Edited by: Frans Verstraten, The University of Sydney, Australia
Reviewed by: Adrian Von Muhlenen, University of Warwick, UK; Todd Horowitz, Harvard University, USA
*Correspondence: Gregory J. Zelinsky, Department of Psychology, Stony Brook University, Psych B Building, Room 240, SUNY Stony Brook, Stony Brook, NY 11794-2500, USA. e-mail:
This article was submitted to Frontiers in Perception Science, a specialty of Frontiers in Psychology.
This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.
Two experiments are reported that further explore the processes underlying dynamic search. In Experiment 1, observers’ oculomotor behavior was monitored while they searched for a randomly oriented T among oriented L distractors under static and dynamic viewing conditions. Despite similar search slopes, eye movements were less frequent and more spatially constrained under dynamic viewing relative to static, with misses also increasing more with target eccentricity in the dynamic condition. These patterns suggest that dynamic search involves a form of sit-and-wait strategy in which search is restricted to a small group of items surrounding fixation. To evaluate this interpretation, we developed a computational model of a sit-and-wait process hypothesized to underlie dynamic search. In Experiment 2 we tested this model by varying fixation position in the display and found that display positions optimized for a sit-and-wait strategy resulted in higher
We often search for things in static displays, situations in which the elements of the scene through which we are searching remain in the same locations over time. The assumption of static search has even pervaded search theory, where it is commonly believed that there is a memory for distractors that have been inspected and rejected, and that this memory is used to improve search efficiency (Treisman and Gelade,
The hypothesis that distractor memory is used during search was challenged most directly by Horowitz and Wolfe (
This bold claim was itself challenged almost immediately by Kristjansson (
The experiments conducted by Horowitz and Wolfe (
One prominent explanation for the Horowitz and Wolfe (
Following von Mühlenen et al. (
The current experiments further explore the hypothesis that a sit-and-wait strategy underlies the search of dynamic displays, focusing on the relationship between eye position and attentional allocation during this search task. In Experiment 1, we examined the role of target eccentricity in dynamic search by using the Geyer et al. (
Our methodology differs from Geyer et al. (
Horowitz and Wolfe (
Considerable work has documented the movement of gaze in static search tasks (see Rayner,
Our premise is that the enlistment of different search processes in these two display conditions might be evidenced by different patterns of eye movements in individual observers. One pattern might involve the active repositioning of gaze during search, presumably accompanied by the active movement of attention to different display positions. Indicating such an active search strategy would be an increase in the number of fixations with both set size and target eccentricity, but little or no change in fixation duration or accuracy. A contrasting pattern, one indicating a sit-and-wait search strategy, would show fewer fixations that were spatially constrained to the sit-and-wait location regardless of set size and target eccentricity. This is the general pattern of oculomotor data reported by Geyer et al. (
One of the authors and three experimentally naïve observers from Stony Brook University’s undergraduate and graduate communities participated in the experiment. The naïve observers were paid $10/h for their participation and all observers had either normal or corrected-to-normal vision.
The target was a T, the distractors were Ls, and all items could appear rotated 0°, 90°, 180°, or 270°. Objects were located on three concentric circles whose radii subtended 1.5°, 3.1°, and 4.6°, and a fourth broken circle banding the left and right sides of the display whose radius subtended 6.2°. Item locations at each eccentricity were equally spaced around the respective imaginary circles; across eccentricities, locations were not aligned. Each eccentricity contained 4, 10, 15, and 10 object locations, respectively. Distractor locations were randomly selected from among these 39 display locations; the target appeared only at the second (3.1°) and fourth (6.2°) eccentricities, appearing at each on 50% of the target-present trials. Observers were therefore not biased by target contingencies into shifting gaze to a particular display eccentricity. Individual stimuli subtended 0.75° × 0.75° and were composed of lines 0.19° in width. The minimum center-to-center distance between objects was 1.55°, and all objects were white, presented on a black background. Mask items replaced each search item after a time-terminated display interval. Individual masks consisted of the superimposed target and distractor line segments (i.e., a “+” sign in a square) and subtended 0.75° × 0.75°. Figure
Participants were instructed to search for the letter T, regardless of its orientation, and to respond as quickly and accurately as possible. A target-present response was indicated by left-clicking a two-button mouse; target absence was indicated by a right-click.
Each trial began with the presentation of a centrally located fixation cross. Observers initiated a trial by pressing a mouse button, which removed the cross and caused the search display to appear. Static trials consisted of a single search display depicting a single configuration of items (item configurations varied from trial to trial). Dynamic trials consisted of 15 sequentially presented displays (configurations varied both within and between trials)
Observers participated in 640 trials over two 1-h sessions, conducted on separate days. These 640 trials were evenly divided into 2 set sizes (9 or 17 items), 2 target conditions (present or absent), 2 search conditions (static or dynamic), and 2 target eccentricities (3.1° or 6.2°), leaving 40 trials per target-present cell and 80 trials per target-absent cell.
Prior to examining the data, approximately 4% of the trials were excluded because of manual RTs falling above or below two SDs from the cell mean. Figure
Experiment 1 |
||||
---|---|---|---|---|
Condition | Static |
Dynamic |
||
Set size | 9 | 17 | 9 | 17 |
RT (ms) | 1257 | 1721 | 1867 | 1911 |
Errors (%) | 0.6 | 3.1 | 23.1 | 20.0 |
Set Size | 9 | 17 | 9 | 17 |
RT (ms) | 1623 | 1678 | 1941 | 1961 |
Errors (%) | 18.1 | 25.3 | 11.1 | 9.5 |
Target-present |
Target-absent |
|||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
Condition | Static |
Dynamic |
Static |
Dynamic |
||||||||
Set size | 9 |
17 |
9 |
17 |
9 | 17 | 9 | 17 | ||||
Eccentricity | Inner | Outer | Inner | Outer | Inner | Outer | Inner | Outer | ||||
RT (ms) | ||||||||||||
G.Z. | 959 | 1006 | 1056 | 1252 | 1138 | 1331 | 1193 | 1546 | 1450 | 1799 | 1831 | 1842 |
S.L. | 1009 | 1210 | 1244 | 1552 | 1883 | 2383 | 2213 | 2383 | 1466 | 2063 | 2644 | 2731 |
R.C. | 712 | 798 | 820 | 1110 | 975 | 1141 | 1173 | 1321 | 1034 | 1454 | 1594 | 1596 |
S.M. | 566 | 827 | 700 | 1036 | 650 | 1274 | 716 | 1356 | 1077 | 1570 | 1398 | 1474 |
Mean | 812 | 960 | 955 | 1238 | 1162 | 1532 | 1324 | 1652 | 1257 | 1722 | 1867 | 1911 |
Errors (%) | ||||||||||||
G.Z. | 2.5 | 12.5 | 12.5 | 12.5 | 10.0 | 22.5 | 20.0 | 47.5 | 0 | 2.5 | 26.3 | 17.5 |
S.L. | 2.5 | 10.0 | 5.0 | 5.0 | 2.5 | 27.5 | 17.5 | 42.5 | 0 | 0 | 22.5 | 18.8 |
R.C. | 2.5 | 2.5 | 5.0 | 0 | 0 | 5.0 | 2.5 | 15.0 | 0 | 0 | 20.0 | 27.5 |
S.M. | 2.5 | 7.5 | 2.5 | 12.5 | 0 | 27.5 | 7.5 | 52.5 | 2.5 | 10.0 | 23.8 | 16.3 |
Mean | 2.5 | 8.1 | 6.3 | 7.5 | 3.1 | 20.6 | 11.9 | 39.4 | 0.6 | 3.1 | 23.2 | 20.0 |
Turning first to the static condition, the data reveal a fairly unremarkable pattern of results. Target-present search averaged 26 ms/item, nearly half the 57 ms/item rate found in the target-absent data. Consistent with the literature, search efficiency in the static condition also interacted with target eccentricity. Search proceeded at a rate of 18 ms/item for inner-eccentricity targets, and 35 ms/item for outer-eccentricity targets. The error data accompanying these RTs were equally unremarkable. Misses occurred on less than 10% of the trials and did not vary appreciably with set size or target eccentricity. False alarms were rare, occurring on less than 4% of the trials.
Data from the dynamic search condition were less straightforward and provided mixed support for the patterns reported in Horowitz and Wolfe (
If observers engaged in a dynamic search task elected to keep their gaze at the display’s center and covertly process the items surrounding fixation, as predicted by a multi-location sit-and-wait search strategy, then outer-eccentricity targets might never be inspected regardless of the number of dynamic frames. The consequence of such a strategy would be a higher miss rate in the outer-eccentricity dynamic condition – the exact pattern observed in the data. Likewise, if observers in the static display condition adopted a more active search strategy, then even outer-eccentricity targets would eventually be inspected. The consequence of such an active search strategy, assuming that covert search originates from the center of gaze and proceeds outward (Wolfe et al.,
Although the selective use of different strategies to search static and dynamic displays would seem consistent with the current data, one might argue that a sit-and-wait strategy does not well describe data from the Horowitz and Wolfe (
Can the current evidence in support of a sit-and-wait strategy be reconciled with the relatively low miss rates reported by Horowitz and Wolfe (
To better assess the possibility that static and dynamic search differences reflect different underlying search processes, we analyzed the number of fixations and the fixation-duration data from individual observers. We expect that a sit-and-wait strategy will be characterized by few, if any, changes in gaze location, and that increases in set size and target eccentricity will instead result in longer fixation durations. The hypothesized effect of set size follows from the assumption that a sit-and-wait strategy involves the monitoring and accumulation of information over a limited display region surrounding current fixation, and that more items will fall within this region as set size increases. The hypothesized effect of eccentricity follows from the assumption that items nearer fixation are processed preferentially relative to more eccentric items, similar to the central attentional bias concept advanced by Wolfe et al. (
Table
Target-present |
Target-absent |
|||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
Condition | Static |
Dynamic |
Static |
Dynamic |
||||||||
Set size | 9 |
17 |
9 |
17 |
9 | 17 | 9 | 17 | ||||
Eccentricity | Inner | Outer | Inner | Outer | Inner | Outer | Inner | Outer | ||||
G.Z. | 4.97 | 5.43 | 5.46 | 6.43 | 1.81 | 1.58 | 1.69 | 1.71 | 8.30 | 9.55 | 1.70 | 1.82 |
S.L. | 4.50 | 6.07 | 5.68 | 7.50 | 1.97 | 2.86 | 3.03 | 3.30 | 7.40 | 9.38 | 3.38 | 3.95 |
R.C. | 3.44 | 4.19 | 3.73 | 5.21 | 2.60 | 3.00 | 3.00 | 3.29 | 5.33 | 7.51 | 3.95 | 3.84 |
S.M. | 1.26 | 1.62 | 1.92 | 2.32 | 1.28 | 2.34 | 1.69 | 3.47 | 1.96 | 3.08 | 2.69 | 2.98 |
Mean | 3.54 | 4.33 | 4.20 | 5.37 | 1.91 | 2.44 | 2.35 | 2.94 | 5.75 | 7.38 | 2.93 | 3.15 |
Target-present |
Target-absent |
|||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
Condition | Static |
Dynamic |
Static |
Dynamic |
||||||||
Set size | 9 |
17 |
9 |
17 |
9 | 17 | 9 | 17 | ||||
Eccentricity | Inner | Outer | Inner | Outer | Inner | Outer | Inner | Outer | ||||
G.Z. | ||||||||||||
Single | 1098 | 1356 | 1230 | 1330 | 1940 | 1816 | ||||||
First | 179 | 183 | 163 | 173 | 178 | 165 | 136 | 203 | 165 | 168 | 214 | 253 |
Second | 154 | 139 | 152 | 154 | 928 | 1058 | 924 | 1171 | 122 | 147 | 1350 | 1223 |
Final | 214 | 236 | 266 | 200 | 1033 | 1165 | 1108 | 1301 | 148 | 149 | 1480 | 1535 |
S.L. | ||||||||||||
Single | 1892 | 2333 | 2173 | 2269 | 2468 | 2554 | ||||||
First | 215 | 211 | 233 | 253 | 398 | 704 | 699 | 669 | 214 | 265 | 1088 | 903 |
Second | 175 | 143 | 133 | 180 | 752 | 529 | 561 | 1052 | 166 | 172 | 524 | 487 |
Final | 313 | 304 | 316 | 289 | 1100 | 703 | 652 | 905 | 176 | 231 | 569 | 547 |
R.C. | ||||||||||||
Single | 970 | 1274 | 758 | 1797 | 1449 | |||||||
First | 206 | 205 | 221 | 210 | 313 | 366 | 367 | 313 | 214 | 227 | 320 | 323 |
Second | 156 | 147 | 155 | 147 | 445 | 367 | 415 | 418 | 170 | 144 | 533 | 476 |
Final | 258 | 240 | 293 | 264 | 397 | 383 | 479 | 484 | 177 | 192 | 562 | 539 |
S.M. | ||||||||||||
Single | 610 | 763 | 625 | 919 | 689 | 992 | 690 | 1003 | 1001 | 1340 | 1212 | 1310 |
First | 454 | 574 | 479 | 645 | 515 | 692 | 439 | 512 | 606 | 647 | 623 | 611 |
Second | 145 | 335 | 226 | 168 | 317 | 409 | 269 | 210 | 351 | 404 | 423 | 429 |
Final | 211 | 308 | 235 | 328 | 335 | 467 | 331 | 285 | 391 | 504 | 579 | 542 |
Comparing the static and dynamic fixation data for observers G.Z., S.L., and R.C. reveals two clear patterns. First, there were fewer fixations in the dynamic (
The oculomotor data patterns from observers G.Z., S.L., and R.C. were most consistent in the static search condition. For each observer, the number of fixations accompanying static search (but not fixation duration) increased as a function of both set size and target eccentricity. In fact, fixation duration remained relatively constant across eye movements (187 ms), with first fixations being only slightly longer than the average duration of all the other fixations (204 vs. 177 ms). Consistent with our suggestion of an active acquisition strategy underlying static search, these observers therefore responded to the increased processing demands associated with the set size and eccentricity manipulations by making more eye movements, not by increasing the duration of individual fixations.
While remaining relatively consistent amongst themselves, observers G.Z., S.L., and R.C. exhibited very different oculomotor patterns when searching dynamic displays. In contrast to their static search behavior, the number of fixations in the dynamic condition did not vary appreciably with either set size or target eccentricity. Rather, the increased processing load expected from these manipulations was expressed as longer fixation durations. Consistent with the use of a sit-and-wait strategy, their average fixation duration in the dynamic condition was 677 ms, 490 ms longer than their average fixation duration in the static condition. Unfortunately, further specifying the variation in fixation duration is complicated by the fact that multiple eye movements were often made even in the dynamic search condition. Fixations in a search trial likely serve different functions, so it is important not to treat all of these dynamic fixations as being equivalent. Some are probably no more than brief stopover points as an observer settles on a display location before actually conducting the search. For these fixations one would expect little or no effect of the independent variables on duration. Of course other dynamic fixations will more directly reflect search processing, and for these fixations one would expect an effect of the search manipulations. To obtain a clearer picture of any strategy engaged during dynamic search, we therefore attempt a brief fixation-by-fixation description of the oculomotor behavior from individual observers.
Turning first to G.Z.’s dynamic condition data, we see no systematic influence of either set size or target eccentricity on the first-fixation durations. Note from Table
Like observer G.Z., S.L. often searched the dynamic displays without making an eye movement. Again consistent with a sit-and-wait strategy, when this observer did not shift gaze, clear effects of set size and target eccentricity emerged in her single-fixation durations. Also like G.Z., this observer made leftward gaze shifts on those trials in which she did move her eyes. These eye movements, however, were more pronounced than G.Z.’s and often resulted in gaze being repositioned quite far from the display’s center. This more extensive use of eye movements complicates an interpretation of the resulting fixation durations. When the display contained only nine items, S.L. used her initial fixation to search for the dynamic target, as indicated by the large effect of eccentricity in the nine-item trials (306 ms). However, when the display was more cluttered (i.e., the 17-item trials), S.L. shifted the bulk of her search processing to the second fixations, resulting again in a sizeable eccentricity effect (491 ms). Regardless of the fixation during which S.L. chose to conduct her search, clear differences again emerged between the static and dynamic conditions. S.L. freely used eye movements to search the static displays, with the durations of these fixations being brief and unrelated to the search manipulations. Fixations were far less frequent in S.L.’s dynamic search data, with the duration of at least one of these fixations showing evidence for an eccentricity or set size effect. For these two observers, we see the very clear emergence of two distinct patterns in the eye data, one indicating the use of a sit-and-wait strategy to search dynamic displays and the other indicating the use of an active acquisition strategy to search static displays.
Observer R.C.’s search of the dynamic displays deviated from the signature sit-and-wait patterns found for G.Z. and S.L. in two respects. First, he moved his gaze more frequently in the dynamic condition, making an average of 2 saccades compared to the 1.2 saccade average from G.Z. and S.L. Second, his fixation durations were shorter than those of the other observers, and these durations showed no clear effects of the search manipulations. Only in the final fixation durations of multi-saccade trials was there a suggestion of a set size effect. These patterns do not mean, however, that R.C. was applying the same search strategy to both the static and dynamic tasks. As can be seen from his scatterplot data, eye movements in the dynamic search condition were still far less frequent and more spatially constrained relative to those in the static condition. Also notice the pronounced tendency to make directly leftward eye movements relative to the display’s midpoint, similar to the pattern observed in S.L.’s eye data. These leftward eye movements, rather than being an attempt to fixate individual display items, appear instead to be a strategic attempt to position gaze at a region in the display deemed favorable to dynamic search. Upon closer analysis of these fixation locations, we found that R.C. was far more likely than the other observers to position gaze halfway between the two target eccentricities. This strategic allocation of gaze suggests that R.C. may have learned, either implicitly or explicitly, where targets were most likely to appear in the display. It might also explain why R.C., unlike G.Z. and S.L., showed no eccentricity effect in his dynamic condition fixation durations or RT × Set Size slopes (25 ms/item for the inner-eccentricity targets; 23 ms/item for the outer-eccentricity targets), as well as the fact that this observer made the fewest errors in the dynamic task. We return to this point in the next section.
Our final observer, S.M., showed no obvious differences in oculomotor behavior between the static and dynamic search tasks. The number of fixations made in the two search conditions did not meaningfully differ, nor were there compelling differences in the spatial dispersal of these fixations. In both tasks, S.M. made relatively few saccades that failed to carry gaze far from the display’s center – patterns that we have been interpreting as evidence for a sit-and-wait search strategy. The reason for this contradictory pattern was made clear during debriefing when S.M. volunteered the fact that he had consciously attempted to perform both the static and dynamic tasks without moving gaze. Returning to Figure
This self-imposed “don’t move your eyes” instruction provides a unique opportunity to compare static and dynamic search behavior in the near absence of eye movement. It is important, however, not to confuse a “don’t move your eyes” search strategy with a sit-and-wait search strategy. Although a sit-and-wait strategy does predict fewer fixations during search, recall that its defining characteristic is a restriction of processing to a region of the search display surrounding fixation. When both of these criteria are considered, we find that S.M.’s dynamic search performance is well described by such a strategy. Like G.Z. and S.L., the other observers who were consistently sitting and waiting for the dynamic target, S.M. was able to successfully search the dynamic displays only when targets appeared at the inner-eccentricity. Indeed, S.M.’s accuracy in the 17-item outer-eccentricity condition was at chance. We interpret this pattern as suggesting that outer-eccentricity targets often went undetected in the dynamic condition because they fell outside the display region monitored as part of the sit-and-wait strategy. If S.M. was using a sit-and-wait strategy to search the static displays, we would therefore expect a similar pattern of misses. However, as indicated in Table
To summarize, the eye data from Experiment 1 revealed distinct oculomotor patterns in the static and dynamic search conditions. With the exception of S.M., observers relied on eye movements to search static displays, with the number and dispersion of these fixations varying as a function of set size and target eccentricity. The number of fixations increased with display set size, and fixations were fewer and more spatially constrained when targets appeared at the inner-eccentricity relative to the outer-eccentricity. Fixation durations also remained fairly constant across these manipulations, suggesting the inspection of a fixed number of items during each fixation. All of these patterns are consistent with what we are calling an active acquisition search strategy. Observers displayed a very different search pattern when searching dynamic displays. Eye movements were relatively infrequent and, when they did occur, failed to move gaze far from the display’s center. The number of fixations during dynamic search also failed to show a clear effect of set size and target eccentricity. Fixation durations, however, were in general much longer than their static condition counterparts and increased as a function of these two manipulations, albeit idiosyncratically both within and across observers. These patterns are consistent with what we are calling a multi-location sit-and-wait search strategy.
Given the existence of these two search strategies, the question remains as to
The general patterns of oculomotor behavior observed in Experiment 1 also correspond to those reported by Geyer et al. (
From Experiment 1 we know that observers in the dynamic search condition often constrained their fixations to the display’s center. We also discussed the consequences of this behavior for search and some of the reasons why observers may have adopted such a sit-and-wait search strategy. What remains to be discussed is an explanation of how this strategy relates to gaze position and the deployment of attention in a dynamic search task, and it is to these questions that we now turn.
We propose a simple relationship between gaze position and the deployment of attention in dynamic displays. Following Wolfe et al. (
The above sketch of our sit-and-wait model suggests that accuracy of target detection in the dynamic search condition should be limited by two interacting factors. First, search accuracy should improve with the number of locations that can be monitored during a dynamic trial (sample size). Second, dynamic search performance will depend on the distance between the target’s location in the search display and the observer’s locus of gaze. Given a central attentional bias (Wolfe et al.,
These sample size and eccentricity constraints on dynamic search can be formally modeled within the context of a multi-location sit-and-wait strategy. The success of a sit-and-wait strategy depends on the probability of the target appearing among the set of monitored display locations, which in turn depends on the size of this set, the position of gaze in the display, and ultimately on the display contingencies governing target location in the dynamic search task. If one knows these contingencies, it is therefore possible to plot the probabilities of successful sit-and-wait target detection as a function of sample size (von Mühlenen et al.,
From the figure it is clear that the probability of correctly detecting a target in Horowitz and Wolfe’s (
Figure
Segregating the simulation data by eccentricity helps to clarify the Figure
Of course higher detection probabilities would be expected as gaze moves closer to the target rings in their respective eccentricity conditions. For example, in the inner-eccentricity condition, only a single-location need be monitored to achieve a 1.0 detection probability when gaze is positioned on the 3.1° target ring (F2). When gaze is positioned at the first (1.5°) eccentricity and two locations are monitored, the nearest-to-fixation constraint requires that one of these locations be on the inner target ring, again resulting in a 1.0 detection probability (F1). For inner-eccentricity targets, the least successful display locations in which to apply a sit-and-wait search strategy are at the 6.2° target ring (F4) and at the display’s center (F0). Predictably, these probabilities differ in the case of outer-eccentricity targets, with high detection probabilities expected only for fixations on the 6.2° target ring (F4). Fixations on the 4.6° locations (F3) should yield an intermediate level of detection, whereas sitting and waiting at the first two eccentricities and at the display’s center should not produce above-chance target detection for the sample sizes used in our simulations.
Returning to Table
A multi-location sit-and-wait strategy therefore provides a good account of the Experiment 1 data, both in terms of the high dynamic condition miss rates when targets appeared at the outer eccentricity and observers elected to keep gaze near the display’s center, as well as the improved accuracy for our one observer who chose to shift his gaze to more eccentric display locations. Whether this observer had learned that targets were more likely to appear at these eccentric locations and was deliberately shifting gaze, or was simply more prone to oculomotor activity and therefore more likely to position gaze nearer the targets, we cannot say. Nor can we say with certainty whether Horowitz and Wolfe’s (
The previous section showed that a multi-location sit-and-wait strategy, when modulated by gaze position, offers a reasonable description of the misses generated by our Experiment 1 observers performing a dynamic search task. However, this demonstration was
In Experiment 2 we seek to examine more systematically this relationship between gaze position and accuracy in a dynamic search task by requiring observers to maintain fixation either at the display’s center or at a point midway between the two target eccentricities, which should be an optimal location to sit and wait for the target (Figure
Twenty four students recruited from the University of Delaware, who were each paid $10/h and had normal or corrected-to-normal vision, participated in this experiment.
Eye movement and manual data were collected using the EyeLink II video-based eye tracking system (SR Research, Ltd.). Eye position was sampled at a rate of 500 Hz, the system’s spatial resolution was estimated to be 0.2°, and changes in gaze position were available to the computer running the display program within 8 ms. All displays were presented on a 21″ Dell SVGA monitor with a refresh rate of 100 Hz. Search displays were presented at a screen resolution of 800 × 600 pixels. Observers’ head position and viewing distance were fixed with a chinrest, and all responses were made with a GamePad controller attached to the computer’s USB port. Search judgments were made with the left and right index-finger buttons; trials were initiated with a button operated by the right thumb. The stimulus and display characteristics were the same as in the previous experiment. We attached a white rectangular cardboard frame to the monitor’s screen so the search displays would fill the screen to the same degree as in Experiment 1.
The methodology was identical to that of Experiment 1 with the following exceptions. First, there were now two dynamic display conditions and no static search condition. As in Experiment 1, the
Fixation position was a between-subjects manipulation, with the central and eccentric groups each having 12 randomly assigned observers. In the eccentric group, half of the observers were shown a fixation cross on the left side of the screen; the other half were shown a fixation cross in the corresponding location on the right side of the screen. Within each fixation condition, trials were divided evenly between two target conditions (present vs. absent), two set sizes (9 or 17 items), and two target eccentricities (3.1° or 6.2° from the display’s center, which we will refer to as
Trials in which the observer’s gaze shifted more than 1° from the fixation cross (3.3%) were not included in any analyses. Before evaluating either of our hypotheses, we converted the miss and false alarm rates for the two conditions to
Our critical hypothesis, that search should be more accurate with an eccentric fixation position than a central fixation position, was supported by the
Our secondary hypothesis, that eccentricity effects should be attenuated in the eccentric fixation condition relative to the central fixation condition, was also supported by the
Observers who fixated midway between the two target eccentricities were able to respond more accurately to targets at the outer-eccentricity relative to the inner-eccentricity. This reverse eccentricity effect was not the by-product of a speed-accuracy tradeoff as an analysis of RTs revealed the same pattern observed in the
At this time, we can only speculate as to the cause of the observed eccentricity effects in the eccentric fixation condition. One possibility is that targets appearing in the outer display ring were more salient under fixed-gaze conditions compared to targets appearing in the inner-ring, perhaps due to differential crowding effects (inner-ring targets are flanked on both sides by distractors, outer-ring targets are flanked on only one side). This differential crowding, combined with the broad distribution of attention likely resulting from observers searching without moving gaze, might have produced an accuracy advantage for outer-eccentricity targets in the eccentric fixation condition. More work will be needed to substantiate this hypothetical relationship between target eccentricity, gaze position, viewing condition (free vs. fixed), and accuracy in a dynamic search task.
Many of our day-to-day search tasks take place in a dynamically changing environment. Whether we are searching for a fly that has invaded our home, or are looking for a fish in an aquarium or a favorite duck in a pond, we are searching for a target that could be almost anywhere at any given moment. The current study advanced understanding of this neglected variety of search task by first documenting the oculomotor behavior accompanying dynamic search and comparing it to static search performance, then by developing and testing a model of dynamic search.
Initially, the dynamic search paradigm was used by Horowitz and Wolfe (
To better understand the processes enlisted during dynamic search, we conducted an individual observer analysis of the eye movement data from Experiment 1. One strikingly clear finding from this analysis was that there is considerable individual variability in how one might conduct a dynamic search. In even our small sample of four observers, only G.Z. and S.M. adopted a similar dynamic search strategy, one that involved holding gaze near the display’s center. Our third observer, R.C., distributed his fixations far more broadly over the search display, and our fourth observer, S.L., seemed to adopt a hybrid strategy of lingering at the display’s center and then making a large-amplitude saccade relatively late in the dynamic search trial. A second and equally clear finding from this analysis was that although observers can, and do, adopt different dynamic search strategies, not all of these strategies are equally effective. Of our four observers, only R.C. was successful in keeping his error rates low regardless of target eccentricity. Given their wildly varying error rates, and the fact that some observers chose to move their eyes whereas other did not, it is misleading to ignore these individual strategic differences by collapsing performance into an “average” measure of dynamic search behavior, as Geyer et al. (
Although such individual differences complicate any simple theoretical account of dynamic search performance, we believe that much of our data, and the data from Horowitz and Wolfe (
In Experiment 2 we manipulated gaze position in order to test our proposed relationship between eye position and the locus of a sit-and-wait strategy, as well as our suggestion that some sit-and-wait display locations are better than others because of target-presentation contingencies. As predicted by our fixation-modulated sit-and-wait model, we found that positioning gaze midway between two target eccentricities resulted in improved accuracy relative to performance when gaze was positioned at the display’s center. We will continue to explore this effect of gaze position on dynamic search with a future version of our model, one that allows the sit-and-wait locus to be updated during a trial. We plan to use the gaze positions from individual observers to dynamically position a sit-and-wait process as it actually existed on each fixation of every dynamic trial, thereby enabling more realistic fits between our sit-and-wait model and human search behavior under free viewing conditions.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Horowitz and Wolfe’s (
To explore the implications of relaxing this single-location constraint on a sit-and-wait strategy, we simulated 9,120 dynamic trials using the display constraints outlined in Horowitz and Wolfe (
In modeling the use of a multi-location sit-and-wait strategy during dynamic search, we assumed that gaze could be located either at the display center or any of the 40 display locations used by Horowitz and Wolfe (
To derive the target-detection probability functions shown in Figure
Gaze position interacts with sample size to determine the actual display locations monitored by the sit-and-wait search strategy. For example, if five display locations were being monitored, then the generation of the F1 function would use all four display locations at the innermost 2° eccentricity and one location from the 4° eccentricity. However, if we again assume that five locations are monitored but consider an F4 simulated gaze position, one location would be used from the second eccentricity, two locations from the third, and two from the fourth. Target detection probabilities can therefore be calculated by adding the probabilities of a target appearing within these monitored locations for each of the tested gaze positions and sample sizes. When a sample included more than one location at the same eccentricity, the probability of the target appearing in either location given its appearance at that eccentricity [
The procedures used to apply a sit-and-wait search strategy to the current dynamic search data (Figure
This work was supported by grants from the NSF (ITR 0082602), NIMH (R01 MH63748), and ARO (DAAD19-03-1-0039) to Gregory J. Zelinsky. We thank Arthur Samuel and Richard Gerrig for their feedback throughout this project, and Jeremy Wolfe for many insightful comments on an earlier version of this manuscript.
1To ensure that no artifact of display presentation was contributing to static and dynamic search differences, a static trial consisted of 15 identical video frames drawn to the screen.
2We believe one factor contributing to this uncommonly shallow slope might have been the time-terminated displays in our paradigm. As discussed more fully elsewhere (Klein and MacInnes,
3Not shown in these scatterplots are the initial and final fixations of a trial. Initial fixations were excluded because they were always aligned with a centrally located fixation cross. Final fixations were excluded because observers in the static search task overwhelmingly chose to fixate the target while making their button press response, a behavior that would not be possible in the dynamic display condition.