Spatial attention in three-dimensional space: A meta-analysis for the near advantage in target detection and localization

Studies have explored how human spatial attention appears allocated in three-dimensional (3D) space. It has been demonstrated that target distance from the viewer can modulate performance in target detection and localization tasks: reaction times are shorter when targets appear nearer to the observer compared to farther distances (i.e., near advantage). Times have reached to quantitatively analyze this literature. In the current meta-analysis, 29 studies (n = 1260 participants) examined target detection and localization across 3-D space. Moderator analyses included: detection vs localization tasks, spatial cueing vs uncued tasks, control of retinal size across depth, central vs peripheral targets, real-space vs stereoscopic vs monocular depth environments, and inclusion of in-trial motion. The analyses revealed a near advantage for spatial attention that was affected by the moderating variables of controlling for retinal size across depth, the use of spatial cueing tasks, and the inclusion of in-trial motion. Overall, these results provide an up-to-date quantification of the effect of depth and provide insight into methodological differences in evaluating spatial attention.


Introduction
The ability to process stimuli appearing within our complex visual environment remains a crucial area of research.Extensive research has explored how human spatial attention appears allocated in twodimensional (2-D) space.It has been shown that, at any given moment, the level of attention forms an attention gradient which peaks at the location of attention focus and gradually reduces for locations away from that focus.These studies ignited various metaphorical examples in an attempt to explain the distribution of attention, such as an attentional spotlight (Posner, 1980), a zoom-lens (Eriksen and St. James, 1986), or an attentional gradient (Downing and Pinker, 1985).However, only a limited number of research have accounted for the allocation of attention across three-dimensional (3-D) space (Chun and Wolfe, 2001).
The inclusion of 3-D space in attention research became popularized following early primate literature.Original primate research began to investigate how depth influenced spatial attention by examining the visual processes underlying the space around the body or at other focal points in the visual environment (Trevarthen, 1968).Specifically, various psychophysical tasks revealed that distinct neural substrates are representative of near-space processing (e.g., area VIP, Duhamel et al., 1997;area 7b, Leinonen et al., 1979) and far-space processing (e.g., area 8, Colby et al., 1996).Berti et al. (2001) summarized the primate literature and began to relate the vast neurophysiological research in primates to human research that demonstrated patient populations could show anatomical distinctions for the coding of near and far space also.For example, Weiss et al. (2000) used a line bisection task within human participants across near and far space and showed using positron emission tomography (PET) that distinct neural activations were present specific to both near and far space-this research provided a direct connection from primate studies to humans.
Progressing forward, human behavioural studies continued to examine neurophysiological and behavioural differences when responding to stimuli appearing at near and far egocentric distances (i.e., near versus far stimuli).Evolutionarily, it would appear sensible to distribute greater attentional resources within near space to serve to protect the observer from potentially unexpected events and allow us to act promptly (Graziano and Cooke, 2006;Rizzolatti et al., 1997).Researchers have included this near space bias in various proposals for the allocation of attention across 3-D space, such as the behavioural urgency hypothesis (Franconeri and Simons, 2003), the self-prioritization effect (Huang et al., 2022;Liu et al., 2021;Rogers et al., 1977), or a viewer-centered (Andersen, 1990;Andersen and Kramer, 1993;Arnott and Shedden, 2000), body-centered (Finlayson and Grove, 2015;Plewan and Rinkenauer, 2016), or an egocentric attention gradient (Karnath, 1997;Plewan and Rinkenauer, 2020;Spaccasassi et al., 2019).These models all comprise the same central idea: attention is prioritized to stimuli within near space compared to far space.
When finding advantages within near space, researchers examined the effect of depth by manipulating egocentric distance through depth cues.Many depth cues exist that can be experimentally manipulated to simulate near and far spaces.One way to afford perceived depth is stereopsis (induced by binocular disparity) where an individual receives slightly altered visual images on one eye compared to the other eye as they are separated by a short distance.This separation in distance allows for the shifted retinal images to produce a reliable disparity signal that helps observers perform tasks such as depth ordering or flow parsing (Warren and Rushton, 2009;Wilcox et al., 2005).Stereoscopic vision can be described separately from the perception of depth in real space (e. g., free viewing) where observers are free from any physical vision manipulations (e.g., shutter glasses).In addition to stereoscopic vision, the alternative way of manipulating depth is by using pictorial depth cues.Pictorial depth cues necessitate only the use of monocular vision to perceive changes across 3-D space.Compared to stereoscopic vision with binocular disparity, monocular viewing conditions can provide better depth discrimination as the spatial range increases past a few meters (McCann et al., 2018).Popular pictorial cues in the literature have included linear perspective, texture gradients, shading and changes in brightness, and object occlusion.Collectively, these methods have demonstrated effective strategic choices for examining how depth can influence cognitive processes- Cutting and Vishton (1995) comprehensively review the human ability to use various depth cues and contextual information to perceive 3-D layouts.
Furthermore, human spatial attention literature in 3-D space has demonstrated distinct behavioural differences across near and far space.The effect of depth has been demonstrated across behavioural tasks including enhanced visual search performance for foreground objects compared to background objects (e.g., Fernandes and Castelhano, 2021), better temporal order judgements for perceived nearer compared to farther surfaces (e.g., Lester et al., 2009;West et al., 2013).One rich set of literature on this topic is the effect of depth in 3-D space for target detection and target localization (See Table 1).

The near advantage for target detection and localization
Over the last forty years, the effect of depth or distance in modulating target detection and localization has become a popular topic.In addition to the effect of depth per se, early research in 2-D space proposed that attention is biased toward stimuli appearing in the lower visual field (Christman, 1993;Nasr and Tootell, 2020;Soballa et al., 2022).Compared to the upper visual field, the lower visual field often corresponds to closer regions of space in depth because typical object affordances exist below eye level (Danckert and Goodale, 2004;Previc, 1990Previc, , 1998;;Szpak et al., 2015;Tipper et al., 1992).However, to study the true effect of depth, experiments often involve the use of targets appearing at the same retinal location across different egocentric distances (along the z-axis).
In the target detection task, participants typically produce one response (e.g., press the space bar, click the mouse) to indicate the onset of the target (regardless of the location).For target localization tasks, participants are required to report the location of the target by selecting one of two responses (e.g., left button for left targets and right button for right targets).For target detection responses, the localization process is likely involved although participants do not need to indicate the location.
A predominant behavioural measure used in the literature has been reaction time (RT).The effect of depth can be quantified by RT differences for conditions where targets appear in either near or far space for target detection and localization tasks.Studies investigating the effect of depth in RT have become a fundamental approach to understanding the distribution of attention in 3-D space.Despite the abundance of literature examining the effect of depth, the literature has yet to be quantitatively analyzed.In addition, the near advantage in target detection and localization has appeared to be widely accepted without further inquiry into the different experimental parameters (e.g., target retinal size) that may generate differences in behavioural outcomes across depth.
In this study, we performed a meta-analysis of the literature to date on the effect of depth in target detection and localization tasks.For the main analysis, we identified five primary factors to be critical for the analysis: (1) was the participant required to localize or just simply detect the target stimuli, (2) the use of a spatial cueing paradigm versus simple (uncued) target detection/localization, (3) whether targets presented in near and far space conveyed identical retinal size, (4) if targets were presented in peripheral or central visual fields, and (5) whether the environment that provided depth information involved full depth cues (i.e., as in real space), stereopsis, or monocular depth cues.These five factors are categorical and were used for our primary moderating factor considerations.We also considered three secondary factors of the extent

Table 1
Target detection and localization studies examining the effect of depth. of depth/distance separation that were continuous.These included the relative near versus far distance/depth separation, the absolute near versus far distance/depth separation, and the absolute far distance/ depth measurements.Lastly, we conclude with a special moderator consideration for studies that do not involve a stationary viewpoint but instead use viewpoint motion at the time of target onset.This was considered to observe if the near advantage substantially changes as a function of viewpoint perspective.

Method
Studies were identified from a literature search across the PsycINFO and Web of Science electronic databases until March 21st, 2024.The search terms used were ('attention') AND ('target detection' OR 'target localization' OR 'spatial cueing') AND ('distance' OR 'depth').The search was refined using literature written in English.In addition to the database search, three studies from our lab that were in the publication process at the time of the literature search were included in the selection process (Britt et al., 2024a;Haponenko et al., 2024aHaponenko et al., , 2024b)).We did receive two additional studies from an email correspondence that did not appear within our search (Qian et al., 2023;Sperandio, 2021).After removing duplicates, 1099 abstracts were screened for selection using the 'study selection' procedure described below.The full text of 90 articles was examined to assess eligibility.After the assessment, we first analyzed 24 studies (including a total of 49 datasets) involving the stationary viewpoints of the participants.These 24 studies were subjected to our meta-analytical model and moderator analyses.In addition, 5 other studies (9 datasets) involving self-motion of the viewpoint were included for a special moderator analysis of motion-datasets were excluded from the three experiments by Andersen et al. (2011) and the first experiment by Song et al. (2021) given their depth manipulation resulted in a difference in the probability of target appearance between near and far depth.Finally, we completed an additional analysis based on a total of 29 studies (24 stationary + 5 motion) (58 datasets) with a stationary or moving viewpoint.
Pertaining to the included studies, if the study did not report or plot the relevant data, we obtained additional information via email correspondence from various authors to assist in calculating relevant effect sizes (Losier and Klein, 2004;Reppa et al., 2010;Song et al., 2021;Sperandio et al., 2009;Van der Stoep et al., 2014;Wang et al., 2016).Additional studies could have been eligible but the necessary data required for our meta-analyses were either unavailable or author contact was unsuccessful (Atchley andKramer, 2000, 2001;Casagrande et al., 2012;Chan et al., 2012;Dvorkin et al., 2012;Gawryszewski et al., 1987; Fig. 1.Flow diagram of the electronic database search and process of the literature analysis.Adapted from the Preferred Reporting Items for Systematic Reviews and Meta-Analyses: The PRISMA Statement (Moher et al., 2009).Heber et al., 2008a;Iavecchia and Folk, 1994;Kasai, 2008;Liu and Rau, 2020;Liu et al., 2023;Maringelli et al., 2001;Plewan and Rinkenauer, 2017;Robertson and Kim, 1999).
Studies that did not involve RT as their dependent measures, used target motion, or were not a detection or localization response (e.g., discrimination) were also revealed in the search but excluded after fulltext assessment.All of these target detection and localization studies that were either included in the analysis or not are listed in Table 1.
Additional criteria that resulted in study exclusion were incompatible reference frames for depth (e.g., manipulations in reference to the observer's hand), different response modalities (e.g., verbal response), alternative tasks (e.g., visual search, target discrimination), and those that involved target motion during a trial.We did not contact authors regarding unpublished studies outside of our lab.The PRISMA process is documented in Fig. 1, and the way of reporting items for meta-analyses laid out by Moher et al. (2009) was followed.

Study selection
Additional and detailed criteria for the inclusion of the dataset in the meta-analysis were: 1. Healthy Participants.2. Task: To accurately analyze studies measuring attention we included datasets that specifically examined reaction time (RT) to the sudden appearance of target stimuli.Specifically, the response requirement needed to be a detection/localization response (i.e., button press, button release).This included the use of target detection paradigms, and spatial cueing paradigms with a detection or localization response.3. Visual Target Stimuli: Studies were included only if the target response was to a stationary visual target stimulus.Other target presentation modalities were not included.4. Depth/Distance Manipulation: Studies were included if they manipulated target depth along the depth axis (z-axis).This included changes in viewing distance or target depth to induce a 'near' target condition compared to a 'far' target condition.A near versus far comparison was required as a within-subject factor in the chosen studies.For studies reporting more than two egocentric distances (e. g., Pierce and Andersen, 2014), we chose 'near' and 'far' distances that most closely resembled other selected studies to attempt to be consistent.5. Age: Average or median ages were below 35.6. Language: Only studies written in English were included.

Moderators
A. Primary Moderator Analysis.The following primary moderator analyses were performed within multivariate models.These models were performed first and independently of the subsequent secondary moderator analyses.

Task: Detection versus Localization. Both detection and localization
responses are believed to recruit similar processing requirements that elicit similar effects on attention (e.g., Prime et al., 2006) and have been grouped together in reviews (Chica et al., 2014).While the literature has often equated the underlying processes when performing target detection (i.e., give a simple button press upon target onset) and target localization (i.e., respond with the corresponding button press that matches the location of the target stimulus), we wanted to see what differences exist within the effect of depth across the literature.It may be that target localization could evoke increased egocentric consideration given the enhanced task considerations for the actual target location-localization could be different from target detection which just requires noticing a sudden abrupt onset of a visual target and a simple manual response irrespective of location.2. Spatial Cueing versus Uncued Tasks.The second consideration that may be critical across studies reporting near versus far target detection differences was the respective use of two different experimental paradigms: simple target detection vs cue-target spatial cueing.Simple target detection studies have simply presented one target stimulus in depth and asked the observer to respond accordingly at each target onset.The original spatial cueing paradigm (Posner and Cohen, 1984) has been modified and adopted in studies investigating spatial attention differences across depth (e.g., Britt et al., 2024a).In a typical 3-D procedure involving spatial cues, a spatial cue would appear at a specific location in depth, then a temporal delay would occur, and finally, the task-relevant target would appear for the observer to respond, either in the same depth as the cue or a different depth (e.g., Casagrande et al., 2012).Given the potential mechanistic differences across paradigms, it remains important to consider whether spatial cueing could attenuate or enhance the near advantage for spatial attention-this could also include audio cues as long the target remained visual (e.g., Van der Stoep et al., 2014).3. Control for Retinal Size.When target stimuli are presented at varying distances/depths across space, researchers have the choice to either control for the retinal size or the apparent (true) size of the targets, across depths.By presenting target stimuli across depths with equal retinal size, the detection performance would not be confounded by the stimulus' low-level visual property (e.g.retinal size).Thus, the effects of visual saliency would be equal across both near and far space.But the interpretation of the scene would be that the far target is presumably larger physically than that of the near target.This might in turn potentially bias the observer to respond faster to a far/perceived larger target and consequently underestimate the near advantage.Conversely, the apparent size of the target could be made to be equal across depth (implying the same physical target appeared either near or far) and consequently with larger retinal size in near space than far space.In this case, the near targets appear larger on the retina and thus perhaps be easier to detect/ localize when compared to farther stimuli that appear smaller on the retina.These issues regarding 'size constancy' have been investigated (Sperandio and Chouinard, 2015) in the same setup and remain pertinent to the current analyses when investigating the near advantage in target detection.4. Target Location: Central versus Peripheral.Another important consideration for visual target presentation is whether the abrupt targets appeared in the central or peripheral visual fields.Many spatial attention paradigms involve a central fixation point which can sometimes be replaced by a central target stimulus (either in front or behind the point of fixation) that requires a response.However, in alternative paradigms, observers can be required to shift their attention overtly or covertly toward targets appearing in peripheral locations in near and/or far-depth planes.These two methods could reveal differences in shifting attention toward central or peripheral target locations as a function of depth because visual signals have been shown to not be processed uniformly across the visual field (e.g., Carrasco et al., 1995).
As an example, Pierce and Andersen (2014) presented target stimuli at various eccentricities across different egocentric distances while undergoing an active driving task.Their results showed that target detectability decreased at farther distances, but more importantly, this decrease was largest for the most peripheral target stimuli.Thus, we intend to see whether the analyzed literature demonstrated any differences in the effect of depth/distance as a function of central or peripheral target presentation.5. Environment Used for the Distance/Depth Manipulations.Earlier research has manipulated the presented locations of physical objects across real-depth (e.g., Downing and Pinker, 1985) or manipulated the distance to the target display by shifting the display device to be closer or further to the participant (e.g., Li et al., 2011).However, other studies have used single displays to simulate depth by presenting stereoscopic information (e.g., Theeuwes and Pratt, 2003) or monocular depth information in a simulated 3D environment (e.g., Haponenko et al., 2024a).It would appear reasonable that visual processing differences exist across virtual and real space that could potentially modulate the near advantage for spatial attention.Specifically, distance/depth in real space tends to provide absolute distance which tends to be in the range of a laboratory environment.However, virtual spaces with only monocular information could simulate rather ambiguous distance/depth that could be difficult to perceive.Alternatively, a larger distance range can be simulated if there is enough contextual (semantic) information (e.g.simulation of highway environment with road and other cars to serve as a reference of the scale of the environment).
While virtual spaces allow for near versus far comparisons to be made in a rather relative term as potential ambiguity in the absolute range of depth, studies using real space can measure the effect of depth for near versus far distances within peripersonal space (or with a far distance in extrapersonal space).We thus reviewed literature examining the effect of depth with absolute depth/distance to compare performance across what the researchers have subjectively chosen as their 'near', 'middle', or 'far' depths/distances.

B. Secondary Moderator Analyses.
The following secondary moderator analyses were performed subsequent to the primary metaanalyses.While the primary moderator analyses focused on the relative comparison between near and far target conditions, the secondary analyses were intended to specifically quantify the effect of the magnitude of depth differences.Secondary analyses were analyzed across different models depending on the depth measurement used for the experimental investigation.The three secondary moderator variables were the relative (Far/Near) distance/depth separation, the absolute (Far-Near) distance/depth separation, and the absolute far distance.
All studies that reported their 'near' and 'far' target locations in centimeters (cm) were included.Studies denoting their manipulations using virtual meters (vm) or visual illusions to simulate depth were not analyzed here due to a lack of a sufficient number of datasets and ambiguity in the scale of distance.
C. Special Consideration: Studies Involving Moving Viewpoint.The main analysis in this study was intended to only factor in tasks where participants' viewpoints remain stationary.In another set of literature, researchers implemented physical or simulated motion of the observers' viewpoint during target processing.Unlike the 24 studies using a stationary viewpoint where targets appear in different depth locations, some studies involve targets appearing in the same depth location, but the viewer undergoes simulated viewpoint motion from farther to nearer viewer-to-target distances (e.g., Song et al., 2023).Alternatively, other studies positioned targets in different depth locations but still involved in-trial viewpoint motion (e.g., Kimura et al., 2009).This experimental manipulation is unique as during the forward motion of the viewpoint, in addition to depth information obtained through optic flow, target motion on the retina would also be different across depth.Specifically, near targets would lead to a greater displacement (away from the central visual field) of the image of the target on the retina compared to far targets.The analysis of the literature involving motion will only be briefly described at the end of the results and discussion.

Effect size estimation
Within our meta-analysis, a positive effect size referenced a faster RT to target stimuli appearing in the near location, while a negative effect size referenced a faster RT to target stimuli in the far location.To attain our measure of effect size, standardized mean differences were obtained using Cohen's d rm (Lakens, 2013) for within-subject designs.We assumed a conservative correlation of 0.7 (r = 0.7) within all studies for the correlation across depth-we also planned to conduct a subsequent sensitivity analysis across correlation coefficients ranging from 0.2, 0.4, 0.6, 0.8 and 1.0.The effect size and effect size variation were calculated using Formula 1 and 2, respectively, and the correlation variation was calculated using Formula 3:

Statistical analysis: meta-analyses and heterogeneity
The robumeta (Fisher et al., 2017) and metafor package (Viechtbauer, 2010) were used in R to perform our meta-analysis.To begin, effect sizes (as described above) and measures of variances were computed for each study.To investigate the effect of spatial attention in near versus far target conditions we used a multivariate model with correlated effects using the robust variance estimation approach (RVE; Hedges et al., 2010),also using a small sample correction to all measures of effect size (Tipton, 2015).The RVE is incredibly accurate in generating mean effect size and confidence interval estimates.
The overall meta-analytic model included all studies with a stationary viewpoint to assess the overall impact of near versus far target conditions on RT.The heterogeneity indexes of I 2 and τ 2 were computed.
Then, to assess the effects of our five primary moderator variables, we added the moderator variables into the multivariate model.These models allowed for accurate testing of the moderator effects of task type, spatial cueing, retinal size, target location, and depth cues.Subsequently, secondary moderator variables were analyzed.Across all moderator variables, the five primary variables were all categorical (or factors) while the three secondary variables were treated as continuous.
Important to note that for the primary moderator variables, data were also then split across the two or three levels of the associated moderator.Then with the individual models representing each level of a moderating factor, we analyzed if each of the individual models resulted in significant near advantages-this method still used multivariate meta-analytic modelling.

Publication bias
As previous research has shown no single publication bias method outperforms others (Carter et al., 2019;McShane et al., 2016), we used two methods that have been used to quantify publication bias in recent meta-analyses (Holgado et al., 2023;Román-Caballero and Lupiáñez, 2022), the Egger's regression test (Egger et al., 1997) for funnel plot asymmetry, and also the skewness test (Lin and Chu, 2018).To avoid the artifactual dependence between Cohen's d rm and its precision estimate (Pustejovsky and Rodgers, 2019), we converted Cohen's d rm into Fisher's z (Borenstein et al., 2021) where d rm scores were converted to correlation coefficients, and then the correlation coefficients 'r' were converted to z values using Formula 4, and standard error using Formula 5: If evidence of potential publication bias is found the trim and fill method will be used to reveal a corrected effect size and confidence interval.

Meta-analytic power analysis
Given the disparity in datasets across certain levels of our moderator analyses, power analysis for the 58 included datasets necessitated inclusion to give insight into the level of power associated with each individual effect size estimate.Using the metameta package (Quintana, 2023), estimates of statistical power were calculated by using the range of hypothetical effect size estimates and the meta-analytical effect size estimate (of the combined stationary and motion datasets) as the true observed effect size.

Overall effect size estimate
Our overall multivariate model with correlated effects using RVE used a total of k i = 24 studies involving stationary viewpoints and a total of k d = 49 separate datasets and their respective effect sizes.These datasets comprised n = 1018 participants.The model resulted in a significant near advantage d rm = 0.13 (p =.003), with a 95 % confidence interval of [0.05, 0.22].The model also included moderate to high heterogeneity (I 2 = 64.17%, τ 2 = 0.05).
Despite using a correlation of 0.7, sensitivity analysis was conducted using the correlation coefficients of 0.2, 0.4, 0.6, 0.8, and 1.0.Sensitivity analysis revealed that identical model parameters were present regardless of the correlation coefficient used (d rm = 0.13, τ 2 = 0.05).
The included studies are listed in the forest plot (Fig. 2) with information regarding the respective moderator variables, and specified

Bias
Publication bias using the Egger's regression test revealed no evidence of potential publication bias (b 1 = − 0.281 [-0.77, 0.21], t(44) = 1.45, p =.153).However, the skewness test revealed potential evidence of publication bias (T s = 1.00 [0.54, 1.48], p =.004).Given that one of the two tests for publication bias revealed a significant result, we used the trim and fill method to observe a new model estimate with a publication bias correction.The trim and fill correction (Duval and Tweedie, 2000) revealed no missing studies using the L0 estimator, but 5 missing studies to the left of the effect size measurement with the R0 estimator.If these 5 additional studies were added to the overall model from the R0 estimator, the new model estimate would be d rm = 0.03 [-0.08, 0.14].A funnel plot for asymmetry is plotted in Fig. 3 with the illustrated study correction for the trim and fill method.

Analysis of primary moderator variables
Important to note that across all five primary moderator analyses, 49 datasets were included in the analyses.Fig. 4 shows the effect size results of all the primary moderator variable analyses.As for the secondary moderator analyses, just 47 datasets were used (see methods).

Task: localization versus detection
A multivariate model was applied including the moderator variable of task type: target detection (k d = 44) versus localization (k d = 5).The inclusion of the moderator revealed no effect on the task type (p =.174).By separating the data across both levels of the moderator variable, the detection model elicited a significant near advantage (d rm = 0.15, [0.05, 0.25], p =.004).However, the localization model did not reveal a significant near advantage (d rm = 0.05, [-0.06, 0.16], p =.247).

Spatial cueing versus uncued tasks
A multivariate model was applied including the moderator variable of whether spatial cueing was used (yes, k d = 34) or not (no, k d = 15).The inclusion of the moderator revealed a significant effect of the use of spatial cueing (p =.013).This significant moderating effect was driven by differences across the data using spatial cueing and the data not.By separating the data across both levels of the moderator variable, the spatial cueing model elicited a significant near advantage (d rm = 0.19, [0.08, 0.30], p =.002).However, the model without spatial cueing did not reveal a significant near advantage (d rm = -0.02,[-0.12, 0.08], p =.698).

Control for retinal size
A multivariate model was applied including the moderator variable of retinal size being controlled for (yes, k d = 19) or not controlled for (no, k d = 30).The inclusion of the moderator revealed a significant effect of the control for retinal size across depth (p =.002).This significant Fig. 3. Funnel plot for the overall model of stationary studies.Plotted are standard errors across the Fisher's z. White circles represent the 5 studies to be added to the left of the mean to achieve a symmetrical model.The figure was computed with the trim and fill method.The vertical dashed lines represent the effect size, corrected for publication bias, with a pseudo-confidence interval region of +/-1.96 standard error.moderating effect was driven by differences across studies controlling for retinal size or apparent size.By separating the data across both levels of the moderator variable, the model including studies that did not control for retinal size elicited a significant near advantage (d rm = 0.24, [0.11, 0.36], p <.001).However, the model for studies that did control retinal size did not reveal a significant near advantage (d rm = − 0.01, [-0.08, 0.06], p =.247).

Target location: peripheral versus central
A multivariate model was applied including the moderator variable of whether targets were presented in peripheral (k d = 31) or central (k d = 18) target locations.The inclusion of the moderator revealed no effect on the moderator (p =.256).By separating the data across both levels of the moderator variable, the central target model elicited a nearsignificant near advantage (d rm = 0.25, [-0.00, 0.51], p =.050), and the peripheral target model did reach a significant near advantage (d rm = 0.09, [0.02, 0.17], p =.015).

Environment: real space, stereopsis, monocular depth
A multivariate model was applied including the moderator variable of whether the experiment manipulated depth/distance in real space (R, k d = 16) or using stereoscopic (S, k d = 26) or monocular depth (M, k d = 7).The inclusion of the moderator revealed no effects on the moderator (p = 468).By separating the data across the three levels of the moderator variable, the model for full-depth cues (real space) trended toward a significant near advantage (d rm = 0.22, [-0.03, 0.48], p =.082), and both the models for stereoscopic and monocular cues revealed significant near advantages, respectively (d rm = 0.11, [0.01, 0.20], p =.035; d rm = 0.10, [0.02, 0.18], p =.025).

Relative distance/depth separation
For studies measuring depth differences in centimeters (k d = 47, the continuous moderator variable of the relative near versus far distance/ depth separation was investigated.For the relative distance/depth separation, the test of the moderator was not significantly associated with the overall effect size (p =.803).

Absolute distance/depth difference
For studies measuring depth differences in centimeters (k d = 47), the test of the absolute difference in distance/depth moderator was not significantly associated with the overall effect size (p =.351).

Absolute far distance/depth
For studies measuring depth differences in centimeters (k d = 47), the test of the absolute far distance/depth moderator was not significantly associated with the overall effect size (p =.171).

Special moderator: the effect of motion
While the main analysis of the present meta-analysis was to quantify the near advantage under stationary viewpoints, other studies analyzed the effect of egocentric distance/depth involving motion of viewpoint in their trials.To analyze studies that used motion in their trials compared to the stationary-viewpoint studies we have analyzed thus far, a model was applied that introduced ten new datasets from seven different studies (k d = 9) (Kimura et al., 2009;Pierce and Andersen, 2014;Song et al., 2021Song et al., , 2023;;Xia et al., 2008Xia et al., , 2009) ) compared to the original stationary datasets in our overall multivariate model (k d = 49).See Fig. 5 for the new motion datasets.
Exclusively when motion was used, the model revealed a large near advantage (d rm = 1.20, [0.61, 1.79]) that was significantly different from zero (p =.001).Also, if we included motion in our overall multivariate model, it would have been composed of a total of k i = 29 studies and a total of k d = 58 separate datasets and their respective effect sizes.The inclusion of motion datasets results in a total of n = 1260 participants.This model resulted in a d rm = 0.29 (p <.001), with a 95 % confidence interval of [0.16, 0.43].
In addition, if motion had been treated as a moderator variable within the overall model including both stationary viewpoint and motion studies, unsurprisingly, the test of moderators would have revealed a significant moderating effect (p <.001).Refer to Fig. 4.

Meta-analytic power analysis
The metameta package (Quintana, 2023) afforded the calculation of statistical power estimates for hypothetical effect sizes ranging from 0.1 to 1.0.The median statistical power that was revealed from the analysis, using the observed overall effect size of d rm = 0.29, was 27 % (min = 8 %, max = 53 %).The entire power matrix across all hypothetical effect size estimates for all 58 included datasets is included online (https://osf.io/q82v9/).

Discussion
The present meta-analysis quantitatively assessed the overall performance of detection/localization of targets within near and far visual space across a variety of studies.Additionally, we felt it necessary to observe if any potential moderating factors greatly impact the aforementioned near advantage for target detection/localization, thus also plausibly affecting the apparent egocentric gradient of spatial attention.
The results of our meta-analysis using studies with a stationary viewpoint revealed a significant overall effect demonstrating a near advantage for spatial attention in target detection and localization tasks.However, this near advantage was not significant across all levels of our moderator factors.The analyses revealed a near advantage for spatial attention that was affected by the moderating variables of controlling for retinal size across depth, possibly the use of spatial cueing tasks, and the inclusion of in-trial We began the moderator analyses by observing if any differences existed across target detection and localization tasks.The current moderator variable analysis showed no significant differences between the two task types (d rm = 0.05 vs d rm = 0.15).It is important to note this analysis was generated off of far fewer localization studies than detection.While we hold the opinion that these two tasks do not differ in how attention should be distributed across depth, any difference statistically would be better quantified with more studies using target localization across depth with a stationary viewpoint.
The second moderator analysis observed if the spatial cueing paradigm would offer unique sensitivity for revealing the depth effect.When spatial cueing was used, it was revealed the near advantage was significantly enhanced compared to when spatial cueing was not used.This likely demonstrates a trend in the direction of spatial cueing paradigms appearing more sensitive in revealing spatial attention differences across near and far space, raising the near advantage compared to uncued tasks (d rm = 0.19 vs d rm = − 0.02).In addition, the models without spatial cueing were essentially a null effect of distance.
From the analysis of spatial cueing, the greater effect size found while involving a more dynamic shift of spatial attention with spatial cueing during the trial requires further discourse.Spatial cueing tasks have been incorporated in experiments to elicit an initial gathering of attention at a given spatial location, followed by a temporal delay, and then the observer performs the subsequent target detection/localization at an abrupt target onset.The original gathering of attention by a cue has been suggested as beneficial in numerous ways, such as by focusing limited attentional resources across more distinct areas of the visual field (Nachmias, 2002), excluding noise or distractor signals (Lu and Dosher, 2008), or by boosting attention toward anticipated target locations (Chen and Wyble, 2018).This proposed method for measuring attention provides an incredible measure that incorporates both attentional engagement, subsequent disengagement, and then further orienting across space (Posner et al., 1987).The use of spatial cueing may still elicit a more sensitive measure for observing the effect of depth by driving participants to allocate attention to a specific depth and then shift attention between different depths or within the same depth.This is different from tasks that do not involve cueing as this would instead force observers to shift attention only to the target location from a more 'neutral' state (i.e., without any initial capture of attention in specific depth).
In addition, while the present analysis focused on RT, spatial cueing tasks also allow for measuring a difference in RT when participants are required to shift attention across depth within a single trial (e.g., shifting attention from a far cue to a near target).These differences in RT (i.e., cueing effects such as facilitation or inhibition of return (IOR)) when shifting attention across depth have also revealed near advantages for target detection and localization tasks (Andersen and Kramer, 1993;Britt et al., 2024a, Downing andPinker, 1985;Gawryszewski et al., 1987;Haponenko et al., 2024aHaponenko et al., , 2024b;;Wang et al., 2016).This demonstrates the spatial cueing task is a better candidate for measuring the effect of depth in 3-D space.
The third moderator variable was if the retinal target size was made to be different across depth (i.e., size-distance scaling), or was controlled to be the same across depth.While both methods exist with various benefits or costs, when retinal size was not controlled for the model effect size was significantly larger compared to when retinal size was controlled for (d rm = 0.24 vs d rm = − 0.01).
By not maintaining the same retinal size, researchers scale the targets to have the same perceived size (i.e., far targets have a reduced retinal image compared to near targets) then near targets may just be easier to detect due to their larger retinal size and likely increased salience (see Osaka, 1976;Teichner and Krebs, 1972), resulting in faster responses.This could explain why RT differences in depth would be more pronounced without controlling for retinal size compared to when near/far targets had the same retinal image.In other words, near-target stimuli appearing larger on the retina (and inherently more salient) compared to far-target stimuli could 'boost' the near advantage for target detection.
Conversely, by maintaining the same retinal image in the near and far depths, one introduces a perceptually larger far target.Larger perceived targets have been shown to facilitate RT (Plewan et al., 2012;Savazzi et al., 2012;Sperandio et al., 2010), including targets appearing larger in perceived size at farther distances/depths compared to near distances/depths (Sperandio et al., 2009).This could indicate the reason the near advantage was not significantly greater than zero when retinal size was controlled for could be due to some participants allocating increased attention to perceived larger far targets instead of the perceived smaller near targets.
The fourth moderator analysis was to see whether peripheral or central target detection was more indicative of spatial attention differences across depth.The separate models revealed near advantages for studies using both peripheral targets and central targets (d rm = 0.09 vs d rm = 0.25) across near and far space.It is important to note that individual eccentricity differences were not factored into the model given the nature of central targets has no deviance in horizontal visual angle, and the peripheral target detection studies have a multitude of eccentricities that were sometimes averaged if multiple eccentricities were presented in the same study.
An important consideration for future studies is that central and peripheral target presentation would not necessarily elicit comparable near advantages unless eccentricities are carefully considered.The literature shows that visual processing areas such as area V1 devote approximately 25 % of the cortex to central processing (De Valois and De Valois, 1988) and that peripheral visual fields rely on differential visual processing areas (see Carrasco, 2011 for a review).The slightly larger effect size measurement for central targets may be potentially due to the ease of attending and processing targets aligned with central fixation compared to peripheral targets that could become more difficult to attend with increasing eccentricity-and this effect of eccentricity has been demonstrated to be greater with increasing egocentric distance (e. g., Pierce and Andersen, 2014;Song et al., 2021).Further examination should be conducted into central versus peripheral target presentations in 3-D space with comparable task difficulty across eccentricities within the same study.
The fifth primary moderator variable was whether the study occurred in an environment that manipulated depth using full-depth cues (i.e., real space), stereoscopic depth cues, or monocular depth cues.The findings revealed no significant differences across the three models for the respective depth cues, however, full-depth cues failed to reach significance.Monocular depth cues generating a significant near advantage suggests the potential advantages of using monocular depth cues to represent a larger depth range for near and far target stimuli compared to real-space and stereoscopic depth manipulations.After all, stereoscopic depth cues are not very informative for larger depth ranges in 3-D space.Although these studies using monocular depth cues do not offer absolute depth information, they still offer a near versus far manipulation and with a possibility for a greater extent across depth.Within these virtual environments 'near' versus 'far' is a rather subjective variable.Any distance relative to another may be considered 'nearer' or 'farther' with respect to the other locations.In addition, by simulating depth in virtual environments, researchers are able to explore an increased number of scenarios that would apply to real-world behaviours in 3-D space (e.g., driving).
Complimentary to the idea of virtual spaces assisting studies investigating attention and perceptual processing across depth, recent research has also shown that images presented to participants that appear to encompass only a 'reachable' space (e.g., a bathroom vanity) versus a full-scale view (e.g., an entire house layout) are perceptually different, with neural network models suggesting mid-(e.g., object textures, surfaces) to high-level stimulus features (e.g., object parts, entire objects) potentially being responsible for the dissociation (Josephs and Konkle, 2019;and see Castelhano and Krzyś, 2020).Thus, it may be more valid to present observers with experimental contexts that encompass full-scale views where near versus far comparisons can be selectively modified-even if these distances exist purely outside peripersonal space.
Finally, we conducted a special moderator analysis to investigate other paradigms that included motion when measuring near versus far target conditions.In studies that included in-trial viewpoint motion, the effect size estimate was incredibly high compared to the overall model with just stationary viewpoint studies (d rm = 1.20 vs d rm = 0.13)-but both remained consistent with a significant near advantage.
Importantly, it is worth noting that motion conditions likely resulted in increased vigilance (or behavioural urgency) while detecting targets that appear similar to looming conditions (Rossini, 2014;Sugarman et al., 2021), resulting in the overall heightened near advantage potentially as a result of enhanced functional connectivity of brain regions correlated with 'looming processes' (Tyll et al., 2013).We believe a similar explanation can likely also account for the large spread in the target localization effect size because, in a number of the experiments involving motion, it was either explicitly a driving-related/simulated experience or perhaps was treated as similar by the participant with oftentimes a localization response.Previous research has shown that increased cognitive load can influence target detection in driving-like conditions (Lee et al., 2007), perhaps leading to behavioural changes as a function of driving experience (Konstantopoulos et al., 2010).Implications for a near advantage within driving contexts may be evident within these studies which could have led to increasingly variable RT differences.However, the overall inclusion of the motion moderator variable should be considered for future detection experiments as it may introduce several other relevant variables that could exacerbate or diminish any observable RT differences due to confounding factors related to the task (e.g., driving experience).
Furthermore, across all of the analyses, this near advantage for spatial attention represents a behavioural difference across depth that has oftentimes been related to the neurophysiological dorsal and ventral visual processing systems (Ungerleider and Mishkin, 1982).The neurophysiological distinction termed the two visual systems hypothesis (Goodale and Milner, 1992;Milner and Goodale, 2008), has been shown to represent functional differences across near and far space (e.g., Heber et al., 2008b;Weiss et al., 2000Weiss et al., , 2003)).Specifically, the dorsal stream has been related to near-vision deficits (Berti and Frassinetti, 2000;Halligan and Marshall, 1991;Guariglia and Antonucci, 1992;Mennemeier et al., 1992;Rapcsak et al., 1988) and the ventral stream has been related with far-vision deficits (Aimola et al., 2012;Butler et al., 2004;Cowey et al., 1994;Shelton et al., 1990;Vuilleumier et al., 1998).Related specifically to a potential near-space advantage demonstrated across the current paper, the dorsal stream has demonstrated significant activations for egocentric coding of the visual environment (Committeri et al., 2004), including peripersonal object processing (e.g., Valdés--Conroy et al., 2012, 2014), near-space grasping tasks (Culham et al., 2003;James et al., 2003), and most importantly for the present paper, target detection and localization tasks (Corbetta et al., 1993;Haxby et al., 1991;Marrett et al., 2011;Rao et al., 2003), including areas responsible for spatial cueing (e.g., Seidel Malkinson et al., 2024).It would appear sensible to suggest this apparent near advantage for attention that was robust across the present analyses could represent the combined processing requirements within the dorsal visual system for (1) stimuli appearing within near space, and (2) the task requirements being target detection/localization.Importantly, we realize limitations exist and various studies are essential to consider for future depth manipulations even if they are outside our analysis.For example, the current study limited the age of participants to be included to relatively young adults instead of any older populations.However, similar near advantages have been found in comparable tasks in older adults (Pierce et al., 2011;Pierce and Andersen, 2014).Another limitation includes our choice of measurement in RT as accuracy represents another popular behavioural measurement across 3D space.Similar to the findings of the meta-analysis, accuracy benefits have been observed in near space for target detection/localization tasks (e.g., Li et al., 2011;Quinlan and Culham, 2007).Also, our chosen 'near' and 'far' distances were sometimes selective to be comparable across the present analysis and excluded other utilized egocentric distances in certain studies that could have varied in RT.Despite our secondary moderator analysis for distances being insignificant, it may be useful for other studies to consider more levels of egocentric distances that were manipulated in certain studies that were not synonymous with choices based on a matter of equivalence in the current study.
One task that was outside the present scope but warrants additional discussion is target discrimination.The effects of depth in discrimination procedures are important to consider in addition to the present detection and localization findings.The present paper highlighted a near advantage for spatial attention through target detection/localization tasks, but the evidence for the effect of depth for discrimination has been mixed.

Table 2
A list of sample literature examining the effects of depth within other closely related behavioural tasks using discrete targets.
Near advantage has also been found in discrimination tasks for RT (Ahsan et al., 2022;Blini et al., 2018;Dureux et al., 2021) and accuracy (Atchley et al., 1997;O'Donnell et al., 2022).However, spatial attention advantages have also been demonstrated for discrimination targets appearing at a far-depth plane in RT (Britt et al., 2024a;O'Connor et al., 2014) and accuracy (Li et al., 2011;Martin et al., 2021;O'Donnell et al., 2022; also see Chien and Watanabe, 2013;Zafarana et al., 2023Zafarana et al., , 2024)).It could be that, unlike the target detection/localization literature, visual processing requirements associated with target discrimination and object recognition/perception have shown a strong relationship with the ventral visual processing stream (e.g., Goodale and Milner, 2023;Lambert and Wootton, 2017;Rao et al., 2003)-appearing to contrast the exclusive visual processing requirements associated with the dorsal stream (but see Freud et al., 2016).Research in our lab is continuing to investigate target discrimination across depth and intends to quantify the near or far advantages of these tasks (with potential moderating factors that are specific to discrimination procedures).
Finally, empirical work has compared near-versus far-space processing has gained momentum in many attention and perceptual tasks.In addition to target detection and localization tasks, other tasks such as target discrimination, visual search, flanker tasks, temporal order judgements, and line bisection literature have all revealed isolated findings concerning differential effects of depth on attention and perception.Table 2 was designed to provide a list of sample literature examining the effects of depth within these specified behavioural tasks that were not included in our meta-analysis but still serve as foundational work when examining the effects of depth in 3-D space.

Conclusion
To conclude, the results of the present meta-analysis revealed a near advantage for spatial attention in target detection and localization tasks that varied across moderators.Spatial attention being prioritized within near space aligns with the ecological explanation of a survival necessity for increased cognitive resources to functionally protect the observer.Moving forward, research should aim to refine the cognitive processes underlying this apparent behavioural advantage for spatial attention within near space-likely beginning with further behavioural implications within the dorsal visual pathway.

Fig. 4 .
Fig. 4. all primary moderator variable analyses.Positive values represent a near advantage for spatial attention (i.e., faster RTs for near targets compared to far targets).

Fig. 5 .
Fig. 5. Forest plot of the nine datasets using motion during their target detection and localization.