Mercier & Sperber's (M&S's) impressive argumentative theory reassesses the function of reasoning, not as involving the noble pursuit of truth, but instead as a Machiavellian communicative art with persuasion and self-interest at its core. A case in point is the infamous confirmation bias, whereby individuals seem motivated to seek confirmatory evidence for their existing beliefs and hypotheses and fail to look for counterevidence or counterarguments (Nickerson Reference Nickerson1998). M&S claim that a quintessential demonstration of confirmation bias that supports their theory can be seen in the phenomenon of belief bias, where some contemporary theories suggest that people try to confirm believable conclusions but disconfirm unbelievable ones (see the selective processing model described by Evans Reference Evans2007, and the multinomial model of Klauer et al. Reference Klauer, Musch and Naumer2000). Thus, in both the case of believable and unbelievable conclusions, M&S claim that people show a motivation “to confirm their initial intuition” (sect. 3.3, para. 3) with unbelievable conclusions effectively promoting an intuition-guided debias (Evans Reference Evans, Garcìa-Madruga, Carriedo and González-Labra2000) leading to improved logical responding. M&S further propose that, when people deal with an unbelievable conclusion, “it is not that they reason more in this case. . . . It is just that the direction reasoning takes is mostly determined by the participants' initial intuitions” (sect 3.3, para. 3).
Our contention, however, is that this latter claim flies in the face of current chronometric evidence in the belief-bias literature, which suggests that a more subtle interpretation of the phenomenon is needed that is not couched purely in terms of the operation of a general confirmation bias. In particular, processing-time data for syllogistic arguments consistently indicate that participants reason most when the conclusion is believable but invalid (Ball et al. Reference Ball, Philips, Wade and Quayle2006; Stupple & Ball Reference Stupple and Ball2008; Thompson et al. Reference Thompson, Striemer, Reikoff, Gunter and Campbell2003). Such a finding is inconsistent with M&S's view that people simply seek support for prior beliefs. Confirmatory mental models of the premises of arguments with believable-invalid conclusions are readily available, so why should significantly longer processing times arise with these problems? We propose instead that many participants show a capacity to inhibit confirmation-oriented processing with such arguments, with the resulting attempt at logical analysis taking time to apply. Of course, the complexity of the required logical processing means that a belief-based response may still often win out, perhaps with people defaulting to such a response under cognitive load (cf. Quayle & Ball Reference Quayle and Ball2000). This would produce a response pattern that looks like confirmatory behaviour, but where the chronometric data support a very different account of the processing that is actually taking place.
To elaborate on our proposals, we outline three possible routes that participants could take through Evans' (Reference Evans2007) selective processing model when confronted with belief-oriented syllogistic arguments (cf. Ball Reference Ball, Manktelow, Over and Elqayam2010; Evans Reference Evans, Evans and Frankish2009). First, participants could employ a pure response bias and respond in accordance with belief without engaging any analytic processing whatsoever, either for a truth-seeking or argumentative purpose. Second, in accordance with argumentative theory, participants could seek confirmatory evidence so as to warrant the acceptance of believable conclusions (including believable-invalid ones) and the refutation of unbelievable conclusions (including unbelievable-valid ones). Finally, participants could attempt to suspend notions of belief and disbelief altogether. For example, rather than searching for a supporting model for an believable-invalid conclusion, they would inhibit a heuristic response as well as confirmatory-oriented analytic response, instead engaging in an exhaustive search for a model of the premises that provides a counterexample to the given conclusion.
The important question that follows from our analysis is this: What if examples of all three reasoning approaches were present in belief-bias data sets? We suggest that the resulting aggregation of these strategies would lead to the pattern of response times that has typically been observed (Ball et al. Reference Ball, Philips, Wade and Quayle2006; Stupple & Ball Reference Stupple and Ball2008; Thompson et al. Reference Thompson, Striemer, Reikoff, Gunter and Campbell2003), with believable-invalid conclusions being associated with extended processing times relative to other problems because of the presence of a subset of reasoners who resist intuitive judgements. This group of reasoners would most likely be those described by Stanovich and West (Reference Stanovich and West2000) as adopting the “normative construal” of the task because they possess the cognitive capacity needed to reason through demanding deductive problems.
This latter interpretation of chronometric findings is supported by data that we have recently acquired (for a preliminary report, see Ball Reference Ball, Manktelow, Over and Elqayam2010) demonstrating that increased response times for believable-invalid problems are predictive of increased overall response accuracies across belief-oriented problems (i.e., these times reflect the behaviour of high-ability reasoners). These data also concur with the observation by Thompson et al. (Reference Thompson, Newstead, Morley, Manktelow, Over and Elqayam2010) that the participants who spend more time reasoning are more likely to reach a logical conclusion. However, Thompson et al. propose an alternative interpretation of the inflated inspection times for believable-invalid problems that is more in keeping with argumentative theory than our account. In their Modified Verbal Reasoning theory they suggest that participants are motivated to find confirmatory support for believable problems and so allow an extended response deadline for such problems (perhaps an attractive proposition for the argumentative theory). Thompson et al. claim that since the reasoning process is more difficult for invalid than valid problems, this combined with the increased motivation to support believable conclusions results in the corresponding increase in response times for believable-invalid problems. We would argue, however, that in adopting this perspective one would expect a far greater acceptance rate for believable-invalid problems than is observed (acceptances are typically in the 50%–70% range), since a confirming model is readily available to those participants who expend the effort trying to find one.
In sum, we disagree with M&S's assertion that a motivated attempt at confirmation necessarily drives belief-bias effects in syllogistic reasoning. Instead, we claim that many participants attempt to seek out the deductive truth of presented arguments and do so at the considerable cognitive expense of inhibiting their intuitive judgements, as is particularly evident in the case of syllogisms with believable-invalid conclusions. On this issue, at least, we would argue against M&S's argumentative theory, but perhaps we cannot rule out having done so in order to confirm a pre-existing hypothesis.
Mercier & Sperber's (M&S's) impressive argumentative theory reassesses the function of reasoning, not as involving the noble pursuit of truth, but instead as a Machiavellian communicative art with persuasion and self-interest at its core. A case in point is the infamous confirmation bias, whereby individuals seem motivated to seek confirmatory evidence for their existing beliefs and hypotheses and fail to look for counterevidence or counterarguments (Nickerson Reference Nickerson1998). M&S claim that a quintessential demonstration of confirmation bias that supports their theory can be seen in the phenomenon of belief bias, where some contemporary theories suggest that people try to confirm believable conclusions but disconfirm unbelievable ones (see the selective processing model described by Evans Reference Evans2007, and the multinomial model of Klauer et al. Reference Klauer, Musch and Naumer2000). Thus, in both the case of believable and unbelievable conclusions, M&S claim that people show a motivation “to confirm their initial intuition” (sect. 3.3, para. 3) with unbelievable conclusions effectively promoting an intuition-guided debias (Evans Reference Evans, Garcìa-Madruga, Carriedo and González-Labra2000) leading to improved logical responding. M&S further propose that, when people deal with an unbelievable conclusion, “it is not that they reason more in this case. . . . It is just that the direction reasoning takes is mostly determined by the participants' initial intuitions” (sect 3.3, para. 3).
Our contention, however, is that this latter claim flies in the face of current chronometric evidence in the belief-bias literature, which suggests that a more subtle interpretation of the phenomenon is needed that is not couched purely in terms of the operation of a general confirmation bias. In particular, processing-time data for syllogistic arguments consistently indicate that participants reason most when the conclusion is believable but invalid (Ball et al. Reference Ball, Philips, Wade and Quayle2006; Stupple & Ball Reference Stupple and Ball2008; Thompson et al. Reference Thompson, Striemer, Reikoff, Gunter and Campbell2003). Such a finding is inconsistent with M&S's view that people simply seek support for prior beliefs. Confirmatory mental models of the premises of arguments with believable-invalid conclusions are readily available, so why should significantly longer processing times arise with these problems? We propose instead that many participants show a capacity to inhibit confirmation-oriented processing with such arguments, with the resulting attempt at logical analysis taking time to apply. Of course, the complexity of the required logical processing means that a belief-based response may still often win out, perhaps with people defaulting to such a response under cognitive load (cf. Quayle & Ball Reference Quayle and Ball2000). This would produce a response pattern that looks like confirmatory behaviour, but where the chronometric data support a very different account of the processing that is actually taking place.
To elaborate on our proposals, we outline three possible routes that participants could take through Evans' (Reference Evans2007) selective processing model when confronted with belief-oriented syllogistic arguments (cf. Ball Reference Ball, Manktelow, Over and Elqayam2010; Evans Reference Evans, Evans and Frankish2009). First, participants could employ a pure response bias and respond in accordance with belief without engaging any analytic processing whatsoever, either for a truth-seeking or argumentative purpose. Second, in accordance with argumentative theory, participants could seek confirmatory evidence so as to warrant the acceptance of believable conclusions (including believable-invalid ones) and the refutation of unbelievable conclusions (including unbelievable-valid ones). Finally, participants could attempt to suspend notions of belief and disbelief altogether. For example, rather than searching for a supporting model for an believable-invalid conclusion, they would inhibit a heuristic response as well as confirmatory-oriented analytic response, instead engaging in an exhaustive search for a model of the premises that provides a counterexample to the given conclusion.
The important question that follows from our analysis is this: What if examples of all three reasoning approaches were present in belief-bias data sets? We suggest that the resulting aggregation of these strategies would lead to the pattern of response times that has typically been observed (Ball et al. Reference Ball, Philips, Wade and Quayle2006; Stupple & Ball Reference Stupple and Ball2008; Thompson et al. Reference Thompson, Striemer, Reikoff, Gunter and Campbell2003), with believable-invalid conclusions being associated with extended processing times relative to other problems because of the presence of a subset of reasoners who resist intuitive judgements. This group of reasoners would most likely be those described by Stanovich and West (Reference Stanovich and West2000) as adopting the “normative construal” of the task because they possess the cognitive capacity needed to reason through demanding deductive problems.
This latter interpretation of chronometric findings is supported by data that we have recently acquired (for a preliminary report, see Ball Reference Ball, Manktelow, Over and Elqayam2010) demonstrating that increased response times for believable-invalid problems are predictive of increased overall response accuracies across belief-oriented problems (i.e., these times reflect the behaviour of high-ability reasoners). These data also concur with the observation by Thompson et al. (Reference Thompson, Newstead, Morley, Manktelow, Over and Elqayam2010) that the participants who spend more time reasoning are more likely to reach a logical conclusion. However, Thompson et al. propose an alternative interpretation of the inflated inspection times for believable-invalid problems that is more in keeping with argumentative theory than our account. In their Modified Verbal Reasoning theory they suggest that participants are motivated to find confirmatory support for believable problems and so allow an extended response deadline for such problems (perhaps an attractive proposition for the argumentative theory). Thompson et al. claim that since the reasoning process is more difficult for invalid than valid problems, this combined with the increased motivation to support believable conclusions results in the corresponding increase in response times for believable-invalid problems. We would argue, however, that in adopting this perspective one would expect a far greater acceptance rate for believable-invalid problems than is observed (acceptances are typically in the 50%–70% range), since a confirming model is readily available to those participants who expend the effort trying to find one.
In sum, we disagree with M&S's assertion that a motivated attempt at confirmation necessarily drives belief-bias effects in syllogistic reasoning. Instead, we claim that many participants attempt to seek out the deductive truth of presented arguments and do so at the considerable cognitive expense of inhibiting their intuitive judgements, as is particularly evident in the case of syllogisms with believable-invalid conclusions. On this issue, at least, we would argue against M&S's argumentative theory, but perhaps we cannot rule out having done so in order to confirm a pre-existing hypothesis.