Socio-Spatial Experience in Space Syntax Research: A PRISMA-Compliant Review

: Characterising and predicting socio-spatial experience has long been a key research question in space syntax research. Due to the lack of synthesised knowledge about it, this review conducts the ﬁrst systematic scoping review of space syntax research on the relationships between spatial properties and experiential values. Adopting the “Preferred Reporting Items for Systematic reviews and Meta-Analyses” (PRISMA) framework, this review of space syntax research identiﬁes 38 studies that examine socio-spatial experiences in architectural, medical, and urban spaces. The data arising from this systematic review are used to identify trends in this sub-ﬁeld of research, including the growth of socio-spatial methods and applications in urban analytics since 2016 and key methodological approaches, characteristics, and factors in space syntax research about socio-spatial experience. The research identiﬁed using the systematic framework employs a mixture of descriptive, correlation, and regression methods to examine the dynamic effects of spatial conﬁgurations on human experiences. Arising from the results of the review, the article further identiﬁes a collective, predictive model consisting of ﬁve syntactic predictors and three categories of experiential values. This article, ﬁnally, examines research gaps and limitations in the body of knowledge and suggests future research directions.


Introduction
Space syntax theorises the relationship between spatial patterns and human behaviours.For example, the space syntax method axial line analysis (ALA) identifies and examines a network of "longest and fewest" lines of sight [1] and access in a plan of the built environment, which also represent human behavioural characteristics such as movement paths and navigational choices.A second space syntax technique, convex space analysis (CSA), develops and then uses a map that describes visibly defined or enclosed spaces and their adjacent accesses in terms of social interaction and inhabitation.Such a map can be used to measure the ways architectural and urban spaces both limit and enable human movement and perception [2].A third technique, isovist analysis (ISA), is based on isovist fields or viewsheds that geometrically represent the spatio-visual properties of locations in an environment.This method can be used for predicting movement as well as controlling observation [3].Additional space syntax techniques such as justified plan graph (JPG) analysis, visibility graph analysis (VGA), and agent-based simulation (ABS) [3] have also been used to study the social, cognitive, and behavioural properties of spaces.From these studies, multiple space syntax measures and indexes have been proposed and applied for measuring and understanding socio-spatial properties.
Using these various syntactic measures, space syntax has been widely applied to explore the relations between spatial organisations and social effects in the built environment.Furthermore, multiple spatial properties are measured to identify spatial topologies and social relations, regardless of the forms that contain them [4].The complicated statistics of spatial configurations have enabled the characterisation and prediction of the socio-spatial experience in the built environment [5].However, while hundreds of space syntax studies have been published in a range of fields over the last 40 years, systematic literature reviews of space syntax methods and results have only rarely been conducted.There are a few notable exceptions [6][7][8], including Nes's and Yamu's "Introduction to Space Syntax in Urban Studies" [9].Such past reviews typically employ descriptive or narrative approaches, and only rarely consider specific research applications, the validity of results, or methodological limitations.An exception is the work of Sharmin and Kamruzzaman [7], who conducted a meta-analysis of the relationship between syntactic measures and pedestrian movements, providing a statistical analysis of the findings and a synthesis of the evidence.Their approach, which both is focussed on a particular application and includes a critical review of the evidence, is one of the few examples of its type.It highlights the lack of other similar systematic reviews in this field and the need for research which critically syntheses complex space syntax methods, indices, and results about the relationships between syntactic properties and human experiences.As such, a major research question in this review is: "how has space syntax research characterised and predicted socio-spatial experience in the built environment?"To address this research problem, this article conducts the first systematic scoping review of space syntax research on socio-spatial relations-which is a sub-set of the much larger body of space syntax applications.This article aims to develop a collective understanding of socio-spatial approaches and factors in space syntax research by identifying and mapping the available evidence in the literature and discussing the relationships between syntactic values and human experiences.
The primary approach used in this article, a systematic literature review, is a rigorous, reproducible methodology that identifies, selects, appraises, and synthesises the relevant studies and evidence [10].In accordance with the best practice in the field, this research adopts both the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) standard [10] and the PECO framework (population or problem, exposure or environment, comparator, and outcome) [11] to frame the research questions and goals of the review.Although these systematic review frameworks were largely developed and applied in medical research, a PRISMA-compliant review has been widely used in a variety of domains as it is an accepted approach for synthesising knowledge about large research fields and specific applications.It has been used, for example, for conceptualising and categorising smart-city projects [12], identifying trends and perspectives in smartcities research [13], and introducing the concepts and techniques of machine learning in urban studies [14].Most of these reviews are, however, limited to identifying the types of available evidence and key characteristics or factors.Thus, they could be considered 'systematic scoping reviews'.In contrast, a full systematic review should identify and synthesis the evidence [15], but due to the complexity in the research scales and measures, only a few studies in architectural and urban sciences have fulfilled the required conditions of a systematic review.For this reason, the present article conducts a systematic scoping review, a precursor to a full systematic review, using the PRISMA framework to provide a holistic overview of experiential variables in space syntax research.In accordance with the principles of meta-analysis, the experiential data, comparative values, and syntactic parameters considered in this article are synthesised and explored to capture the key characteristics and socio-spatial factors in space syntax research.In addition, as the problem addressed in this article is socio-spatial experience in space syntax research, the relevant outcomes identified through this scoping review are focused on the relationship between syntactic and experiential data.
The following section highlights three methodological stages in a PRISMA-compliant review: (1) finding relevant data, (2) data selection, and (3) synthesis.This article then reports the synthesised characteristics and outcomes of the identified space syntax studies and considers their implications.This article concludes with a discussion about the research limitations and contributions.

Finding Relevant Data
In accordance with the PECO framework for systematic reviews, the scope of this review has four elements.First, it is concerned with socio-spatial experience in space syntax research (P: problem).Second, it is focused on syntactic properties of the built environment that were measured using space syntax techniques (E: Environment).Experiential or experimental properties are identified as comparative values to the syntactic parameters (C: Comparison).The last element highlights the relationship between syntactic and experiential data (O: Outcome).From these criteria, search terms are developed for finding relevant articles.These terms are (i) space syntax, (ii) experience, and (iii) experiment or test.This review used a specific search string: "Space syntax" AND experience AND (experiment OR test) This review developed the above search string to produce a set of articles with an appropriate degree of relevance.Nine academic databases-Science Direct, Wiley Online Library, Taylor & Francis Online, SAGE Journals, JSTOR, Web of Science, UCL Discovery, Emerald Insight, and MDPI-were used to identify relevant publications (refereed journal articles only) in late 2021 (November 2021 last search).In addition to the online database search, both forward and backward citation searches were run using Google Scholar.

Data Selection
This review used two rounds of data-selection workflows, online database search, and citation search in Figure 1, extended from the conventional PRISMA templates [10,16].The first, online database search, starts by identifying articles from the nine databases with the chosen search string (n = 1317).After excluding duplicate records and inappropriate formats of publications (conference papers, reports, and chapters), 969 articles were screened by title and abstract using the PECO selection criteria framework.Two authors assessed the quality of the resulting 42 articles for eligibility using an independent full-text review structured around the four PECO questions.A simple rating system (Yes = 1; No = 0; partially = 0.5) was used for the full-text screening and 26 articles from the database search that scored greater than or equal to 3.5 points by each assessor were included in the online database search.The second round of data selection, citation search, involved forward and backward citation searches.The former identified research articles that cited the 26 articles selected in the online database search (n = 422), while the latter selected all the references in the 26 articles (n = 1379).In the screening phase, 1302 articles were screened by title and abstract and 33 articles were then assessed for eligibility by the two authors.Finally, 12 articles were included in the citation search, leading to a total of 38 articles being selected for this scoping review (see also Table 1).

Synthesis
Space syntax research on socio-spatial experience typically consists of three research components, (i) syntactic (or topological map) analysis and (ii) experiential (or experimental) data analysis, closely relating to computational and social research, respectively.The two analytic results are then mapped and compared in (iii) relational analysis such as correlation, where conventional statistical analysis is often applied.Collectively, this review takes a step-by-step synthesis to the three components.Furthermore, based on the PRISMA checklist, the synthesis of the data highlights three aspects of the findings of the 38 articles, (i) syntactic properties of the built environment, (ii) experiential values, and (iii) the relationships between spatial properties and experiential values.The first synthesis addresses specific space syntax techniques (ALA, CSA, ISA, JPG, VGA, ABS, SA) and syntactic parameters or measures, e.g., integration, connectivity, intelligibility, control value, mean depth, choice, isovist properties, visibility, and synergy.Second, experiential values are synthesised with a focus on experimental methods and their results recorded for factors such as behavioural pattern, movement, spatial choice, and human perception.Lastly, this review categorises and discusses the findings of the articles in terms of the relationships between syntactic and experiential data.In addition, this review summarises the spatial properties impacting on socio-spatial experience in the research.

General Characteristics
As shown in Figure 2 (see also Table 1), 20 of the collected 38 articles have a focus on 'urban space' (52.63%), ten address 'medical space' (26.32%), and seven 'architecture' (18.42%).One article investigating mountainous sites was not included in these three categories.The two dominant research areas were neighbourhoods [20,24,28,39,52] and parks [21,43,51,53].The 38 research articles were published between 2007 and 2021.The number of articles records a spike in 2020, largely published about 'urban space'.The review suggests that there is a growth in the number of relevant publications, although this may also be a by-product of database indexing time.The increasing availability of geographical information systems (GIS), global positioning systems (GPS), and state-of-the-art tracking technologies may also be shaping the increase in publications in urban space since 2016 in Figure 2.Each of three journals, Cities, Environment and Planning B: Urban Analytics and City Science, and Sustainability, published three articles.These journals have also regularly published space syntax research on architectural and urban spaces.The other journals, e.g., Journal of Urbanism, Habitat International, Frontiers of Architectural Research, and Open House International, published one or two articles on this topic.Interestingly, space syntax research on "medical space" (e.g., hospital planning, intensive care units (ICU), etc.) has been published at a steady rate of one or two articles per year (Figure 2).As such, health-related journals, e.g., HERD: Health Environments Research & Design Journal, Behavioral Sciences, American Journal of Alzheimer's Disease & Other Dementias, and International Journal of Environmental Research and Public Health, have an interest in space syntax, with ten articles identified in this review.The relationship between syntactic measures of medical spaces and human experiences or responses has become a valuable research topic in recent years [3].

Methodological Approaches
Of the 38 articles, space syntax research about urban space generally used ALA.As such, the collected articles typically generated the syntactic properties of cities using axial maps, focusing on movement paths (typically roads).A wide range of sites was explored in this research, including city areas [23,46], public open spaces [19,30], city centres or central business districts [27,50], tourist destinations [44], and historical sites [33].Several different ranges of radius measures were used for the analysis of movement: an 800 m radius measure for a ten-minute walk [27,39], a 400 m for a five-minute walk [35,44], and a combined measure of 1000 m and 400 m for topological proximity [21].One lesson from these studies is that a buffer-zone radius measure should be carefully defined before the development of a topological map.
UCL DepthmapX (version 0.8.0) software was typically used for the ALA on urban spaces, often in parallel with ArcGIS [21,27,53,54].The QGIS space syntax toolkit was also developed for this purpose [44].ALA provided the dominant technique to measure global integration and local integration from traditional street segments, while it was applied to develop angular syntactic properties in urban networks [27,28,50,51].The 20 urban studies in the collected data tended to use such single syntactic (ALA or SA) methods focusing on natural movements.In contrast, three studies combined it with visuo-spatial analysis (i.e., VGA and ISA) to address social interaction [19], cognitive aspects of a city's imageability [27], and even perceived urban stress [30].Interestingly, instead of ALA, the configurational characteristics of park pathways were captured using a modified CSA focusing on the depth of a space [43].Density, an influential factor on human experience in urban space, was also measured to improve the predictive power of spatial analytic data [30,32,46].Beyond these syntactic approaches, geographic information was developed using Google Earth [23], OpenStreetMap [30,35], and GIS [44,53,54].
The remainder (18 articles) conducted syntactic analyses of plan layouts in architecture (medical and architectural spaces).ALA tended to be used to investigate indoor movement (eight articles), while other architectural studies developed visuo-spatial properties from visibility graph measures such as VGA (11 articles) and ISA (four articles).A few articles used VGA or ISA as a sole technique for spatial analysis to explore psychological seating choices in a three-dimensional (3D) virtual model [45], the visibility and accessibility of medical spaces [18,25,47], and macro-cognitive interactions [37].Interestingly, Alalouch and Aspinall [18] used a small grid size (200 mm × 200 mm) for VGA in a possible attempt to increase the statistical significance in their relational analysis.However, there is no consensus about the spacing across the set of works.For example, one study used 1 foot [36] and another used 1 m [55], while a "human-scale" grid was typically recommended [56].Furthermore, in terms of the combination of VGA and ABS, a Fibonacci retracement calculator was applied to calculate "ideal" integration values and predict human movements in a restaurant [17].
In the set of articles, the combination of ALA and VGA (or ISA) was often used to investigate wayfinding [29,31,41], visual accessibility [22,49], and spatial memory [48].Several architectural studies developed syntactic properties from convex maps (CSA) or plan graphs (JPG).Marquardt et al. [34] addressed the convexity of homes along with clinical variables, and Zeng et al. [42] measured "symmetry-asymmetry" using JPG.To examine both axial and node integrations in hospitals, JPG was associated with ALA [22,40].Furthermore, as a complementary approach to space syntax techniques, Omer and Goldblatt [38] presented "Q-analysis" using a connectivity graph where axial maps for each building floor were transformed into segment maps.Their method could be useful for investigating the multidimensional structure of movement patterns in architecture.
In summary, the collected data revealed that the conventional alpha analysis, along with geographic information, was widely used in urban studies.In contrast, architectural studies tended to adopt more diverse space syntax techniques, highlighting visuo-spatial properties in the built environment.UCL Depthmap was dominantly used in 27 articles (71.05%) in the data, and it is clear that space syntax techniques and approaches have continued to evolve across the body of research captured in this analysis.

Syntactic Properties
The syntactic properties of the built environment can be understood as pertaining to either "environment" or "exposure" (E) in the PECO framework.Figure 3 describes the percentage of articles using specific space syntax techniques and syntactic parameters and their relationships.ALA was a dominant technique, accounting for 23 out of 38 articles or 60.53%.This result aligns with the dominant research area, "urban space", in Figure 2, because it was usually explored with ALA.VGA was the second-most-frequently (36.84%) used to examine indoor spaces.ISA is the third (18.42%) in Figure 3a.Each of CSA, JPG, and SA accounted for 13.16%, while only three articles used ABS.It should be acknowledged, however, that many articles use more than one space syntax technique.VGA is, for example, combined with ALA [19, 27,29,31,48], ISA [19, 26,27], ABS [17,26], and JPG [22].There are also combinations of three techniques-ALA, ISA, and VGA [19]; ALA, JPG, and VGA [22]; ISA, VGA, and ABS [26].Lastly, Güngör and Harman Aslan [27] present a combined method of four techniques, ALA, ISA, VGA, and SA, as a holistic way to define urban design strategies from cognitive and syntactic properties.As shown in Figure 3b, each space syntax technique produced several syntactic parameters.Among them, integration, including global integration and local integration, was measured in 22 out of 38 articles or 57.89%.Since visual integration (39.47%) of VGA is a variation of the same parameter, the number of articles addressing the dominant syntactic parameter could be more than 22.However, this review distinguishes the two, as they use different maps, axial map and visibility map, respectively.The second-most-common parameter was connectivity (47.37%), the third visual integration, and the fourth intelligibility (31.58%).Integration, connectivity, and intelligibility should be fundamental values because research suggests that they are closely related to behaviour or movement patterns in the built environment.In addition to these dominant parameters, choice (23.68%) was frequently used in ALA or SA.Synergy, as a second-order measure like intelligibility, used in ALA, accounted for 5.26%.SA, as an alternative to ALA, also produced these parameters.
In contrast, CSA and JPG developed relatively more control value and mean depth measures but both techniques also weighted integration and connectivity.Again, VGA produced visual integration (39.47%) and visual connectivity (21.05%), while ISA developed several isovist properties (13.16%) such as isovist area and perimeter.Two articles used ABS along with VGA.Interestingly, visibility accounted for 10.53%, but the definitions of visibility were slightly different across the articles.For example, visibility could be regarded as either visual integration or visual access [31], the relative size of isovist areas [30], line of sight [37], and even visual connectivity [47].A few additional parameters-convexity [34], layout complexity [31], and openness [37]-were derived only once in the set of work and are not counted in Figure 3b.Acknowledging these specific parameters, the relationship between space syntax techniques and syntactic parameters in the figure and the synthesised, holistic view should be very useful for space syntax research with a focus on socio-spatial experience.

Discussion of Syntactic Analysis
Although this review has focused on the synthesis of syntactic properties, there is a level of complexity in these which is important to note.All space syntax variables have their own definitions and purposes in terms of socio-spatial epistemology [5].For example, past research has clearly distinguished local integration (R3) from global integration (Rn) [21,23,32,36,57,58].The former is used as the best predictor of "smaller-scale movement" or "pedestrian movement", while the latter is better for "larger-scale movement" or "vehicular movement" [57].Furthermore, normalized angular integration [51,59,60] developed by SA, could be more useful for an accurate prediction of movement than topological integration [61][62][63].Visual integration (local and global) of VGA was also different from integration measured with ALA.That is, different space syntax maps or graphs could produce measures with the same name but meaning different things in different contexts.In addition, integration generally indicates the accessibility of the system, but choice is also sometimes used for this purpose because it can be regarded as an improved measure for human movement prediction [64].Thus, like visibility in this review, each syntactic variable should be clearly defined, even of it is widely used in this domain.
Due to the heterogeneity and complexity of socio-spatial properties, it is reasonable that space syntax research uses more than one syntactic technique and variable.Specifically, many articles combined VGA with another syntactic technique.This is possible because, regardless of whether the topological graph (of urban space or of a large building) is partly or fully manual, the UCL Depthmap software automatically extracts both ALA and VGA from it.Another reason why VGA was preferred by space syntax researchers is the relationship between "visibility and permeability" [65,66].That is, ALA and CSA describe the permeability of the system, while VGA identifies the visibility properties.In this way, their combination could capture the hybrid impact of spatial perceptions and structures on movement and wayfinding in architectural and urban design.
Secondly, ten articles (26.32%) produced behavioural data using questionnaires, interviews, and observations.For example, Sheng et al. [51] adopted field observations to count personal and social activities in urban parks.Askarizad and Safari [19] used gate counting and snapshot observation to identify both movement and behavioural data (e.g., walking, talking, and standing).Snapshot observation was also used in two more articles [20,52].Marquardt et al. [34], furthermore, used an in-home assessment to capture activities of daily living (ADL).In contrast, seven articles (18.42%) dealt with participants' perceptions of built environments.These types of experiential values were also mostly captured with questionnaires and interviews.Interestingly, O'Hara et al. [37] addressed situated macrocognitive activities in a paediatric ICU and El-Hadedy and El-Husseiny [22] observed workplace violence (WPV) locations to apply Crime Prevention Through Environmental Design (CPTED).Lastly, three articles addressed spatial preferences or choices to examine social activity [17], privacy [18], and prospect-refuge theory [45].Two articles that were not categorised in this review produced specific data about spatial memory [48] and cultural identity [42].

Sampling
The development of human experiential data-of physical sites, virtual, or laboratory settings-typically adopts social science techniques.Whilst human data could be automatically collected using tracking technologies such as GPS, most research in this review employed convenient or purposive sampling because of the difficulty of developing population-level data in space syntax research.For example, human data were collected by tracking park visitors [53], counting daily visitors entering a garden [54], interviewing people on a street [28], and conducting visitor surveys [21,32].Students were also conveniently recruited from universities [24,30,48].In addition, questionnaire and observation tended to use self-selection or volunteer sampling [20,23,46,49] and systematic sampling [50].In contrast, only two studies used probability sampling, for example, simple random sampling [44] and stratified random sampling [52], because they knew the population of their research, cruise passengers and housing neighbourhoods, respectively.
To effectively develop human data, the sample size is a key element of the research design.However, the number of participants in the selected articles varied widely from eight to 7319, while the median was 105.These numbers might depend not only on sampling targets, but also on research methods.For example, research using interviews ranged from eight for staff interviews [17] to 75 for street interviews [28].In contrast, field observations recorded the higher numbers of participants, such as 7319 [43] and 4983 [51].GPS tracking also tended to recruit more than 300 participants [53,54].A faceto-face survey or questionnaire that produced movement or behavioural patterns also collected hundreds of samples.For example, Koohsari et al. [46] analysed data from 2591 participants for the Physical Activity in Localities and Community Environments (PLACE) study.Chiang and Li [21] used 964 residents' responses and Ozbil et al. [39] 843 parental surveys.Liu et al. [32] also collected 510 route samples through questionnaires.However, the studies of wayfinding tended to use smaller samples because they recorded routes as well as navigation behaviours.For example, Hölscher et al. [29] conduced wayfinding experiments with 12 participants and Li and Klippel [31] assessed eight participants' wayfinding behaviours in terms of time, distance, and pointing errors.In addition, an in-depth survey used medium-sized samples.For example, Marquardt et al. [34] collected 82 in-home assessments and Rashid et al. [49] 81 staff's perceptions of interaction and communication.In summary, research on movement patterns tended to recruit hundreds of participants, but experimental data such as perceptions and spatial preferences were often developed by small cases.In general, few articles described a clear rationale for determining an effective sample size.

Discussion of Human Data Analysis
A detailed hierarchy of socio-spatial experience, based on the four categories in this review, can be used to support a systematic identification of the appropriate types of syntactic techniques for future research.One of the first considerations when doing this relates to 2D and 3D space.For example, Lu et al.'s work [67] presents 3D visibility graphs associated with social network theory.Despite this example, most space syntax methods are applied to 2D human patterns, although some research captures complex activities [19,34].While it could be argued that 2D "movement" (that is, horizontal rather than vertical) was the dominant category of experiential data, a range of approaches and digital technologies were also used, led by observation studies and questionnaires.In other words, human experience was developed from multi-dimensional factors or "complex dimensions and elements" [68].This is also why multiple social and environmental variables have been considered as predictors in the present article [30,34,39,43].Regardless of the research area, human experience is collectively shaped by many social, economic, cultural, and environmental factors.
Space syntax studies have traditionally addressed the innate or natural socio-spatial properties of architectural and urban space.Thus, a large sample size is essential for this purpose [21,43,46,51,53,54].In contrast, relatively small sample sizes were employed for the in-depth analysis of behavioural or perceptual patterns [29,31,34].That is, the sample size for this comparator data was dependent on the applied syntactic technique(s) as well as the varied quality of data.Furthermore, this review reveals that space syntax research on human preferences, which might need an intensive body of data to study, has been relatively poorly addressed so far.The quality of spatial experiences, probably led by a higher level of cognitive operations [69], interactive aspects of space [68], and socio-economic forces [70], would be further investigated in space syntax communities.

The Relationships between Syntactic and Experiential Data
The findings of the 38 articles can be categorised into three analytic relationships, (i) descriptive, (ii) correlated, and (iii) predictive relationships (regression).The first addresses the descriptive statistics of syntactic and experiential data, separately, and discusses the connections between their results.The correlation and regression methods use statistical relationships, with the latter including the estimation of the relationship through a model in statistics.Generally, the correlations in these articles tend to include descriptive discussions whereas the regression applications are often presented with correlational analysis.

Descriptive Relationship
Articles dealing with descriptive relationships tended to have a small sample size [17,29] or frame an integrated discussion about several types of experiential data describing behavioural patterns [19, 20,52] and perceptions [22,23,27,37].For example, Aknar and Atun [17] argue that integration values are positively related to active tables in a restaurant and can be used for predicting social activity, based on eight manually compiled seating charts.Zerouati and Bellal [52] also find that syntactic values (visual integration, visual connectivity, and intelligibility), are important indicators of social activities, which are related to the frequencies of actions (sitting, standing, playing, buying/selling, and chatting) captured through snapshot observations.As such, connectivity is positively related to long-duration social activities in streets [20].There is also a positive relationship between spatial properties (integration and connectivity) and perceived WPV locations in a healthcare setting, because WPV is largely shaped by patients and their relatives [22].Openness, connectivity, and visibility also had an intricate link to macro-cognitive interactions in healthcare design [37].Syntactic properties are also related to perceptions of Lynch's elements, specifically, paths and nodes, shaping the mental image of the city [27].In contrast, the syntactic data did not always have a relationship with observational data in urban spaces, because pedestrian behaviours are affected by various social, commercial, and environmental factors [19].Human cognition is also too complex to be predicted by space syntax approaches [27].Other design strategies may also be superior to spatial configurations for specific purposes, such as CPTED [22].
Wayfinding research [29,31,41] often employed descriptive analytical methods for human movement and behavioural patterns.For example, novices' routes tended to show relatively higher values of integration, connectivity, and step depth, following a central point strategy, while experienced users employed a more effective plan to navigate environments [29].Integration, intelligibility, and choice were also used to predict movement patterns captured by spatial cognition experiments [24].Some of the findings suggest that low visibility could cause participants' wrong turns and hesitations while wayfinding, and layout complexity is related to wayfinding time and pointing errors [31].It was, however, also suggested that wayfinding behaviours seemed to be impacted more by signage and forms such as "open L-type" spatial plans [41].In summary, research adopting descriptive analysis relied on both syntactic properties and experimental data to examine spatial configurations.

Correlational Relationship
Correlation is the fundamental analytical approach to examining the relationship between variables.Eight articles used this approach, including the Pearson correlation coefficient [18,25,28,33,45], describing r values.For example, considering Appleton's prospectrefuge theory in architecture, Keszei et al. [45] discovered that there were positive correlations between seating choice and visual integration and intervisibility in simulated heritage buildings, e.g., r = 0.351, p < 0.01, and r = 0.381, p < 0.01, respectively, where participants tried to be seen.In a study of movement patterns in shopping malls, Omer and Goldblatt [38] calculate connectivity, topological mean depth, topological choice, and intelligibility at the axial line level and connectivity, topological and metric mean depth, and topological and metric choice at the segment level.Using this approach, they identify multiple correlations between syntactic properties and behaviours.For example, the aggregate movement flow in a shopping mall is found to be significantly (p < 0.05) correlated with topological choice (r = 0.510), connectivity (r = 0.914), metric choice (r = 0.797), and mean depth (r = −0.881).
The architectural studies dealing with "medical" spaces identified particular relationships between syntactic and experiential data.For example, Alalouch and Aspinall [18] reveal a strong negative relationship between preferences and visual integration in hospital wards, r = −0.957,p < 0.05, and a strong positive relationship between ward preference and control value, r = 0.913, p < 0.05, in terms of participants' chosen locations for privacy.Ferdous and Moore [25] also observe that in care facilities, high-level interactions and proximity (or integration) and visibility (or isovist area) are negatively correlated, r = −0.565,p < 0.01 and r = −0.538,p < 0.01, respectively.They argue that the syntactic properties might be positively correlated to low levels of interaction only.The high-level social interactions were better predicted using privacy parameters.Interestingly, the findings of Rashid et al. [40] include significant correlations between four interaction-related behaviours (walking, walking and interacting, standing and interacting, and sitting and interacting), three groups of users (nurse, physician, and others) and three syntactic properties (axial integration, node integration, and control value) in ICUs.For example, axial integration shows a negative correlation with "walking and interacting", r(19) = −0.633,p = 0.004, and a positive correlation with "sitting and interacting", r(19) = 0.575, p = 0.016, among nurses in nurse stations.It also has a significant positive correlation with "sitting and interacting", r(148) = 0.387, p = 0.000, among physicians at the unit level.In contrast, all observed behaviours were found to be negatively correlated with node integration, r(53) = −0.335,p = 0.014, and control value, r(53) = −0.345,p = 0.011, among nurses in patient rooms.
Urban studies, compared to architectural studies, tended to find only limited correlations.In Jakarta, Hidayati et al. [28] revealed that potential vehicular accessibility (normalised angular choice) and three locations (preschool, primary, and tertiary schools) were found to be correlated, r = −0.109,p < 0.05, r = −0.143,p < 0.01, and r = 0.226, p < 0.01, respectively.However, in addition to such weak correlations, their cases did not show any relationships between the realised pedestrian accessibility and school locations.As a particular case using the Spearman correlation, Zhang et al. [54] found that there were multiple significant relationships between temporal-spatial behaviours (first visiting proportion, revisiting proportion, average time, and average speed) and four syntactic indicators (connectivity, mean depth, integration, and intelligibility) in a Chinese garden, ranging from ρ = −0.248,p < 0.05 (average speed and intelligibility in the visible layer) to ρ = 0.525, p < 0.01 (first visiting proportion and connectivity in the passable layer).While there was no common consensus about the range of correlation coefficients, such weak correlations were discussed in some cases.There were also some formulas used to collectively quantify human experience such as social and spatial interactions.Nonetheless, integration and connectivity are fundamental syntactic parameters for correlation.Choice and simple topological depth also show significant correlations with many experiential variables.In summary, the correlational analysis addressed statistically significant relationships between these conventional syntactic properties and various types of experiential data (e.g., movement, behaviour, perception, and spatial preference).
In contrast, a few articles described regression equations.For example, Liu et al. [32] reveal that syntactic values (integration and control value) affect tourist trail flows, y = −75.120+ 20.217S − 191.678L + 716.984G (R 2 = 0.727, F = 22.321, p = 0.001), and tourist visits, y = 1253.037+ 1287.307C− 1567.360L+ 388.846S (R 2 = 0.331, F = 5.779, p = 0.004), where S represents "online scenic attention kernel density value", L local integration, G global integration, and C control value.Zeng et al. [42] also found a linear equation, y = 0.6619x − 2.7874 (R 2 = 0.3124, p < 0.05), describing the influence of syntactic characters of the traditional houses on residents' acculturation results in Beijing.The syntactic characters are quantified from integration and relative ringiness of typical traditional spaces.In contrast, Rashid et al. [49] report only the overall regression (R 2 values with F and p values) to show the significant, positive impact of visual accessibilities (e.g., visual connectivity and visual integration) on interaction behaviours in ICUs.The visibility scores also predict the frequency of use of hand sanitising stations (HSS), adjusted R 2 = 0.479, F(5, 65) = 13.877,p < 0.001 [47].
Again, the dominant reporting format of regression was presenting predictive relationships with R 2 and β-values.The estimated coefficients of variables could be further explained with confidence intervals [46], odds ratios [50], and Wald x 2 tests [39], but were usually reported using t-statistics.Due to the complex nature of these values, however, this review highlights the significant links between predictor and response variables that could collectively generate social-spatial experience.Based on their relationships, this review identifies a collective model that consists of five syntactic predictors (integration, connectivity, control value, choice, and step depth) and three dependent categories (movement, behaviour, and perception).Whilst all predictors were presented in more than one article, integration and connectivity were dominantly examined in 13 and eight articles, respectively.The remainder were used only in two or three articles.Notably, integration was largely associated with both movement and behaviour, while connectivity tended to influence only movement.Choice and step depth were also regarded as a predictor of movement.

Discussion of the Relationships between Syntactic and Experiential Data
It was not unexpected that the collective, predictive model in Figure 4 identified three experiential values (movement, simple behaviour, and perception) because they are closely related to the three socio-spatial phenomena addressed in many space syntax studies [3,71].However, most of the 38 studies identified two or more syntactic predictors in their models, for example, connectivity, choice, integration, and step depth [44] and integration, control value, and mean depths [53].Thus, there was no single syntactic parameter predicting social-spatial experience.A few syntactic properties were simultaneously measured in a study, and they were jointly functioning as predictor variables in the regression model(s).Thus, the dynamic effects of syntactic variables on social experience should be further examined in a future study synthesising β-values.That is, future space syntax studies need to better understand the multi-dimensional socio-spatial relationships identified in this review.
Statistics derived from spatial geometries or forms, from segregation to integration, have enabled the investigation of both the social logic and the consequences of spaces [5].For example, in the research examined in this review, the integration or connectivity values of urban spaces had a strong correlation with human movements such as favourite routes and the frequencies of visits and use.However, the statistical analysis combining syntactic properties and experiential data often revealed the innate limitations of space syntax methods, due to reduced variables, possibly delivering unjustified conclusions about causal relationships.Of course, the original proposition was that syntactic accessibility could reflect the evolution of the system by way of a dynamic feedback cycle, and specifically, land uses and densities in an "organic" city [57,63,[72][73][74].However, the nature of sociospatial phenomena, e.g., society-space relations, social interfaces and encounter, and the temporality of human experience [70], might not be encapsulated in quantitative values only.Thus, an additional in-depth or qualitative survey could be revealing for this purpose.

A Systematic Review on Socio-Spatial Experiences in Space Syntax Research
Hillier and his co-authors argue that spatial configuration and morphology can supersede cognitive, economic, social, and cultural processes in this regard [57,63,74,75], although the reductionism of space syntax theory has been criticised [70,76,77].Syntactic structures (topological) cannot capture the semantic meanings (cognitive and informational) of spaces, events, and acts [70].Furthermore, without a deeper discussion about the complex, recursive phenomena occurring between human and environment, many space syntax researchers have relied on the hybrid theory (social and spatial) and its descriptive statistics, and have simply acknowledged syntactic, epistemological limitations.Likewise, the collected data in this review might be insufficient to demonstrate the quality or complexity of socio-spatial experiences in the built environment.
This review, furthermore, highlights the relationships between syntactic and experiential parameters, systematically examining socio-spatial factors in space syntax research.However, the formalised and codified evaluative framework proposed in this article may be limited to synthesising "nonlocal, or extrinsic properties of space" [72].It is not equally useful for examining the intrinsic, cognitive, and interpretative operations characterising spatial cognition and human experience in a comparative evaluation, because of the complexity of various multi-relational, multi-dimensional factors.That is why a similar meta-analysis [7] was focused on the relationship of just four syntactic measures (integration, choice, connectivity, and control) and pedestrian movement only.Whilst acknowledging this, the limited scope of meta-analysis itself needs additional evaluation components.For example, a heuristic approach such as an analytic hierarchy process (AHP) or an indepth analysis can be added to the existing framework.Particularly, the AHP can capture both the syntax and semantics of spatial experience comprising space, event, and activity, and thereby develop a hierarchical structure of related components, dimensions, and elements [68].Thus, it can be further elaborated from cognitive and social theories, specifically, "visuospatial cognition" or "spatial cognition' [24,27,69].

Limitations of the Data
The findings of the studies reported in this review have multiple limitations associated with sample designs and data collection, e.g., convenience sampling [21,45], limited sampling [18,30,43], limited time of data collection [33,47,52,54], or cross-sectional data [39,43,53].Some studies collected very small samples, e.g., eight participants [31] or 12 [29], although their data were in-depth and rich enough to address their research goals.The sampling of a few studies might also be too limited to generalise their findings [37,39,43,46,53].Lastly, some researchers reported unexpected biases such as participants' environmental changes [34] and GPS data errors [54].These limitations might be common challenges and unavoidable, but the sample design and data collection method should be carefully devised to conduct research particularly on socio-spatial experience.A potential approach to overcoming a limitation of the sample design and data collection would be using a combined and complementary method, where each part enhances the qualities of the other, e.g., observation and questionnaire or interview.An in-depth analysis on a rich data set can also be a promising strategy to examine complex socio-spatial behaviours and perceptions.

Future Research in Socio-Spatial Experience
While the ALA and VGA were dominant techniques, this review has highlighted the value of the combination of syntactic and experiential analyses.Importantly, VGA has been widely associated with the other syntactic techniques such as ALA and JPG, because perceptions are closely related to human movements and behaviours.That is, VGA should be a core method for investigating the socio-spatial experience of both architectural and urban spaces.Conversely, Hillier [72] promoted the combination of line and topological graphs, which were conventionally presented in ALA and CSA (or JPG), respectively.Despite this, only a few examples in urban research used such a hybrid method of axial and convex maps.This combination should be further explored in urban analytics, where convexity is rarely addressed.In this case, either Zhai's and Baran's research [43] or a stroke-based network [53] may provide an appropriate methodological precedent.
Interestingly, there are many syntactic studies using JPG or ALA only, where spatial analyses are largely designed to explore architectural theories or architects' intentions [78][79][80][81][82] and street structures or urban morphologies [32,38,83].However, these approaches may also need VGA to accommodate the types of experiential and experimental approaches suggested in this review.For example, professional interviews [17], snapshot observations [19, 20,52], and wayfinding experiments [29,31] would be very useful for this purpose, if samples are limited.Furthermore, this review has uncovered several wayfinding studies that brought a spatial-cognition-based approach to this domain [24,29,41].Examining individual cognitive processes and behaviours, spatial cognition research can consider landmark or temporal and emotional dimensions that have been largely ignored in conventional space syntax research [24].That is, this type of experimental research into wayfinding can provide important references for both architectural and urban design studies.
Many studies have already investigated the spatial configurations and morphologies of hundreds of cities and buildings.However, their multi-dimensional socio-spatial relationships should be further explored to examine cognitive and spatial experiences.Future research could also reveal the dynamic effects of syntactic properties on socio-spatial experience, examining the various β-values of the explanatory variables captured in this review, which would be able to overcome several fundamental limitations in the simple topological abstractions of space syntax research.In addition, the application of mathematics to space syntax approaches-for example, allometric equations [84], a Fibonacci sequence [17], and a modularity-based algorithm [85]-offer new research directions to test or improve their capabilities of predicting socio-spatial experiences.Likewise, additional ISA measures such as circularity [86] and distance-weighted isovist area [87] are also introduced in this field.In addition, a "weighted and directed JPG" has recently been developed for centrality measures [3,79,88].These new centrality measures can be useful for more precisely examining architectural configurations as well as urban networks.

Conclusions
This research has addressed an important question, how has space syntax research characterised and predicted socio-spatial experience in the built environment, providing the first systematic scoping review of space syntax studies on socio-spatial experience.Syntactic techniques have been widely used in research to explore movement, behaviour, perception, and preference patterns in architectural, medical, and urban spaces, which synthetically form socio-spatial experience.One of the reasons this significant body of empirical research has never been synthesised in a comprehensive manner before relates to the complexity of socio-spatial relationships and the combination of heterogeneous syntactic and experiential measures.
Combining the PRISMA and PECO frameworks, this review rigorously synthesises different types of socio-spatial properties and examines their relations in the built environment, producing new integrated findings.This extension of the traditional scoping review ensures that this article not only classifies and reports past research but synthesises the findings of data in a new way.The proposed collective, predictive model in Figure 4, identifying the three key socio-spatial phenomena (movement, behaviour, and perception), can be used by researchers undertaking systematic reviews.The correlational and regression results of space syntax studies reported in this article also offers a model that can be expanded in a full systematic review to compare the available evidence.
Despite the rigorous use of these frameworks, this approach may not capture all relevant features and insights in the body of research being examined.The data selection process developed from the PRISMA standard will not uncover every theoretical, social, and cognitive study ever completed using space syntax.This is why the present article emphasises critical synthesis, presenting three discussions of results, along with key literature that can augment these findings and provide insights for future space syntax research.In this way, this article contributes to identifying important research areas for future research as well as clarifying key concepts and characteristics in the literature.
Significantly, the synthesis of the findings of the 38 articles has identified descriptive and correlational relationships, as well as a collective model of the predictive relationships between syntactic and experiential variables.Such findings are not only useful for the space syntax community, but also a valuable reference for urban analytics researchers examining various aspects of planning and design.Considering the complex nature of socio-spatial phenomena, a relational mapping such as the proposed collective model, although it could also be a multi-dimensional matrix of spatial and social theories, should be inevitable for a following study to identify the appropriate syntactic and experimental parameters and investigating the various effects of a design.This article contributes to the development of future syntactic, empirical studies to develop research problems and methodologies that positively develop socio-spatial experience in the built environment.

Figure 2 .
Figure 2. The percentage of articles by research area and the number of articles by publication year.

Figure 3 .
Figure 3.The percentage of articles using specific space syntax techniques and syntactic parameters.(KEY-ALA: Axial Line Analysis, CSA: Convex Space Analysis, ISA: Isovist Analysis, JPG: Justified Plan Graph Analysis, VGA: Visibility Graph Analysis, ABS: Agent-based Simulation, and SA: Segment Analysis).

Table 1 .
Thirty-eight articles included in two rounds of data selection.