Evidence for attentional processing in spatial localization

Thirty participants were randomly assigned to one of three conditions in a 2 × 8 × 3 mixed factorial design, crossing the within-subject factors task (single vs. dual) and target stimulus duration (29, 57, 86, 114, 143, 200, 300 or 400 ms) and the between-subject factor distractor load (1, 2, or 3 digits). Thus, there were three different groups of each 10 subjects (1-distractor group; 2-distractor group, and 3-distractor group), with each group performing the same single-task condition (localization of the peripheral target stimulus with eight target duration conditions) but a different dual-task condition (either 1, 2, or 3 to-be-identified distractor items).

All participants performed a visual localization task in two conditions (single- and dual-task). In the single-task condition, participants localized a peripherally presented target stimulus. In the dual-task condition, participants performed two perceptual tasks: identification and localization. Thus, in the dual-task condition there were two simultaneously presented visual stimuli: a distractor stimulus (containing 1, 2, or 3 digits) presented at fixation, which had to be identified; and a target stimulus (a single “*” sign) presented peripherally to the left or right of fixation, which had to be localized. After a variable delay that varied between 29 and 400 ms, the target stimulus (but not the distractor stimulus) was followed by a backward masking stimulus that eliminated its visibility.

At the beginning of each trial in the single task condition (localization only), a fixation sign (‘+’) appeared in the center of the screen (Fig. 1). After 1,000 ms the target stimulus (‘*’) was presented at one of 50 possible stimulus locations on an imaginary horizontal row on either side of the fixation sign (25 stimulus locations to the left and 25 to the right). After a variable delay (target-mask onset delay) a masking stimulus was presented to control the visibility of the target stimulus. The masking stimulus consisted of two horizontal strings of each 29 ‘*’ signs covering all possible target locations on the left and right of fixation (plus four extra, non-target positions). The masking stimulus remained present throughout the remainder of the trial. Eight target-mask onset asynchronies or target durations were employed: 29, 57, 86, 114, 143, 200, 300, and 400 ms. The participants’ task was to indicate the location of the target stimulus as accurately as possible by moving the cursor (a rectangle) from the fixation sign to the observed position of the target stimulus. Movement of the cursor was realized by manipulating the cursor keys on the keyboard with the index and ring fingers of the right hand. When subjects reached the perceived target location they confirmed their response by pressing the space bar with their left hand. No feedback was provided. An intertrial interval of 1.5 s separated the final response in one trial from the start of the next trial. Participants were instructed to fixate their eyes on the fixation sign at the beginning of each trial, but were told that they were free to make eye movements toward the target.

In the dual-task condition there was, in addition to the target stimulus, a briefly (i.e., 29 ms) presented distractor stimulus which had to be identified. Depending on the distractor load, the distractor stimulus contained 1, 2, or 3 digits, which were simultaneously presented and always different from each other. The experimenter wrote down the participant’s verbal response, which was later entered in the computer and analyzed for correctness. The digits were randomly drawn from the set 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 on condition that each number appeared 5 times in a set of 50 test trials.

Participants completed two sessions on separate days. In each session, participants received a series of 50 experimental trials (1 for each of the 50 possible target locations) in each of the eight stimulus durations under either single-task or dual-task condition. Order of task condition between days was counterbalanced. Order of target duration within a day was random. Each series of 50 test trials was preceded by 10 practice trials. In order to asses the effect of retinal eccentricity on localization performance, we established 5 global distances between fixation point and target location by subdividing the 25 possible target locations on each side of fixation into 5 groups of each 5 adjacent target locations. These five distance groups (1, 2, 3, 4, 5) had increasing distances from fixation, namely 1.72, 3.15, 4.58, 6.02, and 7.45° of visual angle, respectively (these values represent the center position within each distance group). The target stimulus, distractor stimulus, and the fixation sign each subtended a visual angle of about 0.23 × 0.29°. The distance between two adjacent target positions was 0.29° of visual angle. Viewing distance was about 60 cm.

Localization performance was quantified by calculating for each participant mean localization error, defined as the (horizontal) absolute distance between the target location and the response location, as a function of target duration and target distance. Figure 2 shows mean localization error for all 29 participants in the single-task condition (localization only) as a function of target duration (i.e., target-mask onset delay) and target distance. A two-way (target duration × target distance) repeated-measures analysis of variance (ANOVA) indicated that localization performance increased (i.e., localization error decreased) significantly as target duration increased and target distance decreased (F(7, 196) = 96.59, p < 0.001 and F(4, 112) = 50.94, p < 0.001, respectively). In Fig. 2c, it can be seen that with short durations, nearby targets were better localized than distant targets, but that this effect of distance diminishes, and eventually disappears, with longer target durations. This was supported by a significant interaction between target distance and target duration (F(28, 784) = 6.10, p < 0.001).

In previous work, we found that the localization performance function exhibited a steep rise in accuracy (or decrease in error) for durations less than 60 ms (Adam et al., 1993), whereas in this study localization performance in this range is constant (i.e., performance at 29 ms is similar to that at 57 ms). The reason for this is unclear but it could be related to the fact that in the present study the targets appeared on a horizontal axis, whereas in the previous work the targets appeared in a two-dimensional grid that moreover contained many more possible target locations, that is, 474 vs. 50 possible positions. Thus, positional uncertainty was much greater in our previous than in the present work, and, moreover, distributed across two dimensions. These procedural differences allow the possibility that the spatial and temporal dynamics of attention allocation in the two paradigms may have differed, with faster allocation in the present, less complex paradigm. Consistent with this idea, Tse, Sheinberg and Logothetis (2003) reported poorer change detection performance along the vertical than the horizontal axis, a finding they attributed to the relatively poor spatial resolution of attention along the vertical dimension (e.g., He, Cavanagh, & Intriligator, 1996). Furthermore, Pellizzer and Hedges (2003, 2004) reported that reaction time of reaching responses towards visual targets increased with positional uncertainty. This finding conformed to the predictions of a capacity-sharing model, which assumes that the processing resources for motor localization are limited, and that they are distributed as a function of the spatial distribution of the possible target locations. Thus, when the target appears, the processing resources must be reallocated to this new location. This reallocation is performed through an adjustment of the dispersion and location of the processing resources, and this affects reaction time. By these arguments, the allocation of attention to the target stimulus was easier to implement in the present uni-dimensional task, which contained relatively few possible target positions, than in the previous two-dimensional task, which contained many more possible target positions.

For target durations longer than 100 ms, localization error systematically decreased, with near-maximal performance achieved at target durations between 300–400 ms. This outcome is similar to the findings obtained in our previous studies, and most likely can be attributed to the eye movement system that with longer target durations is increasingly able to execute a saccade (including secondary correction saccades) toward the target location while it is still visible. In this view, the strong improvement in localization performance between 100 and 300 ms is the result of progressively more saccades being initiated and executed before the target disappears. Note, that this gradual improvement in localization performance between 100 and 300 ms corresponds nicely with response latencies of saccades which typically are in the order of 200 ms with a distribution that ranges between 100 and 300 ms (e.g., Carpenter, 2004).

The main effect of target distance reflected larger localization errors with increasing distance from initial fixation, except for the largest distance, which showed a slight improvement relative to the next largest distance (Fig. 2b). The former finding is caused by the fact that the visual acuity of the retina (i.e., its spatial resolution) deteriorates with increasing retinal eccentricity. The latter finding is probably related to the fact that the end of the masking stimulus acted as some kind of reference point. That is, sometimes localization performance appears to depend more on the distance between the target stimulus and a reference point than upon eccentricity (e.g., White, Levi, & Aitsebaomo, 1992).

In the dual-task condition, where the localization task had to be performed together with the identification task, the performance function showed an initial decrease in error, reaching a stable level of performance after which it decreased further (see Fig. 3). Figure 3a shows mean localization error in single- and dual-task conditions for the 1-distractor group (averaged over target distance). As can be seen in Fig. 3a, localization error was substantially greater in the dual-task (1-distractor) condition than in the single-task (no-distractor) condition (F(1, 8) = 14.59, p < 0.001). This finding shows that the requirement to first identify the central distractor stimulus hampered localization performance and suggests that localization performance depends on the availability of selective visual attention. Importantly, this effect was qualified by a significant interaction with the factor target-mask onset delay (F(7, 56) = 7.46, p < 0.01). This interaction indicated that at the shortest target duration of 29 ms there was a robust (p < 0.001) interference effect that disappeared with the longer stimulus durations of 57 and 86 ms. This important outcome suggests that at the shortest target duration of 29 ms attention was not sufficiently available for localizing the target, as it was allocated to the task of identifying the distractor stimulus. Presumably, this raised localization error at the unattended target location. With the longer target durations of 57 and 86 ms, however, the interference effect disappeared, suggesting that attentional identification of the distractor stimulus had been completed and that visual attention had become increasingly available for, and shifted to, the peripheral target stimulus. Note that this early interference effect and its fast disappearance can not be attributed to the execution of eye movements because these effects occurred within 60 ms of presentation time, a time range far too short to execute saccades.

Interestingly, with still longer target durations, an interference effect appeared again until it disappeared at the longest target duration of 400 ms. This outcome seems to suggest that the identification task interfered not only with the attention system (at short target durations) but also with the eye movement system (at longer target durations). Probably, if attention is delayed in moving to the target stimulus, eye movements are delayed too because shifts of attention are thought to precede (and to be functionally related to) saccadic eye movements (e.g., Godijn & Theeuwes, 2003; Irwin & Gordon, 1998; Rizzolatti, Riggio, Dascola, & Umiltà, 1987). In sum, it appears that the requirement to first identify a single distractor item caused a temporal shift of the full target duration—localization performance function relative to the control, no-distractor function.¹ Importantly, the interference effect present at the shortest target durations reflects the absence of attention for target localization because it can not be allocated to the peripheral target stimulus while it is processing the central distractor stimulus.

To estimate the extent of this temporal shift and test the assertion that the performance function was shifted horizontally, we performed a fitting procedure that calculated the time shift (τ) necessary to produce an optimal fit between the no-distractor and 1-distractor performance functions. First, both functions were non-linearly interpolated (via a piecewise cubic Hermite interpolation procedure). Then, the optimal time shift τ was calculated by the following minimization procedure: (g = 1-distractor function; f = no-distractor function). This procedure yielded a time shift τ of 61 ms, which resulted, after fitting, in a correlation coefficient of 0.99 between the two functions (Fig. 3b).

Localization error as a function of target-mask onset delay for the 2-distractor and 3-distractor groups is presented in Fig. 3c and e, respectively. Separate two-way ANOVAs for the 2- and 3-distractor groups indicated here too large distractor interference effects (F(1, 9) = 17.24, p < 0.01 and F(1, 9) = 43.22, p < 0.001, respectively) that varied as a function of target-mask onset delay (F(7, 63) = 2.25, p < 0.05 and F(7, 63) = 2.73, p < 0.05, respectively). In particular, increasing the number of distractor items shifted the target duration―localization performance function progressively further away from the control, no-distractor function, revealing longer lasting interference effects. In terms of the fitting procedure described above, time shifts τ of 145 and 193 ms were obtained for the 2- and 3-distractor groups, respectively (with correlation coefficients of 0.95 and 0.87, respectively; Fig. 3d,f). Of course, the restricted range and levels of target-mask onset delays used in the present study limits the reliability and accuracy of the fitting procedure, especially in the 2- and 3-distractor conditions.

Figure 4a depicts the mean overall interference effect (i.e., subtracting localization error in the single-task condition from that in the dual-task condition) for the 1-, 2-, and 3-distractor groups. Figure 4b depicts the time shift τ for the 1-, 2-, and 3-element distractor groups. Clearly, increasing the number of to-be-identified distractor items in the distractor stimulus caused increasingly larger and longer lasting localization interference effects. In fact, a linear regression analysis on the individual time-shift data revealed that the attentional system takes about 66 ms to identify each distractor item (F(1, 22) = 26.03, p < 0.001) without a general shift associated with dual-tasking (non-significant intercept = 2.5 ms).

To examine the presence of a systematic bias in localization error we calculated the constant error (CE), which retains the sign or direction of the errors (undershoots or overshoots), as a function of task (single, dual), target-mask-onset delay (eight levels), target distance (five levels), and distractor load (three levels). On average, participants tended to undershoot the target by -2.8 mm, which is consistent with the general finding that localization judgments typically undershoot briefly presented targets (e.g., Müsseler, van der Heijden, Mahmud, Deubel, & Ertsey, 1999). An ANOVA indicated larger undershoots in the dual-task than in the single-task condition (−3.3 and −2.3 mm, respectively; F(1, 26) = 16.59, p < 0.001), larger undershoots with shorter target-mask-onset delays (−0.5, −1.3, −2.4, −3.1, −2.9, −3.4, −4.0, −4.8 mm, for shorter delays, respectively; F(7, 182) = 23.87, p < 0.001), and an U-shaped function relating undershoot to target distance (−1.1, −2.8, −3.7, −3.6, and −2.8 mm, for increasing distances, respectively; F(4, 104) = 13.63, p < 0.001). This latter finding is probably related to the fact that the most peripheral targets fell close to the end of the masking stimulus, which may have acted as a reference point (e.g., White et al., 1992). The above main effects were qualified by a significant 3-way interaction involving all three factors, F(28, 728) = 6.84, p < 0.001. This interaction is depicted in Fig. 5 and indicates that undershoots were disproportionally greater in the dual-task condition than in the single-task condition when targets were presented shorter than 100 ms and at greater distances from fixation. This finding supports the idea that the distractor(s) strongly interfered with the operation of the attentional system, even so much that it eliminated the advantage of the most-distance targets (falling near the end of the masking stimulus). Interestingly, the size of the undershoot effect in the dual-task condition for the shortest target durations (distances 2, 3, 4, and 5) was about 10%, which is very similar to previous estimates of mislocalization (e.g., Van der Heijden, Van der Geest, De Leeuw, Krikke, & Müsseler, 1999).

Although we interpret the current findings within the two-process model as providing supporting evidence for the role of the attentional system in spatial localization, two alternative explanations for the data were addressed in control experiments. The first alternative relates to the confounding of the presence of the distractor(s) and the task requirement (single vs. dual task). We tested the hypothesis that the mere presence of the distractor lead to the change in the performance function. Twelve participants performed the localization task in a condition with no distractor and in a condition with one distractor that did not require identification (but was merely present). Results showed the usual improvement in localization performance with longer target durations, F(7, 77) = 50.12, p < 0.001, and, moreover, that localization performance did not differ between these two conditions (mean localization error: 3.9 and 4.1 mm, respectively; F(1, 11) < 1, p > 0.4). More critically, this was also true for target durations shorter than 100 ms (mean localization error: 6.2 and 6.4 mm, respectively; F(1, 11) < 1, p > 0.6). These findings indicate that the distractor effect is not due passive capture of attention by a foveal onset, but to active allocation of attention for identification of the distractor stimulus.

The second alternative relates to the delay in localization due to uttering the digits, during which the location information could have decayed from visual working memory. We tested this “delayed response hypothesis” with another group of twelve participants performing the localization task before or after identifying three digits. Again, there was the improvement in target localization with longer target durations, F(7, 77) = 14.59, p < 0.001. An analysis of the localization onset times in these two conditions revealed, as expected, a substantially delayed localization onset response in the locate-after condition than in the locate-before condition (mean onset times: 1,770 vs. 605 ms, respectively; F(1, 11) = 88.54, p < 0.001). However, localization performance did not differ between these two response conditions (mean localization error: 5.7 and 5.0 mm, respectively; F(1, 11) = 2.97, p > 0.1).

Together, the results of these control experiments argue against the view that the interference effect found in our main experiment is due to either the mere presence of the distractor or delayed responding in the distractor condition.

Finally, it is interesting to note that the way in which attention is shifted toward the target stimulus may be different for the single- and dual-task conditions. In the single-task condition, the sudden, peripheral onset of the target stimulus may have prompted an exogenous (i.e., involuntarily, automatic) shift of attention, whereas in the dual-task condition the requirement to first identify the central distractor stimulus, which appears simultaneously with the peripheral target stimulus, most likely requires an endogenous (i.e., voluntary) shift of attention (e.g., Posner, 1980). However, this possible difference in attentional control for single- and dual-task conditions can not explain the effect of distractor load, showing greater and longer lasting interference effects at short target durations with increases in the number of distractor items.² Hence, our conclusion still is that the diminished availability of attention for target localization at short target durations is responsible for the observed interference effect.

The two-process model of localization performance emphasizes responses to the stimulus (attention shifts and eye movements) rather than the initial intake or coding of stimulus information by visual cells. This emphasis on attention and eye movements does not negate the fact that there is the initial period prior to the attention movement. Indeed, shifting attention to the target presupposes at least some knowledge of the target location prior to the attention shift.

In line with this observation many theorists (e.g., Logan, 1992; Treisman, 1985) posit a preattentive level of visual input analysis in which the visual scene is coded in parallel along a number of separable dimensions or features such as color, line orientation, motion, but also position. In doing so, initial registration of sensory information by the retina already achieves some form of location coding because retinal visual cells are uniquely sensitive to information from different directions. However, although some degree of spatial localization might be afforded by preattentive analysis, the two-process model and the present findings suggest that subsequent processes (i.e., shifts of attention and eye movements) are needed to precisely localize items.

Although the current experiment was designed to further establish the involvement of the attentional system in object localization, it is interesting to speculate briefly on the implications of the present results for the functional connection between the attentional and saccadic systems. That is, the systematic horizontal shift of 66 ms of the localization performance function for each to-be-identified central distractor item is consistent with the notion of an obligatory or functional link between the attentional and saccadic systems, in that attentional selection of the target stimulus seems obligatory before a saccade toward it can be executed. In this view, the triggering of the saccade in the dual-task condition is delayed by the time it takes the attentional system to identify the distractor stimulus. Our data suggest a time delay of about 66 ms for each distractor item, an estimate that is in close agreement with the results of Kowler, Anderson, Dosher and Blaser (1995, Experiment 2) who reported an increase in saccadic latency by 50–75 ms when subjects, in addition to preparing a saccade, were also required to identify a single letter. Thus, our findings are consistent with a growing body of behavioral (e.g., Deubel & Schneider, 1996; Hoffman & Subramanian, 1995; Kowler et al., 1995), neurophysiological (e.g., Desimone, Wessinger, Thomas, & Schneider, 1989; Kustov & Robinson, 1996; Wardak, Ibos, Duhamel, & Olivier, 2006), and computational (e.g., Clark, 1999; Koch & Ullman, 1985) evidence that supports a strict and functional coupling between spatial attention and saccadic eye movements. This view is most prominent in the premotor theory of attention (e.g., Rizzolatti et al., 1987; Umiltà, Riggio, Dascola, & Rizzolatti, 1991), which has been proposed 20 years ago, postulating a strong, direct coupling between spatial shifts of attention and the preparation of saccadic eye movements. According to the premotor theory, the allocation of attention to a location is intrinsically linked with the preparation to make a saccadic eye movement to that location. In addition, recent studies have demonstrated a close relationship between attentional shifts and the direction of micro-saccades (e.g., Engbert & Kliegl, 2003). However, direct evidence for the interdependency between attention and eye movements in the current localization paradigm awaits future studies that use eye movement recordings.