What makes us think? A three-stage dual-process model of analytic engagement

Highlights

• We develop a three-stage model to explain what causes analytic thinking to occur.

• The model distinguishes conflict detection and decoupling as early and late sources.

• Bias can be caused by failures at either early or late stages.

• Our brains sometimes fail to detect bias and sometimes fail to inhibit bias.

• This work is a falsifiable and clearly defined application of dual-process theory.

Abstract

The distinction between intuitive and analytic thinking is common in psychology. However, while often being quite clear on the characteristics of the two processes (‘Type 1’ processes are fast, autonomous, intuitive, etc. and ‘Type 2’ processes are slow, deliberative, analytic, etc.), dual-process theorists have been heavily criticized for being unclear on the factors that determine when an individual will think analytically or rely on their intuition. We address this issue by introducing a three-stage model that elucidates the bottom-up factors that cause individuals to engage Type 2 processing. According to the model, multiple Type 1 processes may be cued by a stimulus (Stage 1), leading to the potential for conflict detection (Stage 2). If successful, conflict detection leads to Type 2 processing (Stage 3), which may take the form of rationalization (i.e., the Type 1 output is verified post hoc) or decoupling (i.e., the Type 1 output is falsified). We tested key aspects of the model using a novel base-rate task where stereotypes and base-rate probabilities cued the same (non-conflict problems) or different (conflict problems) responses about group membership. Our results support two key predictions derived from the model: (1) conflict detection and decoupling are dissociable sources of Type 2 processing and (2) conflict detection sometimes fails. We argue that considering the potential stages of reasoning allows us to distinguish early (conflict detection) and late (decoupling) sources of analytic thought. Errors may occur at both stages and, as a consequence, bias arises from both conflict monitoring and decoupling failures.

Introduction

A few months after the 2003 invasion of Iraq, current Vice President and then Senator Joe Biden asked President George W. Bush how he can be so sure that the United States was on the right course. Bush responded by putting his hand on the Senator’s shoulder and saying “my instincts” (Suskind, 2004). Bush’s faith in his gut feelings in the face of conflicting or contradictory evidence is, not incidentally, reminiscent of comedian Stephen Colbert’s concept of “truthiness”.¹ There appears to be a great deal of truth to the idea of truthiness and, in fact, it has been known for decades, dating back to Kahneman and Tversky’s heuristics and biases research program, that humans often rely on intuition when making decisions (Tversky & Kahneman, 1974; for a recent overview, see Kahneman, 2011).

An additional point that is rarely emphasized, however, is that gut feelings do not always predominate. Some individuals are less likely to “go with their gut” when reasoning (Stanovich and West, 1998, Stanovich and West, 2000) and problems that cue conflicting response outputs have been shown to lead to deliberative reasoning (De Neys and Glumicic, 2008, De Neys et al., 2008). Investigations of the factors that undermine intuitive decision making may lead to interventions which could be used to avoid future errors; or, in other words, to maximize “truth” and minimize “truthiness”. To that end, it has been suggested that one of psychological science’s most pressing goals should be to “give debiasing away” to the general public (Lilienfeld, Ammirati, & Landfield, 2009).

We argue that basic cognitive research that elucidates how debiasing happens in the absence of explicit top-down intervention could be a fruitful source of practical benefit in the public sphere. In the current work, we attempt to elucidate the cognitive processes that guard against reasoning failures by introducing a three-stage dual-process model of analytic engagement, along with 4 experiments that test predictions generated from the model. Our goal is to integrate perspectives on bias and irrationality that have previously been considered antithetical by breaking the reasoning process into stages and components. We argue that a consideration of the bottom-up sources of analytic thinking offers a new perspective that leads to novel predictions.

Human reasoning and decision-making is thought to involve two distinct types of processes (for reviews, see Evans, 2008, Evans, 2010a, Evans and Stanovich, 2013a, Frankish and Evans, 2009, Sloman, 1996, Stanovich, 2004): Type 1 processes that are intuitive, fast, autonomous, and high capacity; and Type 2 processes that are reflective, slow, and resource demanding. Type 1 processes are thought to provide default outputs that may be acted upon as explicit representations manipulated in working memory via Type 2 processing (Evans and Stanovich, 2013a, Thompson, 2013). However, the question of what leads someone to engage deliberate and effortful reasoning in lieu of more intuitive and automatic cognitive processes is still unclear and, as a result, has been the focus of much recent scholarship and empirical research (e.g., De Neys and Bonnefon, 2013, De Neys and Glumicic, 2008, Evans, 2009, Stanovich, 2009a, Thompson, 2009, Thompson et al., 2011).

One of the criticisms of dual-process theories is that they describe the characteristics of the two processes but are unclear on the question of how they operate (De Neys and Glumicic, 2008, Evans, 2007, Evans, 2010b, Gigerenzer and Regier, 1996, Osman, 2004, Stanovich and West, 2000). A common claim among dual-process theorists is that Type 2 processes monitor the output of Type 1 processes (e.g., Evans, 2006, Kahneman and Frederick, 2005, Stanovich, 1999). This default-interventionist perspective explains how Type 2 processing can be biased by earlier Type 1 outputs. However, the idea that Type 2 processes are themselves responsible for the instantiation of Type 2 processing is clearly problematic. In contrast, parallel form dual-process theories posit that both types of processing operate in parallel from the outset of reasoning (e.g., Sloman, 1996, Sloman, 2002, Smith and DeCoster, 2000). These parallel-competitive models suggest that bias is common because fast Type 1 processes output a response before the slower Type 2 processes can complete, though additional Type 2 processing may occur if the two types of processing output conflicting responses (for a comparison of default-interventionist and parallel-competitive models, see Evans, 2007, Evans and Stanovich, 2013a, Handley and Trippas, 2015). Parallel form theories highlight conflict detection as a source of later Type 2 processing, but still assume that the monitoring of conflict is itself a Type 2 process. Thus, as has been outlined elsewhere (Evans, 2009, Evans and Stanovich, 2013a), neither of the major groups of dual-process theories adequately explain important aspects of cognitive architecture because both assume that Type 2 processing is effectively caused by itself. This is a problem of particular importance because the utility and explanatory value of dual-process theories is thought to depend, at least partially, on our understanding of the sources of analytic reasoning (Evans, 2009, Stanovich, 2009a, Thompson, 2009).

In the current work we introduce a new perspective on the factors that lead to Type 2 engagement. Our goal is to investigate the bottom-up (i.e., stimulus-triggered) processes that lead to increases in deliberative thought, independent from top-down factors such as instructional manipulations (e.g., Evans, Handley, Neilens, Bacon, & Over, 2010) and individual differences in analytic thinking disposition (e.g., Stanovich & West, 2008). We combine insights from recent dual-process models (De Neys, 2012, Evans, 2009, Stanovich, 2009a, Thompson, 2009) into a three-stage model of analytic engagement. Using a version of a classic decision making task, we provide evidence for two core claims derived from the model: (1) The detection of conflicts between intuitive responses is a key determinant of analytic engagement, but sometimes fails, and (2) the deliberative override of an intuitive response in lieu of an alternative is a later source of Type 2 processing that is dissociable from earlier increases in Type 2 processing attributable to conflict detection. Following previous research, we posit that analytic thinking may take the form of either rationalization (i.e., bolstering or verifying an intuitive response) or decoupling (i.e., overriding or falsifying an intuitive response in lieu of an alternative). Moreover, we qualify our findings in meaningful ways with a top-down source of Type 2 processing: individual differences in analytic thinking disposition. Our results indicate that reasoning failures can emerge from two sources: (1) Failing to detect bias (leading to a failure to think analytically; e.g., Evans, 2007, Kahneman, 2003), or (2) successfully detecting bias (e.g., De Neys, 2012), but failing to use analytic thought to override the intuitive response.

Although research has shown that the degree of involvement of Type 2 processing can be affected by top-down factors such as instructions (e.g., Daniel and Klaczynski, 2006, Evans et al., 1994, Vadeboncoeur and Markovits, 1999), the amount of time permitted to think (e.g., Evans and Curtis-Holmes, 2005, Finucane et al., 2000), and individual differences in thinking disposition (e.g., Stanovich and West, 1998, Stanovich and West, 2000) isolating lower level cognitive processes that lead to Type 2 processing are more important for our emerging understanding of the dynamic relation between Type 1 and Type 2 processes in the mind. Bottom-up factors can be used to determine which type of processing will dominate. Consider the following base-rate problem (from De Neys & Glumicic, 2008, adapted from Kahneman & Tversky, 1973):

In a study 1000 people were tested. Among the participants there were 995 nurses and 5 doctors. Paul is a randomly chosen participant of this study. Paul is 34 years old. He lives in a beautiful home in a posh suburb. He is well spoken and very interested in politics. He invests a lot of time in his career. What is most likely?

(a) Paul is a nurse.

(b) Paul is a doctor.

This problem includes two pieces of information that point to alternative responses. The base-rate probability (i.e., 995 nurses versus 5 doctors) indicates that there is a 99.5% chance that Paul is a nurse. In contrast, the personality description contains stereotypes that are strongly diagnostic of a doctor. A great deal of research has demonstrated that participants tend to strongly favor the stereotypical information over the base-rate probability because the stereotype is the more intuitive source of information (see Barbey & Sloman, 2007 for a review). Thus, the base-rate problem is thought to engender an initial response based on the salient stereotypical information.

Recent research has also indicated that people are implicitly aware of the conflict between base-rate and stereotype, despite the apparent neglect or underweighting of the base-rates (De Neys et al., 2011, De Neys and Franssens, 2009, De Neys and Glumicic, 2008, De Neys et al., 2008), perhaps because extreme probabilities (as shown above) can be processed very rapidly (Pennycook et al., 2012, Pennycook and Thompson, 2012, Pennycook et al., 2014). Importantly for present purposes, one of the key pieces of evidence for the conflict detection hypothesis is an increase in response time (RT) for conflict (as above) versus non-conflict (e.g., if there were 5 nurses and 995 doctors above) base-rate problems even when participants give the stereotypical response.² Thus, detection of the conflict between base-rate and stereotype appears to lead to increased Type 2 processing (as reflected by increased RT) even in cases where participants give the response that is more intuitively salient. In support of this claim, De Neys et al. (2008) found increased activation in the anterior cingulate cortex, a region of the brain associated with conflict detection (see Bush, Luu, & Posner, 2000 for review), for stereotypical responses to incongruent problems relative to congruent problems. Given the fact that participants gave the stereotypical response despite the apparent increase in Type 2 processing, it is likely that they spent their time rationalizing the stereotype or, at the very least, weighing the stereotype against the base-rate probability (Pennycook & Thompson, 2012). This leads to the appearance of “effortful beliefs”: i.e., belief processing that is analytic (Handley et al., 2011, Handley and Trippas, 2015, Trippas et al., 2014).

The central role of conflict detection as an initiator of Type 2 processing is evidenced by a wide range of measures across numerous reasoning tasks (see De Neys, 2012 for a review). Indeed, response conflict has long been an important concept in reasoning and decision making research (Evans et al., 1983, Kahneman, 2000, Kahneman and Tversky, 1982, Wilkins, 1928) and much neuropsychological work supports the idea that “conflict” problems are processed differently than “non-conflict” problems (Banks and Hope, 2014, Goel and Dolan, 2003, Liang et al., 2014, Prado et al., 2008, Prado and Noveck, 2007, Sanfey et al., 2003, Stollstorff et al., 2012). Nonetheless, conflict monitoring is not included as a component in most formal dual-process reasoning models (e.g., Evans, 2009, Stanovich, 2009a, Thompson, 2009, but see De Neys, 2012, Handley and Trippas, 2015), perhaps because monitoring has been considered a Type 2 process (and therefore not a separate component). Moreover, the primary dual-process model that does incorporate conflict monitoring – De Neys, 2012, De Neys, 2014 logical intuition model – focuses entirely on the processes that lead to successful conflict detection and therefore does not incorporate differences in the quality of Type 2 processing.

The primary goal of the current work is to develop a dual-process model that includes both a conflict monitoring stage and a Type 2 processing stage that differentiates between different levels of analytic engagement. This model could then accommodate both major perspectives on the primary cause of biased responding: (1) A failure to engage Type 2 processing (e.g., Evans, 2007, Kahneman, 2003), and (2) successfully engaging Type 2 processing following conflict detection, but failing to override the biased response (e.g., De Neys, 2012). These two sources of bias have often been discussed in the context of a debate about the modal biased reasoner (see De Neys, 2014 for a review) and, as such, we will also frame the perspectives as conflicting. However, this should not be taken to mean that authors such as Evans (2007) deny the existence of conflict detection (see Evans, 2009) or that authors such as De Neys (2012) deny the existence of analytic engagement failures (see De Neys, 2014). Our goal here is to assess the models of bias by the respective authors, which include predictions for one or the other source of bias but that do not necessarily preclude other factors.

Conflict monitoring is not the only bottom-up source of analytic thinking. For example, De Neys and Glumicic (2008) also reported an increase in RT for incongruent (i.e., conflict) problems relative to congruent (i.e., non-conflict) when participants gave the base-rate response to the incongruent problems. In this case, the apparent increase in Type 2 processing is potentially a result of a rethinking or decoupling process. Indeed, De Neys and colleagues have postulated that participants engaged additional resources to inhibit the prepotent stereotypical response (De Neys and Franssens, 2009, De Neys et al., 2008, Franssens and De Neys, 2009). That is, given the idea that stereotypes cue intuitive (Type 1) responses, additional Type 2 processing is therefore thought to be necessary to suppress and override the stereotype response in lieu of the base-rate response (Barbey & Sloman, 2007). Again, in support of this claim, De Neys et al. (2008) found increased activation in the right lateral prefrontal cortex (RLPFC) for base-rate responses to incongruent problems relative to congruent. The RLPFC is considered a key area involved in response inhibition (see Aron, Robbins, & Poldrack, 2004, for a review). Base-rate responses, like stereotypical responses, were associated with increased ACC activation. This indicates that participants were able to detect the conflict between base-rate and stereotypes for incongruent problems regardless of their ultimate response, but cases where the base-rate response was given involved an additional deliberative reasoning process relative to when the stereotypical response was given. Considering the association between the RLPFC and response inhibition along with the presumed intuitiveness of stereotypical information, it is plausible that this additional process consisted of participants actively suppressing the stereotypical response. In other words, cognitive decoupling appears to be a later source of Type 2 processing relative to conflict detection.

An additional point needs to be clarified. The claim that base-rate responses are usually accompanied by an active suppression of the salient stereotypical response via Type 2 processing is not the same as claiming that the base-rate response necessarily requires Type 2 processing to enter into reasoning (De Neys, 2007). Indeed, a recent set of experiments using an instruction manipulation illustrated that both base-rates and stereotypes appear to interfere with each other (Pennycook, Trippas, et al., 2014). This cross-interference was evident even when participants were forced to respond within a short time-deadline. This finding indicates that both base-rates and stereotypes cue Type 1 outputs (see also Brenner, Griffin, & Koehler, 2012). Under this account, stereotypes typically dominate reasoning because they cue intuitive responses that come to mind more quickly and fluently than the base-rates (Pennycook, Trippas, et al., 2014). Stereotypes, in other words, are a more salient source of information than base-rates, but both may cue Type 1 outputs. Moreover, decoupling should occur in cases when the base-rate response is provided because an intuitive response based on the stereotypical information is thought to have come first in the reasoning process and therefore needs to be overridden for an alternative response to be given.

Fig. 1 represents our theoretical position. The model was built to describe the reasoning process for a problem or cue that elicits multiple conflicting outputs. It formalizes and combines distinctions made by previous theorists (e.g., De Neys, 2012, Epstein, 1994, Evans, 2006, Evans and Stanovich, 2013a, Handley and Trippas, 2015, Sloman, 1996, Sloman, 2014, Smith and DeCoster, 2000, Stanovich, 2004, Strack and Deustch, 2004, Thompson, 2009) by dividing an individual reasoning event into stages and components. In the first stage, autonomous Type 1 processes generate so-called “intuitive” responses. These Type 1 processes are cued by features of the stimulus, do not require working memory or executive functioning, and operate in parallel (Evans, 2008, Sloman, 1996, Stanovich, 2004). Given these features, we have inferred that some stimuli will cue multiple, potentially competing Type 1 outputs (for a similar perspective, see De Neys, 2012, De Neys, 2014).

A second dimension of the initial stage in our model relates to the idea that some initial responses come to mind more quickly and fluently than others (Thompson, 2009, Thompson et al., 2011, Thompson et al., 2013).³ In the case of base-rate problems, for example, stereotypes are often used as intuitive lures because of the phenomenology of their fluent generation. However, this does not rule out the possibility that alternative sources of information can cue an alternative Type 1 output in parallel. As discussed above, extreme base-rates presented in simple frequency formats influence response time, confidence, and probability estimates in ways diagnostic of Type 1 processing (Pennycook, Trippas, et al., 2014). Thus, base-rate problems serve as an example of a case where two competing sources of information embedded in a problem can elicit competing initial responses (see Section 6.5 for further examples). The base-rate problem example is particularly powerful given the presumed alternative time-course of the stereotype initial response (IR₁) and the base-rate initial response (IR₂). Specifically, stereotypes likely cue initial responses that come to mind more quickly (and, as a consequence, more fluently) than do base-rates. For other types of problems or cues, it is possible that multiple additional initial responses are elicited, hence IR_n (see Fig. 1).

The role of the second stage, then, is to monitor for conflict between Type 1 outputs (De Neys, 2012, De Neys, 2014). If no conflict is detected (either because no conflict existed or because of a conflict detection failure), the first initial response (IR₁) will continue to the third stage where it is accepted with cursory analytic (Type 2) analysis. This is the prototypical way in which bias is thought to arise: unimpeded and with little effort. If a conflict is successfully detected, however, more substantive Type 2 reasoning will be engaged. Thus, conflict detection is a bottom-up source of analytic engagement.

The three-stage model then distinguishes between two very different forms of Type 2 processing, each with different implications for the degree of bias ultimately displayed. Rationalization is a form of Type 2 processing where, despite successful conflict detection, the reasoner focuses on justifying or elaborating the first initial response (IR₁) without seriously considering the Type 1 output that was cued by the stimulus, but that did not come to mind as quickly and fluently (IR₂) as the first initial response (IR₁).⁴ This leads to a response in line with what would typically be considered bias (i.e., one’s strongest intuition, which will often be personally relevant), but that has been bolstered by analytic reasoning (an “effortful” belief-based response; see Handley & Trippas, 2015). This process is traditionally referred to as “rationalization” in the reasoning literature (e.g., Wason & Evans, 1975), to highlight the idea that the additional consideration is focused on verifying, and not falsifying, the Type 1 output. For example, participants typically spend much of their time looking at the card they ultimately select on the Wason card selection task, indicating that they are likely focused on rationalizing their default response (Ball et al., 2003, Evans, 1996). We note, in addition, that rationalization is tied to a substantial body of research on motivated reasoning (see Kunda, 1990). This research indicates that the instantiation of Type 2 reasoning can sometimes lead to the strengthening of a pre-existing belief or attitude, particularly if the belief or attitude is of some personal significance.

The second class of Type 2 processes that could result from conflict detection is cognitive decoupling (Stanovich, 2004, Stanovich, 2009a). This is perhaps the most prototypical “analytic” process and, as such, has dominated the literature on reasoning. Decoupling refers to the additional processing necessary to inhibit and override an intuitive response (primarily, IR₁). There are three obvious possibilities given a decoupling process: (1) IR₁ is suppressed in lieu of IR₂ which, upon reflection, emerges as a stronger alternative, (2) IR₁ is suppressed in lieu of some other initial response (IR_n), and (3) an alternative response (AR) is generated that represents a novel amalgamation of initial responses (see Section 6.7.3 for further comment on AR).

The three-stage model is a novel combination of multiple perspectives, which means that individual aspects of the model are grounded in previous theory. The idea that some intuitive responses come to mind faster than others is an aspect of Thompson’s (e.g., 2009) metacognitive dual-process theory. The idea that conflicting Type 1 outputs may cue analytic thinking is the core of De Neys’ (e.g., 2012) logical intuition model (see also Handley & Trippas, 2015). Rationalization (e.g., Wason & Evans, 1975) and decoupling (e.g., Stanovich, 1999) have long been discussed in the context of dual-process models, though to the best of our knowledge they have not been included as separate classes of Type 2 processing in the same model. Moreover, Stanovich and West (2008) have used stages to determine when and if intuitive response will be overridden under the goal of creating a framework for understanding individual differences in heuristics and biases tasks (see also Kahneman & Tversky, 1982). Here, in contrast, we consider the reasoning process as stages to highlight different bottom-up sources of Type 2 processing. Previous models have highlighted key aspects of the reasoning process, but have largely left unanswered the question of what actually causes Type 2 processing to be engaged.

The goal of the current work is to demonstrate the utility of our three-stage model. This will be done in three ways: (1) By investigating the possibility that conflict monitoring may sometimes fail (Stage 2), (2) by dissociating increases in Type 2 processing that indicate, on one hand, rationalization following successful conflict detection and, on the other hand, cognitive decoupling (Stage 3), and (3) by investigating the locus of individual differences in reasoning.

Prior to outlining our specific predictions, it is necessary to discuss an additional source of Type 2 processing. Research has indicated that the mere willingness to engage deliberative reasoning (i.e., differences in thinking disposition or cognitive style) predicts reasoning performance over and above individual differences in the ability to think analytically (i.e., cognitive ability or intelligence) (for reviews, see Stanovich, 2004, Stanovich, 2009a, Stanovich, 2011, Stanovich and West, 2000). For example, individuals who are actively open-minded are more willing to question and perhaps rethink an initial response (Baron, 2008). This disposition, as assessed by a number of questionnaires, has been linked to a wide range of reasoning and decision-making tasks (Stanovich and West, 1997, Stanovich and West, 1998, Toplak et al., 2011). Taking the base-rate problem as an example, participants who are actively open-minded are more likely to choose the base-rate over the stereotype relative to less analytic individuals, presumably because they were more willing to think analytically about the initial stereotypical response (Pennycook, Cheyne, Barr, Koehler, & Fugelsang, 2014a). Stanovich, 2004, Stanovich, 2009b has argued that thinking disposition is an underappreciated determinant of psychological outcomes. Recent research has supported the idea that cognitive style plays a consequential role in psychological domains that are of some general import: e.g., creativity (Barr, Pennycook, Stolz, & Fugelsang, 2014), moral judgments and values (Paxton et al., 2012, Pennycook et al., 2014b, Rozyman et al., 2014), religious belief (Gervais and Norenzayan, 2012, Pennycook et al., 2012, Pennycook et al., 2013, Pennycook et al., 2014a, Shenhav et al., 2012), and even Smartphone technology use (Barr, Pennycook, Stolz, & Fugelsang, 2015). The research indicating that individual differences in cognitive style have important effects on beliefs and behavior implies that the engagement of Type 2 reasoning processes involves an important dispositional component. Cognitive style has particular relevance for the current discussion as it represents an independent top-down source of Type 2 processing. That is, how much someone values or enjoys analytic thinking may contribute to the probability that they engage Type 2 process, independent of any Type 1 output monitoring process and therefore regardless of the content of the stimulus.

The foregoing highlights an additional source of uncertainty about the factors that elicit Type 2 processing; namely, do individual differences relate to conflict detection? Recently, De Neys and Bonnefon (2013) theoretically integrated research on conflict detection with individual differences in reasoning. Specifically, they asked the question “do biased and unbiased reasoners take different paths early on in the reasoning process or is the observed variance late to arise?” (p. 172). The answer to this question has significant implications: If individual differences only affect reasoning at a relatively late stage (Stage 3 in our model), as De Neys and Bonnefon claim, it would imply that the influence of said individual differences has been greatly overemphasised in the reasoning and decision making literature. To support this argument, De Neys and Bonnefon cited the many cases where even “biased” reasoners appeared to have detected reasoning conflicts, with respect to both RT increases for incongruent base-rate problems (De Neys & Glumicic, 2008), and many other types of problems and measures (De Neys, 2012). These findings suggest that “biased” and “unbiased” reasoners diverge late in the reasoning process, thereby suggesting that both types of reasoners are likely closer in cognitive function to each other than some may have previously considered.⁵

However, while the research cited by De Neys and Bonnefon (2013) does indeed indicate that both biased and unbiased reasoners are able to detect the conflict between base-rates and stereotypes, for example, little research has directly compared reasoners based on the extent of Type 2 processing increase as a function of conflict detection (but see Pennycook et al., 2014a). Do relatively intuitive individuals (i.e., those who are relatively biased) engage in comparable levels of Type 2 processing in the face of conflict as reflective individuals? While it may be the case that intuitive people are able to efficiently detect conflict during reasoning, as suggested by De Neys and Bonnefon, it may also be the case that this conflict detection does not engender much Type 2 processing relative to more analytic individuals. This is an open question that speaks directly to the extent of cognitive processing differences that arise as a function of individual differences.

The utility of dual-process theory is tied largely to the ability to predict when Type 2 processing will be engaged. Here, we have developed a three-stage model of reasoning and applied it to an illustrative class of reasoning problems. Although the model is consistent with a relatively large body of extant research, there are a number of components that must be empirically tested. Here we investigate two core claims derived from the model: (1) Conflict detection is a key determinant of analytic engagement, but sometimes fails, and (2) conflict detection and cognitive decoupling (i.e., expending additional effort to override an intuitive response in lieu of an alternative) are dissociable sources of analytic thinking. Secondarily, we investigate whether responsiveness to conflicts is subject to individual differences.

To do this, we develop a paradigm that is suitable for measuring subtle increases in Type 2 processing. This paradigm uses base-rate problems which, as outlined above, are of particular interest because they reliably elicit RT increases presumed to result separately from conflict detection and cognitive decoupling processes. To reiterate, participants spend more time on problems that contain a conflict between base-rate and stereotype relative to non-conflict control problems (De Neys and Glumicic, 2008, Pennycook et al., 2012). Importantly, this RT increase is evident for both stereotype (IR₁) and base-rate (IR₂) responses. As discussed, these RT increases should reflect different processes in the three-stage model. The RT increase for stereotypical responses relative to non-conflict problems is reflective of successful conflict monitoring because such cases reflect sensitivity to IR₂ even when IR₁ is the chosen response (De Neys et al., 2008). Presumably, the additional time is used to rationalize IR₁. In contrast, following previous research that indicates that stereotypical information is a highly salient source of intuitive responses (see Barbey & Sloman, 2007), the RT increase for base-rate responses relative to non-conflict problems should reflect the use of Type 2 processing to rethink or decouple from the initial stereotypical response (IR₁), leading to the base-rate response (IR₂). This process should take additional time because IR₁ must be inhibited or suppressed in lieu of the alternative.

Response time is a crucial measure for the current purposes as our focus is on measuring relative increases in Type 2 processing as a function of conflict detection and decoupling. Given the presumption that Type 2 processing is typically slower and more resource demanding than Type 1 processing, longer RTs in an experimental condition are thought to reflect an increased level of Type 2 engagement (e.g., Thompson et al., 2011). However, RTs are also notoriously noisy. This is particularly true for typical base-rate problems as mean RTs typically range from 10 to 25 s (De Neys and Glumicic, 2008, Pennycook et al., 2012). Thus, we developed a rapid-response version of the base-rate task wherein participants are presented with the individual components of traditional base-rate problems in succession (Pennycook et al., 2014a). In lieu of the long stereotypical descriptions (see above example), participants are presented with a single trait (e.g., “kind”) that is strongly diagnostic of one group (e.g., nannies) but not the other (e.g., politicians). This allowed us to decrease extraneous variance due to reading times, increase reliability by including a relatively large number of items, and easily manipulate components of the items across conditions and experiments.