Theories of Higher Cognition:
960 Advancing the Debate. Perspectives on Psychological Science, 8 (3), 223–241. MANIPULATING MORAL DUMBFOUNDING 54
961 doi:10.1177/1745691612460685
962 Forsterlee, R., & Ho, R. (1999). An Examination of the Short form of the Need for
963 Cognition Scale Applied in an Australian Sample. Educational and Psychological
964 Measurement, 59 (3), 471–480. doi:10.1177/00131649921969983
965 Fox, J., & Weisberg, S. (2011). An R Companion to Applied Regression (Second.). Thousand
966 Oaks CA: Sage. Retrieved from
967 http://socserv.socsci.mcmaster.ca/jfox/Books/Companion
968 Fox, J., Weisberg, S., & Price, B. (2018). carData: Companion to Applied Regression Data
969 Sets. Retrieved from https://CRAN.R-project.org/package=carData
970 Goodman, J. K., Cryder, C. E., & Cheema, A. (2013). Data Collection in a Flat World: The
971 Strengths and Weaknesses of Mechanical Turk Samples. Journal of Behavioral
972 Decision Making, 26 (3), 213–224. doi:10.1002/bdm.1753
973 Greene, J. D. (2008). The Secret Joke of Kant’s Soul. In W. Sinnott-Armstrong, Moral
974 Psychology Volume 3: The neurosciences of morality: Emotion, brain disorders, and
975 development (pp. 35–79). Cambridge (Mass.): the MIT press.
976 Greene, J. D., Morelli, S. A., Lowenberg, K., Nystrom, L. E., & Cohen, J. D. (2008).
977 Cognitive load selectively interferes with utilitarian moral judgment. Cognition,
978 107 (3), 1144–1154. doi:10.1016/j.cognition.2007.11.004
979 Haidt, J. (2001). The emotional dog and its rational tail: A social intuitionist approach to
980 moral judgment. Psychological Review, 108 (4), 814–834.
981 doi:10.1037/0033-295X.108.4.814
982 Haidt, J., & Björklund, F. (2008). Social Intuitionists Answer Six Questions about Moral
983 Psychology. In W. Sinnott-Armstrong (Ed.), Moral psychology Volume 2, The MANIPULATING MORAL DUMBFOUNDING 55
984 cognitive science of morality: Intuition and diversity (pp. 181–217). London: MIT.
985 Haidt, J., Björklund, F., & Murphy, S. (2000). Moral dumbfounding: When intuition finds
986 no reason. Unpublished Manuscript, University of Virginia.
987 Just, M. A., & Carpenter, P. A. (1992). A capacity theory of comprehension: Individual
988 differences in working memory. Psychological Review, 99 (1), 122–149.
989 doi:10.1037/0033-295X.99.1.122
990 Kruglanski, A. W., & Webster, D. M. (1996). Motivated closing of the mind: "Seizing" and
991 "freezing.". Psychological Review, 103 (2), 263–283. doi:10.1037/0033-295X.103.2.263
992 Kruglanski, A. W., Webster, D. M., & Klem, A. (1993). Motivated resistance and openness
993 to persuasion in the presence or absence of prior information. Journal of Personality
994 and Social Psychology, 65 (5), 861–876. doi:10.1037/0022-3514.65.5.861
995 Landy, J. F., & Goodwin, G. P. (2015). Does Incidental Disgust Amplify Moral Judgment?
996 A Meta-Analytic Review of Experimental Evidence. Perspectives on Psychological
997 Science, 10 (4), 518–536. doi:10.1177/1745691615583128
998 Lenth, R. (2019). Emmeans: Estimated Marginal Means, aka Least-Squares Means.
999 Retrieved from https://CRAN.R-project.org/package=emmeans
1000 Lenth, R. V. (2016). Least-Squares Means: The R Package lsmeans. Journal of Statistical
1001 Software, 69 (1), 1–33. doi:10.18637/jss.v069.i01
1002 Lerner, M. J., & Goldberg, J. H. (1999). When Do Decent People Blame Victims? The
1003 Differing Effects of the Explicit/Rational and Implicit/Experiential Cognitive
1004 Systems. In S. Chaiken & Y. Trope (Eds.), Dual-process Theories in Social
1005 Psychology (pp. 627–640). Guilford Press.
1006 Lüdecke, D. (2018). Sjstats: Statistical Functions for Regression Models. Retrieved from MANIPULATING MORAL DUMBFOUNDING 56
1007 https://CRAN.R-project.org/package=sjstats
1008 Marwick, B. (2019). Wordcountaddin: Word counts and readability statistics in R markdown
1009 documents.
1010 May, J. (2014). Does Disgust Influence Moral Judgment? Australasian Journal of
1011 Philosophy, 92 (1), 125–141.
1012 McGregor, I. (2006a). Offensive Defensiveness: Toward an Integrative Neuroscience of
1013 Compensatory Zeal After Mortality Salience, Personal Uncertainty, and Other
1014 Poignant Self-Threats. Psychological Inquiry, 17 (4), 299–308.
1015 doi:10.1080/10478400701366977
1016 McGregor, I. (2006b). Zeal Appeal: The Allure of Moral Extremes. Basic and Applied Social
1017 Psychology, 28 (4), 343–348. doi:10.1207/s15324834basp2804_7
1018 McGregor, I., Zanna, M. P., Holmes, J. G., & Spencer, S. J. (2001). Compensatory
1019 conviction in the face of personal uncertainty: Going to extremes and being oneself.
1020 Journal of Personality and Social Psychology, 80 (3), 472–488.
1021 doi:10.1037/0022-3514.80.3.472
1022 McHugh, C. (2017). Desnum: Creates some useful functions. Retrieved from
1023 https://github.com/cillianmiltown/R_desnum
1024 McHugh, C., McGann, M., Igou, E. R., & Kinsella, E. L. (2017). Searching for Moral
1025 Dumbfounding: Identifying Measurable Indicators of Moral Dumbfounding. Collabra:
1026 Psychology, 3 (1). doi:10.1525/collabra.79
1027 Michalke, M. (2018a). koRpus: An R Package for Text Analysis. Retrieved from
1028 https://reaktanz.de/?c=hacking&s=koRpus
1029 Michalke, M. (2018b). Sylly: Hyphenation and Syllable Counting for Text Analysis. MANIPULATING MORAL DUMBFOUNDING 57
1030 Retrieved from https://reaktanz.de/?c=hacking&s=sylly
1031 Michalke, M. (2019). koRpus.Lang.En: Language Support for ’koRpus’ Package: English.
1032 Retrieved from https://reaktanz.de/?c=hacking&s=koRpus
1033 Müller, K., & Wickham, H. (2017). Tibble: Simple Data Frames. Retrieved from
1034 https://CRAN.R-project.org/package=tibble
1035 Navarro, D. (2015). Learning statistics with R: A tutorial for psychology students and other
1036 beginners. (Version 0.5). Adelaide, Australia: University of Adelaide. Retrieved from
1037 http://ua.edu.au/ccs/teaching/lsr
1038 Petty, R. E., Cacioppo, J. T., & Kao, C. F. (1984). The efficient assessment of need for
1039 cognition. Journal of Personality Assessment, 48 (3), 306–307.
1040 Petty, R. E., Feinstein, J. A., Blair, W., & Jarvis, G. (1996). Dispositional differences in
1041 cognitive motivation: The life and times of individuals varying in need for cognition.
1042 Psych. Bull, 197–253.
1043 Prinz, J. J. (2005). Passionate Thoughts: The Emotional Embodiment of Moral Concepts.
1044 In D. Pecher & R. A. Zwaan (Eds.), Grounding Cognition: The Role of Perception
1045 and Action in Memory, Language, and Thinking (pp. 93–114). Cambridge University
1046 Press.
1047 Qiu, W. (2018). powerMediation: Power/Sample Size Calculation for Mediation Analysis.
1048 Retrieved from https://CRAN.R-project.org/package=powerMediation
1049 R Core Team. (2017). R: A Language and Environment for Statistical Computing. Vienna,
1050 Austria: R Foundation for Statistical Computing. Retrieved from
1051 https://www.R-project.org/
1052 R Core Team. (2018). Foreign: Read Data Stored by ’Minitab’, ’S’, ’SAS’, ’SPSS’, ’Stata’, MANIPULATING MORAL DUMBFOUNDING 58
1053 ’Systat’, ’Weka’, ’dBase’, ... Retrieved from
1054 https://CRAN.R-project.org/package=foreign
1055 Royzman, E. B., Kim, K., & Leeman, R. F. (2015). The curious tale of Julie and Mark:
1056 Unraveling the moral dumbfounding effect. Judgment and Decision Making, 10 (4),
1057 296–313.
1058 Rozin, P., Haidt, J., & MacCauley, C. (2009). Disgust: The body and soul emotion in the
1059 21st century. na.
1060 Rozin, P., Markwith, M., & McCauley, C. (1994). Sensitivity to indirect contacts with other
1061 persons: AIDS aversion as a composite of aversion to strangers, infection, moral taint,
1062 and misfortune. Journal of Abnormal Psychology, 103 (3), 495–504.
1063 doi:10.1037/0021-843X.103.3.495
1064 Russell, P. S., & Giner-Sorolla, R. (2011). Social Justifications for Moral Emotions: When
1065 Reasons for Disgust Are Less Elaborated Than for Anger. Emotion, 11 (3), 637–646.
1066 doi:10.1037/a0022600
1067 Sabo, J. S., & Giner-Sorolla, R. (2017). Imagining Wrong: Fictitious Contexts Mitigate
1068 Condemnation of Harm More Than Impurity. Journal of Experimental Psychology:
1069 General, 146 (1), 134–153. doi:10.1037/xge0000251
1070 Schmidt, D. (2016). The Effects of Cognitive Load and Stereotyped Groups on Punitiveness.
1071 CMC Senior Theses.
1072 Schnall, S., Haidt, J., Clore, G. L., & Jordan, A. H. (2008). Disgust as Embodied Moral
1073 Judgment. Personality & Social Psychology Bulletin, 34 (8), 1096–1109.
1074 doi:10.1177/0146167208317771
1075 Singmann, H., Bolker, B., & Westfall, J. (2015). Afex: Analysis of Factorial Experiments. MANIPULATING MORAL DUMBFOUNDING 59
1076 Retrieved from https://CRAN.R-project.org/package=afex
1077 Tversky, A., & Kahneman, D. (1983). Extensional versus intuitive reasoning: The
1078 conjunction fallacy in probability judgment. Psychological Review, 90 (4), 293–315.
1079 doi:10.1037/0033-295X.90.4.293
1080 Venables, W. N., & Ripley, B. D. (2002). Modern Applied Statistics with S (Fourth.). New
1081 York: Springer. Retrieved from http://www.stats.ox.ac.uk/pub/MASS4
1082 Wheatley, T., & Haidt, J. (2005). Hypnotic Disgust Makes Moral Judgments More Severe.
1083 Psychological Science, 16 (10), 780–784. doi:10.1111/j.1467-9280.2005.01614.x
1084 Wickham, H. (2007). Reshaping Data with the reshape Package. Journal of Statistical
1085 Software, 21 (12), 1–20. Retrieved from http://www.jstatsoft.org/v21/i12/
1086 Wickham, H. (2009). Ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag New
1087 York. Retrieved from http://ggplot2.org
1088 Wickham, H. (2011). The Split-Apply-Combine Strategy for Data Analysis. Journal of
1089 Statistical Software, 40 (1), 1–29. Retrieved from http://www.jstatsoft.org/v40/i01/
1090 Wickham, H. (2016). Scales: Scale Functions for Visualization. Retrieved from
1091 https://CRAN.R-project.org/package=scales
1092 Wickham, H., Francois, R., Henry, L., & Müller, K. (2017). Dplyr: A Grammar of Data
1093 Manipulation. Retrieved from https://CRAN.R-project.org/package=dplyr
1094 Xie, Y. (2015). Dynamic Documents with R and knitr (2nd ed.). Boca Raton, Florida:
1095 Chapman and Hall/CRC. Retrieved from http://yihui.name/knitr/
1096 Yee, T. W. (2010). The VGAM Package for Categorical Data Analysis. Journal of Statistical
1097 Software, 32 (10), 1–34. Retrieved from http://www.jstatsoft.org/v32/i10/ MANIPULATING MORAL DUMBFOUNDING 60
1098 Yee, T. W. (2013). Two-parameter reduced-rank vector generalized linear models.
1099 Computational Statistics and Data Analysis. Retrieved from
1100 http://ees.elsevier.com/csda
1101 Yee, T. W., & Hadi, A. F. (2014). Row-column interaction models, with an R
1102 implementation. Computational Statistics, 29 (6), 1427–1445.
1103 Yee, T. W., Stoklosa, J., & Huggins, R. M. (2015). The VGAM Package for
1104 Capture-Recapture Data Using the Conditional Likelihood. Journal of Statistical
1105 Software, 65 (5), 1–33. Retrieved from http://www.jstatsoft.org/v65/i05/
1106 Yee, T. W., & Wild, C. J. (1996). Vector Generalized Additive Models. Journal of Royal
1107 Statistical Society, Series B, 58 (3), 481–493.
1108 Zeileis, A., & Croissant, Y. (2010). Extended Model Formulas in R: Multiple Parts and
1109 Multiple Responses. Journal of Statistical Software, 34 (1), 1–13.
1110 doi:10.18637/jss.v034.i01
1111 Zeileis, A., & Grothendieck, G. (2005). Zoo: S3 Infrastructure for Regular and Irregular
1112 Time Series. Journal of Statistical Software, 14 (6), 1–27. doi:10.18637/jss.v014.i06
1113 Zeileis, A., & Hothorn, T. (2002). Diagnostic Checking in Regression Relationships. R News,
1114 2 (3), 7–10. Retrieved from https://CRAN.R-project.org/doc/Rnews/ MANIPULATING MORAL DUMBFOUNDING 61
1115 Appendices
1116 Appendix A: Moral Scenarios
1117 Incest
1118 Julie and Mark, who are brother and sister, are travelling together in France. They
1119 are both on summer vacation from college. One night they are staying alone in a cabin near
1120 the beach. They decide that it would be interesting and fun if they tried making love. At
1121 very least it would be a new experience for each of them. Julie was already taking birth
1122 control pills, but Mark uses a condom too, just to be safe. They both enjoy it, but they
1123 decide not to do it again. They keep that night as a special secret between them, which
1124 makes them feel even closer to each other (Haidt et al., 2000). MANIPULATING MORAL DUMBFOUNDING 62
1125 Appendix B: Sample Statements to Challenge Judgements
1126 Incest
1127 • Do you not agree that any concerns regarding reproductive complications are eased by
1128 their using of two forms of contraception?
1129 • And do you accept that they are both consenting adults, and that they both consented
1130 and enjoyed it?
1131 • And do you concede that nobody else was affected by their actions? MANIPULATING MORAL DUMBFOUNDING 63
1132 Appendix C: Post Discussion Questionnaire
1133 How sure were you about your judgement?
1134 1 2 3 4 5 6 7
1135 Not at all Extremely sure
1136 ______
1137 How much did you change your mind?
1138 1 2 3 4 5 6 7
1139 Not at all Extremely
1140 ______
1141 How confused were you?
1142 1 2 3 4 5 6 7
1143 Not at all Extremely confused
1144 ______
1145 How irritated were you?
1146 1 2 3 4 5 6 7
1147 Not at all Extremely irritated
1148 ______
1149 How much was your judgement based on reason? MANIPULATING MORAL DUMBFOUNDING 64
1150 1 2 3 4 5 6 7
1151 Not at all Extremely
1152 ______
1153 How much was your judgement based on “gut” feeling?
1154 1 2 3 4 5 6 7
1155 Not at all Extremely MANIPULATING MORAL DUMBFOUNDING 65
1156 Appendix D: Need for Cognition and Responses to the Critical Slide
1.00
0.75
Response to Critical Slide
Dumbfounded 0.50 Nothing Wrong
Reasons Predicted Probability
0.25
0.00
−2 0 2 Need for Cognition
Figure 11 . Study 1: Probability of selecting each response to the critical slide depending on Need for Cognition MANIPULATING MORAL DUMBFOUNDING 66
1.00
0.75
Response to Critical Slide
Dumbfounded 0.50 Nothing Wrong
Reasons Predicted Probability
0.25
0.00
−4 −2 0 2 Need for Cognition
Figure 12 . Study 2: Probability of selecting each response to the critical slide depending on Need for Cognition
1.00
0.75
Response to Critical Slide
Dumbfounded 0.50 Nothing Wrong
Reasons Predicted Probability
0.25
0.00
−4 −2 0 2 4 Need for Cognition
Figure 13 . Study 3: Probability of selecting each response to the critical slide depending on Need for Cognition MANIPULATING MORAL DUMBFOUNDING 67
1.00
0.75
Response to Critical Slide
Dumbfounded 0.50 Nothing Wrong
Reasons Predicted Probability
0.25
0.00
−2 0 2 4 Need for Cognition
Figure 14 . Study 4: Probability of selecting each response to the critical slide depending on Need for Cognition
1.00
0.75
Response to Critical Slide
Dumbfounded 0.50 Nothing Wrong
Reasons Predicted Probability
0.25
0.00
−2 0 2 4 Need for Cognition
Figure 15 . Study 5: Probability of selecting each response to the critical slide depending on Need for Cognition