Development of executive function: more than conscious reflection

In arguing for a role of conscious reflection in successful completion of the A not B task, Marcovitch and Zelazo have provided a novel and potentially viable account of the origin, or more accurately the resolution, of the A not B error. The general U-shaped effect of the number of A trials on the probability of error on a B trial (i.e. greater likelihood of error with intermediate numbers of A trials preceding the critical B trial than with few or many A trials: Marcovitch, Zelazo & Schmuckler, 2002; Marcovitch & Zelazo, 2006), together with the systematic nature of performance on B trials (i.e. that performance on a B trial, once correct, remains correct on subsequent responses: Marcovitch et al. , 2002), appears to provide compelling evidence for Marcovitch and Zelazo’s position. Yet the authors go further – they relate their theory to the development of executive function (EF). This is a laudable goal given the limited research to date on EF abilities of infants. However, close inspection of Marcovitch and Zelazo’s hierarchical competing systems model (HCSM) raises a number of troubling issues. For simplicity, let us begin by accepting the evidence at face value. Does the HCSM advance our understanding of the development of EF? A major difficulty in the study of EF is that different researchers adopt different definitions of EF. Thus, Marcovitch and Zelazo take EF to be ‘the cognitive processes underlying the conscious control of behaviour’ (p. 1, emphasis added), and draw on Miyake, Friedman, Emerson, Witzki, Howerter and Wager’s (2000) characterization of EF as comprising (at least) three separable but interacting components: task switching, response inhibition and memory updating. Miyake et al. , however, characterize EF as ‘general-purpose control mechanisms that modulate the operation of various cognitive subprocesses and thereby regulate the dynamics of human cognition’ (Miyake et al. , 2000, p. 50). Critically, Miyake et al. make no claims about EF and consciousness. In contrast, conscious reflection is the central explanatory mechanism in Marcovitch and Zelazo’s account of correct performance on the A not B task. Unfortunately, Marcovitch and Zelazo provide no evidence that the hypothesized process they label ‘conscious reflection’ has any characteristic of consciousness beyond being discrete all-or-nothing. Worse, ‘conscious reflection’ by itself can provide no account of many accepted findings in A not B and related tasks. For example, infants are more likely to perseverate on the first B trial when the delay between hiding and searching is long than when it is short, but the delay that infants can tolerate without producing perseverative errors increases with age (Diamond, 1985; Marcovitch & Zelazo, 1999). Are we to attribute age-related improvements in delay tolerance to developing processes of consciousness? If so, then those processes would appear to possess attributes (such as decay) normally associated with memory. Even more problematic for a conscious reflection account are the dissociations sometimes observed between looking and reaching on the first B trial (e.g. Diamond, 1990). Given our current understanding of EF, these effects would seem to be well accounted for by the development of response inhibition and memory control processes (Diamond, Cruttenden & Neiderman, 1994), i.e. by more traditional (and less ephemeral) EF. It is therefore unclear how Marcovitch and Zelazo’s appeal to conscious reflection extends our understanding of the development of EF. It remains to consider the key evidence for Marcovitch and Zelazo’s hypothesis: the U-shaped effect of number of A trials and the systematic nature of responses on subsequent B trials. First, the data-fitting of Figure 6 is misleading. This involves the use of three parameters to fit three data points and hence is not persuasive. Second, the U-shaped effect requires further empirical investigation. The main study cited by Marcovitch and Zelazo in support of the effect with infants (Marcovitch et al. , 2002) compared the proportion of erroneous A responses after 1, 6 and 11 A trials. While there was a statistically significant increase in A responses from 1 to 6 A trials, the decrease from 6 to 11 A trials was not statistically significant. (Marcovitch et al. (2002) report χ 2 (1) = 2.95,