Age-Related Differences in Saccadic Indices of Top–Down Guidance via Short-Term Memory During Visual Search

Aging has been associatedwith signi ﬁ cant declinesin the speedand accuracyof visual search.These effects have been attributed partly to low-level (bottom – up) factors including reductions in sensory acuity and general processing speed. Aging is also associated with changes in top – down attentional control, but the impact of these on search is less well-understood. The present study investigated age-related differences in top – downattentionalcontrolbycomparingthe speedandaccuracyofsaccadicsamplinginthe presenceand absence of top – down information about target color in young (YA) and older (OA) observers. Displays contained an equal number of red and blue Landholt stimuli. Targets were distinguished from distractors by a unique orientation, and observers reported the direction of the target ’ s gap on each trial. Single-target cues signaled the color of the target with 100% validity. Dual-target cues indicated the target could be present in either colored subgroup. The results revealed reliable group differences in the bene ﬁ ts associated with top – down information on single-target cues compared to dual-target cues. On single-target searches, OA made signi ﬁ cantlymoresaccadesthanYAtostimuliintheuncuedcolorsubset.Single-targetcuesalsoproduceda smaller advantage in the time taken to ﬁ xate the target in OA comparedto YA. These results support an age-relateddeclineinobservers ’ useoftop – downinformationtorestrictsequences ofsaccadestoatask-relevant subset of objects during visual search.

. Age-related deficits in attention and recognition have also been associated with increased rates of driving accidents (Marottoli & Drickamer, 1993), and an extensive literature reflects an ongoing interest in the effect of aging on sensory and cognitive components of visual search (Foster et al., 1995;Madden et al., 1996).Despite the large body of evidence supporting age-related differences in top-down selection, little is known about the way this impacts saccadic sampling and the evaluation of visual objects during search.This is important, because differences in the speed and accuracy of eye movements in young (YA) and older adults (OA) provide a direct comparison of the effect of aging on neural and cognitive processes that integrate sensory input with information in visual short-term memory (VSTM) during the prioritization, maintenance, and evaluation of objects in the scene.Indices of saccadic sampling also provide highly granular measures of components of executive control, which are fundamental to the maintenance of goal-directed behavior and are thought to decline with age (e.g., Kramer et al., 1999;West, 1996).
Visual search is perhaps best characterized in stages.Eimer (2015), for example, used four stages: preparation, guidance, selection, and recognition.Preparation entails the formation of a mental representation or attentional template of the target's visual features (e.g., color and shape), which is maintained in VSTM (Desimone & Duncan, 1995;Woodman et al., 2013).Guidance depends on perceptual processes that segregate the scene based on contrasts between low-level visual features (Itti, 2005;Li, 2002;Zhaoping, 2005).This bottom-up segregation is mediated by topdown mechanisms, which preferentially weight input matching the attentional template (Eimer, 2015;Olivers et al., 2006;Saenz et al., 2002;Serences & Boynton, 2007;Wolfe, 2007).Bottom-up and top-down information converge to signal the location of taskrelevant stimuli, which are subject to selection.Within a single fixation, guidance operates to restrict capacity-limited resources to the locations in the scene most likely to contain the target (Bundesen et al., 2005;Folk et al., 1994;Wolfe, 2007).During oculomotor sampling, guidance also informs the programming of saccades to potential target locations (Bisley & Goldberg, 2010;Rutishauser & Koch, 2007;Zelinsky & Bisley, 2015).The final stage of search entails the verification of selected information against the attentional template (Castelhano et al., 2008).When the degree of similarity passes a decision threshold, the observer reports the target present.If not, the observer reorients to another selected location or reports the target absent (Barrett & Zobay, 2020;Najemnik & Geisler, 2005;Palmer et al., 2000).
The four-stage model proposed by Eimer (2015) highlighted the complexity of visual search as well as the role of top-down processes during preparation, guidance, selection, and target verification.Evidence that aging affects top-down processes during search comes from studies that have compared OA and YA using displays in which targets were differentiated from distractors by values on a single feature or conjunction of features.Searches for feature singleton targets (i.e., a red X among green O) elicit comparable rates of accuracy and response times (RTs) in YA and OA (Humphrey & Kramer, 1997;Müller-Oehring et al., 2013).Where slight differences have been found, they have been attributed to general slowing of sensory processing and manual responses (Salthouse 1996(Salthouse , 2000)).Searches for targets differentiated from distractors by a conjunction of values (i.e., a red X among red Os and green Xs), however, typically elicit significant reductions in the speed and accuracy of responses for OA compared to YA (Hommel et al., 2004;Kramer et al., 1996;Müller-Oehring et al., 2013;Plude & Doussard-Roosevelt, 1989).This asymmetry is found for displays containing identical features, indicating that age-related differences in conjunction search are not caused by an inability to discriminate individual features.Instead, age-related decrements have been attributed to a reduction in the ability to weight relevant over irrelevant features during selection (Dennis et al., 2004;Folk & Lincourt, 1996;Madden et al., 1996;Potter et al., 2012) or to allocate capacity-limited processes to the maintenance and comparison of visual input against attentional templates defined by multiple features (Dennis et al., 2004;C. C. Williams et al., 2009).
Search for features and conjunctions can be described in terms of differences in stimuli as well as the efficiency of top-down processes involved in guidance and the evaluation of objects in the scene (Wolfe, 1994(Wolfe, , 1998)).Search for feature singletons is usually highly efficient, producing RTs that are independent of changes in the number of objects in the display.RTs during conjunction searches in contrast are positively related to set size, with increases in the number of objects in the scene eliciting a concomitant increase in the number and latency of fixations during saccadic sampling (Scialfa & Joffe, 1998;D. E. Williams et al., 1997).In the absence of attentional capture by a unique feature, inefficient search is thought to require the maintenance of a goal state in VSTM and proactive control of sensory and motor processes, which are used to prioritize target features in the display (Gaspelin et al., 2015;Hamblin-Frohman et al., 2022;Hollingworth et al., 2008), track and exclude objects that have already been sampled (Dorris et al., 2002;Klein & MacInnes, 1999), resolve competition for selection (Poole & Kane, 2009), and evaluate foveated objects against information in the attentional template (Castelhano et al., 2008;Yu et al., 2022).These demands may be particularly susceptible to age-related reductions in VSTM and the functional integrity of the frontal parietal network, which are thought to mediate top-down selection and the maintenance of goal-directed strategies during eye movement programming (Menegaux et al., 2020;Zanto & Gazzaley, 2019).Increases in saccadic latencies in the presence of irrelevant but salient singletons during search are larger among OA than YA (Cassavaugh et al., 2003), and aging is associated with a reduced ability to inhibit reflexive eye movements toward visual transients in the antisaccade task (Nieuwenhuis et al., 2004;Olincy et al., 1997;Peltsch et al., 2011).These findings suggest that age-related decreases in the speed and accuracy of inefficient search may reflect a decline in OA's ability to maintain proactive control of the topdown processes used to prioritize objects, guide eye movements, and evaluate potential targets at different stages of search.
Much of the evidence for age-related declines in top-down control during search comes from studies that measured the accuracy and speed of manual responses.Although highly informative, these provide point estimates of cumulative processes that occur across distinct stages of search.This makes it difficult to quantify the impact of age-related changes in the efficiency of top-down control during the specific stages identified by Eimer (2015).The present study is designed to address this limitation by using eye movement recording to quantify age-related differences in the use of top-down information during the preparation, guidance, and verification stages of search.Eye movement recording provides moment-by-moment 2 BARRETT, HUTCHINSON, ZHANG, XIE, AND WANG measures that can be used to identify age-related differences in saccadic sampling that occur before and after target selection.Prior to selection, initial saccadic latencies are thought to index preparatory processes that encode the target's identifying features within the attentional template and perceptual processes that segregate the scene (Malcolm & Henderson, 2009).The time between the initial saccade and the first target fixation is considered an index of the efficiency of proactive prioritization of target-relevant features during the guidance stage of search (Tseng & Li, 2004;Zhao et al., 2012), while the time between the first target fixation and the manual response is considered an index of the efficiency of reactive processes that evaluate selected input against the attentional template (Geng & Witkowski, 2019;Hout & Goldinger, 2015;Yu et al., 2022).
To manipulate the information available to top-down control, we required YA and OA to search displays containing subsets of objects defined by the elementary colors red and blue.Color provides a highly salient cue for top-down selection (Anderson et al., 2010;Bundesen & Pedersen, 1983;J. Lee et al., 2018), and the discrimination of elementary colors across the visual field is largely preserved through the lifespan (Wijk et al., 1999;Wuerger, 2013;Yada et al., 2021).During search, observers were required to detect an orientation singleton among heterogeneously oriented distractors that have previously been shown to elicit inefficient search (Vlaskamp et al., 2005).Prior to search, different cues were used to manipulate the availability of top-down information about the target's color.Single-target cues signaled the subset of colored objects that contained the target with 100% validity.Dual-target cues signaled both colors, precluding the top-down selection of the subset that contained the target during search.Single-and dualtarget cues, therefore, manipulated the presence or absence of topdown information about the target's value on a feature that was orthogonal to target detection (Friedman-Hill & Wolfe, 1995).The order of presentation of single-and dual-target cues as well as the target's color were pseudorandomly assigned on each trial, restricting the informational value of the cue(s) to a single search.Comparing performance on single-and dual-target cues, therefore, provides a direct index of the benefits afforded to YA and OA when top-down information in VSTM can be used to inform preparatory, guidance, and verification stages of search.In the present study, agerelated differences in the use of top-down information are expected to reduce differences in saccadic measures of performance during distinct stages of single-and dual-target search in older relative to younger adults.

Method Transparency and Openness
We report how we determined our sample size and describe all data exclusions and measures used in the study.Summary statistics of eye movement data were computed using custom-built function in MATLAB (Mathworks, Natick, MA, USA) with the Edf2mat toolbox (https://github.com/uzh/edf-converter)from the SR Research.edffile.Summary data were computed using R, Version 4.2.1 (R Core Team, 2021).The MATLAB function used to generate stimuli, summary data, and code for descriptive and inferential statistics are available on the Open Science Framework at https:// osf.io/pbydr/?view_only=56ddc349e59346818bb3bc17b46133ea.The study and analytic plan were not preregistered.

Participants
Eighty observers (40 YA and 40 OA) were recruited to the study.The sample size was predetermined to meet the recommended number of observations (1,600) in each cell for repeated measures factorial analyses of RT (Brysbaert & Stevens, 2018).YA observers were recruited from Tianjin Normal University in Tianjin, and OA participants were recruited from the Tianjin city community.Observers were screened for normal visual acuity (>20/40 Snellen values) using the Tumbling E eye chart (Taylor, 1978) and contrast sensitivity using a Pelli-Robson chart (Pelli & Robson, 1988)

Apparatus
The experiment was run on a DELL OptiPlex XE2 computer with a 24-inch monitor.The display resolution was 1920 × 1,080 pixels, and the frame rate was 85 Hz.Stimulus presentation and data collection were controlled using custom-built software in MATLAB (Mathworks, Natick, MA, USA) with Psychtoolbox (Brainard, 1997;Kleiner et al., 2007) and Eyelink (Cornelissen et al., 2002) functions.Viewing distance was maintained at 75 cm using a fixed chin rest, and eye movements were measured using an EyeLink 1,000 video-based eye tracker (SR Research Ltd., Ottawa, ON, Canada) with a sample rate of 1,000 Hz and spatial resolution of <0.02°.

Stimuli
Displays contained Landholt's C stimuli that subtended 2.1°.The stroke and gap were 20% of the stimulus width, and C-shapes were colored red (RGB 170, 1, 1) or blue (RGB 1, 1, 220) on a midgrey background (RGB 137,137,137).C-shapes were presented at one of 12 equally spaced locations on a virtual circle centered on the display with a radius of 12.8°.Targets and distractors were defined by orientation.Target C-shapes had gaps at 90°or 270°.Gaps for distractors were randomly sampled from target orientations ±45°.Cues were red or blue circles that subtended 1.68°.Cues were presented at two locations centered on the horizontal midline of the screen at ±1.9°from fixation.

Procedure
The experiment used a factorial design to compare YA and OA on two repeated measures: search type (single-or dual-target cue) and set size (2, 4, and 6) of red and blue C-shapes.Experimental blocks contained eight repetitions of this structure (48 trials) and included The order of presentation of trials in each block was randomized, and observers completed five blocks in a single session.Color (red and blue) was used to cue top-down selection of one or two subsets of objects.The use of high-contrast elementary colors in our displays was designed to minimize the potential influence of agerelated differences in color perception on top-down selection.To ensure OA could discriminate between objects in each subset, observers also completed practice trials prior to the experiment.These required observers to identify objects that matched red and blue cues in the display.To ensure that observers understood how they were expected to use the information provided by single-and dual-target cues, they were also asked to verbalize their search strategy for both types of search prior to testing.OA and YA observers were able to identify red and blue objects with 100% accuracy and describe the relationship between single-target cues and the subset of objects that could contain the target on the subsequent search.
Figure 1 illustrates the sequence of events on each trial.Trials began with a fixation cross at the center of the screen.After 500 ms, the fixation was replaced by two cues.On a single-target search, both cues were the same color.On a dual-target search, one cue was red and one blue.On single-target searches, the cues specified the color of the forthcoming target with 100% validity.On dual-target searches, the target was equally likely to be red or blue.Cues were presented for 1,000 ms and followed by a fixation cross on an otherwise blank screen.After 1,000 ms, the cue was replaced by a search display, containing two, four, or six red and blue C-shapes.Search displays remained on the screen until a manual response had been made or 5,000 ms had passed.Observers used the arrow keys on a standard keyboard to report the orientation of the target's gap.Search displays were followed by a screen containing written feedback at the center of the screen ("yes" = correct; "no" = incorrect) for 500 ms.Note.In this display, the set size of red and blue C-shapes is 6.Single-target cues signal the colored subset of C-shapes that contains the target.On dual-target trials, the target is equally likely to be in the red or blue subset.In this display, the target has a gap on the left and appears at the 10 o'clock location.Observers reported the orientation of the target (90°or 270°) using a button press on each trial.See the online article for the color version of this figure.ISI is the interstimulus interval between the cue and search disply.Following feedback, a blank screen of 500 ms preceded the onset of the next trial.Eye movements were recorded monocularly from the right eye during every trial.Deflections in eye position exceeding 0.1°with a minimum velocity of 30°/s and acceleration of 8,000°/second 2 were defined as saccades.Saccadic start, end, and fixation X, Y coordinates corresponded to locations on the screen, and saccadic latencies were defined from the onset of the search display on each trial.The number and latency of saccades to the target and to red and blue distractors were quantified as saccadic end points that landed on or within a circle center on each C-shape with a radius of 1.26°.Each block was preceded by a 5-point calibration sequence to ensure gaze location could be tracked with an accuracy of ≤0.5°, and a drift correction procedure was applied every 16 trials.Observers completed five blocks of 48 trials in a single session.Trials on which no saccades was recorded or RTs were less than 100 ms were excluded from analyses (<4%).Trials on which RTs were longer than 5,000 ms were categorized as incorrect responses.Due to a technical error, one block of eye movement data for an OA observer failed to record (48 trials).

Analyses
To investigate differences in the accuracy of search, we calculated the proportion of trials on which observers correctly reported the target's orientation.To quantify the efficiency of guidance, scan path ratios (SPRs) were computed by dividing the total distance travelled by the eye during search to the linear distance between the observer's fixation at the onset of the display and the center of the target (Brockmole & Henderson, 2006;Castelhano et al., 2008).An SPR of 1 indicates optimal guidance, while decreasing values indicate less direct scan paths during search.In addition, we calculated the proportion of saccades that ended in fixations on targets and distractors in both colored subsets on each trial.To investigate differences in the temporal characteristics of search, we contrasted RTs for manual responses and saccades during early, middle, and late stages of search (Zhao et al., 2012).In the early stage, initial saccadic latencies (ILs) are thought to reflect perceptual segmentation of the display and preparatory processes.In the middle stage, the time from the initial saccade to the first target fixation (target latency; TL) indexes the efficiency of attentional guidance during search.In the late stage, the time between the first target fixation and the manual response (decision latency; DL) indexes decision-making and response execution (Hout & Goldinger, 2015).Aging is associated with reductions in manual (Madden, 1992) and saccadic RTs (Munoz et al., 1998), as well as peak saccade velocity (Bono et al., 1996).Group comparisons revealed very strong evidence for higher RTs (μ = 455.89ms, BF 10 = 2.877 +24 ), ILs (μ = 13.51 ms, BF 10 = 4.852 +6 ), TLs (μ = 243.67,BF 10 = 3.433 +8 ), and DLs (μ = 212.22BF 10 = 1.441 +51 ) in OA compared to YA, which is consistent with an age-related reduction in processing speed (Salthouse, 2000).
To differentiate age-related differences in processing speed from conditional changes in the time course of search in YA and OA, we z-transformed RTs, ILs, TLs, and DLs.z transforms were computed by subtracting individuals' mean in each condition from their overall mean and dividing this value by the standard deviation of their conditional means (Wiegand & Wolfe, 2020).z transforms enable direct comparison of time course distributions that vary by a constant value.Transformed values are centered at zero, with positive and negative values representing conditional changes in time relative to individuals' mean across conditions.Tables listing group means for raw and z-transformed latency measures on trials in which participants correctly reported the target's orientation are provided on the Open Science Framework site listed in the Transparency and Openness section (see Tabulated_Results.pdf).
Accuracy and latency data were entered into Bayesian analyses of variance (ANOVAs).These contrast the predictive performance of competing models for data obtained in factorial designs (Morey et al., 2016;van den Bergh et al., 2020).Data for accuracy measures were compared using five models: main effects of group (OA vs. YA), search type (single-vs.dual-target), set size (2, 4, and 6), an additive, and an interaction model.The additive model included terms for all main effects (i.e., outcome ∼ group + search type + set size).The interaction model specified a group interaction term for within-subjects factors (i.e., outcome ∼ group + search type + set size, + group by search type + group by set size).For z-transformed data, simple comparisons of group index differences in the distribution of the data from group means, which are difficult to interpret in isolation.Analyses of z-transformed data, therefore, exclude the main effect of group and contrast main effects of search type and set size, as well as the additive and group interaction models.Contrasting the main effects and additive model tests the probability that the data reflect one or the combination of independent effects.Contrasting the additive and interaction models provides an explicit test of age-related differences in the effects of search type and set size.
Bayes Factors (BF) were computed using default priors with the BayesFactor R package (v12.4.4;Morey & Rouder, 2022).Subject identity was entered as a random effect, and initial analyses compared all models against the null or subject only model.Subsequent analyses compared the interaction against the main effects and additive models.Additive and interaction models included a search type by set size term to partial differences associated with the multiplicative increase in set size from those associated with changing cognitive demands on single-and dualtarget searches.Estimated BF 10 values indicate the strength of evidence for each model in the comparison.For example, a BF 10 value of 100 for the interaction over the additive model indicates the observed data are 100 times more likely if search type and set size effects are mediated by age.BF 10 estimates were classified using the scheme proposed by M. D. Lee and Wagenmakers (2014), and post hoc contrasts are reported using median-estimated posteriors and highest posterior density (HPD) intervals for the model obtaining the highest support for each comparison.

Accuracy: Manual Responses
Figure 2 plots mean accuracy rates for target orientation judgements by group, search type, and set size.Accuracy for YA was above 81% on both single-and dual-target searches.Accuracy for OA was slightly lower than for YA, falling to 66% on dual-target searches at set size 6.

RT: Manual Responses
Comparison of the main effects, additive, and group interaction against the null model for manual RTs (Figure 3) revealed extreme support for all models (BF 10 = 1.781 +14 ± 1.42% to 4.004 +222 ± 3.58%).Comparison of the group interaction against the additive (BF 10 = 1.981 +5 ± 4.27%) and main effects of search type (BF 10 = 2.248 +208 ± 3.85%) and set size (BF 10 = 8.805 +86 ± 3.82%) yielded the strongest evidence for the interaction model.Post hoc contrasts indicated a larger reduction in zRT on single-target searches compared to dual-target searches for YA (0.401, HPD = [0.364,0.440]) than   ).The results support age-related differences in the effects of search type and set size, with the speeding of manual responses on single-target searches compared to dual-target searches larger in YA than OA.
The results above reveal age-related decreases in the accuracy of single-and dual-target search.Top-down information signaling the color of the cue increased the accuracy of search in OA but not YA, while the benefit associated with single-target cues on transformed RTs was larger in YA than OA.These data suggest top-down information is available to both groups of observers, with the benefit conferred by single-over dual-target cues varying with agerelated differences in the accuracy and speed of search at different set sizes.

Accuracy: Eye Movements
SPRs provide a measure of the efficiency of the guidance stage of search, with decreasing efficiency resulting in an increase in the distance travelled by the eyes before the target is fixated (Figure 4).Comparison of the main effects, additive, and group interaction against the null model for SPRs revealed extreme support for all models (BF 10 = 1.834 +3 ± 1.33% to 9.804 +110 ± 2.95%).Comparison of the group interaction against the additive (BF 10 = 1.041 +5 ± 3.37%) and main effects of group (BF 10 = 5.334 +107 ± 3.24%), search type (BF 10 = 1.815 +99 ± 3.27%), and set size (BF 10 = 5.764 +42 ± 4.95%) yielded the strongest evidence for the interaction model.).These results support more efficient scan paths and a larger advantage on single-target search compared to dual-target search for YA than OA.SPRs were also lower in OA than YA across set sizes, with the largest difference between groups at set sizes 2 and 4.
The analysis above indicates scan paths were longer in OA than YA.To investigate this further, we calculated the proportions of fixations observers made to different objects between the onset of the display and their response.Figure 5 plots the proportion of saccades to targets (T) and distractors from the same (DS) and different (DD) colored subsets by search type and group.On dual-target (DT) search, the distributions of saccades to targets and distractors were similar for OA and YA.On single-target searches, the proportion of saccades to distractors in the uncued subset was higher for OA than YA.Single-target cues signal the subset of objects that contain the target with 100% validity.and the probability of a target in the nontarget subset is zero.Differences in the distribution of saccades to targets and distractors on singletarget search, therefore, reflect differences in the extent that OA and YA used the cue to restrict saccades to objects in the relevant subset.
The proportion of saccades to the nontarget subset (DD in Figure 5) was subject to Bayesian ANOVA.Comparison of the main effects, additive, and group interaction against the null model revealed anecdotal evidence for the set size (BF 10 = 0.189 ± 0.73%) and group (BF 10 = 1.900 ± 0.87%) models and extreme evidence for the search type (BF 10 = 4.420 +115 ± 0.99%), additive (BF 10 = 6.839 +117 ± 9.67%), and group interaction models (BF 10 = 4.811 +127 ± 3.61%).Comparison of the interaction against the additive (BF 10 = 7.036 +9 ± 10.32%) and search type (BF 10 = 1.089 +13 ± 3.74%) models yielded extreme support for the former.Post hoc comparisons indicated a higher proportion of fixations to the nontarget subset in OA than YA on single-target searches (0.060 HPD = [0.040,0.080]).On dual-target searches, the difference in the proportion of fixations to objects in the nontarget subset was much smaller (−0.007HPD = [−0.026,0.012]).OA observers made The results above support an age-related reduction in the exclusivity of spatial sampling on single-target searches.To investigate the use of top-down information on ordering sequences of saccades, we calculated the proportion of fixations that included a switch between colored subsets during search.Object fixations were classified in terms of color on each trial, and the number of switches between color subsets was divided by the number of object fixations.Lower proportions reflect longer sequences of saccades to objects in the same subgroup.On dual-target searches, the mean proportion of switches at set size 4 was .26,indicating that observers made an average of one switch when they sampled displays containing two red and two blue objects.On single-target searches, the proportion of switches reduced, with this decrease larger in YA than OA (Figure 6).
The analyses above reveal age-related reductions in the spatial and temporal selectivity of saccades on single-target searches compared to dual-target searches.To investigate the temporal sequalae of these reductions, we contrasted z-transformed ILs, TLs, and DLs for OA and YA observers by search type and set size.

Initial Saccade Latency
Initial target latencies quantify the time between the onset of the search display and the initial saccade made by observers (Figure 7).Comparison of the main effects, additive, and group interaction against the null model for transformed initial saccadic latencies revealed anecdotal evidence against the search type (BF 10 = 0.798 ± 1.41%) and moderate evidence against the interaction (BF 10 = 0.034 ± 2.75%) model.The additive (BF 10 = 27.042± 2.28%) and set size (BF 10 = 274.299± 1.67%) models received strong and extreme support, respectively.Comparison of the set size against the additive model yielded strong support for the former (BF 10 = 10.143 ± 2.82%).Post hoc contrasts for the set size model yielded lower posterior estimates of initial saccadic latencies at set sizes 2 (−0.038HPD = [−0.061,−0.015]) and 6 (0.010 HPD = [−0.013,0.034]) compared to 4 (0.028 HPD = [0.001,0.052]).These results suggest differences in the latency of initial fixations are influenced by set size independently of search type and differences in observers' age.This indicates that either OA and YA are equally proficient at parsing displays defined by elementary colors or the processes that govern the relative time taken to implement the attentional template and initiate the first saccade are insensitive to bottom-up segregation of colored subsets in our displays.

First-Target Saccade Latency
Figure 8 plots z-transformed first TL by group, search type, and set size.This value is computed as the difference between the initial and first fixations to the target and provides a temporal measure of efficiency of the guidance stage of search.

Decision Latency
Figure 9 plots z-transformed DL by group, search type, and set size.This value is calculated as the time between the first target fixation and the observers' manual response and is considered an index of the time taken to evaluate the selected target against the attentional template during foveation (Castelhano et al., 2008).

Discussion
The present study used single-and dual-target cues to compare the benefit of top-down information signaling the relevant subset of colored objects during search in YA and OA.Analyses of manual responses revealed an age-related decrease in accuracy across single-and dual-target searches.Accuracy was lower on dual-target   searches compared to single-target searches, and the decrease was larger in OA than YA.z-transformed RTs were faster on singlethan dual-target searches, demonstrating a benefit of top-down information when cues signaled the subset of objects containing the target.This benefit was larger in YA than OA, supporting an agerelated reduction in the advantage afforded by single-target cues.These results are consistent with previous evidence of age-related reductions in the speed and accuracy of manual responses (Hommel et al., 2004;Plude & Doussard-Roosevelt, 1989) and evidence an age-related decrease in the benefit afforded on RTs by top-down information during search (Müller-Oehring et al., 2013).
Manual responses provide point estimates of cumulative processes that occur across distinct stages of search.To investigate the impact of age and the availability of top-down information on the accuracy and time course of saccadic sampling, we contrasted the relative benefit of single-and dual-target cues in YA and OA at different stages of search.Comparison of SPRs evidenced an increase in the distance travelled by the eyes during guidance on dual-target searches compared to single-target searches.Scan paths were longer for OA than YA, with the increase larger on dual-than single-target searches.The spatial distribution of saccades also varied significantly by search type.On single-target searches, the mean proportion of fixations on objects in the cued subset was 0.86.On dual-target searches, the mean proportion of fixations on objects of the target color was 0.69, revealing a more equal distribution of saccades across subsets in the absence of information signaling the target's color.Cues also affected the exclusivity of saccadic sequences, with observers switching between subsets more on dualthan single-target searches.On the former, switches occurred an average of one, two, and three object fixations in displays containing four, eight, and 12 objects, respectively.These figures show observers switched between subsets after an average of two object fixations during dual-target searches.On single-target searches, the corresponding averages reduced to 0.5, 1.1, and 1.6, indicating far fewer switches between colored subsets.These results demonstrate cue-related differences in the spatial selectivity and order of saccades in response to changes in the specificity of top-down information about the target's color during search.
In addition to the above, support for the interaction over the additive model in our analyses provides strong or extreme evidence for age-related differences in the benefits afforded by single-target cues compared to dual-target cues at the guidance stage of search.The reduction in SPRs on single-target searches compared to dualtarget searches was smaller in OA than YA, revealing an age-related decrease in the benefit of top-down information on the distance travelled by the eyes prior to target selection.OA were also more likely than YA to fixate objects in the irrelevant subset on singletarget searches and to switch between subsets of objects during sequences of saccades.These findings evidence a relative decline in OA's use of top-down information to restrict sequences of saccades to the relevant subset of objects during single-target search.Complementary effects were observed on z-transformed latencies, with the relative benefit of top-down information on the speed of target acquisition on single-target searches smaller in OA than YA.The time between initial and first target fixations provides a temporal measure of attentional guidance (Castelhano et al., 2008;Hamblin-Frohman et al., 2022;Zhao et al., 2012), and comparisons of z-transformed latencies evidence effects over-and-above those associated with generalized slowing (Salthouse, 2000).The decrease in the benefit afforded by single-target cues on the speed and accuracy of saccades in OA compared to YA, therefore, is consistent with a reduction in the use of top-down information during the planning and execution of saccades during the guidance stage of search.
In contrast to the above, comparisons of DLs supported the additive over the interaction model, indicating a comparable benefit in the time required to evaluate fixated targets on single-target searches compared to dual-target searches for YA and OA.Decision latencies for YA and OA were equally sensitive to top-down information about the target's color, with single-target cues speeding the verification stage of search for both groups.Accuracy for OA was also higher on single-than dual-target cues, highlighting the ability of OA to use top-down information about the target's color to inform their evaluation of foveated objects.Comparisons of initial saccadic latencies also support an age invariant effect of set size across singleand dual-target searches.Variability in the latency of initial saccades is thought to reflect the time taken to instantiate the attentional template as well as perceptual processes that parse the scene based on contrasts between visual features (Malcolm & Henderson, 2009).In the present study, the use of a constant interval between the cue and onset of the search display may have reduced the sensitivity of our task to age-related differences in the time taken to encode the target's features within the attentional template.Evidence against effects of age and cue condition, however, suggests that the incorporation of color information within the attentional template has little or no impact on the relative speed with which YA or OA initiate eye movements when the display appears.
The results above support an age-related reduction in the use of top-down information that is specific to the guidance stage of search.In our study, we also manipulated set size.Mean RT-by-set size slopes for displays containing two, four, and eight objects were 199.74 and 209.23 ms per item for YA and OA, indicating inefficient search for an orientation singleton among heterogenous distractors (Wolfe, 1998).Our results also reveal reciprocal relationships between set size and SPRs and the accuracy and exclusivity of sequences of saccades during search.Set size was also positively associated with the latency of initial saccades, first target saccades, and target decisions.These findings are consistent with an increase in the perceptual and cognitive demands of search as the number of objects in the display gets larger.(Bacon & Egeth, 1997;Egeth et al., 1984;Zohary & Hochstein, 1989).Support for an ageby-set size interaction on the latency of first-target fixations in our analyses is also consistent with previous evidence that set size effects are larger in OA than YA (Plude et al., 1983).
The findings above are consistent with previous evidence of an age-related reduction in the accuracy and speed of conjunction searches (Hommel et al., 2004;Humphrey & Kramer, 1997;Kramer et al., 1999;Müller-Oehring et al., 2013;C. C. Williams et al., 2009).Targets in our displays, however, were defined by a unique value on a single feature (orientation), with cues used to signal the relevance of subsets of objects on a second-order feature (color).Evidence of age-related differences in the accuracy and speed of saccadic sampling, therefore, provide a direct test of top-down guidance in the absence of requirements to bind or evaluate combinations of features against the attentional template during search (Becker et al., 2020).Color provides a highly saliant cue for scene segregation (Anderson et al., 2010;Bundesen & Pedersen, 1983;J. Lee et al., 2018), and observers in our study could distinguish red and blue objects prior to testing.OA and YA were also equally sensitive to color information during the verification stage of search, ruling out a general sensory decline in OAs' ability to distinguish between subsets of objects in the display.Both OA and YA also preferentially sampled the cued subset on single-target searches, with group differences indicating a reduction rather than the absence of color-based selection during the guidance stage of search.This may reflect a decrease in the strength, or an increase in the variability of the control signal linking the attentional template on each search to the planning and execution of eye movements to objects in the display (Ramzaoui et al., 2021;Yu et al., 2023).Previous research has revealed an inverse relationship between VSTM and RTs when observers are required to search a specified subset of locations and ignore objects elsewhere in the visual field.This has been interpreted as evidence of an association between VSTM capacity and the maintenance of proactive control, which preferentially weights goal-relevant visual input during the preparation and execution of action (Poole & Kane, 2009;Sobel et al., 2007).Age-related reductions in the functional integrity of the frontal parietal network (Menegaux et al., 2020;Paxton et al., 2008) and the ability to maintain proactive control during the planning and execution of eye movements (Cassavaugh et al., 2003;Nieuwenhuis et al., 2004;Peltsch et al., 2011) provide a potential mechanism for the decrease in OA on the accuracy and speed of saccadic sampling during the guidance stage of search in our study (McCarley & Mounts, 2017;Schwarzkopp et al., 2016).The decline in the use of top-down information despite awareness of the information provided by the cue is also consistent with previous evidence of "goal neglect" in OA, which has been characterized as the disregard of demands that are understood but not adhered to during tasks requiring the maintenance of goal-directed behaviors (Duncan, 1995;Jong, 2001).Alternatively, age may increase the effort required or reduce the strategic benefit of top-down control via the attentional template in VSTM during the planning and execution of saccades.
The current findings provide evidence of age-related reduction in the use of top-down information during the guidance stage of search.This decreases both the spatial and temporal exclusivity of sequences of saccades in OA and increases the time taken to fixate the target in displays containing subsets of task-relevant and irrelevant objects.Our findings also indicate the use of top-down information at the verification stage of search is preserved in OA.This distinction between different stages of search provides important insights into the impact of aging on processes that occur before and after target selection.In the present study, single-and dual-target cues were selected at random on each trial, preventing top-down guidance based on implicit or explicit knowledge about the target's identity over successive trials.The unpredictable nature of the target on each trial requires the dynamic updating of the attentional template between trials (Kadel et al., 2017;Plude et al., 1983).Results from studies that have investigated search in displays in which target identity is predictable, or semantically congruent with regions of the scene, indicate older adults may be able to use this long-term information to implement top-down guidance during search (Madden et al., 2004(Madden et al., , 2005;;Wynn et al., 2020).The contrast between these findings and our own suggests that the extent that older observers use top-down information during guidance is likely to reflect their ability to apply prior knowledge to predict the target's location across multiple searches.Our data demonstrate a reliable, age-related decline in the use of top-down information in VSTM during the guidance stage of search.This is likely to impair guidance in situations that require the dynamic updating of target information and the flexible allocation of resources in response to changing task demands.Future research to investigate the boundary conditions in which this age-related decline impacts the efficiency of search in everyday life will require studies that manipulate the observers' ability to use prior knowledge to augment information in VSTM, as well as measures that are sensitive to real-time changes in the neural signals that reflect the maintenance of proactive control during the programming and execution of saccades in response to different types of top-down information.
AGING AND TOP-DOWN SELECTION IN VISUAL SEARCH an equal number of targets with gaps at 90°or 270°.Displays contained one target and an equal number of red and blue C-shapes.

Figure 2
Figure 2 Mean Proportion of Correct Responses by Group, Search Type, and Set Size

Figure 4
Figure 4 Mean Scan Path Ratio by Group, Search Type, and Set Size

Figure 5
Figure 5 Mean Proportion of Saccades Directed to T, DS, and DD Distractors by Group, Search Type, and Set Size

Figure 6
Figure 6Mean Proportion of Subset Switches by Group, Search Type, and Set Size 8

Figure 7
Figure 7 Mean z-Transformed Initial Saccade Latencies by Group, Search Type, and Set Size

Figure 9
Figure 9 Mean z-Transformed Decision Latencies by Group, Search Type, and Set Size

Figure 8
Figure 8 Mean z-Transformed Target Saccade Latencies by Group, Search Type, and Set Size 10BARRETT, HUTCHINSON, ZHANG, XIE, AND WANG Approval for the study (The Effects of Age on Different Components of Visual Search in Static and Dynamic Displays) was obtained from the School of Psychology Ethics Committee at the University of Leicester and the Academy of Psychology and Behavior Ethics Committee at Tianjin Normal University.

Table 1
Observer Characteristics in Young and Older Adult Groups Note.YA = young adult; OA = older adult; MoCA-BJ = Montreal Cognitive Assessment Scale.