ANÁLISIS DE LAS CIRCUNSTANCIAS AGRAVANTES EN EL

One empirical chaUenge for task carryover accounts has resulted from the finding that switch costs between tasks of different levels of difficulty are not necessarily 'paradoxical' (i.e. larger in the easier task). Monsell et al. (2000) consider a number of cases (involving tasks with different degrees of stimulus- response compatibihty, for example) where the switch cost was larger for a switch into the more difficult task. Similarly, Rubinstein et al. (2001, Experiments 3-4) have provided evidence that a switch from classifying stimuli on a more

^ The idea that stim uli m ight directly trigger higher-level representations, independently o f o n e’s present intentions or even in conflict with them , has also been put forw ard in the social psychology literature. For exam ples see B argh and G ollw itzer (1994) and Shah and Kruglanski (2002).

familiar dimension to a less familiar dimension can sometimes incur a greater cost than a switch in the opposite direction. Thus, it does not seem to be a universal rule, as predicted by Allport et al. (1994), that a switch from a more familiar to a less familiar task yields a smaller switch cost than a switch in the reverse direction. However, this does not strongly challenge task carryover accounts since there is no reason for such accounts to be incompatible with larger switch costs in less familiar tasks.

A more problematic finding for task carryover accounts of switch costs has come from studies using the alternating runs paradigm with more than two trials before each switch. Experiments with run lengths of four (Rogers & Monsell, 1995, Experiment 6) and eight (Monsell, Azuma, Eimer, Le Pelley, & Strafford,

1000 900 'tn 800 E 700 60 0 50 0 12

I

Figure 2 2 Mean reaction time (RT) and error rate in Rogers andMonselTs (1995) Experiment 6, as afunction of the position in a run of four trials, averaged over both tasks, adapted from Rogers and Monsell (1995, Figure 5)

1998) trials before each switch have found that the cost of task switching is confined to the first trial of a run (see Figure 2.2). In other words, there seems to be no further improvement in RT after the first nonswitch trial. Rogers and Monsell (1995) argue that if switch costs result from a carryover effect from the previous task, they should dissipate gradually over a series of trials, rather than being eliminated after the very first switch trial. This pattern of results seems more easily explained if switch costs reflect a one-off stage-like process of task set reconfiguration at the beginning of switch trials.

Other studies, in contrast to Rogers and Monsell (1995), have found that RT does speed up over successive nonswitch trials (e.g. Meiran et al., 2000; Salthouse, Fristoe, McGurthy, & Hambrick, 1998). Thus, it is not a universal rule that switch costs are confined to the first trial in a run. Nevertheless, it still appears to be a challenge to task carryover accounts that switch costs can be confined to the first trial in a run, at least sometimes. Similarly the finding of larger costs of a switch into a dominant task (e.g. Allport et al., 1994), even if it is not universal, appears to challenge extra process accounts.

2.1 A Multiple *extra processes’

Most versions of the extra process account have posited more than one extra stage of processing on switch trials. These multiple stage-like control processes are typically assumed to have different characteristics and to be affected by different factors. Thus, support for such models has come from evidence suggesting that certain factors selectively affect specific stages involved in task switching, independently of any effect on the lower-level processes required to perform each task. For example, Rogers and Monsell (1995) posit two independent control processes required to switch tasks: a) an endogenous control process, and b) an exogenous control process.

Rogers and Monsell (1995, Experiment 3) supported this distinction by investigating the effects of manipulating the inter-trial interval in the alternating runs paradigm, and hence affecting the amount of time subjects had to prepare for the forthcoming task. They showed that switch costs were reduced over the first 600 ms or so of this preparation interval (from 207 to 130 ms). However, after a further 600 ms of preparation, a 'residual switch cost' of 115 ms remained. Rogers and Monsell (1995) argue that the reduction in switch costs over the first 600 ms or so of the preparation interval reflects the operation of the endogenous control process. The remaining switch cost at the longest preparation interval is hypothesised to correspond with the time taken by the execution of the exogenous control process, which must await stimulus presentation and is therefore insensitive to the preparation interval.

A similar account has been put forward by Rubinstein et al. (2001), who attribute switch costs to the duration of a 'goal shifting' stage, which can be executed before the arrival of a stimulus, and a 'rule activation' stage, which must await stimulus presentation. In order to support their account, Rubinstein et al. (2001) employ the additive factors logic (Sternberg, 1969,1998), which seeks to identify separable processing stages by investigating the effects of more than one factor on performance. If the effect of two factors on reaction time is additive, this is taken as evidence that the two factors affect discrete processing stages. If the factors interact, this is taken as evidence that there is at least one processing stage that is affected by both factors. First, Rubinstein et al. (2001, Experiment 1) seek to establish that control processes involved in task switching are independent of the processes involved in basic task performance. They show that stimulus discriminability affects reaction time equally in alternating-task and pure blocks. In contrast, the complexity of the classification rules (unidimensional or bidimensional) in the basic tasks had a greater effect on the completion time of

altemating-task blocks than repetitive-task blocks. Thus, some factors increase RT independently of the cost of task switching, whereas others boost this cost. Rubinstein et al. (2001) conclude from this that there are control processes involved in task switching which are separable from those involved in performance of the component tasks, since basic task performance and the cost of task switching seem to be affected by different factors (but see section 2.10 below).

Next, Rubinstein et al. (2001) attempt to fractionate the switch cost into components corresponding to goal shifting, which is assumed to be affected by the presence or absence of task cueing, and rule activation, which is assumed to be affected by the complexity of the tasks between which the subject must switch. Rubinstein et al. (2001, Experiment 2) show that task cueing and rule complexity have additive effects on switch costs. Thus, they argue that task cueing and rule complexity affect independent stages of task switching. In a related account, Mayr and Kliegl (2000) have proposed that one important process in task switching is the retrieval of the currently appropriate task from long term memory. This plays a similar role to the goal shifting stage in Rubinstein et al.'s (2001) account.