Spatial abilities at different scales: Individual differences in aptitude-test performance and spatial-layout learning

Abstract

Most psychometric tests of spatial ability are paper-and-pencil tasks at the “figural” scale of space, in that they involve inspecting, imagining or mentally transforming small shapes or manipulable objects. Environmental spatial tasks, such as wayfinding or learning the layout of a building or city, are carried out in larger spaces that surround the body and involve integration of the sequence of views that change with one's movement in the environment. In a correlational study, 221 participants were tested on psychometric measures of spatial abilities, spatial updating, verbal abilities and working memory. They also learned the layout of large environments from direct experience walking through a real environment, and via two different media: a desktop virtual environment (VE) and a videotape of a walk through an environment. In an exploratory factor analysis, measures of environmental learning from direct experience defined a separate factor from measures of learning based on VE and video media. In structural-equation models, small-scale spatial abilities predicted performance on the environmental-learning tasks, but were more predictive of learning from media than from direct experience. The results indicate that spatial abilities at different scales of space are partially but not totally dissociated. They specify the degree of overlap between small-scale and large-scale spatial abilities, inform theories of sex differences in these abilities, and provide new insights about what these abilities have in common and how they differ.

Introduction

There has been a long tradition of research on the measurement and classification of individual differences in spatial abilities (e.g., Carroll, 1993, Eliot and Smith, 1983, Lohman, 1988, McGee, 1979). In this literature, measures of spatial abilities have included tasks such as mental rotation of shapes, solving mazes, imagining the folding and unfolding of sheets of paper, and finding hidden figures. These “small scale” psychometric tests are similar insofar as they are all paper-and-pencil tests, and almost all involve perceptually examining, imagining, or mentally transforming representations of small shapes or manipulable objects, such as blocks or sheets of paper.

In contrast to small-scale spatial abilities, there have been relatively few attempts to assess individual differences in larger-scale or “environmental” spatial abilities. Environmental spatial tasks include learning the layout of new environments, such as buildings or cities, navigation in known environments, and giving and interpreting verbal navigation directions (see reviews by Evans, 1980, Gärling and Golledge, 1987, Liben et al., 1981, Spencer et al., 1989). One goal of the research project reported below is to characterize the sources of individual differences in environmental spatial tasks, assessing whether performance of different environmental spatial tasks reflects a single underlying ability or a disparate set of abilities. The second goal is to examine the extent to which processing of spatial information at different scales of space reflects the same or different underlying abilities, as articulated by Cooper and Mumaw (1985): “To what extent [do] common processes and representations underlie skill in dealing with large-scale space and spatial ability as measured by standard aptitude tests?” (pp. 91–92).

Fig. 1 diagrams different possible models of the relationship between small-scale and large-scale spatial abilities. The Unitary model assumes that spatial abilities at the two scales of space are completely overlapping, as would be the case if they depended on exactly the same cognitive processes. The Total Dissociation model proposes that the two sets of abilities depend on completely distinct cognitive processes. The Partial Dissociation model proposes that the two sets of abilities rely on some common processes, but that ability at each scale of space depends on some unique processes not shared by abilities at the other scale of space. If this is the preferred model, we also need to specify the amount of common variance and the cognitive processes that are shared by the two sets of abilities. The Mediation model proposes that the two sets of abilities are completely dissociated, but that both are related to a third ability that mediates the relationship between large- and small-scale spatial ability.

Although previous studies of spatial abilities have not made strong claims that abilities at different scales of space are either completely overlapping (Unitary model) or completely dissociated (Total Dissociation model), most studies emphasize either the commonalities or the dissociations between these abilities. First, by using “pictorial” scale stimuli, such as images on paper or computer monitors, to study large-scale spatial tasks such as navigation, the majority of studies in the literature have implicitly adopted the Unitary model (Montello, 1993). The fact that almost all tests of spatial abilities involve paper-and-pencil depictions of small objects (Eliot & Smith, 1983) also reflects an implicit assumption that all spatial cognition can be studied with small-scale stimuli. Other research has addressed the unitary model more explicitly. For example, many theories of the evolution of sex differences propose that differences in small-scale spatial abilities (e.g., mental rotation) reflect different selection pressures for navigational (large-scale) abilities between males and females in evolutionary history (Gaulin, 1995, Kimura, 2000). One recent review (Jones, Braithwaite, & Healy, 2003) considers seven theories of the evolution of sex differences in spatial abilities, all of which are based on demands for the sexes to have different amounts of mobility in the environment (e.g., for foraging, warfare, or finding mates). These theories assume that large-scale and small-scale spatial skills reflect the same abilities, and are causally related.

In contrast, some other theorists have emphasized the dissociations between the mental systems for processing spatial information at different scales of space. In comparing environmental space with the object space of traditional space perception, Ittelson (1973) was the first to point out that in contrast with objects, environments are larger than the individual, allow viewing from multiple vantage points, and require locomotion and information integration over time for their apprehension. Several other researchers in cognition and perception (e.g., Acredolo, 1981, Cutting and Vishton, 1995, Freundschuh and Egenhofer, 1997, Gärling and Golledge, 1987, Kuipers, 1982, Mandler, 1983, Montello, 1993, Montello and Golledge, 1999, Tversky et al., 1999, Zacks et al., 2000) and some evolutionary psychologists (e.g. Silverman & Eals, 1992) have echoed this distinction in theorizing that spatial entities at different scales involve distinct cognitive structures and processes. These proposals are inconsistent with the Unitary model in Fig. 1, and consistent with the dissociation models, but do not specify whether the dissociation is partial or complete.

Of particular importance to our work are the distinctions between figural, vista, and environmental space (Montello, 1993, Montello and Golledge, 1999). Figural space is small in scale relative to the body and external to the individual, and can be apprehended from a single viewpoint. It includes both flat pictorial space and volumetric object space (e.g., small, manipulable objects) associated with psychometric tests of spatial ability. Vista space is projectively as large or larger than the body, but can be visually apprehended from a single place without appreciable locomotion. It is the space of single rooms, town squares, small valleys and horizons. Environmental space is large in scale relative to the body and “contains” the individual. It includes the spaces of buildings, neighborhoods, and cities, and it typically requires locomotion for its apprehension (see Montello, 1993, for a discussion of other scales of space).

The neuroscience literature provides considerable evidence that processing spatial information at different scales of space involves different brain structures and mechanisms. Whereas small-scale spatial tasks such as mental rotation are associated primarily with activation of the parietal lobes (see Kosslyn & Thompson, 2003, for a recent review), learning and remembering the layout of large-scale spaces is associated with processing in the hippocampus and surrounding regions in the medial temporal lobes (e.g., Morris & Parslow, 2004). Patients with various forms of topographical disorientation, who are impaired in wayfinding and environmental layout learning following brain lesions, do not typically show impairments in small-scale spatial abilities, such as mental rotation (Aguirre & D'Esposito, 1999). Parietal patients, showing typical impairments in small-scale mental spatial manipulations, in turn, show intact spatial updating during locomotion in large-scale space (Philbeck, Behrmann, Black, & Ebert, 2000). In an extensive review of the neuroscience literature, Previc (1998) proposed four “behavioral realms” for spatial behaviors, based on the sensorimotor systems involved in various actions, such as reaching and locomotion. These realms are quite tightly correlated with spatial scale. They are: peripersonal (near-body space), focal extrapersonal (the space of visual search and object recognition), action extrapersonal (orienting in topographically defined space), and ambient extrapersonal (orienting in earth-fixed space). Previc provides evidence that that the four realms are largely associated with distinct cortical networks: dorsolateral for peripersonal, ventrolateral for focal extrapersonal, ventromedial for action extrapersonal, and dorsomedial for ambient extrapersonal.

Studies of individual differences can provide important insights into the relation between large-scale and small-scale spatial cognition. If large- and small-scale spatial tasks depend on common basic capacities and processes, and there are individual differences in these capacities and processes, then measures of spatial cognition at different scales of space should be highly correlated.² In fact, previous individual differences studies have generally reported only weak, if any, relations between large-scale and small-scale spatial abilities. In a review of 12 studies of the relation between large- and small-scale spatial abilities (Allen et al., 1996, Bryant, 1982, Goldin and Thorndyke, 1982, Juan-Espinosa et al., 2000, Kirasic, 2000, Lorenz, 1988, Meld, 1985, Pearson and Ialongo, 1986, Rovine and Weisman, 1989, Sholl, 1988, Waller, 2000, Walsh et al., 1981), Hegarty and Waller (2005) reported that there were only two studies in which the median correlation exceeded .3. The majority of these correlations were not statistically significant.

Previous factor-analytic studies also support the separability of small-scale and large-scale spatial abilities. Lorenz, 1988, Lorenz and Neisser, 1986 administered several different measures of environmental spatial ability, including giving directions, recalling a route, and pointing to local and distant locations. In exploratory factor analyses, environmental abilities loaded on different factors than pencil-and-paper measures of spatial abilities. A similar dissociation between large- and small-scale spatial abilities was found by Pearson and Ialongo (1986). In the most extensive individual differences study to date, Allen et al. (1996) measured performance on six psychometric spatial tests (Surface Development, Cube Comparison, Hidden Figures, Gestalt Completion, Map Memory and Map Planning) and seven measures of learning of an environment from a walk through a small city. Five of the measures of learning (scene recognition, sequencing scenes along the route, placement of landmarks on a map of the route, intra-route distance estimates and route reversal) defined a large-scale learning factor that Allen et al. called “topological knowledge.” This factor was unrelated to the spatial ability factor defined by the psychometric tests. However another factor called “spatial sequential memory,” defined by measures of maze learning and maze reversal measured in a laboratory task, was related to both the psychometric test factor and the topological knowledge factor and therefore mediated the relationship between these two factors. Other measures of large-scale learning, ability to make straight-line distance estimates and direction estimates, did not load on the topological knowledge factor or on any other factor. There was again no direct relationship between these measures and small-scale spatial ability. However, in one experiment Allen et al. found that a measure of perspective taking mediated the relationship between paper-and-pencil measures of spatial ability and the measure of direction estimation (ability to point to non-visible locations in a learned environment).

In summary, previous individual differences research is inconsistent with the Unitary model in Fig. 1. Some individual differences studies are consistent with the Total Dissociation model while others are consistent with the Partial Dissociation model but suggest that the amount of overlapping variance between small- and large-scale spatial tasks is relatively small. Finally, the results of Allen et al. (1996) are consistent with the Mediation model depicted in Fig. 1, suggesting that other abilities mediate the relationship between small-scale spatial abilities and different aspects of large-scale learning.

An analysis of the perceptual and cognitive processes involved in large-scale spatial tasks can help us identify the possible sources of variance in those tasks, and inform questions of the amount and nature of the dissociation between large- and small-scale spatial abilities. The information processing approach has been used productively in the analysis of small-scale spatial abilities, to reveal that these abilities rely on differences in speed of encoding and transforming spatial information, spatial working memory capacity, and strategies (Hegarty and Waller, 2005, Just and Carpenter, 1985, Lohman, 1988, Pellegrino and Kail, 1982, Shah and Miyake, 1996). Although large-scale spatial cognition involves many complex activities, including learning the layout of new spaces, using existing knowledge of an environment to plan routes and navigate, and communicating about space, it is necessary to restrict our focus in an information-processing analysis. Consistent with previous individual-difference studies (Allen et al., 1996, Bryant, 1982, Goldin and Thorndyke, 1982, Kirasic, 2000, Pearson and Ialongo, 1986), we focus on ability to learn the layout of novel environments.

In studies of large scale learning, participants are given a controlled amount of exposure to a large scale environment, in our study by being led on a route through the environment, being shown a video, or interacting with a desktop virtual environment. They are tested on various “outcome measures” of learning, such as route retracing, pointing to non-visible locations, straight-line distance estimates and map sketching. Fig. 2 shows a preliminary analysis of the representations and processes involved in learning the layout of an environment and performing outcome measures of that learning. First, the layout of the environment must be encoded from the various sensory inputs available. This leads to an internal representation of the environment, which might be a route representation (i.e., a representation of the sequence of landmarks encountered and movements made in locomoting through the environment) or a survey representation (i.e., a two-dimensional representation of the configuration of the environment). The internal representation cannot be measured directly, but must be inferred from performance on the outcome measures. Furthermore the performance of a particular outcome measure (e.g., direction estimation) may involve transformations or inferences from a person's initial representation of the environment that occur either when the person is exposed to the learning medium, or when his or her knowledge of the environment is being tested (Montello, Waller, Hegarty, & Richardson, 2004). A task analysis of large-scale spatial learning must therefore specify the cognitive processes involved both in constructing an internal representation of an environment from exposure to that environment and in transforming that internal representation to perform different outcome measures (see Fig. 2).

Based on the task analysis, we hypothesize three main sources of individual differences in large-scale spatial cognition: ability to encode spatial information from sensory experience, ability to maintain a high-quality internal representation of that information in memory, and ability to perform spatial transformations in order to make inferences from this spatial information. First, learning the spatial layout of an environment depends on the ability to encode spatial information from the various sources of sensory information provided in the learning experience. Vision is probably the main sensory modality in humans for sensing the spatial layout of an environment. However, when one directly navigates through a real environment, one also senses one's own movement through non-visual senses. Movement is sensed by the vestibular system, which provides information about linear and angular accelerations, kinesthesis, which senses movement of the limbs, and efference copy, which is based on signals from the central nervous system to the muscles. Recent research has indicated that these senses contribute to spatial updating and learning of spatial layout independently of vision (Chance et al., 1998, Klatzky et al., 1998, Waller et al., 2004; but see Waller, Loomis, & Steck, 2003), so that it is likely that individual differences in learning from these senses also make an independent contribution to variability in learning of spatial layout. (Spatial layout can also be sensed to some extent by audition, but this will not be addressed in the current study).

Second, various aspects of memory affect the quality of the internal representations or cognitive maps constructed from a given amount of exposure to an environment. Working memory is a key factor underlying environmental learning, because environmental spaces cannot be apprehended in a single view (e.g., Ittelson, 1973, Montello, 1993), and these spaces must therefore be learned by maintaining information over time. The format of the memory constructed from exposure to an environment is another possible source of variance. Information might be retained in verbal working memory, as a sequence of route directions (Allen et al., 1996) or in spatial working memory, as a configuration. Sex differences in large-scale spatial cognition are often characterized as differences in the format of the internal representation, with females depending more on route representations and males depending more on configural representations of space (Lawton, 1994, Lawton et al., 1996, Montello et al., 1999). Finally metric precision is an important way in which spatial memories differ, and can have a large influence on the types of inferences made from memories of environmental spaces.

Third, ability to infer new information from spatial memories is clearly essential to many large-scale cognitive tasks. When one learns spatial layout from direct experience and visual media, one encounters the environment from a sequence of viewpoints as one moves through the environment. A memory of this sequence of viewpoints and movements alone is not sufficient for performing such tasks as estimation of straight-line distances and directions between landmarks that were encountered at different stages of the route. Therefore performance on these tasks requires an inference from the information that was directly perceived.

Encoding, memory, and inference processes are highly related. For example, the ability to infer the straight-line direction or distance between landmarks from a route representation depends on the ability to encode metric information about distances and turns between landmarks and to maintain this information in working memory. Integration over a section of route, in turn, changes the nature of the spatial memory from a sequential route representation to a configural representation. Finally, integration over sections of a route is less effortful when one physically moves in the environment (e.g., Klatzky et al., 1998, Loomis et al., 1999, Waller et al., 2002) so that encoding of self-motion on the basis of body-based senses may facilitate the process of inferring spatial configuration from the sequence of viewpoints encountered as one moves through an environment.

In the present study, we measured people's ability to learn the layout of three different novel environments, one from direct experience walking in the environment, another from watching a video of a route through the environment, and a third by interacting with a desktop virtual environment. After each of these learning experiences, we administered three different measures of learning: estimation of the direction between landmarks in the environment, estimation of the straight-line distance to non-visible landmarks, and map sketching. Compared to other common measures of spatial learning (e.g., scene recognition, landmark sequencing, and route retracing), these tasks cannot be performed on the basis of a route representation alone, and depend more on metric and configural knowledge of the environment (see Kitchen, 1996, Montello et al., 2004, Newcombe, 1985).

We can isolate the different sources of variance identified in our information-processing analysis by comparing patterns of performance across different learning experiences and different outcome measures. Learning spatial layout from direct experience, video, and desktop virtual environments (VEs) is similar in the sense that in all cases learning depends on perception of a sequence of viewpoints as one moves through the environment, memorization of that sequence of viewpoints in memory and updating of one's position in the environment on the basis of visual cues including self-to-object relations and optical flow. In all of these learning experiences, the ability to compute straight-line distances and directions relies on the ability to make inferences from the memorized sequence of viewpoints encountered on the route.

The different learning experiences also vary in important ways. First, although all include visual information, the field of view is greater in real-world navigation than in a video or virtual environment, so that optic flow information is less available in the visual media. Second, when learning (and being tested) in a real environment, as opposed to a video or desktop virtual environment, updating of body position in the environment can be sensed by vestibular input and kinesthesis in addition to vision. Finally, motion is more active and self-directed in a real environment (and to some extent in a desktop virtual environment), whereas in watching a video, movement through the environment is more passive. When motion is self-directed, efference copy of the muscle commands is an additional source of information for updating of self-motion. Because of these differences, learning from different media makes greater or lesser demands on different basic cognitive processes, such as ability to update one's position in an environment from visual information alone. On the basis of our task analysis, therefore, we hypothesize that ability to learn from real world experience and from visual media will be partially dissociated.

Because a correlational study of this type requires that all participants perform all of the same measures, it was necessary to have participants learn three different environments from the three different learning experiences. These environments necessarily had different layouts. To minimize confounds due to the shape of the environments we used three environments that were made up of mostly straight segments and right-angled turns. The environments learned from direct experience and from a video were of similar complexity. The virtual environment was simpler, because previous research (Richardson, Montello, & Hegarty, 1999) and our pilot studies indicated that learning a complex layout from a desktop virtual environment was more difficult and might lead to a floor effect unless the environment was simplified.

Performing different outcome measures based on an internal representation of an environment can also be a source of variance in measures of large-scale learning (see Fig. 2). Whereas all outcome measures depend on how well the learner has memorized the sequence of views that he or she encountered while learning the environment (so that more accuracy and metric precision in that memory will lead to better performance), outcome measures depend on different “readout processes” and can require individuals to make inferences that were not made at the time of learning (Kitchen, 1996, Montello et al., 2004, Newcombe, 1985). For example, ability to point accurately to non-visible locations or to estimate straight-line distances to these locations requires ability to infer a spatial configuration from information that is encoded as one traverses an environment, as well as metric knowledge of the segments and turns along the route. Map sketching may be less dependent on the internal integration of information in memory, in that a relatively accurate map can be sketched from an internal route representation, as long as metric distances and directions are represented, such that the integration over segments and turns of the route occurs when the route representation is externalized in the sketching process. Furthermore, it is possible to draw an accurate map from a verbally encoded memory, as long as suitably precise metric information is included in the verbal memory. Finally ability to represent one's current heading in a memorized environment during testing is necessary for pointing performance, but not for straight-line distance estimates or map drawing.

Because of the cognitive demands made by different outcome measures (Kitchen, 1996, Montello et al., 2004, Newcombe, 1985), we also consider the possibility that different outcome measures may not reflect the same spatial abilities. If all three measures reflect the same underlying ability to acquire or infer configurations from spatial information learned over time, they should load on the same factor. In contrast, if estimating distances, estimating directions, and map drawing rely on abilities associated with different task demands, they should define different factors.

The information processing analysis in Fig. 2 suggests that large-scale spatial learning depends on processes of encoding spatial information from visual input, and maintaining and mentally transforming spatial representations, which may be shared with small-scale spatial tasks (e.g., mental rotation and finding hidden figures), while also depending on processes such as spatial updating that are not shared with small scale tasks. To test this information-processing model, we examined the relation between large-scale learning, and measures of basic capacities and processes in visual–spatial information processing, including psychometric tests of spatial ability and a test of spatial working memory. Although they come from different research traditions, tests of spatial ability are highly related to tests of visual–spatial working memory (Miyake et al., 2001, Shah and Miyake, 1996). In the current study, we are concerned with the variance that they share, i.e., ability to encode, maintain and mentally transform visual spatial information. In addition to psychometric tests and working memory, we include measures of perspective taking ability to examine Allen's alternative model in which perspective taking mediates the relation between small-scale spatial ability and ability to acquire configural knowledge of an environment.

We also included a measure of self-reported sense-of-direction. Such measures are highly correlated with tasks that involve reorienting oneself in the environment (Bryant, 1982, Bryant, 1991, Kozlowski and Bryant, 1977, Sholl, 1988) and measures of spatial knowledge acquired from direct experience (Hegarty et al., 2002, Lorenz and Neisser, 1986, Montello and Pick, 1993). In contrast, they are typically less highly correlated with measures of spatial knowledge acquired from maps and very weakly correlated with paper-and-pencil measures of spatial ability (Bryant, 1982, Hegarty et al., 2002, Sholl, 1988). Self-report sense-of-direction is therefore largely independent of small-scale spatial ability and appears to reflect an ability to update one's location in space as a result of self-motion.

Finally, we measured verbal ability, verbal working memory and abstract reasoning ability. To the extent that environmental information is coded verbally rather than spatially in memory, verbal abilities, including verbal working memory (cf. Shah and Miyake, 1996) should be predictive of performance of large-scale spatial tasks. To the extent that environmental learning reflects general intelligence rather than spatial ability, abstract reasoning ability should be predictive of large-scale spatial tasks.

We hypothesize that all forms of spatial layout learning will be predicted by visual encoding abilities, spatial working memory, and spatial transformation abilities as measured by small-scale spatial ability measures. We expect that learning from direct experience will also be predicted by measures of spatial updating and encoding based on body-based senses. Because spatial updating involves relatively little effort when body-based senses and optic flow information are available (Klatzky et al., 1998, Loomis et al., 1999, Waller et al., 2002) learning from direct experience should be less dependent on individual differences in visual encoding, working memory and effortful spatial transformation abilities than learning from visual media. Therefore we predict that small-scale spatial abilities will be less predictive of performance when learning is based on direct experience. Finally, although the focus of this study is on individual differences in spatial cognition, we also examine sex differences specifically, to inform theoretical debate on their nature and causes.