Acta Psychologica 154 (2015) 69–76

Contents lists available at ScienceDirect

Acta Psychologica journal homepage: www.elsevier.com/ locate/actpsy

Highly reflective reasoners show no signs of belief inhibition Annika M. Svedholm-Häkkinen ⁎ Division of Cognitive Psychology and Neuropsychology, Institute of Behavioural Sciences, University of Helsinki, Finland

a r t i c l e

i n f o

Article history: Received 17 September 2014 Received in revised form 19 November 2014 Accepted 25 November 2014 Available online 10 December 2014 PsycINFO classification: 2340 Keywords: Reasoning Conflict Logic Belief inhibition Dual process theory

a b s t r a c t The processes underlying individual differences in reasoning performance are not entirely understood. What do people who do well on reasoning tasks where beliefs and logic conflict do differently from other people? Because abundant evidence shows that even poorer reasoners detect these conflicts, it has been suggested that individual differences in reasoning performance arise from inhibition failures later in the reasoning process. The present paper argues that a minority of highly skilled reasoners may deviate from this general reasoning process from an early stage. Two studies investigated signs of belief inhibition using a lexical access paradigm (Study 1) and a negative priming paradigm (Study 2). Study 1 showed that while other people exhibited signs of belief inhibition following a belief–logic conflict, people with the highest disposition for cognitive reflection did not. In Study 2, this finding was replicated and similar results were also obtained when comparing groups with higher and lower general cognitive ability. Two possible explanations are discussed. The reasoners with a highly reflective cognitive style or high general cognitive ability may have engaged and inhibited belief processing but if so, they may have been exceptionally efficient at recovering from it, wherefore no belief inhibition effects were found. An alternative account is that these reasoners started Type 2 processing directly, without first engaging in and then inhibiting belief-based processing. Under either explanation, the results indicate that individual differences in reasoning may partly arise from differences that occur early in the reasoning process. © 2014 Elsevier B.V. All rights reserved.

1. The fast lane to logic: highly reflective reasoners bypass belief processing People differ tremendously in their ability to reason logically. In particular, when a reasoning situation calls for a conclusion that is against one's own beliefs, few are able to make that conclusion. In the psychology of reasoning, this is known as the belief bias effect (Evans, Barston, & Pollard, 1983). Despite decades of research on the interplay between intuitive (heuristic, belief-based) and analytical (rational, logical) thinking, the processes underlying the belief bias effect are not yet clear. One possible reason for this is that research has not paid enough attention to possible differences in the processes underlying the performance of subgroups of reasoners. To this end, the present paper hopes to reveal what distinguishes those reasoners who most easily overcome belief bias from other reasoners. Conflicts between beliefs and logic can be explained in terms of dualprocess theories, which consider human reasoning to be the result of an interplay between two kinds of processes, namely contextual, intuitive processes, and decontextualized, analytical processes (Denes-Raj & Epstein, 1994; Evans, 2008; Stanovich & West, 2000). For simplicity, these may be termed Type 1 and Type 2 processes (Stanovich, 1999). ⁎ Corresponding author at: Institute of Behavioural Sciences, P. O. Box 9, 00014 University of Helsinki, Finland. Tel.: +358 50 4484154. E-mail address: annika.svedholm@helsinki.fi.

http://dx.doi.org/10.1016/j.actpsy.2014.11.008 0001-6918/© 2014 Elsevier B.V. All rights reserved.

According to the default-interventionist view of dual-processing, effortless Type 1 processes dominate thinking, and effortful Type 2 processing may intervene upon these when reasoning is leading to outputs that conflict with one's better judgment (De Neys, 2006; Evans, 2008; Evans & Curtis-Holmes, 2005; Stanovich & West, 2000). A general assumption has been that belief processing takes place intuitively, and that logical reasoning is a form of Type 2 processing (Stanovich, 1999). Thus, overcoming belief bias requires that belief processing is inhibited in favor of logical processing. Stanovich (1999, 2009b) has argued that choosing to take Type 2 processing into use is a distinct concept from the ability to do so, and has presented compelling evidence that this disposition predicts avoiding belief bias over and beyond the influence of general cognitive ability (Macpherson & Stanovich, 2007; Toplak, West, & Stanovich, 2011). Accordingly, a person with ample cognitive ability may nevertheless not reason logically, if he or she is not disposed to do so. The disposition to favor Type 2 processing has been termed cognitive reflection, and evidence suggests that it can be assessed using the Cognitive Reflection Test, a simple test that invites intuitively appealing responses that, upon reflection, turn out to be incorrect (Frederick, 2005). Correct responding requires that one withholds the heuristic response long enough to calculate the correct solution. In other words, the test seems to tap into an ability that is similar to the one needed to overcome belief bias. Therefore, to understand the kind of processing that leads to avoiding belief bias, success on this test may be a prime criterion by

70

A.M. Svedholm-Häkkinen / Acta Psychologica 154 (2015) 69–76

which to identify the individuals on whom to focus research. As discussed by Thompson, Prowse Turner, and Pennycook (2011), reasoning task responses alone are inadequate indicators of the quality of the underlying processing, because logically correct responses may be reached by guessing, and failing to give the correct response does not imply that the person has not attempted to reason logically. Recent developments in the reasoning field have emphasized that dual-process theories need to incorporate a separate process that determines when intuition is insufficient; after all, without analytical processing, how could one ever know when something is conflicting with it? Thus, several scholars have suggested the existence of a process that detects when beliefs and logic conflict, and determines whether proper Type 2 processing is to begin (Bonner & Newell, 2010; De Neys & Glumicic, 2008; Evans, 2009; Thompson et al., 2011). Abundant evidence suggests that this conflict detection functions outside of awareness and that it is effortless and fast. Comparing tasks with conflict to tasks without conflict, the presence of conflict has been found to affect numerous indicators, from autonomic nervous system reactions to improved recall of task details (review: De Neys, 2012). De Neys (2012) has put forward the idea that this conflict detection process entails “logical intuitions”, that is, a light Type 1 process which is able to quickly recognize whether a task follows simple rules of logic that the individual has internalized. An important finding is that these effects have been found in all study participants, including those who fail to follow through on logical reasoning. That is, even poorer reasoners do implicitly detect when they are giving illogical responses (De Neys & Glumicic, 2008). Based on the above research, current understanding has it that the reason why people so often fail to reason logically is not that they neglect to detect that their responses are illogical. Relating available evidence to a timeline of the reasoning process, De Neys and Bonnefon (2013) have has concluded that the evidence seems to exclude an early origin of belief bias. That is, those who exhibit belief bias do not differ from those who respond logically in the norms they strive for, or in the types of processes that they initiate. Instead, the authors suggest that everyone detects belief-logic conflicts and attempts to inhibit belief processing. However, the Type 1 response may still be “more strongly activated, salient, or appealing” (De Neys, 2012, p. 35) than the logical response, and engaging Type 2 processing to reach a logically correct response therefore requires effortful belief inhibition (De Neys, 2012). By this account, logical reasoning falters if people fail in this inhibition. Thus, individual differences in reasoning arise from differences in the effectiveness of belief inhibition, which occurs late in the reasoning process. Adding to this discussion, the present paper examines the possibility that for some advanced reasoners, the path to logical processing might be smoother. Specifically, the suggestion is that while most people's processing is a struggle between belief processing and logic, the people who are most inclined to favor Type 2 processing may be able to overcome or avoid the struggle at an earlier stage. This suggestion is based on the results of two studies on samples of exceptionally skilled reasoners. While the samples in reasoning studies typically consist of undergraduates participating for course credit, the present samples were self-selected through advertisements for reasoning experiments that offered no compensation. Consequently, the present studies came about to attract individuals who were older and who had more completed education than is typical in reasoning studies, and who were perhaps more motivated than average to volunteer for studies which they knew would involve effortful reasoning tasks. These samples thus allowed analysis of a population that has so far not received much attention in the literature. The experiments examined indicators of belief inhibition and found that the belief inhibition effects were absent from the reasoners with the strongest tendency for Type 2 thinking. Thus, the present paper suggests that for this minority of individuals, logical reasoning is less disrupted by belief inhibition than for other people.

2. Study 1 Study 1 replicated Experiment 1 from De Neys and Franssens (2009), in which syllogisms were followed by a lexical decision task requiring participants to rapidly distinguish words from nonwords. De Neys and Franssens found that poorer and better reasoners alike were slower to respond to words that were related to the topic of the preceding syllogism if it had involved a conflict between beliefs and logic than if it had not. These results were interpreted as evidence that all participants had detected when a conflict was present, and attempted to inhibit it. In the present study, this effect was compared between reasoners with higher and lower dispositions towards Type 2 thinking. Following the account that those who are most inclined to use Type 2 processing are less disrupted in their reasoning by belief–logic conflicts, the lexical inhibition effect was expected to be weaker or absent among participants with a high reflective disposition than among others (Hypothesis 1). 2.1. Method 2.1.1. Participants Fifty-six Finnish volunteers (45 females, mean age 27 years, age range 19–49) participated in the study. The participants were recruited through invitations distributed through several student mailing lists for a study on “the way people distinguish real words from nonwords when these are presented in conjunction with reasoning problems” and through opportunity sampling. The majority of the participants were university students. 2.1.2. Measures The participants were presented the eight syllogisms from De Neys and Franssens (2009, exp. 1) and another set of syllogisms on moral topics for a study not reported here. Half the syllogisms were logically valid and half were invalid, and half had believable conclusions and half had unbelievable conclusions, resulting in four syllogism types. The syllogisms were presented individually on a computer. Each premise was shown for 3 s, followed by a screen showing both premises and the conclusion, which stayed visible until the participant responded. The participants were given standard instructions to assume that the premises are true and to assess whether the conclusion follows logically from the premises (Evans et al., 1983). Each syllogism was followed by a lexical decision task. In this task, 24 strings of letters were presented individually and the participants were asked to classify whether the strings represented real words or nonwords. Out of the 24 strings, 12 were nonwords, 6 were target words related to the topic of the syllogism and 6 were words unrelated to the topic. The materials were translated into Finnish, and the lists of target and unrelated words were slightly modified after piloting showed that some of the associations between topics and words among the Finnish speakers differed from the associations in Dutch (e.g. while Dutch speakers associate the word ‘canal’ with boats, the Finnish word for ‘canal’ is the same as ‘channel’, and the pilot subjects associated it with television, not boats; thus, it was replaced with the word ‘oar’). The participants were instructed to take the time they needed on the syllogisms, and to respond as quickly as they could on the lexical decision task. The participants were given three practice sets of syllogisms and lexical decision tasks before the experiment began. The order of presentation was randomized for each participant. Responses were given using the buttons of a computer mouse. To assess the disposition favoring Type 2 processing, the participants completed the Cognitive Reflection Test (CRT; Frederick, 2005) at the end of the experimental session. The test consists of three questions such as “A bat and ball cost $1.10. The bat costs one dollar more than the ball. How much does the ball cost?” that require the participant to refrain from giving the heuristic response (10 cents) and to think the problem through before responding (correct response: 5 cents).

A.M. Svedholm-Häkkinen / Acta Psychologica 154 (2015) 69–76

2.1.3. Procedure The participants completed the experiment individually in a laboratory. Upon arrival, they signed an informed consent form. At the beginning of the experimental session, the participants were verbally given the instructions for the syllogisms and for the lexical decision tasks. Instructions for all tasks were also shown on the screen. The experiment lasted about 20 min. 2.2. Results and discussion In order to only include reaction times to correctly classified words in the analyses, 2.5% of the responses on the lexical decision task were excluded. Two reaction time outliers that deviated from the mean by more than 2.5 SD were replaced with this cutoff value. CRT scores were unavailable for two participants who failed to complete the questions. For group comparisons, the participants were median split into a Low-CRT group (n = 36), and a High-CRT group (n = 18). Because the CRT scale is discontinuous, the groups were uneven size (Low-CRT = participants who scored 0 or 1 correct; High-CRT = 2 or 3 correct). Table 1 presents descriptives for the two groups. As the table shows, accuracy on the syllogisms was very high in both the Low-CRT (77% of the conflict problems correct) and the High-CRT group (90% correct), and the groups did not differ significantly on accuracy or response times to the syllogisms. For comparison, in samples that are more typical in reasoning studies, the mean level of correct responses to comparable syllogisms tends to be around half (e.g., Toplak et al., 2011); on this exact same set of syllogisms in De Neys and Franssens (2009), Exp. 1, the mean was 53%. Likewise, CRT performance (mean = 1.09 out of 3 correct) was also higher than is typically observed (e.g., 0.7 in Toplak et al., 2011). Thus, the present comparison of “Low-CRT” and “High-CRT” can in fact be seen as a comparison between groups of more typical reasoners, and exceptionally high-performing reasoners. Fig. 1 shows reaction times on the lexical decision task, broken down by group. A 2 (word type: target or unrelated) ×2 (problem type: conflict or no-conflict) repeated-measures ANOVA with CRT group as a between-subjects variable was run on the reaction times. Full ANOVA results are presented in the Supplementary material. As in the study by De Neys and Franssens (2009), a significant main effect of word type, F(1,25) = 25.74, p b .001, η2p = .33, showed that the target words that had been positively primed by the syllogisms were recognized more easily than unrelated words. In the total sample, the main effect of the problem type and its interaction with word type were not significant, and even though visual inspection of Fig. 1 suggests that the High-CRT group responded more slowly overall, the main effect of group was not significant. The analysis showed no significant interactions with the CRT group. However, analyzing the groups separately shows that the critical interaction effect between word type and problem type, which would indicate that access to target words had been inhibited, approached significance in the Low-CRT group, F(1,35) = 3.27, p = .079, η2p = .09. Pairwise comparisons showed that in the Low-CRT group, the difference

Table 1 Descriptives for the Low-CRT and High-CRT groups, and tests of group differences.

Syllogism accuracy Conflict No-conflict Syllogism response times (ms) Conflict No-conflict Age

Low-CRT

High-CRT

M (SD)

M (SD)

F

p

3.08 (1.27) 3.67 (.72)

3.61 (.70) 3.94 (.24)

2.67 2.54

.108 .117

7491 (5807) 6202 (5053)

9563 (8192) 5835 (4934)

1.15 .06

.288 .801

27.1 (7.3)

26.2 (6.8)

.20

.658

71

between the conflict and no-conflict conditions was significant for the target words, t(35) = 2.46, p = .019, η2p = .15, but not for the unrelated words, t(35) = -.52, p = .607, η2p = .01. In the high-CRT group, the interaction between word type and problem type was not significant, F(1,17) = .01, p = .927, η2p b .01. Thus, solving conflict syllogisms seemed to result in slowed responses to target words among the LowCRT reasoners, but not among the High-CRT reasoners. These results indicate that the effect of conflict syllogisms on subsequent lexical access was different for subgroups of reasoners. Comparisons of the High-CRT participants with the Low-CRT participants support Hypothesis 1 and indicate that the reasoners with the strongest reflective disposition did not seem to exhibit signs of belief inhibition on this task. The present results support the notion that even when beliefs conflict with logic, these reasoners are able to reason effectively, not disrupted by the effect of belief inhibition. However, reasoning performance is influenced by multiple factors, including general cognitive ability. To be better able to tease apart the influence of the reflective disposition from the influence of ability, Study 2 included measures of both. Moreover, it is possible that the strategy indicated by the present data is specific to syllogistic reasoning and that dividing the sample by syllogism performance would reveal differences in strategy more clearly. Thus, Study 2 compared these three ways of grouping participants: by disposition, by general cognitive ability, and by syllogistic reasoning performance. In addition, a potential weakness of Study 1 was that confounds caused by the particular content used in the syllogisms cannot be ruled out (cf. Klauer & Singmann, 2013). Some of the particular topics used in the conflict syllogisms may have evoked different processing strategies than the topics used in the no-conflict syllogisms, and these differences in content, rather than the presence of a belief–logic conflict, might have been responsible for the differences found between the conditions. To avoid any content–condition confounds, Study 2 therefore counterbalanced syllogism content across conditions. 3. Study 2 Study 2 used a negative priming paradigm to reveal the residue of belief inhibition. In negative priming paradigms, a task that requires the person to inhibit one response in favor of another impairs performance on a subsequent task. Among 8–10-year-old children, presenting a valid-unbelievable (VU) syllogism before an invalid-unbelievable (IU) syllogism has been shown to impair performance on the later, presumably because the first syllogism caused the children to inhibit the processing of their intuitive beliefs (Moutier, Plagne-Cayeux, Melot, & Houdé, 2006). In the present study, this paradigm was applied to adults. Participants were presented with two syllogisms on the same topic: first a VU version, and then an IU version. In this constellation, correct responding requires that the participant first puts aside his or her real-world beliefs concerning the unbelievable claim, and accepts an unbelievable conclusion. The next moment, the participant must respond in line with those same beliefs, and reject that same conclusion. If performance in this latter phase is slower or contains more errors as compared with the same type of syllogism presented alone, we can conclude that the participant engaged in belief inhibition. Conversely, the present proposal, by which the reasoners with the highest reflective disposition are able to process the logic efficiently without disruption from belief inhibition, predicts that these reasoners will differ from this general pattern of results. These highly reflective reasoners are expected to show weaker signs of belief inhibition, that is, a weaker negative priming effect (Hypothesis 2) measured in terms of errors and response times, than other participants. To explore the role of general cognitive ability and the role of the ability to solve syllogisms in these processes, comparisons were also made based on general cognitive ability and syllogisms.

72

A.M. Svedholm-Häkkinen / Acta Psychologica 154 (2015) 69–76

Fig. 1. Reaction times on the lexical decision task. Error bars are standard errors of the mean.

3.1. Method 3.1.1. Participants Thirty-six Finnish individuals (23 females, mean age 30 years, age range 15–63) who had participated in an earlier study on personality and cognition and who had indicated an interest to participate in further laboratory studies, participated in the present study. The participants were invited through e-mail to participate in a study on beliefs and reasoning. The invitation message included a link to an online booking calendar where the participants could sign up for an experiment at their chosen time. The participants represented a wide range of occupations. Around half the sample (56%) had completed tertiary level degrees, and all had completed secondary level education. 3.1.2. Measures 3.1.2.1. Pairwise syllogisms. The participants were presented eight pairs of categorical syllogisms on everyday content topics. Each syllogism consisted of two premises and a conclusion. The two syllogisms in a pair differed only on the second premise; the first premise and the conclusion were always the same. Varying the second premise changed the logical status of the syllogism, so that one of the two syllogisms was always VU, and the other syllogism was IU. In four of the pairs, the valid (VU) version had the form “All A are B. All B are C. Therefore, all A are C.” and the invalid (IU) version used the second premise “Some B are C.” In the other four pairs, the valid form was “No A are B. All C are A. Therefore, no C are B” and the invalid version used the second premise “Some C are A.” In the negative priming condition, the conflict syllogism (VU) preceded the no-conflict syllogism (IU), and in the control condition, the order was the reverse. To avoid confounds between syllogism content and condition, the contents were counterbalanced across the conditions so that half of the participants received the syllogism pairs on four of the contents in the negative priming order, and the other four pairs in the control order. For the other half of the participants, the orders were switched.

As the pairwise presentation involved repeating the same content, it was likely that familiarity would facilitate shorter response times independently of the experimental condition. To test this assumption, a separate repeat control condition was included, in which four pairs of two different IU syllogisms were presented. One of the syllogisms was of the form “All A are B. Some B are C. Therefore, All A are C”. In the other syllogism, the premises were the same, but the conclusion was “Some A are C”. Presentation and responding were as in Study 1, except that responses were given using a response box marked with stickers (Yes and No). Following two practice items and feedback, the 12 syllogism pairs (4 in the negative priming condition, 4 in the control condition, 4 in the repeat control condition) were presented. The order of presentation of the pairs was randomized for each participant. Below is an example of a syllogism pair in the negative priming condition. In the first syllogism in the pair (a), there is a conflict because based on logic, the conclusion should be accepted as valid, but because the conclusion is unbelievable, people may want to reject it. The second syllogism (b) contains no conflict, as both the invalidity of the logic and the unbelievability suggest the conclusion be rejected. a). No fruits are sour. All lemons are fruits. Therefore, no lemons are sour. (VU). b). No fruits are sour. Some lemons are fruits. Therefore, no lemons are sour. (IU). 3.1.2.2. Cognitive reflection. The CRT was as in Study 1. The participants had completed the test before they arrived at the laboratory as part of the earlier study from which they were recruited. 3.1.2.3. General cognitive ability. General cognitive ability was assessed using Baddeley's (1968) Grammatical Reasoning Task. In the task, the participants were shown a combination of the letters A and B and a statement about their relation, and asked to indicate as quickly

Fig. 2. Response times on conflict and no-conflict syllogisms in the negative priming condition and in the control condition, and response times in the repeat control condition, for CRT groups. Error bars are standard errors of the mean.

A.M. Svedholm-Häkkinen / Acta Psychologica 154 (2015) 69–76

as possible whether the statement was true or false (e.g. “A does not precede B – AB”). Responses were given using a response box. The task ran for 3 min and the number of correct responses given in this time made up the score. The task has been shown to be a reliable and valid measure of the fluid intelligence component of general cognitive ability (Kyllonen & Christal, 1990). 3.1.2.4. Grouping syllogisms. To gain an independent measure of syllogism performance, the participants were given eight syllogisms. To avoid confounds between content and condition, the contents were counterbalanced across conditions in the following way. First, two valid-believable (VB) syllogism of the form “All A are B. All C are A. Therefore, All C are B” and two of the form “All A are B. No B are C. Therefore, No A are C” were formed using only believable premises, for example, “All mothers are women. All Grannies are mothers. Therefore, all grannies are women”. For each syllogism, an invalid-believable (IB) version was created by switching the order of the second premise and the conclusion. Next, four VU syllogisms were formed using the same two logical forms as above, by using an unbelievable statement as the second premise. For each of these syllogisms, an IU version was created by switching the second premise and the conclusion, as before. The result was two blocks of eight syllogisms. Half of the participants received one block, and half the other block. Mode of presentation and responding were the same as for the pairwise syllogisms. 3.1.3. Procedure The experiment was completed individually in a laboratory. Upon arrival, the participants gave informed consent, and were told that the study concerned how people solve thinking problems, that the problems were purposely devised to be difficult, and that they should not be stressed if they made mistakes. At the start of the experimental session, the participants completed tasks for another study. Between the tasks for the other study and the present study, the participants could take a break if they wanted. When the participant wanted to continue on to the present study, the instructions for the syllogisms were given on screen. The pairwise syllogisms were presented first, followed by the grouping syllogisms, and the Baddeley reasoning task. The session lasted about 30 min in total, and the present experiment lasted about 15 min. 3.2. Results and discussion The participants were divided into groups based on three different criteria. First, to investigate differences in reflective disposition, the participants were median split into a Low-CRT (n = 17) and a High-CRT (n = 19) group, as in Study 1. Second, to investigate group differences based on general cognitive ability, the sample was median split on Baddeley Reasoning Task scores into a Low-General Ability (n = 20) and a High-General Ability (n = 16) group. Third, to investigate differences related to performance on syllogisms, the sample was split into groups based on scores on the grouping syllogisms. The distribution was highly skewed (skewness = -2.51, SD of skewness = .39), with 26 out of 36 participants responding correctly to all eight grouping syllogisms. Thus, the split was made between these participants (High-Syllogism, n = 26) and all others (Low-Syllogism, n = 10). The groupings based on CRT and Grouping Syllogisms overlapped so that 69% of participants were in the same category (Low or High) on both criteria. Between the CRT and General Cognitive Ability groups, the proportion of overlap was 64%, and between General Cognitive Ability and Syllogism groups, 56%. As performance was again overall very high (CRT mean = 1.64 correct; grouping syllogisms 92% correct), the comparisons were effectively between exceptionally high performing participants, and more typically performing participants. Table 1 in the Supplementary material presents additional descriptives for the groups broken down by each grouping criterion.

73

Performance on the pairwise syllogisms was also very high (average 95% correct on the conflict syllogisms). Because of this ceiling effect, it was impossible to detect any negative priming effects on the numbers of correct responses. Therefore, negative priming effects were investigated by examining reaction times only. Fig. 2 shows the response times to conflict and no-conflict syllogisms in the negative priming and control conditions, as well as in the repeat control condition, broken down by CRT group. First, a repeatedmeasures ANOVA on the repeat control syllogisms established that, as expected, response times became shorter when a syllogism was repeated, F(1,34) = 9.39, p = .004, η2p = .22. There were no significant group differences in this effect or in the speed of responding to these syllogisms overall, using any of the three groupings, all analyses ns. Next, negative priming effects were investigated. A 2 (condition: negative priming, control) × 2 (syllogism type: conflict or no-conflict) repeated-measures ANOVA with CRT group as a between-subjects variable was run on the response times in the negative priming and control conditions. Full ANOVA results are presented in the Supplementary material. The analysis revealed a main effect of syllogism type, F(1,34) = 28.82, p b .001, η2p = .46, reflecting overall slower responses to no-conflict syllogisms than to conflict syllogisms. Visual inspection of Fig. 2 suggests that this effect was partly due to differences that were present even before any potential priming effects could have taken place. A pairwise comparison confirmed that responses to the first syllogism in a pair were slower in the control condition (control, no-conflict) than in the negative priming condition (negative priming, conflict), t(35) = -2.02, p = .052. Why there was a difference between these response times is unknown. Inspecting the experimental manipulations, there was no main effect of condition, but the critical interaction between the syllogism type and condition was significant, F(1,34) = 4.32, p = .045, η2p = .11, indicating a negative priming effect. A three-way interaction of syllogism type, condition, and group, F(1,34) = 4.35, p = .045, η2p = .11, indicated that this effect differed between the groups. To investigate the negative priming effects in each group, the analyses were run separately for the groups. In the Low-CRT group, the effects were as in the total sample. The critical interaction of the problem type and condition was significant, F(1,16) = 8.78, p = .009, η2p = .35. As Fig. 2 shows, responses always became faster when a syllogism was repeated within a pair. In line with the effect of repetition on the Repeat control syllogisms, this may be explained as a result of increased familiarity with the content and improved fluency of processing. However, in the negative priming condition, t(16) = 2.17, p = .045, repeating the same words did not facilitate fast responding to the same extent as it did in the control condition, t(16) = 3.65, p = .002. The difference in response times within a pair was significant in both conditions, but importantly, the size of this effect was larger in the control condition (η2p = .46) than in the negative priming condition (η2p = .23). In the High-CRT group, in contrast, the type–condition interaction was not significant,

Fig. 3. Response times on conflict and no-conflict syllogisms in the negative priming condition and in the control condition in General Cognitive Ability groups. Error bars are standard errors of the mean.

74

A.M. Svedholm-Häkkinen / Acta Psychologica 154 (2015) 69–76

F(1,18) b .01, p = .997, η2p b .01, indicating that for these participants, the effect of repetition was equal between the two conditions. Next, the ANOVA was repeated with General Cognitive Ability group as a between-subjects variable. Fig. 3 shows the response times broken down by General Cognitive Ability group. The pattern of results was highly similar as for the CRT groups. Full ANOVA results are presented in the Supplementary material. In this analysis, the three-way interaction of syllogism type, condition, and group fell short of significance, F(1,34) = 2.71, p = .109, η2p = .074, but when analyzing the groups separately, the results were again different for the groups. There was a type–condition interaction in the Low-General Ability group, F(1,19) = 5.51, p = .030, η2p = .23. Follow-up tests of this interaction showed, however, that the difference in response times between conflict and no-conflict syllogisms was significant in both the negative priming condition, t(19) = 3.11, p = .006, η2p = .34, and in the control condition, t(19) = 3.40, p = .003, η2p = .38. In the High-General Ability group, the type–condition interaction was not significant, F(1,15) = b .01, p = .952, η2p b .01. In sum, the pattern of results was largely similar when the participants were compared based on general cognitive ability as when compared on reflective disposition, but the data was less conclusive. Finally, the ANOVA was run with the division based on the Grouping Syllogisms as a between-subjects variable. Full ANOVA results and Supplementary Fig. 1, which shows the response times broken down by Syllogism group, are presented in the Supplementary material. In this analysis, neither the condition–type interaction, F(1,34) = 1.86, p = .181, η2p = .05, nor the three-way condition-type-group interaction, F(1,34) = .28, p = .602, η2p = .01, was significant. For congruence with the analyses on the other two groupings, the analyses were nevertheless repeated separately for the groups. The critical type–condition interaction was not significant in the Low-Syllogism group, F(1,9) = .32, p = .588, η2p = .03, but approached significance in the HighSyllogism group, F(1,25) = 3.38, p = .080, η2p = .12. The follow-up tests on the effects of repetition in the High-Syllogism group, however, revealed no differences between the two conditions. In sum, Study 2 showed that when syllogisms were repeated, the responses tended to become faster. For reasoners with a less reflective disposition, this effect was diminished when a conflict syllogism preceded a conflict-free syllogism, compared with when the order was the reverse. For the reasoners highest on cognitive reflection, the negative priming condition did not present a situation that differed from the control condition. These results may indicate that in line with Hypothesis 2, the processing of the conflict syllogisms was disrupted by belief inhibition in the other reasoners, but not in the reasoners with the most reflective style. In other words, the negative priming condition did not, in fact, seem to cause negative priming effects in this group of high performers. The above findings were highly similar when the high performers were identified with the help of their general cognitive ability. However, as the grouping criteria overlapped considerably, it is difficult to disentangle the effects of these factors. Whether the same results also apply to those who show top performance on syllogistic reasoning was more difficult to conclude, because accuracy on the syllogisms on which this comparison was made approached ceiling, and the division of participants into “Low” and “High” performers based on these syllogisms may be unreliable. Finally, a ceiling effect on the accuracy on the pairwise syllogisms limited the analyses to response times. Thus, it remains for future studies to investigate whether the presence of belief–logic conflicts may also influence the accuracy of reasoning in a negative priming paradigm for adults, as it does for children (Moutier et al., 2006). 4. General discussion Two studies investigated how people think when they must reason against their beliefs. The results indicated that reasoners with a high disposition for reflective thinking or high general cognitive ability do

not show the usual signs of belief inhibition in these situations, raising the question what they did differently. These findings are interesting in light of current dual-processing models of reasoning that view intuitive belief processing as primary (Evans, 2008). Influential models have suggested that logical processing is only possible when an individual has first automatically started belief processing, detected that it conflicts with rules of logic, and then effortfully inhibited the belief processing in favor of Type 2 logical processing (De Neys, 2006; Stanovich, 2009a; Stanovich & West, 2000). Accordingly, De Neys and Bonnefon (2013) suggest a late origin of individual differences in belief bias. In this view, belief bias emerges when people fail to inhibit their beliefs long enough to produce a Type 2 logical response to replace an initial, belief-based answer. An important implication is that people do not differ in their ultimate goals; the people who show belief bias do so because they fail to follow through on logical processing, not because they do not strive for normative logic at all. The results of the present paper suggest that individual differences in reasoning performance may, in a subset of cases, arise early. In Study 1, while other participants' recognition of words related to the belief content of a syllogism was delayed if that syllogism had called for belief inhibition, no such effect was found among the reasoners with the strongest reflective disposition. In other words, the most reflective reasoners seemed to be able to reach logically correct answers without disruption caused by belief inhibition. Likewise, in Study 2, other participants' reasoning on syllogisms was delayed when the previous problem negatively primed the beliefs relevant to that syllogism. However, the highly reflective and high-ability reasoners' performance was independent of any negative priming by belief content. These findings on these subgroups of good reasoners are difficult to reconcile with a model that only recognizes a late origin of individual differences, as such a model should predict the exact opposite of the present findings. If these good reasoners' advantage lay in their stronger belief inhibition, they could be expected to show the most delay on the lexical decision task and the strongest negative priming effects. However, if we allow the possibility that these reasoners were able to overcome or avoid the belief processing at an earlier stage, the present findings are in line with what we could expect. To explain this assumed early origin of individual differences, at least two different accounts are possible. One is that the highly reflective people started Type 1 processing like everyone else did, quickly discovered that beliefs and logic were in conflict, inhibited belief processing, and were exceptionally fast at recovering from this inhibition. In this view, the lack of negative priming effects and the lack of lexical decision slowing would reflect the superior ability of these people to regain access to information that had been temporarily inhibited. By the time that the lexical recognition task or the second syllogism in a pair was presented, these participants would no longer show any negative effects of the inhibition, and respond as fast as usual. This interpretation would be in line with research showing that cognitive reflection and reasoning performance are related to stronger executive functions, such as cognitive inhibition and set shifting, as well as larger working memory capacity (Toplak et al., 2011). The second possibility is that the highly reflective or highly able reasoners did not show any signs of belief inhibition because they had used a processing strategy that did not involve belief processing and belief inhibition. In this view, they took a different processing path immediately when encountering the problems, and focused on the logic of the problems while disregarding their content. In the current experiments, the participants were self-selected, and the mention of reasoning tasks in the recruitment messages likely attracted especially individuals who were confident in their reasoning skills and who therefore thought they might find participation a rewarding experience. Further, they were reminded at the beginning of the session that they would be presented difficult reasoning tasks. In other words, if a strategy of bypassing belief processing is possible, the conditions in these

A.M. Svedholm-Häkkinen / Acta Psychologica 154 (2015) 69–76

experiments would have cued and supported it. In terms of the model of a late origin of belief bias (De Neys & Bonnefon, 2013), this explanation would indicate that the model should be complemented by an alternative route. While the main route to logic likely involves first initiating and then inhibiting belief processing (on which some fail and others succeed), a subgroup of exceptionally skilled reasoners would be able to take “the fast lane”, a direct path from encountering a problem, to Type 2 logical analysis. Importantly, even this second proposal is compatible with the idea that all reasoners strive for logical reasoning, and that they differ in how well they reach that goal (De Neys, 2012). It is also in line with the view that a large proportion of differences in reasoning performance stem from inhibitory failures, which may occur late in the reasoning process (De Neys & Bonnefon, 2013; Lortie Forgues & Markovits, 2010). The present findings concerning the reasoners with the more typical profile in terms of reflective disposition and general cognitive ability in both Study 1 and Study 2 were exactly as could be expected based on the view that conflict detection leads to belief inhibition. The new idea presented here is that for the subgroup of highly reflective or highly able reasoners, logical proficiency may stem from choosing a different processing path immediately when encountering a reasoning problem. However, as the present experiments assessed the occurrence of inhibition indirectly after the event, the data does not help to distinguish between the exceptionally fast recovery account and the bypassing beliefs account. More studies are therefore needed to distinguish between the two explanations. Another proposal concerning early processing differences comes from Pennycook, Cheyne, Barr, Koehler, and Fugelsang (2013), who suggest that there are individual differences in the efficiency of Type 1 conflict detection or logical intuitions. The authors calculated how much longer participants spent on incorrect responses to conflict problems than to no-conflict problems, and took this increase in processing time as an indication of the extra processing involved in conflict detection. The size of this effect was found to be positively related to actively open-minded thinking and to cognitive reflection, and also, in two out of three experiments, to performance on conflict syllogisms. In these authors' view, everyone begins Type 1 processing, and for people with a stronger reflective disposition, this Type 1 process involves better detection of conflict, even when it does not carry through to improved reasoning performance. In light of the present findings, Pennycook and colleagues' findings likely reflect a different range of reasoners than the ones identified here. In that case, poorer reasoners may turn out to have a conflict detection which leads to belief inhibition that often fails, better reasoners have a better conflict detection which more often leads them to better reasoning, and a small subgroup is able to reason effectively without disruption, either because they recover from the disruptive effects of belief inhibition exceptionally quickly, or because they never engage in belief processing at all. It should be stressed that the novel findings presented here likely apply to a small subgroup of people. Even though the response time data indicates that the “Low” and “High” groups differed in their underlying processing, neither group showed hardly any belief bias in their final responses. Thus, compared with samples that are typically encountered in reasoning studies, the present “Low” groups were average to good reasoners, while the “High” groups were highly exceptional. Therefore, typical samples have likely not included many people with this profile, explaining why previous research has not noted it. In studies of conflict detection, which have documented that the presence of belief–logic conflict is implicitly detected and belief inhibition is initiated in case of such conflicts (review: De Neys, 2012), the results of any occasional reasoners like this have likely been averaged out. In the present data, the high reflective and high-ability groups overlapped to a large degree. Therefore, it was not possible to disentangle to what extent the group differences were driven by a reflective disposition and to what extent by general cognitive ability. As these two factors are known to contribute independently to reasoning performance

75

(Stanovich, 2009b), future studies should strive to investigate all possible combinations of these factors. For instance, would people with high general cognitive ability but low reflection show evidence of belief inhibition? It also remains for future research to investigate to what extent the ability to reason without disruption from belief inhibition identified in the present studies generalizes to tasks besides syllogisms. Previous research has shown that avoiding belief bias on one type of task is not always related to avoiding it on another type of task (Thompson & Evans, 2012). Further, the highly reflective reasoners that were here identified by their scores on the CRT might not reason as smoothly on tasks with very different task demands. Whereas many accounts have interpreted the CRT as a domain-general measure of cognitive reflection and avoiding cognitive miserliness (Toplak et al., 2011), other studies suggest that it measures more domain-specific abilities, particularly mathematical skill (Stupple, Gale, & Richmond, 2013; Welsh, Burns, & Delfabbro, 2013). Welsh and colleagues (2013) found that the CRT only predicted avoiding bias on those tasks where an objectively correct answer existed and could be reached through counting. As the present studies only included syllogisms, which have objectively correct logical answers, it is important to replicate the present findings using other types of reasoning tasks and other measures of a reflective disposition. For example, future studies could test whether some people seem to be unaffected by belief inhibition effects even on more difficult tasks than the ones used here. As De Neys (2012, 2014) has stressed, only a simple logic is likely to be internalized to form logical intuitions. Therefore, testing whether the present findings generalize beyond tasks in which logical intuitions are an option may provide some insight. Moreover, the relationships of other cognitive styles to the strategies under discussion are not yet clear. For example, it would be interesting to investigate whether individuals with a high tendency for actively open-minded thinking show signs of belief inhibition disrupting their reasoning. 5. Conclusion Because decades of research testify that people tend to be biased by their beliefs, reasoning against one's own beliefs has been thought of as a difficult task that requires much mental effort. Here, I suggest that for some people, the task might not be as difficult as it is for others. The present studies pointed to the conclusion that when problems pit logic against beliefs, individual differences exist in the degree to which this conflict disrupts people's processing. For people with a highly reflective cognitive style, processing seems to continue as if no conflict had occurred. An interesting question for future studies is to investigate whether these individuals experience conflict problems as requiring effort, either by self-report or by measures of cognitive effort such as pupil dilation (Laeng, Sirois, & Gredebäck, 2012). By current evidence, it is premature to establish the question of what makes a particularly good reasoner. The best reasoners might turn out to be better than others at detecting conflicts, at inhibiting beliefs in favor of logic, at disregarding belief processing altogether, or at solving the specific types of problems used in reasoning studies. Whichever explanation receives the most support, the present results bring attention to individual differences in the processes underlying reasoning performance, and underscore the importance of investigating subgroups of participants separately when trying to understand how reasoning works. Acknowledgments Thank you to Helena Haikonen, Lauri Kangas, Valentina Kieseppä, Riikka Koljonen, Jukka Napola, Aino Niemi, Joel Sammallahti, Marie Sund, Essi Suomela, and Liisa Tulensalo for help in collecting data. Thank you to Marjaana Lindeman and Ville Rantalainen for helpful discussions and comments, and to an anonymous reviewer for helpful comments on an earlier draft of this manuscript.

76

A.M. Svedholm-Häkkinen / Acta Psychologica 154 (2015) 69–76

The preparation of this manuscript was supported by the Research Funds of the Academy of Finland (No. 265518). Appendix A. Supplementary data Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.actpsy.2014.11.008. References Baddeley, A.D. (1968). A 3-min reasoning test based on grammatical transformation. Psychonomic Science, 10, 341–342. http://dx.doi.org/10.3758/BF03331551. Bonner, C., & Newell, B.R. (2010). In conflict with ourselves? An investigation of heuristic and analytic processes in decision making. Memory & Cognition, 38, 186–196. http:// dx.doi.org/10.3758/MC.38.2.186. De Neys, W. (2006). Dual processing in reasoning. Two systems but one reasoner. Psychological Science, 17, 428–433. http://dx.doi.org/10.1111/j.1467-9280.2006. 01723.x. De Neys, W. (2012). Bias and conflict: A case for logical intuitions. Perspectives on Psychological Science, 7, 28–38. http://dx.doi.org/10.1177/1745691611429354. De Neys, W. (2014). Conflict detection, dual processes, and logical intuitions: Some clarifications. Thinking and Reasoning, 20, 169–187. http://dx.doi.org/10.1080/ 1354678.2013.854725. De Neys, W., & Bonnefon, J. -F. (2013). The ‘whys’ and ‘whens’ of individual differences in thinking biases. Trends in Cognitive Sciences, 17, 172–178. http://dx.doi.org/10.1016/j. tics.2013.02.001. De Neys, W., & Franssens, S. (2009). Belief inhibition during thinking: Not always winning but at least taking part. Cognition, 113, 45–61. http://dx.doi.org/10.1016/j.cognition. 2009.07.009. De Neys, W., & Glumicic, T. (2008). Conflict monitoring in dual process theories of thinking. Cognition, 106, 1248–1299. http://dx.doi.org/10.1016/j.cognition.2007.06.002. Denes-Raj, V., & Epstein, S. (1994). Conflict between intuitive and rational processing: When people behave against their better judgment. Journal of Personality and Social Psychology, 66, 819–829. http://dx.doi.org/10.1037/0022-3514.66.5.819. Evans, J. St. B.T. (2009). How many dual-process theories do we need? One, two, or many? In J. St. B. T. Evans, & K. Frankish (Eds.), In Two Minds. Dual Processes and Beyond (pp. 33–54). Oxford & NY: Oxford University Press. Evans, J. St. B.T. (2008). Dual-processing accounts of reasoning, judgment, and social cognition. Annual Review of Psychology, 59, 255–278. http://dx.doi.org/10.1146/ annurev.psych.59.103006.093629. Evans, J. St. B.T., Barston, J., & Pollard, P. (1983). On the conflict between logic and belief in syllogistic reasoning. Memory and Cognition, 11, 295–306. http://dx.doi.org/10.3758/ BF03196976. Evans, J. St. B.T., & Curtis-Holmes, J. (2005). Rapid responding increases belief bias: Evidence for the dual-process theory of reasoning. Thinking and Reasoning, 11, 382–389. http://dx.doi.org/10.1080/13546780542000005.

Frederick, S. (2005). Cognitive reflection and decision making. Journal of Economic Perspectives, 19, 25–42. Klauer, K.C., & Singmann, H. (2013). Does logic feel good? Testing for intuitive detection of logicality in syllogistic reasoning. Journal of Experimental Psychology: Learning, Memory, and Cognition., 39, 1265–1273. http://dx.doi.org/10.1037/a0030530. Kyllonen, P.C., & Christal, R.E. (1990). Reasoning ability is (little more than) workingmemory capacity?! Intelligence, 14, 389–433. http://dx.doi.org/10.1016/S01602896(05)80012-1. Laeng, B., Sirois, S., & Gredebäck, G. (2012). Pupillometry: A window to the preconscious? Perspectives on Psychological Science, 7, 18–27. http://dx.doi.org/10.1177/ 1745691611427305. Lortie Forgues, H., & Markovits, H. (2010). Conditional reasoning under time constraint: Information retrieval and inhibition. Thinking and Reasoning, 16, 221–232. http://dx. doi.org/10.1080/13546783.2010.503606. Macpherson, R., & Stanovich, K.E. (2007). Cognitive ability, thinking dispositions, and instructional set as predictors of critical thinking. Learning and Individual Differences, 17, 115–127. http://dx.doi.org/10.1016/j.lindif.2007.05.003. Moutier, S., Plagne-Cayeux, S., Melot, A. -M., & Houdé, O. (2006). Syllogistic reasoning and belief-bias inhibition in school children: Evidence from a negative priming paradigm. Developmental Science, 9, 166–172. http://dx.doi.org/10.1111/j.1467-7687.2006. 00476.x. Pennycook, G., Cheyne, J.A., Barr, N., Koehler, D.J., & Fugelsang, J.A. (2013). Cognitive style and religiosity: The role of conflict detection. Memory & Cognition. http://dx.doi.org/ 10.3758/s13421-013-0340-7. Stanovich, K.E. (1999). Who is Rational? Studies of Individual Differences in Reasoning. Mahwah, NJ: Lawrence Erlbaum Associates. Stanovich, K.E. (2009a). What Intelligence Tests Miss. The Psychology of Rational Thought. New Haven & London: Yale University press. Stanovich, K.E. (2009b). Distinguishing the reflective, algorithmic, and autonomous minds: Is it time for a tri-process theory? In J. St. B. T. Evans, & K. Frankish (Eds.), In Two Minds. Dual Processes and Beyond (pp. 55–88). Oxford & NY: Oxford University Press. Stanovich, K.E., & West, R.F. (2000). Individual differences in reasoning: Implications for the rationality debate? Behavioral and Brain Sciences, 23, 645–665. Stupple, E.J.N., Gale, M., & Richmond, C.R. (2013, July). Working memory, cognitive miserliness and logic as predictors of performance on the Cognitive Reflection Test. Paper presented at the 35th Annual Conference of the Cognitive Science Society, Berlin. Thompson, V.A., & Evans, J. St. B. T. (2012). Belief bias in informal reasoning. Thinking and Reasoning, 18, 278–310. http://dx.doi.org/10.1080/13546783.2012.670752. Thompson, V.A., Prowse Turner, J.A., & Pennycook, G. (2011). Intuition, reason, and metacognition. Cognitive Psychology, 63, 107–140. http://dx.doi.org/10.1016/j.cogpsych. 2011.06.001. Toplak, M., West, R.F., & Stanovich, K.E. (2011). The Cognitive Reflection Test as a predictor of performance on heuristics-and-biases tasks. Memory & Cognition, 39, 1275–1289. http://dx.doi.org/10.3758/s13421-011-0104-1. Welsh, M.B., Burns, N.R., & Delfabbro, P.H. (2013, July). The Cognitive Reflection Test: How much more than numerical ability? Paper presented at the 35th Annual Conference of the Cognitive Science Society, Berlin.