One Spatial Map or Many? Spatial Coding of Connected Environments

Xue HanNortheast Normal University

Suzanna BeckerMcMaster University

We investigated how humans encode large-scale spatial environments using a virtual taxi game. Wehypothesized that if 2 connected neighborhoods are explored jointly, people will form a single integratedspatial representation of the town. However, if the neighborhoods are first learned separately and laterobserved to be connected, people will form separate spatial representations; this should incur an accuracycost when inferring directions from one neighborhood to the other. Interestingly, our data instead suggestthat people have a very strong tendency to form local representations, regardless of whether theneighborhoods were learned together or separately. Only when all visible distinctions between neigh-borhoods were removed did people behave as if they formed one integrated spatial representation. Thesedata are broadly consistent with evidence from rodent hippocampal place cell recordings in connectedboxes, and with hierarchical models of spatial coding.

Keywords: spatial memory, cognitive map, multiscale hierarchical representations, multiple connectedenvironments, virtual reality

People solve spatial memory tasks using a variety of differentstrategies, likely using multiple different internal representations. Onesuch representation is the cognitive map (Tolman, 1948). Although itis still debated whether nonhuman animals form some sort of inte-grated, maplike representation of space, it is widely accepted thathumans have this ability. And yet, after many decades of research onthis topic, the precise nature of our internal spatial representationsremains a topic of considerable debate. One of many unresolvedissues is when we construct a single spatial representation versusmultiple local representations of large-scale spaces.

Some theories of spatial cognition have emphasized nonhierar-chical, local representations. For example, Worden (1992) pro-posed that mammals store memories of their geographical envi-ronment as a collection of independent fragments, each consistingof a set of landmarks, their geometric relationships, and theirnongeometric properties. However, other theories have empha-sized global representations. Within global theories, there aremany possibilities, such as flat versus hierarchical representations.Some researchers postulate that people’s spatial memories areorganized hierarchically based on global and local properties of anenvironment (e.g., Hirtle & Jonides, 1985; McNamara, 1986;Meilinger, 2008; Stevens & Coupe, 1978). Similarly, Meilinger(2008) proposed a network of reference frames theory, which

assumes that the space is encoded in multiple interconnectedreference frames through the navigator’s locomotion.

Local and global theories of spatial representation make differ-ent predictions about how people will combine information acrossspatial regions. If our representations are based on independentfragments, then integrating knowledge across fragments must oc-cur at retrieval time, incurring a large cost in both accuracy andresponse time. In contrast, with a single flat representation, whenjudging a spatial relation between two points, people should beequally fast and accurate within versus between regions, assumingequal distance judgments. Finally, when using hierarchical represen-tations, people should be able to use their coarse-scale knowledge toinfer directions from one region to another. In a hierarchical scheme,assuming that information is encoded more coarsely at larger spatialscales, when inferring directions from one region to another, theymust rely on coarser spatial information; thus, there should be anaccuracy cost. Whether there is also a speed cost depends on the typeof information encoded at each level. If objects are represented at afine spatial scale but not at a coarser scale, inferring spatial relationsbetween such objects should be slower when the objects are indifferent regions, requiring postretrieval processes. However, for ob-jects that are represented at all spatial scales, there should be no suchspeed cost, as participants can respond on the basis of a coarse scalespanning both regions. Factors such as the object size, type ofattention paid to objects, and their relevance to navigation mayinfluence whether they are encoded locally or globally. Previ-ous research has shown that when objects are relevant tonavigation, they are more likely to be treated as landmarks(Han, Byrne, Kahana, & Becker, 2012) and encoded in thedorsal visuospatial pathway (Janzen & van Turennout, 2004). Inthe studies reported here, we probe people’s memory for highlysalient landmarks that have been visited frequently as naviga-tional targets in a virtual reality (VR) task. We therefore hy-pothesize that these landmarks will be represented at all spatialscales; if this is correct, then participants should be less accu-

This article was published Online First December 23, 2013.Xue Han, Department of Psychology, Northeast Normal University,

Jilin, China; Suzanna Becker, Department of Psychology, McMaster Uni-versity, Hamilton, Ontario, Canada.

This research was funded by a grant from the Natural Sciences andEngineering Council, awarded to Suzanna Becker, and by Natural Sciencesand Engineering Council Scholarship PGS D3, awarded to Xue Han.

Correspondence concerning this article should be addressed to Xue Han,Department of Psychology, Faculty of Education, Northeast Normal Uni-versity, 5268 Renmin Street, Changchun, Jilin, China, 130024. E-mail:hanx100@nenu.edu.cn

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

Journal of Experimental Psychology:Learning, Memory, and Cognition

© 2013 American Psychological Association

2014, Vol. 40, No. 2, 511–5310278-7393/14/$12.00 DOI: 10.1037/a0035259

511

rate but equally fast at inferring directions between landmarksthat span two regions. This would be consistent with the use ofmultiscale hierarchical representations in which the landmarksare represented at all levels of the hierarchy, albeit morecoarsely at larger scales. However, if participants are bothslower and less accurate when landmarks are in different re-gions, they may be using postretrieval processes to integratetheir spatial knowledge across regions. This would be consis-tent with either the use of independent fragments or a hierar-chical representation in which the landmarks are represented atfiner (within-region) but not coarser (between-region) spatialscales. These predictions are summarized in Table 1.

There is extensive empirical support for local representations,but much less evidence that would differentiate between indepen-dent fragments and hierarchical representations. For example,there is evidence that participants are more readily primed bylocations in the same region than locations in different regions(Brockmole & Wang, 2002; McNamara, 1986), direction judg-ments between objects are affected by their superordinate spatialrelations (McNamara, 1986), between-array geometry influencesperformance on within-array judgments (Greenauer & Waller,2010), people make more errors in judging relations betweenlandmarks located at widely separated geographical locations (Ste-vens & Coupe, 1978), and distances between objects in the sameregion tend to be underestimated, whereas distances between ob-jects in different regions tend to be overestimated (McNamara,1986). However, in many of these studies, the distances betweenobjects within the same region versus across regions were not heldconstant (Greenauer & Waller, 2010; Stevens & Coupe, 1978), andreaction times (RTs) for distance and direction judgments were notreported (Greenauer & Waller, 2010; McNamara, 1986; Stevens &Coupe, 1978). In the few studies that have reported RTs for withinversus between region spatial decisions while controlling for dis-tance, evidence for hierarchical versus nonhierarchical theories ismixed. For example, participants were reported to have slowerresponses and larger errors for across-region pointing (Montello &Pick, 1993), consistent with the use of independent fragments oflocal knowledge. However, in a spatially primed object recogni-tion task, RTs were faster for same-region primes compared withbetween-region primes, but similar for same-region distant objectprimes versus different region close object primes (see Table 2,McNamara, 1986).

One factor that may be important in determining whether anenvironment is represented as one global representation or multi-ple independent local representations is whether the local regionswere learned together or separately. Evidence suggests than whentwo routes are learned on different levels of a building, pointing tolandmarks across multiple levels of the building incurs a cost inboth pointing latency and accuracy (Montello & Pick, 1993). TheRT difference suggests that people may have used postretrievalstrategies to integrate spatial information across multiple locally

constructed representations. It could be that combining spatialknowledge across three dimensions (multiple floors of a building)poses an unnatural barrier to spatial integration; within a two-dimensional environment, people may more readily integrate theirspatial knowledge from local regions into a global perspective.However, similar findings have been reported in two-dimensionalenvironments. When people were asked to estimate the directionsof landmarks seen along an S-shaped route and a U-shaped routeon a horizontal plane, they were less accurate in estimating land-marks when they were required to integrate their knowledge acrossthe two routes, regardless of the distance from the current perspec-tive to the landmark (Ishikawa & Montello, 2006). These data areconsistent with either the construction of multiple local represen-tations (independent fragments) or a hierarchy of representationswithin which landmarks are only represented at the local level.

Strong evidence for multiscale spatial representations comesfrom electrophysiological recordings of place cells. Place cells,first discovered by O’Keefe and Dostrovsky (1971) and subse-quently reported in many species including humans (Ekstrom etal., 2003), fire when the animal is within a specific local area andare often insensitive to heading direction (O’Keefe, 1976). Empir-ical evidence suggests that a place cell’s spatial tuning is based oninputs from a multitude of cues, including the location of localboundaries from subicular “boundary vector cells”(Burgess, Jack-son, Hartley, & O’Keefe, 2000; Hartley, Burgess, Lever, Cacucci,& O’Keefe, 2000; Lever, Burton, Jeewajee, O’Keefe, & Burgess,2009), path integration cues from entorhinal grid cells (Hafting,Fyhn, Molden, Moser, & Moser, 2005), and contextual cues (An-derson & Jeffery, 2003). Moreover, recordings made in very largespaces suggest that place cells form the basis of a hierarchical ormultiscale representation, with some place fields spanning theentire length of an 18-m track (Kjelstrup et al., 2008). Similarly,grid cells in the medial entorhinal cortex each fire at multiplelocations arranged in a hexagonal grid (e.g., Fyhn, Molden, Witter,Moser, & Moser, 2004; Hafting et al., 2005; Sargolini et al., 2006),and their firing fields vary in spatial scale (Barry, Hayman, Bur-gess, & Jeffery, 2007).

Given the above evidence for local multiscale representations ofspace, a key question is how larger regions of space are integrated,within very large-scale complex environments such as cities. Onepossibility is that local spatial representations are flexibly combinedduring the learning process into larger scale, more complex represen-tations. If this is the case, then one would expect to see local repre-sentations of connecting paths between regions, and place cells thatfire in one region or the other, but not both. Alternatively, separaterepresentations might be formed for different local regions and flex-ibly combined via postretrieval processes. In the latter case, onewould expect completely distinct sets of place cells firing in differentsubregions, and place cells with multiple unrelated firing fields indifferent regions. A wealth of electrophysiological evidence shedslight on this question. For example, in the hairpin maze, a complex

Table 1Predictions of Pointing Latency and Pointing Error Results by Different Theories for Cross- Versus Same-Region Pointing

Variable Independent fragments Hierarchical representation Flat representation

Pointing latency Slower No difference, assuming objects are represented at all levels No differencePointing error Larger Larger No difference

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

512 HAN AND BECKER

environment with multiple turning points, both place cells and gridcells showed similar remapping patterns at the turning points (Derdik-man et al., 2009), which suggests that regions separated by a turningpoint may be encoded as separate distinct representations; moreover,turning points may be encoded separately, providing a representa-tional bridge between the different regions. Other evidence seems tosupport piecemeal, fragmented representations of space (Derdikmanet al., 2009). For example, when two regions of an environment arelearned separately, the place cells in the two environments bear norelation to each other, and many cells have unrelated place fields inboth regions (Tanila, 1999). Furthermore, opening a connection be-tween the two environments causes many of the place cells to remapand/or develop a single place field in just one part of the environment(Paz-Villagrán, Save, & Poucet, 2004), suggesting that the animal istreating the unified space as a new environment and generating adistinct internal representation in the latter case. In the absence ofdistal cues or task constraints that would differentiate one compart-ment from another, place fields recorded in a regular multicompart-ment environment show repeated firing fields in the correspondinglocation of each compartment; moreover, a contextual change in onecompartment causes only local-remapping (Spiers, Hayman, Jovale-kic, Marozzi, & Jeffery, 2013).

On the basis of the above evidence from human behavioral andanimal electrophysiological studies, we predict that when there islocal spatial structure in a large-scale environment, people willtend to construct multiple local representations, particularly whenthe multiple environments are explored separately. Whether thelocally represented information is also represented at a coarserspatial scale may depend on a number of factors, including how theregions were explored and how salient the information is fororienting and navigation. Therefore, in our study, we had partici-pants explore two connected virtual neighborhoods and investi-gated various conditions in which they might be encoded as oneunitary spatial representation versus multiple piecemeal represen-tations. In the latter case, combining two local representationstogether should incur a cost in accuracy when inferring directionsfrom a location in one neighborhood to a location in the otherneighborhood. Moreover, because we tested spatial memory forlandmarks that were highly relevant to navigation, we predictedthat they would be represented at multiple spatial scales. There-fore, if hierarchical representations are used, we predicted thatpeople should be equally fast at making judgments between re-gions versus within regions. However, if independent fragmentsare used, there should be a cost in RT for integrating across localrepresentations at retrieval time.

Participants in the four experiments reported here learned thelayout of two connected neighborhoods in a town by playing avirtual taxi game, after which they completed a memory task. Thememory task required participants to point in the remembereddirection of each of a set of target locations from viewpoints ineach of the two neighborhoods. The target could either be in thesame neighborhood as the test viewpoint or in the other neighbor-hood, with within- and between-neighborhood trials intermixed.Importantly, in these two types of trials, we controlled for theangles and distances between the viewer and the target locations.The targets for the memory tests were passenger drop-off (PDO)locations, which were expected to be well-learned, salient land-marks, given the goals of the virtual taxi game. In Experiment 1,participants initially learned the two neighborhoods separately;

they were then shown a video clip of the pathway connecting thetwo neighborhoods, after which they explored the neighborhoodsjointly. Because the results were most consistent with participantsforming independent local representations (larger pointing latencyand pointing errors for across neighborhoods pointing), in Exper-iment 2 participants were given more time to learn the town, andwe varied the means by which they learned how the two neigh-borhoods were connected by allowing them to (a) view, but notnavigate, the connection between the neighborhoods; (b) view avideo clip of being teleported along the connection; or (c) freelynavigate along the connection. As in Experiment 1, within-neighborhood pointing to PDO locations was always more accu-rate than between-neighborhood pointing. However, in contrast tothe results of Experiment 1, pointing latency was no different forwithin- versus between-region locations, suggesting that partici-pants may have used hierarchical representations in Experiment 2.In Experiment 3, we examined the possibility that participants’between-neighborhood errors were due to an inability to accuratelyjudge the length of the connection between neighborhoods byremoving the fences that separated the two neighborhoods. Again,participants were more accurate but no faster for within-neighborhood PDO locations. Finally, in Experiment 4 we re-moved all distinct features that differentiated the town’s twoneighborhoods. In this final experiment, there was no difference inmemory between the two types of PDO locations in terms of eitherresponse speed or accuracy. Thus, participants were able to encodethe large town as one single environment when there were nodifferentiating cues to spatially group the features into local re-gions; in all other cases, their behavior was more consistent withthe construction of multiple local representations.

Experiment 1

After viewing images of the target and lure landmarks in apreexposure phase, participants implicitly learned the town layoutand PDO locations by playing a virtual taxi game requiring activenavigation. In subsequent spatial memory tests, they were asked topoint in the remembered directions of well-learned landmarks (thePDO locations) from cued viewpoints. Learning and spatial mem-ory test phases were repeatedly interleaved in five blocks, taking atotal of approximately 1 hr.

Method

Participants. Twenty-six McMaster University students(seven men and 19 women) of ages ranging from 18 to 32 years(mean age � 19.31) participated in the experiment. Participantshad normal or corrected-to-normal vision and received partialcourse credit for taking part in this experiment. Written informedconsent was obtained from all participants. This study was re-viewed and approved by the McMaster Research Ethics Board.

Materials. We used Kahana’s “Yellow Cab” virtual drivingsimulator (see http://memory.psych.upenn.edu/Research) to simu-late the virtual taxi game for the study phase of the experiment.Participants explored a single rectangular-shaped town (21 � 10VR units in size) consisting of two neighborhoods (see Figure 1a)connected by a navigable pathway that was initially occluded byan opaque, nonnavigable barrier. Each of the two neighborhoodsincluded eight distinctly textured and signed PDO locations as well

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

513ENCODING MULTIPLE CONNECTED ENVIRONMENTS

Figure 1 (opposite).

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

514 HAN AND BECKER

as uniformly textured gray background buildings. One of theneighborhoods, hereafter referred to as Neighborhood A (left sideof the town, colored in purple in Figure 1a), was designed as arestaurant district, and the other, which we refer to as Neighbor-hood B (right side of the town, colored in blue in Figure 1a), wasdesigned as a shopping district. Each neighborhood was sur-rounded by distinctly colored and textured fences. To help partic-ipants encode the pathway connecting the two neighborhoods, ineach neighborhood there was a distinct object located at the end ofthe connecting pathway that was visible from the other neighbor-hood. Each neighborhood had nine distinctively marked buildings,including the eight PDO locations and a building marking thestarting points for passenger pickup trials that was never a PDOtarget. On memory test trials, the cued viewpoint was always animage of one of the two starting point buildings, “Mike’s Restau-rant” (see Figure 2a) in Neighborhood A or “Aaron Chang Gal-lery” (see Figure 2b) in Neighborhood B. Thirty images of restau-rants and shops were viewed in the preexposure phase, 18 of whichappeared in the town and the remaining 12 of which served asdistractors for the subsequent memory task. Within each neigh-borhood, the restaurants/shops were located in an irregular con-figuration with the following constraints (see Figure 1a): For asubset of six of the eight restaurants (numbered in Figure 1a) andthe starting point in Neighborhood A, the configuration of theseseven locations was identical to a mirror image of the locations ofthe corresponding six shops and starting point in Neighborhood B.The remaining two restaurants and shops in each neighborhoodwere located irregularly to make the configurations asymmetric;however, only the responses to the symmetrically paired PDOlocations were included in the analyses, in order to control forpointing distances and angles within versus between neighbor-hoods. The identities and locations of the restaurants and shops inthe town remained constant across blocks. Participants were in-structed to press and hold down the arrow keys on the keyboard tocontrol their navigation, allowing them to turn in any direction,control their speed, or do a U-turn.

The memory test was similar to that used by Han et al. (2012).It was implemented in MATLAB with the Psychophysics Toolboxextension (Brainard, 1997; Pelli, 1997). On each memory test trial,a “navigator” appeared on the screen, consisting of a half compass-shaped figure with an image of the town from the test viewpoint(one of the two starting points in the navigation phase) at thebottom of the navigator (see Figure 3a and 3.3b) and a rotatable redcompass pointer. At the tip of the compass pointer, there was animage of the target restaurant or shop for the current trial, whichmoved with the pointer. The target was always one of the 30restaurants/shops shown in the preexposure phase, 18 of which hadbeen in the town and 12 of which were distractors. The participantmoved the mouse to rotate the compass pointer/target to the

remembered direction of the target and clicked the mouse toindicate their final response, or pressed the space bar to indicatethat a landmark was not recognized as having been in the town.

Procedure. There was a total of five blocks. Blocks 1, 2, 4,and 5 each included a preexposure phase, a study phase, and a testphase. Block 3 included a preexposure phase, a video clip viewingphase, and a test phase. In the preexposure phase, each of 30restaurants and shop images appeared for 2 s followed by a blankscreen for 1 s. The purpose of this preexposure phase was toestablish a degree of familiarity with the distractors for the sub-sequent recognition memory test. Participants were informed thatsome of the restaurants and shops would appear in the town. Ineach study phase, the participant was instructed to complete fivepassenger pickup trials. For each of these trials, the participant wasasked to act as a taxi driver, roam through the neighborhood to finda randomly located passenger, and deliver the passenger to therequested location. PDO locations were always restaurants orshops, which were visibly distinct, clearly signed locations in thetown. A passenger pickup trial began with the participant locatedat one of two starting points, facing toward the town. He or shewas asked to navigate freely until a passenger was found and“collected” by bumping into the passenger. A textual cue to thegoal location then appeared (e.g., “Please take me to the ComputerStore, I will give you 100 points”), and the participant was re-quired to navigate as quickly and efficiently as possible to drop offthe passenger to the goal location by bumping into it.

In Block 1, participants explored only half of the town, whereasin Block 2, participants explored the other half. Which of the twoneighborhoods was explored first was counterbalanced betweenparticipants. In the first two blocks, there was no visible or acces-sible connection between the two neighborhoods. In Block 3, therewas no active navigation; instead, after completing the preexpo-sure phase, participants were told that they would view a video clipillustrating how the two neighborhoods were connected. The videoshowed a trajectory of driving from one starting point to the other(from starting point A to B for half the participants and in thereverse direction for the other half). In Blocks 4 and 5, theparticipants were told that the visible barrier between the twoneighborhoods would be removed and the connection would beopen and navigable, allowing them to travel freely back and forthbetween the two neighborhoods and explore the entire space of thetown while locating and delivering passengers.

In each of the active navigation blocks (1, 2, 4, and 5), the studyphase was terminated when either the participant had successfullyfound and delivered five passengers or 10 min had elapsed. InBlocks 1 and 2, after each passenger delivery, the participant wasrelocated to the same starting point within the neighborhood tobegin the next passenger pickup. In the final Blocks 4 and 5 (after

Figure 1 (opposite). Towns’ layouts (21 � 10 VR units in size) used in Experiments 1 and 2. The gray squares are nondistinctive uniformly texturedbuildings at locations where the participants are not able to drive into. The black squares are also places where the participants are not able to drive into.The red “store” squares and red numbered squares are the stores that serve as passenger drop-off locations. The two green “Start” squares are the two startingpoints, locating at the two neighborhoods (“Mike’s Restaurant” and “Aaron Chang Gallery”). The smaller yellow squares are the two objects, one in eachneighborhood. All the white squares indicate locations along the routes that participants can navigate in the town. (a): Town used in Experiments 1 and2. (b): Town used in Experiment 3, similar to the Town used in Experiments 1 and 2, except the fences between the two neighborhoods were removed.(c): Town used in Experiment 4.

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

515ENCODING MULTIPLE CONNECTED ENVIRONMENTS

removal of the barrier), he or she was relocated to the starting pointin the adjacent neighborhood alternatingly between pickups.

Immediately after each study phase, participants completed asimultaneous test of recognition memory and spatial memory. Thetest required participants to point in the direction of the remem-bered location of each PDO building from a given viewpoint asquickly as possible, with no feedback. In Blocks 1 and 2, the testedviewpoint was always an image of the starting point in the sameneighborhood explored in the preceding study block. In Blocks 3,4, and 5, spatial memory for each PDO was tested twice, once fromeach of the two tested viewpoints/starting points (see Figure 3a andb). On each test trial, if the participant thought the restaurant orshop had not appeared in the town, he or she pressed the “spacebar,” and the next restaurant or shop would be displayed. Other-wise, he or she pointed in the remembered direction of the PDOlocation from the displayed viewpoint by using the mouse to move

the compass pointer in the desired direction, and then pressing theleft mouse button.1 After each response, the next target restaurantor shop would be displayed on the compass pointer. The order ofpresentation of the targets and tested viewpoints was randomizedwithin each block for each participant. The recognition responses,pointing directions, and total RT for the combined spatial/recog-nition memory response were recorded during the memory testphase. After five blocks of study and test phases, participants wereasked to draw a map of the town, including the two neighborhoods,restaurants and shops, on a blank piece of paper.

Data analysis. Participants’ data were excluded from furtheranalyses if they misaligned the two neighborhoods on the finalmap-drawing task, indicating that they failed to learn the overalltown layout.

Within-neighborhood pointing responses were defined as re-sponses made from the viewpoint of starting point A to Restau-rants 1a, 2a, and 3a in Neighborhood A, and from starting point Bto Shops 4b, 5b, and 6b in Neighborhood B (see Figure 1).Between-neighborhood pointing responses were defined as re-sponses from the viewpoint of starting point A to Shops 1b, 2b,and 3b in Neighborhood B, and from starting point B to Restau-rants 4a, 5a, and 6a in Neighborhood A.

In a preliminary analysis, to verify that spatial memory tests forthe shop and restaurant neighborhoods were matched for difficulty,the recognition accuracy, pointing latency, and pointing errors(which were recorded prior to participants being shown the con-nection between the two neighborhoods) for the six matchedrestaurants and shops in each neighborhood were compared acrossthe two neighborhoods.

Preliminary analyses included Block as a factor for all fourexperiments. Not surprisingly, significant improvement acrossblocks was seen in terms of both pointing responses and recogni-tion memory. More specifically, pointing latency and recognitionmemory accuracy improved significantly over blocks in all fourexperiments, and pointing errors improved significantly in allcases except Experiment 1. However, to simplify the presentationof results, block effects are not reported except in cases in whicha Block � Location (pointing to within- vs. between-neighborhoodlandmarks) interaction was significant. Bonferroni correctionswere used for all multiple comparisons throughout this article.

Recognition accuracy. A recognition response was counted ascorrect if the participant made a pointing response in any directionto a restaurant or shop that had appeared in the town, or pressed thespace bar for any lure image that had not appeared in the town. Wecalculated percent correct recognition separately for pointing toPDO locations within- and between neighborhoods. A two-tailedpaired sample t test was used to compare the difference in recog-nition memory accuracies between the two types of PDO locations.

Pointing latency. There were two RTs (in seconds) for eachPDO location in each block: one for each of the two testedviewpoints. RTs for correct responses, averaged across blocksand tested viewpoints, were compared for pointing to within-and between-neighborhood PDO locations using a two-tailedpaired sample t test.

1 We did not measure recognition memory reaction time separately, butwe did measure pointing latency, as our memory test combined recognitionand spatial memory.

Figure 2. (a): Starting point A (also a tested viewpoint): “Mike’s Res-taurant.” (b): Starting point B (also a tested viewpoint): “Aaron ChangGallery.”

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

516 HAN AND BECKER

Pointing error (average absolute pointing errors). Each par-ticipant had two pointing responses for each PDO location, onefrom the viewpoint of “Mike’s Restaurant” and one from theviewpoint of “Aaron Chang Gallery,” in each block. A pointingerror was calculated as the absolute value, in degrees, of thedifference between the pointing response direction and the PDOlocation’s actual direction. For each participant, average pointingerrors were calculated separately for within- and between-neighborhood pointing trials. For each of the two trial types, foreach cued viewpoint there were six pointing errors in each block(assuming that participants correctly recognized all PDO loca-tions); these were averaged across blocks and across viewpoints. Atwo-tailed paired sample t test was used to compare the difference

in pointing errors between the two types of PDO locations (within-vs. between neighborhoods).

Results

On the basis of their performance on the map-drawing task, sixparticipants (five women and one man) failed to learn the spatialrelationship between the neighborhoods; their data were excludedon that basis, leaving data from 20 participants (six men and 14women) for further analysis.

To check for differences in encoding difficulty between the twoneighborhoods, we performed a preliminary analysis of the datafrom Blocks 1 and 2. This analysis revealed that there was nodifference in spatial memory for buildings in the two neighbor-hoods in terms of recognition accuracy, t(19) � 1.633, p � .119,Cohen’s d � 0.51 (restaurant neighborhood: M � 92.50%, SE �2.6%; shopping neighborhood: M � 85.83%, SE � 3.3%); point-ing latency, t(19) � 0.554, p � .586, Cohen’s d � 0.13 (restaurantneighborhood: M � 5.45, SE � 0.49; shopping neighborhood:M � 5.17, SE � 0.45); or pointing errors, t(19) � 0.065, p � .949,Cohen’s d � 0.02 (restaurant neighborhood: M � 34.02, SE �2.79; shopping neighborhood: M � 33.75, SE � 3.07). In subse-quent analyses, we averaged data across all study blocks.

Recognition accuracy. A two-tailed paired sample t test ofthe recognition accuracy revealed no significant difference be-tween the two types of PDO locations, t(19) � 1.65, p � .116,Cohen’s d � 0.42 (within-neighborhood locations: M � 95.56%,SE � 1.2%; between-neighborhood locations: M � 92.78%, SE �1.7%; see Table 2).

Pointing latency. A two-way repeated measures Location �Block analysis of variance (ANOVA) of the pointing latenciesrevealed significant main effects of Location, F(1, 19) � 43.643,p � .001, �p

2 � 0.697, and Block, F(2, 38) � 19.772, p � .001,�p

2 � 0.510, and a significant interaction between Location andBlock, F(2, 38) � 3.825, p � .032, �p

2 � 0.168. To investigate theBlock � Location interaction, separate two-tailed paired sample ttests were used to compare within- and between-neighborhoodresponses at each block; these revealed that pointing latency wassignificantly faster for within-neighborhood than for between-neighborhood PDO locations at each block: Block 3,t(19) � �4.76, p � .001; Block 4, t(19) � �3.499, p � .002;Block 5, t(19) � �5.259, p � .001. Because the location effectwas significant at every block, pointing latencies were averagedacross blocks for further analysis.

Averaged across blocks, a two-tailed paired sample t test of thepointing latencies for within- versus between-neighborhood land-marks revealed a significant difference between the two types ofPDO locations, t(19) � �6.61, p � .001, Cohen’s d � �0.76.Responses were faster for pointing to locations within neighbor-hoods (M � 2.87, SE � 0.24) than between neighborhoods (M �3.76, SE � 0.29; see Figure 4 and Table 2).

Pointing errors. A two-tailed paired sample t test of thepointing errors averaged across blocks revealed a significant dif-ference between the two types of PDO locations, t(19) � �4.18,p � .001, Cohen’s d � �1.06. Errors were significantly smallerwhen pointing to PDO locations within neighborhoods (M �24.49, SE � 1.72) than between neighborhoods (M � 37.74, SE �3.57; see Figure 5 and Table 2).

Figure 3. (a) and (b) are the navigators used in the pointing task. At thetip of the compass pointer (red), an image was shown of the target store forthe current trial. It could be moved by moving the pointer. (a): Navigatorfrom “Mike’s Restaurant” point of view. (b): Navigator from “AaronChang Gallery” point of view.

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

517ENCODING MULTIPLE CONNECTED ENVIRONMENTS

Discussion

When the two neighborhoods were first learned separately, andlater explored jointly, participants behaved as if they constructedseparate spatial representations of the two neighborhoods. Wehypothesized that combining two local representations would re-sult in reduced accuracy and increased latency of between-neighborhood pointing responses if independent representationswere formed, both of which were seen in our results. Pointingresponses for PDO locations in the same neighborhoods as thetesting viewpoints were faster than for those in the adjacent neigh-borhoods, even when we only analyzed PDO locations that were,on average, equidistant from the observer (i.e., each restaurant orshop within the same neighborhood was paired with a shop orrestaurant at an equal distance and angle away from the observerin the other neighborhood). The longer time they took did not

improve their accuracies for PDO locations in the adjacent neigh-borhoods; on the contrary, pointing errors were larger for thosePDO locations. Interestingly, although the latency differences ap-peared to be dissipating across blocks (see Figure 4), the accuracydifferences showed a trend toward increasing across blocks (seeFigure 5). This could mean that as learning proceeds, participantsundergo a transition from separate, independent representations tointegrated hierarchical representations. Another possibility is thatparticipants formed a hierarchical representation of the townthroughout the experiment, but the PDO locations were onlyrepresented at the level of local neighborhoods, and not at a coarserspatial scale spanning the entire town. When required to point to aPDO location in a different neighborhood, their coarse level wouldnot be of any use for this task, forcing them to integrate their localrepresentations of buildings in each neighborhood at retrieval time.

Table 2Means and Standard Errors of Recognition Accuracy; Pointing Latency; and Pointing Errors in Experiments 1, 2 (Whole Condition),3, and 4

Experiment N

Recognition accuracy (in percentage) Pointing latency (in seconds) Pointing errors (in degrees)

Within Between Within Between Within Between

Experiment 1 20 95.56% (0.012) 92.78% (0.017) 2.87 (0.24) 3.76 (0.29) 24.49 (1.72) 37.74 (3.57)Experiment 2 (whole

condition) 12 98.27% (0.006) 97.74% (0.007) 3.21 (0.38) 2.94 (0.34) 35.26 (5.07) 50.01 (3.69)Experiment 3 20 96.67% (0.008) 98.61% (0.005) 3.50 (0.21) 3.48 (0.24) 22.20 (1.78) 41.95 (1.63)Experiment 4 20 96.53% (0.008) 97.36% (0.006) 3.91 (0.33) 3.69 (0.30) 35.25 (3.16) 39.29 (2.21)

Figure 4. Pointing latency for pointing to PDO locations within versusbetween neighborhoods in Blocks 3–5 in Experiment 1. White bars are forwithin-neighborhoods PDO locations, and gray bars are for between-neighborhoods PDO locations. The pointing latencies were significantlysmaller for PDO locations within the neighborhoods than for those betweenneighborhoods. PDO � passenger drop-off. Error bars represent �1 SE.

Figure 5. Errors for pointing to PDO locations within versus between neigh-borhoods in Blocks 3–5 in Experiment 1. White bars are for within-neighborhoodsPDO locations, and gray bars are for between-neighborhoods PDO locations. Thepointing errors were significantly smaller for PDO locations within the neighbor-hoods than for those between neighborhoods. PDO � passenger drop-off. Errorbars represent �1 SE.

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

518 HAN AND BECKER

The latter interpretation would not readily explain why the RTdifferences decreased across blocks.

The two neighborhoods may have been treated as separateenvironments because they were explored separately at the begin-ning of the experiment. If instead the two neighborhoods had beenexplored together from the beginning, they might have beentreated as a unified environment. Moreover, as noted above, thedifference in pointing latency between the two types of locationswas shrinking across blocks (see Figure 4). Moreover, in differentphases of Experiment 1, participants actually experienced the twoconnected neighborhoods in three different ways: first separately,then by passively viewing the connection between the neighbor-hoods, and finally, via joint exploration. It is possible that withmore learning time, and the opportunity to explore the two neigh-borhoods together from the outset, participants may have been ableto integrate their knowledge of the two neighborhoods into aglobal representation. Another factor that may be important indetermining whether two regions are integrated in spatial memoryis the manner in which they are explored, whether by directexperience or by simply observing the way regions are connected.

To tease apart the effect of learning by passive viewing versusactive navigation on the formation of integrated spatial represen-tations, in Experiment 2 we allowed them a greater number ofblocks to explore the two neighborhoods and varied the means bywhich they learned how the two neighborhoods were connected. Ina view-only condition, they could look along the pathway con-necting one neighborhood to the other but could not navigate alongit. In a teleport condition, they passively viewed a movie of thetrajectory along the connecting pathway. Finally, in a “wholetown” condition, they could navigate freely along the pathway ineither direction.

Experiment 2

In this experiment, each participant was pseudorandomly as-signed to one of three conditions: View, Teleport, and Whole. Thiswas a 2-hr experiment, with learning and spatial memory testphases repeatedly interleaved in six blocks.

Method

Participants. Sixty McMaster University students partici-pated in the experiment; age ranged from 18 to 37 years, and themean was 20.02. There were 20 participants in each condition (10men and 10 women). Participants had normal or corrected-to-normal vision. Participants received two course credits or $20 fortaking part in this experiment. This study was reviewed andapproved by the McMaster Research Ethics Board. Written in-formed consent was obtained from all participants.

Materials. The materials for the VR component of the studywere the same as those used in Experiment 1. In addition, we hadparticipants complete a questionnaire adapted from the Santa Bar-bara Sense of Direction Scale (SBSOD; Hegarty, Richardson,Montello, Lovelace, & Subbiah, 2002; see the Appendix) to in-vestigate what factors may be associated with performance on thepointing task. We added two questions to the original SBSODquestionnaire. Question 16 had participants rate on a 7-point scalehow likely they were to rely on a GPS when they travel to newplaces, whereas Question 17 asked participants whether they werevideo game players (a yes/no question).

Procedure. The experiment had six blocks, each consisting ofa preexposure phase as in Experiment 1, a study phase and a testphase. Each study phase used similar procedures to those ofExperiment 1, except that there were three conditions, whichvaried between subjects. In the view condition, in the first half ofeach study phase, participants explored one of the two neighbor-hoods but could see the other through a visible but nonnavigablepathway connecting the two neighborhoods. In the second half ofeach study phase, participants explored the other neighborhood,and again could see the first neighborhood along the connectingpath but could not cross through it. The starting neighborhood wascounterbalanced between participants. Throughout the studyphase, after each passenger delivery, participants were relocated tothe starting point in the same neighborhood to start another pas-senger pickup, until they had successfully found and delivered fivepassengers within that neighborhood, after which they completedfive passenger pickup trials in the other neighborhood.

The teleport condition was similar to the view condition exceptthat the pathway along the connection between neighborhoods wasneither visible nor accessible. In each block, after learning bothneighborhoods separately, the participants watched a video clipshowing how the two neighborhoods were connected: movingfrom starting point A to starting point B or from B to A. The videoclips and starting neighborhoods were counterbalanced betweenparticipants.

In the whole condition, the pathway connection was visible andnavigable throughout the experiment, and the starting neighbor-hood alternated between passenger pickup trials. Thus, each timea passenger was delivered, the participant was relocated to thestarting point in the other neighborhood, facing either “Mike’sRestaurant” or “Aaron Chang Gallery,” before being cued tocollect the next passenger. Unlike in the view and teleport condi-tions, the passenger and/or PDO location might be either in thesame neighborhood as the starting point or in the other neighbor-hood, thus requiring the participant to explore both neighborhoodsjointly.

In all three conditions, which neighborhood was explored firstwas counterbalanced between participants, and there were 10passenger deliveries in each of the six blocks, hence a total of 60passenger deliveries.

The same memory test and mapping task were used as in the lastthree blocks in Experiment 1. Additionally, at the end of the finalblock, participants were asked to answer the SBSOD question-naire, augmented with questions concerning their GPS usage andvideo game playing (see the Appendix).

Data analysis. Analyses were the same as those used as inExperiment 1, except we had three conditions. Therefore, weconducted two-way repeated measures ANOVAs, with Location(within vs. between neighborhoods) as a within-subject factor andCondition (view vs. teleport vs. whole) as a between-subjectfactor. We also separately analyzed the final performance in thelast block to investigate whether, with more learning time, the twoneighborhoods would eventually be treated as one environment.

In addition, based on the questionnaire results, we investigatedindividual differences in spatial abilities via correlational analyses.Answers to Questions 1–15 (the questions in the original versionof the SBSOD) were coded such that larger values indicated abetter sense of direction. This sum of ratings yielded final SBSODscores. Correlations were calculated between SBSOD scores and

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

519ENCODING MULTIPLE CONNECTED ENVIRONMENTS

the three memory measures: recognition accuracy, pointing la-tency, and pointing errors. Questions 16 and 17 were analyzedseparately. Question 16 had participants rate on a 7-point scalehow likely they were to rely on a GPS when they travel to newplaces, whereas Question 17 asked participants whether they werevideo game players (a yes/no question).

We also compared SBSOD scores of participants who correctlymapped the two starting points with those who did not to assess thevalidity of the map-drawing task.

Results

On the basis of the map-drawing task, four participants (twowomen and two men) in the view condition, three (two women and1 man) in the teleport condition, and eight (six women and twomen) in the whole condition incorrectly aligned the two neighbor-hoods; their data were excluded from further analyses of spatialmemory measures.

Recognition accuracy. A two-way repeated measures Loca-tion � Condition ANOVA of recognition memory accuracy re-vealed no significant effects or interactions: Location, F(1, 42) �3.85, p � .057, �p

2 � 0.084; Condition, F(2, 42) � 0.48, p � .624,�p

2 � 0.022; Location and Condition, F(2, 42) � 3.09, p � .056,�p

2 � 0.128 (within-neighborhood PDO locations: view: M �96.64%, SD � 3.9%; teleport: M � 97.44%, SD � 4.1%; whole:M � 98.27%, SD � 2.2%; between-neighborhood PDO locations:view: M � 97.71%, SD � 2.5%; teleport: M � 98.73%, SD �2.2%; whole: M � 97.74%, SD � 2.5%).

Pointing latency. A two-way repeated measures Location �Condition ANOVA of the pointing latencies revealed a main effectof Condition, F(2, 42) � 3.87, p � .029, �p

2 � 0.156, but no effectof Location, F(1, 42) � 0.062, p � .805, �p

2 � 0.001, nor anyinteraction between Condition and Location, F(2, 42) � 0.85, p �.434, �p

2 � 0.039 (see Figure 6). Post hoc Bonferroni-correctedt tests showed that pointing responses were significantly faster inthe whole condition than in the view condition (p � .033), but noother pairwise differences were significant (view: M � 4.23, SE �0.34; teleport: M � 4.45, SE � 0.33; whole: M � 3.07, SE �0.40).

Pointing errors. A two-way repeated measures Location �Condition ANOVA revealed main effects of Location, F(1, 42) �29.95, p � .001, �p

2 � 0.416, and Condition, F(2, 42) � 4.34, p �.019, �p

2 � 0.171, but no interaction between Condition andLocation, F(2, 42) � 0.58, p � .565, �p

2 � 0.027. Post hoccomparisons showed that between-neighborhood pointing errors(M � 41.48, SE � 1.99) were significantly larger than within-neighborhood errors (M � 29.84, SE � 1.97, p � .001). Pointingerrors in the whole condition (M � 42.64, SE � 3.20) weresignificantly larger than those in the view condition (p � .017,M � 30.34, SE � 2.77, p � .017), whereas there was no differencein responses between the view and teleport conditions (p � 1.00)or between the teleport and whole conditions (p � .135). Fordetails of pointing errors by blocks, see Figure 7.

Final performance. In all three measures, participants’ per-formance improved over blocks (e.g., see Figure 5 for pointingerrors). Given the trend in Experiment 1 for the Location effect todecrease across blocks, we therefore investigated whether theLocation effect dissipated after six blocks of learning had takenplace by analyzing all three memory measures in the final block

using two-way repeated measures Location � ConditionANOVAs. As in the main analysis, there were no significant maineffects or interactions in terms of recognition accuracy or pointinglatency. However, in terms of pointing errors, even in the finalblock of learning, there was a main effect of Location, F(1, 42) �26.07, p � .001, �p

2 � 0.383, but no effect of Condition, F(2, 42) �2.34, p � .109, �p

2 � 0.10, and no interaction between Locationand Condition, F(2, 42) � 0.128, p � .88, �p

2 � 0.006 (see Figure7). As in the overall analysis across all blocks, in the final block,pointing within neighborhoods was more accurate (mean error �22.65, SE � 1.87) than pointing between neighborhoods (meanerror � 36.71, SE � 2.36) across conditions.

Questionnaire results. T tests revealed that participants whocorrectly mapped the two starting points had higher SBSODscores, t(58) � 2.17, p � .039, Cohen’s d � 0.67, and recalledmore stores in the mapping task, t(58) � 3.372, p � .001, Cohen’sd � 0.92, than those who did not. Moreover, correlational analysesof all participants’ data (including those who did and did notcorrectly map the starting points) revealed that SBSOD scoreswere positively correlated with the number of stores recalled in themapping task, r(60) � .307, p � .017. This significant correlationwith a widely used test of spatial abilities provides validation ofour use of the pointing task and map-drawing task as measures ofspatial ability.

Video game players. Across conditions, participants with videogame experience (gamers) (n � 27, M � 16.04, SE � 0.47) recalled

Figure 6. Pointing latency for pointing to PDO locations within versusbetween neighborhoods in each condition in Experiment 2. White bars arefor within-neighborhoods PDO locations, and gray bars are for between-neighborhoods PDO locations. Left-side bars are for the view condition,middle bars are for the teleport (Tel) condition, and right-side bars are forthe whole condition. The pointing latencies were not significantly differentbetween the two types of locations, but were significantly faster in thewhole condition than in the view condition. PDO � passenger drop-off.Error bars represent �1 SE.

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

520 HAN AND BECKER

significantly more restaurants and shops in the mapping task thanthose without video game experience (nongamers) (n � 18, M �13.83, SE � 0.8), t(43) � 2.53, p � .015, Cohen’s d � 0.74, but wereno better in terms of recognition accuracy, pointing latency, or point-ing errors. Thus, video game experience translated into better recog-nition memory for landmarks but not into better judgments of thedirections of landmarks within or between neighborhoods.

The view condition. In the view condition, SBSOD scoreswere not significantly correlated with any of the measurements(see Table 3). GPS score (Question 16, on a 7-point scale, wherea high score indicates the participant is less likely to use a GPSwhen traveling to new places) positively correlated with recogni-tion accuracy, r(16) � .533, p � .033, and negatively correlatedwith pointing errors, r(16) � �.555, p � .026, when pointingbetween neighborhoods. Thus, people who relied less on a GPS indaily life were more accurate at integrating their knowledge be-tween neighborhoods.

The teleport condition. In the teleport condition, SBSODscores were significantly negatively related to pointing errors forwithin-neighborhood PDO locations, r(16) � �.718, p � .001(see Table 3). In other words, people with a better sense of

direction performed more accurately on the within-neighborhoodtrials. However, there were no significant correlations between anyof the other measures and the SBSOD scores, nor was GPS usagecorrelated with any of the measurements.

The whole condition. In the whole condition, the SBSODscores were significantly negatively correlated with pointing la-tency for between-neighborhood PDO locations, r(12) � �.591,p � .043, and marginally negatively correlated pointing latency forwithin-neighborhood PDO locations, r(12) � �.572, p � .052(see Table 3). No other correlations between the SBSOD scoreswere significant in this condition (see Table 3), nor were there anysignificant correlations between GPS usage and any of the memorymeasures in the whole condition. Thus, people with a better senseof direction, when forced to learn the two neighborhoods together,were faster but no more accurate at judging directions within andbetween neighborhoods.

Discussion

In all three conditions, regardless of whether the two neighbor-hoods were explored separately or jointly, participants were more

Figure 7. Errors for pointing to PDO locations within versus between neighborhoods in each block in eachcondition in Experiment 2. White bars are for within-neighborhoods PDO locations, and gray bars are for between-neighborhoods PDO locations. The top row is for the view condition, the middle row is for the teleport condition, andthe bottom row is for the whole condition. PDO � passenger drop-off. Error bars represent �1 SE.

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

521ENCODING MULTIPLE CONNECTED ENVIRONMENTS

accurate at pointing to PDO locations within neighborhoods thanbetween neighborhoods. This difference persisted even in the finalblock, in all three conditions, suggesting that even after extensivelearning, the two neighborhoods were still encoded as local represen-tations. However, in contrast to our findings in Experiment 1, therewere no pointing latency differences in any of the three conditions inExperiment 2. If people were constructing independent local repre-sentations, and were forced to integrate their knowledge at retrievaltime, we predicted that there would be a reaction time cost. Thus, ourresults suggest instead that participants were using hierarchical rep-resentations, with fine-scale local representations for each neighbor-hood and a coarser scale global representation for the whole town.One difference between these two experiments that might result indifferent strategies was the exploration time. In Experiment 1, partic-ipants had to complete a total of 20 passenger pickups, taking lessthan an hour to complete the experiment, as compared with 60pickups in Experiment 2, which took 2 hr to complete. It would be ofinterest for future research to further investigate the effect of learningexperience on the types of representations people form. It is possiblethat when local regions are initially explored, independent, frag-mented representations are formed, whereas with greater experience,people eventually form either hierarchical multiscale or flat unitaryrepresentations.

There were clear individual differences in how participants carriedout the pointing task. For example, in the view condition, participantswho relied less on a GPS in daily life were better at pointing to PDOlocations between neighborhoods. The view condition poses thegreatest challenge to participants in integrating their knowledge of thetwo neighborhoods. In contrast to the teleport and whole conditions,participants never visually experience moving along the corridor tosee how the neighborhoods are connected. They have to piece the twoneighborhoods together without actually either actively or passivelynavigating between them. Those who do not rely on a GPS may bemore adept at visuospatial imagery and therefore better able to imag-ine moving along the connecting pathway without having directlyexperienced it.

Interestingly, whereas GPS use negatively predicted perfor-mance in the view condition, sense of direction scores positively

predicted performance in the teleport and whole conditions. Thosewho had higher SBSOD scores had the advantage in both condi-tions, as evidenced by greater pointing accuracy in the teleportcondition and shorter pointing latencies in the whole condition. Itis possible that these two conditions favor two distinctly differentresponse strategies, a topic for future research.

Although the results of this experiment are consistent with thenotion that participants treated the two neighborhoods as twoseparate environments in all three conditions, an alternative expla-nation is that participants formed flat global representations of thetwo neighborhoods (or could precisely combine the two separaterepresentations) but made errors in judging the length of thepathway connecting the two neighborhoods, resulting in largererrors in pointing to PDO locations between neighborhoods. Thisalternate hypothesis is weakened by the observation that partici-pants in all three conditions exhibited the same PDO locationeffect even though they had different experiences with the con-nection between the neighborhoods. Nevertheless, to rule out thisalternative explanation, in Experiment 3, the fences between thetwo neighborhoods were removed while all other features of theneighborhoods remained the same. As a result, the possibility ofparticipants misjudging the length of the pathway should be re-duced, and therefore pointing errors for within- and between-neighborhood PDO locations ought to be the same if they are usingone flat representation.

Experiment 3

In Experiment 3, in order to encourage participants to form a singleglobal representation of the environment, they explored both neigh-borhoods jointly from the very beginning, as in the “whole” conditionin Experiment 2. The fences separating the two neighborhoods wereremoved (see Figure 1b), but the town still consisted of two visuallyand semantically distinct neighborhoods, a Restaurant district and aShopping district, each surrounded by differently colored fences.Therefore, we hypothesized that errors in pointing to PDO locationswithin neighborhood would be smaller than errors between neighbor-hoods. This result would suggest that the pointing error differenceobserved in Experiment 2 was not due to misjudgments of the lengthof the pathway, but rather to participants basing their responses ontwo separate local representations.

Method

Participants. Twenty McMaster University students (ninemen and 11 women) participated in this experiment; age rangedfrom 18 to 30 years, and the mean age was 20.9. Participants hadnormal or corrected-to-normal vision. Participants received eithertwo course credits or $20 for taking part in this 2-hr experiment.This study was reviewed and approved by the McMaster ResearchEthics Board. Written informed consent was obtained from allparticipants involved in this study.

Materials. The materials were the same as those used inExperiment 2, except that the fences along the pathway connectingthe two neighborhoods were removed (see Figure 1b). Therefore,there was no distinct pathway connecting the two neighborhoods.However, the two neighborhoods were still visually and semanti-cally distinct; the fences surrounded each neighborhood differed in

Table 3Correlation R Between SBSOD Scores and RecognitionAccuracy, Pointing Latency, and Pointing Errors inExperiment 2

Variable LocationView

(n � 16)Teleport(n � 17)

Whole(n � 12)

Recognition accuracy Within .102 .201 .119p � .707 p � .440 p � .712

Between .143 .247 .523p � .598 p � .338 p � .081

Pointing latency Within �.351 �.196 �.572p � .183 p � .451 p � .052

Between �.362 �.083 �.591p � .169 p � .751 p � .043�

Pointing errors Within .040 �.718�� .125p � .883 p � .001 p � .698

Between .093 �.418 �.135p � .733 p � .733 p � .676

Number of stores recalled .439 .449 �.004p � .089 p � .071 p � .990

Note. SBSOD � Santa Barbara Sense of Direction Scale.

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

522 HAN AND BECKER

color and texture, and one neighborhood contained only restau-rants, whereas the other contained only shops.

Procedure and data analysis. The procedures and analyseswere the same as those used in the whole condition in Experiment2. All participants successfully drew the map.

Results

Recognition accuracy. A two-way repeated measuresBlock � Location ANOVA revealed significant main effects ofBlock, F(5, 95) � 10.086, p � 0.001, �p

2 � 0.347, and Location,F(1, 19) � 6.603, p � .019, �p

2 � 0.258, and a significantinteraction between Location and Block, F(5, 95) � 4.26, p �.002, �p

2 � 0.183. Post hoc comparisons showed that recognitionaccuracy was significantly worse (p � .019) for PDO locationswithin the same neighborhood (M � 96.7%, SE � 0.8%) than forthose in the adjacent neighborhood (M � 98.6%, SE � 0.5%) andwas significantly worse in Block 1 than in Blocks 3 (p � .01), 4(p � .031), 5 (p � .024), and 6 (p � .009). Paired sample t testsat each block, corrected for multiple comparisons, revealed nosignificant differences between the two types of locations in any ofthe blocks.

Pointing latency. A two-tailed paired sample t test of thepointing latency revealed no significant difference between point-ing to PDO locations within versus between neighborhoods,t(19) � 0.122, p � .904, Cohen’s d � 0.02 (within-neighborhoodPDO locations: M � 3.50, SE � 0.21; between-neighborhoodPDO locations: M � 3.48, SE � 0.24; see Figure 8).

Pointing errors. A two-tailed paired sample t test of thepointing errors revealed a significant difference between the twotypes of PDO locations, t(19) � �14.00, p � .001, Cohen’sd � �2.59; errors were smaller for pointing to PDO locationswithin neighborhoods (M � 22.20, SE � 1.78) than between

neighborhoods (M � 41.95, SE � 1.63). For details of pointingerrors by blocks, see Figure 9.

Questionnaire results. Participants with video game experi-ence (n � 10) did not differ from nongamers (n � 10) in terms ofrecognition accuracy, pointing latency, pointing errors, or numberof restaurants and shops recalled in the map-drawing task.

None of the spatial memory measures were significantly corre-lated with the SBSOD scores (see Table 4). GPS usage wassignificantly negatively correlated with pointing latency forbetween-neighborhood PDO locations, r(20) � �.502, p � .024.Thus, less reliance on a GPS in daily life was associated with fasterpointing responses for between-neighborhood PDO locations.

Discussion

The results were similar to those obtained in the wholecondition in Experiment 2: Pointing to PDO locations was moreaccurate within neighborhoods than between neighborhoods,even in the last block (see Figure 7), and even in the face ofbetter recognition memory for between-neighborhood PDO lo-cations. These findings argue against the alternative hypothesisthat larger between-neighborhood pointing errors were due toparticipants misjudging the length of the pathway connectingthe two neighborhoods. Even without fences to separate the twoneighborhoods, participants still behaved as if they treated themas two separately and hierarchically encoded environments.Moreover, participants who relied on a GPS more often in theirdaily lives were slower at pointing to PDO locations in theadjacent neighborhoods.

One possible alternative explanation for the current findings isthat the environment simply was too big to be encoded as onerepresentation. Alternatively, the two types of PDO locations mayhave differed in important ways. For example, it is possible thatthe difference in pointing errors reflects the fact that all within-neighborhood PDO locations were located around the edge of thetown, whereas between-neighborhood PDO locations were locatedin the center of the town. To rule out these possibilities, weconducted a final experiment in which the types of PDO locationswere mixed between the two neighborhoods, and the environmentno longer contained distinct boundaries delineating the two neigh-borhoods.

Experiment 4

Experiment 4 was identical to Experiment 3 except that thedistinctly colored fences surrounding the two neighborhoods werereplaced by uniformly textured walls, and the restaurants andshops were intermixed within the town. Therefore, in contrast tothe three previous experiments, there were no spatial, visual, orsemantic features to differentiate the two neighborhoods. Never-theless, we still analyzed the same six pairs of PDO locations (seeFigure 1c) as in the previous experiments. For cross-experimentcomparison purposes, we therefore still refer to them as within-and between-neighborhood PDO locations. We hypothesized thatthere would be no difference in pointing responses for within- andbetween-neighborhood PDO locations. This result would suggestthat the large pointing error differences observed in Experiments1–3 were caused by participants constructing and using separaterepresentations of the distinct neighborhoods, whenever there were

Figure 8. Pointing latency for pointing to PDO locations within versusbetween neighborhoods in each condition in Experiment 3. White bars arefor within-neighborhoods PDO locations, and gray bars are for between-neighborhoods PDO locations. The pointing latencies were not differentbetween two types of locations. PDO � passenger drop-off. Error barsrepresent �1 SE.

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

523ENCODING MULTIPLE CONNECTED ENVIRONMENTS

features available to differentiate and cluster local landmarks intosubregions.

Method

Participants. Twenty McMaster University students (11 menand nine women) participated in the experiment; age ranged from18 to 35 years, and the mean age was 20.6. Participants had normalor corrected-to-normal vision and received either two course cred-its or $20 for taking part in this 2-hr experiment. This study wasreviewed and approved by the McMaster Research Ethics Board.Written informed consent was obtained from all participants in-volved in this study.

Materials. The materials were the same as those used inExperiment 3, except for the following changes to the town: Thefences distinguishing the two neighborhoods were removed, andrestaurants and shops were intermixed across the town rather thanbeing clustered by category within one or the other neighborhood(see Figure 1c). Therefore, there was no pathway connecting thetwo neighborhoods, and there were no other distinctions betweenthe two neighborhoods. In this case, therefore, the town should beperceived as one environment.

Procedure and data analysis. The procedures and analyseswere the same as those used in Experiment 3. All participantssuccessfully drew the map.

Results

Recognition accuracy. A two-tailed paired sample t test ofthe recognition accuracy revealed no significant difference be-tween the two types of PDO locations, t(19) � �1.19, p � .249,Cohen’s d � �0.25 (within-neighborhood PDO locations: M �96.53%, SE � 0.8%; between-neighborhood PDO locations: M �97.36%, SE � 0.7%).

Pointing latency. A two-tailed paired sample t test of thepointing latency revealed no significant difference between thetwo types of PDO locations, t(19) � 1.43, p � .169, Cohen’s d �0.16 (within-neighborhood PDO locations: M � 3.91, SE � 0.33;between-neighborhood PDO locations: M � 3.69, SE � 0.30; seeFigure 10).

Pointing errors. A two-tailed paired sample t test of thepointing errors revealed no significant difference between the twotypes of PDO locations, t(19) � �1.74, p � .098, Cohen’sd � �0.33 (within-neighborhood PDO locations: M � 35.25,SE � 3.16; between-neighborhood PDO locations: M � 39.29,SE � 2.21; see Figure 11).

Questionnaire results. Participants with video game expe-rience (n � 12) did not differ from nongamers (n � 8) on anyof the performance measures.

None of the memory measures correlated significantly witheither SBSOD scores or GPS usage (see Table 4).

Discussion

Unlike the previous three experiments, in Experiment 4 therewas no indication that errors depended on whether participantswere pointing to within- or between neighborhoods, suggestingthat participants may have encoded the large-scale environmentwithin a single flat spatial representation. Although the differencein pointing errors for within- and between-neighborhood trials wasnot significant (p � .098), it was in the same direction as in theprevious three experiments. This could be due to the positions oftwo types of locations in the town. In our study, the between-neighborhood PDO locations were arranged in the middle of the

Table 4Correlation R Between SBSOD Scores and RecognitionAccuracy, Pointing Latency, and Pointing Errors in Experiments3 and 4

Variable LocationExperiment 3

(n � 20)Experiment 4

(n � 20)

Recognition accuracy Within �.076 �.061p � .749 p � .779

Between �.018p � .940

Pointing latency Within �.075 �.035p � .755 p � .883

Between �.239p � .311

Pointing errors Within .140 .241p � .557 p � .306

Between �.217p � .358

Number of stores recalled .159 .022p � .504 p � .928

Note. SBSOD � Santa Barbara Sense of Direction Scale.

Figure 9. Errors for pointing to PDO locations within versus betweenneighborhoods in each block in Experiment 3. White bars are forwithin-neighborhoods PDO locations, and gray bars are for between-neighborhoods PDO locations. The pointing errors were significantlysmaller for within-neighborhood PDO locations than for between-neighborhood PDO locations. PDO � passenger drop-off. Error barsrepresent �1 SE.

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

524 HAN AND BECKER

town, and the within-neighborhood PDO locations were arrangedcloser to the edges of the town (see Figure 1). It could be that it ismore difficult to estimate the directions of buildings that are closerin the middle of the town than those buildings that are closer to theedges. However, this does not explain why in the previous threeexperiments we observed very large, highly significant differencesin within- versus between-neighborhood pointing errors relative tothe very small, nonsignificant difference observed here.

Cross-experimental comparisons. Our main question in thestudies reported was whether there would be latency and/oraccuracy differences when participants tried to remember thelocations of landmarks in the same versus in a different neigh-borhood from the tested viewpoint. We found a location effect(a difference between within- and between-neighborhood point-ing trials) on response latency in Experiment 1 but in no otherexperiment. However, we found a location effect on responseaccuracy in Experiments 1–3 but not in Experiment 4. Accord-ing to our predictions laid out in Table 1, these results areconsistent with the use of independent representations of thetwo neighborhoods in Experiment 1 and integrated representa-tions in Experiments 2– 4, and the use of hierarchical represen-tations in Experiments 2–3 but not in Experiment 4. Alterna-tively, there may have been other differences between theexperiments that meant the difficulty of within-neighborhoodpointing trials varied, in the absence of qualitative differencesin the type of representation participants may have used. Tofurther explore this possibility, cross-experiment comparisonson both pointing latency and accuracy were conducted amongExperiments 1 (Blocks 3, 4, and 5), 2 (whole condition), 3, and4 (see Figure 12).

Pointing latency. To investigate the possibility that the spa-tial learning problem faced by participants was somehow easierin Experiment 1 relative to Experiments 2– 4, resulting in longer

within-neighborhood response latencies in the latter experi-ments, a cross-experimental comparison was made using atwo-way repeated measures Location � Experiment ANOVAof the pointing latencies. This analysis revealed a significantinteraction between Location and Experiment, F(3, 68) �13.003, p � .001, �p

2 � 0.365, but no main effects of Location,F(1, 68) � 1.489, p � .227, �p

2 � 0.021, or Experiment, F(3,68) � 1.093, p � .358, �p

2 � 0.046 (see Figure 13). Separatepost hoc one-way ANOVAs of pointing latencies on within- andbetween-neighborhood trials revealed that response speed didnot differ across experiments for between-neighborhood trials,F(3, 71) � 1.29, p � .285, �p

2 � 0.054, but was marginallysignificantly different across experiments for within-neighborhood trials, F(3, 71) � 2.60, p � .059, �p

2 � 0.103 (seeTable 2 for means and SEs). Post hoc Bonferroni-correctedpairwise comparisons revealed that within-neighborhood point-ing was marginally slower in Experiment 1 compared withExperiment 4 (p � .051); no other pairwise differences weresignificant (Experiment 1 vs. Experiment 2: p � 1.00; Exper-iment 1 vs. Experiment 3: p � .626; Experiment 2 vs. Exper-iment 3: p � 1.00; Experiment 2 vs. Experiment 4: p � .704;Experiment 3 vs. Experiment 4: p � 1.00).

Pointing errors. To investigate the possibility that the spa-tial learning problem was more difficult in Experiment 4 rela-tive to the other experiments, resulting in larger errors onwithin-neighborhood trials in Experiment 4, a cross-experimental comparison was made using a two-way repeatedmeasures Location � Experiment ANOVA of the pointingerrors. This analysis revealed main effects of Location, F(1,68) � 86.687, p � .001, �p

2 � 0.56, and Experiment, F(3, 68) �3.988, p � .011, �p

2 � 0.15, and a significant interactionbetween Location and Experiment, F(3, 68) � 6.413, p � .001,

Figure 10. Pointing latency for pointing to PDO locations within versusbetween neighborhoods in Experiment 4. White bars are for within-neighborhoods PDO locations, and gray bars are for between-neighborhoods PDO locations. There was no difference between the twotypes of PDO locations. PDO � passenger drop-off. Error bars represent�1 SE.

Figure 11. Errors for pointing to PDO locations within versus betweenneighborhoods in Experiment 4. White bars are for within-neighborhoodsPDO locations, and gray bars are for between-neighborhoods PDO loca-tions. There was no difference between the two types of PDO locations.PDO � passenger drop-off. Error bars represent �1 SE.

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

525ENCODING MULTIPLE CONNECTED ENVIRONMENTS

�p2 � 0.221 (see Figure 14). Post hoc comparisons showed that

pointing errors were larger for between-neighborhood PDOlocations (M � 29.30, SE � 1.43) than for within-neighborhoodPDO locations (M � 42.25, SE � 1.43) across experiments, andwere significantly larger in Experiment 2 (whole condition)than those in Experiment 1 (p � .019) and Experiment 3 (p �.04) across locations, but no other differences were significant(Bonferroni-corrected) (Experiment 1: M � 31.11, SE � 2.31;Experiment 2-whole condition: M � 42.64, SE � 2.98; Exper-iment 3: M � 32.08, SE � 2.31; Experiment 4: M � 37.27,SE � 2.31). Separate one-way ANOVAs of within-neighborhood and between-neighborhood pointing errors wereused to investigate the significant interaction between Locationand Experiment, which revealed a main effect of Experimentfor both within-, F(3, 71) � 6.14, p � .001, �p

2 � 0.213, andbetween-neighborhood trials, F(3, 71) � 2.987, p � .037, �p

2 �0.116 (see Table 2 for means and SEs). Post hoc comparisonsshowed that pointing errors on within-neighborhood trials weresignificantly smaller in Experiment 1 than those in Experiment4 (p � .032) and smaller in Experiment 3 than those in Exper-iment 2 (whole condition; p � .021) and Experiment 4 (p �.005). Pointing errors on between-neighborhood trials weresmaller in Experiment 1 than those in Experiment 2 (whole

condition; p � .035). There were no other significant differ-ences (Bonferroni correction was applied).

Summary

Cross-experimental comparisons revealed that within-neighborhood pointing latencies did not differ between Experi-ment 1 versus Experiment 2 or between Experiment 1 versusExperiment 3, undermining the alternative explanation that the“location effect” in Experiment 1 (i.e., a within- vs. between-neighborhood difference) and total absence of a location effect inExperiments 2 and 3 was simply due to differences in task diffi-culty. However, in Experiment 4, response latencies on within-neighborhood pointing trials were marginally longer than those inExperiment 1, whereas there was no such cross-experiment differ-ence on between-neighborhood trials. This is most likely due to thefact that in Experiment 4, both types of PDOs (shops and restau-rants) were intermixed throughout the town, eliminating the salientfeatures that might differentiate the two neighborhoods, therebymaking within-neighborhood pointing trials just as difficult asbetween-neighborhood trials.

Cross-experiment comparisons of pointing errors presented amore complex picture. For within-neighborhood pointing trials,

Figure 12. Recognition accuracy for within versus between-neighborhoods PDO locations in each experiment.White bars are for within-neighborhoods PDO locations, and gray bars are for between-neighborhoods PDOlocations. Recognition accuracies were significantly worse in Experiment 1 than those in Experiments 2 (wholecondition) and 3 across locations, but there were no other differences among Experiments 2, 3, and 4.Recognition accuracy for between-neighborhood PDO locations was worse in Experiment 1 than in otherexperiments. PDO � passenger drop-off. Error bars represent �1 SE.

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

526 HAN AND BECKER

errors were smaller in Experiment 1 than in Experiment 4, smallerin Experiment 3 than in Experiment 2 (whole condition), andsmaller in Experiment 3 than in Experiment 4. Overall, thecross-experiment comparisons of both pointing latencies andpointing errors suggest that the learning problem faced byparticipants in Experiment 4 was indeed more difficult than thatof the other experiments. This is not surprising, considering thatparticipants were not given any cues that could serve to clusterthe environment into smaller regions.

General Discussion

We hypothesized that when people form spatial representationsof large-scale complex environments, they may use either (a)global, flat representations; (b) multiscale hierarchical representa-tions; or (c) independent, local fragments. Each of these makesdistinct predictions about accuracy and latency for making within-versus between-region judgments (see Tables 1 and 5). Across thefour experiments, we found evidence for all three types of repre-sentations.

One of our main conclusions is that when participants had fewertrials over which to learn the town layout (Experiment 1), theyappeared to form separate, independent representations of the twoneighborhoods, whereas when given more learning time (Experi-ments 2–4), they appeared to form integrated representations of

the town. This conclusion is supported by the significant differencein latency when participants were pointing to locations within-versus between neighborhoods in Experiment 1. Importantly, thiseffect disappeared when participants were given more time toexplore the two neighborhoods jointly (Experiment 2, whole con-dition and Experiments 3 and 4). Another interpretation that wouldbe equally consistent with this pattern of results is that in Exper-iment 1, participants formed hierarchical representations of the twoneighborhoods, but the landmarks were only represented at thelocal, fine-scale level and not at the coarser scale. Thus, whenrequired to point to a landmark in another neighborhood, theywould have to integrate their separate, fine-scale representationsusing postretrieval processes. In either case, the transition to moreglobal, integrated representations may be a natural progression inthe formation of spatial representations of large-scape spaces, asexperience in an environment accumulates. Future work couldexplore this possibility by varying the amount of preexposure toindividual neighborhoods versus joint exploration of the twoneighborhoods.

Another major conclusion from our experiments is that peoplehave a strong tendency to form hierarchical representations. Thusin Experiments 2 and 3, although participants’ response latencieswere no different for pointing within- versus between neighbor-hoods, they were much more accurate on within- compared with

Figure 13. Pointing latency for within versus between-neighborhoods PDO locations in each experiment.White bars are for within-neighborhoods PDO locations, and gray bars are for between-neighborhoods PDOlocations. Pointing latencies were not different among experiments or locations. PDO � passenger drop-off.Error bars represent �1 SE.

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

527ENCODING MULTIPLE CONNECTED ENVIRONMENTS

between-neighborhood pointing trials. However, when we re-moved all visible distinctions between the two neighborhoods inExperiment 4, there was no such effect of location on pointingaccuracy. According to our original hypotheses laid out in Table 1,this pattern of findings is consistent with the use of hierarchicalrepresentations in Experiments 2 and 3 and a single flat represen-tation in Experiment 4. One potential difficulty with this interpre-tation is that in Experiment 4, the notion of a “neighborhood” isill-defined. Our division of the town into two neighborhoods inExperiment 4, using the town midline, was done for convenience,to allow cross-experiment comparisons. From the participants’perspective, there was no obvious basis on which to segregate thetown into two neighborhoods, so the entire town would haveseemed like a single large neighborhood. Thus, our definition of“within-neighborhood pointing” for analysis purposes may havebeen at odds with participants’ perception of what constituted a

neighborhood in the town. Further research is required to disen-tangle the relative contributions of spatially localized distinguish-ing features versus the size of a region to the formation of spatialrepresentations.

Overall, our results suggest that people have a very strongtendency to form hierarchical representations, grouping spaceinto local regions based on common visual, semantic, andgeometric features. Early in the acquisition process, people mayfirst form local independent maps of regions, or less detailedhierarchical representations where landmarks are only repre-sented at fine scales (as suggested in Experiment 1), incurringa reaction time cost when inferring directions between neigh-borhoods. Once sufficient learning has occurred (Experiments 2and 3), people may form global representations at a coarserscale, in addition to fine-scale regional representations. Onlywhen we removed all available cues that could serve to locally

Figure 14. Pointing errors for within versus between-neighborhoods PDO locations in each experiment. Whitebars are for within-neighborhoods PDO locations, and gray bars are for between-neighborhoods PDO locations.PDO � passenger drop-off. Error bars represent �1 SE.

Table 5Supporting Evidence for Each of the Theories for Cross- Versus Same-Region Pointing

VariableIndependentfragments

Hierarchicalrepresentations

Flatrepresentation

Pointing latency Slower No diff. No diff.Pointing error Larger Larger No diff.Results Experiment 1 Experiments 2 & 3 Experiment 4

Note. diff. � difference.

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

528 HAN AND BECKER

segregate the town into two neighborhoods did people behaveas if they formed a single, flat representation of the town, withequal accuracy and latency for within- versus between-regionjudgments.

The setup of the virtual towns used here was inspired by similarstudies in the rodent literature, in which place cells are recorded inenvironments consisting of two connected boxes. It is interestingto compare these place cell data with the behavioral findings in ourown study. In one such study, when rats explored two identicalboxes with a connecting corridor, after removal of the barrier in thecorridor, place cells underwent a partial remapping, such that someof the CA1 pyramidal cells showed similarly shaped spatial firingfields in both boxes, but others showed completely different spatialfiring fields in the two boxes (Skaggs & McNaughton, 1998),suggesting that the rats may have treated the two boxes as separateenvironments.

Initially, we analyzed sex differences but failed to find anysignificant differences between males and females in any of thespatial memory measurements. Therefore, we dropped this fac-tor from our current analyses. However, there were other indi-vidual differences apparent in our results in terms of sense ofdirection and GPS usage. Interestingly, GPS usage and videogame experience have different effects on spatial memory andrecall memory. When the two neighborhoods were distinct,video game players showed better recall memory than nongam-ers in the mapping task. Unfortunately, we did not ask moredetailed questions about the type of video game most oftenplayed by gamers. Future studies could explore, for example,whether the use of first-person perspective versus aerial per-spective is associated with different spatial abilities. Interest-ingly, in the view condition, greater GPS usage predicted pooreraccuracy in making between-neighborhood spatial judgments,suggesting that individuals who frequently rely on a GPS mayhave greater difficulty integrating their knowledge over large-scale spaces. As noted earlier, the view condition places thegreatest demands on participants compared with the whole orteleport conditions, requiring them to integrate knowledge be-tween neighborhoods when they have never directly experi-enced moving from one neighborhood to the other. It could bethat people who rely on a GPS when traveling to new placestend to have poorer mental imagery abilities, and thus greaterdifficulty imagining how the two neighborhoods are connectedin the view condition.

What types of spatial representations did participants use inour tasks? Tolman (1948) proposed that two different types ofmap may underlie spatial cognition: striplike maps and rela-tively broad and comprehensive maps. In a strip map, an ani-mal’s position and the goal position are connected by a singlepath, which is less flexible in the face of changes in theenvironment. In contrast, using a comprehensive map, the an-imal could behave correctly in the face of changes made in theenvironment. An updated version of this view is that routememories and cognitive maps are two separate and distincttypes of spatial representation, with distinct neural substrates(Hartley, Maguire, Spiers, & Burgess, 2003). Although theexistence of place cells is suggestive of a comprehensive map,it is unclear how knowledge from multiple place cells, eachrepresenting a local region of space, is integrated and used tosupport navigation to a goal. McNaughton and colleagues

(1996) suggested that an abstract mental representation of atwo-dimensional environment requires input from place cellsand head direction cells, which convey self-motion informationfor path integration and landmark information, respectively.The landmark information could be used to correct for errorsaccumulated during path integration, but the path integrationsystem should still work without it (McNaughton et al., 1996).Trullier and Meyer (2000) proposed a computational model ofnavigation, in which they consider the hippocampus as a “cog-nitive graph” (i.e., heteroassociative network). The temporalsequences of visited places are learned and an environment’stopological representation is formed by using a “place-recognition-triggered response” strategy, which is stored by thenetwork. In this case, the model could predict the next positionbased on the current position by using place cells, goal cells,and subgoal cells. Burgess and colleagues proposed a modelwhereby Hebbian modification of synaptic strengths betweenplace cell representations generates representations of goal lo-cations (Burgess et al., 2000). Foster, Morris, and Dayan (2000)proposed a hybrid reinforcement learning model that includes amodule for learning goal-independent coordinate representa-tions of space and a second module that gradually learns goal-oriented state–action sequences. All of these models suggestways in which local information from place cells could beintegrated into continuous spatial representations.

Place cells appear to be the basis for representing very smalllocal patches of space, and must be integrated to form regionalspatial representations. At a larger scale, local spatial represen-tations must somehow be integrated to form global maps ofspace. Local representations could be connected hierarchicallyinto a multiscale global representation. Alternatively, theycould be chained together via an associative learning process asparticular trajectories are experienced, resulting in a represen-tation of a single connected pathway. It is possible that whenthe two environments are always connected by a path, peoplemay combine them as a chain of local representations. Incontrast, when there are multiple connections between the twoenvironments or after extensive experience with the singleconnection, people may be able to combine them into a globalrepresentation. These are important avenues for future research.

References

Anderson, M. I., & Jeffery, K. J. (2003). Heterogeneous modulation ofplace cell firing by changes in context. Journal of Neuroscience, 23,8827–8835.

Barry, C., Hayman, R., Burgess, N., & Jeffery, K. J. (2007). Experience-dependent rescalling of entorhinal grids. Nature Neuroscience, 10, 682–684. doi:10.1038/nn1905

Brainard, D. H. (1997). The Psychophysics Toolbox. Spatial Vision, 10,433–436.

Brockmole, J. R., & Wang, R. F. (2002). Switching between environmentalrepresentations in memory. Cognition, 83, 295–316. doi:10.1016/S0010-0277(02)00012-4

Burgess, N., Jackson, A., Hartley, T., & O’Keefe, J. (2000). Predictionsderived from modeling the hippocampal role in navigation. BiologicalCybernetics, 83, 301–312. doi:10.1007/s004220000172

Derdikman, D., Whitlock, J. R., Tsao, A., Fyhn, M., Hafting, T., Moser,M. B., & Moser, E. I. (2009). Fragmentation of grid cell maps in amulticompartment environment. Nature Neuroscience, 12, 1325–1332.doi:10.1038/nn.2396

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

529ENCODING MULTIPLE CONNECTED ENVIRONMENTS

Ekstrom, A. D., Kahana, M., Caplan, J., Fields, T., Isham, E., Newman, E.,& Fried, I. (2003). Cellular networks underlying human spatial naviga-tion. Nature, 425, 184–188. doi:10.1038/nature01964

Foster, D. J., Morris, R. G. M., & Dayan, P. (2000). A model of hippocam-pally dependent navigation, using the temporal difference learning rule.Hippocampus, 10, 1–16. doi:10.1002/(SICI)1098-1063(2000)10:1�1::AID-HIPO1�3.0.CO;2-1

Fyhn, M., Molden, S., Witter, M. P., Moser, E. I., & Moser, M. (2004).Spatial representation in the entorhinal cortex. Science, 305, 1258–1264.doi:10.1126/science.1099901

Greenauer, N., & Waller, D. (2010). Micro- and macro-reference frames:Specifying the relations between spatial categories in memory. Journalof Experimental Psychology: Learning, Memory, and Cognition, 36,938–957.

Hafting, T., Fyhn, M., Molden, S., Moser, M. B., & Moser, E. (2005).Microstructure of a spatial map in the entorhinal cortex. Nature, 436,801–806. doi:10.1038/nature03721

Han, X., Byrne, P., Kahana, M., & Becker, S. (2012). When do objectsbecome landmarks? A VR study of the effect of task relevance on spatialmemory. PLoS ONE, 7, e35940. doi:10.1371/journal.pone.0035940

Hartley, T., Burgess, N., Lever, C., Cacucci, F., & O’Keefe, J. (2000).Modeling place fields in terms of the cortical inputs to the hippocampus.Hippocampus, 10, 369–379. doi:10.1002/1098-1063(2000)10:4�369::AID-HIPO3�3.0.CO;2-0

Hartley, T., Maguire, E. A., Spiers, H. J., & Burgess, N. (2003). Thewell-worn route and the path less traveled: Distinct neural bases of routefollowing and wayfinding in humans. Neuron, 37, 877– 888. doi:10.1016/S0896-6273(03)00095-3

Hegarty, M., Richardson, A. E., Montello, D. R., Lovelace, K., & Subbiah,I. (2002). Development of a self-report measure of environmental spatialability. Intelligence, 30, 425–447. doi:10.1016/S0160-2896(02)00116-2

Hirtle, S. C., & Jonides, J. (1985). Evidence of hierarchies in cognitivemaps. Memory & Cognition, 13, 208–217. doi:10.3758/BF03197683

Ishikawa, T., & Montello, D. R. (2006). Spatial knowledge acquisitionfrom direct experience in the environment: Individual differences in thedevelopment of metric knowledge and the integration of separatelylearned places. Cognitive Psychology, 52, 93–129. doi:10.1016/j.cogpsych.2005.08.003

Janzen, G., & van Turennout, M. (2004). Selective neural representation ofobjects relevant for navigation. Nature Neuroscience, 7, 673–677. doi:10.1038/nn1257

Kjelstrup, K. B., Solstad, T., Brun, V. H., Hafting, T., Leutgeb, S., Witter,M. P., . . . Moser, M. (2008). Finite scale of spatial representation in thehippocampus. Science, 321, 140–143. doi:10.1126/science.1157086

Lever, C., Burton, S., Jeewajee, A., O’Keefe, J., & Burgess, N. (2009).Boundary vector cells in the subiculum of the hippocampal formation.Journal of Neuroscience, 29, 9771–9777. doi:10.1523/JNEUROSCI.1319-09.2009

McNamara, T. P. (1986). Mental representations of spatial relations. Cog-nitive Psychology, 18, 87–121. doi:10.1016/0010-0285(86)90016-2

McNaughton, B. L., Barnes, C. A., Gerrard, J. L., Gothard, K., Jung,M. W., Knierim, J. J., . . . Weaver, K. L. (1996). Deciphering thehippocampal polyglot: The hippocampus as a path integration system.Journal of Experimental Biology, 199, 173–185.

Meilinger, T. (2008). The network of reference frames theory: A synthesisof graphs and cognitive maps. In C. Freksa, N. S. Newcombe, P.Gärdenfors, & S. Wölfl (Eds.), Spatial Cognition VI, LNAI 5248 (pp.344–360). Berlin, Germany: Springer.

Montello, D. R., & Pick, H. L., Jr. (1993). Integrating knowledge ofvertically aligned large-scale spaces. Environment and Behavior, 25,457–484. doi:10.1177/0013916593253002

O’Keefe, J. (1976). Place units in the hippocampus of the freely movingrat. Experimental Neurology, 51, 78–109.

O’Keefe, J., & Dostrovsky, J. (1971). The hippocampus as a spatial map:Preliminary evidence from unit activity in the freely-moving rat. BrainResearch, 34, 171–175. doi:10.1016/0006-8993(71)90358-1

Paz-Villagrán, V., Save, E., & Poucet, B. (2004). Independent coding ofconnected environments by place cells. European Journal of Neurosci-ence, 20, 1379–1390. doi:10.1111/j.1460-9568.2004.03570.x

Pelli, D. G. (1997). The VideoToolbox software for visual psychophysics:Transforming numbers into movies. Spatial Vision, 10, 437–442.

Sargolini, F., Fyhn, M., Hafting, T., McNaughton, B. L., Witter, M. P.,Moser, M. B., & Moser, E. I. (2006). Conjunctive representation ofposition, direction, and velocity in entorhinal cortex. Science, 312,758–762. doi:10.1126/science.1125572

Skaggs, W. E., & McNaughton, B. L. (1998). Spatial firing properties ofhippocampal CA1 populations in an environment containing two visu-ally identical regions. Journal of Neuroscience, 18, 8455–8466.

Spiers, H. J., Hayman, R. M. A., Jovalekic, A., Marozzi, E., & Jeffery, K. J.(2013). Place field repetition and purely local remapping in a multicom-partment environment. Cerebral Cortex. Advance online publication.doi:10.1093/cercor/bht198

Stevens, A., & Coupe, P. (1978). Distortions in judged spatial relations.Cognitive Psychology, 10, 422– 437. doi:10.1016/0010-0285(78)90006-3

Tanila, H. (1999). Hippocampal place cells can develop distinct represen-tations of two visually identical environments. Hippocampus, 9, 235–246. doi:10.1002/(SICI)1098-1063(1999)9:3�235::AID-HIPO4�3.0.CO;2-3

Tolman, E. C. (1948). Cognitive maps in rats and men. PsychologicalReview, 55, 189–208. doi:10.1037/h0061626

Trullier, O., & Meyer, J. (2000). Animat navigation using a cognitivegraph. Biological Cybernetics, 83, 271–285. doi:10.1007/s004220000170

Worden, R. (1992). Navigation by fragment fitting: A theory of hippocam-pal function. Hippocampus, 2, 165–187. doi:10.1002/hipo.450020208

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

530 HAN AND BECKER

Appendix

The Santa Barbara Sense of Direction Scale, Adapted Version

Gender: M F Date:

This questionnaire consists of several statements about yourspatial and navigational abilities, preferences, and experiences.After each statement, you should circle a number to indicate yourlevel of agreement with the statement. Circle “1” if you stronglyagree that the statement applies to you, “7” if you strongly dis-agree, or some number in between if your agreement is interme-diate. Circle “4” if you neither agree nor disagree. For question 17,circle “Yes” or “No” for your answer.

Q1. I am very good at giving directions.(Strongly agree) 1 2 3 4 5 6 7 (strongly disagree)

Q2. I have a poor memory for where I left things.(Strongly agree) 1 2 3 4 5 6 7 (strongly disagree)

Q3. I am very good at judging distances.(Strongly agree) 1 2 3 4 5 6 7 (strongly disagree)

Q4. My “sense of direction” is very good.(Strongly agree) 1 2 3 4 5 6 7 (strongly disagree)

Q5. I tend to think of my environment in terms of cardinaldirections (N, S, E, W).(Strongly agree) 1 2 3 4 5 6 7 (strongly disagree)

Q6. I very easily get lost in a new city.(Strongly agree) 1 2 3 4 5 6 7 (strongly disagree)

Q7. I enjoy reading maps.(Strongly agree) 1 2 3 4 5 6 7 (strongly disagree)

Q8. I have trouble understanding directions.(Strongly agree) 1 2 3 4 5 6 7 (strongly disagree)

Q9. I am very good at reading maps.(Strongly agree) 1 2 3 4 5 6 7 (strongly disagree)

Q10. I don’t remember routes very well while riding as apassenger in a car.(Strongly agree) 1 2 3 4 5 6 7 (strongly disagree)

Q11. I don’t enjoy giving directions.(Strongly agree) 1 2 3 4 5 6 7 (strongly disagree)

Q12. It’s not important to me to know where I am.(Strongly agree) 1 2 3 4 5 6 7 (strongly disagree)

Q13. I usually let someone else do the navigational planning forlong trips.(Strongly agree) 1 2 3 4 5 6 7 (strongly disagree)

Q14. I can usually remember a new route after I have traveledit only once.(Strongly agree) 1 2 3 4 5 6 7 (strongly disagree)

Q15. I don’t have a very good “mental map” of my environ-ment.(Strongly agree) 1 2 3 4 5 6 7 (strongly disagree)

Q16. I use a GPS when I travel to a new place.(Strongly agree) 1 2 3 4 5 6 7 (strongly disagree)

Q17. I play video games.a. Yesb. No

Received November 24, 2012Revision received October 3, 2013

Accepted October 4, 2013 �

Thi

sdo

cum

ent

isco

pyri

ghte

dby

the

Am

eric

anPs

ycho

logi

cal

Ass

ocia

tion

oron

eof

itsal

lied

publ

ishe

rs.

Thi

sar

ticle

isin

tend

edso

lely

for

the

pers

onal

use

ofth

ein

divi

dual

user

and

isno

tto

bedi

ssem

inat

edbr

oadl

y.

531ENCODING MULTIPLE CONNECTED ENVIRONMENTS