Introduction

Everyday visual scenes are often complex and crowded, where many objects compete for visual attention and selection. To successfully interact in such environments, animals (including humans) often perform visual search to detect or identify relevant objects among distractors. Looking for a key on a messy desk, detecting predators, foraging for apples, and security screening at the airport are all examples of visual search. Visual search links what we do in our daily life to neural and behavioral mechanisms of the visual system and has implications in psychology, vision science, neuroscience, and ecology (Nakayama & Martini, 2011). A typical visual search paradigm involves searching for one target (e.g., ‘red’ circle, letter ‘T’ or a ‘tilted’ bar) among many distractors (e.g., ‘green’ circles, letter ‘Ls’ or ‘vertical’ bars). Often, different parameters related to distractors are manipulated such as distractor set size, distractor homogeneity, and distractor similarity to the target (Calder-Travis & Ma, 2020; Palmer, 1994; Wolfe, 2020). However, in real life there exists situations where multiple copies of a target may be present, and sometimes these targets might contain conflicting information (Adamo et al., 2021; Won & Jiang, 2013). For instance, an animal might have to search for multiple predators approaching from different directions or in a social setting a human might have to look for a facial expression amongst other similar or different expressions in the crowd (Won & Jiang, 2013). Another example is in radiology, where detecting one target can interfere with the search for subsequent targets, a phenomenon known as satisfaction-of-search or subsequent-search-misses (Adamo et al., 2021). Previous studies have sought to understand how redundant target information (e.g., multiple targets or features) impacts visual search performance (Egeth & Mordkoff, 1991; Estes & Taylor, 1966; Holmgren et al., 1974; Thornton & Gilden, 2007; van der Heijden, 1975).

Earlier studies investigated the role of redundant target information using either simple detection (i.e., reporting the presence or absence of a target) or go/no-go tasks. When participants are required to simply detect a pre-defined target, responses are faster and more accurate when there are two targets compared with when there is only one target (e.g., Corballis, 2002; Fischer & Miller, 2008; Giray & Ulrich, 1993; Miller, 1982; Miniussi et al., 1998; Mordkoff & Yantis, 1991; Raab, 1962). This performance benefit is generally referred to as a redundancy gain (also known as redundant target effects or redundant signal effects) because the second target produces a gain in performance even though it is redundant. The term redundancy gain has also been used when the number of targets in the display remains the same but the number of target defining features are manipulated (e.g., targets are defined as being either a certain shape or a certain color). In such studies, visual search performance improves when a target consists of more than one target defining feature compared with only one feature (Krummenacher et al., 2001, 2002; Miller, 1982). The two most popular categories of models proposed to explain redundancy gain are race models (Miller, 1982; Raab, 1962; Reuter-Lorenz et al., 1995) and coactivation models (Miller, 1982; Ulrich et al., 2007). On one hand, independent parallel race models postulate that signals from multiple targets (or feature dimensions) are processed independently so that target detection is determined by the signal that wins the race, which is lower than the average time for any single target (see Mordkoff & Yantis, 1993; Raab, 1962). In contrast, co-activation models propose that signals from each target or dimension are summated at a stage before the response and thereby reach the response threshold more quickly (Miller, 1982). However, the exact mechanisms underlying redundancy gain are still debated (Miller, 1982; Fischer & Miller, 2008; Mordkoff & Yantis, 1991). For instance, Giray and Ulrich (1993) showed that participants were not only faster but also responded with greater force when two targets were presented compared with one, supporting a motor coactivation hypothesis. An electroencephalographical study by Miniussi et al. (1998) revealed that peak latencies for event related potentials were earlier when two targets (bilateral) were presented compared with one target (unilateral), supporting a perceptual coactivation hypothesis. Furthermore, most of these models were used to explain redundancy gain in studies that primarily involved either target detection task without distractors or go/no-go task, wherein multiple targets had identical perceptual features and associated responses.

A redundancy gain has also been reported in feature singleton detection tasks wherein multiple feature dimensions are used to define a target (Krummenacher et al., 2001, 2002). For instance, Krummenacher et al. (2001) observed that reaction times in trials including a pop-out target defined by two features (orientation and color) were faster than those including a target defined individually by either feature. In various previous studies, a similar reaction time advantage has been observed for cases when two identical targets (i.e., redundant targets) are present or when targets are defined by intra-dimensional redundancy (Eriksen & Eriksen, 1974, 1979; Holmgren et al., 1974; Miller, 1982; van der Heijden et al., 1984). Redundancy gain has also been observed for detection of tumors in simulated X-ray images (Hebert et al., 2020). For instance, Hebert et al., (2020) showed that displaying multiple identical or similar images yields significantly lower false-negative rates. They suggested that the redundancy gain may reflect a combination of enhanced perception, an alteration in search procedure, and a change in the threshold for when to quit search.

Many of these previous studies demonstrating redundancy gains have primarily employed target detection tasks where both redundant target yield the same target present or absent response. Much less is known about how redundant targets defined by one feature (e.g., color) interact during visual search when fine-detail discrimination (e.g., cut-off side of the target) is required for response. The present study aimed to examine whether the redundancy gain observed in detection tasks also occurs during discrimination tasks when one vs. two copies of a target are present.

Previous studies have demonstrated that while pop-out detection is thought to suffice with distributed attention across a wide range of the visual field, pop-out discrimination requires focused attention to a stimulus to resolve a perceptual feature. Consequently, pop-out search performance differs as a function of distractor number for detection compared with discrimination task (Bravo & Nakayama, 1992; Maljkovic & Nakayama, 1994; McPeek et al., 1999; Nakayama, 1990; Song & Nakayama, 2006). For instance, Bravo and Nakayama (1992) revealed distinctive visual search patterns associated with target detection and discrimination. They presented an odd-colored diamond target among homogeneous colored distractors, where target and distractor colors were randomly switched from trial-to-trial. When participants were required to detect the presence or absence of a target, reaction times were relatively fast and did not vary with the number of distractors, demonstrating a characteristic flat slope. This result suggested that when there are salient perceptual differences between the target and distractors, a broad scope of distributed attention is sufficient for target detection. However, when participants were asked to discriminate a detailed feature of the odd-colored target such as a tiny cut-off corner side, search time decreased as the number of distractors increased. This diverging pattern of reaction times in detection and discrimination pop-out search tasks have been consistently reported in both humans and non-human primates (Kristjánsson et al., 2001; Nakayama & Mackeben, 1989; Song et al., 2008; Song & Nakayama, 2006). They proposed that the perceptual grouping process of segregating the odd-colored target from distractors is more efficient with larger numbers of homogenous distractors, leading to faster allocation of focused attention to the target (Julesz, 1986; Koch & Ullman,1985). Here, we modified Bravo and Nakayama’s paradigm to examine how varying the number of odd-colored targets (one vs. two) influences visual search in target detection and discrimination task.

In accord with prior work, we expected to observe a redundancy gain for two targets compared with one target during detection because more targets in the search display leads to faster detection (Giray & Ulrich, 1993; Krummenacher et al., 2001; Miller, 1982; Miniussi et al., 1998; Mordkoff & Yantis, 1991; Raab, 1962). However, we expect that discrimination task performance may differ based on the perceptual and attentional requirements of feature discrimination (Bravo & Nakayama, 1992; Flowers & Garner, 1971; Garner & Flowers, 1969; Schöpper et al., 2019) (Experiment 1), the strength of perceptual grouping (Experiment 2) and conflict between differential perceptual features or response activation associated with different targets (Experiment 3).

Experiment 1: How do two odd-colored targets affect visual detection and discrimination?

In Experiment 1, we examined how target detection and discrimination are impacted by the presence of two redundant odd-colored targets among homogenous distractors. Following Bravo and Nakayama (1992), we asked participants to perform both a detection task and a discrimination task. On a subset of trials, two odd-colored targets were present. Note that when we presented an additional odd-colored target with the same color, we maintained the total number of stimuli in the display constant, in accord with previous studies (e.g., Akyürek & Schubö, 2013; Eriksen & Eriksen, 1979; Krummenacher et al., 2001, 2002). To our knowledge, performance in detection and discrimination tasks in the context of redundancy gain including two targets has never been directly compared using the same display.

Methods

Participants

The sample size was predetermined on the basis of previous work examining redundancy gain (Grubert et al., 2011; Hebert et al., 2020; Krummenacher et al., 2001; Miniussi et al., 1998; Won & Jiang, 2013). The estimated power was greater than .95 with a sample size of 15.

Fifteen participants (seven female, mean age = 21 years) from the Brown University community volunteered to take part in this experiment for one hour in exchange for course credit or monetary compensation. All participants were right-handed and had normal or corrected to normal vision and normal color vision. They were naïve to the goals of the experiment. The protocol was approved by the Brown University Institutional Review Board.

Apparatus

Stimuli were displayed at 72 Hz on a ViewSonic G90fB monitor running Windows XP (19-inch display, 1,152 by 864 resolution). Eye position was measured using an EyeLink 1000 eye tracker (SR Research, Ottawa, Ontario, Canada).

Stimuli and procedure

Participants performed 3 blocks each of the detection (180 trials/block) and discrimination (120 trials/block) tasks. Three participants completed only two blocks of the detection task due to time constraints. The order of blocks alternated and was counterbalanced across participants. Each participant practiced a block of each task to start. In parallel with previous studies investigating attentional constraints on search (Bravo & Nakayama, 1992; Song et al., 2008; Song & Nakayama, 2006), we required participants to maintain fixation at a central fixation cross. It enables us, in turn, to minimize the impact of different eye movement strategies on visual search (e.g., serial vs. parallel) as well as to create more comparable contexts to examine differences between target detection and discrimination tasks.

Detection task (Fig. 1a, left column)

At the beginning of each trial a gray cross appeared at the center of the monitor. The cross subtended 0.5° by 0.5° and had a luminance of 5 cd/m2 presented against a black background of 0.03 cd/m2. Participants were instructed to fixate the cross throughout the trial. They initiated a trial by pressing a key (‘5’), which turned the cross white (26 cd/m2), and continued to hold it until they made a response. Once a trial began, after 500 ms, the stimulus array with six diamonds subtending 1° by 1° were displayed. On each target present trial, target color was randomly selected to be red or green (equiluminant at 29 cd/m2) with distractors presented in the other color. During target-absent trials all stimuli were displayed in the same color. Within a block there were an equal number of target-absent, one target (Fig. 1a, left top), and two-target trials (Fig. 1a, left bottom). Participants were asked to release the ‘5’ key and press an assigned key with the same finger to report whether any odd-colored target (defined as the color that appeared less) was present (‘8’) or absent (‘2’). Auditory feedback on response correctness was provided after each trial. Participants were instructed to respond as soon as they found the first target while being as accurate as possible. We note that this task is different from visual search tasks in subsequent-search-misses or satisfaction-of-search phenomenon where participants are expected to search all possible targets in the display (Adamo et al., 2021). We discarded trials in which participants released the ‘5’ key before the stimulus onset or failed to respond within 1,500 ms and repeated them later in the block.

Fig. 1
figure 1

Tasks and results of Experiment 1. a Representative displays. In both detection (left column) and discrimination (right column), one (top row) or two (bottom row) odd-colored targets were randomly presented among homogenously colored distractors. During detection, participants reported whether at least one odd-colored target was present or absent. Target-absent trials were also included, where all six stimuli were presented in the same color. During discrimination, participants reported whether the top or bottom corner was cut off from one odd-colored target. On two-target trials, each target shape was randomly selected, resulting in trials where targets were identical (same top or bottom cut-off corner) or opposite (one top and one bottom cut-off corner). Target color was randomly switched between red and green on each trial, with distractors presented in the other color. Stimuli positions were also randomized on each trial. b Mean reaction time as a function of the number of targets. Results from the detection task are plotted on the left side and results from the discrimination task are plotted on the right side. Performance in target-absent trials (only in detection) are presented in gray, one-target trials are presented in blue, and two-target trials are presented in purple. While reaction time decreased for two-target trials during detection, it increased for two-target trials during discrimination. Error bars represent the between-participants standard error of the mean (SEM)

The position of each stimulus was randomized within the following constraints: (1) stimuli had to be within a 10° by 10° invisible square surrounding the center of the screen, (2) stimuli could not appear within 1.0° of each other, (3) no stimuli were presented within 1.5° of the vertical midline of the display, (4) three stimuli were presented to the left and right of the vertical midline, and (5) when a second target was presented, the distance between both targets was randomly selected with equal probability to be 3°, 5°, or 7° to prevent anticipation of the second target location. If a participant blinked or moved their eyes further than 1° from the cross the trial was immediately discarded and replaced later in the block.

Discrimination task (Fig. 1a, right column)

The procedure was identical to the detection task except for the following. On each trial, one (Fig. 1a, right top) or two (Fig. 1a, right bottom) odd-colored targets of the same color were presented, the number of which was randomly selected with equal probability. Each diamond had a 0.25° corner cut off from the top or bottom that was randomly selected for each stimulus. Participants reported which corner was cut off from an odd-colored target by pressing the ‘8’ key (top corner cut off) or ‘2’ key (bottom corner cut off). They were instructed to report as soon as they found a target even if two were displayed. On two-target trials, we randomized the cut-off corner of each target (top or bottom) so that the two targets had either the same or different cut-off side with equal probability. This led to two trial types where the target shapes and potential responses were identical (Tsame) and opposite (Tdifferent).

Data analysis

For each participant, we excluded trials from data analysis where the reaction time was more than 3 standard deviations away from the mean of each condition. Using this criterion, we excluded an average of 1.7% ± 0.1% (standard error of the mean [SEM]) of detection trials and 1.4% ± 0.1% of discrimination trials from each participant. Reaction time was measured as the difference in time between the onset of the stimulus array and the subsequent press of the ‘2’ or ‘8’ key to report their response. When comparing reaction times, we excluded trials where the response was incorrect. We conducted repeated measure ANOVAs and applied Bonferroni correction for planned pairwise comparisons. Greenhouse-Geisser correction was applied whenever sphericity assumption was violated. Effect size was estimated using η2G and Cohen’s d. η2G is known to provide more conservative and reliable estimates for within-subject design as in the present study than η2 in which 0.01, 0.06 or 0.14 corresponds to a small, medium and large effect (Bakeman, 2005; Cohen,1988; Lakens, 2013; Olejnik & Algina, 2003). A Cohen’s d of 0.2, 0.5, 0.8 is considered a small, medium, and large effect (Cohen, 1988; Lakens, 2013).

Results and discussion

Detection task

Overall, participant accuracy was high in all conditions: target-absent (96.5% ± 0.6% SEM), one target (98.7% ± 0.3%), and two target (99.4% ± 0.2%). Accuracy was higher when at least one target was present compared with when there was no target. This was confirmed with a one-way repeated-measures ANOVA that revealed a significant main effect of number of targets (F1.19,16.68 = 20.16, p < .001, η2G = 0.379) and pairwise comparisons between target-absent and one- target trials (t14 = 5.04, p <.001, d = 1.3) and target-absent and two- target trials (t14 = 4.55, p < .001, d = 1.17). However, one-target and two-target conditions did not significantly differ from each other after correction for multiple comparisons (t14 = 2.31, p > .1, d = 0.6). On two-target trials, the distance between targets was randomly chosen to be 3°, 5°, or 7°. There was no main effect of distances in accuracy (F2,28 = 0.13, p = .8, η2G = 0.005).

When comparing reaction times, we excluded trials where the response was incorrect. Figure 1b (left) demonstrates that reaction times differed depending on the number of targets presented (absent, one, or two), which was confirmed by a one-way repeated-measures ANOVA (F1.18,16.52 = 45.24, p < .001, η2G = 0.168). Further planned analysis indicated that reaction time on target-absent trials (gray) was slower than on one-target present trials (blue; t14 = 5.44, p < .001, d = 1.4) consistent with previous visual search studies (e.g., Bravo & Nakayama, 1992; Treisman & Gelade, 1980). Of interest was whether two targets (purple) facilitate or deteriorate visual search. We observed faster reaction time in the two-target trials compared with one-target trials (blue; t14 = 4.65, p = .001, d = 1.20). This result is consistent with prior studies using a similar singleton pop-out detection task (e.g., Krummenacher et al., 2001, 2002, 2014; Töllner et al., 2011; Zehetleitner et al., 2009). Two-target trials (purple) were also faster than target-absent trials (gray; t14 = 8.46, p < .001, d = 2.18). On two-target trials, we observed a significant effect of distance (F2,28 = 6.54, p = .005, η2G = 0.010). This effect was driven by faster reaction times when targets were presented 7° apart (500 ± 14 ms) than 3° (512 ± 14 ms; t14 = 3.37, p = .014, d = 0.872). However, there was no significant difference between the 3° and 5° (511 ± 15 ms; t14 = 0.121, p > .9, d = 0.03), and between 5° and 7° (t14 = 2.56, p = .067, d = 0.663).

Discrimination task

We compared accuracy in one-target trials with two-target trials in which both targets share the same cut-off side (Tsame). We limited our analysis to these trials because when each target had a different cut-off side (Tdifferent), any of the two responses (‘top’ or ‘bottom’) would be correct, resulting in an inflated accuracy estimate. We observed significantly higher accuracy for one target (94.2% ± 1.7%) compared with Tsame trials (91.1% ± 1.9%; t14 = 3.22, p = .006, d = 0.833).

Figure 1b (right) shows the average reaction time for the one target and two target conditions in the discrimination task. Overall reaction time was slower in the two-target (purple; 696 ± 19 ms) than in the one-target condition (blue; 651 ± 15 ms), which was the opposite of what we observed in the detection task. To assess differences between conditions, we compared one-target against Tsame and Tdifferent trials in two-target condition and observed a significant effect (F1.43,20 = 16.26, p < .001, η2G = 0.147). One-target trials were significantly faster than both Tdifferent trials (721 ± 24 ms; t14 = 4.9, p < .001, d = 1.26) and Tsame trials (676 ± 15 ms; t14 = 3.26, p = .017, d = 0.843). For the two-target condition, one might think that the Tdifferent condition would be easier than the Tsame condition because participants would be correct with either of the two responses (‘top’ or ‘bottom’). However, reaction time in the Tdifferent condition was significantly slower compared with in the Tsame condition (676 ± 15 ms; t14 = 3.169, p = .020, d = 0.818).

We further assessed whether reaction time varied in two-target trials based on inter-target distance. To do so, we conducted an 3x2 repeated-measures ANOVA with factors distance (3°, 5°, or 7°), and target type (Tsame or Tdifferent). While the main effect of target type was significant (F1,14 = 8.9, p = .01, η2G = 0.059), neither distance (F2,28 = 2.15, p =.135, η2G = 0.013) nor the interaction (F2,28 = 0.37, p = .695, η2G = 0.002) was significant.

To summarize, in Experiment 1 we demonstrated that when more than one odd-colored target was present, target detection was facilitated. This result is consistent with prior studies that have demonstrated reaction time and accuracy gains when extra target stimuli are presented (Eriksen & Eriksen, 1979; Krummenacher et al., 2001, 2002; Miller, 1982). However, we also showed that target discrimination was hindered by an additional odd-colored target, resulting in a redundancy cost. In subsequent experiments, we further investigated what contributed to this inefficiency redundancy cost related during discrimination.

Here, we held the total number of stimuli constant at six so that there was always one less distractor present on two-target trials than one-target trials, following prior studies with two targets that kept a constant display size (Akyürek & Schubö, 2013; Eriksen & Eriksen, 1979; Krummenacher et al., 2001, 2002). That said, the one less distractor on two-target trials may have affected search efficiency during discrimination. As discussed earlier, previous studies with one target have shown that as the number of homogenous distractors increases, the strength of perceptual grouping due to distractors is enhanced, which leads to faster allocation of focused attention to a target (Bravo & Nakayama, 1992; Julesz, 1981, 1986; Koch & Ullman, 1985; McPeek et al., 1999; Song & Nakayama, 2006). Thus, increased perceptual grouping facilitates the efficiency of odd-colored target discrimination but does not affect detection during pop-out search (Bravo & Nakayama, 1992; Nakayama & Joseph, 1998). In Experiment 2, we assessed whether this one less distractor weakened perceptual grouping during discrimination, resulting in less efficient allocation of attention to a target and contributing to redundancy cost.

Experiment 2: Does perceptual grouping contribute to redundancy cost during discrimination?

Here, we examined whether the redundancy cost observed in Exp. 1 during target discrimination was led by weakened perceptual grouping during discrimination. Perceptual grouping is known to be a complex process that takes into account many aspects of stimuli, including proximity (Bacon & Egeth, 1991), shape (Duncan & Humphreys, 1989), color (Bundesen & Pedersen, 1983; Farmer & Taylor, 1980), and orientation (Julesz, 1981). Therefore, we attempted to equate the strength of perceptual grouping between one target and two target displays during discrimination by matching the targets to distractors ratio during discrimination. For example, at a target to distractors ratio of 1:2, displays would contain either one target and two distractors or two targets and four distractors. If an unequal strength of perceptual grouping between one target and two-target trials primarily contributed to the longer reaction times in Experiment 1, we expected to observe a diminished difference between one target and two-target trials.

Methods

Participants

Fifteen participants (nine female, mean age = 19.65 years) from the Brown University community volunteered to take part in this experiment for one hour in exchange for course credit or monetary compensation. All participants were right-handed and had normal or corrected to normal vision and normal color vision. They were naïve to the goals of the experiment. The protocol was approved by the Brown University Institutional Review Board.

Apparatus

The same apparatus was used as in Experiment 1.

Stimuli and procedure

The stimulus and task procedure were the same as in the discrimination task of Experiment 1 except for the following. On each trial the number of distractors varied randomly. With equal probability, one target was presented with 2, 3, 5, 10, or 14 distractors and two targets were presented with either 3, 4, 6, 10, or 13 distractors. Figure 2a shows this manipulation equated the targets to distractors ratio on a subset of one target and two-target trials. Specifically, targets to distractors ratios of 1:2, 1:3, and 1:5 were present during both one-target and two-target conditions, which consisted of displays containing one target with 2, 3, or 5 distractors (Fig. 2a, top row) or two targets with 4, 6, or 10 distractors (Fig. 2a, bottom row). Stimulus position was randomized under the following constraints (1) stimuli had to be within a 10° by 10° invisible square surrounding the center of the screen, (2) stimuli could not appear within 1.0° of each other, (3) no stimuli were presented within 1.5° of the vertical midline of the display. When two targets were present the distance between them was always 5°. In Experiment 1, we did not observe a significant effect of distance between targets in reaction time for two target trials during discrimination. Given that null effect, we affixed distance at 5 degrees as it was not a variable of interest. While the distance between targets was always the same, both the angular direction of the location of the second target relative to the first and the actual positions of both targets in the display were randomized. Moreover, participants were instructed to discriminate and report the shape of only one target, meaning that successful task performance did not require finding both targets. If a participant blinked or moved their eyes further than 1.25° from the cross the trial was discarded and replaced later in the block. Participants completed six blocks (90 trials/block) following a practice block. One participant completed only five blocks due to time constraints.

Fig. 2
figure 2

Task and results of Experiment 2. a Representative displays for equated targets to distractors ratios. Participants reported whether the top or bottom corner was cut off from one odd-colored target. Either one (top row) or two (bottom row) targets were presented on each trial. On two-target trials, each target shape was randomly selected, resulting in trials where targets were identical (Tsame) or opposite (Tdifferent). We manipulated the number of distractors to equate the targets to distractors ratio between one-target and two-target trials on a subset of trials. In each column here, we present example displays for each matched targets to distractors ratio, 1:2 (left), 1:3 (middle), and 1:5 (right). b Mean reaction time as a function of targets to distractors ratio. One-target trials are presented in blue and two-target trials are presented in purple. Reaction time did not differ between one-target and two-target trials when the targets to distractors ratio was equated. c Mean reaction time in two-target trials. Reaction time differed depending on whether targets were identical (Tsame, pink) or opposite (Tdifferent, dark purple) as well as the number of distractors increased. Error bars represent the between-participants standard error of the mean SEM)

Data analysis

For the comparison between one-target and two-target trials, only trials where the targets to distractors ratio was matched (1:2, 1:3 and 1:5; Fig. 2a) were included. Using the same exclusion criteria as in Experiment 1, an average of 1.1% ± 0.1% of one-target trials and 1.2% ± 0.2% of two-target trials per subject were excluded from this analysis. When comparing trials where the target shapes and potential responses were identical (Tsame) to when they were opposite (Tdifferent), trials from all numbers of stimuli used were included. Using the same exclusion criteria as in Experiment 1, an average of 1.3% ± 0.2% of Tsame trials and 1.1% ± 0.1% of Tdifferent trials per subject were excluded from analysis.

Results and discussion

Effect of perceptual grouping: One versus two-target discrimination

We first assessed the effects of the number of targets and perceptual grouping on accuracy. Tdifferent trials were not included for the accuracy analysis as in Experiment 1. We submitted the accuracy scores to a two-way repeated-measures ANOVA with factors number of targets (one vs. Tsame) and targets to distractors ratio (1:2, 1:3, and 1:5). We did not observe an overall difference between one target (94.5% ± 1.1%) and Tsame trials (94.2%± 1.2%; F1,14 = 0.05, p = .81, η2G < 0.001). However, the manipulation of targets to distractors ratio significantly affected accuracy (F2,28 = 4.50, p = .02, η2G = 0.042): 92.7% ± 1.0% (1:2), 95.1% ± 1.1% (1:3), and 95.1% ± 0.9% (1:5). There was no significant interaction between number of targets and targets to distractors ratio (F1.37,19.17 = 1.33, p =.27, η2G = 0.018). Altogether, these results suggest that discrimination accuracy increases as perceptual grouping gets stronger but is not affected by the number of targets when perceptual grouping is matched.

We next compared reaction times for correct trials with a two-way repeated-measures ANOVA with factors number of targets (one vs. two) and targets to distractors ratio (1:2, 1:3, and 1:5). As Fig. 2b demonstrates, reaction time decreased as the targets to distractors ratio decreased from 1:2 to 1:5. This result is consistent with prior studies demonstrating decreasing reaction times as the number of homogenous distractors increase (e.g., Bravo & Nakayama, 1992; Duncan & Humphreys, 1989; Song et al., 2008; Song & Nakayama, 2006). This decrease was confirmed by a significant main effect of targets to distractors ratio (F1.21,16.97 = 7.18, p = .012, η2G = 0.049). Post hoc pairwise comparisons with Bonferroni correction showed that reaction time were faster for 1:5 (676ms) condition compared with 1:3 (686ms; t14 = 2.73, p = .048, d = 0.70) and 1:2 (707ms; t14 = 3.16, p = .021, d = 0.81) conditions. However, we did not observe a significant difference between the one- target (blue markers) and two target conditions (purple markers; F1,14 = 2.53, p = .134, η2G = 0.005) and no interaction with the targets to distractors ratio (F2,28 = 2.08, p = .144, η2G = 0.006). These results suggest that when perceptual grouping was matched between one-target and two-target trials, reaction time was similar. Therefore, to some extent, redundancy cost during discrimination shown in Exp. 1 is determined by the strength of perceptual grouping.

Effect of same or different targets on two-target discrimination

Next, we shifted our focus to the two-target conditions to determine how both targets sharing the same cut-off side and potential response modulates performance as perceptual grouping increases by comparing Tsame and Tdifferent conditions.

We confirmed that participants overall performed the two target conditions well (Tsame: 93.9% ± 1%). Since the perceptual grouping was always equated between Tsame and Tdifferent trials, we included performance at each total number of distractors used during the two- target conditions (3, 4, 6, 10, or 13 distractors) rather than restricting our analysis to just the subset of conditions used to equate target to distractor ratio with the one-target condition.

A two-way repeated-measures ANOVA with factors target shape (Tsame vs. Tdifferent) and number of distractors (3, 4, 6, 10, or 13 distractors) was conducted on reaction time. There was a main effect of target shape with reaction time in Tsame trials (overall mean: 681 ± 15 ms, dark purple markers) being faster than Tdifferent trials (721 ± 15 ms, pink markers) across each number of stimuli presented (F1,14 = 65.1, p < .001, η2G = 0.086) (see Fig. 2c). We also confirmed that increasing perceptual grouping facilitated search, as indicated by decreasing reaction times for increasing total number of distractors shown (F4,56 = 17.33, p <.001, η2G = 0.118), which is consistent with prior research (e.g., Bravo & Nakayama, 1992; Duncan & Humphreys, 1989; Song & Nakayama, 2006). There was no significant interaction (F4,56 = 1.96, p = .113, η2G = 0.022). Taken together, these results suggest slower reaction times on two-target trials when the target shapes and responses are different compared with when they are the same, independent of the strength of perceptual grouping.

To summarize, Experiment 2 suggested that unmatched perceptual grouping between one-target and two-target trials in part might have contributed to the longer reaction times for two-target trials during discrimination in Experiment 1. In addition, we also demonstrated in two- target trials that when targets differed in shape and were thus associated with different potential responses (Tdifferent, pink markers in Fig. 2c), performance was slower than when target shapes and potential responses were the same (Tsame, dark purple markers).

While the overall redundancy cost due to two targets during discrimination disappeared when perceptual grouping was equated, the difference between Tsame and Tdifferent for the two-target condition was still observed even after matching perceptual grouping, hinting at the role of perceptual-response compatibility in this effect. Similar to our results, Fournier and Eriksen (1990) also reported that when discriminating the identity of a single target, the presence of two predefined targets associated with different responses (e.g., the left lever for an ‘O’ vs. the right lever for an ‘X’) lead to slower reaction times compared with when two identical targets were presented. They reasoned this occurred because both potential responses were activated, causing a competition between responses that had to be resolved before an appropriate response was executed (e.g., Eriksen & Eriksen, 1979; Eriksen & Schultz, 1979; Gratton et al., 1988).

While response competition could have resulted in slower reaction times in Tdifferent than Tsame trials, we are not able to completely separate out the effect of response and target perceptual features (e.g., cut-off side). This is because the cut-off side of a target (top or bottom corner) determined the potential response to each target (press top or bottom button). Thus, in Experiment 3, we assessed the relative contributions of competition at the level of perceptual features and responses on two-target discrimination by dissociating these two features.

Experiment 3: Does perceptual or response competition between targets modulate two- target discrimination performance?

In Experiments 1 and 2, two types of two-target trials were included (1) two identical targets (Tsame), which shared the same shape (Ssame) and response (Rsame) or (2) two different targets (Tdifferent), which had different shapes (Sdifferent) associated with different responses (Rdifferent). In order to determine the relative contributions of perceptual and response competition in modulating two-target discrimination performance, we introduced a new two- target trial type, where both targets had different shapes (Sdifferent) but were associated with the same response (Rsame). Thus, we included the following three types of trials: 1) same shape-same response (Ssame–Rsame), 2) different shape-different response (Sdifferent–Rdifferent), and 3) different shape-same response (Sdifferent–Rsame). We reasoned that comparing performance in the new condition (Sdifferent–Rsame) with the other two (Ssame–Rsame and Sdifferent–Rdifferent) would provide further insight into how perceptual and/or response competition affected performance in two- target trials. The perceptual competition hypothesis would predict Sdifferent–Rsame is slower than Ssame–Rsame while comparable in performance with Sdifferent–Rdifferent because different shapes should incur competition relative to the same shape condition. However, the response competition hypothesis would predict Sdifferent–Rsame is faster than Sdifferent–Rdifferent while comparable in performance to Ssame–Rsame because different responses should incur competition relative to the two same response conditions.

According to the Theory of Event Coding (TEC), perceptual features and its associated responses get automatically associated and stored in the brain as ‘event files’ (Frings et al., 2020; Hommel, 2004, 2005; Hommel et al., 2001). These stimulus–response associations are assumed to be so strong that mere perception of a particular stimulus automatically activates the associated response codes and vice versa. TEC predicts that performance in both the Ssame–Rsame, and Sdifferent–Rsame conditions will be similar as in both of these conditions the same response code gets activated by associated perceptual features. For the Sdifferent–Rdifferent condition, it predicts that the two targets activate different and competing response codes, which might lead to relatively longer RT compared with the Rsame conditions.

Methods

Participants

Nineteen participants (12 female, mean age = 21 years) from the Brown University community volunteered to take part in this experiment for one hour in exchange for course credit or monetary compensation. All participants were right-handed and had normal or corrected to normal vision and normal color vision. They were naïve to the goals of the experiment. The protocol was approved by the Brown University Institutional Review Board.

Apparatus

The same apparatus was used as in Experiments 1 and 2.

Stimuli and procedure

The stimuli and task procedure were the same as in Experiment 2 except for the following. Because our primary focus was comparisons among the two target trials and not between one vs. two targets, we fixed the total of stimuli to six, in which either one or two targets were included. We modified the stimuli used in Experiments 1 and 2 by rotating them 45° (Fig. 3a). Thus, for each stimulus, either the top-left, top-right, bottom-left, or bottom-right corner could be cut off. The discrimination response remained the same as in Experiments 1 and 2, requiring participants to respond to a ‘top’ or ‘bottom’ cut-off corner regardless of whether it was cut off from the left or right side of the target. Target shapes varied in whether their shapes and potential responses were the same or different, resulting in three conditions (Fig. 3a): Ssame–Rsame, Sdifferent–Rsame and Sdifferent–Rdifferent. In the Ssame–Rsame condition, target shapes were always the same (e.g., top right and top right cut off), that corresponded to the same response (‘top’; Fig. 3a, left). In the Sdifferent–Rsame condition, both targets had different shapes (e.g., top left vs. top right cut off) that corresponded to the same potential response (‘top’; Fig. 3a, middle). Finally, in the Sdifferent–Rdifferent condition, each target had a different shape (e.g., bottom right vs. top right cut off) that corresponded to a different response (‘top’ and ‘bottom’; Fig. 3a, right).

Fig. 3
figure 3

Task and results of Experiment 3. a Representative displays. Participants reported whether the top or bottom corner was cut off from one odd-colored target, regardless of whether it was cut off from the left or right side. Either one or two targets were presented on each trial. On two-target trials, target shapes were randomly selected to create three trial types in combination of whether the shape (S) or response (R) between the two targets were the same or different: Ssame–Rsame, Sdifferent–Rsame or Sdifferent–Rdifferent. First, Ssame–Rsame refers to trials when identical targets were presented that were associated with the same response (left). Second, Sdifferent–Rsame refers to trials when targets had different cut-off corners that were associated with the same response (middle). Finally, Sdifferent–Rdifferent refers to trials when targets had different cut-off corners that were associated with the opposite responses (right). Both targets are highlighted by a dashed white line for display purposes only that was not presented in the experiment. b Mean reaction time for the three trial types. In accord with the response competition hypothesis, Sdifferent–Rsame (middle) is faster than Sdifferent–Rdifferent (right), while comparable in performance to Ssame–Rsame (left). Error bars represent the between-participants standard error of the mean (SEM)

Each two-target condition occurred an equal number of times in each block. During two-target trials, a distractor with each of the four unique shapes were presented on every trial. For one-target trials, the fifth distractor was selected to be each possible shape an equal number of times within a block. Participants completed three blocks each after a block of practice. Within each block, one target was presented on 32 trials (47%) and two targets were presented on 36 trials (53%).

Data analysis

One participant was excluded from analysis because of poor performance. For the remaining 18 participants we used the same exclusion criteria for each trial as in the previous experiments. This resulted in a mean of 1.1% ± 0.2% of one-target trials, and 1.1% ± 0.4% , 0.8% ± 0.3%, and 0.6% ± 0.3% of Ssame–Rsame, Sdifferent–Rsame, and Sdifferent–Rdifferent trials excluded from analysis. During analysis, subsequent pairwise comparisons were Bonferroni corrected.

Results and discussion

We first assessed how accuracy varied across conditions. Sdifferen–Rdifferent trials were dropped from the accuracy analysis because participants could not be wrong. Accuracy between the one target (94.9% ± 0.8%), Ssame–Rsame (92.3% ± 1.2%), and Sdifferent –Rsame (90.1% ± 1.7%) conditions differed significantly in a one-way repeated-measures ANOVA (F2,34 = 9.40, p < .001, η2G = 0.118). Pairwise comparisons revealed percent correct was higher for one-target trials than Sdifferent–Rsame trials (t17 = 4.00, p = 0.003, d = 0.94) and Ssame–Rsame trials (t17 = 2.85, p = 0.03, d = 0.671). However, the difference between Ssame–Rsame and Sdifferent–Rsame trials (t17 = 1.83, p = .253, d = 0.432) was not significant.

We next examined whether reaction time for correct trials differed between one target and two-target trials. The average reaction time was slower for two target (764ms ±22ms) relative to one target (733ms ±20ms) trials (t17 = 4.81, p < 0.001, d = 1.13). This is consistent with the results of Experiment 1, where the total number of stimuli was kept constant at six as in this experiment. The critical comparison was whether the Sdifferent–Rsame condition differed significantly from the Ssame–Rsame and Sdifferent–Rdifferent conditions. We reasoned that response competition would result in faster reaction times in the Sdifferent–Rsame condition, where the response is the same, than in the Sdifferent–Rdifferent condition, where the response is different, along with comparable performance to the Ssame–Rsame condition. In contrast, perceptual competition would result in slower reaction times in the Sdifferent–Rsame condition, where the shapes are different, than in the Ssame–Rsame condition, where the shapes are the same, and comparable performance to the Sdifferent–Rdifferent condition.

Figure 3b depicts mean reaction time for each two- target trial type. A one-way repeated-measures ANOVA revealed a significant main effect of trial type (F2,34 = 12.36, p <.001, η2G = 0.045). We were interested in comparing the reaction time difference between the Sdifferent–Rsame and the Sdifferent–Rdifferent conditions, which would tell us about the role of response competition, and between the Ssame–Rsame and the Sdifferent –Rsame condition, which would tell us about the role of stimulus competition. In accord with the predictions of a response competition, pairwise comparisons after Bonferroni correction revealed that reaction times were significantly faster for the Sdifferent–Rsame (middle) relative to the Sdifferent–Rdifferent condition (right; t17 = 6.43, p < .001, d = 1.52), with no significant difference between the Sdifferent–Rsame (middle) and Ssame–Rsame condition (left; t17 = 2.24, p = .115, d = 0.53). Although, the reaction time difference between the Ssame–Rsame and the Sdifferent–Rdifferent conditions (t17 = 2.29, p = .104, d = 0.54) did not reach significance it showed the trend in the predicted direction. Overall, these results suggested that responses were faster when both the presented targets were associated with the same response code compared with when both the presented targets were associated with different response.

General discussion

Previous studies have investigated how redundant targets influence visual search during detection. Here, we extended these investigations to discrimination of a target’s features. Pop-out detection is thought to suffice with distributed attention across a wide range of the visual field, while pop-out discrimination requires focused attention to a stimulus to resolve a perceptual feature (Bravo & Nakayama, 1992; McPeek et al., 1999; Nakayama, 1990; Nakayama & Joseph, 1998; Song & Nakayama, 2006). We hypothesized that a redundant target might influence visual search performance differently depending on whether the task requires detection or discrimination, and whether the information provided by the redundant target is congruent or incongruent.

First, we observed that oddity target detection reaction times were faster when two targets were present compared with one target. This result successfully replicated the redundancy gain found in previous studies (Krummenacher et al., 2001, 2002; Miller, 1982) for two targets in the pop-out search paradigm. In our task, targets were defined by being an odd color, and target and distractor colors were randomized on each trial. This task design requires the use of salient perceptual differences between targets and distractors to detect a target, rather than a search strategy that allows one to search for a specific feature. Contrary to our design, a series of studies by Krummenacher and colleagues investigated the effects of multiple targets on oddity target detection when target features were pre-specified to participants, thereby allowing the use of target feature information to bias search (Krummenacher et al., 2001, 2002, 2014; Töllner et al., 2011). In their tasks, targets were defined in two feature dimensions (e.g., a red stimulus and a right-oriented line) and a multiple target trial would consist of both feature dimensions (e.g., a red right-oriented line). The authors demonstrated faster reaction times when multiple targets were present compared with when either target was present alone due to an intensified target-present signal (Krummenacher et al., 2001, 2002). Even with the differences in task, the common result found in our study and by Krummenacher and colleagues suggests that multiple salient targets facilitate detection regardless of foreknowledge about the target defining feature.

Second, when participants performed a pop-out discrimination task, we initially observed that reaction time was slower when two targets were present relative to one. This result suggests that redundant targets can incur a redundancy cost for discrimination. In subsequent experiments, we further examined what factors contributed to this redundancy cost. We first found that matching the perceptual grouping efficiency between one-target and two-target displays reduced the reaction time cost for two targets, suggesting that perceptual grouping partly modulates the impact of multiple targets. It has been proposed that perceptual grouping efficiency determines pop-out discrimination performance because of the need to allocate focused attention to a target, whereas pop-out detection does not require focused attention (Bravo & Nakayama, 1992; Julesz, 1986; Koch & Ullman, 1985; McPeek et al., 1999; Song & Nakayama, 2006). This may partly explain the asymmetry in the effects of multiple targets across tasks that we observed in this study. Though perceptual grouping appears to partly explain the redundancy cost observed in our pop-out discrimination task, it remains unclear whether perceptual grouping can also explain the redundancy gain in detection. Further experiments are required to confirm the role of perceptual grouping for target detection.

We also demonstrated that performance in pop-out discrimination with two targets was modulated by the competition of potential responses associated with each target, rather than perceptual features. In Experiments 1 and 2 and when comparing the Sdifferent–Rsame and Sdifferent–Rdifferent conditions in Experiment 3, reaction time when either target was associated with opposite responses was slower than when the responses were the same, regardless of whether the target shapes matched or not. This pattern of results suggests that both responses associated with either target were activated and that interference driven by the opposing nature of the responses (press ‘top’ or ‘bottom’) incurred slowed reaction times. Many studies suggest that simultaneously active responses can compete with each other, which causes conflict that must be resolved prior to one response being executed (Eriksen & Eriksen, 1974; Eriksen & Schultz, 1979; Fournier & Eriksen, 1990; Gratton et al., 1988). Those studies and the findings reported here are consistent with a motor coactivation model of redundant target decision making where the activity of processing nodes associated with multiple activated responses are summated (Giray & Ulrich, 1993; Miller, 1982). In such a processing architecture, response conflict could incur a redundancy cost by either slowing down the rate of evidence accumulation for the eventually executed response or else halting the evidence accumulation process until the conflict has been resolved. Future work should be conducted to differentiate between these possibilities. Regardless of the underlying information processing architecture, our results suggest that response competition may arise during pop-out discrimination when multiple targets differ in their potential responses.

The response competition that we observed is also consistent with the notion of “event files” (Frings et al., 2020; Hommel, 2004, 2005). According to Hommel (2004, 2005), when a participant encounters a perceptual event and responds with a specific action, a transient “event file” is created in which a representation of the perceptual event, task context, and associated action are bound. These files can be retrieved during future encounters with that same perceptual event and task context, thereby reducing the demand on limited cognitive resources required for action selection. Perhaps, in our task, the presence of two targets with different perceptual features associated with different responses elicited response competition by simultaneously activating both event files associated with a target missing a top and bottom corner, thereby priming both responses.

Our results may also shed light on the mechanisms of attentional selection during visual search. Classic theories of visual attention propose that the feature information of a target is available only after attention is allocated to the target (Eriksen & St James, 1986; Nakayama & Joseph, 1998; Treisman & Gelade, 1980; Wolfe, 1994; Wolfe, 2007). It follows that a response predicated on a target feature, such as during the discrimination tasks in our study, is available only after the target is attended. Thus, according to these theories our finding that target responses modulate two target pop-out discrimination performance suggests that both targets were attended in this task. There is considerable evidence that attention may be split between multiple stimuli (Cavanagh & Alvarez, 2005; McMains & Somers, 2004; Pylyshyn & Storm, 1988), including during visual search (Eimer & Grubert, 2014; Grubert & Eimer, 2015, 2016). Given the highly salient nature of the targets used in our pop-out task, it is plausible that attention may have been directed to both targets, either inadvertently or as part of a strategy that participants employed. Under this assumption, it is unclear whether both targets would be attended simultaneously (i.e., parallel selection) or each target was serially attended (or some combination of these strategies occurred across trials). While parallel selection has often been assumed in redundant target detection tasks (e.g., Eriksen & Eriksen, 1974; Eriksen & Schultz, 1979; Fournier & Eriksen, 1990; Gratton et al., 1988), serial selection can in theory also result in response conflict if the response associated with the first target remains activated after attention has been directed to the second target. Moreover, a serial selection account would predict that at least some of the redundancy cost we observe could be driven by an additional time cost associated with re-directing attention from the first to the second target. Future work should be conducted to determine the dynamics of attentional allocation during two target discrimination.

It is also worth noting that in principle, participants did not need to resolve the response conflict in our experiment as any of the associated responses activated by the redundant targets would always be correct. Our result may imply that as soon as two targets are attended, co-activation of opposite responses is automatic and inevitable. Therefore, the observed delay might have arisen due to the process of conflict resolution because participants must select one response only on each trial. One possible future experiment might be able to throw some light on this issue: if participants have the option to press both keys simultaneously whenever both response codes are activated then there would not be any need to inhibit any particular response, resulting in no redundancy cost if contributed by the inhibitory processes.

To summarize, we believe that a full understanding of redundant target effects in visual search involves consideration of a variety of perceptual and cognitive factors including the level of feature processing required by the task (e.g., mere detection of a target’s presence vs. discrimination of a target’s shape), attentional selection (of either one or both targets), perceptual grouping processes (Bravo & Nakayama, 1992; Duncan & Humphreys, 1989; Song et al., 2008; Song & Nakayama, 2006) and response conflict (or the absence thereof). Future studies should further investigate the role of each of these factors and develop newer models that can provide a coherent theory to explain both redundancy gain and redundancy cost in a common framework.