Comprehension of Layout Complexity: Effects of
Architectural Expertise and Mode of Presentation
Christoph Hölscher
University of Freiburg, Germany
Ruth Conroy Dalton
University College London, United Kingdom
This paper presents an experiment on judgments of design complexity, based on
two modes of stimuli: the layouts of corridor systems in buildings shown in plan
view and movies of simulated walkthroughs. Randomly selected stimuli were presented to 166 subjects: ‘experts’ (architects or students currently enrolled on an
architectural course) and ‘lay people’ (all others). The aims were to investigate
whether there were differences between these two groups in terms of their judgments of building complexity, effects of modality of stimuli and if any environmental measures (geometric or complexity-based) correlated with the assessments.
The results indicate that indeed complexity and wayfinding ratings show distinct
patterns. Architects are more sensitive to differences between complexity and
wayfinding ratings in the plan views, while lay-people provided more distinctive
ratings for movies. Similarly, lay-people judged the same materials to be simpler
and easier when seen as ego-centric movies, with architects showing the opposite
pattern. The judgments of both groups correlated highly with a number of environmental measures, with architects providing greater differentiation regarding
layout symmetry.
Aims
Three separate and distinct aims underpin the experiment presented in this
paper. First is the investigation of the differences in how architects and
non-architects view building-layout designs with respect to perceptions of
complexity and judged ease of wayfinding; Montello recently suggested
J.S. Gero and A.K. Goel (eds.), Design Computing and Cognition ’08,
© Springer Science + Business Media B.V. 2008
159
160
C. Hölscher, R.C. Dalton
(2007) that “architects will generally see greater differentiation in the built
environment than non–architects”.
The second aim of the experiment is to determine whether the mode of
presentation of the design influences such judgments; two modes of stimuli
were used in this experiment, an abstracted plan view and a ‘walkthrough’
movie (with ego-centric perspective, i.e., the eye-level view of a person
moving through the layout). Navigating through buildings is a common
regular activity for most people and they are familiar with experiencing
such environments from an ego-centric movement perspective. Consequently, assessing complexity or ease-of-orientation could be more intuitive for lay-people than providing such judgments from plan views. In fact,
architects are trained to rapidly switch between different perspectives and
draw inferences from plan views of buildings as part of their formal education. It was initially hypothesized that differences should be found in the
lay-people’s judgments of complex environments, depending upon the
mode of stimulus (the assumption being that lay-people would find the
movie-mode easier to comprehend than the plan-mode, whereas architects
should find both equally as easy).
The final aim is to determine whether the subject’s judgements of design complexity correlate with a set of objective, environmental measures.
Detecting the variables designers and non-designers – consciously or unconsciously – take into account for their judgments and especially their
differences in doing so would provide a more detailed understanding of the
cognitive processes involved in expert design performance.
Significance
The process of design frequently involves the continuous assessment of
many aspects of the developing design solution (Cross 2006). This paper
focuses on two particular types of design-criterion that may play a role in
the process of architectural design, that of ‘design complexity’ and the allied judgment of ‘ease of wayfinding’. These judgements are of importance, not only to the architect engaged in the process of design, but equally to the end-user of any building that is ultimately realized as the product
of such a process. In fact, assessing the ease of wayfinding in a building
requires the designer to anticipate the cognitive limitations of building users (lay-people) (Brösamle & Hölscher 2007), and if designers and
building users differ widely in their assessment criteria such a difference
might be a source of mismatch in communication between these groups
(Bromme, Rambow & Nückles 2001).
Comprehension of Layout Complexity
161
Previous work on judgements of complexity have tended to fall predominantly into one of two groups: those primarily concerned with subjective assessments of design and those focussed on computational measures
of complexity, an area strongly connected to the field of computational
aesthetics (Sha and Bod 1993). This paper attempts to consider both the
subjective assessments of complexity as well as objective, computational
measures and to determine the relationship between them.
Measures of Design Complexity
As Montello comments, “…defining psychological complexity is difficult
because humans organize information into meaningful units in a way that
reduces complexity in an information-theory sense to relative simplicity in
a psychological sense.” (Montello 2007). This difficulty has not deterred
investigation; with respect to architectural design, the key paper on subjective judgments of building layout complexity is Weisman’s 1981 paper,
upon which this study is based (and which will be discussed further in the
Method Section). O’Neill (1991) also employed human judges to assess
floor plans, but he only reports on a subset of materials and the results are
mostly similar to Weisman’s. Finally, Butler et al. (Butler, Acquino, Hissong and Scott 1993) use a number of different environment-variables, including some complexity-measures, to examine the route-preference for
paths through a notoriously hard-to-navigate building: certain measures
correlated highly with expressed preference, however, none of these were
the measures of complexity.
Papers on methods of assessing complexity of abstract visual stimuli
include those by Leeuwenberg (1968) and Attneave (1957) in the area of experimental psychology. The ability to compute the complexity of a building can be of great help in evaluating design decisions, in comparing and
classifying plans as well as being particularly useful as a fitness function
for generative design (Jackson 2001). More recently, information theoretic
measures have been applied to building designs such as the interesting paper by Gero and Jupp, (2002), in which building plans are assessed using
an entropy-based measure applied to string-encodings of building layouts.
Many of the calculations of design complexity focus on either linear
elements (corridor segments, routes, paths, lines of sight etc.) or onedimensional elements such as junctions, turns or decisions points.
162
C. Hölscher, R.C. Dalton
Gestalt Forms
In a paper by Passini (1996), he suggests that two principles of building
design exert most influence on wayfinding performance. The first are configurational measures: the network, spatial hierarchies and symmetries.
Passini’s second principle concerns the contribution to environment comprehension made by gestalt forms (Köhler 1929; Koffka 1935). Passini
suggests that it is common for building layouts to be designed around gestalt, or ‘good form’, principles: the square, the cross, the T and L-shape
etc. He suggests that these are the most easily apprehendable of layouts
and that once the ordering principle has been grasped, it can be of significant assistance in understanding the complexity of the layout and then employed in navigation. This view is reinforced by Montello who suggests
that, “… the overall shape or “gestalt” of a path layout can determine
whether a particular element is disorienting… Layouts may be said to vary
in their closeness to a ‘good form’; comprehending a layout is easier when
the layout has an overall pattern that can be apprehended as a single simple
shape, perhaps allowing easy verbal labelling. A square has better form
than a rhombus; a circle has better form than a lopsided oval. People tend
to understand and remember layouts as having good form.” O’Neill (1991)
also reports that his subjects appeared to respond to ‘good form’ when rating architectural plans for their complexity. A number of the stimuli used
in this study can be considered to have ‘good form’ (see Figure 1) and the
effect of these special cases is discussed in Section Individual Layouts with
High Prägnanz.
Method
The study by Weisman (1981) provided the first systematic assessment of
floor plan complexity by human judges. He used thirty simplified building
layouts that spanned a wide variety of building styles. We opted to use his
original materials as the starting point for this study to allow for later comparison with the additional results of both Weisman (1981) and O’Neill
(1991) about wayfinding performance in some of these settings.
Comprehension of Layout Complexity
Fig. 1. Layout-stimuli as abstracted corridor systems in plan mode
163
164
C. Hölscher, R.C. Dalton
Stimuli Selection
Thirty simplified building layouts from Weisman (1981) were used, spanning a range of complexity features. These stimuli, both in plan and movie
mode, are reduced to corridors, with no indication of the buildingenvelope, rooms or other spatial subdivisions. The corridor-layouts were
then assigned to a number of classes or ‘bins’ from which stimuli could be
selected randomly. The layouts were grouped into the bins by attributes of
their environmental features. This was intended to later permit the comparison several spatial factors directly with modality and professional
background using an Analysis of Variance (ANOVA) test and to identify
any interactions between spatial factors and non-spatial factors (see Section Results). Having established 16 bins based on the number of axial
lines, the number of spatial symmetries and O’Neil’s ICD measure (1991;
see Section Measures of Environmental Properties), it became evident that
two additional building layouts were required to balance the number of
layouts in each ‘bin’. These were added to the sample, ensuring that each
bin contained between 1 and 3 layouts. The full set of layout-stimuli can be
seen in Figure 1; the two additional plans are located on the bottom left.
Method of Route Path Construction for Movies
One crucial difference between a plan view of a building and an egocentric movie is that the plan view permits the viewer to make a global
judgement of the whole setting irrespective of the specific route through
the building. An ego-centric movie necessarily follows a particular route
through the building layout (see Figure 2 for examples of each mode). The
complexity-assessment is thus based on a sequence of local impressions.
The choice of route through the building will almost certainly have an impact on how the participant assesses the building. This difference between
the modalities means that the task for the judge is quite a different one for
the plan views versus the ego-centric movies.
The construction of each walkthrough movie required the selection of
navigational paths for each building layout. The paths aimed to traverse the
maximally possible distance (i.e. passing directly through as much of the
building as possible), but without navigating any corridor segment twice
(i.e. eliminating instances of backtracking), whilst containing the minimal
number of turns. In order to ensure that all of the building was viewed, a
number of stationary ‘head turns’, also kept to a minimum, were included
at strategic junctions. In addition to these head turns, each movie commenced with a stationary 360° revolution.
Comprehension of Layout Complexity
165
Fig. 2. Layout-stimuli as Ego-centric Movie (left) & Abstracted Plan (right)
Complexity & Wayfinding Judging Task
The complexity- and ease-of-wayfinding-judging task was administered in
the form of an online questionnaire, which took approximately 20-30 minutes to complete. Each participant is presented with 16 layouts, one from
each bin. Each layout is presented both as a movie and as a plan view.
Presentation format and order is balanced and randomized into six blocks
of 5-6 stimuli, with each block containing either movies or plans. This
complex procedure ensures maximum balancing of elements, full randomisation and a minimum distance of 10 elements between the repetition of
members of pairs (same element/different modality). Modality is realized
as a within-subject factor. Prior to presentation of the first stimulus, participants view a page containing a set of five sample layouts, derived from
the study materials and illustrating the range of layout variation in the
study: this familiarizes the participants with the expected materials and
minimizes the danger of any temporal drift in the ratings of participants
over the course of the experiment.
In the main part of the study each stimulus was presented on a single
page, with two questions accompanying each layout. Subjects were instructed to view each plan or movie and were asked to make two judgments: first, of the complexity of the layout (ranging from ‘simple’ to
‘complex’) and second, of the projected ease or difficulty of finding one’s
way around a building with such a plan configuration (ranked between
‘easy’ and ‘difficult’). Both ratings used 9-point, bulleted scales. The rating values were used in their original state and were also submitted to a
process of z-standardization (separately for each participant and rating
166
C. Hölscher, R.C. Dalton
scale, but across all layouts and presentation modalities) in order to improve comparability of ratings between participants for the analysis of layout features. Without z-standardization participants with a low range of
ratings would likely be under-represented in the correlations (and those
with high judgments become overly influential).
In total 166 subjects successfully completed all parts of the questionnaire and were included in the following analyses. Of these 52 were architects or had an architectural education and 114 could be considered ‘nonexperts’ or laypersons.
Measures of Environmental Properties
Prior to the study’s inception, we identified a number of measures that could
be used as objective evaluations of the stimuli. These were introduced to
determine if any objective factors of the built environment correlated with
people’s subjective judgements of complexity as well as ‘ease-of-wayfinding’
and could thus be used predictively. Many of these were straightforward
geometric measures such as a layout’s area, perimeter or its number of
walls and polygon vertices. Other measures were included due to evidence
in the wayfinding literature that they may play a role in how easily people
navigate (O’Neill, 1991, Passini, 1996). Measures such as the number of
corridor segments and decision points were included for this reason. In
particular, the number of symmetries was included as it was hypothesized
that these might have a strong influence on the complexity-task: the number
of lines of symmetry, rotational symmetries and their sum were evaluated.
A pair of measures are included which arose from the literature on space
syntax (Hillier & Hanson 1984). These are the number of axial lines and
convex spaces in the layout. A convex space is a discrete spatial unit whose
bounding polygon contains no reflex-angled corners and therefore all
points within such a space are visible from all others. A building’s decomposition into a set of discrete convex spaces is based upon a judgment of
the fewest and ‘fattest’ spaces required to fully encompass all occupiable
space. An axial line is an unimpeded line of sight and the equivalent axial
description of a building consists of the fewest and longest lines of sight
that pass through every convex space and complete all circulation rings.
Then, there are a number of miscellaneous measures that do not fit into
any ‘family’ or class of measures: one of these measures is the number of
‘topological loops’ in a layout, i.e. a cross would contain zero ‘loops’, a
square one ‘loop’ and a figure-eight two ‘loops’.
Comprehension of Layout Complexity
167
Convexity is a measure developed by Batty (2001) and was originally
intended for use in studying the convexity of a polygon representing a potential field of view or ‘isovist’. As Batty describes, “ΨI [or convexity],
originates from area and perimeter, and is defined as the ratio of the radius
of an idealized circle associated with the actual area of the [polygon] to the
radius of an idealized perimeter from the actual perimeter in question...
[Ψi] varies from a value of 0 for a straight line [polygon]… to 1 for a circle… This measure falls within [the range 0 to 1 and] appears to covary
with the convexity of the space.” Since it is derived from area to perimeter
and is clearly a measure of the ratio between the two, convexity correlates
well with both measures. However, it appears to outperform both in terms
of correlating with the judgment tasks (see Section Environmental Variables and Complexity). The last measure to be included is O’Neill’s measure of ‘ICD’ (1981) or interconnection density. For each location in which
a change of direction must be made (a node) the number of directional
choices is noted: this is 1 for a dead-end (the only choice is to turn around
and go back), 2 for a ‘corridor-turn’ and 3 for a T-junction etc. These values are summed for the entire layout and then divided by the total number
of nodes in the building. O’Neill found a strong and significant correlation
between judged complexity and average ICD value (r=0.78, p<0.01). It
was clear that any follow-on complexity-judging task should attempt to reproduce this result. Note that all of the environmental properties discussed
in this section capture aspects of “layout complexity”. Generally, any feature that increases layout complexity could at the same time also foster
wayfinding difficulty, but only for a subset of measures the wayfinding literature actually points to a direct connection (namely corridor segments,
decision points, axial lines and ICD).
The Interdependency of the Layout Measures
The analysis described in the previous paragraph originally included 27
variables of layout characteristics. As these variables were highly redundant a
factor analysis was performed to identify a core set of features capturing a
maximum variability between layouts. Note that the factor analysis was
performed on the layout variables only and was thus independent of the
ratings provided by the study participants. Four independent factors were
identified, explaining 93% of the variance of these variables across the 32
layouts.
The seven variables listed in Table 1 formed a minimal set capturing all
relevant categories identified from the literature: symmetry and decision
points load uniquely in the factor structure, convex spaces and axial lines
168
C. Hölscher, R.C. Dalton
(although having a high direct correlation) show divergent patterns of relations to the remaining factors; topological loops, convexity and ICD load
highly on one common factor, yet topological loops and convexity are
clearly distinct in both their correlation as features in our sample and certainly represent different levels of granularity as spatial variables.
Table 1 Correlations between measures (Pearson’s r); highest loading on factor
Axial Lines
Convex
Spaces
Topological Loops
Decision
Points
Convexity
ICD
Symmetry
Factor No.
Axial
Lines
1.000
-
Convex
Spaces
0.924
1.000
Top.
Loop
0.680
0.559
Dec.
Points
0.169
0.225
Convexity
- 0.661
-0.652
ICD
0.750
0.674
Symmetry
0.082
0.184
-
-
1.000
0.194
-0.532
0.857
0.294
-
-
-
1.000
-0.550
0.391
0.097
1
1
2
3
-1.000
2
0.790
1.000
2
0.199
0.267
1.000
4
Results
First, we investigated the average rating of each participant aggregated
across all layouts that the participant had experienced, as a global indicator
of the impact of modality and architectural expertise as well as highlighting potential differences between assessing complexity versus wayfinding
difficulty (see Table 2).
Across all layouts, the mean rating of complexity did not differ significantly between laypeople and architects (p>.65 for plans, p>.80 for movies).
Yet, for the mean rating of wayfinding difficulty, laypeople and architects show clear differences: architects give higher ratings of wayfinding
difficulty both for plans (t(164)= -2.23, p < .03) and for movies (t(164)=
-1.74, p < .083) than laypeople.
This pattern is further reflected in differences between the average ratings of complexity versus wayfinding difficulty. The architects give significantly higher ratings for wayfinding difficulty than for complexity of
the same stimuli (for plans: t(51)= -5.30, p < .03; for movies: t(51)= -3.93,
p < .083). For laypeople this difference is numerically smaller yet, at least
Comprehension of Layout Complexity
169
for one modality, the movies it is statistically reliable (for plans: t(113)= 1.86, p < .066; for movies: t(113)= -2.84, p < .005).
Table 2 Ratings across all materials (means & standard deviations)
Laypeople
Rating Type
Complexity
Wayfinding
Architects
plan
Standard
Standard
Mean
Deviation
Deviation
3,46
1,11
3,41
1,30
movie
3,42
1,02
3,50
1,20
plan
3,57
1,17
4,09
1,46
movie
3,59
1,18
3,95
1,37
Modality
Mean
By contrast, we find no global difference between the modality plan
versus movie: within each level of rating type (complexity; wayfinding) or
architectural expertise (laypeople; architects) no statistically reliably difference was found between the average rating of plans versus movies (all
t < 1.4, p < .172).
An analysis of variance (with rating type and modality as withinsubject- and expertise as between-subjects-factors) identified a significant,
primary effect of rating type (complexity versus wayfinding; F(1,164) =
35.15; p <.001) and a significant interaction of rating type by expertise
(F(1,164) = 13.75; p < .001) as well as a three-way interaction of these factors with modality (F(1,164) = 8.85; p =.003). In order to interpret this interaction, we repeated the analysis of variance separately for laypeople and
architects. This revealed a significant interaction of rating type and modality (F(1,51) = 5.64; p =.021) for architectural experts, but not for laypeople
(F(1,113) = 1,43; p > .05): While experts see more extreme differences between complexity and wayfinding difficulty in plans rather than movies,
the layperson’s ratings of complexity versus wayfinding differ more distinctly when rating movies, with smaller differences in rating plan-view
images.
To test the stability of ratings between the beginning and end of the experiment, we compared the mean ratings of layouts in the first, middle and
last two blocks of the experiment, separately for each two levels of expertise and modality. We observe a main effect of presentation block within
each of the four combinations of modality and expertise (all F > 4.64, all
p < .01), and, more importantly, a significant interaction of presentation
block and expertise for the plan-view ratings of both complexity (F(1.85;
310.77) = 5.01, p < .01) and wayfinding (F(2; 336) = 6.81, p < .01);
For the ratings of plan views the architects provide stable ratings from
170
C. Hölscher, R.C. Dalton
beginning to end of the experiment session, while the laypeople initially
give lower scores that steadily increase during the session. No such interaction of presentation block and expertise is found for movies.
Environmental Variables and Complexity
To assess the impact of layout features on participants’ ratings we calculated the average ratings for each of the 32 layouts, separately for the eight
possible combinations of expertise (layperson, architect), modality (plan,
movie) and rating type (complexity, wayfinding).
First, we observed high correlations between these eight dependent
measures across layouts, ranging from r = .825 to r = .982. Some remarkable differences between experts and laypeople are also found: the median
correlation between measures is r= .896 for architects and r = .947 for laypeople. This difference is clearly attributable to rating type rather than
presentation modality: for non-experts the agreement between complexity
and wayfinding rating is significantly higher than for experts (plans: r =
.964 versus r = .885, p < .05; movies: r = .982 versus r = .896).
The zero-order correlations of all layout features with participants’ ratings (see Table 3), except for symmetry, proved to be statistically significant for all subgroups of ratings. So each feature variable, by itself, explained a significant portion of the variance in the ratings. Two sets of
multiple regressions were then conducted, either including or excluding the
dominant variable ‘convexity’, separately for each group of ratings. Table
3 documents the resulting standardized beta-weights for these two multiple
regression models. Those environmental features with a significant contribution to the prediction in each final model variant are marked with an asterisk (*).
Convexity, when included in the model, is the strongest predictor of all
ratings, irrespective of expertise, modality or rating type. Removing convexity from the regression model (model 2) reveals that it picks up variance from other spatial factors, especially ICD and topological loops.
Interestingly, as soon as ICD is included in the regression model, the betaweight of topological loops turns sharply negative. Upon closer inspection
of the individual materials we find that the ICD measure is extremely sensitive to the presence of topological loops. And while participants do judge
layouts with such loops as more complex and difficult, they do this to a
lesser extent than the ICD measure by itself would suggest.
Comprehension of Layout Complexity
171
Table 3 Correlations and multiple regression weights of environmental variables
and complexity judgments
Laypeople
Architects
Plan view
Movie
Plan view
Movie
waywaywaywayfinfinfinfinzero-order comcomcomcomcorrelation plexity ding
plexity ding
plexity ding
plexity ding
Axial Lines
.507
.541
.564
.596
.590
.650
.518
.651
Convex
.535
.568
.615
.657
.587
.686
.525
.637
Spaces
Decision
.612
.606
.537
.509
.574
.570
.442
.538
Points
Topological
.288
.382
.379
.445
.457
.554
.331
.524
Loops
ICD
.593
.685
.640
.693
.667
.790
.606
.800
Symmetry
-.077
.065 -.054
.066 -.134
.215 -.275 -.001
Convexity
-.752 -.802 -.725 -.781 -.778 -.887 -.701 -.862
beta wt.s
model 1
Axial Lines
.106
.019
.150
.141
.108
.112
.059 -.042
Convex
Spaces
Decision
Points
Topological
Loops
ICD
.174
.079
.248
.257
.168
.187
.159
.053
.280 *
.237
.198
.113
.205
.118
.073
.112
-.061
-.063
-.010
.040
.147
.113
.058
-.288
.203
.236
.236
.274
.383 *
.103
.139
.181
Symmetry
-.232 *
-.098
-.206
Convexity
-.644 * -.802 * -.725 * -.781 * -.838 * -.887 * -.787 * -.604 *
beta wt.s
model 2
Axial Lines
Convex
Spaces
Decision
Points
Topological
Loops
ICD
Symmetry
-.093 -.301 *
.040 -.432 * -.224 *
.416 *
.208
.250
-.392
.237
-.281
.086
.172
.205
.198
.316
.330
.281
.304 *
.197
.161
.542 *
.297 *
.255
.227
.367 *
.323 *
.152
.181
-.188 -.613 * -.639 * -.539 *
-.170
-.298 -.601 * -.605 *
.612 *
.459 * 1.235* 1.318*
.190 1.095* 1.187*
-.165
-.083
-.201
.932 *
-.094 -.333 *
.006 -.429 *
-.193
172
C. Hölscher, R.C. Dalton
The variable decision points represents the number of true intersections
(and not just turns in corridors) in a layout. Contrary to our expectation,
the number of decision points had a significant impact beyond the other
environmental variables for the plan mode, but not for the movies.
Although symmetry showed the lowest zero-order correlation with ratings, the multiple regression models reveal an important independent contribution of symmetry to the assessment of layout complexity and wayfinding difficulty. Again we identify a distinct difference between laypeople
and experts: Laypeople respond significantly to symmetry but only for
plans and this is especially true of their complexity ratings. When assessing ease of wayfinding, laypeople appear not to pick up this property. By
contrast, the architects in our sample were extremely sensitive to symmetry, with significant contributions of symmetry in three out of four rating
groups. This is a notable difference in the role of experts and laypersons
and a clear finding of this study.
Interaction of Layout-specific and Global Factors
Next we investigate how two pre-selected layout-features (axial lines and
symmetry) interact with presentation modality, rating type and architectural expertise of the participants. While the regression analysis in the previous paragraph looks at correlations of variables, we now return to comparisons of mean differences between groups of stimuli. Layouts were
categorized into bins according to the number of axial lines (low / high)
and symmetry (asymmetrical, 1 line of symmetry, 2 or more lines of symmetry). An analysis of variance (ANOVA) was conducted with the factors
axial lines, lines of symmetry, modality and architectural expertise, separately for ratings of complexity and wayfinding difficulty (see Table 4 for
the results).
As was to be expected from the literature (especially Peponis et al.,
1991) and the previous regression analysis, layouts with a high number of
axial lines were rated as significantly more complex (F(1,164) = 239,30; p
<.001) and difficult for wayfinding (F(1,164) = 256,67; p <.001) compared
to the layouts with only 1-3 axial lines.
Furthermore, we observed a fundamental effect of symmetry (complexity: F(2,328) = 33,81; wayfinding F(2,328) = 63,71; all p <.001): layouts
with a low (one) level of symmetry are rated as least complex and least difficult for wayfinding. The complexity ratings of no-symmetry and highsymmetry layouts do not differ reliably, while for wayfinding assessments
the highly symmetrical layouts are rated as most difficult.
Comprehension of Layout Complexity
173
To understand this rather counter-intuitive finding, we must also consider the higher-order interactions identified in the ANOVA: there is a significant two-way interaction of symmetry and modality (complexity:
F(2,328) = 5,46; wayfinding F(2,328) = 8,08; all p <.01) and a significant
three-way-interaction of modality, expertise and symmetry (wayfinding:
F(2,328) = 2,97; p <.53; complexity: F(2,328) = 2.25; p <.107). While laypeople show no such interaction at all, architects react differently to symmetry depending on the presentation modality.
In addition to the effects of layout variables, this fine-grained analysis
revealed another global difference between laypeople and architects: an interaction of presentation modality and expertise for complexity ratings
(F(1,164) = 4.07; p <.045). Architectural experts judge the same materials
as more simple in plan mode, while laypeople judge the layouts as more
simple when presented as movies.
Individual Layouts with High Prägnanz
Two layouts were isolated from the above analysis, namely the crossshaped and square-shaped layouts. This was done primarily because they
are prototypical examples of shapes with high “Prägnanz” (engl.: conciseness) as described in the literature on Gestalt psychology (Köhler 1929;
Koffka 1935). They represent named-shapes, most likely to be highly familiar to laypeople and architects alike and we expected that they might
form their own category, potentially obscuring the interaction of other layout variables in the full sample. And indeed we observe yet another sharp
contrasting interaction of architectural expertise, rating type and presentation modality for the cross-shaped layout (see Figure 3). The three-way
interaction is statistically highly reliable (F(1,164) = 6.22; p <.014), statistical trends for all two-way interactions (all F(1,164) > 2.58; p <.109),
main effects of rating type (F(1,164) = 24.50; p <.001) and expertise
(F(1,164) = 6.17; p <.014) and a statistical trend for modality (F(1,164) =
2.07; p <.087). It is currently not clear why this distinct pattern is observed
only for the cross-shape but not for the square-shaped layout.
Discussion
Previous studies by Weisman (1981) and O’Neill (1991) have only looked
at zero-order correlations of judgments and have consequently interpreted
the assessments of ’complexity‘ and ’wayfinding‘ to be largely equivalent,
combining them in their analyses. In this study we find substantial differ-
174
C. Hölscher, R.C. Dalton
ences between these measures when comparing movies versus plans or experts versus laypeople.
Table 4 Mean ratings for groups of layouts with varying numbers of axial lines
and symmetry (z-standardized scores)
1-3 axial lines
4-9 axial lines
symmetry
none
medium high
symmetry
none
medium high
Complexity
plan_layperson
-0,153
-0,532
-0,011
0,369
0,244
0,419
movie_layperson
-0,211
-0,687
-0,294
0,303
0,220
0,417
plan_architect
-0,097
-0,502
-0,061
0,282
0,138
0,377
0,133
-0,499
-0,225
0,540
0,296
0,289
plan_layperson
-0,269
-0,615
0,064
0,322
0,133
0,415
movie_layperson
-0,301
-0,675
-0,246
0,281
0,135
0,424
plan_architect
-0,356
-0,585
0,150
0,238
0,038
0,583
movie_architect
-0,068
-0,707
-0,219
0,275
0,007
0,449
movie_architect
Wayfinding
Fig. 3. Mean ratings for the cross-shaped layout
Comprehension of Layout Complexity
175
Our results show that the laypeople’s ratings of complexity versus wayfinding differed more distinctly when rating movies, and with smaller differences in rating plan-view images. This can be contrasted to the performance of the experts who appear to perceive greater differences between
complexity and wayfinding difficulty in plans rather than movies. For the
ratings of plan-views the architects provide stable ratings from beginning
to end of the session, while the laypeople initially give lower scores that
increase during the session. It is as if architects can immediately employ a
consistent use of the rating scale, based on the 5 examples provided on the
pre-test page. In contrast, the non-experts need to view a larger exampleset before developing a consistent rating scale. On ratings for movies, architects and laypeople alike need time to adjust to the rating scale, possibly
because such an assessment rarely forms part of their daily routine.
Architectural experts judge the same materials as being simpler in plan
mode, while laypeople judge the layouts as simpler when presented as
movies. A tentative interpretation of this finding is that experts are more
familiar with assessing plan views, while laypeople have greater difficulties interpreting plans and thus find movies easier to comprehend. This difference does not extend to rating wayfinding difficulties per se: a further
indicator that architects and laypeople interpret the two rating tasks in a
different manner.
Correlation between Judgments and Environmental Factors
A number of environmental variables are shown to correlate highly with
participants’ judgements. Architects react differently to symmetry depending on the presentation modality. Against a general trend of judging movie
versions of identical stimuli as easier to navigate, experts appear to be distinctly critical of the complete lack of symmetry in the low-symmetry
group when presented in movie-mode. In plan-view, the high-symmetry,
high-number-of-elements stimuli are judged as rather complex and difficult to navigate, while in movie-mode these elements receive relatively
positive ratings and here the non-symmetric elements are considered complex and difficult, despite their low number of parts/axes/corridors. For
complexity, the pattern is similar: highly-symmetric elements are judged as
simple and easily navigable in the movie modality, but in plans, the experts
rely less on symmetry and attribute high complexity to layouts with many
elements, despite high symmetry. Furthermore, while participants judge
layouts containing a larger number of ‘topological loops’ as being more
complex and difficult, they do this to a lesser extent than the ICD measure,
a high correlate, would suggest by itself.
176
C. Hölscher, R.C. Dalton
In conclusion, the fact that the variables initially identified through factor analysis appear to be particularly relevant for predicting human assessments of complexity and navigability can be taken as an indication that
our stimuli covered a considerable range of the potential feature space. Yet
there is still considerable co-linearity between some features, especially
convexity, which is correlated with all other features (except symmetry).
The Effect of High Prägnanz
The effects of ‘Prägnanz’ or ‘good form’ were considered by examining
the judgments of two of the stimuli that could be considered to fit into this
category: the cross- and square-shaped layouts. One of these layouts, the
cross revealed a distinct pattern: The experts considered this layout as even
less complex than the laypeople did. Interestingly, this was against their
general trend not to do so across all other layouts. Furthermore, they
clearly differentiated between complexity and wayfinding ratings: the architects judge the cross to be simple in complexity, yet potentially quite
difficult for wayfinding and they are particularly aware of this in plan
view, the mode they frequently use for assessing building-layouts. Once
again, laypeople show no such differentiated pattern. One interpretation of
this is that it might be because architects are sensitive to the need for ‘local
visual differentiation’ for ease of wayfinding and, in the case of the cross,
the potential for confusing one location with another, visually identical,
one is quite high. This consideration simply might not be shared by laypeople.
Future Work
It is intended that work on this study will continue by including a set of
‘information theoretic’ measures of shape and thus layout properties
(based on Leeuwenberg (1968), Shannon & Weaver (1949) etc.). Especially Leeuwenberg’s approach tries to capture how difficult a stimulus is
to encode for human cognition and experts and laypeople may substantially differ in their ability to deal with materials of varying complexity in
this sense.
The lack of a prevalent or universal effect of symmetry was also a puzzling finding of this study, as symmetry is generally considered as a main
factor of reducing visual complexity. It is intended that this be further investigated by examining the effect, if any, of Hillier’s ‘Symmetry Index’,
which provides a continuous scale of symmetry ranging from strongly
Comprehension of Layout Complexity
177
asymmetrical to highly symmetrical in contrast to the concept of symmetry
as a discrete property which a layout either possesses or does not (Hillier
2007). Future work will also include a more detailed analysis of other individual layouts (focusing initially on the ‘good form’ layouts) to further
pinpoint contributing factors of layout assessments.
The present study could also be a starting point for a new line of behavioral experiments on actual wayfinding performance, rather than subjective
judgments on the matter. Such experiments would likely extend the movie
stimuli into interactive Virtual Reality simulations of building layouts. The
studies by O’Neill (1991) indicate that perceived complexity and performance indeed correlate substantially, but it is a fully open question whether
the subjective judgments of architects or lay-people better predict the behavior of building users. One might of course suspect that architects are
better trained to make such judgments properly. But it could also be the
case that lay-people – when presented with materials in the egocentric modality of the movie – actually provide more accurate predictions of their
fellow peer’s performance in wayfinding, while architects might misjudge
the spatial abilities of lay-people (see Bromme et al. 2001).
In conclusion, this study has already yielded interesting results about the
design cognition of architects and lay-people, but it is anticipated that more
will be forthcoming.
Acknowledgements
To the EPSRC Platform Continuation Grant (EPSRC GR/S64561/01) and
to SFB/TR8 Spatial Cognition for co-funding this study. To Kinda
Al_Sayed for the construction of the movie stimuli and to Gregor Wilbertz
for the creation and programming of the online questionnaire. To our participants without whom this study would not have been possible.
References
1. Attneave F: (1957) Physical determinants of the judged complexity of shapes.
J. Exp. Psych 53: 221-227
2. Batty M (2001) Exploring isovist fields: space and shape in architectural and
urban morphology. Environment and Planning B: Planning and Design 28(1):
123 – 150
3. Bromme R, Rambow R, Nückles M (2001) Expertise and estimating what
other people know: The influence of professional experience and type of
knowledge. Journal of Experimental Psychology: Applied 7: 317-330
178
C. Hölscher, R.C. Dalton
4. Brösamle M, Hölscher C (2007) Architects seeing through the eyes of building users. In Barkowsky T, Bilda, Z Hölscher, C Vrachliotis, G (Eds.), Spatial
cognition in architectural design. Melbourne, Australia
5. http://www.sfbtr8.spatial-cognition.de/SCAD/
6. Butler D, Acquino A, Hissong A, Scott P (1993) Wayfinding by newcomers
in a complex building. Human Factors 35(1): 159-173
7. Cross N (2006) Designerly Ways of Knowing. Springer-Verlag London
8. Gärling T, Böök A, Lindberg E (1986) Spatial orientation and wayfinding in
the designed environment: A conceptual analysis and some suggestions for
postoccupancy evaluation. Journal of Architectural Planning Resources 3: 5564
9. Gero JS, Jupp JR (2002) Measuring the information content of architectural
plans. In Hippolyte PL & Miralles E (eds), SIGraDi Caracas 2002, Ediciones
Uni. Central de Venezuela: 155-158
10. Hillier B, Hanson J (1984) The social logic of space. Cambridge University
Press, Cambridge
11. Hillier B (2007) Space is the machine: A configurational theory of architecture. Space Syntax, London
12. Jackson H (2001) Towards a symbiotic coevolutionary approach to architecture. In Creative Evolutionary Systems, Bentley PJ and Corne DW (eds),
Morgan Kaufmann Publishers
13. Koffka K (1935) Principles of gestalt psychology. Harcourt Brace Javanovich,
New York
14. Köhler W (1929) Gestalt psychology. Liveright, New York
15. Leeuwenberg ELJ (1968) Structural information of visual patterns. Mouton &
Co
16. Montello D (2007) The contribution of space syntax to a comprehensive theory of environmental psychology. Proceedings of 7th International Space Syntax Symposium, Istanbul
17. O'Neill MJ (1991) Effects of signage and floor plan configuration on wayfinding accuracy. Environment and Behavior 23(5): 553-574
18. Passini R (1996) Wayfinding design: logic, application and some thoughts on
universality. Design Studies 17(3): 319-331
19. Peponis J, Zimring C, Choi YK (1990). Finding the building in wayfinding.
Environment and Behavior 22(5): 555–590
20. Sha R, Bod R (1993) Computational esthetics (English translation of “Computationele Esthetica”, originally published in Dutch in Informatie en Informatiebeleid 11(1): 54-63l)
21. Weaver W, Shannon CE (1949) The mathematical theory of communication.
Urbana, Illinois, University of Illinois
22. Weisman J (1981) Evaluating architectural legibility: Way-finding in the built
environment. Environment and Behavior 13: 189-204