In psychology, evidence of dissociation comes in one of two forms: synchronic or diachronic. Two capacities are synchronically dissociable if there are adults in whom the first capacity is defective and the second capacity is intact, or vice versa. Evidence for dissociation of this sort comes from studies of clinical mental disorders, such as autism and schizophrenia. Two capacities are diachronically dissociable if the first capacity emerges before the second capacity does, or vice versa. Evidence for dissociation of this sort comes from studies of normally developing children. A central issue in the target article is the dissociability – or lack thereof – of metacognition (first-person metarepresentation) from mindreading (third-person metarepresentation). After waving the issue of diachronic dissociation mostly to one side (sect. 4), Carruthers argues that there is no good evidence of synchronic dissociation (sects. 9 and 10). With respect to the first point, his dismissal of the developmental evidence is premature; on the second point, he makes a slightly better case, but the evidence isn't all in – so the jury is still out. Or so I argue.

Let's start with the case of synchronic dissociation and the view from developmental psychology. Carruthers opens his discussion here by citing a large-scale meta-analysis that found no evidence of developmental self-other asymmetries on metarepresentation tasks (Wellman et al. Reference Wellman, Cross and Watson2001). This is grist for the “mindreading is prior” mill but only up to a point. That's because the meta-analysis by Wellman et al. was exclusively concerned with studies using standard false-belief tasks and no other measures of metarepresentational capacity. And once we take into account the results of studies using such nonstandard measures, the idea that metacognition is developmentally prior to mindreading becomes more plausible (Nichols & Stich Reference Nichols and Stich2003; see also Robbins Reference Robbins2006).

A key piece of evidence here comes from a study by Wimmer et al. (Reference Wimmer, Hogrefe and Perner1988) using a version of the “seeing leads to knowing” paradigm. (Curiously, Carruthers never cites this study, though he does cite a later study of children with autism that employed much the same paradigm, albeit with different – and for his purposes, more favorable – results; see Kazak et al. Reference Kazak, Collis and Lewis1997.) In this study, done with normally developing 3-year-olds, subjects were divided into two groups. In the first group, the children were instructed to look inside a box; in the second group, visual access to the interior of the box was denied. Subjects in both groups were then asked whether or not they knew what was in the box, and most of the children in each group gave the correct answer: those in the first group said yes, those in the second group said no. Subsequently, children in both groups observed another person either looking inside of the box or not looking into it. They were then asked whether or not the person they had observed knew what was inside the box. The answers they returned were surprising (especially for anyone familiar with the literature on egocentric biases). For example, 14 children represented their own epistemic state correctly while misrepresenting the epistemic state of the other person, whereas only 2 children displayed the opposite pattern of responses (i.e., correct for other, incorrect for self) – a highly significant contrast (p<.01). In short, results from the Wimmer et al. study suggest that young children are better at reporting their own knowledge state than the knowledge state of others with whom they knowingly share access to the relevant information. Because knowledge is a propositional attitude par excellence, this looks like good support for the synchronic dissociability of metacognition from mindreading in precisely the domain of interest, namely, the metarepresentation of propositional attitudes.

Let's turn now to the issue of synchronic dissociation. Carruthers points out that passivity symptoms in schizophrenia, such as thought insertion and delusions of control, need not be (and probably are not) due to a breakdown in metacognition, and hence that studies of passivity-symptomatic schizophrenia – a condition in which mindreading appears to be relatively intact – do not support the idea that metacognition is defective in this condition. But this is a double-edged sword. For if the “mindreading is prior” view is correct, then metacognition should be impaired in those subtypes of schizophrenia in which mindreading is defective. Passivity-symptomatic schizophrenia is unrepresentative in this respect, in that patients with this subtype of the disorder tend to perform normally on standard first- and second-order false-belief tasks (Corcoran Reference Corcoran, Baron-Cohen, Tager-Flusberg and Cohen2000). By contrast, individuals in the paranoid subtype of schizophrenia perform poorly on a wide range of mindreading tasks, including tasks involving the attribution of intentions and the understanding of jokes, hints, and conversational implicatures (Brüne Reference Brüne2005). Indeed, the defining symptoms of paranoid schizophrenia include persecutory delusions and delusions of self-reference, both of which involve misattributing to other people intentions toward oneself. Exactly what explains this tendency is a matter of controversy, but its characterization in terms of a deficit in the mindreading system is not (Blackwood et al. Reference Blackwood, Howard, Bentall and Murray2001). Hence, if the “mindreading is prior” view is right, then we should expect to find impaired metacognition in these patients, especially in the domain of intention attribution; otherwise, we should conclude that metacognition and mindreading are dissociable after all. As things stand, however, there appears to be no credible evidence to this effect.

There are at least two possible explanations for this fact, only one of which is immediately damaging to Carruthers' position. The first possibility is that evidence of metacognitive impairments in paranoid schizophrenia is lacking because metacognition is spared in this disorder. That would be bad news for the “mindreading is prior” view. A second possibility, however, is that evidence of this sort is lacking simply because the requisite empirical studies have not been done yet. This second possibility seems at least as likely as the first, especially as far as the metacognition of propositional attitudes like intention is concerned. Schizophrenia researchers have paid more attention to the metacognition of emotions, that is, nonattitudinal mental states. But here the news isn't good for Carruthers either, for it appears that third-person deficits, such as difficulties with face-based emotion recognition, need not be accompanied by first-person deficits, such as difficulties with recognizing one's own emotions (Brunet-Gouet & Decety Reference Brunet-Gouet and Decety2006). It is entirely possible, then, that future studies of the metacognition of intention in schizophrenia will point toward a similar dissociation. And in that case, proponents of the “mindreading is prior” view will have some explaining to do.