Acta Psychologica 145 (2014) 1–9

Contents lists available at ScienceDirect

Acta Psychologica journal homepage: www.elsevier.com/ locate/actpsy

Cognitive control in context: Working memory capacity and proactive control Thomas S. Redick ⁎ Georgia Institute of Technology, United States Purdue University, United States

a r t i c l e

i n f o

Article history: Received 16 May 2012 Received in revised form 8 October 2013 Accepted 23 October 2013 Available online 14 November 2013 PsycINFO Codes: 2340 2343 2346 Keywords: Cognitive control Working memory Individual differences

a b s t r a c t Working memory is important for maintaining critical information in an active state to guide future behavior. The executive-attention theory of working memory capacity (WMC; Engle & Kane, 2004) argues that goal maintenance is important for response selection when stimuli are associated with competing responses. Braver, Burgess, and Gray (2007) have labeled this type of preparatory activity proactive control. Previous WMC studies have not allowed individuals to use goal information to prepare a specific response in advance of the stimulus. The current experiment used different versions of a cue-probe task to examine the relationship between individual differences in WMC and proactive control. Across three versions of the AX version of the Continuous Performance Test, the proportion of targets was manipulated to affect both the predictive validity of the A cue and the prepotency of the target response to X probes. The results indicated that the high-WMC individuals used the cue information to prepare responses in advance only when a specific probe was likely to occur. In contrast, the performance of the low-WMC individuals was less dependent upon the cue and more contingent upon overall response frequencies. The results indicate that individual differences in WMC are related to proactive control and anticipation, and important for translating cognition into action. © 2013 Elsevier B.V. All rights reserved.

1. Introduction

2. Executive-attention theory of working memory capacity

The notion of cognitive control has been invoked to explain performance in a number of activities. Broadly defined, cognitive control is the set of mental processes by which information is maintained in a temporary format to guide behavior towards task success, especially if there are competing alternative actions that could be selected instead of the desired target behavior. Braver, Gray, & Burgess (2007) (see also Braver, 2012) proposed a mechanistic account that attempts to synthesize the cognitive control literature. Specifically, their dualmechanism theory of cognitive control provides a framework for understanding both person- and task-related variations in controlled behavior. Their model is also influenced by knowledge of neurotransmitter and neuroanatomical properties of the human cortex observed in both typical and atypical biological functioning. The current research sought to apply Braver et al.'s model to cognitive control variation observed in individuals varying in working memory capacity (WMC). More specifically, (a) do individuals high in WMC use proactive control to anticipate and prepare a response more often than low-WMC individuals, and (b) do high-WMC individuals adjust their use of proactive control based on the predictive nature of the cue-related information?

Engle and colleagues (Engle & Kane, 2004; Kane, Conway, Hambrick, & Engle, 2007; Unsworth & Engle, 2007) have provided evidence that performance on complex working memory span tasks is predictive of behavior in a variety of situations. In a typical complex span task, such as Operation Span (Unsworth, Heitz, Schrock, & Engle, 2005), subjects must mentally solve math problems while also remembering letters for later recall. Variation in the ability to complete these types of tasks is used to measure individual differences in WMC. Not only do individuals high in WMC outperform those low in WMC on a variety of memory tasks, but they also show improved performance on several attention and inhibition tasks (for review, see Redick, Heitz, & Engle, 2007). According to the executive-attention theory (Engle & Kane, 2004; Kane et al., 2007), individual differences in WMC reflect variation in goal maintenance and response-conflict resolution.

⁎ Purdue University, Department of Psychological Sciences, 703 Third Street, West Lafayette, IN 47907, United States. Tel.: +1 765 494 6061. E-mail address: [email protected]. 0001-6918/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.actpsy.2013.10.010

3. The dual-mechanism theory of cognitive control While the executive-attention theory has been examined using young adults varying in WMC, Braver et al. (2005, 2007) proposed the dual-mechanism theory of cognitive control initially to account for cognitive aging deficits. The theory derives its name from the two modes of control that are assumed to be responsible for flexible, goaldriven behavior. Proactive control involves the active maintenance of information that will help to respond appropriately to upcoming stimulus

2

T.S. Redick / Acta Psychologica 145 (2014) 1–9

events. This context information could be general task instructions, the identity of previous stimuli, or the relevant information conveyed by previous stimuli or cues that are salient for later behavior. The second control type is reactive control. Reactive control involves the reactivation or retrieval of context information that is imperative for the current decision-making; however, reactive control is only engaged in response to the probe or critical stimulus. Proactive control is important for sustaining prior information to bias future responding in a way consistent with expectancies and the schedule of rewards and punishments. However, there may be instances where there is either no predictive information available to help prepare one action versus another, or the cue or warning information that is available for use is unreliable. In these situations, the individual should rely upon reactive control to respond accurately. Braver et al. (2005) used the AX version of the Continuous Performance Test (AX-CPT) to examine proactive and reactive control modes. Although in the AX-CPT, individual letters are presented visually one at a time, the task is best understood as a series of letters presented in a cue-probe sequence. Subjects are instructed to make a target keypress when the probe letter X immediately follows the cue letter A (Fig. 1). AX target sequences occur on the majority of trials (70% of all cueprobe sequences), so an expectancy to make a target response is created when the letter A is presented as a cue. Because AX trials occur on 70% of all trials, this version is denoted hereafter as the AX-CPT-70. There are three other possible nontarget trial types on the AX-CPT-70, each occurring on 10% of trials. On an AY trial (where Y stands for all non-X letters as probes), the cue A is not followed by an X probe. In addition, on BX trials (where B stands for all non-A letters as cues), the probe letter X follows a letter other than A. Finally, on BY trials, letters other than A and X are presented sequentially to serve as a baseline condition, because neither the cue nor the probe is associated with the target response (see Table 1 for more information about the probability of the different types of cues, probes, and conditional probabilities for different cue-probe sequences and responses). As Braver et al. (2007) state, “In the AX-CPT, proactive control means control engaged by the cue, whereas reactive control means control driven by the probe” (p. 81). Thus, subjects using proactive control more often will lead to fewer errors specifically on AX and BX trials, and will also speed correct responses on AX and BX trials, because they have used the cue information to anticipate and prepare a response to the subsequent probe during the cue-probe interval. However, if a subject is engaging in proactive control on an AY trial, he must stop the prepared target response and execute a nontarget response. Thus, AY trials should be more error-prone, and slower for correct trials, because the expected target stimulus does not occur. In contrast,

1000 ms Wait

A

500 (1000) ms

Cue 4500 ms Delay

X

500 (1000) ms

Probe Fig. 1. Example AX target trial sequence from the AX-CPT. For cue and probe screens, the letter was displayed for 500 ms, and the numbers in parentheses (1000 ms) indicate the amount of time since the onset of the letter that the subject had to respond to each letter (a response deadline of 500 ms after the offset of the letter).

Table 1 Stimulus probabilities and predicted control mode for the versions of the AX-CPT. Type Freq p(cue) p(probe) p(probe| cue)

p(targ| cue)

p(targ| probe)

Optimal Mode

AX-CPT-70 AX 70%

.8

.8

.875

.875

.875

AY

10%

.8

.2

.125

.875

.000

BX

10%

.2

.8

.500

.000

.875

BY

10%

.2

.2

.500

.000

.000

Proactive (targ) Proactive (targ) Proactive (non) Proactive (non)

AX-CPT-10 AX 10%

.8

.2

.125

.125

.500

AY

70%

.8

.8

.875

.125

.000

BX

10%

.2

.2

.500

.000

.500

BY

10%

.2

.8

.500

.000

.000

AX-CPT-40 AX 40% AY 40% BX 10%

.8 .8 .2

.5 .5 .5

.500 .500 .500

.500 .500 .000

.800 .000 .800

BY

.2

.5

.500

.000

.000

10%

Proactive (non) Proactive (non) Proactive (non) Proactive (non)

Reactive Reactive Proactive (non) Proactive (non)

Note. Type: Trial Type; Freq: Frequency; p(cue): probability that the first letter in the Type column appears as a cue; p(probe): probability that the second letter in the Type column appears as a probe; p(probe|cue): conditional probability that the type of probe appears given the type of cue for that trial type; p(targ|cue): conditional probability that a target response is the correct response given the type of cue for that trial type; p(targ|probe): conditional probability that a target response is the correct response given the type of probe for that trial type; targ: target response; non: nontarget response.

individuals not engaging in proactive control do not prepare a response during the cue-probe delay, and thus must rely upon a transient reactivation of the cue information when the probe appears. Therefore, subjects engaging less often in proactive control will commit more errors specifically on BX trials, and have slower correct RTs on AX and BX trials. However, on AY trials, not preparing a target response based on the A cue should actually help performance. Thus, when a letter such as F appears as the probe, the subject can and should relatively quickly and accurately respond that it is not a target. 4. Individual differences in WMC as variation in proactive control Across multiple studies (Braver, Paxton, Locke, & Barch, 2009; Braver et al., 2001, 2005; Paxton, Barch, Racine, & Braver, 2008; Paxton, Barch, Storandt, & Braver, 2006), older adults' performance on the AX-CPT-70 was consistent with reactive control (spared AY performance in either errors or correct RTs relative to young adults, but impaired BX performance), while the young adult groups exhibited performance consistent with engaging in proactive control (fast and accurate AX and BX performance). In addition, Braver et al. (2007) suggested that “individuals with high-WM span… should thus show an increased tendency to use proactive control strategies, but only in the task demands that most require and benefit from such strategies” (p. 89). Moreover, Kane et al. (2007) argued that the “executive-attention view…parallels the dual mechanisms of cognitive control proposed by Braver et al.” (p. 44). That is, although previous research suggests that young adults use proactive control more often than older adults, individual differences in WMC may explain variation in cognitive control observed within young adults. Previous research using a cued-antisaccade task (Unsworth, Schrock, & Engle, 2004), cued-visual search task (Poole & Kane, 2009), and versions of a go/no-go task where subjects must use previous stimuli to determine future targets (Redick, Calvo, Gay, & Engle, 2011) all

T.S. Redick / Acta Psychologica 145 (2014) 1–9

demonstrated that high-WMC young adults performed in a manner indicating they were more likely than low-WMC individuals to use cues to guide response selection to subsequent stimuli. More germane for the current research, Redick and Engle (2011) administered a version of the AX-CPT-70 that intermixed short (1000 ms) and long (5000 ms) intervals between the cue and probe letters. The low-WMC group made more errors and was slower to respond on AX and BX trials compared to the high-WMC group, but did not differ in AY or BY performance. The AX and BX results suggest that low-WMC individuals were hurt by not maintaining the cue information to respond appropriately to X probes, and the spared AY performance confirms that the low-WMC individuals were less likely to use proactive control to prepare a response based on the A cue. 5. Current research Despite the evidence just reviewed, there are still several unanswered questions in regard to how individual differences in WMC are related to the use of proactive control. Redick and Engle (2011) examined how the length of cue-probe interval affected high- and low-WMC individuals' AX-CPT-70 performance. In the current research, young adults who had been previously classified as high- or low-WMC were tested on three different versions of the AX-CPT. I manipulated the proportion of AX trials to affect both the predictive validity of the A cue and the prepotency of the target response to X probes. The AX-CPT provides a way to systematically manipulate target frequency, which should affect the use of the cue to anticipate and plan responses to subsequent probes. A specific advantage of the AX-CPT design is that the cue information in the AX-CPT allows biasing of distinct responses to the various probes across the different trial types, and provides a way to measure proactive control more specifically compared to other cognitive control tasks (e.g., Stroop, antisaccade, task-switching) where preparation of a specific response in advance is not possible. At best in the latter type of tasks, the subject can sustain the goal information and perhaps activate all possible responses, but not plan a particular motor response. 5.1. AX-CPT-70 Data obtained via computational simulations and empirical studies indicate that optimal performance on the AX-CPT-70 is produced via a proactive strategy (Braver et al., 2007). Cue information guides the appropriate responding to probes, and maintaining this context over the cue-probe delay will aid subjects in responding quickly and accurately on all trial types except for the rare occasion when a Y probe follows an A. Overall, the AX-CPT-70 should induce a global context of proactive control (Table 1), because the A cue is strongly predictive of a target response to the subsequent probe (conditional probability = .875), and the B cue is perfectly predictive of a nontarget response to the subsequent probe (conditional probability = 1.000). I predicted that lowWMC individuals would have difficulty responding correctly on BX trials because they failed to maintain the B cue information, and because the X probe is strongly associated with a competing target response. In addition to poorer BX performance, the failure to prepare a response during the cue-probe interval should manifest as decreased performance on AX trials by low-WMC individuals compared to high-WMC individuals. The two WMC groups were expected to show comparable AY and BY performance on the AX-CPT-70 (Redick & Engle, 2011). 5.2. AX-CPT-10 A separate group of high- and low-WMC subjects completed the AXCPT-10 (Table 1), in which the frequencies for AX and AY trials were switched, such that now AX targets occurred on 10% of trials, whereas AY nontargets occurred on 70% of all trials. BX and BY nontargets each occurred on 10% of all trials (see Dias, Foxe, & Javitt, 2003, for similar design). In this way, the AX targets occur much less frequently than the

3

AX-CPT-70 just discussed, but in both versions the A cue is equally predictive of a specific response. The global context and local context of the AX-CPT-10 should lead to the use of proactive control to prepare a nontarget response during all cue-probe intervals (Table 1). The A cue is strongly predictive of a nontarget response to the subsequent probe (conditional probability = .875), and the B cue is perfectly predictive of a nontarget response to the subsequent probe (conditional probability = 1.000). If this interpretation of the AX-CPT-10 is correct, the high-WMC group should use the cues in advance on all trials to prepare nontarget responses, and thus show better AY and BX performance than the low-WMC group, who is not preparing responses based on the cue information. However, a different possibility exists for the AX-CPT-10. Nontargets occur on 90% of all trials (AY, BX, and BY), and thus individuals could respond correctly on probes 90% of the time even if they were ignoring the cue information. If low-WMC subjects are less likely to use the cue information to prepare responses anyway, they could habitually make a nontarget response without processing the cue. If this account of low-WMC behavior is correct, the low-WMC subjects would only show a performance deficit on the AX target trials. The low-WMC subjects would not be classified as being proactive because they are not using the cue to prepare their responses, they are simply responding according to the overall response probabilities — a target response is correct on only 4 out of 40 trials in every block.

5.3. AX-CPT-40 A third version of the AX-CPT was designed to force high-WMC individuals into a situation in which preparing an advance response based on the A cue was counterproductive. In the AX-CPT-40, AX targets and AY nontargets each occurred on 40% of trials, and BX and BY nontargets each occurred on 10% of all trials (Table 1). The advantage of the AXCPT-40 is that the A cues carry different predictive validity compared to the previous two versions of the AX-CPT. Here, A cues predict a target or nontarget response with equal probability (conditional probability = .500). Contrast this with A cues from the previous two task versions, which served as cues that predicted a specific response with a conditional probability = .875, and with B cues in each task version, which are 100% predictive of a subsequent nontarget response. In the AX-CPT-40, there is no consistent control strategy that should be employed to produce optimal performance, and although the frequency of A cues here is the same as in the previous two versions, now reactive control is expected to be engaged in response to A cues. This condition critically assesses the aspect of the dual-mechanism theory that “cognitive flexibility can also be achieved by modulating the manner in which a particular control mechanism is deployed in response to changing task demands or internal goal states” (p. 7351, Braver et al., 2009). In the AXCPT-40, the cue-probe contingencies are set up to study within the same task whether or not cognitive control can dynamically shift between the optimal control states, and whether or not this ability differs for highand low-WMC individuals. That is, are high- and low-WMC individuals sensitive to the differential predictive nature of the A and B cues, and can they adapt their cognitive control on a trial-by-trial basis as necessary? On the AX-CPT-40, the high-WMC group is predicted to be able to use a reactive control mode in response to A cues and a proactive control mode in response to B cues. Thus, the results should show that performance on AX and AY trials do not differ for the high-WMC group, and BX performance should continue to be highly accurate. For the lowWMC group, BX trials should be more error-prone than the high-WMC group. In addition, the low-WMC group is predicted to have more difficulty switching between proactive and reactive control as necessary. If the WMC groups are equivalent when using reactive control, the lowWMC group should show performance on AX and AY trials that is qualitatively similar to that of the high-WMC group. However, if the high-

4

T.S. Redick / Acta Psychologica 145 (2014) 1–9

WMC group is also better able to use reactive control, the low-WMC group will show impaired performance even after an A cue.

the task. The dependent variable for Operation Span is the number of letters recalled in the correct serial position across all trials.

6. Method

6.2.2. Symmetry Span The task structure is similar to that of the Operation Span just described, with the following exceptions. First, instead of remembering letters, subjects are presented with a 4 × 4 matrix of blank squares, with one square colored in red on a given trial. At recall, the subject must indicate the location of the squares within the matrix that were colored for that trial. Second, instead of solving math problems, subjects make vertical symmetry decisions about an 8 × 8 figure composed of black and white squares. Finally, the list length can vary between two and five, for a total of 42 symmetry figures and square locations on the task. The dependent variable for Symmetry Span is the number of square locations recalled in the correct sequential order across all trials.

6.1. Subjects All subjects were between 18 and 30 years of age. Out of 122 young adults that completed the study, 60 high- and 60 low-WMC individuals were included in the final sample, with 20 high- and 20 low-WMC subjects in each version of the AX-CPT. One low-WMC subject in the AXCPT-70 and one low-WMC subject in the AX-CPT-40 was classified as an outlier due to extremely low accuracy on AX trials, which led to an overall accuracy on probe trials of less than 70%. Trial type frequencies were manipulated between subjects and kept constant within an individual to prevent possible task order effects that could complicate the interpretation of the results (Kane, Bleckley, Conway, & Engle, 2001; Kane & Engle, 2003; Unsworth et al., 2004). Demographic information about the final sample is presented in Table 2. All subjects were examined for the following exclusionary criteria: (a) currently suffering from a major illness that affected the subject's attention or memory; (b) currently taking medication that impaired the subject's attention or memory; (c) history of head injury or trauma; (d) non-English native speaker; and (e) poor visual acuity (less than 20/50 corrected). 6.2. WMC screening In a previous session, subjects were classified as having high- or lowWMC on the Operation and Symmetry Span tasks. Performance on each task was transformed into a z-score based on a database of over 4000 scores from the laboratory over the previous five years. A WMC composite was created by averaging across the two complex span tasks' zscores for each subject. If a subject's WMC composite score fell within the upper or lower quartiles compared to the database, the individual was invited to participate in a second session in which one version of the AX-CPT was administered. In both sessions, subjects were tested in individual rooms and seated at a comfortable distance from the computer screen. Each WMC task is described briefly below; for further information, see Redick et al. (2012). 6.2.1. Operation span In this task, individuals must mentally compute the results of mathematical problems while also concurrently remembering to-be-presented letters for later recall. After completing the practice, the subject completes 15 arithmetic-and-letter trials. Within each trial, between three and seven math problems and letters are presented, with the exact number of items on any given trial unknown to the subject. There are three trials of each list length, for a total of 75 math problems and letters on

Group

Z-WMC

AX-CPT-70 High Low

0.88 (0.18) −1.18 (0.55)

9/11 5/15

23.0 (3.1) 23.1 (4.2)

AX-CPT-10 High Low

0.82 (0.14) −0.97 (0.43)

11/9 8/12

21.8 (2.6) 22.3 (3.1)

AX-CPT-40 High Low

0.92 (0.16) −1.14 (0.56)

7/13 8/12

23.3 (3.2) 23.7 (3.9)

Note. SDs are in parentheses.

The general materials and stimuli presentation for the three versions of the AX-CPT utilized are described first, and then the information specific to each version is presented subsequently. As seen in Fig. 1, each trial began with the presentation of a blank screen for 1000 ms, followed by a letter (cue) for 500 ms. Next, a blank interstimulus interval of 4500 ms occurred, followed by another letter (probe) for 500 ms. A beep served as auditory feedback if the subject did not make a response to stimuli within 1000 ms of onset of each letter. All letters except vowels were used as possible cue and probe stimuli. A serial response box (Psychology Software Tools) was used to collect responses. All subjects used the index fingers of the left and right hand to respond to nontargets and targets, respectively. Each subject performed a practice block of 40 trials similar to the experimental blocks that followed. The only exception was that each practice trial was followed by auditory accuracy feedback. During the practice block, subjects had to achieve 75% accuracy to move on to the experimental blocks, which all subjects achieved on the first attempt. Each of the 10 experimental blocks contained 40 trials, containing the specific target and nontarget trial types according to the version of the AX-CPT being performed (Table 1). The instructions were identical for all versions of the AX-CPT, except for one instruction screen that varied. In the AX-CPT-70, this instruction screen stated: “The letter X will follow the letter A most, but not all, of the times that the letter A appears. Therefore, you should expect an X to appear next if you see an A.” In the AX-CPT-10, the instruction screen stated: “The letter X will NOT follow the letter A most, but not all, of the times that the letter A appears. Therefore, you should NOT expect an X to appear next if you see an A.” In the AX-CPT-40, the instruction screen stated: “The letter X will follow the letter A half of the times that the letter A appears. Therefore, you should expect an X to appear next if you see an A half of the time.” 6.3.1. AX-CPT-70 In the AX-CPT-70, AX targets occurred on 70% of all cue-probe trials. The remaining nontarget trial types (AY, BX, BY) each occurred on 10% each of cue-probe trials.

Table 2 Demographic information for the high- and low-WMC subjects. M/F

6.3. AX-CPT

Age (in years)

6.3.2. AX-CPT-10 In the AX-CPT-10, the frequency of AX targets and AY nontargets was switched. That is, each block contained 28 AY nontargets and 4 trials each of AX targets and BX and BY nontargets. 6.3.3. AX-CPT-40 In the AX-CPT-40, the frequencies of AX target and AY nontargets were equal. Each of the 10 blocks contained 16 trials each of AX targets and AY nontargets and 4 trials each of BX and BY nontargets.

T.S. Redick / Acta Psychologica 145 (2014) 1–9

5

6.4. Design and analyses Although the results focus on the performance on probes, cue accuracy was also assessed to ensure general compliance with task instructions.1 For both the probe accuracy and correct mean RT results, an omnibus analysis of variance (ANOVA) was first conducted with WMC (Braver, 2012) and Version (Braver et al., 2001) as between-subjects factors, and Trial Type (Braver et al., 2007) as the within-subjects factor. Because the predictions were tailored to examine the effect of manipulating the cue and target frequencies and conditional probabilities as a function of WMC, follow-up split-plot ANOVAs with WMC (Braver, 2012) as the between-subjects factor, and Trial Type (Braver et al., 2007) as the within-subjects factor, were conducted separately for AX-CPT-70, AX-CPT-10, and AX-CPT-40.2 In addition, follow-up analyses were conducted for each WMC group separately to examine the relative performance on the different trial types across the AX-CPT-70, AX-CPT-10, and AX-CPT-40. Accuracy performance was assessed via split-plot ANOVAs with Version (Braver et al., 2001) as the between-subjects factor, and Trial Type (Braver et al., 2007) as the within-subjects factor, and follow-up tests were used to examine significant differences in accuracy on each trial type across task versions. An alpha of p b .05 was used for all statistical tests. Partial eta-squared (η2p) is provided as index of effect size. 7. Results 7.1. Accuracy comparison within each AX-CPT version The full omnibus ANOVA output is provided in Table A.1. The threeway interaction was significant, F(6, 342) = 4.03, p = .001, partial η2 = .066, and follow-up analyses were conducted by examining performance separately for each level of Version and subsequently for each level of WMC to decompose this interaction. 7.1.1. AX-CPT-70 Errors on the AX-CPT-70 are presented in Fig. 2a. The high-WMC group appeared to make fewer errors on AX trials and BX trials than the low-WMC group. The main effect of Trial Type was significant, F(3, 114) = 18.77, p b .001, η2p = .331, and although the main effect of WMC was significant, F(1, 38) = 6.14, p = .018, η2p = .139, the WMC × Trial Type interaction did not approach significance (F b 1). Although the WMC × Trial Type interaction was not significant, in order to fully compare the AX-CPT-70 results to both previous work (Redick & Engle, 2011) and the other AX-CPT versions in the current study, the high- versus low-WMC group was contrasted for each trial type separately. Independent-samples t-tests indicated that the highWMC subjects were more accurate than low-WMC subjects on AX trials, t(38) = 2.53, p = .016, marginally more accurate on BX trials, t(38) = 2.00, p = .053, and more accurate on BY trials, t(38) = 2.08, p = .044. However, the WMC groups did not differ in accuracy on AY trials (t b 1). Note that the WMC group difference on BY trials reflects a total of 12 errors for the low-WMC group and 4 errors for the highWMC group out of a total of 800 BY trials for each group; therefore, this result will not be discussed further. Overall, the accuracy results of AX-CPT-70 are consistent with the notion that high-WMC individuals

1 Cue accuracy was 99% for each WMC group across the three task versions. In order to maximize the number of observations, performance was evaluated on all trials regardless if the cue was correct or not. Restricting analyses to only those trials in which the cue was responded to correctly did not change the results. 2 Probe accuracy analyses were also conducted after applying an arcsine transformation. Statistical results were the same as those presented in the text using the untransformed accuracy rates. Specifically, significant main effects of Trial Type and WMC were observed for each version of the AX-CPT, and significant WMC × Trial Type interactions were observed for the AX-CPT-10 and AX-CPT-40. Given the similarity of the results, the untransformed analyses are reported in the text.

Fig. 2. Errors in AX-CPT-70 (A), AX-CPT-10 (B), and AX-CPT-40 (C) for each WMC group as a function of trial type. Error bars represent ±1 SE of the mean. *Indicates p b .05; †indicates p = .05.

are more likely to proactively control behavior by preparing responses in advance of critical stimulus events.

7.1.2. AX-CPT-10 Errors on the AX-CPT-10 are presented in Fig. 2b. Few errors were made across all trial types, except that the low-WMC group made substantially more AX errors than the high-WMC group. The main effect of Trial Type was significant, F(3, 114) = 91.06, p b .001, η2p = .706, as was the main effect of WMC, F(1, 38) = 13.39, p = .001, η2p = .261. However, these effects were qualified by a significant WMC × Trial Type interaction, F(3, 114) = 15.04, p b .001, η2p = .284. Independent-samples t-tests indicated that high-WMC individuals were significantly more accurate than low-WMC individuals on AX trials, t(38) = 3.93, p b .001. However, the WMC groups did not differ in accuracy on all other trial types (all t's b 1). The accuracy results of the AX-CPT-10 showed that high-WMC individuals will continue to use proactive control, even when a much simpler strategy not based on cue information would lead to accurate performance on nearly all trials.

6

T.S. Redick / Acta Psychologica 145 (2014) 1–9

7.1.3. AX-CPT-40 Accuracy results for probe trials on the AX-CPT-40 are presented in Fig. 2c. Low-WMC subjects appeared to make more errors than the high-WMC group on AX and BX trials. Significant main effects of Trial Type, F(3, 114) = 28.17, p b .001, η2p = .426, and WMC, F(1, 38) = 7.94, p = .008, η2p = .173, were qualified by a significant WMC × Trial Type interaction, F(3, 114) = 7.13, p b .001, η2p = .158. Independent-samples t-tests indicated that the high-WMC group was significantly more accurate than the low-WMC group on AX trials, t(38) = 2.91, p = .006, and BX trials, t(38) = 2.04, p = .048. However, the two WMC groups did not differ in accuracy on AY or BY trials (both p's N .184). In the AX-CPT-40, even when advance preparation via proactive control was disadvantageous, high-WMC individuals still produced fewer AX errors than the low-WMC individuals. In addition, low-WMC subjects made more BX errors than high-WMC subjects, despite the fact that the B cue was entirely predictive of the response to the subsequent probe. 7.2. Accuracy comparison across AX-CPT versions 7.2.1. High WMC Accuracy for the high-WMC group was examined as a function of Version and Trial Type. Looking at Fig. 2, errors varied as a function of the predictive validity of the A cue, whereas errors did not differ when the cue was a B. Although the main effect of Version was not significant, F(2, 57) = 1.86, p = .165, η2p = .061, the main effect of Trial Type was significant, F(3, 171) = 19.98, p b .001, η2p = .260. Critically, the Version × Trial Type interaction was significant, F(6, 171) = 18.69, p b .001, η2p = .396. For AX trials, AX-CPT-10 errors were significantly higher than both the AX-CPT-40 and AX-CPT-70 (both p's b .001), which did not differ from each other (p = .231). For AY trials, AX-CPT-70 errors were significantly higher than both AX-CPT-40 and AX-CPT-10 (both p's b .001), which did not differ from each other (p = .653). When the cue was a B, and thus perfectly predictive of the subsequent response, BX errors did not differ across the task versions (all p's N .183). If one is using the cue to prepare a response in the cue-probe interval, as the highWMC individuals appear to be doing, then they would be primarily affected by the cue validity manipulation. Thus, the high-WMC subjects' BX performance across all three AX-CPT versions is the same, and their AX and AY performance is essentially reversed across the AXCPT-70 and AX-CPT-10. 7.2.2. Low WMC Accuracy for the low-WMC group was analyzed as a function of Version and Trial Type. Looking at Fig. 2, the low-WMC group made many AX errors on the AX-CPT-10 and AX-CPT-40. They also appeared to make fewer BX errors on the AX-CPT-10 than on the other two task versions. Although the main effect of Version was not significant, F(2, 57) = 1.12, p = .332, η2p = .038, the main effect of Trial Type was significant, F(3, 171) = 38.40, p b .001, η2p = .402. Again, the Version × Trial Type interaction was significant, F(6, 171) = 21.77, p b .001, η2p = .433. For AX trials, AX-CPT-10 errors were significantly higher than both the AX-CPT-40 (p = .002) and AX-CPT-70 (p b .001), and the difference between AX-CPT-40 and AX-CPT-70 was marginally significant (p = .052). For AY trials, AX-CPT-70 errors were significantly higher than both the AX-CPT-40 and AX-CPT-10 (both p's b .001), which did not differ from each other (p = .905). For BX trials, AX-CPT-10 errors were significantly lower than AX-CPT-70 (p = .002) and AX-CPT-40 (p = .048), and there was no difference between AX-CPT-70 and AXCPT-40 (p = .514). Overall, the low-WMC individuals showed accuracy differences that varied as a function of the probability that the X probe required a target response, despite the fact that for the B cue, the predictive validity of the cue was unchanged across the versions of the AXCPT.

7.3. Response times The full omnibus ANOVA output is provided in Table A.1. The threeway interaction was not significant (F b 1). In addition, when examining each version of the AX-CPT separately, WMC was not involved in any significant main effects or interactions with Trial Type. Mean RTs on the AX-CPT are presented in Fig. 3, but given the lack of WMC effects, the RT results for each AX-CPT version are presented in the Appendix A. 8. Discussion Across three experiments, the accuracy analyses produced several significant effects involving WMC. The analyses indicated that, compared to high-WMC individuals, low-WMC individuals made more: (a) AX errors on the AX-CPT-70, AX-CPT-10, and AX-CPT-40; and (b) BX errors on the AX-CPT-70 and AX-CPT-40. That is, across the various versions of the AX-CPT, the significant WMC differences in accuracy occurred when the probe decision involved the letter X. In all task versions, correctly responding to the probe letter X depends upon remembering the most recent cue letter (A or B), whereas the correct

Fig. 3. Mean RTs in AX-CPT-70 (A), AX-CPT-10 (B), and AX-CPT-40 (C) for each WMC group as a function of trial type. Error bars represent ±1 SE of the mean. †Indicates p = .05.

T.S. Redick / Acta Psychologica 145 (2014) 1–9

response to any other probe letters (Y) can be made without regard to the specific cue that preceded the probe letter. The findings are consistent with a specific impairment for low-WMC individuals, instead of a generalized cognitive deficit (more error-prone or slower across all trial types). The implication of the current results is that what differentiates the two WMC groups is the ability to use memory of local context (cue stimuli) to guide behavior according to global context (task instructions and goals). 8.1. BX performance and individual differences in WMC Low-WMC individuals made more BX errors than high-WMC individuals on both the AX-CPT-70 and AX-CPT-40, but not on the AXCPT-10. Importantly, BX performance is particularly indicative of whether or not an individual is maintaining the cue and/or using the cue to prepare a response in advance of the probe (Braver et al., 2001; Cohen, Barch, Carter, & Servan-Schreiber, 1999). The B cue was always perfectly predictive of a nontarget response to be made to the subsequent probe. Using this information in advance of the probe stimulus is important on BX trials, where the letter X appears as the probe. Across all task versions, the letter X is associated both with making a target and a nontarget response. Thus, there is response uncertainty associated with the letter X, and it is only in conjunction with the cue that there is absolute information to respond correctly. However, the amount of response uncertainty varies across AX-CPT versions, which helps explain the specific pattern of BX errors for the high- and low-WMC subjects. On the AX-CPT-70, BX errors could be a consequence of either: (a) an inability to maintain or retrieve the representation for the preceding cue (local context); or (b) difficulty in preventing a target response, given the high frequency of AX target responses (global context). Thus, the worse BX performance by the low-WMC group on this task version is insufficient to confirm that they are less likely to maintain or remember the cue. However, the increased BX errors by the low-WMC group on the AX-CPT-40 do differentiate between these two alternatives. BX errors on this task version should be specifically due to not maintaining or remembering the cue, because the overall response frequency actually favored a nontarget response (AX targets on 40% of trials, nontargets on the other 60%). If BX errors indicate a failure of maintaining the cue information, why do the high- and low-WMC groups not differ in BX accuracy on the AX-CPT-10? The answer is related to the response frequencies associated with the various stimuli. On the AX-CPT-10, a target response is correct on only 10% of all probe stimuli (AX targets on 10% of trials, nontargets on the other 90%). Thus, the response that is likely to be executed by default by low-WMC individuals would be a nontarget response, even if the B cue information has not been used to specifically prepare a nontarget response in advance. The high-WMC individuals also produce few BX errors, but that is because they are using the B cue to prepare a nontarget response during the cue-probe interval. Therefore, although BX performance is similar for the two WMC groups on the AX-CPT-10, the means by which the subjects achieve their comparable performance is different. The comparison of BX performance across AX-CPT versions for each WMC group separately is also informative. The high-WMC subjects showed similar BX error rates across the three task versions. In all AXCPT versions, B cues were 100% predictive of a subsequent nontarget response to the upcoming probe. Thus, high-WMC subjects used the B cue to prepare a nontarget response in all three AX-CPT versions. In contrast, low-WMC subjects committed significantly more BX errors on the AX-CPT-70 and AX-CPT-40 compared to the AX-CPT-10. If lowWMC individuals use proactive control less often, and thus ignore or fail to maintain the cue information, then why would BX performance differ across the AX-CPT versions? My interpretation is that, in the absence of preparing a nontarget response during the cue-probe interval, low-WMC subjects were more influenced than high-WMC subjects by

7

the overall target/nontarget response frequencies. Therefore, in the AX-CPT-70, 70% of all probe stimuli should get a target response, and in the AX-CPT-40, 40% of all probe stimuli should get a target response. In contrast, in the AX-CPT-10, only 10% of all probe stimuli should get a target response, meaning there is less target response prepotency associated with the probe letter X in the AX-CPT-10 compared to the other two versions. Again, the B cue validity does not change across task versions, and so the high-WMC subjects' BX performance is constant — but the low-WMC group, which is suspected of less frequently using the B cue to prepare advanced responses, shows varied BX performance across the AX-CPT versions used here. 8.2. AX performance and individual differences in WMC The low-WMC subjects made more AX errors than the high-WMC subjects across all three experiments. Across AX-CPT versions, AX targets occurred on 70%, 10%, or 40% of all trials. AX errors reflect making an incorrect nontarget response to the probe letter X. Correct AX trials reflect a mixture of sometimes using the cue to prepare a target response in advance, other times correctly retrieving the cue upon presentation of the probe, and the influence of guessing based on the probability of making a target response to an X (Table 1). Because the probe X almost always follows the cue A in the AX-CPT70, AX errors should reflect not actively maintaining the cue and instead an incorrect retrieval of the cue upon presentation of the probe X. This is especially the case in the AX-CPT-40, where there is no advantage of selectively preparing a target or nontarget response based on the A cue. Note that even when the two WMC groups are assumed to be behaving similarly during the cue-probe interval (i.e., not preparing a response in advance), the low-WMC group still produces more AX errors than the high-WMC group. This suggests that not only are high-WMC individuals more likely to use proactive control to prepare a response based on cue validity when available (AX-CPT-70 and AX-CPT-10), they are also more likely to either maintain the cue information in a ready state to guide behavior and/or effectively retrieve the cue in order to select the appropriate probe response when forced to do so (AX-CPT-40). The comparison of AX performance across AX-CPT versions for each WMC group separately illustrates further differences between high- and low-WMC individuals. Specifically, only low-WMC subjects committed more AX errors on the AX-CPT-40 (19.0%) compared to the AX-CPT-70 (6.9%). Again, if one views low-WMC individuals' behavior as less reliant on the cue information, and more influenced by overall target/nontarget response frequencies, then the low-WMC subjects more than the highWMC subjects should be affected by the differences between these two AX-CPT versions in the proportion of target responses. AX targets occur on 70% of trials in the AX-CPT-70, so guessing or making a target response to a probe letter X will produce the correct answer more often than not. However, in the AX-CPT-40, AX targets happened on 40% of trials compared to 60% of trials that were nontargets, and thus lowWMC subjects' would be expected to be more likely to make a nontarget response in the AX-CPT-40 versus the AX-CPT-70. 8.3. WMC and the dual-mechanism control theory The pattern of AX-CPT performance is consistent with existing theories of individual differences in WMC. Engle and Kane (2004) argued that high-WMC individuals are better at both goal maintenance and response conflict resolution. In addition, Unsworth and Engle (2007) have shown that high-WMC individuals are better at maintaining goalrelevant information in primary memory and also retrieving specific information from secondary memory. The wrinkle provided by applying the dual-mechanism theory of context-processing to WMC is that this view additionally predicts that high-WMC individuals translate the goal maintenance into a prepared action before the critical event, when possible.

8

T.S. Redick / Acta Psychologica 145 (2014) 1–9

The accuracy results of the high-WMC group across all three experiments are largely consistent with the idea that these individuals are engaging in control according to the predictive information conveyed by the cue. In the AX-CPT-70, the high-WMC group made the most errors on AY trials, consistent with the idea that they used the A cue to expect an X as the subsequent probe, and thus prepared a target response during the cue-probe interval. In the AX-CPT-10, the pattern reversed, with the high-WMC group producing the most errors on AX trials. In this version, the A cue was strongly predictive of a subsequent Y as the probe. Thus, the high-WMC subjects appeared to prepare a nontarget response during the cue-probe interval, and the proportion of AX errors committed in the AX-CPT-10 was strikingly similar to the proportion of AY errors committed in the AX-CPT-70. Finally, in the AX-CPT-40, the high-WMC group made a similar amount of errors on AX and AY trials. Although the number of AX errors was statistically greater than the number of AY errors, the difference between the two trial types was much smaller compared to the AX-CPT-70 and AX-CPT-10. Because the A cue was equally predictive of a subsequent target or nontarget response in the AX-CPT-40, differentially preparing one response during the cue-probe interval would be largely counterproductive. This manipulation was introduced to prevent high-WMC individuals from preparing a specific response during the cue-probe interval, and compared to the AX and AY performance on the AX-CPT-70 and AX-CPT-10, appears consistent with this interpretation. The AX-CPT-70 and AX-CPT-10 results can be interpreted as supporting the proposal that the high-WMC individuals used proactive control more often than the low-WMC individuals. The results on the AX-CPT-40 provide evidence as to whether the high- and low-WMC groups can switch between the two control modes as necessary for successful task performance. On the AX-CPT-40, the low-WMC group made more AX and BX errors than the high-WMC group. As a group, the highWMC subjects made slightly over two times more AX errors than AY errors, whereas the low-WMC subjects made over nine times more AX errors than AY errors. The higher amount of AX errors for the low-WMC subjects versus the high-WMC subjects on the AX-CPT-40 is consistent with a prediction by Braver et al. (2007) — namely, that high-WMC individuals appear to use proactive control as the task situation allows, and can effectively switch from trial-to-trial between preparing a response in advance of the probe and waiting until the probe appears to begin the response selection process. The current results are consistent with previous WMC investigations of cognitive control that showed that high-WMC individuals keep abstract goal-related information active more than low-WMC individuals. The current results are novel in that the probability manipulations across the versions of the AX-CPT demonstrate that those with higher WMC also translate the cognitive representation of the cue into a specific, planned motor response in advance of upcoming stimuli. The increased readiness of high-WMC subjects is also consistent with previous research using a simple psychomotor vigilance task (Unsworth, Redick, Lakey, & Young, 2010).

In Redick and Engle (2011), AY errors were higher than typically observed in previous AX-CPT-70 studies with healthy young adults. One reason likely contributing to the higher frequency of AY errors for both WMC groups in Redick and Engle (2011) is that the delay manipulation was intermixed within trial blocks, which has been shown previously to increase errors (Barch, Carter, MacDonald, Braver, & Cohen, 2003). In the current AX-CPT-70 study, the frequency of AY errors was more in-line with the results of previous research with young adults (e.g., Braver et al., 2005). For both WMC groups, the overwhelming proportion of AY errors was choice errors (making a target response) and not misses (failing to make a target or nontarget response), as might be expected if an individual had prepared a target response but then ran out of time before executing the correct nontarget response. Future research to examine the nature of the response process (target/nontarget vs. target-only) might be informative about the preparatory processes in which both high- and low-WMC individuals engage on the AX-CPT-70. One area that needs further clarification is the role of individual differences in WMC as related to task-switching. Previous research has shown that task-switching costs are not related to individual differences in WMC (Kane et al., 2007). Yet, in the AX-CPT-40, I observed that highWMC individuals were more adept than low-WMC subjects at switching between using the B cue in advance of a probe stimulus to prepare a response, and on other trials not being able to use the A cue to plan a specific response. An important consideration is the nature of the tasks. In a typical task-switching experiment, an advance cue signals which type of decision to make about an upcoming stimulus (e.g., is a number even or odd). The cue does not allow the subject to prepare a specific response. In addition, when the target digit is presented, there is no uncertainty associated with the response. One prediction is that performance on task-switching experiments that allow the subject to anticipate a certain response in advance (e.g., Ruge, Braver, & Meiran, 2009) would be related to individual differences in WMC — if found, the result would be consistent with the current results, and clarify why previous task-switching results were not dependent upon the subject's WMC.

9. Conclusion The current research investigated individual differences in WMC in relation to the use of proactive control to maintain and translate cognition into preparatory action. The results indicated that high-WMC individuals are more likely than low-WMC individuals to use information conveyed by a cue to prepare a response in advance, especially when the cue information is predictive of subsequent action. In contrast, the performance of the low-WMC individuals was less dependent upon the cue and more contingent upon overall response frequencies. Overall, individuals high in WMC behave in a manner consistent with proactive control when possible, whereas individuals low in WMC are less likely to engage in proactive control.

8.4. Limitations and future directions Acknowledgments Somewhat surprising was the lack of WMC differences in mean RT across the versions of the AX-CPT. The WMC effects on RT were significant in Redick and Engle (2011) with a larger sample size, and the pattern in the current AX-CPT-70 was qualitatively similar. An additional difference was that in Redick and Engle (2011), target and nontarget responses were made with the index and middle fingers of the same hand, whereas in the current work, the two responses were mapped to the index fingers of the left and right hands. Consistent with previous research showing that two-choice RTs are faster when mapped to separate hands versus separate fingers on the same hand (e.g., Kornblum, 1965), the overall mean RT was faster in the current AX-CPT-70 compared to Redick and Engle (2011).

Thomas S. Redick, Purdue University. This research was completed in partial fulfillment of the author's requirements for the degree of Doctor of Philosophy at the Georgia Institute of Technology. I thank the dissertation committee members (Randy Engle, Paul Corballis, Eric Schumacher, Dan Spieler, and David Washburn) for their helpful guidance throughout the project. I also thank Nash Unsworth, Whitney Hansen, Todd Braver, Gregory Burgess, Nelson Cowan, and the members of the Georgia Tech Attention & Working Memory Lab for feedback on an earlier version of this manuscript. Finally, I thank Ashley Coogan for assistance with data collection.

T.S. Redick / Acta Psychologica 145 (2014) 1–9

References

Appendix A

Table A.1 Complete omnibus ANOVA output for error and mean RT data. Effects and interactions Error rates Main effects WMC Version Trial Type 2-Way interactions WMC × Version WMC × Trial Type Version × Trial Type 3-Way interaction WMC × Version × Trial Type Mean RTs Main effects WMC Version Trial Type 2-Way interactions WMC × Version WMC × Trial Type Version × Trial Type 3-Way interaction WMC × Version × Trial Type

9

Partial η2

p

26.24 2.37 55.75

.187 .040 .328

b.01⁎ .10 b.01⁎

0.28 11.73 37.95

.005 .093 .400

.75 b.01⁎ b.01⁎

4.03

.066

b.01⁎

1.47 0.70 110.14

.013 .012 .491

.23 .50 b.01⁎

1.33 2.57 53.76

.023 .022 .485

.27 .05† b.01⁎

0.95

.016

.46

F

⁎ Indicates p b .05. † Indicates p = .05.

Mean RTs on the AX-CPT-70 are presented in Fig. 3a. The two WMC groups do not appear to differ. Although the main effect of Trial Type was significant, F(3, 114) = 125.31, p b .001, η2p = .767, both effects involving WMC were not significant: WMC, F(1, 38) = 2.41, p = .129, η2p = .060; WMC × Trial Type, F(3, 114) = 1.47, p = .226, η2p = .037. Independent-samples t-tests indicated that the high-WMC subjects were faster than low-WMC subjects on BY trials, t(38) = 2.00, p = .053. However, the WMC groups did not differ in mean RT on the other trial types (all p's N .140). Mean RT results for probe trials on the AX-CPT-10 are presented in Fig. 3b. The two WMC groups did not appear to differ in RTs across trial types. Although the main effect of Trial Type was significant, F(3, 114) = 69.75, p b .001, η2p = .647, both the main effect of WMC (F b 1) and the WMC × Trial Type interaction, F(3, 114) = 1.15, p = .331, η2p = .029, were not significant. Independent-samples t-tests on each trial type indicated that the high and low-WMC subjects did not significantly differ in mean RT to any of the trial types (all p's N .253). Mean RT results for probe trials on the AX-CPT-40 are presented in Fig. 3c. The two WMC groups did not appear to differ in RTs across trial types. Although the main effect of Trial Type was significant, F(3, 114) = 26.47, p b .001, η2p = .411, both effects involving WMC were not significant: WMC, F(1, 38) = 1.56, p = .219, η2p = .039; WMC × Trial Type, F(3, 114) = 1.80, p = .151, η2p = .045. Independent-samples t-tests on each trial type indicated that the high-WMC subjects were marginally faster than low-WMC subjects on BY trials, t(38) = 1.94, p = .060. However, the WMC groups did not differ in mean RT on any other trial types (all p's N .211).

Barch, D.M., Carter, C. S., MacDonald, A. W., III, Braver, T. S., & Cohen, J.D. (2003). Context-processing deficits in schizophrenia: Diagnostic specificity, 4-week course, and relationships to clinical symptoms. Journal of Abnormal Psychology, 112, 132–143. Braver, T. S. (2012). The variable nature of cognitive control: A dual mechanisms framework. Trends in Cognitive Sciences, 16, 106–113. Braver, T. S., Barch, D.M., Keys, B.A., Carter, C. S., Cohen, J.D., Kaye, J. A., et al. (2001). Context processing in older adults: Evidence for a theory relating cognitive control to neurobiology in healthy aging. Journal of Experimental Psychology: General, 130, 746–763. Braver, T. S., Gray, J. R., & Burgess, G. C. (2007). Explaining the many varieties of working memory variation: Dual mechanisms of cognitive control. In A.R. A. Conway, C. Jarrold, M. J. Kane, A. Miyake, & J. N. Towse (Eds.), Variation in working memory (pp. 76–106). New York: Oxford University Press. Braver, T. S., Paxton, J. L., Locke, H. S., & Barch, D.M. (2009). Flexible neural mechanisms of cognitive control within human prefrontal cortex. Proceedings of the National Academy of Sciences, 106, 7351–7356. Braver, T. S., Satpute, A.B., Rush, B. K., Racine, C. A., & Barch, D.M. (2005). Context processing and context maintenance in healthy aging and early stage dementia of the Alzheimer's type. Psychology and Aging, 20, 33–46. Cohen, J.D., Barch, D.M., Carter, C. S., & Servan-Schreiber, D. (1999). Context-processing deficits in schizophrenia: Converging evidence from three theoretically motivated cognitive tasks. Journal of Abnormal Psychology, 108, 120–133. Dias, E. C., Foxe, J. J., & Javitt, D. C. (2003). Changing plans: A high density electrical mapping study of cortical control. Cerebral Cortex, 13, 701–715. Engle, R. W., & Kane, M. J. (2004). Executive attention, working memory capacity, and a two-factor theory of cognitive control. In B. Ross (Ed.), The psychology of learning and motivation. , Vol. 44. (pp. 145–199). New York: Elsevier. Kane, M. J., Bleckley, M. K., Conway, A.R. A., & Engle, R. W. (2001). A controlled-attention view of working memory capacity. Journal of Experimental Psychology: General, 130, 169–183. Kane, M. J., Conway, A.R. A., Hambrick, D. Z., & Engle, R. W. (2007). Variation in working-memory capacity as variation in executive attention and control. In A.R. A. Conway, C. Jarrold, M. J. Kane, A. Miyake, & J. N. Towse (Eds.), Variation in working memory (pp. 21–48). New York: Oxford University Press. Kane, M. J., & Engle, R. W. (2003). Working-memory capacity and the control of attention: The contributions of goal neglect, response competition, and task set to Stroop interference. Journal of Experimental Psychology: General, 132, 47–70. Kornblum, S. (1965). Response competition and/or inhibition in two-choice reaction time. Psychonomic Science, 2, 55–56. Paxton, J. L., Barch, D.M., Racine, C. A., & Braver, T. S. (2008). Cognitive control, goal maintenance, and prefrontal function in healthy aging. Cerebral Cortex, 18, 1010–1028. Paxton, J. L., Barch, D.M., Storandt, M., & Braver, T. S. (2006). Effects of environmental support and strategy training on older adults' use of context. Psychology and Aging, 21, 499–509. Poole, B. P., & Kane, M. J. (2009). Working-memory capacity predicts the executive control of visual search among distractors: The influences of sustained and selective attention. Quarterly Journal of Experimental Psychology, 62, 1430–1454. Redick, T. S., Broadway, J. M., Meier, M. E., Kuriakose, P. S., Unsworth, N., Kane, M. J., & Engle, R. W. (2012). Measuring working memory capacity with automated complex span tasks. European Journal of Psychological Assessment, 28, 164–171. Redick, T. S., Calvo, A., Gay, C. E., & Engle, R. W. (2011). Working memory capacity and go/no-go task performance: Selective effects of updating, maintenance, and inhibition. Journal of Experimental Psychology: Learning, Memory, and Cognition, 37, 308–324. Redick, T. S., & Engle, R. W. (2011). Integrating working memory capacity and context-processing views of cognitive control. Quarterly Journal of Experimental Psychology, 64, 1048–1055. Redick, T. S., Heitz, R. P., & Engle, R. W. (2007). Working memory capacity and inhibition: Cognitive and social consequences. In D. S. Gorfein, & C. M. MacLeod (Eds.), Inhibition in cognition (pp. 125–1420). Washington, DC: APA. Ruge, H., Braver, T., & Meiran, N. (2009). Attention, intention, and strategy in preparatory control. Neuropsychologia, 47, 1670–1685. Unsworth, N., & Engle, R. W. (2007). The nature of individual differences in working memory capacity: Active maintenance in primary memory and controlled search from secondary memory. Psychological Review, 114, 104–132. Unsworth, N., Heitz, R. P., Schrock, J. C., & Engle, R. W. (2005). An automated version of the operation span task. Behavior Research Methods, Instruments, & Computers, 37, 498–505. Unsworth, N., Redick, T. S., Lakey, C. E., & Young, D. L. (2010). Lapses in sustained attention and their relation to executive control and fluid abilities: An individual differences investigation. Intelligence, 38, 111–122. Unsworth, N., Schrock, J. C., & Engle, R. W. (2004). Working memory capacity and the antisaccade task: Individual differences in voluntary saccade control. Journal of Experimental Psychology: Learning, Memory, and Cognition, 30, 1302–1321.