Yesterday I reviewed evidence showing that set switching (e.g., your ability to suddenly switch behaviors) and rule representation (your ability to represent rules in a game, for example), may be distinct processes, at least insofar as they may show distinct developmental trajectories and rely on distinct neural substrates. Today's post will review a new study from Developmental Neuropsychology that also aims to show distinct developmental trajectories for set switching and rule maintenance, and how these claims hold up to a deeper analysis.
Huizinga & van der Molen administered four tasks to 237 subjects varying in age from 6 to 26 years (subjects were categorized as "7 year-olds", "11 year-olds," "15 year-olds" or "21 year-olds"). One of these tasks was the Wisconsin Card Sorting (WCST), an "executive function" test that is widely used with developmental and neuropsychiatric populations because of its sensitivity to the integrity of prefrontal cortex. The task itself involves sorting cards along the dimensions of shape, color or number according to unspoken rules; subjects must determine the current correct sorting rule by trial and error. After some number of trials, the current sorting rule will change, and subjects must then again determine the correct rule through trial and error.
The WCST is particularly useful because it provides a variety of different measures: "set maintenance" abilities can be indexed by the number of spontaneous errors a subject makes after they have "stumbled upon" the currently correct sorting rule (these are called "distraction errors"). Likewise, switching ability can be indexed by the number of times a subject responds according to an old sorting rule that is no longer correct (these are called "perseverative errors," which are particularly common in young and frontally-damaged populations), and also by the proportion of errors which are made in attempting to determine the current sorting rule. Finally, gross measures of performance include the total # of correct trials and the number of different sorting rules achieved.
The authors also administered tests of working memory, inhibition, and shifting, in order to examine the correlations between these capacities and performance on WCST, and how these correlations may change with age. The test of working memory was a somewhat unusual "tic tac toe" task, in which subjects had to indicate when a specific spatial sequence was reproduced on a 3x3 matrix. Inhibition was measured with the Stop Signal task, which measures the amount of time required for a subject to inhibit a motor response by comparing reaction time on "go" trials with that on "stop" trials (this task is described in more detail here). Finally, shifting was measured with the local/global task, in which subjects must switch between attending to either the individual features that make up a larger shape, or the identity of the larger shape itself (if you're not familiar with this task, this image should clear things up).
So, what did the authors find?
Not surprisingly, the number of rules achieved in the WCST, the # of distraction errors in the WCST, and # of trials correct in WCST increased with each successive age group, although there were no differences between the two oldest groups in terms of trials correct. Only the youngest group differed in terms of perseverative errors. This was sufficient for Huizinga & van der Molen to conclude that set-switching and maintenance processes follow distinct developmental trajectories: perseveration (a measure of switching) reaches adult levels at 11, whereas distraction errors (maintenance) continues to improve up to 21 years of age.
There are several problems with this conclusion. First of all, 21 year-olds actually committed more distraction errors than 15 year-olds (contrary to the claim made in the paper's text), and furthermore 21 year-olds commit numerically less perseverative errors than 15 year-olds. So it is at least plausible that set switching continues to improve beyond age 11. Finally, the same problem I raised in yesterday's post also applies here: the measures used to index "set switching" and "set maintenance" may differ in sensitivity or difficulty rather than in the inherent developmental trajectory of "set switching" and "set maintenance."
Anyway, Huizinga & van der Molen conducted a principal components analysis (PCA) on these measures, a data reduction technique that is useful in consolidating the variance present in several measures to relatively fewer measures. In this case, PCA identified two underlying factors in the youngest age group: one consisting of both perseverative errors and efficient errors (which supports the authors' hypothesis that these measures both index "set switching" ability), and a second factor consisting only of distraction errors (which the authors had hypothesized to index "set maintenance"). In contrast, only one factor was present in 11 year-olds and above.
There are several interesting things about this analysis which go unmentioned by the authors. First is a technical point: principal axis factoring (similar to latent factor analysis) may have been preferable in this case, because it retains only the variance shared by multiple measures. Second, if "set switching" and "set maintenance" are in fact dissociable, then why should PCA identify only a single factor in older age groups among measures of these purportedly distinct process? Third, the number of "efficient errors" was apparently unused in the principal components analysis, despite the authors' hypothesis that this also indexes set maintenance.
Disregarding for the moment these additional problems, these principal components were regressed on the shifting, inhibition, and working memory tasks. The results showed that among 7 year-olds the "set switching" factor was best predicted by performance on the shifting task, whereas the "maintenance" factor was best predicted by performance on the inhibition taks. For 11 year-olds, the single factor was best predicted by shifting, whereas in 15 year olds, a combination of shifting and working memory was the best predictor. By 21 years of age, however, working memory alone was the best predictor. Although these effects were significant, the total amount of variance explained by these models was startlingly small (as low as 6.9% and never higher than 15%).
As with the WCST shifting measures, the authors note that the tasks measuring "shiffting & inhibition appeared to reach young-adult levels" by 11 years, whereas working memory performance didn't reach adult levels until later (similar to the WCST set maintenance measures). The same caveats apply here as to the previous analysis that suggested age differences in performance reflect differing developmental trajectories, as opposed to differing sensitivity/difficulty, between measures.
This study provides somewhat equivocal evidence on the claim that set switching and rule representation are in fact different processes. The evidence reviewed yesterday was similarly controversial. And there are completely distinct reasons for thinking that rule representation and set switching are not different; switching can be realized in computational models simply by strongly representing the current rule. Perhaps the updating of working memory is distinct from working memory maintenance processes, but standard task-switching paradigms may not be capable of measuring these two quantities independently.