Running head : Remembering to prepare 1 Remembering to prepare : The benefits ( and costs ) of high working memory capacity

The dual mechanisms of control framework postulates that cognitive control can operate in two distinct modes: a ‘proactive’ preparatory mode and a ‘reactive’, wait-andsee mode. Importantly, the two modes are associated with both costs and benefits in cognitive performance. Here we explore this framework, in terms of its relationship with working memory capacity (WMC). We hypothesize that high WMC individuals are more likely to utilize proactive control yielding not only benefits, but also specific costs to performance. Across two separate, large-sample experiments, healthy young adults performed different variants of the AX-CPT context processing task, a well-established probe of proactive and reactive cognitive control. In two experiments, WMC predicted both improvements and relative impairments in task performance in a manner that was consistent with usage of proactive control. These findings suggest that individuals differ in the degree to which they utilize proactive control based on WMC. Individual differences in WMC and strategy 3 Remembering to prepare: The benefits (and costs) associated with high working memory capacity Executive functions, such as cognitive control and working memory abilities, are central to many everyday activities (P. Burgess, Alderman, Evans, Emslie, & Wilson, 1998; Cahn-Weiner, Boyle, & Malloy, 2002; Isquith, Gioia, & Espy, 2004). For example, working memory capacity (WMC) has been found to predict performance in a wide variety of cognitive domains including reasoning (Kyllonen & Christal, 1990), reading comprehension (Daneman & Carpenter, 1980) and fluid intelligence (Engle, Kane, & Tuholski, 1999). Similarly, cognitive control abilities have been shown to be related to theory of mind performance in preschool children (Carlson, Moses, & Claxton, 2004). In addition, a key component of cognitive control, interference resolution, has been shown to mediate the relation between WMC and fluid intelligence (G. Burgess, Gray, Conway, & Braver, 2011). However, there is still relatively little known about the specific subcomponents of cognitive control that can be explained by WMC. Specifically, the work presented here aims to investigate the extent to which WMC might modulate cognitive control in terms of the way that contextual information is strategically deployed in a proactive, or preparatory manner. The AX-Continuous Performance Test (AX-CPT) has previously been utilized to study cognitive control. In this task, contextual contingencies are built into the task such that specific cues indicate the appropriate response to be made to a subsequent stimulus. Specifically, the stimulus “X” is only a target when it follows an “A” cue (‘AX’ trial type). In contrast, an “X” stimulus following any non-A cue warrants a non-target response (termed ‘BX’ indicating a non-A-X stimulus sequence). Similarly, an “A” cue Individual differences in WMC and strategy 4 prior to any non-“X” stimulus warrants a non-target response (‘AY’ indicating an A-nonX sequence). Lastly, there are trials in which neither A nor X are present and these trials also warrant a non-target response (i.e. ‘BY’ trials indicating a non-A, non-X sequence). In other words, it is only the context provided by the cue that determines whether or not a target response to the X is appropriate. Thus, the task probes the processing and utilization of contextual cues in terms of how they lead to expectancies and response biases to upcoming stimuli (see Braver, et al., 2001 for an extended discussion of this task). Previous work utilizing this paradigm has shown that there are two primary control strategies that one can adopt within the AX-CPT (Braver, 2012). A proactive strategy is characterized by preparing a response to the probe (X or Y) based on information gleaned from the cue (A or B). Utilization of a proactive strategy on this task will lead to high accuracy for AX and BX trials, but may lead to increased interference on AY trials (e.g., lower accuracy or slower reaction times), given the increased difficultly in overriding the prepared target response. An alternative strategy has been termed reactive control. This strategy involves a sort of ‘wait-and-see’ mentality; participants wait to prepare a response until the probe is presented (X or Y) and then think back to the context set up by the cue (A or B) to determine the correct response. Here, participants are expected to perform Y-probe trials relatively successfully, but may exhibit poorer performance on AX and BX trials because sometimes the cue context will be incorrectly (or slowly) retrieved. A more familiar example may help solidify the main properties of the proactive versus reactive control distinction. Imagine that you are driving your vehicle at a high Individual differences in WMC and strategy 5 rate of speed when you notice that a car appears ready to pull out from a side street into your lane of traffic. In order to avoid an accident, you could engage in proactive control by preparing to swerve into the other lane based on the expectation that the car will in fact turn in front of you. Alternatively, you could just mentally note that the car is about to pull out without actually translating that into an avoidant maneuver. In this case, even if you have not prepared to swerve in advance, it may still be possible to use reactive control to switch lanes at the last second. However, it is likely that there is more of a chance of an accident in this latter case, as you would need to react very quickly to avoid hitting the turning car. Other variables may influence the likelihood of engaging in proactive or reactive control in this situation. For example, if you are driving a motorcycle, the consequences for not using predictive information from the driving environment could be more hazardous. In addition, you may have driven on this stretch of road previously and had an accident or near-accident in a similar situation. Accessing this previous event history may bias you to engage in proactive control mode to ensure that an error (an accident) does not occur (see Braver, 2012, for another outside-of-thelab example of proactive and reactive control). Healthy young adults with intact executive functioning generally exhibit a proactive control strategy in the AX-CPT (Braver, Cohen & Barch, 2007; Paxton, Barch, Storandt & Braver, 2006). Conversely, previous research has demonstrated that those with reduced executive functioning ability, such as children (Chatham, Frank & Munakata, 2009; Lorsbach & Reimer, 2008), older adults (Braver, et al., 2001; Paxton, et al., 2006) and people with schizophrenia (Barch, Carter, MacDonald, Braver & Cohen, Individual differences in WMC and strategy 6 2003; Dias, Butler, Hoptman & Javitt, 2011), generally exhibit a reactive strategy in response to this task. In general, high WMC has been widely associated with other positive cognitive outcomes (cf. Unsworth & Engle, 2007, Table 4). However, there is some evidence that even within healthy young adults, individual differences in WMC may account for differences in the use of proactive and reactive control; these differences may not only lead to performance benefits, but also performance costs. For example, compared to lowWMC individuals, participants with high WMC: (a) showed more forgetting of items from a ‘forget’ list in a directed forgetting task (Delaney & Sahakyan, 2007); (b) were more likely to miss hearing their name in the unattended channel of a dichotic listening task (Conway, Cowan, & Bunting, 2001); (c) exhibited a significantly smaller facilitation effect on the Stroop task (Kane & Engle, 2003, Experiment 1); and (d) performed worse on a surprise memory test for the neutral word stimuli from a previously completed Stroop task (Shipstead & Broadway, 2013). Interestingly, even in school-aged children this pattern of costs associated with high WMC has been observed. With respect to math achievement, students with high-WMC tend to rely on WM-intensive solution strategies. However, when these students also exhibit high levels of math anxiety, these strategies become less effective (Ramirez, Gunderson, Levine & Beilock, 2013; see also Beilock, 2008 for theoretical grounding of this account). Taken together, these results are largely consistent with the idea that individual differences in WMC might predispose some participants to use the proactive mode and others the reactive control mode. For example, within the context of the Stroop task (Kane & Engle, 2003), if high-WMC individuals are actively using a proactive strategy and keeping the task instruction in Individual differences in WMC and strategy 7 mind in advance of each trial, this is expected to result in selectively slowed performance on congruent trials relative to low-WMC individuals who appear to not engage in such preparation and instead quickly read the word when it appears onscreen. More germane for the current work, Redick and Engle (2011) administered the AX-CPT and found that young adults with low WMC made more AX and BX errors compared to high-WMC individuals. In addition, low-WMC participants were slower than high-WMC participants to respond on AX, BX and BY trials. Importantly, highand low-WMC participants did not differ in AY response times (RT), indicating that high-WMC participants were disproportionately slowed by the unexpected non-target stimulus (Redick & Engle, 2011). Additional work in this vein (Redick, 2014) tested a number of permutations of the traditional AX-CPT and found increased error rates (although no RT differences) specifically on AX and BX trials for lowversus highWMC participants. These results support the interpretation that low-WMC individuals are less likely than high-WMC individuals to maintain the cue information across time, leading to an increased error rate when an X stimulus is presented (Redick, 2014). The current research differs from, and extends Redick (2014) and Redick and Engle (2011) in several ways: (1) we manipulated the AX-CPT trial-type frequencies in a novel manner (see below for specific manipulations); (2) we utilized cue-based and probe-based signal detection indices of sensitivity and response bias; and (3) instead of using an extreme-groups approach (i.e. top 25% and bottom 25% of WMC), here we evaluate individual differences in WMC as a continuous variable (see Conway et al., 2005, for a discussion of the strengths and weaknesses of both approaches). The manipulation of trial-type frequencies is similar to the approach taken in a previous set of Individual differences in WMC and strategy 8 experiments investigating individual differences in WMC and their relation to Stroop interference (Kane & Engle, 2003). In addition, a new trial type is introduced in Experiment 2 to isolate preparatory processes. Specifically, the current investigation was aimed at investigating whether individuals with better cognitive control (as indexed by WMC) preferentially adopt a proactive strategy to respond quickly but accurately when presented with information that allows advance preparation. We predicted that such a strategy would reveal clear benefits in task performance on most trial types, but also relative costs on specific trial types, demonstrating theoretical specificity (i.e., as opposed to a pattern of generally superior performance).