Atten Percept Psychophys DOI 10.3758/s13414-014-0713-4

Effects of spatial configurations on the resolution of spatial working memory Aysu Mutluturk & Aysecan Boduroglu

# The Psychonomic Society, Inc. 2014

Abstract Recent research demonstrated that people represent spatial information configurally and preservation of configural cues at retrieval helps memory for spatial locations (Boduroğlu & Shah, Memory & Cognition, 37(8), 1120– 1131 2009; Jiang, Olson, & Chun, Journal of Experimental Psychology: Learning, Memory, and Cognition, 26(3), 683– 702 2000). The present study investigated the effects of spatial configurations on the resolution of individual location representations. In an open-ended task, participants first studied a set of object locations (three and five locations). Then, in a test display where available configural cues were manipulated, participants were asked to determine the original location of a target object whose color was auditorially cued. The difference between the reported location and the original location was taken as a measure of spatial resolution. In three experiments, we consistently observed that the resolution of spatial representations was facilitated by the preservation of spatial configurations at retrieval. We argue that participants may be using available configural cues in conjunction with the summary representation (e.g., centroid) of the original display in the computation of target locations. Keywords Configurations . Representation resolution . Spatial representations . Ensemble representations In the last 2 decades, growing interest in visual short-term/ working memory (WM) has resulted in the accumulation of evidence regarding the capacity limitations on visual WM and the nature of visual representations (for a review, see Luck, 2008). However, most of this past research does not speak to A. Mutluturk : A. Boduroglu (*) Bogazici Universitesi, Psikoloji Bolumu Bebek, 34342 Istanbul, Turkey e-mail: [email protected]

the nature of visuospatial representations. The present research directly investigates the nature of spatial representations in WM. Even though it is typically thought that objects in visual memory are represented in a configural manner and that such representations are effortless (Jiang, Olson, & Chun, 2000; Simons, 1996), more recent research has demonstrated that configural representations are not obligatory (Gmeindl, Nelson, Wiggin, & Reuter-Lorenz, 2011) and the ability to represent locations in a configural manner may be related to WM capacity (Boduroglu & Shah, 2009). Thus, alternate modes of spatial representation must be further investigated. There are two logical possibilities about how objects may be represented in WM (Boduroglu & Shah, 2014). Configural representations suggest that locations of objects may be represented in a relational manner, possibly as part of a global layout. It is also possible that people represent location information in absolute terms, where each represented object is coded independently from others. A similar dichotomy has been suggested in some prior research on location memory (Boduroglu & Shah, 2009, 2014; Chun & Jiang, 1998; Jager & Postma, 2003; Jiang & Wagner, 2004; Kosslyn et al., 1989). Since it has been shown that disruption of configural cues at the time of retrieval results in poorer location memory for individual targets (e.g., Jiang et al., 2000, Experiment 2A), in the present experiments we further investigated the interaction between these two possible modes of representations (see also Jiang & Wagner, 2004). We specifically tested the effects of configuration-based representations on the resolution of independent representations in spatial WM by manipulating the availability and congruency of configural information at retrieval. One key aspect of absolute positional representations that may be influenced by configural changes is their resolution. Resolution of representations basically refers to the fidelity of representations, and it has been operationalized in the visual WM literature in two ways: (1) comparison of recognition

Atten Percept Psychophys

errors in a change detection task that requires making fine or coarse discriminations between complex objects (i.e., within or between category changes; Awh, Barton, & Vogel, 2007; Barton, Ester, & Awh, 2009; Scolari, Vogel, & Awh, 2008) or (2) precision of responses measured by open-ended tasks requiring retrieval of exact features of studied objects (e.g., Bays, Catalao, & Husain, 2009; Bays & Husain, 2008; Zhang & Luck, 2008, 2011). Given that the former approach provides a more indirect measure, in the present experiments, we chose to measure spatial resolution by using an open-ended task. Specifically, we used a spatial recall task that had already been used in our lab to measure spatial WM resolution (Boduroglu, Mueller, Ng, & Shah, under review). In this task, participants study a set of randomly placed colored objects. Then one of the objects is presented at the center, and participants are asked to drag this target object to its original position. The distance between the original and the reported location of the target object is taken to be a measure of spatial representation resolution. In order to investigate how configural representations impact absolute location representations, we manipulated the nature of configuration information presented to participants at retrieval. At retrieval, the color of the target that was to be placed in its original location was specified via an auditory cue; the locations of nontarget objects were manipulated, which resulted in either the preservation or the disruption of the initially studied configuration. In the same-configuration condition, the positions of nontargets in the probe display remained identical with their positions in the study display. In the different-configuration condition, the positions of the nontargets randomly changed. In the no-configuration condition, none of the nontargets was presented, resulting in the absence of configural cues. As participants placed the target in its original location, the original configuration was regenerated only in the same-configuration condition. If congruency of spatial configurations between study and retrieval influences the resolution of individual location representations, we would expect participants to report the location of targets more precisely in the same-configuration condition, as opposed to the different- and no-configuration conditions. The literature does not allow us to make specific predictions on how the different- and no-configuration conditions would compare with one another.

Experiment 1 Method Participants Twenty-six Boğaziçi University undergraduates with normal or corrected-to-normal vision participated in the experiment in

return for course credit. Data from 1 participant were excluded because he or she reported having astigmatism, leaving us with 25 participants (21 female; mean age = 20.24 ± 2.35). All participants provided informed consent, and they were debriefed at the end of the session. Apparatus The participants were tested in a well-lit room. A computer with an Intel Core 2 Duo processor, an ATI Radeon X300/X550/X1050 Series graphics card, and a 17-in. CRT Philips 107S6 monitor was used to present stimuli. The screen resolution was set to 640 × 480, with a refresh rate of 75 Hz (refresh duration = 13.33 ms). The experiment was programmed in E-Prime (Psychology Tools, Inc.). Participants viewed the computer screen from approximately 57 cm, where 1 cm corresponds to 1°. Materials In the spatial recall task, each trial began with a warning cross that stayed on the screen for 500 ms signaling the appearance of the next study display. Participants were instructed to keep their eyes on the fixation cross while it was visible. Then participants were presented with either three or five randomly positioned colored squares for 500 ms. After a brief delay (900 ms), they were presented with the probe display. In the probe display, participants had to determine the location of one of the colored squares; the target object’s color was heard over headphones. The onset of the sound cue was simultaneous with that of the probe display. On one third of the trials, no other objects were presented in the probe display (noconfiguration condition). On the remaining trials, we manipulated the locations of nontarget objects; they were presented either in their original locations (same-configuration condition) or in a randomly determined new positions (different-configuration condition). The probe display was shown until participants responded, and instructions emphasized accuracy over speed. Participants responded by clicking on the location of the target object with a mouse. The target object appeared where the mouse was clicked. After each response, participants were asked whether they had remembered or had guessed the answer; they were instructed that the “guess” option was to be used only when they did not have any idea where the target was. The stimuli consisted of a set of colored squares with sides of 1 cm, subtending a visual angle of 1°. In the set size 3 condition, three squares colored in red, green, and blue were displayed on a gray background. In the set size 5 condition, purple and yellow squares were added to the stimulus set. The target object was randomly chosen. The repetition of colors was not allowed in a particular display (Fig. 1a).

Atten Percept Psychophys

Fig. 1 a Probe types (same, no, and different configuration) for set size 3 in Experiment 1. b In Experiment 2, the no-configuration condition was replaced with the color swap condition. Different shadings represent different colors

On each trial, the locations of the squares were pseudorandomly generated within a 12° × 12° square region. Objects never appeared in the central foveal region (3° × 3°), and they were separated by a minimum interitem distance of 3°. In the different-configuration condition, nontargets were presented in a new set of locations in the probe display, with the above-mentioned constraints. In addition, each new location in the probe display was separated from the original location of the target object by a minimum distance of 3°. Thus, an overlap between an object at a new location and the target location was prevented. Each participant completed a total of 360 experimental trials. Trials were blocked by set size. Each block included 180 trials, and within each block, there were 60 trials per condition that were randomly intermixed. After completing 60 trials, participants were required to take a 2-min break. Before the beginning of the experimental trials, each participant received 15 practice trials (5 practice trials for each condition in both set sizes). Half of the participants first received the set size 3; the other half first received the set size 5 condition.

Results For Experiment 1, we analyzed the effects of configural cues and set size on accuracy of retrieval (indicated by Euclidean error) and reaction times (RTs) for “remembered” trials. Both independent variables impacted the two mentioned dependent variables. Below, we describe these relationships in greater detail.

For each trial, to determine how accurately each location was represented, we calculated the Euclidean distance (in visual degrees) between the target and reported location. Lower error meant higher precision. For each participant, we calculated the median error across remember trials for each condition, for each list length. We chose to report median errors because each participant’s data were nonnormally distributed; the results of the Kolmogorov–Smirnov (K–S) test of normality revealed that the distributions deviated from a normal distribution (all ps < .001). We excluded “guess” responses from further analyses,1 2; in all conditions, participants’ errors were significantly higher on the guess (same configuration, M = 3.42°, SD = 2.52°; no configuration, M = 2.75°, SD = 1.75°; different configuration, M = 3.15°, SD = 1.37°) than on the remember (same configuration, M = 1.07°, SD = 0.33°; no configuration, M = 1.36°, SD = 0.36°; different configuration, M = 1.37°, SD = 0.43°) trials (all ps < . 001).3

1

When we collapsed guess and remember responses, the pattern of results was the same as the results from the “remember” responses only (same configuration, M = 1.13°, SD = 0.36°; no configuration, M = 1.40°, SD = 0.37°; different configuration, M = 1.58°, SD = 0.49°. 2 Percentage of guess responses in each condition was as follows. For set size 3, it was 3 % (±5), 2 % (±3), and 8 % (±10) in the same-, no-, and different-configuration conditions, respectively. For set size 5, it was 13 % (±16), %13 (±12), and 29 % (±17) in the same-, no-, and different-configuration conditions, respectively. 3 When we took an objective rather than a subjective criterion to exclude guess responses (for each participant, trials on which error was greater than mean error + 3 SDs were excluded), the pattern of results remained the same (the amount of error on remember trials were as follows: same configuration, M = 1.06, SD = 0.32; no configuration, M = 1.36, SD = 0.36; different configuration, M = 1.35, SD = 0.43).

Atten Percept Psychophys

Further analysis of the no-configuration condition In the no-configuration condition, participants were not presented with any of the nontarget objects at retrieval, which may, in turn, have misled them to report the location of a 4 All pairwise comparisons were Bonferroni-corrected here as well as throughout the article.

1.80

Distance from Original Locations (degrees)

In order to assess whether configuration had an effect on the precision of responses, a 2 × 3 repeated measures ANOVA was conducted with probe type (same, no, different configuration) and set size (3 and 5) as within-subjects factors. The results revealed that there was a main effect of probe type, F(2, 48) = 41.10, MSE = .04, p < .001, η2p = .63. This main effect was driven by the larger amount of error in the different- and no-configuration conditions than in the same-configuration condition (both ps < .001).4 There was no difference between the different- and no-configuration conditions. Set size had a main effect, F(1, 24) = 9.11, MSE = .09, p = .006, η2p = .28; this was due to the higher amount of error in set size 5 (M = 1.34°, SD = 0.40°) than in set size 3 (M = 1.20°, SD = 0.34°). There was no interaction between probe type and set size, F(2, 48) =0.31, MSE = .01, p = .69, η2p = .01( see Fig. 2). Analyses of RT data also revealed that there was a main effect of probe type, F(2, 48) = 377.56, MSE = 10,462.74, p < .001, η2p = .94. Post hoc comparisons showed that RT was longest in the no-configuration condition (M = 2,193.62, SD = 207.07; M = 1,677.85, SD = 251.07, and M = 1,742.07, SD = 263.34, for the same- and different-configuration conditions, respectively; both ps < .001). The RT difference between the different- and same-configuration conditions was also significant (p = .02). This RT pattern suggests that the disruption of configuration cues brought an additional cost during the retrieval of individual object locations; this cost was largest in the no-configuration condition. As in the error data, in the RT data, there was also a main effect of set size; participants were slower on the set size 5 trials (M = 1,947.1, SD = 236.68) than on the set size 3 trials (M = 1,795.26, SD = 244.31), F(1, 24) = 24.65, MSE = 35,072.38, p < .001, η2p = .51. There was no interaction between probe type and set size, F(2, 48) = 1.24, MSE = 5,678.83, p = .30, η2p = .05. Overall, these data indicate that changing configurations or absence of configural cues at retrieval reduced the precision of individual location judgments. Consistent with findings of load effects on spatial memory (for a more detailed discussion of this issue, see Boduroglu & Shah, 2014), the increase in set size resulted in a higher amount of error in recalling the location of the target.

1.60

1.44

1.43 1.30

1.40

1.31

1.16

1.20

0.98

1.00

Set Size 3

0.80

Set Size 5

0.60 0.40 0.20 0.00

Same Config.

No Config. Probe Type

Different Config.

Fig. 2 Median error as a function of probe type in Experiment 1. Config. = configuration

nontarget object. If so, this would have caused an artificial increase in error in the no-configuration condition, in comparison with the same-configuration condition. To rule out this explanation, we calculated the minimum distance between nontarget objects and the reported location. As is illustrated in Fig. 3, in all experimental conditions, the minimum distance between nontarget locations and response was significantly greater, as compared with the distance between target and response, (all ps < .001). The large distance between responses and nontarget objects suggests that participants were unlikely to report a nontarget instead of a target location. Discussion In the present experiment, we demonstrated that spatial configurations impact the precision of individual location representations. Specifically, we demonstrated that disruption or absence of configuration cues at retrieval resulted in higher error for individual object locations. We also observed an increase in the amount of errors from set size 3 to 5, suggesting that such an increase in set size may result in lower memory resolution. In general, the RT data yielded a similar pattern of results with the error data. Critically, RT was longest when no configural cues were present at retrieval. It is possible that participants were trying to visualize the original configuration while trying to retrieve the target location, and in the no-configuration condition, the absence of configural cues may have interfered with the effective utilization of visual imagery in determining the target location. The greater error in the no-configuration condition is unlikely to be driven by color–location binding errors; in the no-configuration condition, responses were actually closer to targets than to distractors, suggesting that participants were indeed trying to retrieve the item whose color was heard over the headphones.

Atten Percept Psychophys 3 Objects Distance from Response (degrees)

6.00

5 Objects

5.80

5.47

5.40

5.00

4.22

4.04

4.03

4.00

Target

3.00 2.00 1.00

0.98

1.30

1.31

1.16

1.43

1.44

Non-Target

0.00

Same

No

Diff

Same

No

Diff

Fig. 3 Distances between the response and the target and nontarget locations in Experiment 1. Diff = different

Experiment 2 Experiment 1 demonstrated that configural information impacts the resolution of spatial representations. However, it is possible that the results from Experiment 1 could be attributed to the existence of a change at test, not necessarily a configural change, in the two critical conditions (no- and different-configuration conditions). In other words, any change at retrieval that potentially distracted participants may have resulted in greater error. In Experiment 2, we directly tested this possibility by using a color swap condition instead of the no-configuration condition (see Fig. 1b). In the color swap condition, at the retrieval phase, nontarget objects switched their colors, allowing the overall spatial configuration to remain the same, yet the relative location cues to become uninformative. The same- and different-configuration conditions were identical with those in Experiment 1. If the amount of error were similar in the swap and different conditions, this would suggest that location representations do not solely rely on configural information. On the other hand, if participants’ resolution for target locations is selectively impaired only when configural changes are introduced (i.e., in the different condition), then it would suggest that configural information is more readily available and heavily relied on for determining the locations of individual objects. Method Participants Twenty-six Boğaziçi University undergraduates with normal or corrected-to-normal vision participated in the experiment in return for course credit (18 female; mean age = 19.85 ± 1.64). All participants provided informed consent, and they were debriefed at the end of the session. Materials Procedures were identical to those in Experiment 1, except for the following changes. In Experiment 2, instead of the noconfiguration condition, we used the color swap condition to

test whether any change at retrieval may equally decrease the precision of reports as configural changes. There were two within-subjects variables: probe type (same, color, different) and set size (3 and 5). In the color swap condition, in the probe display, nontarget objects switched colors. The same and different conditions were identical with the same- and different-configuration conditions in Experiment 1.

Results In Experiment 2, the data showed that the errors were again the highest in the different-configuration condition. Critically, the amount of error in the swap condition was significantly less than in the different-configuration condition. Consistent with Experiment 1, in Experiment 2 the amount of error was least in the same-configuration condition. The data were analyzed as in Experiment 1. At the individual level, the K–S tests revealed that errors were nonnormally distributed for each condition (all ps < .001). Thus, for each participant, we calculated median as opposed to mean error for each condition. As in Experiment 1, participants’ errors were significantly higher for guess responses (same configuration, M = 2.46°, SD = 1.58°; color swap, M = 2.51°, SD = 1.38°; different configuration, M = 3.11°, SD = 1.48°) than on remember trials (same configuration, M = 0.99°, SD = 0.21°; color swap, M = 1.07°, SD = 0.24°; different configuration, M = 1.41°, SD = 0.31°; all ps < .001).5 With the amount of errors as the dependent variable, a 2 (set size) × 3 (probe type) repeated measures ANOVA was conducted. The results yielded a main effect of probe type, F(2, 50) = 87.22, MSE = .03, p < .001, η2p = .78. Post hoc comparisons revealed that the amount of error in the different-configuration condition (M = 1.41°, SD = 0.31°) was significantly higher than that in both the color swap (M = 1.07°, SD = 0.24°) and the same-configuration 5 Percentage of guess responses in each condition was as follows For set size 3, it was 2 % (±4), 6 % (±6), and 10 % (±13) in the sameconfiguration, color swap, and different-configuration conditions, respectively. For set size 5, it was 12 % (±11), 21 % (±15), and 31 % (±17) in the same-configuration, color swap, and different-configuration conditions, respectively.

Atten Percept Psychophys

Distance from Original Locations (degrees)

1.80 1.60

1.47 1.35

1.40 1.20

1.06

1.07 1.06

0.91

1.00

Set Size 3

0.80

Set Size 5

0.60 0.40 0.20 0.00

Same Config.

Color Swap Probe Type

Different Config.

Fig. 4 Median error as a function of probe type in Experiment 2. Config. = configuration

(M = 0.99°, SD = 0.21°) conditions (all ps < .001) (Fig. 4). The amount of error in the color swap condition was significantly higher than that in the same configuration condition (p = .006). Also, there was also a marginal effect of set size on error, F(1, 25) = 3.33, MSE = .09, p = .08, η2p = .12, with greater error when five, as opposed to three, locations were studied (M = 1.11°, SD = 0.25°, and M = 1.20°, SD = 0.26°, for set size 3 and 5, respectively). Finally, there was a marginal interaction between probe type and set size, F(2, 50) = 2.84, MSE = .034, p = .09, η2p = .10. This marginally significant yet small interaction was due to the different pattern of error in the same and swap conditions across the two set sizes. While for set size 3, the amount of error in the color swap condition was larger than that in the same-configuration condition (p < .001), in set size 5, the amount of error was similar in the same and color swap conditions (p >.05). For both set sizes, error was significantly larger in the different-configuration than in the other conditions (all ps < .001). Analysis of the RT data yielded a main effect of probe type, F(2, 50) = 14.97, MSE = 5,564.22, p < .001, η2p = .37. Post hoc comparisons revealed that this effect was due to the longer RTs in the different-configuration condition (M = 1,724.58, SD = 226) than in the same-configuration (M = 1,644.83, SD = 212.64, p < .001) and color swap (M = 1,678.84, SD = 219, p = .002) conditions. RTs were significantly longer for set size 5 (M = 1,731.14, SD = 238.13) than for set size 3(M = 1,634.36, SD = 200.28), F(1, 25) = 4.56, MSE = 80,083.51, p = .043, η2p = .15. The interaction between probe type and set size did not reach significance, F(2, 50) = 2.30, MSE = 3,910.01, p = .11, η2p = .08. Discussion In Experiment 2, both the error and RT data demonstrated that people were influenced by configural changes the most.

Interestingly, in the swap condition, error was significantly more than that in the same condition, suggesting that relative location changes impacted performance to a certain degree. These results suggest that configural changes at retrieval impact spatial resolution above and beyond that of relative location changes. This finding is novel in that it demonstrates that both configural and relative location information is represented in spatial WM and that configural representations cannot be reduced to relative location information. More critically, error in the swap condition was less than that in the different condition, suggesting that preservation of global configuration despite relative location changes impacts spatial memory. Previous research had demonstrated that global configural information impacts visual change detection. For instance, Treisman and Zhang (2006) reported that even in visual change detection tasks, people were biased to respond “match” when global configurations of displays were preserved in whole, as opposed to single-probe, displays. Similarly, Boduroglu and Shah (2009) demonstrated that preservation of global configural structure resulted in a bias to report no color changes (i.e., responding “same”). The findings of Experiment 2 suggest that global configural structure directly impacts the resolution of individual location representations.

Experiment 3 In Experiments 1 and 2, we consistently observed that the resolution of location memory was worse when configural information was disrupted at retrieval (different-configuration condition) and best when configural information was preserved between study and probe displays (sameconfiguration condition). However, the results from these two experiments do not allow us to determine whether the observed pattern of results is due to the reduction of resolution of location memory because of disruption of global configuration or a facilitation of location memory because of preservation of global configurations. The goal of Experiment 3 was to directly test these two possibilities. In the different-configuration condition, the configural cues appeared at random at previously unoccupied locations. The mismatch between initially presented configuration and the one observed at test could reduce the resolution of the target location, since neither the represented global configuration nor the relative interitem relationships are present in the probe display. Thus, it could be argued that there are no effective anchors to utilize while generating the target location. Also, it is possible that the different partial configuration presented at test may interfere with the maintenance (or override) of the initially represented configuration because it is processed as something entirely new. Consequently, identifying the target

Atten Percept Psychophys

location may become difficult in the different-configuration condition. It is equally possible that the pattern of results observed in Experiments 1 and 2 may be due to the facilitatory effects of preserved configural cues presented at test. In the sameconfiguration condition, the viewers may effectively utilize partial yet matching configural cues and be directed to the target location. This would then require performance to be negatively impacted in all circumstances when cues do not directly match the study display. The data from the no configuration condition of Experiment 1 is consistent with the latter view of facilitation. Specifically, even though there were no partial configural cues to interfere with the maintained representation, in the no-configuration condition, the amount of error was equally high as that of the different-configuration condition. Thus, interference caused by the processing demands of the test display cannot fully explain the pattern of results observed. To empirically demonstrate whether preserved configural cues at retrieval result in facilitation, in Experiment 3, we modified our paradigm and provided the auditory cue identifying the target object prior to the onset of the response screen where the configural cue was present. This allowed participants to retrieve the target location prior to the onset of the probe display and, in turn, enabled us to use the noconfiguration condition as a baseline while interpreting the data. Then the pattern of results could specifically be attributed to either the disruption caused by the different or the facilitation caused by the preserved configural cues at test. In other words, sequentially presenting the auditory and configural cues at retrieval allowed us to eliminate any potential biases introduced by presenting these simultaneously. If participants perform better in the same-configuration than in the no-configuration condition and performance is similar across the no and different conditions, one could conclude that preserved configurations facilitate location memory. On the other hand, if performance in the different-configuration condition is worse than that in the no-configuration condition, this would be consistent with the idea that different configural cues at retrieval reduce the precision of target representations.6

Method Participants Twenty-four Boğaziçi University undergraduates with normal or corrected-to-normal vision participated in the experiment in return for course credit. Data from 1 participant were excluded 6 Even though there are other logical possibilities regarding the data patterns (e.g., the same- or/and different-configuration conditions is/ are worse than the no-configuration condition), results from both Experiments 1 and 2 suggest that these are unlikely.

because of equipment malfunction, leaving 23 participants (14 female; mean age = 21.83 ± 2.44). Materials and procedure The only difference between Experiments 1 and 3 was the temporal separation of the response prompt and configural cues in Experiment 3, instead of their simultaneous onset as in Experiment 1. After the presentation of the stimulus display (500 ms) and a brief delay (900 ms), participants heard the response prompt within the next 600 ms.7 Immediately after the response prompt offset, the probe display was presented. Hence, participants were presented with the auditory cue after a 900-ms delay, and the probe display with (or without) configural cues after a total delay of 1,500 ms. Results The general pattern of results was similar to that in Experiment 1. As in Experiment 1, participants’ errors were significantly higher for guess responses (same configuration, M = 4.43°, SD = 3.07°; no configuration, M = 3.67°, SD = 2.83°; different configuration, M = 3.76°, SD = 2.42°) than for remember trials (same configuration, M = 1.16°, SD = 0.34°; no configuration, M = 1.56°, SD = 0.44°; different configuration, M = 1.54°, SD = 0.50°; all ps < .001).8 Our results, as illustrated in Fig. 5, revealed that facilitation was more likely when configural cues were preserved at retrieval. The results showed a main effect of probe type, F(2, 44) = 45.51, MSE = .053, p < .001, η2p = .67. Performance was significantly better in the same-configuration condition, as opposed to the no- and different-configuration conditions, p < .001, Bonferroni corrected (for Ms and SDs; see above). There was no difference between the different- and no-configuration conditions (p > .05). A 2 (set size) × 3 (probe type) repeated measures ANOVA on RT data also revealed a main effect of probe type, F(2, 44) = 5.52, MSE = 6,740.85, p = .007, η2p = .20. Participants were fastest in the no-configuration condition (M = 1,267.15, SD = 272.95), as compared with the different- (M = 1,319.39, SD = 294.07, p = .02) and same-configuration (M = 1,312.78, SD = 235.62, p = .05; all Bonferroni corrected) conditions. This RT 7 The duration of each auditory cue was slightly different due to word length differences. However, by presenting the cues immediately at the offset of the delay interval, we ensured that for all cues, participants received the cue exactly at the same spot in the trial timeline; there were slight variations regarding the ending of each auditory cue. Since all colors were cued equally often across all conditions, we believed that this was unlikely to cause any systematic biases in the data. 8 Percentage of guess responses in each condition was as follows. For set size 3, it was 2 % (±5), 7 % (±16), and 5 % (±11) in the same-, no-, and different-configuration conditions, respectively. For set size 5, it was 15 % (±11), 19 % (±19), and 19 % (±19) in the same-, no-, and different-configuration conditions, respectively.

Atten Percept Psychophys 2.00

1.73

Distance from Original Locations (degrees)

1.80 1.60

1.28

1.40

1.40

1.66 1.41

1.03

1.20 1.00

Set Size 3

0.80

Set Size 5

0.60 0.40 0.20 0.00

Same Config.

No Config. Probe Type

Diff Config.

Fig. 5 Median error as a function of probe type in Experiment 3. Config. = configuration

pattern suggests that participants may have started retrieving the location of the target prior to the onset of the probe screen, and unless they were provided with additional partial cues (as in the same- and different-configuration conditions), they reported their response as soon as they were given the chance to do so. However, when additional information was presented at the probe screen, they may have taken some additional time to adjust their responses in light of the new information provided. In line with previous experiments, set size had a main effect on both error, F(1, 22) = 36.75, MSE = .074, p < .001, η2p = .63, and RT, F(1, 22) = 4.70, MSE = 80,476.57, p = .041, η2p = .18. The effect of set size on error was driven by greater error for set size 5 (M = 1.56°, SD = 0.45°) than for set size 3 (M = 1.28°, SD = 0.41°), p < .001, Bonferroni corrected. Similarly, the set size effect on RT was due to longer RTs for set size 5, as compared with set size 3. There was no significant interaction between probe type and set size for either error, F = 1.05, p = .36, or RT, F = 2.41, p = .10. Discussion In Experiment 3, we demonstrated that presenting cues preserving the configural structure at retrieval facilitates memory for individual locations. In this experiment, we separated the onsets of the response prompt and the configural cue, allowing participants to retrieve the target location before they finalized their responses on the response screen. In other words, we were able to directly compare the impact of same and different cues on the resolution of individual target locations to determine whether configural cues available at retrieval increased the resolution of target locations. Consistent with both Experiments 1 and 2, in this experiment, participants performed best when spatial configural information was partially present and preserved at test. As in Experiment 1, there were no significant differences in the amount of error in the no- and differentconfiguration conditions.

Interestingly, the RT data revealed a different pattern from Experiment 1; participants were fastest in the no-configuration condition, and response speed did not vary in the same- and different-configuration conditions. It is likely that presenting the auditory cue identifying the target prior to the onset of the probe screen allowed participants to retrieve the target location. When additional information was not presented, they responded faster. However, when additional information was presented during the test phase, these pieces of information were further processed before a response was made. Specifically, in the same- and different-configuration conditions, trying to match the represented global configurations and interitem relationships with the cues present at the response screen may have increased RT. Whenconfigural cues were different, the mismatch between the initially represented configuration and the cues present at retrieval possibly let observers disregard them, resulting in similar levels of error in the no- and different-configuration conditions. However, in the same-configuration condition when configural cues presented at the probe screen were consistent with the represented configuration, the amount of error was reduced, suggesting that participants incorporated this information in their computation of the target location. This error pattern, together with the RT pattern, favors the facilitative effects of the preserved configuration on the location memory.

General discussion These three experiments demonstrate that people rely on configuration information while recalling the exact position of a target. In all experiments, we demonstrate that the resolution of spatial WM representations is better in the same-configuration condition, suggesting that participants indeed represent individual target objects within larger configurations. These findings are consistent with previous studies indicating that spatial configurations are encoded along with individual objects (Simons, 1996) and visuospatial WM is organized in a configural manner (Boduroglu & Shah, 2009; Jiang et al., 2000). More important, the evidence presented in this article is the first to show the effects of configuration information on the resolution of individual spatial representations. In all experiments, we consistently demonstrated that the precision of reports was facilitated by the preservation of configural cues at retrieval. Furthermore, Experiment 2 demonstrated that in addition to configural information, spatial representations include relative location information. These results point to the possibility that different modes of spatial representations coexist and are not mutually exclusive (see also Boduroglu & Shah, 2014). Future research needs to define

Atten Percept Psychophys

the underlying mechanisms by which these different modes of spatial representations interact. What remains unclear from the research on configural spatial representations is the exact nature of configural representations. There are a few possibilities through which configuration information may be represented. One such possibility is that locations are represented as forming a virtual polygon, with each object becoming a vertex of this polygon (Yantis, 1992). The selective interference effects on location memory from visual shapes and patterns (for a review, see Jiang, Makovski, & Shim, 2009) may make this option seem viable. However, it is not clear how one could reconcile this possibility with our observation that configural representations cannot be reduced to relative location cues. It is highly likely that configural representations in the form of virtual polygons inherently contain relative location information, and thus the separability of the above-described modes of representation seems unlikely. Another possibility is that viewers represent spatial configurations as ensembles. Growing research in the visual domain has indicated that viewers can easily extract summary statistics from visual displays, and their ability to do so seems to extend to various features and stimuli (e.g., Alvarez, 2011; Alvarez & Oliva, 2008; Ariely, 2001; Brady, Konkle, & Alvarez, 2009; Brady & Tenenbaum, 2012; Oliva & Torralba, 2006, 2007). Ongoing work from our lab suggests that participants are able to accurately and effortlessly represent the statistical summary of a set of spatial locations in the form of a centroid (i.e., the center of mass of the visual display) (Mutlutürk & Boduroglu, 2013). It is possible that visual information is represented in a constructive manner, encoding displays at multiple levels rather than only representing information about individual objects (Brady & Alvarez, 2011). People may build the spatial representation by combining the summary information of a display with the information on individual and relative object locations. These modes of spatial representations may coexist, possibly at different levels of a hierarchy. The facilitatory effects we observed in the same-configuration condition may be due to this hierarchical representation that allows people to compute the location of an individual item precisely when configural cues are available at retrieval. Specifically, if people can easily and accurately extract the centroid of a display (Mutlutürk & Boduroglu, 2013), then the present cues congruent with the initially studied display can help narrow down the options where the target could possibly be. Thus, the error would be reduced with congruent configural cues. In a similar vein, in the different condition, the cues would mislead observers, resulting in greater error, as compared with the same condition. In our data, in the same-configuration condition, the available configural cues at test (possibly indicating the vertices of

a virtual polygon) and summary information (e.g., centroid) represented in spatial WM may allow people to compute the exact location of a missing target. However, when configural cues change or when they are absent, this may hinder their attempts to compute the target location, resulting in higher amounts of error in responses.

References Alvarez, G. A. (2011). Representing multiple objects as an ensemble enhances visual cognition. Trends in Cognitive Sciences, 15, 122– 131. Alvarez, G., & Oliva, A. (2008). The representation of simple ensemble visual features outside the focus of attention. Psychological Science, 19(4), 392–398. Ariely, D. (2001). Seeing sets: Representation by statistical properties. Psychological Science, 12(2), 157–162. Awh, E., Barton, B., & Vogel, E. (2007). Visual working memory represents a fixed number of items regardless of complexity. Psychological Science, 18(7), 622–628. Barton, B., Ester, E. F., & Awh, E. (2009). Discrete resource allocation in visual working memory. Journal of Experimental Psychology: Human Perception and Performance, 35, 1359–1367. Bays, P., Catalao, R., & Husain, M. (2009). The precision of visual working memory is set by allocation of a shared resource. Journal of Vision, 9(10), 1–11. Bays, P., & Husain, M. (2008). Dynamic shifts of limited working memory resources in human vision. Science, 321(5890), 851–854. Boduroglu, A., & Shah, P. (2009). Effects of spatial configurations on visual change detection: An account of bias changes. Memory & Cognition, 37(8), 1120–1131. Boduroglu, A. & Shah, P. (2014). Configural representations in spatial working memory. Visual Cognition,1–25 doi:10.1080/13506285. 2013.875499 Brady, T. F., & Alvarez, G. A. (2011). Hierarchical encoding in visual working memory: Ensemble statistics bias memory for individual items. Psychological Science, 22(3), 384–392. Brady, T. F., Konkle, T., & Alvarez, G. A. (2009). Compression in visual short-term memory: Using statistical regularities to form more efficient memory representations. Journal of Experimental Psychology: General, 138, 487–502. Brady, T. F., & Tenenbaum, J. B. (2012). A probabilistic model of visual working memory: Incorporating higher order regularities into working memory capacity estimates. Psychological Review. doi:10.1037/ a0030779. Advance online publication. Chun, M. M., & Jiang, Y. (1998). Contextual cueing: Implicit learning and memory of visual context guides spatial attention. Cognitive Psychology, 36(1), 28–71. Gmeindl, L., Nelson, J. K., Wiggin, T., & Reuter-Lorenz, P. A. (2011). Configural representations in spatial working memory: Modulation by perceptual segregation and voluntary attention. Attention, Perception, & Psychophysics, 73, 2130–2142. Jager, G., & Postma, A. (2003). On the hemispheric specialization for categorical and coordinate spatial relations: A review of the current evidence. Neuropsychologia, 41, 504–515. Jiang, Y. V., Makovski, T., & Shim, W. M. (2009). Visual memory for features, conjunctions, objects, and locations. In J. R. Brockmole (Ed.), The visual world in memory (pp. 33–65). Hove, England: Psychology Press.

Atten Percept Psychophys Jiang, Y., Olson, I., & Chun, M. (2000). Organization of visual short-term memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26(3), 683–702. Jiang, Y., & Wagner, L. (2004). What is learned in spatial contextual cuing-configuration or individual locations? Perception and Psychophysics, 66(3), 454–463. Kosslyn, S., Koenig, O., Barrett, A., Cave, C., Tang, J., & Gabrieli, J. (1989). Evidence for two types of spatial representations: Hemispheric specialization for categorical and coordinate relations. Journal of Experimental Psychology: Human Perception and Performance, 15(4), 723–735. Luck, S. J. (2008). Visual short-term memory. In S. J. Luck & A. Hollingworth (Eds.), Visual memory. New York: Oxford University Press. Mutlutürk, A., & Boduroglu, A. (2013). Resolution of centroid and individual location representations in spatial working memory. Manuscript in preparation. Oliva, A., & Torralba, A. (2006). Building the gist of a scene: The role of global image features in recognition. Progress in Brain Research, 155, 23–36.

Oliva, A., & Torralba, A. (2007). The role of context in object recognition. Trends in Cognitive Science, 12, 520–527. Scolari, M., Vogel, E. K., & Awh, E. (2008). Perceptual expertise enhances the resolution but not the number of representations in working memory. Psychonomic Bulletin & Review, 15(1), 215–222. Simons, D. (1996). In sight, out of mind: When object representations fail. Psychological Science, 7(5), 301–305. Treisman, A., & Zhang, W. W. (2006). Location and binding in visual working memory. Memory & Cognition, 34(8), 1704 – 1719. Yantis, S. (1992). Multielement visual tracking: Attention and perceptual organization. Cognitive Psychology, 24, 295–340. Zhang, W., & Luck, S. (2008). Discrete fixed-resolution representations in visual working memory. Nature, 453(7192), 233–235. Zhang, W., & Luck, S. J. (2011). The number and quality of representations in working memory. Psychological Sciences, 22, 1434–1441.