Measuring the Allocation of Attention in the Stroop Task

Embed Size (px)

Citation preview

  • 8/10/2019 Measuring the Allocation of Attention in the Stroop Task

    1/10

    Psychological Research

    DOI 10.1007/s00426-011-0405-9

    1 3

    ORIGINAL ARTICLE

    Measuring the allocation of attention in the Stroop task: evidence

    from eye movement patterns

    Bettina Olk

    Received: 6 September 2011 / Accepted: 6 December 2011

    Springer-Verlag 2011

    Abstract Attention plays a crucial role in the Stroop task,

    which requires attending to less automatically processed

    task-relevant attributes of stimuli and the suppression of

    involuntary processing of task-irrelevant attributes. The

    experiment assessed the allocation of attention by monitor-

    ing eye movements throughout congruent and incongruent

    trials. Participants viewed two stimulus arrays that diVered

    regarding the amount of items and their numerical value

    and judged by manual response which of the arrays con-

    tained more items, while disregarding their value. DiVerent

    viewing patterns were observed between congruent (e.g.,

    larger array of numbers with higher value) and incongruent

    (e.g., larger array of numbers with lower value) trials. The

    direction of Wrst saccades was guided by task-relevant

    information but in the incongruent condition directed more

    frequently towards task-irrelevant information. The data

    further suggest that the diVerence in the deployment of

    attention between conditions changes throughout a trial,

    likely reXecting the impact and resolution of the conXict.

    For instance, stimulus arrays in line with the correct

    response were attended for longer and Wxations were longer

    for incongruent trials, with the second Wxation and consid-

    ering all Wxations. By the time of the correct response, this

    latter diVerence between conditions was absent. Possible

    mechanisms underlying eye movement patterns are dis-

    cussed.

    Introduction

    Research on attentional control shows that we are able to

    ignore some features and stimuli in our environment, but not

    others. One of the most prominent tasks to study control is

    the Stroop interference task (Stroop, 1935), which requires

    the suppression of involuntary processing of task-irrelevant

    attributes of a stimulus and preferred attending to less auto-

    matically processed task-relevant attributes (Macleod, 1991).

    These demands make the Stroop task very informative for

    the examination of attentional control (Banich et al., 2000).

    The wordcolour task is probably the best known Stroop

    task. A variant of the task is that participants are required to

    report the colours in which words are presented. The words

    can be congruent or incongruent with the colour. When the

    colour and the word are congruent, i.e., the word GREEN is

    shown in green, naming the colour is faster than when the

    colour and the word are incongruent, i.e., the word GREEN

    is shown in red. In the latter situation, the conXict slows

    responding, which is reXected in the Stroop eVect. Other

    variants of the Stroop task use numbers as items. In a numer-

    ical version participants judge on which side of a display

    more items are shown, irrespective of the identity of the

    items (Washburn, 1994). In the congruent condition the

    larger array of numbers consists of numbers with higher

    value than the numbers of the smaller array. In the incongru-

    ent condition the larger array of numbers is composed of

    numbers with lower value than the numbers of the smaller

    array (see Fig. 1for examples). Participants manual reaction

    times (RT) are slower in the incongruent than in the congru-

    ent condition, indicating that the task-irrelevant attribute

    the value of the numbersis processed involuntarily and

    slows down responding (Wolach, McHale, & Tarlea, 2004).

    The allocation of attention is proposed to play a large

    role in mediating the Stroop eVect. According to theoretical

    B. Olk (&)

    School of Humanities and Social Sciences,

    Jacobs University Bremen, Campus Ring 1,

    28759 Bremen, Germany

    e-mail: [email protected]

  • 8/10/2019 Measuring the Allocation of Attention in the Stroop Task

    2/10

    Psychological Research

    1 3

    frameworks of attention, e.g., the biased competition

    framework (Desimone & Duncan, 1995; Duncan, 2006),

    attention is limited in capacity and selective, i.e., unwanted

    information can be Wltered out to a certain extent. It is

    understood that stimuli are competing for processing and

    that this competition is biased, for instance by task rele-

    vance and by stimulus characteristics. It is suggested that a

    short-term description (attentional template) of currently

    required information is used to bias competition (Desimone

    & Duncan, 1995). Competitive processing as proposed by

    the framework is seen as a stronger response to any one

    object [] bought at the cost of weaker response to others

    and this competition is shown in relative enhancement ofthe response to task-relevant objects, or relative suppres-

    sion of objects to be ignored (Duncan, 2006, p. 5). These

    mechanisms are also reXected in models that grant attention

    an important role in the Stroop task. For instance, the paral-

    lel distributed processing account (Cohen, Dunbar, & McC-

    lelland, 1990; see also Botvinick, Braver, Barch, Carter, &

    Cohen, 2001; Cohen, Aston-Jones, & Gilzenrat, 2004;

    Cohen, Servan-Schreiber, & McClelland, 1992) illustrates

    how attention may contribute to the Stroop eVect and the

    resolution of conXict.1The model suggests two processing

    pathways, one for the processing of the task-relevant attri-

    butes of the stimulus, i.e., in the numerical version thiswould be the amount of numbers in an array, and one for

    the processing of the task-irrelevant attributes, i.e., the

    value of the numbers, which both converge on a common

    response mechanism. Each pathway consists of input units,

    intermediate units and output units, along which activation

    is spread. When suYcient activation has accumulated to

    exceed a response threshold, a response is given. Impor-

    tantly, the model assumes two task demand/attention units

    that are linked to the intermediate units in the two pathways

    and their function is to allocate attention to one of the two

    pathways. The instructions given to a participant, for exam-

    ple, to assess on which side of the display more items are

    shown, will activate the task demand/attention units. The

    activation of a given task demand/attention unit will thensensitize the processing in the task-relevant pathway, sup-

    porting eVective information Xow along it. In contrast,

    accumulation of information in the other pathway is attenu-

    ated. Thus, information, on which attention is allocated, is

    processed to a greater degree than information on which no

    or less attention is allocated. Attention is understood to act

    as a modulator which selects one of the two competing pro-

    cesses based on the task instructions. Later versions of the

    model (e.g., Cohen et al., 2004) also recognize that features

    of the stimuli and their representation may have an impact

    on attentional allocation as well and that competition in

    attention may occur. Thus, the presentation of a stimulus

    will divert attention away from another stimulus (Cohen

    et al., 1992) and the presentation of a distracting stimulus

    can activate the task demand/attention units for the location

    containing the distracting stimulus. Taken together, in this

    model attention is deWned as the inXuence that activity in

    the task-demand units has on processing in the two path-

    ways and the representations of task instructions and of a

    stimulus inXuence which dimensions of a stimulus are pro-

    cessed (Cohen et al., 2004).

    Considering the prominent role that is attributed to atten-

    tion in the Stroop task, the present study set out to investi-

    gate where attention is actually allocated in the course of a

    trial in this task and whether the distribution of attention

    diVers between congruent and incongruent trials, as would

    be expected based on the theoretical considerations and

    model outlined above. A powerful way of measuring where

    attention is allocated is to monitor saccadic eye movements

    and Wxations. It has been established that attention and eye

    movements are ultimately linked (e.g., Deubel & Schnei-

    der, 1996) and that the direction of saccades and the loca-

    tion and duration of Wxations are informative with respect

    to the allocation of attention during the solution process of

    tasks and cognitive processing (Charness, Reingold, Pom-

    plun, & Stampe, 2001; Grant & Spivey, 2003; Thomas &

    Lleras, 2007; Olk, Harvey, & Gilchrist, 2002). Eye move-

    ment patterns provide information that is not obtained by

    measuring manual reaction times and error rates alone. As

    but one example, a study by Grant and Spivey (2003) was

    successful in revealing that attention and the thought pro-

    cess during problem solving are related by measuring eye

    movement patterns. In a similar manner, the present study

    sought to investigate whether the conXict participants expe-

    rience in the Stroop task is reXected in the allocation of

    1 The parallel distributed processing account has originally been

    applied to the wordcolour task.

    Fig. 1 Examples of displays of trials in the (a) congruent condition,

    and (b) incongruent condition in Experiment 1. Displays are not drawn

    to scale

  • 8/10/2019 Measuring the Allocation of Attention in the Stroop Task

    3/10

    Psychological Research

    1 3

    attention during a trial, as assessed with eye movements. To

    this end, saccades and Wxations in the course of a trial were

    analysed.

    In order to be able to analyse the direction of a saccade

    to and the location of a Wxation on task-relevant and task-

    irrelevant attributes, a numerical version of the Stroop task

    was chosen because it provides the possibility to spatially

    separate attributes of the stimuli that are critical for the con-Xict in the incongruent condition. The latter is not possible

    in the wordcolour version as here word and colour occupy

    the same location. The experiment contained two condi-

    tions. In the congruent condition, an array containing more

    numbers was composed of numbers with higher value than

    the numbers of an array with fewer numbers, e.g., Wve 3s

    versus four 2s. In the incongruent condition, an array con-

    taining more items was composed of numbers of smaller

    value than an array with fewer items, e.g., four 3s versus

    Wve 2s (see Fig. 1). Participants judged by button press

    which stimulus array was more numerous, irrespective of

    the identity of the numbers. Importantly, in the incongruent

    condition the two conXicting pieces of information related

    to the concept of more are presented in separate visual

    Welds. In one Weld task-relevant numerosity is larger and in

    the other task-irrelevant value is higher. This allows assess-

    ing which feature (task-relevant or task-irrelevant) may

    guide attention and the eyes. By comparing eye movement

    patterns between the congruent and incongruent condition

    information with respect to the allocation of attention dur-

    ing task completion can be gained as well as its relation to

    the Stroop eVect. Participants were not instructed where to

    look and for how long in order to assess their natural view-

    ing behaviour.

    The allocation of attention in the congruent and incon-

    gruent condition may vary throughout the trial. Therefore,

    diVerent stages of each trial were evaluated. First, it was

    assessed whether already at the beginning of a trial the allo-

    cation of attention would diVer between congruent and

    incongruent trials. It was analysed how fast participants

    would leave the starting positionthe centre of the

    screento explore the arrays, as reXected by the saccadic

    reaction time (SRT) of the Wrst saccade after display onset.

    It was hypothesised that a diVerence between the congruent

    and incongruent conditionthe Stroop conXictcould

    already be mirrored in the SRT, i.e., participants may take

    longer to leave the start position in incongruent than in con-

    gruent trials, in line with longer manual reaction times that

    are typically found in the incongruent compared to congru-

    ent condition. Second, it was analysed whether a diVerence

    between conditions would be apparent when considering

    the location to which attention is directed immediately after

    leaving the central starting position. If so, a diVerence with

    respect to the direction of the Wrst saccade after stimulus

    onset should be found between conditions. Participants are

    supposed to judge which array contains more items; hence,

    considering the impact of task instructions, more Wrst sac-

    cades should be directed to the stimulus array with more

    items (task-relevant). However, this may mainly be the case

    in the congruent condition and the presence of a conXict in

    the incongruent compared to congruent condition may be

    mirrored by fewer Wrst saccades to the array with more

    items in the incongruent than the congruent condition. Infact, considering that the processing of the task-irrelevant

    value of the numbers should not be suppressed completely

    in this task (Washburn, 1994), leading to conXict in the

    incongruent condition and the Stroop eVect, it is likely that

    attention is attracted by numbers with a higher value and

    hence within the incongruent condition more Wrst saccades

    should be directed towards the stimulus array with numbers

    with higher value (task-irrelevant). Third, the duration of

    the Wrst and second Wxation after the Wrst saccade,2 sepa-

    rately for array and condition, and fourth, the average dwell

    time (Wxation duration) on a stimulus array considering all

    Wxations during an entire trial were measured to assess

    whether those Wxation durations would reXect the Stroop

    eVect. In accordance with the task instructions, participants

    should spend more time on the location that contains the

    stimuli corresponding to the correct response, i.e., the

    stimulus array with more items. However, such a pattern

    may again be modulated by condition. In the incongruent

    condition, attention may be attracted by the higher value of

    numbers in an array and therefore Wxation duration may be

    longer on the stimulus array with fewer items but numbers

    of higher value in the incongruent condition. Finally, the

    Wxation location and its duration at response were assessed

    to elucidate the locus of attention when the response is

    given. It is expected that by that time participants have

    decided for the correct response and are therefore attending

    more often/longer to the stimulus array with more items,

    irrespective of condition.

    Methods

    Participants

    Fifteen participants, ranging from 18 to 27 years of age

    (M= 21 years), with normal or corrected vision gave

    informed consent and received course credits or were paid

    for their participation. The experiment was conducted in

    accordance with the ethical standards laid down in the 1964

    Declaration of Helsinki.

    2 As the number of saccades and Wxations diVered between trials not

    every saccade/Wxation made in a trial was analysed separately.

  • 8/10/2019 Measuring the Allocation of Attention in the Stroop Task

    4/10

    Psychological Research

    1 3

    Stimuli and procedure

    At the beginning of a trial a Wxation dot (diameter 0.5) was

    presented until Wxation was stable and the trial was initi-

    ated. A new, identical Wxation dot was then presented for a

    variable, random interval of another 8001,200 ms, and

    then two arrays of stimuli were shown (Fig. 1). The stimu-

    lus arrays appeared on the left and right on 50% of the trialsand above and below the centre in 50% of the trials. The

    order of locations was randomized within an experimental

    block. The arrays appeared simultaneously and until a man-

    ual response was given; maximally for 3,000 ms. Each

    array consisted of three to six items.3 The two arrays

    diVered in numerosity by one or two items. The distance

    between the central Wxation dot and the inner edge of a

    number array was 2 of visual angle. Stimuli were black

    and presented on a white background in the font Arial.

    Each stimulus extended 1.51.9 in height and 0.91.1

    in width. A stimulus array extended between 34.5 in

    height and 34 in width. Participants heads rested on a

    chin rest at a distance of 57 cm. Displays were presented on

    a 21CRT monitor.

    Of the 320 trials per participant 50% of trials were con-

    gruent, 25% were incongruent. Stimulus arrays were com-

    posed of the numbers 2, 3, 4 or 5. In congruent trials the

    larger array was composed of the numbers with higher

    value. In incongruent trials the larger array was composed

    of the numbers of smaller value. The diVerent types of trials

    with respect to amount of items and value were counterbal-

    anced between visual Welds. The remaining trials were Wller

    trials, in which letter stimuli (A, B, C or D) were shown to

    prevent participants from generating hypotheses with

    respect to a relationship between the number of items in an

    array and the value. All types of trials (congruent, incon-

    gruent, Wller) were randomized within a block. The task

    was to judge by button press, which array contained more

    items (pressing the " key with the right index Wnger, #

    with the right thumb, ! with the right middle Wnger and

    with the left index Wnger on a keyboard), while ignor-

    ing the value of the items. The spatial eccentricity of the

    displays allowed solving the task without looking at the

    arrays. No instructions were given regarding eye move-

    ments. Accuracy and manual RT were recorded in addition

    to eye movements. Participants obtained detailed written

    instructions and examples before the experiment began.

    Eye movement recording and data analysis

    Eye movements were recorded with an EyeLink II eye

    tracker (SR Research, Canada). Testing was preceded by a

    9-point grid calibration and validation for which partici-

    pants saccaded to a black circle (0.5) on white back-

    ground, which appeared sequentially at 9 points in a square

    array. Eye movements were recorded at 500 Hz sample rateat a spatial resolution of about 0.3. Participants viewed the

    displays with both eyes, but only the data of the right eye

    were recorded.

    In all subsequent manual and saccadic reaction time as

    well as saccade direction analyses, trials were rejected if

    (a) a blink (3%) or recording error occurred (0.5%), (b) the

    initial Wxation location at the start of a trial was not within

    1 around the centre of the Wxation dot (9.1%), (c) a sac-

    cade began before the stimuli appeared (2.9%) or (d) a but-

    ton other than one of the assigned response buttons was

    pressed (0.1%). For the eye movement analyses only trials

    with correct responses were considered.

    Results

    Manual responses

    One additional trial had to be excluded due to anticipation

    (RT < 100 ms). No responses were slower than 3,000 ms.

    The analysis of manual reaction times for correct responses

    showed that participants responded signiWcantly faster in

    the congruent (M= 656 ms) than in the incongruent

    (M= 705 ms) condition, t(14) = 7.6,p < .001. The analysis

    of the accuracy data showed that participants were also sig-

    niWcantly more accurate in the congruent (M= 99.3% cor-

    rect) than incongruent (M= 96% correct) condition,

    t(14) = 3, p < .01, reXecting the Stroop eVect and showing

    that task-irrelevant value information interfered with their

    responses.

    Trials without saccade

    On 7% of the trials no saccade was made. Participants were

    still able to provide a correct judgement. On congruent tri-

    als they were 100% correct (out of 146 trials), on incongru-

    ent trials in 90.6% of the trials (out of 64 trials). The

    responses were also given quite fast (congruent:

    M= 511 ms; incongruent: M= 550 ms), indicating that

    making no saccade was not detrimental for solving the task.

    It was thus possible to solve the task without making an eye

    movement or, in other words, with deploying attention

    covertly. However, trials without saccades were rare, sug-

    gesting that participants preferred to allocate attention

    overtly and to move their eyes. Because trials without

    3 No Wller items were used (see Pansky & Algom, 2002) to allow easy

    identiWcation of the amount of items without the need for detailed

    visual search of a stimulus array as this would implicitly encourage

    strategic eye movements, which would not be triggered by the amount

    or the value of the numbers but simply the need to search.

  • 8/10/2019 Measuring the Allocation of Attention in the Stroop Task

    5/10

    Psychological Research

    1 3

    saccades were rare and not present for every participant,

    those data were not analysed further.

    Latency of theWrst saccade

    The distribution of saccadic reaction times (SRT) was uni-

    modal (M= 217 ms, SD = 72 ms). For the analysis the

    mean SRT of the Wrst saccade was calculated separately forcongruent and for incongruent trials. On average SRT was

    M= 221 ms (SD = 55 ms) on congruent trials and

    M= 228 ms (SD = 71 ms) on incongruent trials. The diVer-

    ence in SRT between conditions was not signiWcant,

    t(14) = 1.3, p = .227. Because SRT was generally rather

    short and hence some of the included saccades could be

    anticipatory in nature, the data were also analysed including

    only the Wrst saccades with an SRT > 200 ms, leaving

    1,754 trials for the analysis. Including only those saccades,

    mean SRT on congruent trials was M= 254 ms (SD =

    41 ms) and on incongruent trialsM= 261 ms (SD = 60 ms).

    Again, the diVerence between conditions was not signiW-

    cant, t(14) = 1.1,p = .313.

    Direction of theWrst saccade

    Next, it was analysed in which percentage of trials in the

    congruent and in the incongruent condition the Wrst saccade

    was directed towards the array with more items and to the

    array with fewer items. Trials in which a saccade was not

    directed towards any of the arrays, i.e., when the stimuli

    were shown on the left and right but a saccade was made

    up- or downward or when the stimuli were shown above

    and below the centre and a saccade was made to the left or

    right, were rare and excluded (0.9%).

    A repeated measures ANOVA showed that partici-

    pants looked more frequently to the array with more

    items, F(1, 14) = 25.6, p < .001. This eVect was modu-

    lated by an interaction between congruency amount of

    items (more, fewer), F(1, 14) = 23.6,p < .001. To inves-

    tigate the interaction further, paired-samples ttest were

    conducted and showed that in the congruent and in the

    incongruent condition more Wrst saccades were directed

    to the visual Weld that contained the stimulus in line with

    the correct response, i.e., with more items (congruent:

    68.1 vs. 31.9%, t(14) = 6.7,p < .001; incongruent: 59.7

    vs. 40.3%, t(14) = 3.2,p < .01, see Fig. 2). The analysis

    shows that the task (where is more?) guided the Wrst sac-

    cades in both conditions. Importantly though, the inter-

    action and following comparisons between both

    conditions indicate that the value of the numbers had

    also an eVect. The percentage of the Wrst saccades to the

    array with more items was signiWcantly higher in the

    congruent condition (more items/higher value: 68.1%)

    than in the incongruent condition (more items/lower

    value: 59.7%), t(14) = 4.9, p < .001. In the incongruent

    condition, signiWcantly more Wrst saccades went to the

    array with fewer items (fewer items/higher value:

    40.3%) than in the congruent condition (fewer items/

    lower value: 31.9%), t(14) = 4.9, p < .001. Thus, the

    task-irrelevant higher value of the numbers did attract

    the Wrst saccades in the incongruent condition. The anal-

    ysis only including saccades with SRT > 200 ms showed

    the same results, a main eVect of amount of items,

    F(1, 14) = 35.5, p < .001, modiWed by an interaction,

    F(1, 14) = 13, p < .005. In the congruent and in the

    incongruent condition more saccades were directed to

    the array with more than with fewer items (congruent:

    71 vs. 29%, t(14) = 7.4, p < .001; incongruent: 62 vs.

    38%, t(14) = 3.9, p < .005). Comparing the conditions,

    more saccades went to the array with more items in the

    congruent condition (higher value) than in the incongru-

    ent condition (lower value), t(14) = 3.6, p < .005. The

    opposite was true for saccades to fewer items with more

    saccades to those arrays in the incongruent condition,

    t(14) = 3.6,p < .005.

    Duration of theWrst Wxation

    Duration of the Wrst Wxation (following the Wrst saccade)

    showed that participants Wxated longer on the array that

    contained more items (congruent: M= 235 ms; incongru-

    ent:M= 229 ms) than fewer items (congruentM= 164 ms;

    incongruent M= 172 ms), F(1, 14) = 64.2, p < .001. This

    eVect was not modulated by an interaction with congru-

    ency,p > .05, showing that this pattern was true for congru-

    ent as well as incongruent trials. There was also no

    diVerence between congruent and incongruent trials with

    respect to longer Wxation durations on the arrays with more

    or with fewer items, allp > .05. The results are illustrated in

    Fig. 3a.

    Fig. 2 Percentage of Wrst saccades directed to the stimulus arrays,

    considering the amount of items (more/fewer) and value of the num-

    bers (higher/lower), shown separately for the congruent and incongru-

    ent condition.Error barsindicate standard errors

    68 %60 %

    32 %40 %

    0

    20

    40

    60

    80

    100

    congruent incongruentPercentage

    offirstsaccades

    Condition

    Percentage of first saccades to arrays

    more items fewer items

    highervalue

    highervalue

    lowervalue

    lowervalue

  • 8/10/2019 Measuring the Allocation of Attention in the Stroop Task

    6/10

    Psychological Research

    1 3

    Duration of the second Wxation

    With respect to the second Wxation, participants Wxated

    longer on incongruent than congruent trials, F(1, 14) = 6,

    p < .05, and Wxated longer on the array that contained more

    than fewer items, F(1, 14) = 26.4, p < .001. This pattern

    was true in the congruent and incongruent condition (con-

    gruent more: M= 243 ms vs. fewer: 203 ms; incongruent

    more: M= 272 ms vs. fewer: 212 ms), as the interaction

    with congruency was not signiWcant, p > .05, illustrated in

    Fig. 3b.

    Fixation duration on stimulus arrays during the trial

    In order to assess for how long during a trial participants

    dwelled on the stimuli, it was calculated how much time

    they spent, on average, in each condition (congruent, incon-

    gruent), on each of the two stimulus arrays averaged over

    all Wxations until the correct response button was pressed.

    For trials on which the stimulus arrays were shown to the

    left and right of the centre, participants looked only rarely

    above or below the centre, some participants not at all. For

    trials on which the stimulus arrays were shown above and

    below the centre, participants looked only rarely to the left

    or right, with some participants not doing so at all. Thoserare Wxations were excluded (0.2% of trials). Fixation dura-

    tion was signiWcantly longer on incongruent than congruent

    trials, F(1, 14) = 5.9, p < .05, and on the stimulus array

    with more items and corresponding to the correct response,

    F(1, 14) = 32.4,p .001. As the interaction was not signiWcant,

    this pattern was true for congruent as well as incongruent

    trials. On congruent trials participants spent on average

    235 ms on the array with more items and 192 ms on the

    array with fewer items. On incongruent trials they spent on

    average 244 ms on the array with more items and 194 ms

    on the array with fewer items. The results are illustrated in

    Fig. 3c.

    Fixation location and duration at response

    The percentage of trials participants Wxated the array with

    more or fewer trials while they gave their manual response

    was analysed. Here, no diVerence in percentage of Wxations

    at response between the arrays with more or fewer items or

    between congruency of conditions was found, all p > .05.

    In congruent trials, the Wnal Wxation was on the array with

    Fig. 3 Average Wxation duration on stimulus arrays for (a) the Wrst

    Wxation, (b) the second Wxation, (c) all Wxations, and (d) at response,

    considering amount of items (more/fewer) and value of the numbers

    (higher/lower), depicted separately for the congruent and incongruent

    condition.Error barsindicate standard errors

    235 ms 229 ms

    164 ms172 ms

    0

    50

    100

    150

    200

    250

    300

    350

    congruent incongruent

    Fixationduration(inms)

    Condition

    congruent incongruent

    Condition

    congruent incongruent

    Condition

    congruent incongruent

    Condition

    Average duration on arrays (first fixation)

    more items fewer items

    highervalue

    highervalue

    lowervalue

    lowervalue

    243 ms272 ms

    203 ms

    212 ms

    0

    50

    100

    150

    200

    250

    300

    350

    Fixationduration

    (inms)

    Average duration on arrays (second fixation)

    more items fewer items

    highervalue

    highervalue

    lowervalue

    lowervalue

    235 ms 244 ms192 ms

    194 ms

    0

    50

    100

    150

    200

    250

    300

    350

    Fixationduration(inms)

    Average duration over all fixations on arrays

    during a trial

    more items fewer items

    highervalue

    highervalue

    lowervalue

    lowervalue

    282 ms 289 ms

    210 ms 213 ms

    0

    50

    100

    150

    200

    250

    300

    350

    Fixationduratio

    n(inms)

    Average fixation duration on arrays at response

    more i tems fewer items

    highervalue

    highervalue

    lowervalue

    lowervalue

    (a)

    (b) (d)

    (c)

  • 8/10/2019 Measuring the Allocation of Attention in the Stroop Task

    7/10

    Psychological Research

    1 3

    more items on 49% of the trials and on the array with fewer

    items on 51% of the trials. For incongruent trials, the array

    with more items was Wxated in 53% and with fewer items

    on 47%. With respect to the duration of the last Wxations

    (see Fig. 3d), participants Wxated not longer on incongruent

    than congruent trials, p > .05, but longer on the array with

    more than with fewer items, F(1, 14) = 65.2,p < .001. The

    interaction was not signiWcant, p > .05, showing that thiseVect was true in the congruent (M= 282 ms,M= 210 ms,

    respectively), and incongruent condition (M= 289 ms,

    M= 213 ms, respectively).

    Discussion

    The goal of the present study was to characterize the alloca-

    tion of attention in the course of a trial in the Stroop task by

    analysing and comparing eye movement patterns in the

    congruent and incongruent condition. Participants indicated

    by button press which of two stimulus arrays contained

    more stimuli and their eye movements were recorded while

    they did so. The manual reaction time and accuracy mea-

    surements showed that a conXict occurred, and that task-

    irrelevant information was processed and interfering and

    hence RT was longer and accuracy lower in the incongruent

    condition. This Wnding is in line with a vast amount of liter-

    ature (Macleod, 1991) and proving the eVectiveness of the

    materials used in the present study.

    To demonstrate where attention was deployed during

    this task eye movement patterns between the congruent and

    incongruent condition were compared. The Wrst analysis

    focused on the beginning of a trial, i.e. how fast participants

    are leaving the central starting position and are initiating

    their Wrst saccade to explore the arrays. No diVerence

    between congruent and incongruent trials was found, show-

    ing that participants are not slower to initiate a move of

    attention away from the centre at the beginning of the trial

    in the incongruent condition when the correct response is

    given. These data suggest that the Stroop eVect observed in

    the manual response times with slower manual RT in the

    incongruent condition is not related to the speed with which

    the Wrst eye movement is initiated.

    Further, it was analysed whether the Stroop conXict

    would be apparent when considering the location to which

    attention is directed immediately after leaving the central

    starting position. It was expected that participants may look

    more often towards the array with more items, in line with

    task demands, but that a diVerence between the congruent

    and incongruent condition may be apparent. More speciW-

    cally, the presence of a conXict in the incongruent com-

    pared to congruent condition could be reXected by fewer

    Wrst saccades to the array with more items in the incongru-

    ent than the congruent condition. In line with the hypothe-

    sis, in both conditions more Wrst saccades were directed to

    the visual Weld with the stimulus with more items but the

    percentage of Wrst saccades to this array was higher in the

    congruent than in the incongruent condition. In the incon-

    gruent condition, more Wrst saccades were directed to the

    array with fewer items/higher value than in the congruent

    condition, indicating that the task-irrelevant attribute

    (higher value) attracted attention in the incongruent condi-tion. The direction of the Wrst saccade thus reXects the con-

    Xict and may demonstrate that the task (where is more?)

    activates a concept or sets up an attentional template

    (Desimone & Duncan, 1995) that is initially still indistinct

    enough to bias attention not only towards task-relevant

    (more items) but also task-irrelevant (higher value) attri-

    butes or, alternatively, initiallyat the beginning of a

    trialhas less impact. In relation to the parallel distributed

    processing model (e.g., Cohen et al., 2004), this Wnding

    shows that also information in the task-irrelevant pathway

    is attended to, albeit to a lesser degree than information in

    the task-relevant pathway, because overall, more Wrst sac-

    cades were directed to the array with more items. The latter

    conclusion is supported by the Wxation durations that were

    analysed, as described further down below.

    The data of the Wrst saccade thus shows that participants

    look more often towards the array with stimuli with higher

    value in the incongruent compared to the congruent condi-

    tion but are equally fast to initiate the Wrst saccade in both

    conditions. It had been expected that they may be slower to

    initiate an eye movement in the incongruent condition. It

    thus seems that the conXict is not apparent in the latency

    and direction data alike. Considering that saccadic

    responses in both directions in the incongruent condition

    (to more items as well as to higher value) are in line with

    the concept or template that may be created, it is possible

    that the speed with which a saccade is initiated does not dis-

    tinguish between task-relevant (more) and task-irrelevant

    (higher). Alternatively, bearing in mind that saccades were

    initiated quite fast, it cannot be ruled out that a Xoor eVect

    occurred and that an SRT diVerence was only present

    numerically but not signiWcant.

    The analyses of the durations of the Wrst and second Wxa-

    tion after the Wrst saccade, the average Wxation duration

    considering all Wxations of a trial and Wxation duration at

    response, separately for array and condition, showed that

    participants spent more time Wxating the array with more

    items and this was the case in both conditions.4 Thus,

    although attention is directed to both arrays, it dwells for

    4 It is unlikely that the direction of saccades and Wxation durations were

    simply directed to the array containing more items because those were

    physically slightly larger. In an experiment with similar stimuli but

    diVerent instructions participants did not look more often to the physi-

    cally larger arrays (Olk, 2007).

  • 8/10/2019 Measuring the Allocation of Attention in the Stroop Task

    8/10

    Psychological Research

    1 3

    longer on the stimulus array that is in line with task

    demands and on which the stimuli corresponding to the cor-

    rect response were shown. It therefore appears that even

    though the Wrst sweep of attention as indicated by the direc-

    tion of Wrst saccades was aVected by number value and may

    be directed towards the location that does not correspond

    with the correct response in the incongruent condition, the

    nature of a stimulus array is discerned quickly and thenattention is preferably allocated to the array corresponding

    to the correct response and accordingly, dwells there for

    longer. This Wnding is in line with the idea that attention is

    allocated with respect to task demands and instructions and

    that it supports the correct response (e.g., Desimone &

    Duncan, 1995; Duncan, 2006; Cohen et al., 2004). Such a

    behaviour is also in agreement with Wndings reported in the

    eye movement literature, which show that high-level fac-

    tors such as goals, knowledge and attentional and working

    memory capacity (Henderson, 2003; Kane, Bleckly, Con-

    way, & Engle, 2001; Kappas & Olk, 2008; Olk & King-

    stone, 2003, 2009; Yarbus, 1967) aVect where we attend

    and look. In the present case, it is further possible that the

    impact of task demands increases throughout a trial. For

    instance, the concept or attentional template may gain

    more impact or it may be reWned more in the course of the

    trial, i.e., referring more to larger array than higher

    value.

    Interestingly, with their second Wxation and when all

    Wxations of a trial were considered participants spent more

    time Wxating in the incongruent condition. What could be

    the reason as to why participants looked longer in the

    incongruent condition? A viable answer may be that after

    some initial processing (after the Wrst saccade and the Wrst

    Wxation) the actual conXict is mostly perceived in the

    incongruent condition. The time that participants spent

    Wxating in this condition may partly reXect the time

    required to verify which array corresponds to the correct

    response. In the congruent condition, both attributes of an

    array, numerosity and number value, converge, i.e., for the

    array with more items they are in line with the activated,

    reWned concept or template more, but in the incongruent

    condition this is not the case. A further mechanism in play

    could be the inhibition of irrelevant information. For

    instance, the Selective Attention Model (SLAM, Phaf, van

    der Heijden, & Hudson, 1990) suggests that at Wrst all stim-

    uli may be processed and in the following irrelevant infor-

    mation may be actively inhibited (Phaf et al., 1990).

    Applying this idea to the present task, as such processes,

    may take time, it is conceivable that longer Wxations in

    incongruent trials are the result. For instance, after initial

    processing the impact of task demands may increase and

    the array with more items is activated and inhibition may be

    imposed on the numbers with higher value (of the array

    with fewer items). Or, inhibition may be imposed within

    the array with more items/lower value on the attribute

    lower value or on number value as such because it is in

    conXict with the task demands. The longer Wxation duration

    in the incongruent condition may thus reXect such a resolu-

    tion of the conXict.

    Finally, the Wxation location and its duration at response

    were assessed to monitor the locus of attention when the

    response was given. Here no diVerence in percentage ofWxations on arrays with more or less items was found in

    either condition but participants spent more time on the

    array with more items in both conditions. Note that the time

    spent Wxating in the incongruent compared to congruent

    condition was not longer anymore. It thus seems that by the

    time of the response participants have decided for the cor-

    rect response, are attending more often/longer to the stimu-

    lus array with more items, irrespective of condition. A lack

    of a diVerence between conditions may indicate that at this

    point the conXict in the incongruent condition was resolved

    and inhibition may dissipate.

    Taken together, the present experiment shows that atten-

    tion is mostly deployed to the array in line with the correct

    response. Initially though, it is also guided towards the

    task-irrelevant attribute higher value in the incongruent

    condition, as indicated by the direction of the Wrst saccade.

    Considering Wxation duration, particularly in the course of a

    trial until conXict resolution, more time is spent on the

    array with more items and in the incongruent condition.

    The present experiment is the Wrst that measured the

    allocation of attention, as indicated by eye movement pat-

    terns, throughout a trial in a Stroop task. The Wndings are

    important because they not only demonstrate that monitor-

    ing of eye movements is a very informative method for the

    investigation of cognitive conXicts but also because the

    obtained data suggest potential underlying mechanisms and

    lay the foundation for much further work. For instance,

    future studies may wish to incorporate eye movement mea-

    surements to investigate eVects of varying discriminability

    and salience of numerosity and numerical value (Pansky &

    Algom, 2002) or the similarity between the representations

    of relevant and irrelevant information (Pavese & Umilta,

    1998). Further, the role of working memory (WM) could be

    addressed. Recently, Kane and Engle (2003) have shown

    that working memory span is related to maintaining a goal

    in the wordcolour Stroop task. High-WM-span partici-

    pants demonstrated less interference than did low-span par-

    ticipants. It would thus be most instructive to assess

    whether low-span participants allocation of attention

    would diVer from attentional allocation of high-span partici-

    pants. For example, low-span participants may look less

    often towards the stimulus array in line with the task goal.

    Additional research could also explore whether persons

    with lapses of intention (West, 1999), with deWcient inhibi-

    tory control or who may have diYculties to suppress the

  • 8/10/2019 Measuring the Allocation of Attention in the Stroop Task

    9/10

    Psychological Research

    1 3

    task-irrelevant information and may thus be prone to

    enhanced interference (Kaufmann et al., 2008; Spieler,

    Balota, & Faust, 1996) would show diVerent eye movement

    patterns, i.e., would be attracted more by the value of the

    numbers. Further, it could be explored how guiding atten-

    tion in a Stroop task could aVect the conXict. In line with

    Grant and Spivey (2003) who showed that guiding attention

    towards the problem-relevant features promoted problemsolving, in the presently used task guiding attention

    towards the array with more items at the beginning of a trial

    could reduce the Stroop eVect. Interestingly, Perret and

    Ducrot (2010) could show that the Stroop eVect in the

    wordcolour version could be reduced by presenting the

    word stimuli in a way that participants Wrst Wxation was on

    a position in the word that is not optimal for reading, i.e.,

    towards the end of the word, reducing the eYciency of

    word processing. Attentional eVects could also be investi-

    gated further by varying the percentage of congruent and

    incongruent trials and thereby altering participants strate-

    gies of dividing attention and allocating attentional weight

    to the relevant and irrelevant dimension (Logan &

    ZbrodoV, 1979). Future work may also wish to apply a very

    diYcult Stroop task in which error rates would be high

    enough to be able to explore the diVerences in attentional

    allocation between trials in which correct and incorrect

    responses were given. Last but not least, the proposed

    mechanisms regarding the role of the impact of a concept

    and attentional template throughout a trial and inhibition

    of irrelevant information should be investigated further.

    Acknowledgements I would like to thank Bernhard Hommel, Wer-

    ner Schneider and Artem Belopolsky for very constructive commentson previous drafts of this manuscript.

    References

    Banich, M. T., Milham, M. P., Atchley, R., Cohen, N. J., Webb, A.,

    Wszalek, T., et al. (2000). FMRI studies of Stroop tasks reveal

    unique roles of anterior and posterior brain systems in attentional

    selection.Journal of Cognitive Neuroscience,12, 9881000.

    Botvinick, M., Braver, T., Barch, D., Carter, C., & Cohen, J. (2001).

    ConXict monitoring and cognitive control. Psychological Review,

    108(3), 624652.

    Charness, N., Reingold, E. M., Pomplun, M., & Stampe, D. M. (2001).

    The perceptual aspect of skilled performance in chess: evidencefrom eye movements.Memory and Cognition, 29, 11461152.

    Cohen, J. D., Aston-Jones, G., & Gilzenrat, M. S. (2004). A systems-

    level perspective on attention and cognitive control: guided acti-

    vation, adaptive gating, conXict monitoring, and exploitation vs.

    exploration. In M. I. Posner (Ed.), Cognitive neuroscience of

    attention(pp. 7190). New York: Guilford Press.

    Cohen, J. D., Dunbar, K., & McClelland, J. L. (1990). On the control

    of automatic processesa parallel distributed-processing account

    of the Stroop eVect. Psychological Review,97, 332361.

    Cohen, J. D., Servan-Schreiber, D., & McClelland, J. L. (1992). A par-

    allel distributed processing approach to automaticity. The American

    Journal of Psychology, 105, 239269.

    Desimone, R., & Duncan, J. (1995). Neural mechanisms of selective

    visual attention.Annual Review of Neuroscience,18, 193222.

    Deubel, H., & Schneider, W. X. (1996). Saccade target selection and

    object recognition: evidence for a common attentional mecha-

    nism. Vision Research,36, 18271837.

    Duncan, J. (2006). EPS Mid-Career Award 2004: brain mechanisms of

    attention. Quarterly Journal of Experimental Psychology, 59,

    227.

    Grant, E. R., & Spivey, M. J. (2003). Eye movements and problem

    solving: guiding attention guides thought. Psychological Science,

    14, 462466.

    Henderson, J. M. (2003). Human gaze control during real-world scene

    perception. Trends in Cognitive Science,7, 498504.

    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, 169183.

    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, 4770.

    Kappas, A., & Olk, B. (2008). The concept of visual competence as

    seen from the perspective of the psychological and brain sciences.

    Visual Studies,23, 162173.

    Kaufmann, L., Ischebeck, A., Weiss, E., Koppelstaetter, F., Sieden-topf, C., Vogel, S. E., et al. (2008). An fMRI study of the numer-

    ical Stroop task in individuals with and without minimal cognitive

    impairment. Cortex,44, 12481255.

    Logan, G. D., & ZbrodoV, N. J. (1979). When it helps to be misled:

    Facilitative eVects of increasing the frequency of conXicting stim-

    uli in a Stroop-like task. Memory and Cognition,7, 166174.

    Macleod, C. M. (1991). Half a century of research on the Stroop eVect

    an integrative review. Psychological Bulletin,109, 163203.

    Olk, B., (2007). Task-set guides eye movements: evidence from the

    Stroop eVect. Poster presented at the European Conference on

    Eye Movements, Potsdam, Germany.

    Olk, B., Harvey, M., & Gilchrist, I. D. (2002). First saccades reveal bi-

    ases in recovered neglect.Neurocase,8, 306313.

    Olk, B., & Kingstone, A. (2003). Why are antisaccades slower than

    prosaccades? A novelW

    nding using a new paradigm. Neurore-port,14, 151155.

    Olk, B., & Kingstone, A. (2009). A new look at aging and performance

    in the antisaccade task: the impact of response selection. Euro-

    pean Journal of Cognitive Psychology,21, 406427.

    Pansky, A., & Algom, D. (2002). Comparative judgment of numerosi-

    ty and numerical value: Attention preempts automaticity.Journal

    of Experimental Psychology. Learning, Memory, and Cognition,

    28, 259274.

    Pavese, A., & Umilta, C. (1998). Symbolic distance between numeros-

    ity and identity modulates Stroop interference.Journal of Exper-

    imental Psychology: Human Perception and Performance, 24,

    15351545.

    Perret, P., & Ducrot, S. (2010). Viewing position eVects in the Stroop

    task: Initial Wxation position modulates Stroop eVects in fully col-

    ored words. Psychonomic Bulletin and Review,17, 550555.Phaf, R. H., Van der Heijden, A. H. C., & Hudson, P. T. W. (1990).

    SLAM: a connectionist model for attention in visual selection

    tasks. Cognitive Psychology,22, 273341.

    Spieler, D. H., Balota, D. A., & Faust, M. E. (1996). Stroop perfor-

    mance in healthy younger and older adults and in individuals with

    dementia of the Alzheimers type. Journal of Experimental Psy-

    chology: Human Perception and Performance,22, 461479.

    Stroop, J. R. (1935). Studies of interference in serial verbal reactions.

    Journal of Experimental Psychology: General,121, 1523.

    Thomas, L. E., & Lleras, A. (2007). Moving eyes and moving thought:

    On the spatial compatibility between eye movements and cogni-

    tion. Psychonomic Bulletin and Review,14, 663668.

  • 8/10/2019 Measuring the Allocation of Attention in the Stroop Task

    10/10

    Psychological Research

    1 3

    Washburn, D. A. (1994). Stroop-like eVects for monkeys and humans

    processing speed or strength of association. Psychological Sci-

    ence,5, 375379.

    West, R. (1999). Age diVerences in lapses of intention in the stroop

    task. Journal of Geronto logy B: Psychologi cal Sciences and

    Social Sciences,54, 3443.

    Wolach, A. H., McHale, M. A., & Tarlea, A. (2004). Numerical Stroop

    eVect. Perceptual and Motor skills,98, 6777.

    Yarbus, A. L. (Ed.). (1967). Eye movements and vision. New York:

    Plenum Press.