ChapterPDF Available

Distinguishing Sketch Map Types: A Flexible Feature-Based Classification: 11th International Conference, Spatial Cognition 2018, Tübingen, Germany, September 5-8, 2018, Proceedings

Authors:

Figures

Content may be subject to copyright.
Distinguishing Sketch Map Types: A Flexible Feature-
based Classification
Jakub Krukar, Stefan Münzer, Lucas Lörch, Vanessa Joy Anacta, Stefan Fuest
and Angela Schwering
Abstract. Sketch maps are often used as a means of assessing participants' knowledge
of spatial environments. However, the evaluation of sketch maps is challenging as they
differ in many aspects and can be scored on many possible criteria. In particular, the
classification of sketch maps into different types can be problematic, because partici-
pants rarely follow any of the identifiable formats consistently. This paper presents a
set of criteria that can be used to score a sketch map on two dimensions simultaneously:
its "route-likeness" and its "survey-likeness". The scoring is based on the presence or
absence of six features for route-conveying information and six features of conveying
survey information. In the present study, reliability estimates and factor structure of the
approach were examined with 460 sketch maps with a high variability of spatial ele-
ments included. Results show that the two dimensions are largely independent. Sketch
maps are found that score high on the route dimensions but low on the survey dimension
and vice versa, as well as sketch maps that score high (or low) on both dimensions. It
is concluded that the proposed two-dimensional scoring is useful for analysing sketch
maps, however, results will also depend on the task and instruction when assessing
participants' knowledge of spatial environments.
Keywords: sketch maps, factor analysis, route knowledge, survey knowledge
1 Introduction
Sketch maps are drawings of spatial environments, most often based on spatial memo-
ries of the drawer. Many readers might have drawn a sketch map before - for instance,
when asked to describe a route in an area to their visiting friend. The task of drawing a
sketch map is often used in spatial cognition studies as a measure of participant's spatial
knowledge about the relevant area. While sketch maps are not exactly equivalent to the
state of that knowledge, they can be interpreted to inform a number of commonly asked
research questions.
Among them, special attention has been given to interpreting sketch map types.
Since there are many ways to draw a single spatial environment, it is interesting to
observe repeatable patterns in the way the environment can be depicted. For example,
while some sketch maps remain reminiscent of classic metric maps, others contain ego-
centric views of encountered locations, or choose to abstract locations and their con-
nectivity by depicting a graph. Some sketches are rich in landmarks, while others use
them scarcely; some provide more than necessary information, while other only depict
2
the necessary minimum; some sketches schematize turns, distances, and area extents,
while others try to scale the real-world spatial relations accurately. Depending on the
theory and the investigated research question, sketch maps can be classified into infi-
nitely many combinations of types. This makes it difficult to decide about the number
and criteria of categories when analysing sketch maps in each new study.
Another common problem is the fact that a single drawing rarely follows any of the
identifiable formats consistently. Thus, classifying sketch maps into types has been typ-
ically performed in a subjective manner, by multiple raters who must agree on the con-
troversial cases. This approach is not only time-consuming, but also likely to yield very
different results in a potential replication, if only a small change occurs in the subjective
interpretation of a single minor aspect of the sketches. For this reason, the current ap-
proach proposes a set of features, which can be scored based on the presence or absence
of the given feature in each drawing. The proposed sketch map analysis is performed
by following a check list and by interpreting the resulting sum scores. The features are
related to two dimensions, and accordingly, two sum scores will result for each sketch
map. One dimension (score) represents the extent to which a sketch map conveys in-
formation that is typical for visually describing a route. The other dimension (score)
represents the extent to which a sketch map conveys spatial survey information. These
dimensions are thought to be largely independent of each other.
In the following sections, we first review existing approaches for interpreting and
classifying sketch map types in previous research. Then, we provide two check lists
that can be used to score sketch maps on two dimensions: its route-likeness and survey-
likeness (reflective of the route-related and survey-related information present in the
sketch). The check list for route-likeness has six features, and the check list for survey-
likeness has six features. We examine the approach by performing a reliability analysis
and a factor analysis. Furthermore, the paper provides suggestions for researchers wish-
ing to apply the scoring method in their analysis of sketch maps.
2 Previous Work
2.1 Existing Classification Schemes
The study of sketch maps was inspired by the exploratory analysis of "imageability" of
city elements (Lynch 1960). Appleyard (1970) presented an extensive analysis of
sketch maps drawn by the residents of the Venezuelan Ciudad Guayana. He distin-
guished eight sketch types classified along two dimensions: level of accuracy (topolog-
ical - positional) and predominant element type (sequential - spatial). In his approach,
each map has been manually classified into one of the eight categories. This method
has been long influential in the development of approaches to analyse sketch maps. For
example, drawing sequences have been recently studied within this framework (Huynh
et al. 2008).
Another classification was inspired by the developmental approach to spatial cogni-
tion. Moore (1976) suggested a three-level classification of the development of envi-
ronmental representations. In order to demonstrate the differences between them, he
described how sketch maps produced at each of these three levels would differ from
3
each other. Sketch maps produced at Level I would only consist of egocentric view-
points, reflecting a single, experience-based viewpoint on the subsets of the environ-
ment. At Level II, sketch maps would consist of clusters so that the quality of infor-
mation within the clusters would be higher than the quality between the clusters. At
Level III, sketch maps would consist of hierarchically organized clusters, related to
each other on a consistent reference system.
This scheme inspired the analysis performed by Aginsky et al. (1997) where sixteen
sketch maps were classified into one of three types: 0-D Place type, 1-D Place type,
and 2-D Place type. These correspond to Moore's Levels I, II, and III. Nineteen percent
of sketches analysed by the authors presented only egocentric views of – at most topo-
logically related – locations (0-D Place type); half of all of the observed maps success-
fully connected visited places but often only with a straight line, while the individual
locations were often enlarged and line distances were distorted (1-D Place type); 31%
of maps had a consistent global structure, including accurate segment lengths, without
disproportional enlargement of individual locations (2-D Place type). Further data anal-
ysis presented by the authors linked the 2-D Place type maps to a different learning
strategy compared to the participants who produced 0-D or 1-D Place type sketches.
This work inspired the analysis later conducted in an indoor study described by
Blajenkova et al. (2005). In this work, three types of sketch maps were named: 1-D, 2-
D, and 3-D, based on the accuracy of spatial relations, as well as the type of topological
features present in the sketches. A 1-D sketch map type was described as consisting of
some turns present along the travelled route, by not including accurate relations be-
tween individual segments, and between the two floors on which the experiment took
place. The 2-D drawings contained some relations between route segments, as well as
approximately correct shape of the route, but did not differentiate between the two
floors. The remaining drawings, named 3-D sketch maps, depicted the correct overall
shape of the route and the two floors. Maps were classified as 3-D even if some of the
turns they contained were incorrect. The authors use the classification into 1-D, 2-D,
and 3-D sketch maps in order to derive the type of mental representation of the envi-
ronment employed by the participants and use it for comparison with other wayfinding
performance and memory measures.
A similar classification scheme was used by Zhong & Kozhevnikov (2016). In their
experiment, participants were asked to sketch a route through a multi-level building.
The authors classified 62 sketch maps into three categories: procedural route maps,
allocentric-survey maps, and egocentric-survey maps. The first type consisted of maps
including sequential information but no overall consistent layout of the environment.
Those maps which included consistent global layout were classified as either allocen-
tric-survey (if they employed a birds-eye view without clearly distinguishing between
floors), or as egocentric-survey (if they included first-person views or clearly distin-
guished between the floors). The authors provide an empirical evidence for differences
in cognitive strategies employed by the participants drawing these three different sketch
map types, in the indoor context.
In the work presented above, the analysis of sketch map types has been proven useful
in answering diverse research questions. It is possible, however, that sketch maps form
a hybrid between more than one of the distinguished types. There seem to be at least
4
two reasons for this. First, in the above analyses accuracy of information is analysed
jointly with the form (type of elements) with which the information is being depicted.
Second, these classification schemes force each map into a single category. This often
suggests the superiority of the last category even when the quality of low-level route
information in some of its maps is poorer, compared to the maps belonging to simpler
categories.
2.2 Theoretical Constraints for the Two-Dimensional Scoring Approach
First, it appears essential to distinguish the accuracy of the information being depicted
on a sketch map from the type of information that is being depicted. The accuracy of
sketch maps is a separate research problem, for which some analytical methods have
been provided both for quantitative (Friedman & Kohler 2003) and for qualitative as-
pects (Schwering et al. 2014; Wang & Schwering 2015). However, it can be questioned
whether accuracy of sketch maps should at all be evaluated with respect to a metric
correspondence with the real-world configuration, because the cognitive function of
sketches is not to reflect the spatial reality, but to externalize the mental representation
of an environment for the specific goal of the drawing. This includes the cognitive pro-
cessing of spatial features and results in, among other characteristics, schematizing,
omitting, and distorting of spatial information (Tversky 2002). Sketch maps in particu-
lar are subject to many systematic changes (Tversky 1992; Wang & Schwering 2015).
For example, communicating the presence of a long, straight stretch of the route does
not necessitate drawing it to scale and with the level of the surrounding detail equivalent
to other parts of the sketch. Moreover, constructing a sketch enforces self-consistency
in the model as well as the use of Euclidian metrics between its sub-parts, neither of
which might be necessarily present in the mental representation of the environment
(Montello 1991; Kirsh 2010). Therefore, it can be misleading to evaluate sketch maps
based on their metric accuracy in relation to the real-world configuration. This issue is
particularly problematic when the sketches are subjectively evaluated by raters, since
there are multiple possible sources of errors and the source of error cannot be objec-
tively distinguished based on the final drawing. An error can result, for instance, from
an erroneous mental representation, from erroneous inferential processing during the
construction of the sketch, or from limited drawing experience and poor drawing ability
(which might be particularly problematic when the restricted size of the available paper
sheet requires planning the drawing in advance). For these reasons, analysing sketch
maps based on the presence or absence of particular features representing qualitative
aspects of conveying spatial information might be a good alternative to analysing ac-
curacy - especially when sketch map types are the focus of the analysis.
Second, there seem to be two dimensions that repeatedly form the basis for the dif-
ferentiation of sketch map types. One of them is characterized by the egocentric disjoint
experience of individual locations. The other is the survey-oriented way of depicting
distant parts of the environment on a single reference system (although important dif-
ferences exist within diverse survey depictions, as demonstrated by Zhong &
Kozhevnikov [2016]). This is in line with the well-established classification of spatial
knowledge types where landmark and route knowledge are distinguished from survey
5
knowledge (Siegel & White 1975; Montello 1998). Within this framework, integrating
information about separately experienced places is considered the most challenging as-
pect of learning an environment. Importantly, however, information relevant to route
knowledge and to survey knowledge is gathered simultaneously (Montello 1998;
Ishikawa & Montello 2006). Therefore, the aim of this work is to propose a scoring
approach for sketch maps which considers these two dimensions simultaneously and
accounts for cases in which both route and survey information co-exists within a single
sketch map. This would be difficult to achieve when each sketch map is forced into
only one pre-defined category type.
The current paper does not recommend to always substitute the analysis of sketch
maps' accuracy by the analysis of their types. Instead, it argues that the typology of
sketch maps and their accuracy are two separate variables, that should be assessed with
methods that do not intermix them. The aim of this work is to propose a method for
measuring the typology of sketch maps that is independent of their accuracy. In any
experimental dataset, it is possible that many sketch maps that are classified as a "sim-
pler" type are highly accurate, or that sketch maps that are classified as a "richer" type
contain profound and multiple accuracy errors. The choice of an appropriate analytical
method (i.e., focused on accuracy, on typology, or on the combination of the two)
should be dictated by the investigated research question.
3 Defining Route-likeness and Survey-likeness of Sketch Maps
This paper focuses on evaluating sketches on two dimensions simultaneously: their
route-likeness as well as their survey-likeness. These can be interpreted as the extent to
which indications of route knowledge and survey knowledge information are present
in the drawing, taking into account the possibility that a single map can contain ele-
ments indicative of only one or both of these dimensions. The scoring is based on the
presence or absence of specific features on a sketch map, and was inspired by the stud-
ies of elements used in route instructions given by people to other people (Denis 1997;
Schwering et al. 2013; Schwering et al. 2017; Anacta et al. 2016). The following check-
lists provide descriptions of the specific features (criterions) – six for route information,
and six for survey information – and can be applied by answering the question "Does
this element exist in the sketch map?". The scoring results from the sum of points given
for the identified criteria, with one point given for each criterion identified in the sketch
map (theoretical range 0 – 6 for both dimensions). The procedure implies that only one
point is scored if the criterion is present on the map, regardless of the number of its
instances. At this stage of analysis, it is not considered whether the depicted spatial
features and spatial relations are accurate and whether they match the reality. Simply,
a single point is given if the question concerned one specified criterion can be answered
positively. With respect to the route-likeness dimension, the scoring procedure makes
sense for sketch maps describing a path through the environment, which might depend
on the participant’s task and instruction. Multiple categories of landmarks, including
their distinction into local and global landmarks, are defined based on the definitions
presented in (Anacta et al. 2016). The way in which landmarks are depicted (Anacta et
6
al. 2017) is not important for this assessment as long as the landmark can be uniquely
identified by a potential sketch-map user unfamiliar with the area (e.g., a word label is
enough if it describes a visible property of the landmark object).
3.1 Route-likeness
Route-relevant information can be communicated by depictions of turns, landmarks,
and side streets - information that helps a potential future user of the map to follow the
visually indicated route and to make correct turning decisions. The route-likeness di-
mension is proposed to be reflected by the following six criteria.
1) Continuous route: are there no "gaps", interruptions or "holes" in the depicted
path (such that segments of the path are missing, making it impossible to de-
termine how to move from one fragment of the path to the next)? Is the route
continuous and not fragmented?
2) Turns included: does the sketch depict clear turns indicative of approximate
turning directions?
3) Side streets at decision points: does the sketch include some indication (at least
a single line or arrow) of possible choice alternatives at junctions? A rounda-
bout is treated as a regular junction.
4) Side streets outside decision points: does the sketch depict route alternatives
along the straight stretches of the route, for instance indicating the number of
junctions that need to be passed before turning?
5) Local landmarks at decision points: does the sketch depict local landmarks at
junctions?
6) Local landmarks not at decision points, but along the route: does the sketch
depict local landmarks along the route?
3.2 Survey-likeness
Survey-relevant information can be communicated by depictions of various global
landmarks, as well as hierarchical and configurational details and relations between
elements not constituting a part of the main path. The survey-likeness dimension is
proposed to be reflected by the following six criteria.
1) Global landmark - point: does the sketch include a point-like landmark located
off-route or visible from many parts of the route? Example: a city cathedral.
2) Global landmark - line: does the sketch depict a line which does not constitute
an integral path of the street network but provides structure to the sketch or a
global spatial reference for other objects? This feature can include barriers to
movement. Examples: a highway disjoint from the city streets, a river, a rail-
road.
3) Global landmark - region: does the sketch include a region, either with clearly
depicted, or vague boundaries, or with a label making it a uniquely identifiable
area? Examples: a zoo, a city centre.
7
4) Street network: are at least two streets connected outside the main path, so that
taking an alternative route or a shortcut would be possible, at least at a short
stretch of the route?
5) Containment hierarchy: does the sketch depict containment of one object in
another object, or in a region? Examples: a tower inside a zoo, a cathedral in
a marked city centre.
6) Spatial relation between distant objects: does the sketch depict an object which
has a clear spatial relation to two other objects, which would be otherwise not
directly connected to each other? Examples: A U-shaped street network with
a building in the centre, where the building is clearly located between two
otherwise opposite and disconnected streets.
Table 1 presents the summary of the scoring features. Figure 1 presents examples of
sketch maps drawn by human participants varying on the two dimensions.
Table 1. Route-likeness and Survey-likeness scoring features.
dimension
criterion
route-likeness
r1 - Continuous route
r2 - Turns included
r3 - Side streets at decision points
r4 - Side streets outside decision points
r5 - Local landmarks at decision points
r6 - Local landmarks outside decision points
survey-likeness
s1 - Global landmark - point
s2 - Global landmark - line
s3 - Global landmark - region
s4 - Street network
s5 - Containment hierarchy
s6 - Spatial relation between distant objects
Fig. 1. Sample sketch maps scoring: a) low on route-likeness and high on survey-likeness; b)
high on both dimensions; c) low on both dimensions; d) high on route-likeness and low on sur-
vey-likeness.
9
4 Examination of the Two-Dimensional Scoring Approach
The main goal of the analysis was to examine whether the assessment of spatial features
in sketch maps on the proposed twelve criteria are reflected by two proposed dimen-
sions (factors): route-likeness and survey-likeness. Therefore, we applied an explora-
tory factor analysis.
4.1 Data
The data included binary (yes/no) scores for all twelve criteria, manually coded for 460
sketch maps. The sketch maps were sourced from three studies not reported in this pa-
per in detail. They depicted one American, two German, and two fictional cities either
already known to or learned by the participants. The instructions of all three studies
asked participants to draw a map for a friend visiting the city, including a path between
specified locations. Being sourced from multiple experimental conditions and from par-
ticipants with large individual differences, the sketches are diverse: they employ differ-
ent styles to depict the elements of the urban layout. Table 2 presents raw count data of
the criteria within the analysed 460 sketch maps.
Table 2. Number of maps (out of 460) containing each criterion.
route-likeness
survey-likeness
r1
r2
r3
r4
r5
r6
s1
s2
s3
s4
s5
s6
352
366
171
215
253
235
129
210
213
172
255
253
4.2 Inter-rater reliability
Three raters (one of the authors and two student assistants) independently assessed a
subset of the sketch maps. Sixty-one sketch maps were randomly chosen from the 460
sketch maps. Each rater scored the same chosen 61 sketch maps based on a document
detailing the scoring procedure that included descriptions and examples of criteria
presented above (Table 1; the full document is available at: http://osf.io/3d97m). Inter-
rater reliability was assessed using a two-way random, agreement-based, average-
measures intra-class correlation (Hallgren 2012), calculated separately for route-
likeness scores and survey-likeness scores using the irr R package (Gamer et al. 2012).
Intra-class correlations (ICCs) were in the "good" range, ICC = 0.68 for the route-
likeness dimension and ICC = 0.61 for the survey-likeness dimension.
4.3 Results
We employed exploratory factor analysis to analyse the sketch map classification. A
scree plot inspection of Eigenvalues indicated that the data coming from twelve criteria
might be described best using a three-factor solution. Further inspection revealed that
two out of three suggested factors were sub-categories of the survey-likeness scale
10
(i.e., criteria r1-r6, s1-s3, and s4-s6 loaded separate factors). On theory-motivated
grounds, we performed factor analysis restricted to two factors, examining whether
each criterion assumed to describe the survey-likeness factor indeed correlates the
strongest with that factor, and not with the route-likeness factor. The analysis was per-
formed in the psych R package (Revelle 2017), using the weighted least squares solu-
tion, the oblimin rotation, and the tetrachoric correlation method (suitable for binary
data, such as the yes/no responses). This analytical method does not pre-impose any
structure on the data and is thus suitable for examining whether the correlations found
in the dataset are similar to those assumed on the theoretical grounds by the researchers.
Raw correlations are presented in Table 3. Results of the factor analysis are presented
in Table 4 and Figure 2. Guttman's Lambda 6 was used to calculate reliability, as it is
appropriate for binary data and for the joint assessment of multiple scales. Interpretation
is similar to Cronbach's alpha with its values ranging from 0 to 1; the score of .70 is
considered satisfactory for preliminary research, and scores above .90 are expected for
decision-making tools (Nunnally 1978). Guttman's Lambda 6 was 0.82 for the route-
likeness dimension and 0.73 for the survey-likeness dimension.
Table 3. Tetrachoric correlations between individual items.
r1
r2
r3
r4
r5
r6
s1
s2
s3
s4
s5
s6
r1
1.00
r2
0.93
1.00
r3
0.52
0.58
1.00
r4
0.68
0.75
0.69
1.00
r5
0.39
0.51
0.43
0.50
1.00
r6
0.50
0.64
0.51
0.61
0.69
1.00
s1
0.03
-0.03
0.07
0.23
0.18
0.20
1.00
s2
-0.05
-0.06
0.17
0.24
0.31
0.15
0.37
1.00
s3
0.02
-0.04
0.24
0.39
0.32
0.03
0.56
0.53
1.00
s4
0.36
0.30
0.47
0.55
0.14
-0.05
0.18
0.38
0.44
1.00
s5
0.21
0.21
0.51
0.53
0.15
0.29
0.46
0.31
0.56
0.65
1.00
s6
-0.20
-0.31
0.17
0.23
-0.02
-0.29
0.21
0.48
0.52
0.72
0.40
1.00
Table 4. Standardized factor loadings. The correlation between the two factors equals 0.14.
criterion
factor 1
factor 2
r1
0.872
-0.097
r2
0.964
-0.170
r3
0.689
0.289
r4
0.786
0.368
r5
0.649
0.133
r6
0.819
-0.088
s1
0.022
0.564
s2
-0.024
0.685
s3
0.003
0.822
s4
0.214
0.737
s5
0.274
0.704
s6
-0.305
0.851
Fig. 2. The twelve analyzed criteria and their factor loadings.
4.4 Discussion
The analysis confirms that twelve previously presented criteria can be seen as repre-
sentative of two factors, in a structure as suggested in Table 1. It bears noting, that the
purpose of this analysis was not to find the factorial structure reducing the data best to
12
the fewest components, but rather to verify the assumption that twelve criteria can be
used as indicators of two theoretically-motivated factors.
The correlation between both factors was small (0.14) and there was some moderate
correlation between individual criteria belonging to two separate factors (Table 3,
e.g., criteria r4 and s4). This is not surprising, given the fact that the characteristics of
sketch maps tends to correlate with individual abilities, the knowledge of the environ-
ment, and the experimental task at hand. It is therefore likely that highly-skilled indi-
viduals draw information-richer maps in general ("richer" meaning sketches with more
points on the presented classification), that higher familiarity contributes to richer maps
on both dimensions, and that easier tasks result in richer maps on both dimensions.
Overall, however, the correlation of 0.14 between the two factors is small. It is appar-
ently meaningful to analyse the two dimensions – route-likeness and survey-likeness
separately, despite the fact that they are not completely orthogonal. The advantage of
scoring sketch maps on two largely independent dimensions lies in the opportunity for
making finer distinctions between sketch maps. It is possible, for instance, to consider
maps that are high on one dimension, but low on the other, high on both, or low on both
dimensions; and this can be done without creating new categories or manually re-clas-
sifying maps.
The two-dimensional scoring also contributes to the solution of the problem of hy-
brid sketches. In our dataset of 460 sketches, 327 scored at least one point on both
dimension. Thus, most of the sketch maps are hybrids: they contain information rele-
vant to both the information concerning a specified route and to the allocentric over-
view of the broader environment. The presented two-dimensional scoring approach
does not involve forcing each hybrid map into a single category.
Nevertheless, more discrete categorizations can be derived, if needed. For instance,
it is possible to calculate the mean route-likeness and mean survey-likeness scores and
classify sketch maps into four relative categories, depending on their score below or
above the two means (Figure 1 contains some examples). Performing this operation on
our 460 sketches resulted in 116 maps falling into the "low - low" category, 84 sketches
in the "low route - high survey" category, 95 sketches in the "high route - low survey"
category, and 165 in the "high - high" category. The spread of these numbers highlights
the fact that the two-dimensional scoring approach can capture the diversity of sketch
maps and does not result in a disproportionally large number of "poor" maps.
The implication for broader spatial cognition studies lies in the fact that the presented
classification method appreciates the diversity in drawing strategies among individuals.
For instance, it could be intuitively expected that higher amount of survey information
is always associated with high amount of route detail. However, even poorly perform-
ing participants differ in their drawing strategies and it is possible that some participants
aim to convey large amount of configurational information even when they do not have
detailed knowledge of the route they are required to describe. This, as well as other
nuanced relations, would be difficult to capture using a pre-defined set of sketch map
categories, unless such a category is explicitly expected to occur among the sketches.
13
5 Limitations and Conclusion
The paper presented an approach that can be used for scoring sketch map types. Twelve
criteria have been shown to load two separate factors, which correspond to the theoret-
ically-supported concepts of route-likeness and survey-likeness. The scoring approach
does not require forcing hybrid maps into a single category and makes it possible to
distinguish higher and lower scores within each dimension.
The inter-rater agreement was not perfect in a situation when three raters scored the
sketches based on a single document with descriptions and examples. An active com-
munication between the raters might be necessary to ensure a shared understanding of
all criteria. Since the sketches evaluated in our study came from three separate experi-
ments, their diversity (and dissimilarity to the prototypical examples) might have been
larger than it is typically the case for sketches derived from a single study.
It is important to note that the scores resulting from using the checklist are on the
ordinal, and not on the continuous ratio scale. Limitations similar to likert-scale
measures apply. Moreover, the distance of "1 point" should not always be interpreted
in the equivalent way. If to consider three maps scoring 2, 3, and 3 points on the survey-
likeness scale, the latter two maps are not necessarily "more survey-like" from the first
map to the same extent, as their points might derive from different criteria. Likewise,
two maps should not be considered disposing completely identical characteristics when
they score the same number of points, as the points might be derived from distinct cri-
teria. Researchers should consider relevant statistical tools for further analyses, depend-
ing on the case-specific application of the approach.
It also bears noting that the presented two-dimensional scoring approach does not
consider the accuracy of information contained in sketch maps, but it offers the possi-
bility of including this aspect in the analysis. This could be done, for instance, by scor-
ing the maps twice: once for the presence/absence of the criteria listed in Table 1, and
the second time for the presence of correct vs erroneous instances of each criterion.
A researcher investigating the accuracy of sketch maps would then be interested in the
relation between these two values. Yet another alternative is to impose a threshold value
of instances of each criterion that needs to be reached before a point is awarded on the
checklist. This could either be a generic number (e.g. a point is awarded only if more
than two landmarks at decision points are present on the sketch map), or it could be
linked to a particular location of interest (e.g. a point is awarded only if a landmark at
decision point other than the cathedral is present on the sketch map). The scoring can
be also filtered by the informational value of the elements included in the sketch maps:
for example, landmarks that are not visible from the route can be ignored in the scoring
process. The presented scoring approach is flexible enough to support multiple appli-
cation scenarios, without the need for deriving new category sets for each experimental
dataset.
6 References
Aginsky, V. et al., 1997. Two Strategies for Learning a Route in a Driving Simulator.
14
Journal of Environmental Psychology, 17(4), pp.317331. Available at:
http://linkinghub.elsevier.com/retrieve/pii/S0272494497900700.
Anacta, V.J.A. et al., 2017. Investigating Representations of Places with Unclear Spatial
Extent in Sketch Maps. In A. Bregt et al., eds. Societal Geo-innovation. AGILE
2017. Lecture Notes in Geoinformation and Cartography. Cham: Springer, pp. 3
17. Available at: http://link.springer.com/10.1007/978-3-319-56759-4_1.
Anacta, V.J.A. et al., 2016. Orientation information in wayfinding instructions:
evidences from human verbal and visual instructions. GeoJournal, pp.1–17.
Appleyard, D., 1970. Styles and methods of structuring a city. Environment and
Behavior, 2(1), pp.100117.
Blajenkova, O., Motes, M.A. & Kozhevnikov, M., 2005. Individual differences in the
representations of novel environments. Journal of Environmental Psychology,
25(1), pp.97109.
Denis, M., 1997. The Description of Routes: A Cognitive Approach to the Production
of Spatial Discourse. Cahiers Psychologie Cognitive, 16(4), pp.409458. Available
at: citeulike-article-id:7753896.
Friedman, A. & Kohler, B., 2003. Bidimensional Regression: Assessing the Configural
Similarity and Accuracy of Cognitive Maps and Other Two-Dimensional Data
Sets. Psychological Methods, 8(4), p.468.
Gamer, M., Lemon, J. & Singh, I.F.P., 2012. irr: Various Coefficients of Interrater
Reliability and Agreement. Available at: https://cran.r-project.org/package=irr.
Hallgren, K.A., 2012. Computing Inter-Rater Reliability for Observational Data: An
Overview and Tutorial. Tutorials in Quantitative Methods for Psychology, 8(1),
pp.2334. Available at: http://www.tqmp.org/RegularArticles/vol08-1/p023.
Huynh, N.T. et al., 2008. Interpreting urban space through cognitive map sketching
and sequence analysis. The Canadian Geographer/Le G{é}ographe canadien,
52(2), pp.222240.
Ishikawa, T. & Montello, D.R., 2006. Spatial knowledge acquisition from direct
experience in the environment: Individual differences in the development of
metric knowledge and the integration of separately learned places. Cognitive
psychology, 52(2), pp.93129.
Kirsh, D., 2010. Thinking with external representations. AI and Society, 25(4), pp.441
454.
Lynch, K., 1960. The Image of the City, London: MIT Press.
Montello, D.R., 1998. A new framework for understanding the acquisition of spatial
knowledge in large-scale environments. In M. J. Egenhofer, ed. Spatial and
Temporal Reasoning in Geographic Information Systems. New York: Oxford
University Press, pp. 143154.
Montello, D.R., 1991. The measurement of cognitive distance: Methods and construct
validity. Journal of Environmental Psychology, 11(2), pp.101122.
Moore, G.T., 1976. Theory and research on the development of environmental
knowing. In G. T. Moore & R. G. Golledge, eds. Environmental Knowing: Theories,
Research, and Methods. Stroudsburg, PA: Dowden, Hutchinson & Ross, pp. 138
164.
15
Nunnally, J.C., 1978. Psychometric theory 2nd Ed., New York: McGraw-Hill.
Revelle, W., 2017. psych: Procedures for Psychological, Psychometric, and Personality
Research. Available at: https://cran.r-project.org/package=psych.
Schwering, A. et al., 2014. SketchMapia: Qualitative Representations for the
Alignment of Sketch and Metric Maps. Spatial Cognition and Computation,
14(3), pp.220254.
Schwering, A. et al., 2017. Wayfinding Through Orientation. Spatial Cognition &
Computation, pp.131. Available at:
https://www.tandfonline.com/doi/full/10.1080/13875868.2017.1322597.
Schwering, A., Li, R. & Anacta, V.J.A., 2013. Orientation information in different forms
of route instructions. In Short Paper Proceedings of the 16th AGILE Conference
on Geographic Information Science, Leuven, Belgium.
Siegel, A.W. & White, S.H., 1975. The development of spatial representations of large-
scale environments. Advances in child development and behavior, 10, pp.955.
Tversky, B., 1992. Distortions in cognitive maps. Geoforum, 23(2), pp.131138.
Tversky, B., 2002. What do Sketches say about Thinking. In Proceedings of AAAI spring
symposium on sketch understanding. pp. 205210. Available at:
http://www.aaai.org/Papers/Symposia/Spring/2002/SS-02-08/SS02-08-
022.pdf.
Wang, J. & Schwering, A., 2015. Invariant spatial information in sketch maps a study
of survey sketch maps of urban areas. Journal of Spatial Information Science,
11(11), pp.3152. Available at:
http://josis.org/index.php/josis/article/view/225.
Zhong, J.Y. & Kozhevnikov, M., 2016. Relating allocentric and egocentric survey-based
representations to the self-reported use of a navigation strategy of egocentric
spatial updating. Journal of Environmental Psychology, 46, pp.154175.
Available at: http://linkinghub.elsevier.com/retrieve/pii/S0272494416300263.
... These tasks provide insight into the adaptability of one's spatial skills but do not necessarily give an accurate reflection of their mental representation of their lived environment. For many decades, self-generated maps have offered insight into mental representations of the known environment by providing opportunities to explore what elements of the environment and relations are most salient (Krukar et al. 2018;Matthews, 1981;Tversky, 2003). These factors are crucial in navigating the real world but are seldom explored in tasks focussed on spatial skills. ...
... In their seminal work, Siegel and White (1975) proposed a framework of map development that began with the use of landmarks, which over time and experience became connected into a route and ultimately resulted in a survey representation of an environment that incorporated connections between objects and paths beyond the immediate route. Subsequent studies of children's sketch maps have shown that they often contain elements of multiple categories (Harwood & Usher, 1999;Krukar et al. 2018;Matthews, 1984a;Thommen et al. 2010) but the distinguishing features of a route representation is that landmark and road inclusions are arranged around a central path and decision points along the way, with a focus on the start and end points. By contrast, survey representations include a more global network of landmarks within an environment. ...
... These studies used different metrics for comparing maps, which were fundamentally based on precision and task performance, not understanding of children's spatial representations. More recently, studies have shifted to the digital medium, focussing on children's ability to recall and navigate controlled, virtual environments (Nazareth et al. 2018) or dimensionality of map characteristics for fictional locations (Krukar et al. 2018). However, sketch mapping studies of children's local environments have shown that even young children are able to represent their known environment with a remarkable degree of accuracy. ...
Article
Large-scale spatial representations are rarely a perfect replication of the corresponding environment. Factors such as age, experience and environmental association play a significant role in each person’s cognitive map. Yet, traditional spatial tasks remove these contextual elements in determining a person’s spatial orientation skill, which has contributed to the apparent widening gap in spatial skills for children in non-urban areas. In this study we examined children’s sketch maps of their route between home and school from four geographically diverse sites (urban, regional, rural, and remote Indigenous communities). Our findings suggest children from rural and remote communities possess highly detailed and sophisticated mental representations of their local environment, outside of traditional developmental frameworks. This work provides insights into these representations through student work samples. We suggest future work consider the contextual nature of spatial skills when comparing students.
... In cognitive science, several ways are used to measure spatial learning. Some examples are the accuracy of pointing directions to previously experienced landmarks (Credé et al., 2019), distance estimations (Richardson et al., 1999), and metal map depiction (drawing a sketch from memory) Krukar et al., 2018). The current paper uses landmark recall and landmark placement tests, which are also standard methods to test spatial learning (Credé & Fabrikant, 2018;Evans et al., 1984;Frankenstein, 2015;Huang et al., 2012;Münzer et al., 2006), and is suitable to address research hypotheses (see Section 3). ...
Full-text available
Article
Today’s navigation systems use topographic maps to communicate route information. Being general-purpose maps, topographic maps lack optimal support for the specific task of route reading and navigation. In the public transportation domain, research demonstrated that topographic maps do not support planning of routes as good as schematic maps. Our current paper applies this idea to the domain of in-car navigation. Schematic maps emphasize functional aspects of geography and direction information by highlighting information relevant to navigation actions and orientation. However, there is a lack of systematic studies researching the usability of schematic cartography in wayfinding tasks. This article evaluates schematic route maps, created with an algorithm developed in our previous work, regarding user interaction, navigation performance, and spatial memorability. We compare these schematic maps with correspondent non-schematic ones in two different tasks: prospective and situated (driving simulator) route reading. The schematic map and the corresponding non-schematic map are identical in terms of their elements and topology; they vary only in their geometric shape: on the schematic maps, features are highly generalized, following schematic simplification rules for clarity. The experimental data shows that participants using the schematic route maps require fewer map interactions to complete the tasks, orientation information is more visible and leads to more accurate spatial knowledge acquisition. This result contributes to a better understanding of schematic route visualizations’ benefits to support users in wayfinding and orientation tasks.
Full-text available
Article
Dominant approaches in computer-assisted wayfinding support adhere to the deeply problematic principles of turn-by-turn navigation. In this article, we suggest a new approach called “Wayfinding Through Orientation,” which supports the acquisition of spatial knowledge and cognitive mapping for advancing the user’s spatial orientation. Being oriented on one’s way is a prerequisite to enabling people to verify instructions and to incorporate new spatial information into their existing knowledge structure. In three studies described in this article we first present empirical evidence that people can be supported in survey knowledge acquisition through suitable wayfinding instructions. Consequently, we explore orientation information in human wayfinding instructions. Finally, we outline how orientation information can be communicated within a prototypically implemented navigation assistance system.
Full-text available
Article
Although the importance of landmarks for human navigation and for orientation is well accepted, most of today’s navigation systems hardly incorporate any landmark information or information supporting orientation. Different from previous research that only addressed turn-by-turn instructions in verbal forms and landmarks at decision points, the present study provides empirical evidence that human-generated wayfinding instructions are not solely turn-by-turn but the majority of instructions provide orientation information. We propose a new classification scheme for identifying information in wayfinding instructions that will not only support orientation but also facilitate construction of mental map. We explored and compared two forms of representations—visual sketch map and verbal instructions. Results revealed that landmark information is important in human wayfinding instructions with particular importance of local landmarks along the route and global landmarks that support orientation. In addition, sketch maps contained more global landmarks than verbal instructions. In contrast to turn-by-turn navigation, we found that many instructions in human route descriptions do not always refer to turning actions but to orientation. We conducted interviews with a set of raters on what they consider helpful information in both forms of representations. These results confirmed our findings and supported that future route descriptions can be more meaningful and helpful when they are enriched with orientation information as they conform to how humans structure wayfinding instructions.
Full-text available
Article
It is commonly recognized that free-hand sketch maps are influenced by cognitive impacts and therefore sketch maps are incomplete, distorted, and schematized. This makes it difficult to achieve a one-to-one alignment between a sketch map and its corresponding geo-referenced metric map. Nevertheless, sketch maps are still useful to communicate spatial knowledge, indicating that sketch maps contain certain spatial information that is robust to cognitive impacts. In existing studies, sketch maps are used frequently to measure cognitive maps. However, little work has been done on invariant spatial information in sketch maps, which is the information of spatial configurations representing correctly the real world. We aim to study such information from a cognitive perspective. This paper first presents basic spatial objects identified in sketch maps and then introduces sketch aspects that capture invariant spatial information. The accuracy and reliability of these aspects were evaluated by a human study. We collected sketch maps from participants, extracted and measured spatial relations of identified spatial objects, and in the end analyzed the accuracy and statistical significance of these relations. Based on the statistical survey, we propose in this paper a set of seven sketch aspects that constitute invariant spatial information, along with a spatial analysis method to measure them. The findings of these aspects help to understand which spatial information is preserved under the transformation from the physical world to human sketch maps.
Full-text available
Article
More and more private citizens collect and publish environmental data via web-based geographic information systems. These systems face two challenges: the user interface must be intuitive and the processing of geographic information must account for cognitive impacts. We propose to use sketch maps as the medium for interaction, because they reflect a person's spatial knowledge. Information from sketch maps is distorted, schematized, incomplete, and generalized while metric maps are not. This paper employs qualitative representations for the alignment of sketch and metric maps. We suggest a set of cognitively oriented aspects in sketch maps stably computed by people and evaluate qualitative representations to formalize these aspects. This allows us to align and integrate geographic information from sketch maps.
Conference Paper
This study analyzes different ways of representing vaguely defined places from a set of sketch maps specifically when used in giving route instructions. A total of 30 participants who are familiar with the study area were asked to sketch a route map consisting of pre-identified set of places. The task involved two groups: intra-city route and inter-city route. Sketch maps were analyzed using a previously developed classification scheme to investigate how places with unclear spatial extent are represented. These were then classified into different category of places: district, site and neighborhood. Results showed that labels and regular shapes are the most preferred, as opposed to other types of sketch representations, regardless of the category of place. It also occurred that a specific place can be classified under one or more categories, which influences the type of sketch representation used.