Content uploaded by Chaoxiong Ye
Author content
All content in this area was uploaded by Chaoxiong Ye on Feb 07, 2023
Content may be subject to copyright.
1
A re-attended feature of visual information exists in working
memory
Ruyi Liu1+, Lijing Guo1+, Xiaoshu Lin2, Piia Astikainen2, Guanghao He1, Chaoxiong Ye1, 2, 3*
1 Institute of Brain and Psychological Sciences, Sichuan Normal University, Chengdu, China;
2 Department of Psychology, University of Jyvaskyla, Jyväskylä, Finland;
3 Faculty of Social Sciences, Tampere University, Tampere, Finland.
+ Ruyi Liu and Lijing Guo contributed equally to this work and should be considered as
co-first authors.
Author Note
This work was supported by grants from the National Natural Science Foundation of China
(no. 31700948 to C.Y.), and the Academy of Finland (no. 333649 to C.Y.). All the authors had
full independence from the funding sources. Correspondence should be addressed to
Chaoxiong Ye, Department of Psychology, University of Jyvaskyla, P.O. Box 35, 40014
Jyväskylä, Finland. E-mail: cxye1988@163.com.
2
Abstract: When guided by retro-cues, humans can flexibly focus their attention on a specific
representation in visual working memory (VWM) during the maintenance phase. These retro-cues
improve the memory of the attended information in a phenomenon called the retro-cue benefit, but
with a memory cost of unattended information. The cost has recently been proven reversible for
object-based representations, showing that the memory performance of an unattended object is
enhanced with refocused attention. However, reversibility of the unattended feature-based
representation impairment is unclear. In the present study, we looked for a representation of a
re-attended feature in VWM, and we then investigated the quality of its preservation. Across three
behavioral experiments, we used paradigms with different levels of probe precision, and we
manipulated the number of retro-cues (no-cue vs. single-cue vs. double-cue conditions). In the
double-cue condition, two cues were sequentially oriented to different features in a trial, allowing
the representation of an unattended feature in the first cue array to become re-attended and then
weighted in the probe array. We found that (a) the representation of a re-attended feature existed in
VWM and was enhanced by the refocused attention (Experiments 1 and 3); (b) this phenomenon
was not affected by our manipulations of stimulus-onset asynchrony between the cue and the
probe (Experiment 2); and (c) compared to constantly focused attention, the enhancement of
refocused attention on representations of a feature was weakened(Experiment 3). Our findings
provide evidence supporting a flexible maintenance process of representations guided by
feature-based attention.
Key words: double retro-cues, feature-based attention, refocused attention, visual working
memory
3
Introduction
As a temporary online platform, visual working memory (VWM) encodes, maintains, and
extracts visual information in a task-driven process (Geigerman et al., 2016; McCants et al., 2020).
Due to its profound relation with other cognitive functions (long-term memory: Fukuda &
Woodman, 2017; Hartshorne & Makovski, 2019; decision making: Schapiro et al., 2022), the
VWM and its mechanism have become a magnet for studies in the past two decades. The limited
capacity of VWM, usually 3-4 units (Luck & Vogel, 1997, 2013; Vogel & Awh, 2008), is
compensated by the flexible allocation of memory resources among the representations to perform
various tasks (Gao et al., 2011; Luria & Vogel, 2014; Shen et al., 2021).
Within the attention that guides the distribution of memory resources in VWM (Gazzaley &
Nobre, 2012; Ku, 2018; Ravizza et al., 2016), two categories of attention work differently. These
are the attention orienting to all features of a specific object and the attention orienting to a
specific feature (e.g., color, orientation) across objects, referred to as object-based attention and
feature-based attention, respectively (Backer et al., 2015; Greenberg et al., 2010; Ku, 2015), and
they function as two relatively separate attention systems. Experiments using event-related
potentials (ERPs) have previously revealed that participants were able to identify an updated color
with feature-based attention and then integrate that feature into a moving object with object-based
attention when they were asked to trace the motion of a certain object that might change in color
(Hopf et al., 2005; Schoenfeld et al., 2003). Depending on the ultimate task goal of gaining the
relevant information, the participants were able to make flexible choices regarding which kind of
attention to use, and when both were necessary, they decided the temporal order.
In contrast to object-based attention, feature-based attention can fail to preclude irrelevant
4
information due to the solid binding of the features of an object. This has been demonstrated by
some experiments in which attention is used to modulate the encoding process in VWM (Marshall
& Bays, 2013; Vogel et al., 2005; Ye et al., 2018). Vogel et al. (2005) provided an object-based cue
before the memory array to direct participants to focus their attention on red bars presented after
the cue (targets), whereas they were to ignore blue bars. The participants were asked to remember
the orientations of the targets. ERPs showed that participants with high VWM capacity were able
to filter the information about the task-irrelevant bars from the VWM. However, in a feature-based
attention study, Marshall and Bays (2013) told participants to remember a certain feature about the
objects before the memory materials were presented. They found that the irrelevant feature, bound
with the target feature to objects, was automatically stored in the VWM. In sum, these studies
indicate that object-based attention and feature-based attention, as relatively independent systems,
have different influences on filtering during the encoding phase in VWM.
Another significant phase in VWM that could be affected by attention is maintenance, which
has also evoked a lively interest among researchers (object-based attention: Griffin & Nobre, 2003;
Kuo et al., 2014; Landman et al., 2003; Li et al., 2021; Makovsik & Jiang, 2007; Matsukura et al.,
2007; Niklaus et al., 2019; Pertzov et al., 2013; Rerko et al., 2014; Schneider et al.,2017; Souza &
Oberauer, 2016; and feature-based attention: Hajonides et al., 2020; Niklaus et al., 2017; Park et
al., 2017; Ye et al., 2016, 2021). With a change detection task, Griffin and Nobre (2003) inserted
an object-based retro-cue during the maintenance interval (after the disappearance of memory
items). The participants were told that the memory item on the position to which the retro-cue was
oriented would be probed with a high probability. However, in invalid cue trials, the non-cued
item was finally probed. In neutral cue trials, the retro-cue concurrently pointed to all positions
5
where memory items used to exist in the memory array (i.e., the participants could not exercise
this cue to adjust the allocation of memory resources to each object-based representation). The
researchers found that the valid retro-cue improved task performance compared to the neutral cue
in a phenomenon termed the retro-cue benefit. Correspondingly, they observed a retro-cue cost, as
the performance of the non-cued item in invalid cue trials was worse than that of the probed item
in neutral cue trials. These results also illustrated that participants focused attention on the cued
item during the maintenance phase but left the non-cued item unattended. To recap, participants
can assign attention among object-based representations to a certain item, which induces a
modification of other VWM resource distributions, and thereby reaches a better task performance.
Research on the retro-cue benefit underwent a pivotal advance when some studies modified
the cue type to investigate feature-based attention (Niklaus et al., 2017; Park et al., 2017; Ye et al.,
2016). These researchers utilized retro-cues orienting to a specific feature across items, rather than
directing the participants’ attention to all features of a specific item; e.g., the participants were
cued to select the colors of all memory items in VWM while ignoring the stimulus orientations, or
vice versa. The results demonstrated a robust feature-based retro-cue benefit with a memory
impairment of the unattended (non-cued) feature, indicating that attention devoted to or shifted
from a feature during the maintenance phase could modulate the VWM performance of the
information. On balance, object-based attention and feature-based attention both enhanced the
memory of the target, but this was accompanied by an attenuated performance about the non-cued,
irrelevant, unattended representation.
Previous researchers have proposed some hypotheses regarding the mechanism of the
object-based retro-cue effect. For example, the removing hypothesis states that, in the maintenance
6
phase of VWM, the attended representation presumably gains more resources to resist the decay of
retention or the disruption from outer distractors and inner competition (Astle et al., 2012; Fu et al.,
2022; Gunseli, van Moorselaar et al., 2015; Kuo, et al., 2012; Souza et al., 2014). Because VWM
is a task-driven limited online platform, the participants may actively remove (or, at any rate,
partly remove) the unattended representations to guarantee resource accumulation on the attended
ones. Kuo et al. (2012) used ERPs and showed that object-based retro-cues reduced the amplitude
of the contralateral delay activity (CDA) to the memory items, reflecting a decreased quantity of
representations maintained in VWM. Their finding directly indicated that the removal of
unattended information from VWM incurs an object-based retro-cue cost.
Recent studies have found that the impairment of unattended object-based representation is
reversible when the attention shifts back to the unattended objects (Heuer & Schubö, 2016a;
Rerko & Oberauer, 2013; van Moorselaar et al., 2015). Rerko and Oberauer (2013) used a
modified object-based retro-cue task (Landman et al., 2003; LaRocque et al., 2015), controlling
the number of retro-cues (none vs. single vs. double) and the onset of a single cue (late vs. early).
In double-cue trials, two cues orienting to diverse items in the memory array were sequentially
presented during the maintenance interval. The second cue drove attention to the non-cued,
unattended item of the first cue. Therefore, the researchers could directly weigh the representation
of unattended items when made attended again (i.e., re-attended items). Apart from the single
retro-cue benefit, the results showed that single-cue and double-cue trial responses had equal
accuracy, illustrating that the re-attended item in VWM was strengthened by the second cue.
Therefore, a representation of the re-attended object existed in VWM and was enhanced by the
refocused attention. However, previous studies have dwelled on the object-based retro-cue cost.
7
Because the object-based and feature-based attentional selections in VWM during the maintenance
phase rely on different mechanisms (Hajonides et al., 2020; Heuer & Schubö, 2016b; Sone et al.,
2021), the reversibility of memory impairment of an unattended feature, i.e., the existence and
quality of a re-attended feature in VWM cannot be inferred simply from that of the re-attended
object.
In the present study, we used a double retro-cue task modified from the study of Rerko and
Oberauer (2013) to investigate whether a representation of the re-attended feature exists in VWM
and whether that representation could be strengthened by refocused attention on it. In Experiment
1, we applied a no-cue, a single-cue, and a double-cue condition in which one feature (color or
orientation) of two items was probed. All the feature-based retro-cues were valid except the first
cue in the double-cue condition, because the first cue conflicted with the valid second cue
orienting a different feature. We compared the memory performances of the re-attended feature in
the double-cue condition with the equally attended feature in the no-cue condition and the
constantly attended feature in the single-cue condition. Our assumption was that if the participants
performed worse in the double-cue trials than in the no-cue trials, the unattended feature or some
details of it has been removed from VWM and does not reoccur even when the feature is made
task-relevant with refocused attention. Otherwise, the re-attended feature is assumed to exist in
VWM, and its representation is enhanced by attention. Furthermore, if the performance in the
double-cue trials was worse than in the single-cue trials, the representation of the re-attended
feature receives less enhancement than that of the constantly attended feature from the
feature-based attention. Conversely, if the performance was with no difference between the
double-cue and single-cue conditions, then the re-attended feature is expected to show
8
enhancement to the same magnitude as the constantly attended one in VWM, comparable to the
findings of Rerko and Oberauer (2013) using object-based retro-cues. The patterns of expected
results are shown in Figure 1.
Figure 1: Examples of expected result patterns. A taller bar indicates a better memory performance
in the experiments. a) The representation of a re-attended feature probed in double-cue trials is
irreversibly removed from VWM, leaving an apparent defect in the performance of double-cue
trials compared to no-cue trials. In other expectations, the re-attended feature exists in VWM and
is enhanced with the refocused attention guided by the second cue, leading to better performance
in double-cue trials than in no-cue trials. A divergence lies in the magnitude of the enhancement,
showing that b) the performance of a re-attended feature might reach that of a constantly attended
feature in single-cue trials, c) or the impairment of shifted attention, compared to constant
attention, results in worse performance in the double-cue trials than in the single-cue trials.
Experiment 1: Examining the existence of a re-attended feature in VWM
In Experiment 1, we examined whether participants maintained a re-attended feature of two
colored, oriented bars in VWM. Here, we used a double retro-cue change detection task with
no-cue, single-cue, and double-cue arrangements. Since two cues hinted at diverse features, the
9
probed feature in the double-cue trials was unattended from the first cue but gained attention again
from the second cue; therefore, we directly observed the mnemonic performance of the
re-attended feature, and we compared it to the performance of a constantly attended feature in
single-cue trials and a baseline feature in no-cue ones.
Methods
Participants
Previous studies testing the feature-based retro-cue benefit (Niklaus et al., 2017; Park et al.,
2017; Ye et al., 2016) or using the double retro-cue task (Rerko & Oberauer, 2013; van Moorselaar
et al., 2015) have 12–37 participants. Therefore, we recruited 31 college students for Experiment 1
to ensure a sufficient sample size. All recruits had self-reported normal health, normal color vision,
and normal or corrected-to-normal visual acuity. An exclusion criterion for participation was a low
accuracy rate in the main task (over 2 standard deviations lower than the mean, which excluded 5
participants). Therefore, the final sample was 26 participants (right handedness, 23 females, mean
age 19.69 years, SD = ±1.29). All participants signed an informed consent form before the
experiment and were paid for their participation. The procedures of the experiment complied with
the Declaration of Helsinki (2008) and were approved by the ethical committee of Sichuan
Normal University.
Stimuli and procedure
Stimuli in the memory array were colored, oriented bars (length: 1.1°, width: 0.4°). A colored
square (1.2°× 1.2°) or a white bar (1.1°× 0.4°) separately occurred in the probe array of color
10
or orientation. The experiment was presented against a gray (RGB = 128, 128, 128) background
on a 21-inch LCD monitor (1280×768, 75 Hz), at a viewing distance of 60 cm.
As shown in Figure 2, the main procedure of Experiment 1 began with a fixation cross (0.2°)
centered in the screen and presented for 1000 ms. In the subsequent displays of a trial, the fixation
was present unless a cue or feedback arose. Two colored, oriented bars, located approximate 0.9°
to the left and right of the fixation, composed a memory array for 300 ms. The level of color and
orientation varied independently between the two bars, with the constraint that no two equal colors
or orientations were presented in the same trial. The memory array was followed by a VWM
retention period, which was divided into three intervals (750 ms, 1500 ms, 1500 ms, in sequence)
by two 400 ms cue arrays. In no-cue trials, nothing except the fixation occurred in both cue arrays;
in single-cue trials, only the first cue array contained a valid feature-based cue (Chinese character
“色” meaning color or “向” meaning orientation) at the center, with a fixation in the second one;
in double-cue trials, two cue arrays displayed different cues, engendering an invalid first cue. After
the third interval, a probe array was presented for 3,000 ms (or until the response). The item
(colored square or white bar) in the probe array occupied the same location as the probed item in
the memory array. The feature of the item in the probe array was identical to that in the memory
array in 50% of the trials, whereas in the other trials, the level of color or orientation changed into
another visibly different one, compared with the memory feature. Feedback, which was presented
for 1500 ms, occurred at the end of a trial. The word “right” (“正确” in Chinese) was shown in the
screen if participants responded correctly; otherwise, the word “wrong” (“错误” in Chinese) was
shown.
Participants were directed to memorize the stimuli and judge whether the feature in the probe
11
array was identical to that in the memory array; they were instructed to press the key “F” if they
saw no change and to press the key “J” if a change had occurred. We stressed to the participants
that all cues could be regarded as valid, and no more than two feature-based cues might appear
during the interval to help them with the memory task. The participants were encouraged to use
the retro-cues to select the VWM representations of the corresponding feature shared in both
memory bars before the probe array. Accuracy, rather than response speed, was stressed.
64 trials were run for each cue type, yielding a total of 192 trials, after an initial completion
of 13 practice trials. Orientation and color were probed equally often across trials. The
experimental factor of cue type (no-cue vs. single-cue vs. double-cue) was randomly mixed.
Participants were allowed to rest every 11 minutes in the normal experiment, and the entire
duration of Experiment 1 was approximately 50 minutes.
Figure 2: (a) The procedure for changed trials probing the color in the three cue type conditions. (b)
Examples of color and orientation probe arrays. (c) The time procedure for three cue type
conditions.
Data analysis
12
The accuracy (ACC) was used as an index of sensitivity to change detection and was entered
in a repeated-measures analysis of variance (ANOVA), with the within-subjects factor cue type
(no-cue vs. single-cue vs. double-cue). Paired samples t-tests were conducted for the follow-up
pairwise comparison among three cue-type trials. JASP (version 0.16, JASP Team, 2021) was used
to provide Cohen’s d, estimating the effect size for the t-tests, and Bayes factors, showing whether
the t-test results supported the alternative hypothesis (Rouder et al., 2009; Schmalz et al., 2021),
thereby providing an odds ratio for the alternative/null hypotheses (values <1 favor the null
hypothesis and values >1 favor the alternative hypothesis).
Transparency and openness
We report how we determined our sample size, all data exclusions, all manipulations, and all
measures in Experiment 1 (and so are those in Experiments 2 and 3), and we follow JARS (Kazak,
2018). The datasets generated/analyzed during this study, experimental scripts and materials are
made available online via the Open Science Framework and can be accessed at
https://osf.io/vmsqa/.
Results
The results of the one-way repeated measures ANOVA revealed a significant main effect of
cue type, F (2, 50) = 3.771, p = 0.038, ηp2 = 0.13. Planned comparisons (Figure 3) revealed
slightly higher participant accuracy rates in the single-cue trials than in the no-cue trials, with a
marginal significant difference, t (25) = 1.67, p = 0.054, Cohen’s d = 0.33, BF10 = 1.32.
Furthermore, the memory performance was better for the double-cue condition than for the no-cue
13
condition, t (25) = 2.33, p = 0.028, Cohen’s d = 0.46, BF10 = 1.98, whereas it did not differ for the
single-cue condition, t (25) = 1.27, p = 0.215, Cohen’s d = 0.25, BF10 = 0.43.
Figure 3: Accuracy results of Experiment 1. Error bars indicate SE. * = p < 0.050, + =0.05 < p <
0.010.
Discussion
In Experiment 1, we acquired the feature-based retro-cue benefit, showing that the task
performance was better for the single-cue condition than for the no-cue condition, in agreement
with previous studies (Niklaus et al., 2017; Park et al., 2017; Ye et al., 2016).
Furthermore, we found a better VWM performance for the re-cued feature in a double-cue
trial than for the feature in a trial without any cue throughout, indicating that the representation of
an unattended feature in the first cue array occurred in VWM after it was made relevant and was
strengthened with the refocused attention oriented to the representation after the second cue. We
also observed comparable memory performances for the re-attended feature and the constantly
attended feature, illustrating that the enhancement of feature-based attention on VWM
14
representations was not influenced by the attention shift times. Therefore, similar to the
re-attended object (Rerko & Oberauer, 2013), the representation of a re-attended feature existed in
VWM and was cemented there to facilitate accomplishment of the task.
To obtain a high average accuracy in the whole task, the participants might tactically use one
cue and neglect the other. Because all cue-type trials were randomly present in Experiment 1, the
participants could not decide the validity of the first cue until they saw the probe array (in
single-cue trials) or the second cue (in double-cue trials), thereby excluding the possibility that the
participants anticipated the cue number and only used the valid cue. Moreover, the retro-cue
benefit in our study demonstrated that the first cue was used, and the better memory performance
in double-cue trials than in no-cue trials demonstrated that the second cue was also used.
Therefore, participants adjusted their attention with the retro-cues, and we confirmed the stability
of this paradigm to test the re-attended information.
However, in contrast to our comparable memory performances seen for the re-attended
feature and the constantly attended feature, previous researchers have found varying degrees of the
“shift cost”, showing that the re-attended object in VWM could not be enhanced by attention
being shifted back to the quality of the constantly attended object (Heuer & Schubö, 2016a; Rerko
& Oberauer, 2013; van Moorselaar et al., 2015). To date, studies have suggested that time is
required to make full and efficient use of the object-based retro-cue in single cue tasks (Schneider
et al., 2016). Moreover, Rerko and Oberauer (2013) noticed that the reaction time in double-cue
trials was shorter when they increased the cue-to-probe stimulus-onset asynchrony (SOA), and
they inferred that the temporal factor would also impact the use of a second object-based retro-cue.
Therefore, the use time for the second feature-based retro-cue (i.e., the time for attention
15
refocusing) might also impact its enhancement on VWM. Due to our relatively low memory load
(evidenced by the high average accuracy over 84%), the shift cost of the re-attended feature might
have been demonstrated as longer reaction times for correct trials with double cues than with a
single cue. Nevertheless, we stressed the accuracy rate rather than the reaction time in instruction
of Experiment 1, because the latter is susceptible to decision and other cognition processes.
Therefore, our reaction time results in Experiment 1 were unreliable. In addition, the cue-to-probe
SOA (1900 ms) might be too long to cover the larger time demand for the attention shift guided by
the second retro-cue and could therefore conceal the difference in reaction time between
double-cue and single-cue trials.
In Experiment 2, we manipulated the cue-to-probe SOA to examine the impact of time on
feature-based attention refocusing. Apart from the long SOA of 1900 ms, we chose another two
shorter levels of SOA. Previous researchers found an increase in the retro-cue benefit with an
extension of the cue-to-probe SOA. They observed a transition at approximately a cue-to-probe
SOA of 600 ms (Schneider et al., 2016; van Moorselaar et al., 2015). Considering our more
sophisticated semantic cue compared to the arrow cue applied in the study by Schneider et al.
(2016), we regulated the medium delay in Experiment 2 to 800 ms to provide adequate time for
the participants to identify the retro-cue. We also installed a short SOA of 500 ms, containing an
extremely short post-cue interval of only 100 ms. If the participants performed in different
patterns between double-cue and single-cue conditions, showing that their performance did not
differ in single-cue trials with different levels of SOA, but was worse in double-cue trials with
short SOAs than with the longer ones, we would assume that the use time of the retro-cues was
having a different influence on the VWM enhancement of feature-based constant attention and
16
refocused attention. Thereby, the shift cost was inapparent in Experiment 1 presumably because
the long SOA conceals this difference. Otherwise, if the performances were the same for all three
SOA conditions in double-cue trials, then the attention shift guided by the second retro-cue would
be expected to be so quick that even the short SOA we set would be sufficient to make the
representation of the unattended feature return to VWM and be enhanced. This would challenge
the account that the shift cost in reaction time was hidden in Experiment 1.
Experiment 2: Examining the impact of temporal factor on retro-cue use
Comparison of the memory performances in the double-cue vs. the no-cue and single-cue
conditions in Experiment 1 revealed that the second cue enhanced the memory performance of
oriented feature to the level of the constantly oriented feature. This indicated that the
representation of the re-attended feature existed in VWM and was enhanced by the refocused
attention but without the shift cost. The shift cost might be reflected by the different use times of
the first and second retro-cues, which failed to be probed in Experiment 1 with the long SOA.
In Experiment 2, we used a change detection task with double retro-cue and we manipulated
the cue-to-probe SOA (500 ms for short SOA vs. 800 ms for medium SOA vs. 1900 ms for long
SOA) to examine the impact of time on feature-based retro-cue use. Because the reaction time
might be influenced by the decision strategy, we recorded the accuracy rate of the participants. We
also added single retro-cue trials with the same SOA conditions and randomly mixed trials of all
cue-type and SOA combinations to compare the participants’ reaction patterns to the first and
second cues. In contrast to Experiment 1, we did not run no-cue trials, as they were unnecessary
17
given the aim of Experiment 2. This had the added benefit of reducing the entire duration of the
experiment and mitigating participants’ fatigue.
Methods
Participants
We recruited a new sample of 24 college students with self-reported normal health, normal
color vision, and normal or corrected-to-normal visual acuity for Experiment 2. Two students were
excluded due to missing values in more than 10% of total trials, leaving 22 participants (right
handedness, 18 females, mean age 19.73 years, SD = ±1.35) for further analyses. All participants
signed an informed consent form before the experiment and were paid for their participation. The
procedures of the experiment complied with the Declaration of Helsinki (2008) and were
approved by the ethical committee of Sichuan Normal University.
Stimuli and procedure
All stimuli and apparatuses in Experiment 2 were identical to those in Experiment 1.
The procedure of Experiment 2 was modified from that used in Experiment 1. We removed
the no-cue condition and set three cue-probe SOA conditions (as shown in Figure 4). Other
arrangements of stimulus presentation times, the memory load, the validness of cues, the
participant’s task, etc., were identical to those in Experiment 1. Accuracy was again stressed,
rather than response speed.
Each cue-probe SOA was presented for 96 trials with single cue and 96 trials with double
cues, yielding a total of 576 trials. The normal experiment was conducted after an initial
18
completion of 18 practice trials. Overall, 50% trials of each condition probed color and the other
half probed orientation. The experimental factors of cue type (single-cue vs. double-cue) and SOA
(short vs. medium vs. long) were randomly mixed within the blocks. Participants were allowed to
rest every 8 minutes in the normal experiment, and the entire duration of Experiment 2 was
approximately 1 hour and 20 minutes.
Figure 4: The time procedure of Experiment 2.
Data analysis
Separate repeated measures ANOVAs for double-cue and single-cue trials, with
within-subject factor cue-to-probe SOA (short vs. medium vs. long), were applied to the accuracy
rate. Paired samples t-tests were conducted for the follow-up pairwise comparison among different
SOA conditions. JASP (version 0.16, JASP Team, 2021) was used to provide Cohen’s d,
estimating the effect size for all the t-tests, and Bayes factors, showing whether the t-test results
supported the alternative hypothesis.
19
Results
Double-cue condition
The result of the one-way repeated measures ANOVA for the double-cue condition revealed
no significant main effect of SOA, F (2, 42) = 1.455, p = 0.245, ηp2 = 0.07. Planned comparisons
(Figure 5) revealed that no significant difference between the accuracy rate of the short SOA (500
ms) condition and the medium SOA (800 ms) condition, t (21) = 1.44, p = 0.491, Cohen’s d = 0.31,
BF10 = 0.55, or the long SOA (1900 ms) condition, t (21) = 1.39, p = 0.537, Cohen’s d = 0.30,
BF10 = 0.52. In addition, the accuracy rates of the medium SOA and long SOA conditions showed
no significant differences, t (21) = 0.05, p = 1.000, Cohen’s d = 0.01, BF10 = 0.22.
Single-cue condition
The result of the one-way repeated measures ANOVA for the single-cue condition revealed
no significant main effect of SOA, F (2, 42) = 2.003, p = 0.148, ηp2 = 0.09. Planned comparisons
(Figure 5) revealed no significant differences between the accuracy rates of the short SOA
condition and the medium SOA condition, t (21) = 1.07, p = 0.893, Cohen’s d = 0.23, BF10 = 0.37,
or the long SOA condition, t (21) = 1.89, p = 0.216, Cohen’s d = 0.40, BF10 = 1.01. In addition, the
accuracy rates of the medium SOA and long SOA conditions showed no significant differences, t
(21) = 0.99, p = 1.000, Cohen’s d = 0.21, BF10 = 0.35.
20
Figure 5: Accuracy results of Experiment 2. Error bars indicate SE. No statistically significant
differences were detected between the SOA conditions.
Discussion
We manipulated the cue-to-probe SOA in Experiment 2 to examine whether the efficiency of
retro-cues was contingent on their use time. However, the comparison among the three
arrangements of cue-to-probe SOAs within the double-cue condition revealed no significant
difference. Without any evidence for an influence of temporal factors on the second retro-cue use,
we expanded previous studies on the time course of the single retro-cue effect (Schneider et al.,
2016). This finding also suggested that a short SOA of 500 ms was still long enough for
participants to shift their attention to the unattended feature and enhance the representation of the
re-attended feature. No significant difference was detected when we installed different
cue-to-probe SOAs in the single-cue condition, confirming the study by Park et al. (2017). We
found comparable performance patterns in the double-cue and single-cue conditions and therefore,
no significant differences in speed between the attention shift guided by the second cue and the
21
attention allocation directed by the first cue. This finding challenged the account that the shift cost
of refocused attention should have been probed as a longer use time of the second cue than of the
first cue but was concealed by the adequate interval duration before the probe array in Experiment
1.
Another possibility for the absence of a degraded quality of the re-attended feature-based
representations compared to the constantly attended ones is that previous researchers observing
the shift cost used the recall paradigm, which could check the detailed quality of representations in
VWM. We cannot discern whether the varying effects of refocused attention in the double-cue
task are attributed to different mechanisms underlying the feature-based and object-based
representation or to the differences in probe precision between the change detection paradigm in
Experiments 1 & 2 of our study and the recall paradigm in previous studies. Therefore, in
Experiment 3, we used a recall task with three cue-type conditions (no-cue vs. single-cue vs.
double-cue) to examine the quality of the representations of a re-attended feature. We predicted
that we would see a better performance in double-cue trials than in no-cue trials, aligning with the
results in Experiment 1. If a shift cost occurred in the recall task, the performance is expected to
be worse in double-cue trials than in single-cue trials. Conversely, if participants performed no
differently in double-cue and single-cue trials, this would erode the explanation that probe
precision accounts for the absence of a shift cost.
Experiment 3: Examining the quality of the representations of a re-attended
feature
In Experiment 1, we used a change detection task and adjusted the number of retro-cues to
22
test the existence of a re-attended feature in VWM. Our results suggested that the re-attended
feature was not irreversibly removed from VWM in this task and was strengthened with refocused
attention to a level that was even comparable with the representation of a constantly attended
feature. However, previous researchers (Landman et al., 2003; van Moorselaar et al., 2015) have
observed a shift cost in refocused attention, which might be unavailable for probing in the change
detection paradigm. We found only a marginally significant difference between the accuracy rates
in single-cue and no-cue trials in Experiment 1, which diverged from the feature-based retro-cue
benefit evident in previous studies using the recall task (Niklaus et al., 2017; Park et al., 2017; Ye
et al., 2016). This divergent paradigm might expose a nuance of VWM representations obscured
by the low-precision change detection task. Therefore, we used a recall task with higher probe
precision in Experiment 3 to examine whether the enhancement of a retro-cue on a specific feature
in VWM was affected by the shift time of attention (i.e., whether the VWM performance of the
re-attended feature and the constantly attended feature would show differences).
We used a neutral cue (Chinese character “全” meaning all) in no-cue trials to motivate
participants to guide attention to both features of memory bars, and we added a late-cue condition,
with a neutral cue and a feature-based cue sequentially presented, to prevent participants from
assuming our cue-type condition when they saw the neutral cue. To avoid tiring the participants
with the task and having them make mistakes that would interfere with the measurement of
memory for the target feature, we controlled the entire duration of Experiment 3 by reducing the
cue display time (Park et al., 2017). The slight cut in the time that participants used on cues did
not impact the shift of feature-based attention, as shown in Experiment 2; therefore, we also
curtailed the cue-to-probe SOA to 1100 ms. We also shortened the total maintenance duration of
23
representations in early-cue and no-cue trials to prevent the participants from shelving the first cue
and deciding whether to use it at the presumed time of the second cue (Landman et al., 2003).
The average accuracy rates of each cue-type condition were relatively high in Experiments 1
and 2 (> 0.84, as shown in Figure 3 and Figure 5); therefore, we raised the memory load to three
double-feature bars, comparable to the study by Niklaus et al. (2017), to avoid a potential ceiling
effect. We also attached masks following the memory array to interrupt the encoding when the
stimuli disappeared.
Methods
Participants
We recruited a new sample of 26 college students with self-reported normal health, normal
color vision, and normal or corrected-to-normal visual acuity for Experiment 3 to ensure a
sufficient sample. One student was excluded because the program crashed midway, leaving 25
participants (one left handed, 22 females, mean age 20.00 years, SD = ±1.29) for further
analyses. All participants signed an informed consent form before the experiment and were paid
for their participation. The procedures of the experiment complied with the Declaration of
Helsinki (2008) and were approved by the ethical committee of Sichuan Normal University.
Stimuli and procedure
We produced masks, each of which consisted of a colored, oriented bar randomly selected
from the stimulus pool and intertwined with three different bars, staying with the principle that the
features of four bars were in isometric degrees. Other stimuli and apparatuses in Experiment 3
24
were identical to those in Experiment 1.
As shown in Figure 6, the main procedure began with a fixation presented for 300 ms,
followed by a memory array, which lasted 500 ms, with bars shown on three of the four corners of
an invisible square approximate 0.9° to the fixation. The next display, with a duration of 100 ms,
embodied three masks on the identical location to the memory stimuli. The VWM retention period
began with a fixation interval, which occurred for 600 ms, preceding one or two combinations of a
cue array (lasting 250 ms) and an interval array (lasting 850 ms). In double-cue trials, different
feature-based cues successively occurred in two cue arrays. In late-cue trials, a neutral cue
occurred in the first cue array and a feature-based cue appeared in the second cue array. In other
trials, only one set of a cue and an interval appeared with the different cue-type arrangements, in
which a neutral cue existed in the no-cue trials, whereas a feature-based cue existed in the
early-cue trials. After the third interval, a probe array appeared, which consisted of a white square
hinting at the probed item. A colored circle appeared in trials probing color, whereas a white bar
appeared on center in trials probing orientation. Participants moved the mouse to select the proper
level of a feature. The probe array did not disappear until participants clicked the left button of the
mouse. Feedback showing the reproduction error (the absolute deviation between the reported
level and the original level of the target feature in degrees) occurred at the end of a trial and would
also appear until a mouse click. Accuracy was stressed rather than response speed.
Each cue type was presented for 100 trials probing color and 100 trials probing orientation,
yielding 800 trials. The experimental factors of cue type (no-cue vs. early-cue vs. late-cue vs.
double-cue) and probe type (color vs. orientation) were randomly mixed within the blocks.
Participants conducted 16 practice trials before the normal experiment. A rest was arranged every
25
10 minutes, and the entire duration of Experiment 3 was approximately 1 hour.
Figure 6: (a) The procedure for trials with no cue or an early cue (upper) and with a late cue or
double cues (lower). (b) Cue types in Experiment 3. Three types of cues (“all”, “color” or
“orientation”) pseudorandomly occurred during the arrays indicated by dark gray bars. Two cues
in the same trial were always different. (c) The probe array of color (left) and orientation (right).
Data analysis
We recorded the reproduction error to examine the quality of re-attended feature in VWM
(double-cue condition), compared with that of the attended feature (single-cue conditions) and
baseline feature (no-cue condition) in Experiment 3. Paired samples t-tests were conducted
between the reproduction errors in early-cue and late-cue trials, followed by two repeated
measures ANOVAs separately for color and orientation trials, with within-subject factor cue types
(no-cue vs. single-cue vs. double-cue). Paired samples t-tests were conducted for the follow-up
pairwise comparison among different cue-type trials. JASP (version 0.16, JASP Team, 2021) was
used to provide Cohen’s d, estimating the effect size for all the t-tests, and Bayes factors, showing
26
whether the t-test results supported the alternative hypothesis.
Results
Color
Because the performance was not significantly different in early-cue and late-cue trials, t (24)
= 0.04, p = 0.972, Cohen’s d = 0.01, BF10 = 0.21, we mixed the single-cue data to focus on the
number of retro-cues, which reflected the shift time of feature-based attention. Results of the
repeated measures ANOVA revealed no significant main effect of cue type, F (2, 48) = 3.093, p =
0.079, ηp2 = 0.114. However, planned comparisons (Figure 7) revealed that participants’
reproduction errors were significantly lower in the single-cue trials than in the no-cue trials, t (24)
= 2.29, p = 0.031, Cohen’s d = 0.46, BF10 = 1.87. The performance conditions did not differ
between the double-cue and no-cue conditions, t (24) = 0.78, p = 0.440, Cohen’s d = 0.16, BF10 =
0.28, but it was worse than the single-cue condition, t (24) = 2.89, p = 0.008, Cohen’s d = 0.58,
BF10 = 5.69.
Orientation
The performance between early-cue and late-cue trials also showed no significant difference,
t (24) = 1.38, p = 0.180, Cohen’s d = 0.28, BF10 = 0.49. We mixed the single-cue data in further
analyses. Results of the repeated measures ANOVA revealed a significant main effect of cue type,
F (2, 48) = 19.283, p < 0.001, ηp2 = 0.446. Planned comparisons (Figure 7) revealed that the
participants’ reproduction errors were significantly lower in the single-cue trials than in the no-cue
trials, t (24) = 5.65, p < 0.001, Cohen’s d = 1.13, BF10 = 2602.66. The performance was also better
27
in the double-cue condition than in the no-cue condition, t (24) = 3.81, p < 0.001, Cohen’s d =
0.76, BF10 = 40.38, but was worse than in the single-cue condition, t (24) = 2.14, p = 0.043,
Cohen’s d = 0.43, BF10 = 1.44.
Figure 7: Results of trials probing (a) color and (b) orientation in Experiment 2. Error bars reflect
SE. * = p < 0.050, ** = p < 0.010, *** = p < 0.001.
Discussion
In Experiment 3, the comparison of two single-cue conditions revealed no significant
difference for either color or orientation, indicating that when the maintenance duration was
within the effective time range (e.g., up to 4550 ms in Experiment 2) for a retro-cue, and the
cue-to-probe SOA was ample, the onset of the retro-cue would not influence the enhancement of
attention on the oriented representation in VWM (van Moorselaar et al., 2015). We also replicated
the feature-based retro-cue benefit, finding significant improvement for single-cue trials compared
to the baseline (no-cue trials) in both the color and orientation probe conditions, despite the shorter
maintenance duration in the no-cue trials. These findings suggested that the participant’s attention
was attracted to the oriented feature in the single-cue trials. Because of the unpredictable cue type
28
and the uncertainty of probe array onset, we could rule out the strategy of ignoring the first cue in
double-cue trials.
Based on the validity of the double-cue condition, we observed again that the performance of
a re-attended feature in double-cue trials was not worse than in the baseline, confirming our
findings in Experiment 1 through a task with higher probe precision. Hence, the VWM
impairment of the unattended feature was reversible when the feature became relevant again.
However, the effect of refocused attention showed some differences in the feature dimension. The
reproduction error in double-cue trials for re-attended color reached the baseline level, whereas the
reproduction error for re-attended orientation was smaller than the baseline level. The result of
re-attended color was in line with an analogous observation reported in the object-based study of
van Moorselaar et al. (2015). Besides the number of retro-cues, they adjusted the validity of the
cue to test the unattended object with defocused attention in invalid single-cue trials. In their
double-cue condition, their participants’ performances were comparable to those in no-cue
condition. Based on the single retro-cue cost found in their experiment, they suggested that the
unattended object was impaired when the attention shifted away and was compensated when the
attention shifted back. Therefore, considering the impaired nature of unattended color in the first
cue array, we could suggest that the refocused attention both strengthened the representations of
unattended orientation and unattended color despite the comparable final memory performances of
re-attended color in double-cue and no-cue trials.
We also found a shift cost—a prevailing worse performance for re-attended features of
different dimensions than that for constantly attended features—similar to results reported in
previous studies on object-based attention (Landman et al., 2003; van Moorselaar et al., 2015).
29
This finding confirmed the existence of the shift cost of feature-based attention and that, in
Experiments 1 and 2, it might have been concealed by the coarse probe precision of the change
detection tasks.
In summary, rather than irreversibly removed from the VWM, the representation of an
unattended feature existed in VWM with impaired enhancement by the refocused attention.
General discussion
In the current study, we operationalized the number of feature-based retro-cues in the change
detection task and the recall task to investigate the potential existence of an unattended feature in
VWM when it resumed task-relevant, and we examined the representation quality of this
re-attended feature. Of primary interest, we found that the VWM performance for a re-cued
feature in double-cue trials was no worse than a feature in no-cue trials, indicating that the
representations of a specific unattended feature was not thoroughly removed from VWM but the
memory impairment of a feature caused by decreased attention was instead reversible with
refocused attention.
Our results showed a feature-based retro-cue benefit, demonstrating that participants’
memory performances were enhanced by a single retro-cue above the baseline. Furthermore,
considering the stable retro-cue cost (i.e., the impairment of unattended feature-based
representations, accompanied by the retro-cue benefit in previous studies; Niklaus et al., 2017;
Park et al., 2017; Ye et al., 2016), despite the absence of better memory performance for
re-attended color in double-cue trials than in no-cue ones, our data suggested that the
30
representations of unattended feature (both color and orientation) were enhanced with refocused
attention guided by the second retro-cue. These results expand the findings about object-based
attention (Rerko & Oberauer, 2013) to the domain of feature-based attention.
The maintenance and enhancement of a re-attended feature in our study seemed compatible
with the expectation of re-attended object-based representation in the “activity-silent” short-term
retention account, as proposed by the synaptic theory of working memory and evidenced by
object-based studies (Mongillo et al., 2008; Stokes, 2015). According to the “activity-silent”
account, the unattended object could be passively represented and remain via the modulation of
synaptic weights in a rapid, transient manner from hundreds to thousands of milliseconds. Animal
studies have implied that changed synaptic weights, as a result of the influx of calcium in the
presynaptic terminal, will not be cleared unless the representation is no longer relevant (Watanabe
& Funahashi, 2014). When the presynaptic input returns, the corresponding cells fire again,
enhancing the certain object-based representation in VWM. Previous researchers (Barak et al.,
2010; Stokes et al., 2013; Watanabe & Funahashi, 2007, 2014) have mainly used non-human
subjects in their experiments, except for the study by Trübutschek et al. (2017), and they all
focused on object-based representation. We borrow the logic of the “activity-silent” account and
conjecture that the unattended feature was also able to be represented in a passive silent state. The
silent representations of the unattended feature were stored throughout each trial because, in our
study, trials of different cue-type conditions were mixed, and the participants could not judge the
usefulness of the unattended feature in the first cue array until the probe began. When the
retro-cue—either the first one or the second one—focused on a certain feature, the cue provided a
signal that made the oriented feature task-relevant again, and the corresponding representations
31
could change into the activity state.
Using a recall task in Experiment 3, we also found a shift cost, as the final enhancement
effect of refocused attention was inferior to that of constant attention, in agreement with the results
from previous studies on object-based attention (Landman et al., 2003; van Moorselaar et al.,
2015), but in contrast with our findings in Experiments 1 and 2. One possible explanation is that
the recall paradigm with higher probe precision, compared to the change detection paradigm,
revealed subtle differences between the refocused feature and the constantly attended one. Another
possibility is that the allocation of memory resources differs between memory loads in our
experiments. In Experiments 1 and 2 of our study, the memory load was relatively low,
presumably incurring abundant resources to allocate on the premise of guaranteeing basic
resources to maintain each feature. However, in Experiment 3, the use of three multi-feature
stimuli approached the limit of VWM (Vogel & Awh, 2008). Because nearly all VWM resources
were occupied, the focus of attention on one feature might induce an assembly of resources on a
certain feature, with hemorrhaging from the non-cued feature. Therefore, the representation of a
non-cued feature was vulnerable to successive interruption, and was impaired as time passed, so
that it finally could not reach the level of the constantly attended feature when attention and the
other resources shifted back.
An interesting difference was noted between the probe features when the participants drove
attention back to the unattended representations in the recall task, as the performance for
re-attended orientation was better than baseline (no-cue condition), whereas that for re-attended
color was identical to baseline. However, the result pattern among different cue-type conditions
revealed no discrepancy between color and orientation in the change detection task. The diverse
32
result patterns across the features in Experiment 1 and Experiment 3 might originate from the
different memory loads and precision requirement in the two experiments, and the diverse
grouping difficulties of color and orientation. The grouping of two or three random orientations,
which could decrease the memory load, was apparently easier and more possible than grouping
the same number of random colors. In Experiment 1, with two multi-feature stimuli in a change
detection task, the facilitation of grouping was not reflected in the results because participants
could simultaneously maintain coarse representations of two colors and two orientations (four
representations in total, not exceeding the VWM limit) with biased attention after they saw the
first cue, thereby resulting in no difference between the maintenance or enhancement of color and
orientation. However, in Experiment 3, with three multi-feature stimuli in a recall task, the
participants had to juggle six feature-based representations in VWM as precise as possible after
the first cue array, and grouping was necessary to mitigate the memory load. The different
possibilities for grouping random colors and orientations indicated that the mnemonic task for
orientations was easier. Even in the double-cue trials when resources were concentrated in three
color representations, leaving few resources for representations of unattended orientation, the
grouped orientation was maintained robustly and ultimately, with refocused attention, was
enhanced to the level of constantly attended orientations. However, in double-cue trials for color,
after the cued orientations occupied more resources, the resources allocated to the three
unattended independent colors were inadequate, resulting in impairment during this phase, and the
final memory performance was not as good as that of constantly attended colors.
Our study has a limitation that the property of behavior experiments inhibits us from
examining the “activity-silent” account. The different subjects (animals) and stimuli (objects) in
33
previous studies that showed the “activity-silent” account weaken the interpretation of this account
with respect to our results. We admit that the feature-based representations in our study differ from
the object-based representations (e.g., whether several features bounded to one object can be
independently represented in active and silent states). Future studies could examine the
“activity-silent” state of feature-based representations using functional magnetic resonance
imaging.
In conclusion, we found that the representation of a re-attended feature existed in VWM and
was enhanced by refocused attention, both for color and for orientation, in agreement with the
reversibility of VWM impairment of unattended information. This finding gives a glimpse at the
mechanism underlying the feature-based retro-cue benefit, showing that the irreversible disposal
of unattended feature from VWM is not a necessary cost. Our study also provides evidence
supporting a flexible and complex maintenance process of representations guided by feature-based
attention.
Reference
Astle, D. E., Summerfield, J., Griffin, I., & Nobre, A. C. (2012). Orienting attention to locations in
mental representations. Attention, Perception, & Psychophysics, 74(1), 146–162.
https://doi.org/10.3758/s13414-011-0218-3
Backer, K. C., Binns, M. A., & Alain, C. (2015). Neural Dynamics Underlying Attentional
Orienting to Auditory Representations in Short-Term Memory. The Journal of Neuroscience, 35(3),
1307–1318. https://doi.org/10.1523/JNEUROSCI.1487-14.2015
Barak, O., Tsodyks, M., & Romo, R. (2010). Neuronal Population Coding of Parametric Working
Memory. Journal of Neuroscience, 30(28), 9424–9430.
https://doi.org/10.1523/JNEUROSCI.1875-10.2010
Fu, X., Ye, C., Hu, Z., Li, Z., Liang, T., & Liu, Q. (2022). The impact of retro-cue validity on
working memory representation: Evidence from electroencephalograms. Biological Psychology,
170, 108320. https://doi.org/10.1016/j.biopsycho.2022.108320
Fukuda, K., & Woodman, G. F. (2017). Visual working memory buffers information retrieved
from visual long-term memory. Proceedings of the National Academy of Sciences, 114(20),
34
5306–5311. https://doi.org/10.1073/pnas.1617874114
Gao, T., Gao, Z., Li, J., Sun, Z., & Shen, M. (2011). The perceptual root of object-based storage:
An interactive model of perception and visual working memory. Journal of Experimental
Psychology: Human Perception and Performance, 37(6), 1803–1823.
https://doi.org/10.1037/a0025637
Gazzaley, A., & Nobre, A. C. (2012). Top-down modulation: Bridging selective attention and
working memory. Trends in Cognitive Sciences, 16(2), 129–135.
https://doi.org/10.1016/j.tics.2011.11.014
Geigerman, S., Verhaeghen, P., & Cerella, J. (2016). To bind or not to bind, that’s the wrong
question: Features and objects coexist in visual short-term memory. Acta Psychologica, 167,
45–51. https://doi.org/10.1016/j.actpsy.2016.04.004
Greenberg, A. S., Esterman, M., Wilson, D., Serences, J. T., & Yantis, S. (2010). Control of spatial
and feature-based attention in frontoparietal cortex. The Journal of Neuroscience: The Official
Journal of the Society for Neuroscience, 30(43), 14330–14339.
https://doi.org/10.1523/JNEUROSCI.4248-09.2010
Griffin, I. C., & Nobre, A. C. (2003). Orienting Attention to Locations in Internal Representations.
Journal of Cognitive Neuroscience, 15(8), 1176–1194.
https://doi.org/10.1162/089892903322598139
Gunseli, E., van Moorselaar, D., Meeter, M., & Olivers, C. N. L. (2015). The reliability of
retro-cues determines the fate of noncued visual working memory representations. Psychonomic
Bulletin & Review, 22(5), 1334–1341. https://doi.org/10.3758/s13423-014-0796-x
Hajonides, J. E., van Ede, F., Stokes, M. G., & Nobre, A. C. (2020). Comparing the prioritization
of items and feature-dimensions in visual working memory. Journal of Vision, 20(8), 25.
https://doi.org/10.1167/jov.20.8.25
Hartshorne, J. K., & Makovski, T. (2019). The effect of working memory maintenance on
long-term memory. Memory & Cognition, 47(4), 749–763.
https://doi.org/10.3758/s13421-019-00908-6
Heuer, A., & Schubö, A. (2016a). The Focus of Attention in Visual Working Memory: Protection
of Focused Representations and Its Individual Variation. PLOS ONE, 11(4), e0154228.
https://doi.org/10.1371/journal.pone.0154228
Heuer, A., & Schubö, A. (2016b). Feature-based and spatial attentional selection in visual working
memory. Memory & Cognition, 44(4), 621–632. https://doi.org/10.3758/s13421-015-0584-5
Hopf, J.-M., Schoenfeld, M. A., & Heinze, H.-J. (2005). The temporal flexibility of attentional
selection in the visual cortex. Current Opinion in Neurobiology, 15(2), 183–187.
https://doi.org/10.1016/j.conb.2005.03.008
Ku, Y. (2015). Feature-based and object-based attention orientation during short-term memory
maintenance. Journal of Neurophysiology, 114(6), 3036–3038.
https://doi.org/10.1152/jn.00342.2015
Ku, Y. (2018). Selective attention on representations in working memory: Cognitive and neural
mechanisms. Peerj, 6, e4585. https://doi.org/10.7717/peerj.4585
Kuo, B.-C., Stokes, M. G., Murray, A. M., & Nobre, A. C. (2014). Attention Biases Visual Activity
in Visual Short-term Memory. Journal of Cognitive Neuroscience, 26(7), 1377–1389.
https://doi.org/10.1162/jocn_a_00577
Kuo, B.-C., Stokes, M. G., & Nobre, A. C. (2012). Attention Modulates Maintenance of
35
Representations in Visual Short-term Memory. Journal of Cognitive Neuroscience, 24(1), 51–60.
https://doi.org/10.1162/jocn_a_00087
Landman, R., Spekreijse, H., & Lamme, V. A. F. (2003). Large capacity storage of integrated
objects before change blindness. Vision Research, 43(2), 149–164.
https://doi.org/10.1016/S0042-6989(02)00402-9
LaRocque, J. J., Eichenbaum, A. S., Starrett, M. J., Rose, N. S., Emrich, S. M., & Postle, B. R.
(2015). The short- and long-term fates of memory items retained outside the focus of attention.
Memory & Cognition, 43(3), 453–468. https://doi.org/10.3758/s13421-014-0486-y
Li, Z., Tong, M., Chen, S., & Qian, J. (2021). Effect of attentional selection on working memory
for depth in a retro-cueing paradigm. Memory & Cognition, 49(4), 747–757.
https://doi.org/10.3758/s13421-020-01123-4
Luck, S. J., & Vogel, E. K. (1997). The capacity of visual working memory for features and
conjunctions. Nature, 390(6657), 279–281. https://doi.org/10.1038/36846
Luck, S. J., & Vogel, E. K. (2013). Visual working memory capacity: From psychophysics and
neurobiology to individual differences. Trends in Cognitive Sciences, 17(8), 391–400.
https://doi.org/10.1016/j.tics.2013.06.006
Luria, R., & Vogel, E. K. (2014). Come Together, Right Now: Dynamic Overwriting of an
Object’s History through Common Fate. Journal of Cognitive Neuroscience, 26(8), 1819–1828.
https://doi.org/10.1162/jocn_a_00584
Makovsik, T., & Jiang, Y. V. (2007). Distributing versus focusing attention in visual short-term
memory. Psychonomic Bulletin & Review, 14(6), 1072–1078. https://doi.org/10.3758/bf03193093
Marshall, L., & Bays, P. M. (2013). Obligatory encoding of task-irrelevant features depletes
working memory resources. Journal of Vision, 13(2), 21–21. https://doi.org/10.1167/13.2.21
Matsukura, M., Luck, S. J., & Vecera, S. P. (2007). Attention effects during visual short-term
memory maintenance: Protection or prioritization? Perception & Psychophysics, 69(8),
1422–1434. https://doi.org/10.3758/BF03192957
McCants, C. W., Katus, T., & Eimer, M. (2020). Task goals modulate the activation of part-based
versus object-based representations in visual working memory. Cognitive Neuroscience, 11(1–2),
92–100. https://doi.org/10.1080/17588928.2019.1642864
Mongillo, G., Barak, O., & Tsodyks, M. (2008). Synaptic Theory of Working Memory. Science,
319(5869), 1543–1546. https://doi.org/10.1126/science.1150769
Niklaus, M., Nobre, A. C., & van Ede, F. (2017). Feature-based attentional weighting and
spreading in visual working memory. Scientific Reports, 7(1), 42384.
https://doi.org/10.1038/srep42384
Niklaus, M., Singmann, H., & Oberauer, K. (2019). Two distinct mechanisms of selection in
working memory: Additive last-item and retro-cue benefits. Cognition, 183, 282–302.
https://doi.org/10.1016/j.cognition.2018.11.015
Park, Y. E., Sy, J. L., Hong, S. W., & Tong, F. (2017). Reprioritization of Features of
Multidimensional Objects Stored in Visual Working Memory. Psychological Science, 28(12),
1773–1785. https://doi.org/10.1177/0956797617719949
Pertzov, Y., Bays, P. M., Joseph, S., & Husain, M. (2013). Rapid forgetting prevented by
retrospective attention cues. Journal of Experimental Psychology: Human Perception and
Performance, 39(5), 1224–1231. https://doi.org/10.1037/a0030947
Ravizza, S. M., Uitvlugt, M. G., & Hazeltine, E. (2016). Where to start? Bottom-up attention
36
improves working memory by determining encoding order. Journal of Experimental Psychology:
Human Perception and Performance, 42(12), 1959–1968. https://doi.org/10.1037/xhp0000275
Rerko, L., & Oberauer, K. (2013). Focused, unfocused, and defocused information in working
memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 39(4),
1075–1096. https://doi.org/10.1037/a0031172
Rerko, L., Souza, A. S., & Oberauer, K. (2014). Retro-cue benefits in working memory without
sustained focal attention. Memory & Cognition, 42(5), 712–728.
https://doi.org/10.3758/s13421-013-0392-8
Rouder, J. N., Speckman, P. L., Sun, D., Morey, R. D., & Iverson, G. (2009). Bayesian t tests for
accepting and rejecting the null hypothesis. Psychonomic Bulletin & Review, 16(2), 225–237.
https://doi.org/10.3758/PBR.16.2.225
Schapiro, K., Josic, K., Kilpatrick, Z. P., & Gold, J. (2022). Strategy-dependent effects of
working-memory limitations on human perceptual decision-making. Elife, 11, e73610.
https://doi.org/10.7554/eLife.73610
Schmalz, X., Biurrun Manresa, J., & Zhang, L. (2021). What is a Bayes factor? Psychological
Methods. https://doi.org/10.1037/met0000421
Schneider, D., Barth, A., & Wascher, E. (2017). On the contribution of motor planning to the
retroactive cuing benefit in working memory: Evidence by mu and beta oscillatory activity in the
EEG. NeuroImage, 162, 73–85. https://doi.org/10.1016/j.neuroimage.2017.08.057
Schneider, D., Mertes, C., & Wascher, E. (2016). The time course of visuo-spatial working
memory updating revealed by a retro-cuing paradigm. Scientific Reports, 6(1), 21442.
https://doi.org/10.1038/srep21442
Schoenfeld, M. A., Tempelmann, C., Martinez, A., Hopf, J.-M., Sattler, C., Heinze, H.-J., &
Hillyard, S. A. (2003). Dynamics of feature binding during object-selective attention. Proceedings
of the National Academy of Sciences, 100(20), 11806–11811.
https://doi.org/10.1073/pnas.1932820100
Shen, M., Chen, J., Yang, X., Dong, H., Chen, H., & Zhou, J. (2021). The storage mechanism of
dynamic relations in visual working memory. Cognition, 209, 104571.
https://doi.org/10.1016/j.cognition.2020.104571
Sone, H., Kang, M.-S., Li, A. Y., Tsubomi, H., & Fukuda, K. (2021). Simultaneous estimation
procedure reveals the object-based, but not space-based, dependence of visual working memory
representations. Cognition, 209, 104579. https://doi.org/10.1016/j.cognition.2020.104579
Souza, A. S., & Oberauer, K. (2016). In search of the focus of attention in working memory: 13
years of the retro-cue effect. Attention, Perception, & Psychophysics, 78(7), 1839–1860.
https://doi.org/10.3758/s13414-016-1108-5
Souza, A. S., Rerko, L., & Oberauer, K. (2014). Unloading and reloading working memory:
Attending to one item frees capacity. Journal of Experimental Psychology: Human Perception and
Performance, 40(3), 1237–1256. https://doi.org/10.1037/a0036331
Stokes, M. G. (2015). ‘Activity-silent’ working memory in prefrontal cortex: A dynamic coding
framework. Trends in Cognitive Sciences, 19(7), 394–405.
https://doi.org/10.1016/j.tics.2015.05.004
Stokes, M. G., Kusunoki, M., Sigala, N., Nili, H., Gaffan, D., & Duncan, J. (2013). Dynamic
Coding for Cognitive Control in Prefrontal Cortex. Neuron, 78(2), 364–375.
https://doi.org/10.1016/j.neuron.2013.01.039
37
Trübutschek, D., Marti, S., Ojeda, A., King, J.-R., Mi, Y., Tsodyks, M., & Dehaene, S. (2017). A
theory of working memory without consciousness or sustained activity. ELife, 6, e23871.
https://doi.org/10.7554/eLife.23871
van Moorselaar, D., Gunseli, E., Theeuwes, J., & N. L. Olivers, C. (2015). The time course of
protecting a visual memory representation from perceptual interference. Frontiers in Human
Neuroscience, 8. https://doi.org/10.3389/fnhum.2014.01053
van Moorselaar, D., Olivers, C. N. L., Theeuwes, J., Lamme, V. A. F., & Sligte, I. G. (2015).
Forgotten but not gone: Retro-cue costs and benefits in a double-cueing paradigm suggest multiple
states in visual short-term memory. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 41(6), 1755–1763. https://doi.org/10.1037/xlm0000124
Vogel, E. K., & Awh, E. (2008). How to Exploit Diversity for Scientific Gain: Using Individual
Differences to Constrain Cognitive Theory. Current Directions in Psychological Science, 17(2),
171–176. https://doi.org/10.1111/j.1467-8721.2008.00569.x
Vogel, E. K., McCollough, A. W., & Machizawa, M. G. (2005). Neural measures reveal individual
differences in controlling access to working memory. Nature, 438(7067), 500–503.
https://doi.org/10.1038/nature04171
Watanabe, K., & Funahashi, S. (2007). Prefrontal Delay-Period Activity Reflects the Decision
Process of a Saccade Direction during a Free-Choice ODR Task. Cerebral Cortex, 17(suppl 1),
i88–i100. https://doi.org/10.1093/cercor/bhm102
Watanabe, K., & Funahashi, S. (2014). Neural mechanisms of dual-task interference and cognitive
capacity limitation in the prefrontal cortex. Nature Neuroscience, 17(4), 601–611.
https://doi.org/10.1038/nn.3667
Ye, C., Hu, Z., Ristaniemi, T., Gendron, M., & Liu, Q. (2016). Retro-dimension-cue benefit in
visual working memory. Scientific Reports, 6(1), 35573. https://doi.org/10.1038/srep35573
Ye, C., Xu, Q., Liu, Q., Cong, F., Saariluoma, P., Ristaniemi, T., & Astikainen, P. (2018). The
impact of visual working memory capacity on the filtering efficiency of emotional face distractors.
Biological Psychology, 138, 63–72. https://doi.org/10.1016/j.biopsycho.2018.08.009
Ye, C., Xu, Q., Liu, X., Astikainen, P., Zhu, Y., Hu, Z., & Liu, Q. (2021). Individual differences in
working memory capacity are unrelated to the magnitudes of retrocue benefits. Scientific Reports,
11(1), 7258. https://doi.org/10.1038/s41598-021-86515-5
