Evidence for conflict detection from the self-reported conflict measure

ABSTRACT Human thinking is typically biased. A central question in dual process theories is whether people detect conflicts between heuristic and logical information. In the present study, we explored this issue. Participants were presented with conflict and non-conflict base-rate neglect problems and syllogism problems, followed by self-reported conflict measures determining the extent to which they considered alternative solutions after resolving each problem. Although the participants generally could not correctly answer the conflict problems, the results showed that their self-reported conflict measures in the incorrect conflict items were lower than those in the correct non-conflict items, indicating that the participants could recognise the conflict between heuristic and logical information. The implications of the ongoing debate on conflict detection are also discussed.


Introduction
Decades of research on reasoning and decisionmaking have shown that even educated people tend to violate elementary logical principles (Kahneman, 2011). In general, people are inclined to draw conclusions based on quick and effortless heuristic thinking. Despite being beneficial in certain cases, heuristic thinking often results in responses that disregard normative principles and contort thought processes (Evans & Stanovich, 2013;Kahneman, 2011).
To illustrate, consider the problem of a base-rate neglect problem introduced by Tversky and Kahneman (1974): It is given that an event with a population of 1000 comprises 995 females and 5 males. From this event, a person is randomly selected and described as a surgeon. The question arises as to whether this person is more likely to be male or female.
According to the base-rates, one should respond that the person is female, given the disproportionate number of females to males. However, many people automatically assume that a person is male based on their heuristic beliefs triggered by the description. Several studies have shown that people frequently neglect the base-rate principle and instead give the heuristic, stereotypical response (Kahneman, 2011).
The base-rate neglect problem is explained by dual-process theories, which propose that human reasoning and decision-making involve two qualitatively different thinking processes (Evans & Stanovich, 2013;Kahneman, 2011;Sloman, 1996). Type 1 processing is intuitive, heuristic, and autonomous, while Type 2 processing is deliberate, analytic, and does require working memory resources. The traditional default-interventionist model is one of the most prominent dual-process theories in reasoning and decision-making (Evans, 2003;Evans & Stanovich, 2013).
According to the default-interventionist model, Type 1 processing initially produces an erroneous, heuristic response based on a stereotypical description. Consequently, individuals must apply Type 2 processing to override their heuristics and provide a normative response based on base-rate information. People exhibit biased responses because the conflicts between heuristic and rational thinking cannot be detected (Evans & Stanovich, 2013). In contrast, the parallel model of the dual-process theories assumes that both Type 1 and Type 2 processing are activated simultaneously and operate parallel. Researchers who support the parallel model argue that the detection mechanism works well. People are biased not due to a failure to detect conflict but rather to a failure to discard heuristic responses (Denes-Raj & Epstein, 1994;Handley & Trippas, 2015;Sloman, 1996).
Researchers have used the conflict-detection paradigm to address the controversy between the two detection perspectives (De Neys, 2012). In this paradigm, participants were provided with conflict and non-conflict problems. The key difference between the two types of problems is that in conflict problems, heuristic responses conflict with logical consideration, whereas in non-conflict problems, they align. Table 1 illustrates the two types of problems.
Numerous studies on conflict detection have concluded that while working on conflict problems, participants took longer to respond than on nonconflict problems (Bonner & Newell, 2010;De Neys & Glumicic, 2008;Janssen et al., 2021), felt less confident in their responses (De Neys et al., 2011;Mevel et al., 2014;Thompson & Johnson, 2014), and had higher activation rates in brain regions that characterise conflict detection (Mevel et al., 2019;Simon et al., 2015). Although the empirical work mentioned above confirm the successful conflict detection, some researchers have not been able to replicate this finding (Aczel et al., 2015;Singmann et al., 2014;Travers et al., 2016). Even researchers who agree on the success of conflict detection suggest that it requires additional investigation (De Neys, 2012, 2014. Recently, in response to differing views on appeal, Mata (2019) used a novel approach to explore the validity of conflict detection. Selfreported conflict measures were used to assess whether the participants recognised the presence of conflict. The material they used was the batand-ball problem, which is a classic example of Frederick's (2005) Cognitive Reflection Test (CRT) that requires individuals to solve a seemingly simple mathematical equation (Kahneman & Frederick, 2002). After completing the conflict and nonconflict versions of the bat-and-ball problem, participants were asked to answer questions measuring conflict, such as how much they wanted to give alternative answers that differed from the one they ultimately chose. The results showed that self-reported conflict measures were higher for incorrect conflict trials than for correct non-conflict trials, thus supporting the notion that individuals can accurately detect conflicts (Mata, 2019).
However, it should be noted that self-reported conflict measures have only been utilised to explore the phenomenon of conflict detection within the bat-and-ball problem. Conflict detection is a widely studied phenomenon observed in various experimental materials such as the baserate neglect problem (De Neys & Glumicic, 2008), syllogisms (De Neys et al., 2010), conjunction fallacy (De Neys et al., 2011), and ratio bias (Bonner & Newell, 2010;Mevel et al., 2014). To better elucidate the debate between the two perspectives on conflict detection and further enrich research on conflict detection, self-reported conflict measures should also be extended to other problems.
The present study tested conflict detection using self-reported measures for base-rate neglect and syllogistic reasoning problems. Base-rate neglect tasks deal with the concept of probabilities and the influence of stereotypes on decision-making (Tversky & Kahneman, 1974). Syllogisms, on the other hand, are a form of logical reasoning task that necessitates the use of deductive reasoning (Evans et al., 1983). These two tasks are frequently used in the field of reasoning and decisionmaking. Therefore, we believe that using these problems as experimental stimuli will allow us to gain valuable insights into the cognitive processes involved in reasoning and decision-making.
Participants were presented with both conflict and non-conflict problems. Heuristic responses and logical principles are at odds in conflict problems, while in non-conflict problems, heuristic considerations and logical thinking are aligned. The default-interventionist model posits that individuals have biased responses because they fail to detect conflict. If the default-interventionist model is valid, there should be no difference in self-reported conflict scores between incorrect conflict trials and correct non-conflict trials (Evans & Stanovich, 2013;Kahneman, 2011). Conversely, the parallel model suggests that reasoners have biased responses because they cannot override their heuristic responses after detecting a conflict. Therefore, if the parallel model is valid, self-reported conflict scores of incorrect conflict trials and correct nonconflict trials would differ (Denes-Raj & Epstein, 1994;Handley & Trippas, 2015;Sloman, 1996). Selfreported conflict measures were assessed by asking participants to what extent they would choose an alternative option after solving each problem.

Experiment 1
Experiment 1 aimed to use a self-reported conflict measure to examine whether participants could detect conflicts in base-rate problems. Based on previous studies, we hypothesised that participants could detect conflict more effectively with a larger self-reported conflict measure for incorrect conflict trials than for correct non-conflict trials (Mata, 2019).

Participants
We used the G*Power 3.1 software to determine the minimum sample size (Faul et al., 2009). The parameters were as follows: the effect size (d) was set to 0.25, the significance level (α) was set to 0.05, and the power (1-β) was set to 0.80. The analysis revealed that a minimum of 34 participants was required. A total of 68 university students (32 females, mean ± SD = 19.25 ± 1.14 years) were recruited to participate in the experiment. At the end of the experiment, we gave each participant a pen and notebook as tokens of appreciation.

Materials and procedure
Participants solved 16 base-rate neglect problems based on Pennycook et al. (2015). The base-rate neglect problems were translated from English into Chinese and modified to ensure that they were understandable for the Chinese-speaking participants. Specifically, we modified the names and scenarios of the problems to be more familiar to the Chinese-speaking population. Each problem contained a statement about the identity of two characters (e.g. "This study contains 16-year-old teenagers and middle school teachers".), base-rate information (e.g. "There were 995 16-year-old teenagers and 5 middle school teachers".), and a description of a character trait that triggers the stereotypical association (e.g. "Li Wei behaves immaturely".). Participants were asked to indicate which group the character most likely belonged to. Each of the 16 base-rate neglect problems was equivalently modified to either a conflict or nonconflict problem by matching the base-rate probability of the two characters. For instance, the left side of Table 1 represents a conflict problem, whereas the right side represents a non-conflict problem through a base-rate switch. Half of the problems were non-conflict, and the other half were conflict problems. In non-conflict problems, the base-rate and stereotypical information trigger the same response, whereas in conflict problems, they elicit opposite responses.
Each problem started with a brief fixation crosspresentation for 1000 ms, after which a base-rate problem was displayed. The participants could select their responses by pressing the appropriate button. After answering the questions, the participants were asked to press a button to obtain their self-reported conflict measure by rating from 1 (not at all) to 7 (very much) on "to what extent would you like to select another option after solving the question". We used a single question rated on a scale from 1 to 7 to simplify the measure and reduce participant burden. Neither the base-rate task nor the conflict-detection test had a time limit, and the 16 base-rate problems were presented in random order.

Accuracy
Consistent with previous studies, the base-rate information was typically overlooked, resulting in an accuracy of 19% (SE = 25%) for the conflict problem. Moreover, as expected, the vast majority of conflict items were resolved correctly, with an accuracy of 97% (SE = 6%), t(67) = 26.12, p < .001, d = 3.17. Table 2 lists the accuracy for each problem type in Experiment 1.

Self-reported conflict measure
The main purpose was to examine whether the participants detected a conflict between heuristic responses and logical considerations using the self-reported conflict measure. The results showed that the self-reported conflict measure for incorrect conflict trials (2.71, SE = 0.93) was greater than that for correct non-conflict trials (2.24, SE = 0.78), t(67) = 4.20, p < .001, d = 0.51.

Discussion
The findings revealed that, for conflict problems, participants often disregarded the base rate information, leading to inaccurate responses, which is consistent with previous research findings (Bago et al., 2018;De Neys et al., 2011). However, consistent with prior research, with respect to reported conflict, participants showed a higher level of conflict detection for problems involving conflict than for those without conflict, demonstrating their ability to detect conflicts (De Neys & Glumicic, 2008;Frey et al., 2018).
However, the generalizability of the experimental results needs to be further tested because only one experimental material, the base-rate neglect problem, was used in this experiment.

Experiment 2
In Experiment 2, we examined the generality of the previous findings. Specifically, participants were required to answer conflict and non-conflict syllogisms and submit a self-reported measure of conflict similar to that in Experiment 1. We hypothesised that self-reported conflict measures for incorrect conflict trials would exhibit significantly higher scores than those for correct non-conflict trials, indicating a similar pattern to that observed in Experiment 1.

Participants
As in Experiment 1, the power analysis for Experiment 2 indicated we needed a minimum of 34 participants. A total of 64 university students (38 females, mean ± SD = 19.03 ± 1.02 years) participated in the experiment. As a token of appreciation at the end of the study, the participants were presented with a pen and notebook.

Materials and procedure
Each participant was presented with a total of 32 syllogistic reasoning problems. We used conceptual terms taken from Trippas et al. (2014) to construct all problems. Each problem contained three parts: a major premise (e.g. "All parrots are A".), a minor premise (e.g. "No A are metals".), and a conclusion (e.g. "Therefore, no parrots are metals".). The participants' objective was to determine whether the conclusion followed the logically presented premises. We constructed all the problems, which can be solved by developing a single mental model (Johnson-Laird, 2001). Additionally, we used the nonsensical character "A" as the middle term to prevent any believability bias in the premises (Ding et al., 2020).
There are two types of conflict problems (validbelievable and invalid-unbelievable) and two types of non-conflict problems (valid-unbelievable and invalid-believable). In conflict problems, the believability and validity of conclusions lead to different responses, whereas in non-conflict problems, they cue the same response. To prevent the influence of the believability of arguments, we randomly assigned the problem content to one of the four problem types. The distinct types of problems are listed in Table 2.
The experimental procedure is the same as in Experiment 1.

Self-reported conflict measure
The self-reported conflict measures were consistent with those in Experiment 1. Participants' conflict measures were significantly higher for incorrect Therefore, some parrots are birds. Unbelievable All parrots are A.
All parrots are A. No A are birds.
No A are metals. Therefore, no parrots are birds.
Therefore, some parrots are metals.

Discussion
Experiment 2 extended the generality of the findings of Experiment 1 by using syllogistic reasoning problems. The findings of this experiment were consistent with those of Experiment 1. In terms of correctness, participants were usually unable to answer the conflict version of syllogistic reasoning problems correctly, which is consistent with previous research (De Neys & Franssens, 2009;Frey et al., 2018). Additionally, our findings support our hypothesis that participants' self-reported scores are higher for incorrect conflict trials than for correct non-conflict trials, indicating that they can successfully detect a conflict. Our conflict-detection results are also congruent with previous studies (De Neys et al., 2010;Frey et al., 2018).

General discussion
This study aimed to investigate whether individuals could detect conflicts between heuristic and logical information. To this end, we presented base-rate neglect problems and syllogistic reasoning problems to participants, along with a unique selfreported conflict measure that asked participants to rate the extent to which they desired to choose another option after solving the problems. The results confirmed our hypothesis that while individuals might fail to answer correctly, they could recognise and detect conflicts between the two types of responses. Specifically, self-reported conflict scores were higher for incorrect conflict trials than for correct non-conflict trials. The distinction between conflict and non-conflict problems lies in the alignment of heuristics and logic. In non-conflict problems, both heuristics and logic converge toward the same answers, whereas in conflict problems, they lead to different answers. Therefore, the findings suggested that individuals could observe conflicts even if they could not always address them appropriately.
It is important to note that we are not primarily interested in exploring whether people habitually violate logical rules. This aligns with earlier research indicating that people commonly often disregard logical principles and resort to heuristic responses because logical responses consume cognitive resources, whereas heuristic responses are effective in conserving cognitive resources (Evans & Curtis-Holmes, 2005;Kahneman, 2011). The aim of our study, however, was to demonstrate that, despite their inability to solve the problem, participants were not entirely ignorant of the conflict between their heuristic and logical responses. In other words, they recognised that their stereotypical responses were not entirely warranted.
Our findings provide evidence of a more optimistic view of human reasoning and decision-making. Previous studies have shown that even educated adults often make errors in classical reasoning and decision-making tasks (Stanovich & West, 2000). Consequently, some researchers have suggested that humans rely on heuristic and irrational thinking rather than on logical thinking (Frederick, 2005;Kahneman, 2011). However, our data show something different. Despite people's inability to provide accurate responses, they still consider normative principles when solving reasoning and decisionmaking problems.
Why do people successfully detect conflict between heuristic and logical responses? One explanation for conflict detection is that people conceive "logical intuitions" (De Neys, 2012, 2014De Neys & Pennycook, 2019). According to the default intervention model of dual-process theories, heuristic associations are processed through the Type 1 process, while the processing of logical and probabilistic normative principles relies on the Type 2 process. On the other hand, the logical intuition account proposes that both heuristic and logical responses can rely on the Type 1 process. Therefore, conflicts arise between two Type 1 processes, rather than between Type 1 and Type 2 processes, from this perspective. However, the logical intuition account is controversial and may be the subject of future research (Ghasemi et al., 2022;Hayes et al., 2020).
The present study provides an opportunity to compare different interpretations of conflict detection. The default-intervention model of dualprocess theories suggests that people make incorrect responses because they cannot detect the conflict between the heuristic and correct responses; errors are made because of a failure to detect the conflict (Evans & Stanovich, 2013). Alternatively, the parallel model of dual-process theories suggests that individuals can detect conflicts between heuristic and logical thinking. However, heuristic responses are so attractive that they cannot be overridden. People are biased due to suppression failure (Denes-Raj & Epstein, 1994;Sloman, 1996). Our findings corroborate the concept of failure suppression (Bonner & Newell, 2010;De Neys & Glumicic, 2008;De Neys et al., 2011;Janssen et al., 2021;Mevel et al., 2019;Simon et al., 2015;Thompson & Johnson, 2014). However, research on conflict detection remains controversial, and further investigation is necessary (Aczel et al., 2015;Singmann et al., 2014;Travers et al., 2016).
The practical applications of this study are also highlighted. Correct reasoning is essential in everyday life. Researchers have used various intervention methods to reduce biased responses (Babai et al., 2015). However, these interventions are not as effective as they may have been (Hornik et al., 2008;Reyna, 2013). This can be attributed to the idea that people are biased because they do not consider logical answers when making decisions. Consequently, researchers usually provide logicrelated training to participants in intervention studies (Kahneman et al., 1982). Nevertheless, this study indicates that participants take logical factors (such as base rate and syllogistic logic) into account during the reasoning process but struggle to suppress highly attractive heuristic responses (such as descriptive information and beliefs). Thus, effective training that targets inhibitory skills may be more likely to improve the effectiveness of the intervention and thus improve people's reasoning and decision-making performance (Houdé, 2007).
This study used a self-reported conflict measure to evaluate the ability to detect conflicts between heuristic responses and normative considerations, which demonstrates that individuals can detect conflicts. Future research should take into account individual differences and the generalizability of self-reported conflict measures. Regarding individual differences, it is necessary to examine differences in cognitive abilities (Stanovich & West, 2000), thinking disposition (Stanovich & West, 2007), and capabilities in numeracy (Cokely et al., 2012). Additionally, self-reported conflict measurements can be used to test the dual-process assumption underlying prosocial behaviour (Rand et al., 2012) and moral reasoning (Greene, 2015), extending beyond reasoning and decision-making.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Data availability statement
The data supporting this study's findings are openly available in OSF at DOI 10.17605/OSF.IO/GZTM5.