Selective attention can enhance some aspects of our visual world while filtering others from awareness. Given our limited cognitive resources, such filtering is essential when viewing complex scenes, but it also applies to simple scenes. Eitam, Yeshurun, and Hassan (2013) observed better performance for the attended color than the ignored color in a simple, two-colored object even though both colors were salient and the complexity of the display did not tax the capacity of visual memory. Our goal was to replicate this finding while addressing a potential task demand that could have contributed to the results. Specifically, participants might have misread the instructions and mistakenly reported the attended color when asked to report the ignored color first. Experiment 1 (n=67) replicated Eitam et al.’s (2013) finding while measuring memory precision. We found that people had worse memory for the ignored than the attended feature of a single, simple object. Experiment 2 (n=69) replicated the pattern while again addressing the potential task demand, although the effect was smaller. Experiment 3 (n=186) provided visual feedback to eliminate any remaining risk of response error and again replicated the original finding. Attended information was stored with greater precision than unattended information, even for a simple object.

Selective attention allows us to focus on some aspects of our visual world while ignoring others. For example, when searching for apples among other produce in a market, having an attention set for red helps us find the apples more efficiently (Most et al., 2005; Neisser, 1976). In addition to improving processing of attended features, attention can also enhance spatial resolution for attended areas while decreasing it for unattended areas (Yeshurun & Carrasco, 1998).

Attentional selection also contributes to the perception of simple displays. For example, people often fail to notice an unexpected object appearing at fixation when engaging in another task (Mack & Rock, 1998). They even can miss an unexpected face in an otherwise empty display (Persuh & Melara, 2016). Given that even basic objects often have many features including shape, size, hue, texture, and luminance (Swan et al., 2016), it is important to understand how task relevance influences the way we store that information.

Eitam, Yeshurun, and Hassan (2013) tested whether attentional selection would affect memory for a two-colored object in a simple display. Participants viewed a colored disk surrounded by a differently colored ring. When asked to focus attention on the whole object, they accurately selected both colors in subsequent forced-choice decisions (only 8% errors). However, when asked to focus on one color (the ring or the disk), they performed better for the attended color than the ignored one. The display itself did not exceed their capacity for attention, perception, or memory [as might happen with more objects or more complex displays; see Zhang and Luck (2008)].

However, the studies tested recognition rather than recall; participants were asked to select the color they had seen from a fixed set of options (either three or four), so the task did not measure the precision of memory. The use of a forced-choice task could amplify an otherwise subtle difference in memory precision for the attended and ignored feature because participants are required to select from the options available; a small difference in the mental representation favoring one option could yield a large difference in choice accuracy. Even if a participant were almost entirely uncertain about the color, if their memory trace slightly favored red over blue, they would select red most of the time when forced to make a decision. Therefore, the proportion of participants selecting a color does not necessarily indicate the precision of their memory for that color, but only that they thought it was more likely to be that color than any other. To address this issue, our replication used a recall task in which participants selected the color they remembered using a color wheel, thereby allowing us to measure the precision of their memory (e.g., see Zhang & Luck, 2008).

Eitam and colleagues (2013) presented the two forced-choice recognition tasks sequentially, with order counterbalanced across participants; some people were tested first on the attended ring, and others were tested first on the ignored ring. That approach introduces a possible confound. Participants might not expect a memory test for the unattended feature, which could cause them to mistakenly report the attended color when asked to report the unexpected color first. Even if participants perfectly remembered both colors, this error could lead to better memory performance for the attended color than the ignored color.

To make the concern concrete, imagine that a participant is asked to focus on the inner disk. They likely will expect a memory test for the disk, so when they see a forced-choice task, they might not read the instructions carefully and therefore select the attended color (which is incorrect). Then, when they unexpectedly see another forced-choice question, they might realize their mistake, read the question thoroughly, and this time report the attended color as requested. Those participants would show better accuracy for the attended color than the ignored color. In contrast, participants who were asked first about the attended color could respond with the correct color on both questions, leading to fewer errors. To address this issue, our second experiment presented both color memory tests in a single display rather than sequentially, and our third experiment asked participants to recreate the image itself (and provided real-time feedback as they adjusted the color wheels) to eliminate any confusion about which feature they were reporting.

Eitam et al. (2013) used a single-trial design to prevent participants from paying more attention to the unattended feature in subsequent trials. Other studies have addressed this concern by only asking about the unattended feature on a final, “surprise” trial. For example, Chen and Wyble (2015a) asked subjects to report a letter on 11 trials and only asked about the letter’s color on the 12th trial. Only 25% of participants could recall the color even though they had focused on the letter. Unlike Eitam et al. (2013), the letters were presented only briefly (200ms), making the representations more susceptible to proactive interference or rapid forgetting (Nee & Jonides, 2013; Oberauer, 2002).

In a similar “surprise trial” design, participants were asked to report an arrow’s color across multiple trials and then were unexpectedly asked to report its orientation (Swan et al., 2016). People were able to recall the orientation, but recalling the orientation impaired their memory for the color when they reported it immediately after. In a larger study, participants who completed 30 trials followed by a surprise question about either orientation or color showed worse precision for the previously irrelevant features (Shin & Ma, 2016). Collectively, these findings suggest that we encode both relevant and irrelevant features, but with varying degrees of precision.

Our studies add to this growing literature on memory for task-irrelevant features. Whereas most studies have presented objects with multiple features and asked about an unattended feature dimension, Eitam et al. (2013) used a simple two-feature object and asked about an unattended feature on the same dimension. Given that their finding is foundational in this literature and may be affected by a task-related confound, we aimed to replicate their finding. Finally, Eitam et al. (2013) also found that participants’ subjective perceptions aligned with their objective performance, so we included confidence measures to replicate that aspect of their study as well.

All three experiments were preregistered, and all materials, code, and data are available in the separate “Study” links under components at https://osf.io/pzavd/. Note that the preregistration plan for Experiment 1 was posted to OSF before data collection, but we neglected to formally register it before starting the study. The project history shows that the preregistration documentation was posted before the start of data collection, with the registration during data collection.

Power Analysis

We preregistered a plan to test at least 60 participants, which a sensitivity analysis shows would give us 95% power to detect an effect size of dz = 0.430 (and 80% power to detect and effect of dz = 0.325) for a paired t-test comparing error rates for the attended and ignored colors (with alpha = 0.05). Eitam et al. (2013) reported observing a difference in accuracy for the attended and ignored feature corresponding to Z = 4.36 for study 1a and Z = 2.86 for study 1b in their forced-choice test. Given that we are unsure how their Z test was computed, we estimated their effect size for our planned test statistic (a paired t-test) by calculating the one-tailed p-value corresponding to their Z statistics and converted those to t. The corresponding t(96) = 4.598 for their study 1a and t(53) = 2.989 for study 1b represent effect sizes of dz = 0.467 and dz = 0.407 respectively. Our target sample size of N = 60 would have approximately 97% power to detect the effect in Study 1a and 93% power for the effect in Study 1b, not taking into account the increased precision of our continuous measure.

Method

Participants. Students (n= 67, not including pilot participants) were recruited from the Psychology Subject Pool at the University of Illinois at Urbana-Champaign and participated in exchange for course credit. The protocol (#09441) was approved by the Institutional Review Board at the University of Illinois at Urbana–Champaign, and participants provided signed, informed consent. Participants self-reported normal/corrected-to-normal vision in a screening questionnaire given to all participants in the subject pool at the start of the semester. We did not exclude any data from recruited participants. Prior to starting the experiment, we collected pilot data to ensure that the program recorded data appropriately and did not crash. These data were automatically excluded from all analyses (they are distinguished from actual subjects in the data files via missing ID numbers or non-standard IDs such as “pilot”). Since this was a one-trial experiment, subjects completed the task prior to participating in an unrelated study that examined individual differences in pseudoscientific beliefs. Participants were informed that they would complete this brief task before participating in a longer study.

Testing Setting. Participants were tested individually or in small groups of up to seven people during the spring and fall semesters of 2018. Testing took place in one of two windowless laboratory rooms, one with eight computer testing stations and one with four. In each room, the computers were separated by dividers so that participants could not see each other’s screens. Each testing station included an identical model of Apple Mini computer with a BenQ XL2420TX 24” LED monitor. Participants rested their chin on an adjustable, custom-made chin rest positioned 57cm from the monitor (such that 1cm on the screen subtends 1 degree of visual angle).

Stimuli and Procedure. In this one-trial experiment, each participant studied a target display consisting of a colored disk (1.87° diameter) surrounded by a differently colored ring (1.87° thick) against a gray background. The ring and disk colors were randomly selected to be 90-180 degrees apart on the color wheel to ensure that they were distinctive.

Before viewing the display, participants were instructed to focus on either the inner disk or outer ring portion of the target display (determined randomly). They pressed the space bar when ready, and the display appeared for 500ms, followed by a 500ms blank screen, and then by two color recall tasks. For each, a ring-shaped, RGB color wheel with the same dimensions as the original stimulus appeared at the center of the screen, and participants clicked on the remembered hue (see Figure 1). After selecting the remembered color, participants rated their confidence in their judgment on a 1-100 slider (cursor started at 50). The order of the two recall tasks was determined randomly for each participant, with approximately half of the participants asked to report the attended color first, and half asked to report the ignored color first.

Figure 1.
Schematic illustration of the task for Experiment 1.
Figure 1.
Schematic illustration of the task for Experiment 1.
Close modal

Results and Discussion

Analyses were conducted in R 4.3.0 (R Core Team, 2019a) using the following packages: tidyverse 2.0.0 (Wickham et al., 2019), data.table 1.14.8 (Dowle & Srinivasan, 2019), tidystats 0.5.2 (Sleegers, 2020), afex 1.3.0 (Singmann et al., 2020), lsr 0.5.2 (Navarro, 2015), PupillometryR 0.0.4 (Forbes, 2020), and ez 4.4.0 (Lawrence, 2016). The manuscript was written using in RMarkdown 2.21 (Allaire et al., 2020) using the following packages: papaja 0.1.1 (Aust & Barth, 2020), knitr 1.42 (Xie, 2020), kableExtra 1.3.4 (Zhu, 2019), grid 4.3.0 (R Core Team, 2019b), and gridExtra 2.3 (Auguie, 2017).

Confirmatory Analyses - Memory Precision. We calculated the absolute difference in degrees between the actual hue and the participant-reported hue (their error in degrees ranging from 0-180°) separately for the ring and the disk. Like Eitam et al. (2013), we are uninterested in memory differences for the inner disk and outer ring. Instead, we are interested in memory differences for the attended and ignored features. Consequently, we recoded each difference score to reflect whether it was for the attended or ignored color, regardless of whether the attended item was the ring or disk. An exploratory analysis suggested by a reviewer confirmed that there was little difference in the average error for the inner disk (m=45°, sd=52°) and the outer ring (m=41°, sd=56°; t[66]=0.38, p=.708, d = 0.06; where d=(X¯1X¯2)/(σ^12+σ^22)/2), suggesting that the visual eccentricity difference between the ring and disk likely was unimportant to task performance.

Consistent with Eitam et al.’s (2013) results, our participants demonstrated more precise memory for the attended color (m=28°, sd=43°) than the ignored color (m=58°, sd=59°; mdiff=30°, sd=60°), t(66)=4.10, p.001, d = 0.58). Overall, about two thirds of the participants (45 out of 67) were more accurate for the attended than the ignored color.

To examine whether response order led to greater errors when participants were asked to report the ignored color first, we separated the attended and ignored errors based on the testing order (see Figure 2). For the attended color, participants were equally accurate whether that color was tested first (m=29°, sd=49°) or second (m=27°, sd=39°), t(52.54)=0.12, p=.901, d = 0.03. In contrast, for the ignored color, participants were highly inaccurate when it was tested first (m=79°, sd=64°) and relatively accurate when it was tested second (m=30°, sd=37°), t(61.13)=3.99, p.001, d = 0.95.1

When the ignored color was tested second, participants were nearly as accurate as they were for the attended color. Only when the ignored color was tested first did they show worse memory for the ignored color. That pattern suggests that the poorer recall performance for the ignored color than the attended color might result from an order effect rather than from a difference in memory.

Figure 2.
Absolute error as a function of which color was tested and testing order in Study 1. Dots represent individual participants.
Figure 2.
Absolute error as a function of which color was tested and testing order in Study 1. Dots represent individual participants.
Close modal

Confirmatory Analyses - Confidence. Participants reported greater confidence in their recollection of the attended color (m=72, sd=27) than the ignored color (m=61, sd=29), t(66)=2.55, p=.013, d = 0.41. For the attended color, participants were similarly confident when it was tested first (m=75, sd=27) and second (m=70, sd=28), t(61.65)=0.77, p=.442, d = 0.19. For the ignored color, participants were somewhat more confident when it was tested first (m=69, sd=25) than second (m=50, sd=30), t(53.98)=2.66, p=.010, d = 0.66. That confidence difference might also reflect confusion. If participants mistakenly reported the attended color when asked about the ignored one, they might not realize their error until the second memory test appeared. If so, they would respond that they were confident in their accuracy, thinking that they had correctly reported the attended color. In contrast, the somewhat reduced confidence when the ignored color was tested second is consistent with weaker memory for the ignored color.

We replicated Eitam et al.’s (2013) finding of better memory for the attended than the ignored color, but the pattern was affected by question order. For the attended color, participants were equally accurate when tested first or second. For the ignored color, participants were inaccurate when tested first and relatively accurate when tested second. That difference in pattern is consistent with an alternative explanation: Participants remember the attended and ignored colors equally well, but they expect a test on the attended feature.

Exploratory Analyses. In a set of exploratory analyses, we further examined the possibility that participants mistakenly responded with the attended color when first asked about the ignored one. We first tallied how often the recalled hue was closer to the other target color. For example, when participants were asked to recall the color of the inner disk, how often was their response closer to the outer ring’s color than to the inner disk’s color?

In general, we would expect memory errors to be centered on the to-be-recalled color, with some spread around that. However, when participants were asked to report the ignored color first, their errors were approximately uniformly distributed, suggesting that their errors were unrelated to the actual hue of the ignored part of the display. Perhaps they were related to the attended color instead. If participants were responding with the attended hue rather than the ignored one, their “errors” might not be uniformly distributed if we computed the difference between their reported hue and the attended color. More specifically, if participants were reporting the ignored color as requested, the difference between their reported color and the attended color should be large and centered on the average difference between the attended and ignored hues in the actual display (approximately 135 degrees given that the two colors were randomly selected to be 90-180 degrees apart for each participant). However, if some participants were mistakenly reporting the attended color rather than the ignored color, then those participants would have a small difference between the reported color and the attended color. That would lead to a bimodal distribution, with some responses clustered near zero and others clustered near 135. Figure 3 plots the difference between the reported color and the attended hue when participants were asked to report the ignored color.

The pattern shown in the figure confirms the possibility that some participants were reporting the incorrect color when first asked to report the ignored color. When asked to report the ignored hue first, some responded with a hue close to that of the attended color and others reported a hue close to what would have been the ignored color. The bimodal pattern of errors suggests that some participants mistakenly reported the attended hue rather than the ignored one when they were asked first about the ignored color. In contrast, when asked about the ignored color second, responses were centered around what would have been the ignored color; when they first reported the attended color and only then reported the ignored one, they reported the intended hue for both.

Figure 3.
Difference between the reported color for the ignored item and the actual color of the attended item in Study 1. Dots represent individual participants.
Figure 3.
Difference between the reported color for the ignored item and the actual color of the attended item in Study 1. Dots represent individual participants.
Close modal

Of the 38 participants in Study 1 who were tested first on the ignored color, 18 responded with a hue that was closer to the attended color. In contrast, only 2 of 29 participants who were asked to report the ignored color second responded with a hue closer to the attended color. This pattern is consistent with confusion due to task demands; participants expected to be asked about the attended hue. When they were asked about the ignored hue first, they mistakenly reported the attended one.

For the 45 participants who showed no confusion on either response (i.e., their response to each color question was closer to the requested color than to the other color), the mean error for the ignored color (m=23°, sd=25°) was roughly comparable to the uncorrected error for the attended color (28°). The mean error for the ignored color also was greatly reduced from the 58° estimated before eliminating potentially confused responses. However, after eliminating data from participants who apparently responded to the attended color test with a hue closer to the ignored one, the attended advantage re-emerged (m=13°, sd=10°), t(44)=2.78, p=.008, d = 0.55. That is, after correcting for erroneous responses to both the attended and ignored color, participants were more precise for the attended color. That pattern might emerge if, after realizing that they had responded incorrectly on the first screen, they deliberately responded incorrectly on the second screen so that they recalled each of the colors once. After accounting for potentially confused responses to either color test, the size of the attended advantage was reduced, and performance for the ignored color was substantially more precise than it was without eliminating confused responses.

Experiment 2 presented stimuli simultaneously rather than sequentially to determine whether the attended advantage persists after eliminating response order as a potential source of confusion.

If the greater memory precision for the attended color resulted from confusion due to question order, that difference might be reduced or eliminated by testing both colors on the same screen. With that approach, participants should realize that they are being asked about both the inner disk and outer ring.

Power Analysis

We preregistered a target sample size of 60 or more participants based on the same sensitivity analysis as in Experiment 1. Note that our effect size in Study 1 corresponds to dz = 0.501, which is larger than the effect we calculated for Eitam et al.’s (2013) studies. With our target sample size of N=60, Study 2 would have approximately 99% power to detect the effect we observed in Study 1.

Method

This experiment was identical to Experiment 1 except that both color wheel memory tests appeared on the same screen rather than on separate screens. Whether the “inner” or “outer” recall task appeared on the left or right of the display was counterbalanced across participants (see Figure 4). The confidence rating sliders appeared together on a subsequent screen, with their left/right position matching the position of the corresponding color response from the prior screen. Participants (n=69, not including pilot participants) were recruited from the Psychology Subject Pool at the University of Illinois at Urbana-Champaign and participated in exchange for course credit.

Figure 4.
Schematic illustration of the task for Experiment 2.
Figure 4.
Schematic illustration of the task for Experiment 2.
Close modal

Results and Discussion

Confirmatory Analyses - Memory Precision. As in Experiment 1, participants showed lower error for the attended (m=40°, sd=51°) than the unattended color (m=56°, sd=55°; mdiff=16°, sd=46°), t(68)=2.85, p=.006, d = 0.30. And again, about two thirds of the participants (43 out of 69) were more accurate for the attended than the ignored color. Memory accuracy did not differ significantly as a function of left/right position for either the attended (mleft=41°, sd=49°; mright=38°, sd=54°; t[65.11]=0.22, p=.829, d = 0.05) or ignored color (mleft=47°, sd=54°; mright=63°, sd=56°; t[66.81]=1.18, p=.240, d = 0.29). This pattern replicates the original Eitam et al. (2013) result showing better memory for the attended than the ignored feature. And, it eliminates the confound that could have contributed to the “attended advantage” both in the original experiments and in our Experiment 1 (see Figure 5).

Figure 5.
Absolute error as a function of where in the display each color was tested in Study 2. Dots represent individual participants.
Figure 5.
Absolute error as a function of where in the display each color was tested in Study 2. Dots represent individual participants.
Close modal

A mixed-design ANOVA (Type-III SS) with Study (1 vs. 2) as a between-participants factor and Attended/Ignored as a within-participant factor revealed a non-significant interaction (F(1, 134)=2.34, MSE=1416.13, p=.128), showing that the difference in precision for attended and ignored colors did not differ significantly across the studies. Across both studies, errors were lower for the attended than the ignored feature (F(1, 134)=25.08, MSE=1416.13, p=<.001). Overall performance, averaging across attended and ignored colors, did not differ significantly between the studies (F(1, 134)=0.36, MSE=4100.65, p=.550).

Confirmatory Analyses - Confidence. Participants reported higher confidence in their recollection of the attended color (m=79, sd=25) than the ignored color (m=56, sd=31), t(68)=5.33, p.001, d = 0.82.

Exploratory Analysis. Although simultaneous testing eliminates the testing order confound in Experiment 1 and in the original study, some participants still responded in a way that might reflect response confusion. Across all participants, 9 swapped their responses, responding closer to the ignored hue for the attended color question and closer to the attended hue for the ignored color question.

If we look only at the 46 participants who showed no confusion on either response, the error for the ignored color (m=23°, sd=18°) was still greater than the error for the attended color (m=16°, sd=15°), t(45)=2.06, p=.046, d = 0.43, but the effect is reduced. Note, though, that excluding all participants who gave responses that were closer to the wrong color on either question is overly conservative. Imagine, for example, that the difference between the attended and ignored hue was 90°. In some cases, participants might have poor memory for the ignored hue, responding with a hue that was off by 50° in a random direction. In about half of those cases, their error would result in a selected color closer to the attended hue even though that was not due to response confusion. By treating those cases as confusion rather than error and excluding them from this analysis, we artificially increase the estimated memory precision. Moreover, if such errors are more likely for the ignored color than for the attended one, excluding them would artificially decrease the difference in memory precision between the attended and ignored colors. Even with this overly conservative procedure, attended colors were recalled more precisely than ignored ones.

To further test whether confusion contributed to the difference in memory precision for the attended and ignored features, Experiment 3 presented the response display in the same configuration as the original stimulus. That is, by adjusting the hue of the inner disk and outer ring, participants could attempt to reproduce the exact stimulus they saw. One concern with these reproduction tasks is the unequal spread of perceived colors along the color wheel. For example, the range of hues that appear red is larger than the range that appear green. Consequently, errors might be larger for participants that happen to be assigned a color in the green range, and if more of them happen to be assigned green for the ignored color, that might inflate the difference between conditions. For Experiment 3, we recruited a larger sample which should diminish the role of such randomization differences.

Power Analysis

A power analysis showed that a sample size of 175 participants would give us greater than 99% power to detect the effect size observed in Experiment 2 (dz = .34) with α = 0.05 (using the same paired t-test). We preregistered a plan to continue scheduling participants until we had data from at least 175 participants.

Method

The design of Experiment 3 was identical to Experiment 2 except that participants received visual feedback that helped them recreate the original stimulus and answer the question prompts. The inner disk was 100 pixels in diameter and the outer ring was 50 pixels thick. When reporting what they had seen, participants saw the inner and outer shapes arranged exactly as they had been during the initial presentation, and they used separate color wheels to adjust the inner and outer hues. The two color wheels appeared to the left and right of the central shape (with the position of the inner and outer control wheel randomized). Participants could use each color wheel to adjust the inner and outer hue until they were satisfied that they had reproduced what they had seen. Both the color wheels and the image that participants recreated were the same size as the original stimulus (see Figure 6).

Figure 6.
Schematic illustration of the task for Experiment 3.
Figure 6.
Schematic illustration of the task for Experiment 3.
Close modal

Participants were recruited from Prolific’s worldwide participant pool with the following constraints applied from Prolific’s pre-screening: (a) must be 18 or older, (b) must report normal or corrected-to-normal acuity, and (c) must report no problems seeing colors. The study took approximately 2 minutes and participants were compensated $0.40. Participants read an online description and consented to participate by entering the study (under University of Illinois IRB protocol #09441 which includes a waiver of signed consent for online studies). After pilot testing to ensure that our code functioned correctly, we tested a total of 247 participants. Although our pre-screening criteria required normal color vision, 61 participants either incorrectly reported the number for the Ishihara plate or did not report normal color vision when asked. After excluding data from these participants in accordance with our preregistered plan, the final sample included 186 participants.

Results and Discussion

Figure 7.
Absolute error as a function of where in the display the color wheel appeared in Study 3. Dots represent individual participants.
Figure 7.
Absolute error as a function of where in the display the color wheel appeared in Study 3. Dots represent individual participants.
Close modal

Confirmatory Analyses - Memory Precision. Overall, participants in Experiment 3 were substantially more accurate (and less variable) than in Experiments 1 and 2, likely because they could recall the inner and outer hue in the same configuration that they saw it. Despite this better recall performance, participants again showed lower error for the attended (m=19°, sd=24°) than the unattended color (m=33°, sd=39°; mdiff=14°, sd=36°), t(185)=5.23, p.001, d = 0.43 (Figure 7). And, as in the first two experiments, about two thirds of the participants (120 out of 186) were more accurate for the attended than the ignored color. Thus, Experiment 3 again replicates the original Eitam et al. (2013) result showing better memory for the attended than the ignored feature.

Figure 8.
Absolute error as a function of study and condition. Dots represent individual participants.
Figure 8.
Absolute error as a function of study and condition. Dots represent individual participants.
Close modal

A mixed-design ANOVA (Type-III SS) with Study (1, 2, and 3) as a between-participants factor and Attended/Ignored as a within-participant factor revealed two significant main effects and a significant interaction effect. Averaging across attended and ignored colors, overall performance differed significantly across the studies (F(2, 319)=12.21, MSE=2528.15, p=<.001), with the most accurate performance in Study 3 and the least accurate in Study 1. Consistent with all three studies showing an accuracy advantage for the attended over the ignored hue, the main effect of attended versus ignored was significant as well (F(1, 319)=52.29, MSE=977.09, p=<.001). Finally, the size of the difference between attended and ignored varied across studies (Study 1: M=30°, SD=60°; Study 2: M=16°, SD=46°; Study 3: M=14°, SD=36°), resulting in a significant interaction (F(2, 319)=3.27, MSE=977.09, p=.039).

Confirmatory Analyses - Confidence. Participants again reported higher confidence in their memory for the attended color (m=82, sd=23) than the ignored color (m=59, sd=32), t(185)=9.29, p.001, d = 0.83.

Our studies provide evidence that attended colors are remembered more precisely than are ignored colors, replicating and extending Eitam et al.’s (2013) finding that relied on forced-choice recognition performance. Even though the display contained only a single object composed of only two colors—well within the capacity of attention—attending to one color led to superior memory for that color. Across all studies, confidence ratings reflected an awareness of greater memory precision for the attended hue.

The memory advantage for the attended color varied with question order, raising the possibility that the original results (Eitam et al., 2013) arose from confusion about which color to recall; some participants reported the attended color when asked about the ignored color first. If participants failed to read the instructions carefully, they might have assumed we were asking about the attended color. Confirming our suspicion, Experiment 1 found that recall of the unattended color was significantly worse when participants were asked about the unattended color first. Experiment 2 examined whether the attended advantage would be eliminated by testing both the attended and ignored colors on the same display rather than sequentially. Again, we replicated the original pattern. Even with the simultaneous color-wheel presentation, though, a few participants still might have swapped their responses.

In our exploratory analyses for Experiments 1 and 2, we excluded participants whose responses to one or both questions were closer to the incorrect target color. The remaining participants showed a smaller advantage for the attended color and somewhat improved memory precision for the ignored color. However, this approach might be too conservative as it eliminates genuinely incorrect responses. These exploratory analyses highlight the need for care when inferring memory differences between conditions, as imprecise measurement and participant expectations might spuriously introduce effects.

In Experiment 1, we observed more errors when the questions were asked sequentially, suggesting that task confusion might have amplified the difference in precision for the attended and ignored color. In Experiment 2, where the response wheels were presented simultaneously to the left and right, it seems plausible that participants would respond using the left wheel first, even if it were the one for the unattended feature. If so, that might make the subsequent report for the attended color less accessible. Unfortunately, Experiment 2 did not record which wheel they used first, so we cannot be certain whether response order affected performance.

Experiment 3 was designed to eliminate response confusion by asking participants to recreate the original disk/ring stimulus while giving participants real-time feedback as they adjusted the color wheels. We again replicated the original result: Participants demonstrated greater memory precision for the attended color than the ignored color. The reduced effect size in Experiment 3 might have resulted from the online data collection, but it could also indicate that the results of Experiments 1 and 2 were partly affected by response confusion. Still the results of all three experiments were consistent with the original finding of a memory advantage for the attended color (Eitam et al., 2013).

Finally, it is possible that some participants ignored our instructions and looked at the entire figure instead of a part. Many of our participants performed with approximately equal precision for the attended and ignored color, a pattern consistent with that strategy and with the results of Eitam et al’s (2013) control experiment in which participants were instructed to focus on both colors. Our results are consistent with the possibility that a third of participants followed our selective-attention instructions, and they were the ones who performed substantially more accurately for the attended than the unattended feature (note the skewed distributions in Figure 8).

Constraints on Generality

Experiments 1 and 2 were conducted with undergraduate participants in a laboratory environment in which the displays were viewed from a fixed position using a chin rest. Experiment 3 tested participants online, which reduced the degree of control over viewing conditions and possible distractions. The slightly reduced size of the effect for Experiment 3 could have resulted from these differences in addition to the elimination of possible response confusion.

With a single-trial design, we had only one measure of memory for the attended and ignored feature for each participant, making our measure of memory precision noisy. Ideally, we could measure memory precision on multiple trials for both the attended and ignored features. But, with repeated testing, participants would know that the ignored feature would be tested, so the task would be one of divided attention rather than selective attention. It might be possible to use infrequent tests of the ignored feature rather than a single unexpected test, but that would not necessarily produce the same results given the difference in expectations and practice with the task. Prior studies have incorporated multiple trials prior to a “surprise” trial (Chen & Wyble, 2015a, 2015b; Shin & Ma, 2016; Swan et al., 2016). Our studies show that the advantage for attended over ignored objects does not depend on building up familiarity with and expectations for a particular stimulus over multiple trials.

Finally, participants might have expected that they should be less accurate for the unattended color, leading them to intentionally perform worse. Although we cannot eliminate such a “demand characteristic” explanation, we find it unlikely given that our results held across multiple types of responses (sequential presentation, simultaneous presentation with separate responses, and simultaneous presentation with feedback).

Conclusion

In sum, we replicated Eitam et al.’s (2013) finding of better memory for attended than ignored features even after controlling for a potential confound due to task expectations. Although the confound induced by testing memory sequentially inflated the difference in memory performance between the attended and ignored colors, eliminating that confound did not eliminate the effect. We also replaced the forced-choice measure of memory accuracy with a color recall task and demonstrated that attended colors were recalled with greater precision than ignored ones. Even for a single object in a simple display, people recalled attended features more precisely than ignored ones.

Studies 1 and 2 were completed as part of an undergraduate honors thesis by VH, advised by DJS. Study 3 was conducted later, in response to reviewer suggestions.

VH and DJS jointly designed and preregistered the studies. VH conducted Experiments 1 and 2, analyzed the data, and wrote the first draft of the manuscript as part of the undergraduate thesis. DJS provided feedback and guidance throughout. DJS and VH jointly revised the manuscript for publication. As part of that process, VH and DJS independently re-analyzed the data in R to verify that they obtained the same results for the primary analyses. (This re-analysis revealed that data from one participant had been mislabeled as part of the undergraduate thesis analyses, so the total sample size in Experiment 2 is 66 rather than 67 people.) DJS then converted the manuscript to RMarkdown format to ensure reproducibility and added additional exploratory (not preregistered) analyses.

The authors thank Frances Wang for her guidance as director of the undergraduate honors thesis program at Illinois and Jake Miller and Owen Haupt for assistance with programming the experiments in Python and JavaScript.

The authors do not have any conflicts of interest for the research presented in this article.

All three experiments were preregistered, and all materials, code, and data are available in the separate “Study” links under components at https://osf.io/pzavd/.

1.

For the condition in which the attended inner ring was the first queried, participants were unusually accurate for the ignored outer-ring color. That condition had the fewest participants, and the pattern was not replicated in Experiments 2 or 3. It most likely was a fluke, perhaps due to the random selection of color pairs for each participant.

Allaire, J., Xie, Y., McPherson, J., Luraschi, J., Ushey, K., Atkins, A., .., & Iannone, R. (2020). Rmarkdown: Dynamic documents for r. https://github.com/rstudio/rmarkdown
Auguie, B. (2017). gridExtra: Miscellaneous functions for “grid” graphics. https://CRAN.R-project.org/package=gridExtra
Aust, F., & Barth, M. (2020). papaja: Create APA manuscripts with R Markdown. https://github.com/crsh/papaja
Chen, H., & Wyble, B. (2015a). Amnesia for object attributes: Failure to report attended information that had just reached conscious awareness. Psychological Science, 26(2), 203–210. https://doi.org/10.1177/0956797614560648
Chen, H., & Wyble, B. (2015b). The location but not the attributes of visual cues are automatically encoded into working memory. Vision Research, 107, 76–85. https://doi.org/10.1016/j.visres.2014.11.010
Dowle, M., & Srinivasan, A. (2019). Data.table: Extension of ‘data.frame‘. https://CRAN.R-project.org/package=data.table
Eitam, B., Yeshurun, Y., & Hassan, K. (2013). Blinded by irrelevance: Pure irrelevance induced “blindness.” Journal of Experimental Psychology: Human Perception and Performance, 39(3), 611–615. https://doi.org/10.1037/a0032269
Forbes, S. H. (2020). PupillometryR: An r package for preparing and analysing pupillometry data. Journal of Open Source Software, 5(50), 2285. https://doi.org/10.21105/joss.02285
Lawrence, M. A. (2016). Ez: Easy analysis and visualization of factorial experiments. https://CRAN.R-project.org/package=ez
Mack, A., & Rock, I. (1998). Inattentional blindness. MIT Press. https://doi.org/10.7551/mitpress/3707.001.0001
Most, S. B., Scholl, B. J., Clifford, E. R., & Simons, D. J. (2005). What you see is what you set: Sustained inattentional blindness and the capture of awareness. Psychological Review, 112(1), 217–242. https://doi.org/10.1037/0033-295x.112.1.217
Navarro, D. (2015). Learning statistics with r: A tutorial for psychology students and other beginners (Version 0.5). University of Adelaide. http://ua.edu.au/ccs/teaching/lsr
Nee, D. E., & Jonides, J. (2013). Trisecting representational states in short-term memory. Frontiers in Human Neuroscience, 7, 796. https://doi.org/10.3389/fnhum.2013.00796
Neisser, U. (1976). Cognition and reality: Principles and implications of cognitive psychology. WH Freeman/Times Books/Henry Holt Co.
Oberauer, K. (2002). Access to information in working memory: Exploring the focus of attention. Journal of Experimental Psychology: Learning, Memory, and Cognition, 28(3), 411.
Persuh, M., Melara, R. D. (2016). Barack obama blindness (BOB): Absence of visual awareness to a single object. Frontiers in Human Neuroscience, 10, 118. https://doi.org/10.3389/fnhum.2016.00118
R Core Team. (2019a). R: A language and environment for statistical computing. R Foundation for Statistical Computing. https://www.R-project.org/
R Core Team. (2019b). R: A language and environment for statistical computing. R Foundation for Statistical Computing. https://www.R-project.org/
Shin, H., Ma, W. J. (2016). Crowdsourced single-trial probes of visual working memory for irrelevant features. Journal of Vision, 16(5), 10. https://doi.org/10.1167/16.5.10
Singmann, H., Bolker, B., Westfall, J., Aust, F., Ben-Shachar, M. S. (2020). Afex: Analysis of factorial experiments. https://CRAN.R-project.org/package=afex
Sleegers, W. (2020). Tidystats: Save output of statistical tests. https://CRAN.R-project.org/package=tidystats
Swan, G., Collins, J., Wyble, B. (2016). Memory for a single object has differently variable precisions for relevant and irrelevant features. Journal of Vision, 16(3), 32. https://doi.org/10.1167/16.3.32
Wickham, H., Averick, M., Bryan, J., Chang, W., McGowan, L. D., François, R., Grolemund, G., Hayes, A., Henry, L., Hester, J., Kuhn, M., Pedersen, T., Miller, E., Bache, S., Müller, K., Ooms, J., Robinson, D., Seidel, D., Spinu, V., … Yutani, H. (2019). Welcome to the tidyverse. Journal of Open Source Software, 4(43), 1686. https://doi.org/10.21105/joss.01686
Xie, Y. (2020). Knitr: A general-purpose package for dynamic report generation in r. https://yihui.org/knitr/
Yeshurun, Y., Carrasco, M. (1998). Attention improves or impairs visual performance by enhancing spatial resolution. Nature, 396(6706), 72–75. https://doi.org/10.1038/23936
Zhang, W., Luck, S. J. (2008). Discrete fixed-resolution representations in visual working memory. Nature, 453(7192), 233–235. https://doi.org/10.1038/nature06860
Zhu, H. (2019). kableExtra: Construct complex table with ’kable’ and pipe syntax. https://CRAN.R-project.org/package=kableExtra
This is an open access article distributed under the terms of the Creative Commons Attribution License (4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Supplementary data