You are on page 1of 12

Journal of Environmental Psychology (1992) 12, 237-248 0272-4944/92/030237+12508.

00/0
01992 Academic Press Ltd

ENVIRONMENTAL
PSYCHOLOGY
A C O G N I T I V E A N A L Y S I S OF P R E F E R E N C E F O R U R B A N S P A C E S

THOMAS R. HERZOG
Department of Psychology, Grand Valley State University,
Allendale, MI 49401, U.S.A.

Abstract

Preferences for urban spaces were studied as a function of spatial category and nine predictor variables:
spaciousness, refuge, enclosure, coherence, legibility, complexity, mystery, typicality, and age. A non-metric
factor analysis of the preference ratings yielded four categories of urban spaces: Open-Undefined, Well-
Structured, Enclosed Settings, and Blocked Views. These categories are similar to the spatial categories
proposed for natural environments by S. Kaplan (1979, Assessing Amenity Resource Values. USDA Forest
Service General Technical Report RM-68), who stressed openness and spatial definition as bases for
categorization. The Well-Structured category was best liked, with the other categories not very well liked and
about equal in preference. Regression analyses revealed three variables as consistent predictors of preference.
Coherence and complexity were positively related to preference, and age was negatively related. Overall, the
results support the Kaplans' proposal that both spatial and non-spatial factors are important in categorizing
environments and in explaining environmental preferences.

Introduction settings contain a screened or otherwise protected


area which could serve as a hiding place. The
A cognitive analysis of preference for urban spaces influence of Appleton's (1975) concept of refuge is
apparent. Finally, Blocked Views contain visual
The purpose of this study was to investigate the role obstructions near the observer that prevent visual
of spatial configuration in determining both percep- access.
tual categories and preference reactions for urban Kaplan proposed t h a t the spatial variables of
settings. Specifically, the study focused on whether openness and definition would not only define envi-
certain spatially defined .categories, known to be ronmental categories but would also affect relative
salient in natural settings, would also be relevant in preferences among the categories. Because of their
urban environments. The point of departure for the evolutionary history, h u m a n s constantly evaluate
study was a paper by S. Kaplan (1979) on what settings to prepare for effective action. Settings t h a t
the Kaplans (R. Kaplan & S. Kaplan, 1989) now are readily organized spatially aid in this process
prefer to call category-identifying methodologies. and should therefore be preferred. There is an obvi-
Summarizing several studies involving primarily ous similarity here to Gibson's (1979) concept of af-
natural settings, Kaplan noted t h a t the basis for fordance and to ideas discussed earlier by Gregory
environmental categories derived from preference (1969) and S. Kaplan (1975). These considerations
ratings could be either the content or the spatial lead to the prediction t h a t the Spacious-Structured
configuration of the settings. He went on to identify category should be relatively high in preference and
four types of spatially based categories. Open- both the Open-Undefined and the Blocked-Views
Undefined settings are typically flat, open, and lack- categories should be relatively low. Some enclosed
ing in spatial definition. In contrast, Spacious- settings should be preferred, but those t h a t are too
Structured settings contain elements (trees, edges, cramped or lacking in spatial definition would not
landmarks, etc.) arranged to facilitate the cognitive be preferred. The research reviewed by Kaplan
organization of the setting in depth. Enclosed generally supported these predictions.
237
238 T.R. Herzog

Ten years later not much had changed. In their Three of them may be considered general affor-
book, R. Kaplan and S. Kaplan (1989) provided es- dances of settings in the sense described earlier.
sentially the same analysis of spatial configuration Spaciousness, defined as the feeling of spaciousness
except for the omission of the Enclosed category. or room to wander, permitted an empirical assess-
Apparently, in those few instances where enclosure ment of predictions concerning openness. Refuge,
or refuge has b e e n investigated empirically it has the extent to which the setting contains possible
not worked well as either a basis for categorization hiding places viewed from the outside, corresponds
or a preference predictor (Woodcock, 1982; Herzog, closely to Appleton's (1975, 1984) use of the term.
1989). Enclosure, the feeling of viewing the environment
Because most of the research on the salience of from inside a hiding place, was borrowed from
spatial configurations involved natural settings, the Woodcock (1982) who called it primary refuge.
question in this study was how well the findings Thus, both possible senses of refuge, hiding places
would transfer to urban settings. Participants rated viewed from without or within, were assessed. En-
color slides of urban spaces for preference. The closure, as defined in this study, was disavowed by
validity of color slides as surrogates for natural Appleton (cited in Woodcock, 1982) as an appropri-
environments was endorsed in a thorough review of ate definition of refuge. Nonetheless, because the
landscape simulation research (Zube et al., 1987). enclosed settings used in this study were primarily
The focus of each slide was on the space itself, not views from within, it seemed prudent to obtain rat-
on specific structures. Examples of each of Kaplan's ings of enclosure.
four spatially defined categories were included in Another group of four rated predictors was de-
the sample of settings. In addition, a fifth category rived directly from the Kaplan's model of environ-
that seemed prevalent in urban environments was mental preference. Their model (S. Kaplan & R. Ka-
included. This involved corridors, views into or out plan, 1978, 1982; S. Kaplan, 1987; R. Kaplan & S.
from pathways (usually streets or alleys) bordered Kaplan, 1989) fits within the cognitive paradigm of
by tall buildings or other structures. Subjectively, Zube et al. (1982), the psychological model of Daniel
such views typically convey a strong sense of linear and Vining (1983), and the behavioral viewpoint of
perspective and are therefore spatially well defined. Zube (1984). The Kaplans propose that environmen-
Categories as seen by the participants were deter- tal preference reactions are mediated by two cogni-
mined empirically via non-metric factor analysis of tive processes, understanding and exploration. Un-
the preference ratings. It was thus possible to see derstanding refers to comprehending or making
how well the empirically derived categories corre- sense of an environment, exploration to having
sponded to the spatial categories proposed by one's interest aroused and held by an environment
Kaplan. while being attracted toward sources of additional
It might seem trivial to have people rate settings information. Both processes have survival value,
containing certain spatial qualities and then show and thus environments that engage both should be
that the settings are categorized by those very qual- preferred. Variables promoting understanding in-
ities. However, the exercise is far from trivial for clude the degree of order or organization in the im-
two reasons. First, there is no guarantee that what mediate setting (coherence) and the apparent ease
seems salient to the researcher will also seem of wayfinding in the larger environment that in-
salient to raters. Second, for urban settings, there is cludes the immediate setting (legibility). Explo-
good reason to believe that non-spatial variables ration is promoted by the richness or diversity of
may either combine with or even override spatial the immediate setting (complexity) and by the
variables as bases for categorization. Two non-spa- promise of new information if one could travel
tial variables that have repeatedly figured promi- deeper into the environment (mystery). The effec-
nently in urban categories have been the presence tiveness of mystery and coherence has been demon-
of natural elements and the age of structures de- strated many times (R. Kaplan & S. Kaplan, 1989;
picted (R. Kaplan & S. Kaplan, 1989). So it is mean- R. Kaplan et al., 1989), but the other two variables
ingful, both theoretically and practically, to pose the have received less consistent empirical support, per-
question of whether people, when given the oppor- haps because of difficulties in defining them clearly
tunity to categorize on the basis of spatial configu- to raters.
ration, will do so. The last two rated predictors were included
To aid in characterizing categories and in the fur- primarily because of demonstrated relevance in the
ther analysis of preference, the settings were also environmental preference literature. Typicality,
rated independently on nine predictor variables. defined as how good an example a setting is of
P r e f e r e n c e for U r b a n S p a c e s 239

whatever category it belongs to, was an a t t e m p t to Procedure


assess the resemblance of specific settings to envi-
ronmental categories the rater knows well. Presum- Participants rated each of the 70 settings on one of
ably such familiarity should aid understanding but ten variables. All ratings used a five-point scale
might detract from exploration. As defined here, the ranging from 1 = 'not at all' to 5 = 'a great deal'.
variable was ineffective in Herzog's (1989) study of There were nine predictor variables. Spaciousness
urban nature. However, Purcell (1986) found typi- was defined as 'the feeling of spaciousness or depth
cality useful in accounting for preferences for the the scene conveys, how much room there is to wan-
specific reference category of churches. Under the der into it. To what extent does the structure of the
name 'identifiability', the sense of familiarity scene suggest t h a t one would have to go a long way
evoked by a setting has been a generally positive to reach its farthest point'? Refuge involved 'fea-
predictor of preference (Herzog et al., 1982; Herzog, tures t h a t you could use to hide yourself. In other
1984, 1987). The last predictor was the rated age of words, to w h a t extent does the scene contain possi-
the elements in each setting. If age is a non-spatial ble hiding places t h a t you are viewing from the out-
basis for category formation in urban environments, side'? By contrast, enclosure involved 'viewing the
as documented earlier, ratings of age would provide environment from inside a hiding place ... a feeling
supportive evidence for such an interpretation. Age of being 'inside looking out' ... a strong feeling of
might also be a predictor of preference, independent being enclosed in a hiding place'. Coherence was
of category effects, although it too failed to predict 'how well the scene 'hangs together'. How easy is it
preference in Herzog's (1989) u r b a n - n a t u r e study. to organize and structure the scene'? Legibility was
'how easy it would be to find your way around in the
environment depicted ... to figure out where you are
Method at any given moment or to find your way back to
any given point in the environment'. Complexity
Participants
was 'how much is going on in the scene, how much
The sample consisted of 326 undergraduate stu- there is to look at', how much 'the scene contains a
dents, 195 females and 131 males, at Grand Valley lot of elements of different kinds'. Mystery was pre-
State University. The students received extra sent when a setting 'promises more to be seen if you
course credit for participation. Twenty-four sessions could walk deeper into it'. Typicality referred to the
of from five to 22 participants were run. 'extent the scene seems to be a representative exam-
ple of its class. How good an example is the scene of
Stimuli whatever category it belongs to'? Age was 'how old
the elements in the scene seem to be'. The criterion
The settings consisted of 70 color slides of urban variable was preference, defined as 'how much you
environments containing examples of the five like the scene, for whatever reason'.
categories of spaces discussed in the introduction: Sessions proceeded as follows. First, five sample
open, undefined (n = 8); spacious, well structured slides were rated to help participants get used to
(n = 19); enclosed (n = 11); blocked views (n = 9); the task and the rating scale. Then participants
and corridors (n = 23). In the author's judgement, rated 80 slides, presented in two sets of 40 each,
goodness of fit of the settings to the indicated with a brief intermission between sets. In both sets,
categories ranged from very high to marginal. the first three and the last two slides were consid-
Undoubtedly, informed observers could reasonably ered filler slides. Their purpose was to absorb any
disagree on some of the a priori category place- beginning- or end-of-set effects that might have af-
ments. Although n a t u r a l and built structures were fected the ratings. Both sample slides and filler
typically plentiful in the settings, the focal point of slides were chosen from the two largest a priori spa-
the scene for compositional purposes was on the tial categories, spacious-structured and corridors,
space it contained. None of the settings contained because there were excess slides available in those
people since they have been found to be powerful categories but not in the other categories. The re-
distractors (Herzog et al., 1976). Age of structures, maining 70 slides from both sets yielded the data
known to be a salient factor in urban preference for analysis. These 70 slides were presented in three
caegories (Herzog et al., 1976, 1982; Herzog, 1989) different orders, with each order used for 8 of
ranged from very old to very new. All settings were the 24 sessions. Each order of slide presentation
from the cities of Grand Rapids and Kalamazoo, was used in one-third of the sessions in which each
Michigan, and their suburbs variable was rated. One of the orders of slide
240 T.R. H e r z o g

presentation was generated randomly. The second non-metric version of principal-axes factor analysis.
presentation order was the reverse of the first order, It uses squared multiple correlations in the diago-
and the third presentation order was devised by nal of the input correlation matrix and a varimax
interchanging the halves of the first order. rotation of the final solution. The procedure finds a
Eighteen of the sessions were devoted to the pre- solution of the user-specified dimensionality t h a t
dictor variables. In such sessions, three predictor best fits the rank order of the original correlation
variables were rated simultaneously, with each par- matrix rather than the more stringent linear trans-
ticipant rating for only one variable and approxi- formation of the original correlations required by
mately one-third of the group rating each variable. metric factor analysis. Proponents of non-metric
Participants were aware of the other two variables analysis argue t h a t more stable solutions in fewer
being rated during these sessions. For each order of dimensions can be found. The input matrix for this
slide presentation, three sets of three predictors study contained correlations ranging from -0-29 to
were chosen randomly without replacement from 0.69, and 86 of them (or 4%) were negative. For
the pool of nine predictor variables, with the excep- descriptive purposes, dimensional composition was
tion t h a t refuge and enclosure were not allowed determined by including all settings with a factor
within the same set of three predictor variables. loading greater t h a n ] 0-501 on one dimension and
Each such set of three predictors was used in two no loading greater t h a n ]0-451 on any other dimen-
sessions with its assigned order of slide presenta- sion. With this criterion, for each dimension con-
tion. Six sessions were devoted exclusively to the taining enough settings to provide a basis for inter-
criterion variable, preference. Each order of slide pretation, all factor loadings for included settings
presentation was used in two of these sessions. had the same sign. Hence, such dimensions may be
Viewing time was 15 s per slide in all sessions. treated as clusters or categories of similar settings.
Final sample sizes were 105 for preference, 26 for Solutions in five to 11 dimensions were examined,
refuge, 25 each for spaciousness, enclosure, coher- and all yielded four interpretable dimensions. The
ence, complexity, and mystery, 24 for typicality, and first four dimensions of the 11-dimensional solution
23 each for legibility and age. The larger sample for are described here. That solution yielded commu-
preference was necessary for the factor analysis de- nalities ranging from 0.45 to 0.77.
scribed below. The first two dimensions or categories each
contained the same number of settings (13) but
contrasted sharply in both spatial and non-spatial
Results characteristics. Specifically, the first category
tended toward large, open, unstructured spaces in
Reliability of measurement the foreground and older structures in the back-
ground. In contrast, the second category featured
To evaluate reliability of measurement for each smaller spaces t h a t were well structured in depth
variable, final samples were divided into half sam- and populated by more contemporary structures.
ples, and m e a n ratings for each setting were com- The two categories were named Open-Undefined
puted based on each h a l f sample. The two sets of 70 (Figure 1) and Well-Structured (Figure 2), respec-
mean-per-setting scores were intercorrelated for tively, although clearly age was also a factor in dis-
each variable. The resultant correlations were cor- tinguishing them. The last two categories were also
rected by the Spearman-Brown formula to yield distinguishable both spatially and with respect to
split-half reliability coefficients for each variable. age. The third category (nine settings) consisted
The reliability coefficients ranged from 0.79 for typi- primarily of views from within enclosed surround-
cality to 0.98 for preference, spaciousness, and age. ings and hence was named Enclosed Settings
The average value was 0.90. Thus, reliability of (Figure 3). Because the enclosing surround involved
measurement seems adequate. an alley in several instances, the structures tended
to appear noticeably older. The final category (seven
Environmental categories settings) contained of a variety of blocked views, but
the majority of them were not blocked by an ob-
To discover the categories embodied in the partici- struction in the immediate foreground. Typically,
pants' preference reactions, the preference ratings one could see some distance into the setting and
were analyzed by non-metric factor analysis, was then confronted by a noticeable obstruction,
specifically the Guttman-Lingoes Smallest Space greatly reducing further visual access. An example
Analysis III (SSA-III) (Lingoes, 1972). SSA-III is a would be a view down an alley with the wall of a
P r e f e r e n c e for U r b a n S p a c e s 241

FIGURE1. Scenefrom the Open-Undefinedcategory. (a)

building extending partially or completely across


the end of the alley. Sometimes the obstructing
surface appeared to seal off the end of the view (a
true cul-de-sac), but just as often it was clear that
the obstruction did not seal off the end of the view
although further visual access was severely limited.
Ages of structures varied. This category was named
B l o c k e d V i e w s (Figure 4), but clearly this is not
meant to imply foreground blocking.
Because of differences in the nature of the empiri-
cal categories, described above, and those postu-
lated a priori, and because one of the postulated
categories (corridors) did not appear in the empiri-
cal set, the correspondence between the postulated
placement of specific settings into categories and
(b)
their actual placement was not very good. Table 1
contains the frequency distribution of settings in FIGURE3. Scenes from the Enclosed-Settingscategory.
each empirical category as a function of postulated
categories. F u r t h e r discussion of this lack of corre-
spondence appears in the Discussion section.

FIGURE2. Scenefrom the Well-Structuredcategory. FIGURE4. Scenefromthe BlockedViews category.


242 T. R. H e r z o g

TABLE 1
Frequency distribution of settings in each empirical category as a function of postulated categories

Postulated categories Empirical categories

Open-Undefined Well-Structured Enclosed Settings Blocked Views

Open-Undefined 3 0 0 0
Spacious-Structured 8 4 0 0
Enclosed 0 4 2 1
Blocked Views 1 2 0 1
Corridors 1 3 7 5
Totals 13 13 9 7

One way to c h a r a c t e r i z e t h e empirical categories Table 2 contains m e a n s a n d s t a n d a r d deviations for


is to e x a m i n e differences across categories in the all nine predictor variables as a function of environ-
m e a n r a t i n g s on predictor variables. For this pur- m e n t a l category. Results for both category a n d
pose, two t y p e s of scores were c o m p u t e d as raw d a t a setting scores are included. T h e m e a n s are the s a m e
for analysis. T h e first, a category score, was simply for e i t h e r type of score.
the m e a n r a t i n g for all settings comprising a cate- Two a n a l y s e s of v a r i a n c e were carried out for
gory. Thus, for each r a t i n g variable, e v e r y partici- each predictor variable. The first used c a t e g o r y
p a n t h a d a category score for each of the environ- scores as the d e p e n d e n t variable, a n d the second
m e n t a l categories r e s u l t i n g from the n o n - m e t r i c used setting scores. The first analysis allows conclu-
factor analysis. T h e second type of score was a sions to be generalized to the p o p u l a t i o n of partici-
setting score, t h e m e a n for each setting based on all pants, the second analysis to the population of set-
p a r t i c i p a n t s who r a t e d each variable. Thus, for each tings. F o r both analyses, the i n d e p e n d e n t v a r i a b l e
r a t i n g variable, e v e r y setting h a d a setting score. was e n v i r o n m e n t a l category. Only effects w i t h

TABLE2
Means and standard deviations (in parentheses, for category scores, left, and setting scores, right) as a function of
environmental category for each rated variable

Variable Environmental categories

Open-Undefined Well-Structured Enclosed Settings Blocked Views

Spaciousness 3-65a 2.675 2.445 2.095


(0.60, 0.68) (0.50, 1.01) (0.72, 0.69) (0.61, 0-43)
Refuge 2.99 a 2.46 a 2.50 a 1.665
(0-83, 0.78) (0.54, 0.41) (0.53, 0.65) (0.44, 0.25)
Enclosure 1.90a 2.50ab 3.71 c 3.325c
(0.62, 0.40) (0.38, 0.81) (0.43, 1.02) (0.77, 0.64)
Coherence 3.01~ 3.655 2"64a 2"62a
(0.51, 0"29) (0.60, 0.39) (0"64, 0.42) (0.68, 0.34)
Legibility 3.29a 3.48a 2.815 3"05a5
(0.63, 0-28) (0.59, 0.42) (0.82, 0.39) (1.00, 0.31)
Complexity 3"28a 2"79ab 2.375c 1.68c
(0.58, 0.87) (0.60, 0.64) (0.50, 0.79) (0.33, 0.34)
Mystery 2-53 a 3.425 3.805 3.665
(0-79, 0.38) (0.54, 0.38) (0.67, 0.53) (0.68, 0-26)
Typicality 3.42 a 2.99a5 2.745 2.665
(0"45, 0"37) (0"80, 0"29) (0"76, 0"83) (0.78, 0"50)
Age 3.66 a 1.485 4.14 a 3.09 c
(0.38, 0-50) (0.21, 0.40) (0.34, 0.51) (0-38, 0.60)
Preference 1.98~ 3.36 b 1.68a 1.62~
(0-62, 0.33) (0.67, 0-38) (0.68, 0.48) (0.57, 0.22)

Note. For each rating variable, means not having a common subscript differ significantly at p < 0.05 in Tukey-B tests
for both category and setting scores.
P r e f e r e n c e for U r b a n S p a c e s 243

p < 0-05 in both category- and setting-score analyses ing variables, the sharpest category differences oc-
were considered statistically significant. The analy- curred for age. The Enclosed-Settings and Open-
ses of variance for the category effect yielded Undefined categories were highest in age (denoting
significant differences at p < 0.01 for each predictor older structures) and did not differ from each other.
variable and both types of scores. In addition, for The Well-Structured category was lowest in age,
setting scores a multivariate analysis of Variance with the Blocked-Views category at an intermediate
was performed on all nine predictor variables as a level. On complexity, the Open-Undefined and Well-
set. The multivariate test statistics provide further Structured categories did not differ and were each
protection against Type I errors. Such an analysis higher t h a n the Blocked-Views category. The Open-
was not possible for category scores. The multi- Undefined category was also higher than the En-
variate test of the category effect yielded significant closed-Settings category. Finally, the Open-
results at p < 0.001 for all multivariate test statis- Undefined category was higher in typicality than all
tics (Pillai, Hotelling, Wilks). but the Well-Structured category, and no other
The nature of the category effect for each predic- pairs of categories differed significantly.
tor variable was explored further by means of the
Tukey-B test for pairwise multiple comparisons The prediction of preference
(Wike, 1971), with Kramer's (1956) modification for
unequal sample sizes where appropriate. The pat- Categories. The last two rows of Table 2 contain
tern of significant (p < 0.05) pairwise differences for preference means and standard deviations, for both
both category and setting scores is summarized in category and setting scores, as a function of envi-
Table 2. As might be expected, the Open-Undefined ronmental category. Analyses of variance yielded a
category was highest in spaciousness and lowest in significant effect of categories (p < 0.01) for both
mystery. For both rating variables, the other three types of scores. Tukey-B tests of the category effect
categories did not differ from each other. Also as revealed t h a t the Well-Structured category was
might be expected, the Well-Structured category higher in preference than the other categories,
was seen as high in spatial organization. It was which did not differ from each other.
highest in coherence, with the other three cate- Predictor variables. Table 3 presents intercorrela-
gories not differing from each other. As for legibil- tions among the rated variables based on setting
ity, both the Well-Structured and Open-Undefined scores for all 70 settings. The majority of the predic-
categories were higher than the Enclosed-Settings tor variables were significantly correlated with pref-
category, but they failed to differ from each other. erence, but 25 of the 36 intercorrelations among
Not surprisingly, the Enclosed-Settings category predictor variables were also significant. To under-
was higher in enclosure t h a n all other categories ex- stand how the variables of the study worked to-
cept, curiously, the Blocked-Views category. The gether, multiple regression analyses, which take
latter category was higher in enclosure t h a n the correlations among predictors into account, were
Open-Undefined category. The Blocked-Views cate- performed on the setting scores with preference as
gory was lowest in refuge, with the other three cate- the criterion variable. The first analysis included all
gories not differing from each other. Of the remain- 70 settings, and the nine rated predictor variables

TABLE3
Intercorrelations among the rating variables for setting scores
Variable 2 3 4 5 6 7 8 9 10

(1) Preference 0.14 0.01 -0.28 0-81 0.52 0.24 0.07 0.11 -0.76
(2) Spaciousness 0.48 -0.44 0.31 0.42 0.72 -0.49 0-56 0.22
(3) Refuge -0.24 0.05 -0-10 0.74 -0.28 0.42 0.19
(4) Enclosure -0.32 -0.50 -0.43 0.53 -0.36 0-19
(5) Coherence 0.52 0.26 -0.09 0.41 -0.56
(6) Legibility 0.20 -0.37 0.28 0.32
(7) Complexity -0.43 0.51 0.13
(8) Mystery -0.43 -0.13
(9) Typicality 0-20
(10) Age

Note. For Irl > 0-23,p < 0.05; for Irl > 0.30,p < 0.01.
244 T.R. Herzog

TABLE4
Multiple regression results for preference setting scores: regression weights (B), partial correlations (r), and squared
multiple correlations (R-Sq)
All settings Category settings
(n -- 70) (n = 42)
Predictor B r B r B r

Spaciousness -0.08 -0.15 -0-06 -0.14 -0.00 -0.01


Refuge -0.03 -0-04 -0.09 -0.16 -0.09 -0.16
Enclosure 0.03 0.09 0-08 0.23 0.06 0.18
Coherence 0.67 0.59** 0.63 0.65** 0.46 0.53**
Legibility 0.36 0.39** 0.20 0.26 0.19 0.26
Complexity 0.31 0.45** 0-52 0.64** 0.41 0.55**
Mystery 0.21 0.35** 0.12 0.24 0-17 0.30
Typicality -0.12 -0.17 -0.34 -0-48** -0.30 0.45*
Age -0.29 -0.56** -0.34 -0.72** -0.24 -0.41"
Categories -- 0.47
R-Sq 0-87** 0.93** 0.95**

* p < 0.05. ** p < 0.01.

served as predictors. The second analysis included just described except t h a t it included only the four
only the 42 settings in the environmental categories significant rated predictors. The purpose was to
t h a t emerged from the non-metric factor analysis. see if an increase in power from the elimination of
The nine rated predictors were entered first as ineffective predictors would yield a significant effect
a block of variables followed by a set of d u m m y for environmental categories when they were added
vectors representing the effect of environmental cat- in the second step of the analysis. The answer was
egories. The third analysis was identical to the sec- yes. The partial correlation for categories rose to
ond except t h a t it included only the significant rated 0.52 (p -- 0.012).
predictors from the second analysis. The results of
the first two regression analyses are presented in
Table 4. Discussion
In the analysis of all 70 settings (left third of
Table 4), the nine rated predictors accounted for Environmental categories
87% of the preference variance, with significant con-
tributions from coherence, legibility, complexity, A central question motivating this study was
mystery, and age. Settings high in the first four whether the spatial categories proposed by S.
variables but low in age (i.e. contemporary struc- Kaplan (1979) for natural environments could also
tures) were most preferred. In the analysis of the 42 be found in urban environments. More generally,
category settings, summarized in the rest of the would the spatial dimensions of openness and
table, two points are noteworthy. First, when only definition figure prominently in the categorization
the rated predictors were entered into the analysis of urban spaces? The answer to both questions ap-
(middle third of Table 4), the pattern of significant pears to be yes. The empirically derived preference
preditors shifted somewhat. Legibility and mystery, categories of this study were sufficiently similar to
although still positively related to preference, were Kaplan's t h a t it was possible to borrow his category
no longer significant predictors. In contrast, the names without doing violence to his category de-
magnitude of the negative relationship between scriptions. On the whole it is clear t h a t openness
typicality and preference increased to the point t h a t and spatial definition can be as salient in urban set-
typicality became a significant predictor. Second, tings as they are in natural settings. Of course,
when environmental categories were added to the finding t h a t spatial features are noticed when
analysis (right third of Table 4), there was a slight spaces are depicted does not answer the question of
(2%) but not quite significant (p = 0.065) improve- whether such features typically have an impact on
ment in prediction, and the same four rated predic- people's perceptions of urban settings. However, the
tors were significant. The final regression analysis, results agree with those of Lynch and Rivkin (1959)
not shown in Table 4, was identical to the analysis in suggesting t h a t an affirmative answer to t h a t
P r e f e r e n c e for U r b a n S p a c e s 245

question is plausible. The extent of such impact also showed the expected difference between the
remains an open but researchable question. Well-Structured and Open-Undefined categories in
Some discrepancies between the categories of this this study.
study and the spatial categories originally proposed The second anomaly involves the equality of the
by S. Kaplan (1979) are instructive. For example, Enclosed-Settings and Blocked-Views categories
although Kaplan's well-structured category for nat- with respect to rated enclosure. This may have been
ural environments was also described as spacious, a matter of statistical power because the categories
the Well-Structured category of this study was nei- did differ in the more powerful analysis of category
ther high nor low in rated spaciousness (mean = scores but not in the analysis of setting scores.
2.67). It is clear from the Kaplans' more recent dis- What really needs explaining is the high enclosure
cussion of these issues (R. Kaplan & S. Kaplan, rating of the Blocked-Views category. In fact, many
1989) that structure or spatial definition is possible of the blocked-view settings do convey a feeling of
at all levels of openness except for the limiting case enclosure in the sense of being surrounded in an
of blocked foreground views. Thus, the discrepancy enclosed space. The view is into an enclosed space
m a y simply reflect a reality concerning urban envi- from a vantage point near but inside the entrance
ronments: Structure tends to occur or be noticed at (e.g. the mouth of an alley). In contrast, the typical
intermediate levels of openness. view in the Enclosed-Settings category is oriented
A more intriguing discrepancy concerns the toward the outside from a vantage point within the
Blocked-Views category. The Kaplans' discussion of enclosed space. Although the definition given to
this category (S. Kaplan, 1979; R. Kaplan & S. raters stressed the latter type of view, it is possible
Kaplan, 1989) and their photographic examples em- that they m a y have been responding to the sense of
phasize foreground obstructions. However, in this being inside an enclosed area, which is about equal
study most of the blocked-view settings contained in either type of view.
an obstruction in the middle distance. Denial of Although the categories can be interpreted
visual access m a y be such an important concern in in terms of spatial factors, it is clear that non-
urban settings that it becomes the focal point for spatial factors are also involved. In particular,
categorization even when it occurs in the middle the categories contrasted strongly in the rated age
distance. If so, the psychodynamics behind such a of the elements depicted, with the Enclosed-
concern would provide a fruitful direction for fur- Settings and Open-Undefined categories at the
ther research. An intriguing alternative is that aged end of the continuum and the Well-Structured
viewers failed to distinguish between true cul-de- category at the contemporary end. Because age
sacs (sealed-off end points) and false ones. Instead has repeatedly figured in the categorization of
they interpreted both configurations as true cul-de- urban settings in past research (e.g. Herzog et al.,
sacs, potential traps, and therefore highly salient 1976, 1982; Herzog, 1989) it would have been
spaces. The low refuge rating of the Blocked-Views surprising if it had not played a role in this study.
category is consistent with this interpretation. In a sense, the mix of spatial and non-spatial factors
Future research will be necessary to determine involved in categorizing the urban spaces of this
whether viewers typically distinguish between true study represents an ideal outcome. The phrase
and false cul-de-sacs. 'urban space' denotes both content (urban) and
Two apparent anomalies in the present categories configuration (space).
involve a lack of category differences in predictor The fate of the proposed corridor category should
variables where they might have been expected. be noted. Corridors were included because the
First, the Well-Structured and Open-Undefined cat- author had the intuition that they seemed prevalent
egories were equal in legibility. Because legibility is in urban settings, but no corridor category emerged
supposed to tap wayfinding potential as reflected in from the analysis of the preference ratings. Instead,
three-dimensional spatial structure, the Well-Struc- the corridor settings were perceived in terms of
tured category should have been rated higher in leg- openness and spatial definition, which led to their
ibility. Either the concept or its definition may be distribution across empirical categories as sum-
at fault. In discussing its poor record in recent marized in Table 1. The theory-based categories
environmental-preference research, the Kaplans (R. were confirmed; the intuitive category was not. So
Kaplan & S. Kaplan, 1989) noted that legibility much for intuition.
needs further development. Coherence, the other Most of the points just made may be invoked to
predictor variable involving spatial organization, account for the lack of correspondence between
not only has the better research track record but postulated and actual category placement of specific
246 T.R. Herzog

settings, as summarized in Table 1. For example, Blocked Views, Enclosed Settings, and Open-
if postulated corridor placements are considered Undefined. The Well-Structured settings, by con-
correct' p l a c e m e n t s in Table 1 (add the frequency trast, tend to have higher visual access and are cor-
in the Corridors row to t h e frequency of t h e respondingly higher in preference. Thus, visual
'correct' row within each column), t h e n t h e penetration or access m a y be worth pursuing as a
expected category is the modal category for all predictor variable in future studies of preferences
b u t the Open-Undefined category. The m a i n for urban spaces.
problem w i t h the Open-Undefined category is
t h a t several of t h e settings classified as spacious- Predictor variables. Preference differences across
s t r u c t u r e d by t h e a u t h o r were seen i n s t e a d as categories are largely accounted for by corresponding
open-undefined b y the preference raters. Inspec- differences in predictor variables. The most notable
tion reveals t h a t t h e v a s t majority of those parallels involve coherence and age. The Tukey-B
settings f e a t u r e a relatively large open a r e a in tests on category means yielded identical results
the foreground even t h o u g h some spatial struc- for preference and coherence. The category effect
t u r i n g is also present. A p p a r e n t l y the open a r e a for age was very nearly the inverse of that for
w a s more salient to t h e r a t e r s t h a n t h e cues for preference, the sole discrepancy being the inter-
spatial definition. The result is t h a t the empirical mediate age mean for Blocked Views.
Open-Undefined category contains i n s t a n c e s in A broader view of the role of the predictor
which o p e n n e s s d o m i n a t e s s t r u c t u r a l cues, a n d variables reveals a pattern unlike that in earlier
the p o s t u l a t e d s p a c i o u s - s t r u c t u r e d category h a s studies of preference involving regression analyses
to be reconceptualized as W e l l - S t r u c t u r e d b u t not (Herzog, 1985, 1987, 1989). Specifically, the effec-
necessarily spacious. tive predictors were different for the entire set of
70 settings than for the 42 category settings (left
The prediction of preference two-thirds of Table 4). This must mean that the
28 uncategorized settings, considered as a fifth
Categories. The categories differed in preference, 'category' (the 'remainder' category), differ some-
with the Well-Structured category most preferred what from the regular categories in the relative
and the other three categories less preferred and effectiveness of the predictors or in the location of
roughly equal to each other. Apparently the cate- the entire category in the predictor-preference space
gories predicted preference independently of the or both. In any event, the rated predictors can b e
rated predictors although their independent contri- divided into three groups: ineffective predictors (not
bution was marginal. The effect of categories was effective in either analysis), inconsistent predictors
significant only when non-significant rated predic- (effective in one analysis but not the other), and
tors were dropped from the regression analysis. The consistent predictors (effective in both analyses).
small additional effect of categories may be a matter The changes that produced the inconsistent group
of interactions among the rated predictors or it m a y of predictors were fairly modest. The most dramatic
involve some variable(s) not included in the study. was the rise in the partial correlation for typicality
An intriguing candidate is visual penetration, from - 0 . 1 7 in the full analysis to - 0 . 4 8 in the
defined by Ruddell et al. (1989, p. 395) as 'the extent analysis of category settings.
to which a line of sight extends uninterrupted into The ineffective predictors were the three affor-
or through a landscape scene'. They showed that dances: spaciousness, refuge, and enclosure. As a
visual penetration was a potent predictor of scenic preference predictor, spaciousness has an incon-
beauty estimates in near-forest settings. Blocked sistent track record (R. Kaplan & S. Kaplan, 1989).
views clearly lower visual penetration by limiting Its successes have tended to occur in studies of
visual access. Woodcock (1982) suggested that natural environments, its failures in studies of urban
enclosure, in the sense of looking out from a hiding environments, a point perhaps worth pursuing in
place through a constricted opening (his primary future research. Enclosure, as primary refuge, was
refuge), may do much the same thing. Although a negative predictor of preference for hardwood-
Ruddell et al. did not discuss large unstructured forest settings in Woodcock's (1982) study. Refuge
open areas with relatively impenetrable borders in has proved ineffective in two studies (Woodcock,
the background, such configurations presumably 1982; Herzog, 1989) besides this one. Although
would be judged low in visual penetration and in inconsistent or ineffective as preference predictors,
preference. What has just been described are the affordances clearly play a role defining preference
three lower-preference categories of this study: categories, as noted earlier.
P r e f e r e n c e for U r b a n Spaces 247

The inconsistent predictors were legibility, preference, it need not be. Frewald (1989) found
mystery, and typicality. As currently defined, that well-maintained older buildings were generally
legibility tends to be strongly correlated with coher- more preferred than modern buildings, and this was
ence (Table 3) and to be dominated by coherence in particularly true for older buildings with visually
the prediction of preference (Table 4). The same rich and highly articulated facades. Clearly, age
pattern occurred in Herzog (1989). Thus, legibility interacts with design features and factors like
needs either or both of the following: (a) a definition perceived maintenance or cleanliness in affecting
that will distinguish it empirically from coherence, preference. Further exploration of the complex
or (b) a measurement procedure that will tap into relationship between age and preference should
the conceptual distinctions between the two con- provide a fruitful agenda for future research.
cepts. The inconsistency of mystery may stem from
several factors. The first is the low nature content of References
the settings in this study. Mystery has clearly been
the most effective of the Kaplans' predictors for Appleton, J. (1975). The Experience of Landscape. London: John
natural settings (R. Kaplan & S. Kaplan, 1989), but Wiley.
its effectiveness in primarily urban settings is less Appleton, J. (1984). Prospect and refuge re-visited. Landscape
Journal, 3, 91-103.
well established. Second, two of the categories in Daniel, T. C. & Vining, J. (1983). Methodological issues in the as-
this study (Enclosed Settings and Blocked Views) sessment of landscape quality. In 1. Altman & J. F. Wohlwill,
were high in mystery but low in preference (Table Eds., Behavior and the Natural Environment (pp. 39-83). New
2). Such categories will dilute otherwise positive York, NY: Plenum.
relationships between mystery and preference in a Frewald, D. B. (1989). Preferences for older buildings: a psycho-
logical approach to architectural design. Doctoral dissertation,
regression analysis including only category settings. University of Michigan.
Third, safety concerns may override mystery in Gibson, J. J. (1979). The Ecological Approach to Visual Percep-
determining preference reactions for settings such tion. Boston, MA: Houghton Mifflin.
as urban alleys (Herzog & Smith, 1988). It is not Gregory, R. L. (1969). On how little information controls so much
clear why typicality was an effective predictor only behavior. In C. H. Waddington, Ed., Towards a Theoretical
Biology, 2 Sketches Edinburgh: Edinburgh University Press,
in the analysis of category settings. What is more pp. 236-247.
important i s that typicality was negatively related Herzog, T. R. (1984). A cognitive analysis of preference for field-
to preference. As noted in the introduction, typicality and-forest environments. Landscape Research, 9, 10-16.
was ineffective in Herzog's (1989) urban-nature Herzog, T. R. (1985). A cognitive analysis of preference for water-
study, but ratings of 'sense of familiarity' have scapes. Journal of Environmental Psychology, 5, 225-241.
Herzog, T. R. (1987). A cognitive analysis of preference for
generally been positive related to preference. Thus, natural environments: mountains, canyons, deserts. Land-
typicality, as defined in this study, probably is not scape Journal, 6, 140-152.
getting at the same thing as familiarity. It seems Herzog, T. R. (1989). A cognitive analysis of preference for urban-
prudent to withhold judgement on typicality as a nature. Journal of Environmental Psychology, 9, 27-43.
Herzog, T. R., Kaplan, S. & Kaplan, R. (1976). The prediction of
predictor of preference.
preference for familiar urban places. Environment and Behav-
The consistent predictors were coherence, com- ior, 8, 627-645.
plexity, and age. Coherence and complexity impli- Herzog, T. R., Kaplan, S. & Kaplan, R. (1982). The prediction
cate both of the Kaplan's cognitive processes, of preference for unfamiliar urban places. Population and
understanding and exploration, as significant in the Environment: Behavioral and Social Issues, 5, 43-59.
prediction of preferences for urban spaces. In Herzog, T. R. & Smith, G. A. (1988). Danger, mystery, and
environmental preference. Environment and Behavior, 20,
general, it seems that the most preferred urban 320-344.
spaces are well organized and yet contain enough Kaplan, R. & Kaplan, S. (1989). The Experience of Nature: A
diversity to hold one's attention. Coherence has Psychological Perspective. New York, NY: Cambridge Univer-
established a track record second only to mystery as sity Press.
Kaplan, R., Kaplan, S. & Brown, T. (1989). Environmental
an effective predictor of preference (R. Kaplan &
preference: a comparison of four domains of predictors.
S. Kaplan, 1989). This study suggests that in urban Environment and Behavior, 21,509-530.
environments largely devoid of natural elements Kaplan, S. (1975). An informal model for the prediction of
coherence may well be the more important predic- preference. In E. H. Zube, R. O. Brush & J. G. Fabos, Eds.,
tor. Finally, the negative relationship between age Landscape Assessment: Values, Perceptions, and Resources.
and preference in this s t u d y contrasts with the Stroudsburg, PA: Dowden, Hutchinson and Ross, pp. 92-101.
Kaplan, S. (1979). Concerning the power of content-identifying
ineffectiveness of age as a preference predictor methodologies. In T. C. Daniel & E. H. Zube, Eds., Assessing
in Herzog's (1989) s t u d y of u r b a n n a t u r e . amenity resource values, pp. 4-13. USDA Forest Service
Thus, although age may be associated with lower General Technical Report RM-68.
248 T.R. Herzog

Kaplan, S. (1987). Aesthetics, affect, and cognition: environmental schema discrepancy model. Environment and Behavior, 18,
preference from an evolutionary perspective. Environment and 3-30.
Behavior, 19, 3-32. Ruddell, E. J., Gramann, J. H., Rudis, V. A. & Westphal, J. M.
Kaplan, S. & Kaplan, R. (Eds.). (1978). Humanscape: En- (1989). The psychological utility of visual penetration in near-
vironments for People. Belmont, CA: Duxbnry (Division of view forest scenic-beauty models. Environment and Behavior,
Wadsworth) (Ann Arbor, MI: Ulrichs). 21,383-412.
Kaplan, S. & Kaplan, R. (1982). Cognition and Environment: Wike, E. L. (1971). Data Analysis. Chicago, IL: Aldine-Atherton.
Functioning in an Uncertain World. New York, NY: Praeger. Woodcock, D. M. (1982). A functionalist approach to environmental
Kramer, C. Y. (1956). Extension of multiple range tests to group preference. Doctoral dissertation, University of Michigan.
means with unequal numbers of replications. Biometrics, 12, Zube, E. H. (1984). Themes in landscape assessment theory.
307-310. Landscape Journal, 3, 104--110.
Lingoes, J. L. (1972). A general survey of the Guttman-Lingoes Zube, E. H., Sell, J. C. & Taylor, J. G. (1982). Landscape perception:
nonmetric program series. In R. N. Shepard, A. K. Romney & research, application, and theory. Landscape Planning, 9, 1-33.
S. B. Nerlove, Eds., Multidimensional Scaling. New York, NY: Zube, E. H., Simcox, D. E. & Law, C. S. (1987). Perceptual land-
Seminar Press, vol. 1, pp. 52-68. scape simulations: history and prospect. Landscape Journal, 6,
Lynch, K. & Rivkin, M. (1959). A walk around the block. Land- 62-80.
scape, 8, 14-24. Manuscript received 10 September 1991
Purcell, A. T. (1986). Environmental perception and affect: a Revised manuscript received 31 March 1992

You might also like