Consciousness and Cognition 21 (2012), 277–291
Authors: Paul Seli, James Allan Cheyne, Daniel Smilek
University of Waterloo, Waterloo, Ontario, Canada
Contact: firstname.lastname@example.org (P. Seli)
In two studies of a GO–NOGO task assessing sustained attention, we examined the effects of (1) altering speed–accuracy trade-offs through instructions (emphasizing both speed and accuracy or accuracy only) and (2) auditory alerts distributed throughout the task. Instructions emphasizing accuracy reduced errors and changed the distribution of GO trial RTs. Additionally, correlations between errors and increasing RTs produced a U-function; excessively fast and slow RTs accounted for much of the variance of errors. Contrary to previous reports, alerts increased errors and RT variability. The results suggest that (1) standard instructions for sustained attention tasks, emphasizing speed and accuracy equally, produce errors arising from attempts to conform to the misleading requirement for speed, which become conflated with attention-lapse produced errors and (2) auditory alerts have complex, and sometimes deleterious, effects on attention. We argue that instructions emphasizing accuracy provide a more precise assessment of attention lapses in sustained attention tasks.
Here we explore how altering speed of responding by instructing subjects to respond fast (without sacrificing accuracy) or slow (specifically emphasizing accuracy) influences performance in a theoretically informative manner on a task designed to measure sustained attention abilities: the Sustained Attention to Response Task (SART: Robertson, Manly, Andrade, Baddeley, & Yiend, 1997). The SART has become a popular paradigm for assessing lapses of sustained attention and reverses the more common GO/NOGO procedure by requiring repeated responding (key press) to a series of digits (1–9) and withholding responding when a rare (NOGO) stimulus appears (e.g., “3”). Subjects are instructed to respond as quickly as possible while maintaining high accuracy. The critical SART measure is the proportion of NOGO trials in which a subject fails to withhold a response (i.e., SART Commission Errors). Other measures of performance include response time to GO trials (RT) and the within subjects variability of RTs such as SD or, when mean RTs and RT variances are positively correlated, the RT coefficient of variation (i.e., RT CV = RT SD/Mean RT) over trials. RT variability has become a widely used dependent measure in the SART literature as it is believed to reflect subtle differences in RTs that are produced by lapsing attention (Bellgrove et al., 2006 , Cheyne, Solman, et al., 2009 , Johnson, Kelly, et al., 2007 , Johnson et al., 2008 , McVay and Kane, 2009 , Molenberghs et al., 2009 , O’Connell et al., 2008 and van der Linden et al., 2005).
We sought to explore the influence of instruction-induced speed–accuracy trade-offs in the SART for two reasons. First, SART performance is becoming a very popular measure of individual differences in sustained attention abilities. The SART has been used to assess sustained attention abilities in numerous special populations such as patients with traumatic brain injury (TBI; Chan, 2001 , Chan, 2005 , Manly et al., 2002 , Manly et al., 2003 , McAvinue et al., 2005 , O’Keeffe et al., 2007 and O’Keeffe et al., 2004 ; but see Willmott, Ponsford, Hocking, & Schönberger, 2009), attention-deficit/hyperactivity disorder (ADHD; Bellgrove et al., 2006 , Dockree et al., 2004 , Greene et al., 2009 , Johnson, Kelly, et al., 2007 , Johnson, Robertson et al., 2007 , Johnson et al., 2008 , Mullins et al., 2005 , O’Connell et al., 2004 and Shallice et al., 2002), depression (Farrin, Hull, Unwin, Wykes, & David, 2003), cortical lesions (Molenberghs Gillebert, Schoofs, Dupont, Peeter, 2009), affective disorders (Smallwood, O’Connor, Sudbery, & Obosawin, 2007), schizophrenia (Chan et al., 2009) and stress-related burnout (van der Linden et al., 2005). Performance on the SART has also been assessed as a function of intelligence (Farrin et al., 2003), exposure to natural disasters (Helton, Head, & Kemp, 2011) and normal development and aging (Carriere, Cheyne, Solman, & Smilek, 2010). A variant of the SART has also been used to assess the loss of self-agency during attention lapses (Cheyne, Carriere, & Smilek, 2009). Additionally, the SART has been used to investigate neurophysiological correlates of sustained attention, implicating brain areas such as the dorsomedial and ventromedial prefrontal cortices (both of which are linked to the default network; Christoff, Gordon, Smallwood, Smith, & Schooler, 2009), as well as the anterior cingulate cortex (ACC; Cheyne, Cheyne, Bells, Carriere, & Smilek, 2009). A general assumption of these studies is that the SART provides an accurate measure of sustained attention abilities. Consistent with this assumption, a number of studies have reported significant associations between SART performance and self- and other-reported everyday cognitive and attention-related failures (for review see, Smilek et al., 2010a and Smilek et al., 2010b). An advantage of the SART over older, traditional vigilance tasks (Rosvold, Mirsky, Sarason, Bransome, & Beck, 1956) is that it provides a positive behavioral measure on trials leading up to critical NOGO trials that have provided valuable information on changes in attentional state leading up to NOGO errors (Cheyne and Carriere, 2011 and Cheyne, Carriere, et al., 2009).
There remains, however, a concern that SART performance might, in part, reflect strategic choices in responding along a speed–accuracy trade-off curve (see Helton, 2009 and Helton et al., 2009). One of the more venerable observations of experimental psychology is that errors tend to increase with response speed (Woodworth, 1899). This phenomenon has been observed in numerous tasks and at many levels of psychological functioning from the simplest motor movement to complex semantic processing (MacKay, 1971). Nonetheless, some studies provide evidence of limits and exceptions. The relation between speed and accuracy has been found, for example, to be reversed for highly practiced actions, skilled performers, and familiar tools (Beilock et al., 2008 , Beilock et al., 2004 , MacKay, 1982 and Smith-Chant and LeFevre, 2003) and for extremely slow responding (Newell, 1980). Indeed, it was noted rather early on that there can be an optimal speed for accuracy, with deviations in either direction adversely affecting performance (Rupp, 1932). Notwithstanding these complications, it remains true that experimentally induced alterations of speed of responding, such as by instruction, can significantly affect accuracy (Ridderinkhof, 2002a and Wylie et al., 2009). An instruction-induced slowing strategy minimizing such trade-offs and hence providing a more accurate assessment of attention lapses would potentially increase the power of the SART to assess attention abilities.
The second reason to explore the impact of instruction-induced speed–accuracy trade-offs in the SART arises from recent efforts to test assumptions about top-down monitoring and control of attention and to remediate poor sustained attention performance in the context of the SART (e.g., Manly et al., 2004 and O’Connell et al., 2004). Manly et al. (2004) have reported, for example, that SART performance, and by extension, sustained attention performance, can be improved by presenting subjects with periodic auditory alerts that putatively bring them back on task. However, as we explain below, the addition of alerts in these studies is often confounded with explicit or implied alterations of instructions that might shift responding along the speed–accuracy trade-off curve. Thus, in the present studies, we also explore the combined and separate influences of instruction-induced speed–accuracy trade-offs and periodic auditory alerts.
Speed–accuracy trade-offs in the SART
Typical SART instructions to the subjects state that speed and accuracy are of equal importance in the successful performance of the task. It has been remarked that such commonly used instructions (to equally focus on speed and accuracy) are contradictory as each often requires a mode of responding that is incompatible with the other (Edwards, 1961). In the context of the SART, this instruction is quite misleading as high accuracy (low rates of commission errors on NOGO trials) is considered successful and speeding is taken as an index of a failure of sustained attention. Although evidence exists that subjects sometimes interpret such instructions to emphasize accuracy (Howell & Kreidler, 1963), there are reasons to suspect important individual differences in interpretation of such speed–accuracy instructions according to such diverse variables as skill (Imbo & Vandierendonck, 2010) and age (Carriere et al., 2010).
Importantly, the inverse relation between speed and accuracy is central to the rationale underlying the SART. Robertson and colleagues (1997) argue that the tedium of the SART leads to lapsing attention expressed behaviorally as the automatic, and hence rapid, triggering of responses to stimulus onset prior to a detailed analysis of the stimulus. Rather obviously, more rapid responding provides less time for inhibitory processes to intervene (cf. van den Wildenberg et al., 2010). An important consequence of very rapid responding is therefore that even the briefest of attention lapses that delay, even minimally, the identification of the NOGO stimulus will allow the automatic response to terminate the trial in error (Seli, Cheyne, Barton, & Smilek, in press). Supporting this claim, numerous SART studies have consistently reported robust associations between NOGO commission errors and both global (overall means) and local (immediately preceding and following errors) response times (See Cheyne, Carriere, et al., 2009).
It is important to note that the SART is concerned with very different attentional processes from those assessed in traditional vigilance tasks, which typically involve long tedious tasks that promote the waning of attention over extended periods of time (Davies and Parasuraman, 1982 , Matthews et al., 2000, See et al., 1995 , van der Linden et al., 2005 and Warm et al., 2008). In standard versions of the SART, however, commission errors begin early in the task and increase little, if at all, within the time constraints of the task (Cheyne et al., 2011). The SART appears to be sensitive to moment-to-moment fluctuations in attention potentially associated with limitations of working memory capacity (McVay & Kane, 2009) and typically referred to as absent-mindedness (Cheyne, Carriere, & Smilek, 2006) or mind-wandering (Smallwood & Schooler, 2006). This claim is consistent with a fluctuating top-down control hypothesis of impairments of sustained attention (Bellgrove et al., 2006 , Castellanos et al., 2005 and Stuss et al., 2003).
Unfortunately, by the foregoing reasoning, there is the real possibility that rapid responding will lead to increased SART commission errors resulting not from failures of sustained attention but from other sources. In particular, the standard instructions to respond quickly could lead subjects to fail to inhibit their responses on NOGO trials even though they are adequately attending to the task, simply because the motor program for the response was initiated before the critical stimulus was identified. This would mean that SART errors are not only a measure of sustained attention failures, but that they are contaminated by an instruction-induced artefact, namely, failures to inhibit very rapidly initiated responses caused by the instruction to respond as fast as possible. Consistent with this concern, we recently observed that an auditory version of the SART produced significantly fewer SART errors than the standard visual version. This difference was fully attributable to an overall slowing of response time to presentation of digits in the auditory mode rather than changes in attention lapse rate in the two modalities (Seli et al., in press). Thus, NOGO errors attributable to lapses of sustained attention may become confounded with factors influencing the speed–accuracy trade-off function.
The effects of alerts and instructions on sustained attention
The foregoing considerations of the SART also have implications for the rationale behind the use of non-predictive alerts in a SART investigation of remediation of attentional deficits in clinical populations reported by Manly et al. (2004). Manly and colleagues (2004, Study 1) reported that periodic alerts, contingent on response speeding, significantly improved SART performance of TBI patients with attention deficits. Manly and others have hypothesized that the alerts provide periodic bottom-up activation to compensate for deficits in endogenous top-down management of alertness and/or served as a training procedure to encourage an “executive stance” of active self-control (Bellgrove et al., 2006 , Manly et al., 2004 and O’Connell et al., 2004).
Unfortunately, the instructions accompanying the alerts changed the relative emphasis on speed and accuracy. According to Manly et al. (2004): “The subjects were initially given the standard SART instructions, namely to press the mouse key as quickly as possible following each digit with the exception of the nominated no-go target.” The instructions subsequently associated with the alerts had, however, a very different emphasis: “The link between the presentation of the tone and speeding of reaction times was made explicit. Following discussion about attention wandering and speeding responses, subjects were told: ‘To help you avoid this, the computer will play you a sound to warn you if you are pressing too quickly’”(p. 96). Thus, although the initial standard instructions encourage subjects to respond “as quickly as possible,” subjects were subsequently told that the alerts are presented to warn them that they are responding “too quickly” with the clear implication that (1) they should slow down, presumably in the service of (2) becoming more accurate. Thus, an alert, rather than being a nonspecific cue, became a specific meaningful signal to sacrifice speed, slow down, and focus on accuracy.
In subsequent studies, Manly et al. (2004, Study 2 and 3) attempted to minimize confounding of differential emphasis on speed versus accuracy and the presence of alerts. The instruction associated with cues was altered and subjects were instructed simply to be “very aware of what you are doing in the task.” However, although this second study removed the explicit reference to response speed, alerts continued to be associated with slower response times along with reduced NOGO errors. In a third study, the presentation of alerts produced response slowing over the four post-alert GO trials, and reduced errors on the post-alert NOGO trials. Thus, it remains unclear whether alerts directly affected sustained attention by bringing people back on task, or whether they simply caused an automatic reactive post-alert slowing, which shifted responding along the speed–accuracy trade-off curve, incidentally increasing accuracy.
Alerts were subsequently employed in a SART study involving ADHD children (O’Connell, Bellgrove, Dockree, & Robertson, 2006). The O’Connell et al. study used less explicit instructions than did Manly and colleagues in their Study 1, but perhaps less neutral than for the Manly et al. Studies 2 and 3: “Every now and then, you are going to hear some beeps coming from the speakers. Try to use them as a reminder to concentrate even more on what you are doing” (p. 658). One must wonder what the subjects made of the instruction to “concentrate” on what they were doing. The instruction to “concentrate” could have been construed by the subjects to be more careful, as responding to the GO signal hardly requires concentration. Thus, the alerts could be taken to mean that they should be more cautious about upcoming NOGO trials. In any case, O’Connell and colleagues failed to find alerts to reduce the overall error rate for their ADHD sample.
Given the foregoing considerations, it remains an open question whether alerts simply lead to nonspecific slowing, which would decrease errors by permitting some delay in identification of the NOGO stimulus, or whether they directly affect the rate of attention lapses. Alerts, as Manly et al. (2004) acknowledge, have complex effects. On the one hand, alerts provide reminders to reengage with a task. On the other, as rare critical events, they command attention and demand cognitive resources to consider a long-term strategy that can reduce attention to the immediate, occurrent aspects of the task and lead to errors (Cheyne, Carriere, et al., 2009; Cheyne et al., 2011). These issues can be directly addressed by orthogonally varying instructions to shift responding along the speed–accuracy trade-off function and by varying the presence or absence of alerts.
The present studies
In two studies, we examined the effects of changing SART instructions from the double-edged “be fast and accurate” to providing the more conceptually accurate goal of maintaining high accuracy by responding slowly and carefully. In Study 1, we evaluate (1) whether subjects can maintain slow rates of response within the temporal constraints of the task, and (2) whether slowing, if it occurs, results in fewer commission errors on NOGO trials. Demonstrating that instructing subjects to respond slowly and accurately can lead to a reduction in SART errors would suggest that some of the standard SART errors are caused by an instruction-induced procedural artefact of responding quickly. In addition, such a result would show that the contaminating instruction-induced errors can be removed by instructing people to respond slowly. In Study 2, we investigate the effects of noncontingent alerts on SART measures and their possible interaction with instruction effects. In both Studies, we also analysed performance across first- and second-half blocks of trials to examine the stability of any effects of experimental conditions.
In addition to these goals, we evaluate whether possible instruction-induced slowing has a more direct effect on an increasingly important index of failures of sustained attention, response time variability. Response time variability has become an important marker for genetic, biochemical, and neurophysiological indicators of deficits in management of attentional control (Bellgrove et al., 2006 , Castellanos and Tannock, 2002 , Castellanos et al., 2005 , Kuntsi and Stevenson, 2001 , O’Connell et al., 2008 , O’Connell et al., 2004 and Stuss et al., 2003). Global response time variability may, however, be too nonspecific and hence, recent efforts have been made to parse global response time variability into different temporal frequencies (Castellanos et al., 2005 and Johnson, Kelly, et al., 2007) or temporal intervals (Ridderinkhof, 2002b, Stins et al., 2007 and van den Wildenberg et al., 2010). Although response times are consistently found to have global negative correlations with SART NOGO errors, there is also evidence that both extremely fast and extremely slow response times are associated with high rates of errors, suggesting an underlying U-shaped function in the distribution of the association of response times and errors (Cheyne, Carriere, et al., 2009). Such findings were predicted by a three-state model of mind-wandering-induced attention lapses consisting of: focal task inattention, global task inattention, and response disengagement (Cheyne, Carriere, et al., 2009). Focal task inattention consists of brief and partial waning of detailed moment-to-moment stimulus processing, as when we drift away from, but retain the gist of, a conversation. The first state is likely brief and unstable, and often sufficient to detect incipient slips before they become overt errors. During the second state, moment-to-moment processing has ceased and the individual is now keying in and out of the global task-relevant aspects of the environment, relying on well-practiced automatic responding. In everyday terms, we characterize this as “going through the motions.” In this state, we sometimes find ourselves well beyond the last point of recall of what we have been reading, but find that we have continued to scan down the page. It is in these two states that, during repetitive tasks such as the SART, automated speeded responding takes over and errors are associated with response speeding. In the third state, response disengagement, however, even automatic task responding falters. When reading in this state, we cease to even move down the page and stare – blank, unmoving, and unseeing – at the page. During SART performance, responding ceases, though the flashing stimuli often rapidly bring subjects back on line, but with slowed response times. Thus, the available SART data and the three-state model suggest that failures of sustained attention can be accompanied by both very fast and very slow responding, which, across trials, is characterized by relatively high variability. The present Study therefore examines the distribution of response times across the entire inter-stimulus interval to more precisely specify the sources of response time variability as well as the critical ranges for effects on commission errors. This approach also provides an opportunity to address the issue of an optimal range of responses times (Rupp, 1932).
Subjects were 60 (39 females) University of Waterloo psychology undergraduate students with self-reported normal or corrected-to-normal visual acuity participating in a session lasting approximately 25 min. Participation was voluntary and subjects received course credit.
The SART program used in both studies was created using E-prime software (Schneider, Eschman, & Zuccolotto, 2002). Stimuli presentation was controlled by a Dell Latitude D800 laptop. Displays were presented on a Viewsonic G225F 21” CRT. Responses were collected on a Dell RT7D50 keyboard.
The dependent variables of interest were the proportion of commission errors on NOGO trials, mean RT and RT CV. RT and RT CV measures included all responses that were 100 ms or greater.
The Sustained Attention to Response Task (SART)
On each SART trial, a single digit (1–9) was presented in the center of a computer monitor for 250 ms, followed by an encircled “x” mask for 900 ms. The digits appeared in 48, 72, 94, 100, and 120 point size (randomly selected) Symbol font, in white, on a black background. Digits were randomly distributed on all 630 trials with equal frequency of each. Subjects viewed displays at a distance of approximately 50 cm. Following 18 practice trials, which included the presentation of 2 NOGO targets (the digit “3”), were 630 continuous experimental trials, which included the presentation of 70 NOGO targets.
Subjects were randomly assigned to one of two instruction conditions: (1) standard instruction condition, or (2) go–slow instruction condition. Each set of instructions was visually presented on the monitor and was read aloud by the experimenter. Subjects were provided with either the standard speed–accuracy or go–slow instructions. Subjects assigned to the standard instruction condition were instructed to give equal importance to speed and accuracy when completing the task. Subjects assigned to the go–slow instruction condition were instructed to take their time and respond slowly so as to reduce the number of errors they make. All subjects were instructed to respond to GO stimuli by pressing the spacebar on the keyboard. If subjects in either condition had any questions about the instructions, the researcher provided clarification.
NOGO errors, GO RT, and GO RT CV
The proportions of NOGO commission errors were analyzed with a mixed ANOVA with 2 between (instruction: standard versus go–slow) by 2 within (block 1 (first-half) versus block 2 (second-half)) factors. The analysis yielded a significant effect of instruction, F(1, 58) = 4.28, MSE = .112, p < .043, and blocks, F(1, 58) = 20.19, MSE = .007, p < .001, as well as an instruction by block interaction, F(1, 58) = 4.26, MSE = .007, p < .044 (see Table 1). There were significantly fewer errors under go–slow instructions in block 1, t(58) = 2.69, SE = .059, p < .01, but not block 2, t(58) = 1.41, SE = .067, p > .05. Errors under the go–slow instructions condition increased significantly from block 1 to block 2, t(29) = 1.73, SE = .022, p < .001.
Parallel analysis with RT as the dependent variable yielded significant effects only for instructions, F(1, 58) = 10.46, MSE = 21461.28, p < .002. Response times were significantly slower under the go–slow instruction condition (Table 1). The analysis for RT CV yielded a significant effect of block, F(1, 58) = 12.79, MSE = .006, p < .001. RT CV increased significantly between blocks (Table 1).
Individual differences were stable between the two blocks. Pearson product-moment correlations were significant, rs = 0.88, 0.87, and 0.56, for NOGO errors, GO RT, and GO RT CV, respectively, with the latter having significantly smaller magnitude than for errors or RTs, p < .01, via a Williams test for related samples. NOGO errors were significantly correlated with both mean RT and RT CV respectively, rs = -0.84 and 0.39, ps < .02, though the latter correlation is significantly less than the former, p < .01.
Parsing the RT distribution: Proportion of RTs within 100 ms intervals
To further elucidate the RT and RT CV effects, we examined, for each subject, the proportion of GO trials on which their RT fell within each of 11 intervals from 1 to 1150 ms (each interval was 100 ms except the last which was 150 ms) plus the proportion of GO trials in which no response was made within the available 1150 ms of the trial (omissions). We conducted a mixed ANOVA with instructions as the between factor and RT intervals as the repeated measure. Given the ipsative nature of the interval data, only the interaction was of interest. Greenhouse–Geisser corrections were used for all repeated measures analyses reported. The analysis revealed a significant interaction between instructions and intervals, F(2.00, 115.91) = 7.00, MSE = .056, p < .002. The proportion of response times within the 201–300 ms interval under standard conditions was much higher than it was under the go–slow condition and this was reversed in the 501–600 ms interval (Fig. 1A). Moreover, under the standard condition, proportions peaked markedly in the 201–300 ms interval, whereas proportions in the go–slow condition were distributed evenly across the intervals from 201 ms to 600 ms. That is, the effect of instructions was not merely to shift the peak to a slower range but to flatten the distribution. The distribution under the standard instruction condition was significantly leptokurtic, K = 2.97, s. e. = 1.23, whereas the distribution under the go–slow condition was mesokurtic or mildly platykurtic, K = -1.30, s. e. = 1.23) with a major mode between 201–400 ms and a minor mode in the 501–600 ms range.
We next examined the Pearson product-moment correlations between proportions of responses within intervals and mean NOGO errors for each subject within each group. These correlations are plotted in Fig. 1B. The distribution of correlations across intervals is distinctly U-shaped. Analysis of the quadratic components of the regression of the correlation coefficients on proportions of responses within intervals yielded significant quadratic coefficients for both groups, r2QUAD, p < .001, for both standard and go–slow conditions. Proportions of RTs less than 300 ms were significantly positively correlated with errors for both groups; those between 400 ms and 700 ms were significantly negatively correlated for both groups. In addition, omissions for all groups were positively associated with errors. The left limb of the U-function is also shifted slightly to the right for the go–slow instruction group. For both groups, however, the steep slope, approaching a step function, of the left limb reveals an abrupt shift from highly positive to highly negative correlations with errors.
The magnitude of the correlations at the extremes of the U-function is remarkable given the small probabilities and skewed distributions within these intervals potentially severely attenuating the correlations. To ameliorate such constraints, and to test the hypothesis that wandering out of the optimal range of response times, a “Goldilocks Zone,” is critical in predicting errors, we combined intervals from 1 to 300 ms and those from >900 ms, to create two extreme ranges on either side of the Goldilocks zone and entered these as predictors in a multiple regression predicting mean NOGO errors. The linear combination of these intervals was highly and significantly correlated with errors for both conditions: both multiple Rs = 0.91, ps < .001, though the long RTs, >900 ms, made a significant contribution only under the standard instruction condition (Table 2). NOGO errors are almost completely determined by wandering outside of the optimal limits, though the effect of slower intervals was significant only under the standard condition.
Go–slow instructions, relative to standard speed–accuracy trade-off instructions, increased response times and reduced errors. Subjects were able to modulate their response times within the constraints of the SART. The reduction in errors was, however, somewhat short lived, fading in the second half of the task, despite the relative stability of mean response time.
Mean response time did not change over time, consistent with prior research (Cheyne et al., 2011; McVay & Kane 2009). Also consistent with previous research, response time variability increased over blocks (McVay & Kane, 2009). Thus, although general response slowing was associated with decreased errors overall, it cannot account for the increase in errors across time. Increased RT CV would appear to more accurately reflect fluctuations in trial-by-trial attention to the task and is necessarily associated with, and indeed, very much determined by, extreme values of RT at both ends of the distribution (Cheyne, Carriere, et al., 2009; Cheyne et al., 2011).
Moreover, rather than simply shifting the distribution of response times into longer response time intervals (i.e., shifting the function rightward in Fig. 1A), go–slow instructions flattened the distribution across intervals. One interpretation of this result is that, despite efforts to maintain a slower response rate, subjects under the go–slow conditions repeatedly wandered outside of the optimal “Goldilocks” zone. The finding is consistent with the hypothesis that, whereas standard instructions appeared to encourage responding within response time ranges positively associated with high proportions of errors, go–slow instructions encouraged response times within ranges negatively associated with errors. The difference in proportions of response times falling in the 201–300 ms range between instruction conditions may therefore roughly represent the difference between the frequency of attention lapses (under go–slow conditions) and the frequency of attention lapses plus the explicit attempts of subjects to conform to instructions to respond rapidly (under standard instruction conditions). By this reasoning, approximately half of the errors in the 201–300 ms interval represent attention lapses and half misplace diligence. Response time variability measures would be more selectively responsive to the latter (drift or wandering) component, whereas mean response time would be affected by both components. On the assumption that very rapid response times lead to more errors in either case, this line of reasoning would also explain the consistently higher association of errors with mean response time than with response time variability (see also Seli et al., in press). This assumption receives strong support in the robust positive associations between NOGO errors and GO response times less than 300 ms (See Fig. 1B and Table 2).
Expanding this line of reasoning a little more, we believe that the findings of Study 1 suggest that the SART with go–slow instructions provides a better measure of attention lapses than does the standard SART. It has been widely assumed in the sustained attention literature that the proportion of SART errors reflects a subject’s propensity to experience attentional failures and not the subject’s strategic decision to respond quickly or slowly. However, we have shown that SART errors are roughly cut in half simply by instructing subjects to respond slowly and that slowing instructions substantially reduce the proportion of RTs in the 201–300 ms range compared to the standard SART. These findings imply that errors in the standard SART reflect not only attention lapses, but also an unwanted procedural artefact of the instructions (i.e., “respond quickly”) leading to failures of response inhibition. The unwanted errors due to failures of response inhibition without attention failures likely occur when subjects are in a very fast response mode, contributing a large number of RTs in the 201–300 ms range. Critically, the fact that instructions to slow down reduce the number of RTs in the 201–300 ms range while also reducing errors suggests the conclusion that at least some of the SART errors due to the procedural artefact inherent in the standard SART have been removed. Based on these empirical findings, we claim that SART errors under go–slow instructions provide a better measure of sustained attention than standard SART errors. In addition, the go–slow instructions eliminate the confusion inherent in the equal emphasis of speed and accuracy in the standard SART instructions, which could lead some subjects to focus on speed and others on accuracy.
Finally, and perhaps most striking, was the finding of a U-shaped function in the association of proportions of response times within different ranges rather than a linear function consistent with a conventional speed–accuracy trade-off argument. Wandering outside the optimal zone (300–800 ms) had dramatic effects on commission errors, accounting for over 80% of the error variance, though largely attributable to excessively rapid rather than slow responding (Table 2). Note that the right limb of the function never reaches the levels of the left, likely because very slow responses are very rare, possibly because they reflect a deeper stage of attention lapse (response disengagement) less likely to occur within the brief time span of the SART (Cheyne, Carriere, et al., 2009).
In summary, the results of Study 1 suggest that the standard SART instructions produce two kinds of errors: (1) those arising from attempts to conform to the misleading requirement for speed, and (2) those reflecting lapses in attention. Hence, given that instructions to respond slowly eliminate the misleading requirement for speeded responses, these results support the hypothesis that the errors produced under the go–slow condition are less contaminated by speed–accuracy trade-off effects. Thus, instructing subjects to respond slowly and accurately would provide a more accurate assessment of sustained attention.
Having found that subjects are indeed capable of following instructions to slow their response tempo, but that this appears to wane rapidly, we next addressed the question of whether periodic noncontingent alerts might help subjects sustain the slower tempo for a longer period or, alternatively, reduce attention lapses independently of response slowing as suggested by Manly et al. (2004). In this study we cross instructions, standard and go–slow, with the presence of periodic alerts and assess the effects on NOGO commission errors, mean RT, RT CV and the distribution of response time across 100 ms intervals. In addition, to better standardize the instructions across individuals, we formalized and recorded the instructions for Study 2 and presented them to subjects via headphones.
Subjects were 120 (83 females) University of Waterloo psychology undergraduate students with self-reported normal or corrected-to-normal visual acuity participating in a session lasting approximately 25 min. Participation was voluntary and subjects received course credit.
Study 2 included the same two instruction conditions from Study 1 (go–slow and standard instruction conditions) as well as two alert conditions. Thus, Study 2 consisted of a total of four conditions: (1) standard instruction, (2) go–slow instruction, (3) standard instruction with alerts, and (4) go–slow instruction with alerts. Subjects were randomly assigned to one of the four conditions. All subjects were instructed to respond to GO stimuli by pressing the spacebar.
At the beginning of the experiment, pre-recorded auditory instructions were presented via Sony MDR-XD200 Headphones. Subjects assigned to the standard instruction condition were instructed as follows.
Standard speed–accuracy instructions
“Please give equal importance to SPEED and ACCURACY when completing this task. We would like you to respond as FAST as possible while maintaining a high level of ACCURACY.”
Following the practice session subjects were reminded:
“Please remember to respond as FAST as possible while maintaining a high level of ACCURACY.”
Subjects assigned to the go–slow instruction condition were presented the following instructions.
“The point of this task is to make as few errors as possible. That is, to respond to all numbers except 3 and to avoid hitting the space bar when the 3 appears. So please DO NOT RUSH but respond carefully so that you make as few errors as possible.
I want to emphasize the importance of responding SLOWLY on this task. We would like you to SLOW DOWN so that you reduce the number of errors that you make. Now, you have approximately one second to respond before the next digit appears, so you’ll still have to respond fairly quickly, but we would like you to take as much time as you can before responding to the digit. As long as you respond to one digit before the next appears, your response will count.”
After completing the practice session, these subjects were presented the following brief reminder instruction:
“The practice session is now complete and you are about to begin the task. Please remember that the point of the task is to make as few errors as possible; that is, to respond to all numbers except 3 and avoid hitting the space bar when the 3 appears. So please DO NOT RUSH, but take your time, SLOW DOWN, and respond carefully so that you make as few errors as possible.”
Subjects in both conditions were instructed to keep the headphones on throughout the experiment.
The alert conditions were identical to the instruction conditions except auditory alerts were periodically presented and subjects in these conditions were provided one additional instruction. Specifically, immediately following the post-practice instructions, subjects in the alert conditions were told:
“Every now and then, you will be presented with an auditory tone which is a reminder for you to try and be very aware of what you are doing in the task.”
In the alert conditions, one auditory alert (261.63 Hz at approximately 60 dB) was presented at a random point within each sequential block of 30 experimental SART trials. Thus, Over the 630 SART trials, 21 auditory alerts were presented. Each alert was presented at the onset of the mask. The alerts were delivered via the same Sony MDR-XD200 Stereo Headphones used to deliver the instructions in all conditions. Subjects in all conditions were instructed to leave the headphones on for the entirety of the experiment.
NOGO errors, GO RT, and GO RT CV
A mixed ANOVA with 2 (instruction: standard versus go–slow) by 2 (no alerts versus alerts) as between factors and blocks (first-half versus second-half) with proportion errors as dependent variable yielded a significant effect of instruction, F(1, 116) = 51.42, MSE = .085, p < .001, alerts, F(1, 116) = 5.26, MSE = .086, p < .024, and blocks, F(1,116) = 17.83, MSE = .011, p < .001, but no interactions (Fs < 1). There were significantly fewer errors under go–slow instruction and no alerts conditions and errors increased between blocks (see Table 3). Parallel analysis for RT yielded significant effects only for instructions, F(1, 116) = 35.65, MSE = 18502.27, p < .001. Response times were slower under the go–slow instruction conditions. A parallel analysis for RT CV yielded significant effects for instruction, F(1, 116) = 7.78, MSE = .024, p < .006, blocks, F(1, 116) = 26.40, MSE = .006, p < .001, and an alert by block interaction, F(1, 116) = 4.84, MSE = .006, p < .030. Response times were more variable under standard instructions than under go–slow instructions. RT CV increased significantly across blocks only under the alerts conditions, t(29) = 3.85, SE = .03, p < .001, (Table 3).
Local effects of alerts
To evaluate the immediate and direct effects of alerts we examined RTs for four lags (i.e., trials) preceding and four lags following the presentation of alerts, via a one-way mixed 2 by 9 ANOVA with standard and go–slow conditions as a between-groups factor and lag as the within-subject variable. This analysis yielded significant effects for group, F(1, 58) = 15.27, MSE = 94230.95, p < .001, and lag, F(4.05, 235.10) = 14.95, MSE = 1835.02, p < .001, as well as a significant interaction, F(4.05, 235.09) = 2.73, MSE = 1835.02, p < .024. Both groups slowed briefly following alerts and response times were significantly slower under the go–slow instruction condition at all lags (Fig. 2).
Given the limited sampling of NOGO trials at each lag, proportions of errors of the four trials preceding and following alerts were combined to increase stability of parameter estimation. A mixed 2 by 2 ANOVA with alert conditions (no alert versus alert) by timing (pre-alert versus post-alert) as factors and proportion NOGO errors as dependent variable yielded significant alert condition effect, F(1, 58) = 34.22, MSE = .091, p < .001 and a marginal pre-post alert effect, F(1, 58) = 3.32, MSE = .028, p < .074. Mean proportion error was significantly higher in the standard instruction condition than in the go–slow condition, 0.63 (SD = 0.24) versus 0.30 (SD = 0.23). The mean for errors was slightly higher in the four trials prior to the alert, 0.49 (SD = 0.29) than in the four trials following the alerts, 0.44 (SD = 0.33).
Proportion of RTs within RT intervals
We again examined, for each subject, the proportion of GO trials in which their RTs fell within each of 11 intervals from 1 to 1150 ms (each interval was 100 ms except the last which was 150 ms) plus the proportion of GO trials in which no response was made (omissions). We conducted a mixed ANOVA with instructions and alerts as between factors and RT intervals as repeated measures. There was one significant interaction: between instructions and intervals, F(2.5, 290.17) = 25.85, MSE = .040, p < .001. Distribution of proportions across intervals varied between instruction conditions and consistently for each alert condition (Fig. 3A). RT proportion within the 201–300 interval under standard conditions was much higher than that under go–slow conditions and this was reversed in the 501–600 interval. Under standard conditions, proportions peaked markedly in the 201–300 ms interval, whereas under go–slow conditions, proportions were distributed evenly across the intervals from 201 ms to 600 ms. As in Study 1, the effect of instructions was not to shift the peak to a slower range but to flatten the distribution. Both standard instruction conditions produced leptokurtic distributions, K = 2.38 and 5.66, whereas go–slow instructions yielded mildly platykurtic distributions, K = -1.39 and -1.422, for no alerts and alerts groups respectively (s. e. = 1.23) with modes in the 301–400 interval and 501–600 interval.
We next examined the Pearson product-moment correlations between proportions within intervals and mean NOGO errors for each subject within each group. These correlations are plotted in Fig. 3B. The distribution of correlations across intervals is, as in Study 1, U-shaped. Regressing the correlation coefficients on interval order yielded significant quadratic coefficients, r2QUAD, 0.84, 0.71, and 0.74, all ps < .004, for standard no alert and alert, and go–slow no alert and alert, respectively. RTs less than 300 ms are significantly and positively correlated with errors for all groups; those between 400 ms and 700 ms significantly and negatively correlated for all groups. In addition, several very long RTs as well as omissions for all groups are positively associated with errors. The left limb of the U-function is also shifted slightly to the right for the go–slow instruction groups. A high proportion of RTs in the 301–400 ms interval is significantly positively correlated with errors for the go–slow instruction groups, whereas it is marginally negatively correlated with errors for the standard instruction groups. Thus, the suggestion of right shift under go–slow conditions in Study 1 became significant in Study 2.
As in Study 1, to test the hypothesis that wandering out of the Goldilocks Zone is critical in predicting errors, we combined intervals 1–300 ms and >900 ms and entered these combinations as predictors in a multiple regression predicting mean NOGO errors. The linear combination of these interval ranges was highly and significantly correlated with errors, with multiple Rs ranging from 0.82 to 0.93.
Instructions to respond slowly and carefully again produced significantly fewer errors, longer response times, and decreased response time variability relative to instructions emphasizing speed and accuracy. In addition, correlations between errors and increasing RTs again produced a U-function. As in Study 1, responses that “wandered” beyond the optimal Goldilocks range accounted for a great deal of the variance in errors.
With regard to the hypotheses regarding alerts, the results were unequivocal. We not only failed to replicate the finding of beneficial effects of alerts when controlling for speed–accuracy trade-offs, we found strong effects in the opposite direction. Not only did the presentation of alerts fail to reduce errors when unaccompanied by instructions to respond slowly, but also resulted in increased errors under both standard and go–slow instructions. Standard instructions without alerts did produce more errors than go–slow instructions with alerts, consistent with the concern that sometimes explicit or implicit instruction effects may be responsible for apparent effects of alerts on remediation in prior research. Most daunting for the hypothesis of the remediating effects of alerts, however, was the finding that, even with instructions encouraging slow responding, alerts led to increased errors overall and no evidence of this improving over time. Moreover, alerts also produced greater response time variability and did so increasingly between blocks. Rather than maintaining sustained attention over time, therefore, alerts appear to lead to greater fluctuations of attention over time. The SART and the experimental context in which it is investigated is in some ways considerably removed from the use of electronically assisted “reminders to attend” as might be implemented in everyday “real world” contexts. Nonetheless, it reminds us that the use of such periodic and inherently unpredictable alerting reminder might well introduce yet another source of distraction from ongoing tasks and, in many real world contexts, concurrent interpersonal interactions (Cheyne et al., 2011).
Instructions and the purposes of the SART
A positive implication of the present findings is that the use of instructions more consistent with the intent of the SART may generate commission errors that more accurately reflect failures of sustained attention independently of errors arising from explicit attempts to comply with the instruction to respond quickly. On the present view, the standard SART instructions introduce “errors of planning” (Reason, 1977 and Reason, 1979), based on mistaken task goals arising from misinformation about the real goals of the task, in addition to the conceptually more relevant “actions not as planned” explicitly targeted by the developers of the SART. The original use of the speed–accuracy instruction for the SART likely reflected a desire to place individuals at the threshold of inhibition given delayed identification of the NOGO signal arising from very brief lapses of attention. The standard instructions may very well do this, but appear to do so by introducing instruction-induced speeding and errors caused by misplaced diligence rather than attention lapses.
Of particular interest is the significant and dramatic difference between the correlations between proportion of response times and overall commission errors for the standard and go–slow conditions in the 301–400 ms RT range (Fig. 3B). For the standard conditions, the correlation is significantly positive, whereas, for go–slow conditions, the correlation is marginally negative. It seems that the Goldilocks zone itself has shifted, suggesting that increased errors in this zone may not be simply the result of responding too quickly to detect the NOGO event and/or inhibit the dominant response. Rather, the alternative hypothesis suggested is that responding in the 301–400 RT range reflects a differential contribution of attentional state to errors under each of the two conditions. On this hypothesis, under the standard condition, responses within this range critically represent not only lapses of attention but also the response tempo of the diligent subject attempting to comply with the joint requirement to respond quickly and correctly. Under the go–slow condition, in contrast, responding within this range represents a failure to maintain a slow tempo of responding arising from lapses of attention uncontaminated by misplaced diligence (See Fig. 3A). The optimal range of response times is therefore not fixed but depends on the subject’s task set and its implication for the meaning of responses within a given range.
In the present paper we present a novel method of presenting response time data over the inter-stimulus interval, a technique which allowed us to specify rather precisely critical ranges of response times that predict errors. Of particular interest in Figs. 1B and 3B is the abrupt switching of the signs of the correlations between 100 ms bins. This near step-function is consistent with the view that the SART is sensitive to extremely brief fluctuations in attention to features of the immediate environment (particularly mind wandering states 1 and 2, Cheyne, Carriere, et al., 2009; see also McVay & Kane, 2009) and that such brief “absences” are remarkably common. Such attention lapses are consistent with a view that we constantly “blink” out of awareness (perhaps sometimes quite literally: See Smilek et al., 2010a and Smilek et al., 2010b) or, at least of explicit processing of task relevant features. That this occurs in the quiet non-distracting context of the rather brief and undemanding SART, which requires little beyond sustaining attention to elementary stimuli, further suggests very persistent low level endogenous sources of attention lapses.
Fluctuations of attention and waning of strategic responding
Given the more appropriate instructions, maintaining a slow rate of response on GO trials would seem to be an obvious and easy-to-implement strategy. It might be thought to be especially effective if, as has been suggested, strategic responding plays a significant role in SART performance (Helton, 2009 , Helton et al., 2010 and Helton et al., 2009); although, this group appears to have recently abandoned this particular critique in favor of a response inhibition interpretation (Helton et al., 2005 and Stevenson et al., 2011). It might seem surprising, therefore, that the effects on errors of the instructions encouraging the strategy to respond slowly and cautiously were so partial and temporary. First, although go–slow instructions significantly shifted a substantial proportion of response times to the 501–600 ms range, there remained an equal proportion of responses in the 201–300 ms range (see Figs. 1A and 3A). Second, the effects of the go–slow instructions on commission errors declined significantly between the first- and second-half of the SART session. That response times slowed under the go–slow condition is scarcely surprising; that they did so inconsistently requires explanation. The regular tempo of the stimulus presentation might induce a persistent reversion to basic response time. It has long been known that perception and action mutually modulate one another (James, 1890 and Prinz et al., 2009) within the constraints of phase and period error variance of internal timekeepers (Wing & Kristofferson, 1973). The features of response time on the SART suggest that strategic efforts must continuously be brought to bear to modulate such intrinsic mechanisms of entrainment. Consistent modulation would require continuous monitoring and control of response rates, as lapses of monitoring and control will lead to a reversion to natural rhythms of responding. The reliable maintenance of this simple strategy is, as previously noted, precisely the definition of sustained attention.
Wandering attention: Wandering Behavior
Examination of the distribution of individual response times across 100 ms bins over the duration of a trial corroborate previous research findings that extreme response times at both ends of the continuum are predictive of errors (Cheyne, Carriere, et al., 2009, Figs. 5 and 7). Behavioral wandering, as an embodiment of mind wandering, expressed in the SART by response times outside of the Goldilocks Zone, either as extremely rapid or extremely slow responding, is highly predictive of NOGO commission errors, accounting for between 68% and 86% of the variance in errors across all six conditions of the two studies reported (see Table 2 and Table 4). The SART and particularly the brief duration of the SART, likely militate against long lapses of attention and hence against finding strong effects of the slowing of response times.
The paradox of omissions
The positive association of GO trial omissions and NOGO errors of commission, on the face of it, would appear to be an obvious logical and empirical paradox. Clearly, omissions can be predictors of errors of commission only in the global and indirect sense that they indicate a propensity to wander markedly off-task (Cheyne, Carriere, et al., 2009). Conceptually, if omissions represent a deep level of off-task processing, whereby subjects have temporarily stopped responding to the occurrent aspects of the task at the time of presentation of a NOGO trial they will not make commission errors. In order to assess this hypothesis, one would need to use an alternate task, such as the response switching task (RST: Cheyne, Carriere, et al., 2009), wherein, having made error of commission, subjects have the opportunity perform a corrective alternative action thereby indicating an awareness of their failure to withhold. Failures to take corrective action on the RST would suggest absent-minded withdrawal from the task. It would, however, likely require rather longer sessions to detect deep level attention lapses as healthy subjects appear to be rarely unaware of their commission errors within the normal time of the SART, although TBI patients have been found to be unaware of up to a quarter of their errors (McAvinue, O’Keefe, McMackin, & Robertson, 2005).
The deleterious effects of alerts
It remains unclear why alerts had the deleterious effects observed. The modest local benefits of alerts observed, though of small magnitude and brief duration, should have improved, however modestly, rather than impaired, overall performance. One speculative explanation for the global effect is that the procedure of introducing alerts did indeed lead subjects to think about what they were doing, but in global strategic terms, thereby withdrawing cognitive resources from immediate task demands. Thus, the interference effect of the alerts may not coincide with their presentation but rather act more pervasively by introducing or augmenting an interfering stream of task-related thought (Smallwood, Baracaia, et al., 2003 , Smallwood et al., 2004 and Smallwood, Obonsawin et al., 2003). The repeated administration of the alerts might well sustain the interfering train of thoughts originally engendered by the instructions defining the meaning of the alerts. We have recently argued and presented supporting evidence that attempts to behave strategically on SART-like tasks are often without benefit or are counterproductive for similar reasons of resource misallocation (Cheyne et al., 2011).
In previous papers we have shown that the SART is related to self-reported attention lapses as assessed by questionnaires such as the ARCES and MAAS-LO (See Cheyne, Carriere, et al., 2009 , Cheyne et al., 2006 , Smilek et al., 2010a and Smilek et al., 2010b). Thus, one way to further assess the hypothesis that the SART with go–slow instructions provides a less contaminated measure of sustained attention abilities is to explore the relationships between self-reported attention lapses and SART errors under both go-slow and standard instructions. If this hypothesis is correct, then the correlations between SART errors and self-reported attention lapses should be significantly greater under go–slow relative to standard instructions.
In a very direct sense, the standard instructions for the SART are misleading with regard to the ultimate interpretation of performance. Although subjects are encouraged to respond both quickly and accurately, the major dependent variable is accuracy (rate of errors of commission on NOGO trials) and, in addition, fast response times are taken to reflect inattention to the task. Given the double-edged standard instructions, it is not unlikely that subjects may vary in their interpretation of the joint emphasis on speed and accuracy. In previous research, for example, we have found younger subjects to make more errors on the SART than older subjects, but that this difference is largely accounted for by younger subjects’ much more rapid response style (Cheyne et al., 2006 and Carriere et al., 2010). Similar results across a comparable age-range have been reported for a variant of the Simon Task (Juncos-Rabadán, Pereiro, & Facal, 2008). Younger individuals appear to be willing, strategically, to trade away accuracy for speed, whereas older subjects may strive more for accuracy (Salthouse, 1979). By advising subjects to slow down and exercise caution, it appears that we can reduce, minimize, or eliminate individual differences in such response styles and obtain a more accurate estimate of individual differences in sustained attention. Such an improved version of the SART could provide a better measure of sustained attention abilities in various special populations, a more accurate understanding of the neurophysiology of sustained attention, as well as more generally providing more reliable measures of attention lapses and their relation to mind-wandering.
This research was supported by a Discovery Grant from the Natural Sciences and Engineering Council (NSERC) awarded to DS and an NSERC CGSM awarded to PS.
Beilock et al., 2008 S.L. Beilock, B.I. Bertenthal, M. Hoerger and T.H. Carr, When does haste make waste? Speed–accuracy tradeoff, skill level, and the tools of the trade. Journal of Experimental Psychology: Applied, 4 (2008), pp. 340–352.
Beilock et al., 2004 S.L. Beilock, B.I. Bertenthal, A.M. McCoy and T.H. Carr, Haste does not always make waste: Expertise, direction of attention, and speed versus accuracy in performing sensorimotor skills. Psychonomic Bulletin & Review, 11 (2004), pp. 373–379.
Bellgrove et al., 2006 M.A. Bellgrove, Z. Hawi, M. Gill and I.H. Robertson, The cognitive genetics of attention deficit hyperactivity disorder (ADHD): Sustained attention as a candidate phenotype. Cortex, 42 (2006), pp. 838–845.
Carriere et al., 2010 J.S.A. Carriere, J.A. Cheyne, G.J.F. Solman and D. Smilek, Age trends for failures of sustained attention. Psychology and Aging, 25 (2010), pp. 569–574.
Castellanos et al., 2005 F.X. Castellanos, E.J. Sonuga-Barke, A. Scheres, A. Di Martino, C. Hyde and J.R. Walters, Varieties of attention-deficit/hyperactivity disorder-related intra-individual variability. Biological Psychiatry, 57 (2005), pp. 1416–1423.
Castellanos and Tannock, 2002 F.X. Castellanos and R. Tannock, Neuroscience of attention deficit/hyperactivity disorder: The search for endophenotypes. Nature Reviews Neuroscience, 3 (2002), pp. 617–628.
Chan, 2001 R.C.K. Chan, A further study of the sustained attention to task (SART): The effect of age, gender and education. Brain Injury, 15 (2001), pp. 819–829.
Chan, 2005 R.C.K. Chan, Sustained attention in patients with traumatic brain injury. Clinical Rehabilitation, 19 (2005), pp. 188–193.
Chan et al., 2009 R.C.K. Chan, Y. Wang, E.F.C. Cheung, J. Cui, Y. Deng and Y. Yuan, et al. Sustained attention deficit along the psychosis proneness continuum. Cognitive and Behavioral Neurology, 22 (2009), pp. 180–185.
Cheyne and Carriere, 2011 Cheyne, J. A., Carriere, J. S. A, Solman, G. J. F., & Smilek, D. (2011). Challenge and error: Critical events and attention-related errors. Cognition doi:10.1016/j.cognition.2011.07.010.
Cheyne et al., 2006 J.A. Cheyne, J.S.A. Carriere and D. Smilek, Absent-mindedness: Lapses of conscious awareness and everyday cognitive failures. Consciousness and Cognition, 18 (2006), pp. 481–493.
Cheyne, Carriere, et al., 2009 J.A. Cheyne, J.S.A. Carriere and D. Smilek, Absent minds and absent agents: Attention-lapse induced alienation of agency. Consciousness and Cognition, 15 (2009), pp. 578–592.
Cheyne, Cheyne, et al., 2009 Cheyne, D. O., Cheyne, J. A., Bells, S., Carriere, J. S. A., & Smilek, D. (2009). Neuromagnetic imaging of cortical dynamics associated with response switching and response errors in a speeded motor task. In Poster presented at NCM annual meeting, Waikoloa, Hawaii, April 28–May 3.
Cheyne, Solman, et al., 2009 J.A. Cheyne, G.J.F. Solman, J.S.A. Carriere and D. Smilek, Anatomy of an error: A bidirectional stage model of task engagement/disengagement and attention-related errors. Cognition, 111 (2009), pp. 89–113.
Christoff et al., 2009 K. Christoff, A.M. Gordon, J. Smallwood, R. Smith and J.W. Schooler, Experience sampling during fMRI reveals default network and executive system contributions to mind wandering. Proceedings of the National Academy of Science, 106 (2009), pp. 8719–8724.
Davies and Parasuraman, 1982 D.R. Davies and R. Parasuraman, The psychology of vigilance, Academic Press, London (1982).
Dockree et al., 2004 P.M. Dockree, S.P. Kelly, R.A. Roche, M.J. Hogan, R.B. Reilly and I.H. Robertson, Behavioural and physiological impairments of sustained attention after traumatic brain injury. Cognitive Brain Research, 20 (2004), pp. 403–414.
Edwards, 1961 W. Edwards, Costs and payoffs are instructions. Psychological Review, 68 (1961), pp. 275–284.
Farrin et al., 2003 L. Farrin, L. Hull, C. Unwin, T. Wykes and A. David, Effects of depressed mood on objective and subjective measures of attention. Journal of Neuropsychiatry and Clinical Neurosciences, 15 (2003), pp. 98–104.
Greene et al., 2009 C.M. Greene, M.A. Bellgrove, M. Gill and I.H. Robertson, Noradrenergic genotype predicts lapses in sustained attention. Neuropsychologia, 47 (2009), pp. 591–594.
Helton, 2009 W.S. Helton, Impulsive responding and the sustained attention to response task. Journal of Experimental and Clinical Neuropsychology, 31 (2009), pp. 39–47.
Helton et al., 2011 Helton, W. S., Head, J., & Kemp, S. (2011). Natural disaster induced cognitive disruption: Impacts on action slips. Consciousness and Cognition. doi:10.1016/j.concog.2011.02.011.
Helton et al., 2005 W.S. Helton, T.D. Hollander, J.S. Warm, G. Matthews, W.N. Dember and M. Wallaart, et al. Signal regularity and the mindlessness model of vigilance. British Journal of Psychology, 96 (2005), pp. 249–261.
Helton et al., 2009 W.S. Helton, R.P. Kern and D.R. Walker, Conscious thought and the sustained attention to response task. Consciousness and Cognition, 18 (2009), pp. 600–607.
Helton et al., 2010 W.S. Helton, L. Weil, A. Middlemiss and A. Sawers, Global interference and spatial uncertainty in the sustained attention to task (SART). Consciousness and Cognition, 19 (2010), pp. 77–85.
Howell and Kreidler, 1963 W.C. Howell and D.L. Kreidler, Information processing under contradictory instructional sets. Journal of Experimental Psychology, 65 (1963), pp. 39–46.
Imbo and Vandierendonck, 2010 I. Imbo and A. Vandierendonck, Instruction and load effects on high-skill and low-skill individuals: A study in the domain of mental arithmetic. European Journal of Cognitive Psychology, 22 (2010), pp. 964–989.
James, 1890 W. James, Holt, New York (1890).
Johnson et al., 2008 K.A. Johnson, E. Barry, M. Bellgrove, M. Cox, S.P. Kelly and A. Dáibhis, et al. Dissociation in response to methylphenidate in response variability in a group of medication naïve children with ADHD. Neuropsychologia, 46 (2008), pp. 1532–1541.
Johnson, Kelly, et al., 2007 K.A. Johnson, S.P. Kelly, M.A. Bellgrove, E. Barry, M. Cox and M. Gill, et al. Response variability in attention deficit hyperactivity disorder: Evidence for neuropsychological heterogeneity. Neuropsychologia, 45 (2007), pp. 630–638.
Johnson, Robertson et al., 2007 K.A. Johnson, I.H. Robertson, S.P. Kelly, T.J. Silk, E. Barry and A. Dáibhis, et al. Dissociation of performance of children with ADHD and high-functioning autism on a task of sustained attention. Neuropsychologia, 45 (2007), pp. 2234–2245.
Juncos-Rabadán et al., 2008 O.J. Juncos-Rabadán, A.X. Pereiro and D. Facal, Cognitive interference and aging: Insight from a spatial stimulus–response consistency task. Acta Psychologica, 127 (2008), pp. 237–246.
Kuntsi and Stevenson, 2001 J. Kuntsi and J. Stevenson, Psychological mechanisms in hyperactivity: II. The role of genetic factors. Journal of Child Psychology and Psychiatry, 42 (2001), pp. 211–219.
MacKay, 1971 D.G. MacKay, Stress pre-entry in motor systems. American Journal of Psychology, 84 (1971), pp. 35–51.
MacKay, 1982 D.G. MacKay, The problems of flexibility, fluency, and speed–accuracy trade-off in skilled behaviour. Psychological Review, 89 (1982), pp. 483–506.
Manly et al., 2002 T. Manly, K. Hawkins, J. Evans, K. Woldt and I.H. Robertson, Rehabilitation of executive function: Facilitation of effective goal management of complex tasks using periodic alerts. Neuropsychologia, 40 (2002), pp. 271–281.
Manly et al., 2004 T. Manly, J. Heutink, B. Davidson, B. Gaynord, E. Greenfield and A. Parr, et al. An electric knot in the handerchief: “Content free cueing” and the maintenance of attentive control. Neuropsychological Rehabilitation, 14 (2004), pp. 89–116.
Manly et al., 2003 T. Manly, A.M. Owen, L. McAvenue, A. Datta, G.H. Lewis and S.K. Scott, et al. Enhancing the sensitivity of a sustained attention task to frontal damage: Convergent clinical and functional imaging evidence. Neurocase, 9 (2003), pp. 340–349.
Matthews et al., 2000 Matthews, G., Davies, D. R., Westerman, S. J., & Stammers, R. B., (2000). Human performance: Cognition, stress and individual differences.UK, Psychology Press, East Sussex..
McAvinue et al., 2005 L. McAvinue, O’Keefe, D. McMackin and I.H. Robertson, Impaired sustained attention and error awareness in traumatic brain injury: Implications for insight. Neuropsychological Rehabilitation, 15 (2005), pp. 569–587.
McVay and Kane, 2009 J.C. McVay and M.J. Kane, Conducting the train of thought: Working memory capacity, goal neglect, and mind wandering in executive control task. Journal of Experimental Psychology: Learning, Memory, and Cognition, 35 (2009), pp. 196–204.
Molenberghs et al., 2009 P. Molenberghs, C.R. Gillebert, H. Schoofs, P. Dupont, R. Peeter and R. Vandenbergh, Lesion neuroanatomy of the sustained attention to response task. Neuropsychologia, 47 (2009), pp. 2866–2875.
Mullins et al., 2005 C. Mullins, M.A. Bellgrove, M. Gill and I.H. Robertson, Variability in time reproduction: Difference in ADHD combined and inattentive subtypes. Journal of the American Academy of Child and Adolescent Psychiatry, 44 (2005), pp. 169–176.
Newell, 1980 Newell, K. M. (1980). The speed–accuracy paradox in movement control: Errors in time and space. In G. E. Stelmach & J. Requin (Eds.), Tutorials in motor behaviour. New York: North Holland..
O’Connell et al., 2008 R.G. O’Connell, M.A. Bellgrove, P.M. Dockree, A. Lau, M. Fitzgerald and I.H. Robertson, Self-alert training: Volitional modulation of autonomic arousal improves sustained attention. Neuropsychologia, 46 (2008), pp. 1379–1390.
O’Connell et al., 2004 R.G. O’Connell, M.A. Bellgrove, P.M. Dockree and I.H. Robertson, Reduced electrodermal response to errors predicts poor sustained attention performance in attention deficit hyperactivity disorder. Neuroreport, 15 (2004), pp. 2535–2538.
O’Connell et al., 2006 R.G. O’Connell, M.A. Bellgrove, P.M. Dockree and I.H. Robertson, Cognitive remediation in ADHD: Effects of periodic non-contingent alerts on sustained attention to response. Neuropsychological Rehabilitation, 16 (2006), pp. 653–665.
O’Keeffe et al., 2007 F.M. O’Keeffe, P. Dockree, P. Moloney, S. Carton and I.H. Robertson, Awareness of deficits in traumatic brain injury: A multidimensional approach to assessing cognitive knowledge and online-awareness. Journal of the International Neuropsychology Society, 13 (2007), pp. 38–49.
O’Keeffe et al., 2004 F.M. O’Keeffe, P.M. Dockree and I.H. Robertson, Poor insight in traumatic brain injury mediated by impaired error processing? Evidence from electrodermal activity. Cognitive Brain Research, 22 (2004), pp. 101–112.
Prinz et al., 2009 Prinz, W., Aschersleben, G., & Koch, I. (2009). Cognition and action. In E. Morsella, J. A. Bargh, & P. M. Gollwitzer (Eds.), Oxford handbook of human action. New York: Oxford University Press.
Reason, 1977 J.T. Reason, Skill and error in everyday life, M. Howe, Editor, Adult learning, Wiley, London (1977).
Reason, 1979 J.T. Reason, Actions not as planned: The price of automatization, G. Underwood, R. Stevens, Editors , Aspects of consciousness, Academic Press, London (1979), pp. 67–89.
Ridderinkhof, 2002a K.R. Ridderinkhof, Micro- and macro-adjustments of task set: Activation and suppression in conflict tasks. Psychological Research, 66 (2002), pp. 312–323.
Ridderinkhof, 2002b Ridderinkhof, K. R. (2002b). Activation and suppression in conflict tasks: empirical clarification through distributional analyses. In W. Prinz, & B. Hommel (Eds.), Attention and performance, common mechanisms in perception and action (Vol. XIX, pp. 494–519). Oxford: Oxford University Press.
Robertson et al., 1997 I.H. Robertson, T. Manly, J. Andrade, B.T. Baddeley and J. Yiend, ‘Oops!:’ Performance correlates of everyday attentional failures in traumatic brain injured and normal subjects. Neuropsychologia, 35 (1997), pp. 747–758.
Rosvold et al., 1956 H.E. Rosvold, A.F. Mirsky, I. Sarason, E.D. Bransome and L.N. Beck, A continuous performance test of brain damage. Journal of Consulting Psychology, 20 (1956), pp. 343–350.
Rupp, 1932 H. Rupp, Ueber Arbeitsschnelligkeit und Arbeitsgüte/Concerning speed and accuracy or work. Psychotechnisches Zeitschrift, 7 (1932), pp. 161–175.
Salthouse, 1979 T.A. Salthouse, Adult age and speed–accuracy trade-off. Ergonomics, 22 (1979), pp. 811–821.
Schneider et al., 2002 W. Schneider, A. Eschman and A. Zuccolotto, E-prime user’s guide, Psychology Software Tools, Pittsburgh, PA (2002).
See et al., 1995 J.E. See, S.R. Howe, J.S. Warm and W.N. Dember, Meta-analysis of the sensitivity decrement in vigilance. Psychological Bulletin, 117 (1995), pp. 230–249.
Seli et al., in press Seli, P., Cheyne, J. A., Barton, K. R., & Smilek, D. (in press). Consistency of sustained attention across modalities: Comparing visual and auditory versions of the SART. Canadian Journal of Experimental Psychology. doi:10.1037/a0025111.
Shallice et al., 2002 T. Shallice, G.M. Marzocchi, S. Coser, M. Del Savio, R.F. Meuter and R.I. Rumati, Executive function profile of children with attention deficit hyperactivity disorder. Developmental Neuropsychology, 21 (2002), pp. 43–71.
Smallwood, Baracaia, et al., 2003 J.M. Smallwood, S.F. Baracaia, M. Lowe and M. Obonsawin, Task related thought whilst encoding information. Consciousness and Cognition, 12 (2003), pp. 452–484.
Smallwood, Obonsawin et al., 2003 J. Smallwood, M.C. Obonsawin and S.D. Heim, Task unrelated thought: The role of distributed processing. Consciousness and Cognition, 12 (2003), pp. 169–189.
Smallwood et al., 2004 J.M. Smallwood, R.C. O’Connor, M.V. Sudbery, C. Haskell and C. Ballantine, The consequences of encoding information on the maintenance of internally generated images and thoughts: The role of meaning complexes. Consciousness and Cognition, 13 (2004), pp. 789–820.
Smallwood et al., 2007 J.M. Smallwood, R.C. O’Connor, M.V. Sudbery and M. Obosawin, Mind-wandering and dysphoria. Cognition and Emotion, 21 (2007), pp. 816–842.
Smallwood and Schooler, 2006 J. Smallwood and J.W. Schooler, The restless mind. Psychological Bulletin, 132 (2006), pp. 946–958.
Smilek et al., 2010a D. Smilek, J.S.A. Carriere and J.A. Cheyne, Failures of attention in life, lab, and brain: Ecological validity of the SART. Neuropsychologia, 48 (2010), pp. 2564–2570.
Smilek et al., 2010b D. Smilek, J.S.A. Carriere and J.A. Cheyne, Out of mind, out of sight: The eyeblink as indicator and embodiment of mind wandering. Psychological Science, 21 (2010), pp. 786–789.
Smith-Chant and LeFevre, 2003 B.L. Smith-Chant and J.-A. LeFevre, Doing as they are told and telling like it is: Self reports in mental arithmetic. Memory and Cognition, 31 (2003), pp. 516–528.
Stevenson et al., 2011 Stevenson, H., Russell, P. N., & Helton, W. S. (2011). Search symmetry, sustained attention, and response inhibition. Brain and Cognition. doi:10.1016/j.bandc.2011.08.007.
Stins et al., 2007 J.F. Stins, J.C. Polderman, D.I. Boomsma and E.J.C. de Geus, Conditional accuracy in response interference tasks: Evidence from the Eriksen flanker task and the spatial conflict task. Advances in Cognitive Psychology, 3 (2007), pp. 389–396.
Stuss et al., 2003 D.T. Stuss, K.J. Murphy, M.A. Binns and M.P. Alexander, Staying on the job: The frontal lobes control individual performance variability. Brain, 126 (2003), pp. 2363–2380.
van den Wildenberg et al., 2010 W.P.M. van den Wildenberg, S.A. Wylie, B.U. Forstmann, B. Burle, T. Hasbroucq and K.R. Ridderinkoff, To head or heed? Beyond the surface of selective action inhibition: A review. Frontiers in Human Neuroscience, 4 (2010), pp. 1–13.
van der Linden et al., 2005 D. van der Linden, G.P.G. Keijsers, P. Eling and R. Van Schaijk, Work stress and attentional difficulties: An initial study on burnout and cognitive failures. Work & Stress, 19 (2005), pp. 23–36.
Warm, 1984 J.S. Warm, Editor, Sustained attention in human performance, Wiley, Chichester, UK (1984).
Warm et al., 2008 J.S. Warm, R. Parasuraman and G. Matthews, requires hard mental work and is stressful. Human Factors, 50 (2008), pp. 433–441.
Willmott et al., 2009 C. Willmott, J. Ponsford, C. Hocking and M. Schönberger, Factors contributing to attentional impairments after traumatic brain injury. Neuropsychology, 23 (2009), pp. 424–432.
Wing and Kristofferson, 1973 A.M. Wing and A.B. Kristofferson, Response delays and the timing discrete motor responses. Perception and Psychophysics, 14 (1973), pp. 5–12.
Woodworth, 1899 Woodworth, R. S. (1899). The accuracy of voluntary movement. Psychological Monographs, 3 (2) (whole no. 3)..
Wylie et al., 2009 S.A. Wylie, W.P.M. van den Wildenberg, K.R. Ridderinkhof, T.R. Bashore, V.D. Powell and C.A. Manning, et al. The effect of speed–accuracy strategy on response interference control in Parkinson’s disease. Neuropsychologia, 47 (2009), pp. 1844–1853.