Content uploaded by Thomas Herzog
Author content
All content in this area was uploaded by Thomas Herzog on Oct 27, 2020
Content may be subject to copyright.
www.elsevier.com/locate/jep
Journal of Environmental Psychology 23 (2003) 159–170
Assessing the restorative components of environments
Thomas R. Herzog*, Colleen, P. Maguire, Mary B. Nebel
Department of Psychology, Grand Valley State University, Allendale, MI 49401, USA
Received 7 September 2001; received in revised form 9 September 2002; accepted 11 November 2002
Abstract
We used a direct rating approach based on definitions of each construct to measure the four components of a restorative
environment proposed by attention restoration theory (ART): being away, extent, fascination, and compatibility. We used the same
approach to measure two criterion variables, perceived restorative potential (PRP) of a setting and preference for the setting, as well
as four additional predictor variables (openness, visual access, movement ease, and setting care). Each participant rated 70 settings,
35 each from urban and natural environments, for only one of the variables. Mean ratings were higher for the natural than the urban
settings for both criterion variables and all four restorative components, with differences significant in all cases except for
fascination. Correlations across settings generally followed the predictions of ART, but collinearity appeared among several sets of
variables, most notably being away and setting category, PRP and preference, and extent and fascination. Despite these problems,
regression analysis showed that being away and compatibility predicted PRP and that the pattern of prediction for PRP and
preference was somewhat different.
r2002 Published by Elsevier Science Ltd.
1. Introduction
Attention restoration theory (ART) holds that
intensive or prolonged use of directed attention, the
kind that requires effort, leads to the fatigue of the
mechanisms that serve it (Kaplan & Kaplan, 1989;
Kaplan, 1995, 2001). The person is then said to be in a
state of mental fatigue. The consequences of mental
fatigue can be serious: inaccuracy, impulsivity, irrit-
ability, and incivility. Recovery of effective functioning
is enabled by settings that have certain key properties
discussed below. Such settings are known as restorative
settings. The benefits of a deeply restorative experience
include clearing away of mental noise, recovery of
directed attention capacity, and enhanced ability to
reflect on issues of importance. Ordinary natural
environments are thought to be especially effective as
restorative settings.
A fairly extensive body of research evidence has
accumulated in support of ART. Much of it has dealt
with the distinctive ability of natural settings to foster
effective functioning and well-being. Earlier work is
reviewed by Kaplan (1995). More recent studies include
Kaplan (2001),Korpela, Hartig, Kaiser, and Fuhrer
(2001),Kuo, Bacaicoa, and Sullivan (1998);Kuo and
Sullivan (2001), and Taylor, Wiley, Kuo, and Sullivan
(1998). There have been studies dealing with the
distinctive benefits of restorative settings (Herzog,
Black, Fountaine, & Knotts, 1997) and, as we shall see
below, with the proposed features of such settings. Some
of the most compelling work has linked the beneficial
effects of nature with its effects on attentional capacity
(Wells, 2000;Taylor, Kuo, & Sullivan, 2001), including
two studies pinpointing a mediating role for directed
attention in the relation between natural settings and
beneficial outcomes (Kuo, 2001;Kuo & Sullivan, 2001).
ART asserts that there are four properties or features
of restorative settings. The first such property or
component is being away. This refers to settings that
call on mental content different from that ordinarily
elicited. The idea is that avoiding well-worn mental
content allows one to avoid the use of directed attention
required to support the activation of such content
(Kaplan, 2001). Thus, fatigued directed attention can
rest. This is the basic rationale for the beneficial effects
of ‘‘getting away from it all.’’ For many people who deal
with and think about urban contents and concerns,
natural settings fulfill the requirement of being away
(Knopf, 1987).
The second proposed component of restorative
settings is extent. A setting has extent if it has sufficient
*Corresponding author.
E-mail address: herzog@gvsu.edu (T.R. Herzog).
0272-4944/03/$ - see front matter r2002 Published by Elsevier Science Ltd.
doi:10.1016/S0272-4944(02)00113-5
content and structure that it can occupy the mind for a
period long enough to allow directed attention to rest.
Such settings are characterized as being ‘‘whole other
worlds’’ (Kaplan, 1995). They engage the mind and
support extended exploration. A setting that is small in
physical scale can have extent if it has sufficient content
and structure to occupy the mind. Japanese gardens are
an excellent example. According to ART, natural
settings are relatively well endowed with extent (Kaplan
& Kaplan, 1989, pp. 190–192;Kaplan, 1995).
The third proposed component of a restorative setting
is fascination. Fascination refers to effortless attention.
A setting that can hold one’s attention without effort
has fascination. Here, however, an important distinction
is drawn. Some settings may be so fascinating that they
rivet one’s attention, leaving little room left for thinking
about other things. Such settings will not allow one to
achieve the deeper benefits of a restorative experience
such as reflection on important issues. A more peaceful
kind of fascination, characterized by a moderate level of
effortless attention coupled with aesthetic beauty in the
setting, will foster a more deeply beneficial restorative
experience. Such a setting is said to have ‘‘soft’’ or
‘‘quiet’’ fascination (Kaplan, 1995;Kaplan, Kaplan, &
Ryan, 1998). Natural settings typically have the optimal
combination of moderate fascination and aesthetic
beauty. Herzog et al. (1997) provide evidence in support
of the distinction between ‘‘hard’’ and ‘‘soft’’ fascina-
tion, their different restorative benefits, and the associa-
tion of soft fascination with natural settings.
The last component of a restorative setting is
compatibility. A setting is compatible if there is a good
fit between an individual’s purposes or inclinations and
the kinds of activities supported, encouraged, or
demanded by the setting. This component is complex
because of its explicit reference to the individual’s goals
and inclinations which are many and can be conceptua-
lized as falling on a continuum ranging from very
general (to move freely, to be able to see clearly) to very
specific (to get gas, to play basketball). Thus, a setting
could be compatible on one level and incompatible on
another. One might also have several inclinations at
roughly the same level, and the setting could be
compatible for some of them but incompatible for
others. Nonetheless, despite this complexity, natural
settings are distinctive for the wide range of activities
they support that coincide with the inclinations of
people who visit them.
Empirical attempts to measure the four components
of a restorative setting and to relate them to the
beneficial effects of such settings have not been as
common as the more general studies reviewed earlier.
One set of studies has been carried out by Hartig,
Korpela, and colleagues (Korpela & Hartig, 1996;
Hartig, Kaiser, & Bowler, 1997;Hartig, Korpela, Evans,
&G
.
arling, 1997), another by Laumann, G .
arling, and
Stormark (2001). Both have produced the same kind of
measuring instrument, a multi-item seven-point Likert-
scale questionnaire. Hartig et al. named their instrument
the Perceived Restorativeness Scale, or PRS. Laumann
et al. have not named their instrument, but following the
title of their published paper, we will refer to it as the
Restorative Components Scale, or RCS.
The research strategy for the development and
validation of both the PRS and the RCS has been the
same. Several Likert-scale items were generated to
measure each of the four components of a restorative
setting. The items had face validity based on the
researchers’ understanding of ART as well as input
from other researchers in the area. After refinement
based on the results of pilot studies (and, in the case of
the PRS, on the results of previously published studies),
the final instrument in both cases consisted of 22 items.
A major difference between the two scales is that the
latest version of the PRS measures four constructs while
the RCS measures five. For the RCS, the being away
component is split into novelty (the perception of the
physical setting and activities within it as different from
the usual) and escape (the psychological feeling of being
away from the usual routines and demands). A second
difference is that the RCS items measuring extent deal
with both the amount of content in the setting (scope)
and the structure or organization of the setting elements
(Laumann et al., 2001). The analogous items on the PRS
deal only with structure, and the construct measured has
been named ‘‘coherence’’ by the authors of the PRS
(Hartig et al., 1997). Both instruments have yielded
satisfactory reliability for their subscales. In both cases,
validation has been achieved by showing that natural
environments have significantly higher scores on each of
the component measures than urban environments and
that this is true for a variety of presentation mediums
(site visit, video tape, color slides, imagining being in the
setting). In addition, correlations between the restora-
tive components and various other variables (e.g.
preference, relaxation, complexity, originality, enclo-
sure, positive and negative affect) have been predicted
and verified.
The study reported here was also an attempt to
develop and validate a useful measure of the four
components of a restorative setting, but the research
strategy was different from the one described above.
There were five major differences in our approach. First,
each of the variables in our study, including the four
restorative components, was measured by a single item
rather than a multi-item scale. That single item was a
five-point rating of each setting based on a definition of
the construct to be measured. The advantage of this
approach is that it requires much less time and effort to
obtain measures of the variables. The challenge is to
devise appropriate definitions for the raters. In our case,
we went through several rounds of refining our
T.R. Herzog et al. / Journal of Environmental Psychology 23 (2003) 159–170160
definitions based on feedback from colleagues deeply
conversant with ART.
Two of the final definitions deserve comment. First,
our definition of being away emphasized the psycholo-
gical aspect of being away from everyday thoughts and
concerns. This is closer to the escape component of
being away as measured by Laumann et al. (2001) than
to the novelty component. Given that their escape
component yielded results more in conformity with the
predictions of ART than did their novelty component,
this seemed like a wise choice on our part. Second, given
our earlier analysis of compatibility, we had to choose at
what level of generality to focus our definition of this
construct. We opted for a very high level of generality,
knowing full well that this would leave much unfinished
business for future research. Our line of reasoning was
that we wanted a definition that could apply to a very
broad array of settings and that was not closely tied to
specific goals one might pursue in a setting. In this way,
we hoped to maximize our chances of obtaining the
predicted positive relation between compatibility and
restoration. Thus, our definition focused on how
comfortable and at ease the setting made the rater feel.
The second major difference in our research approach
was that each participant rated the settings for only one
construct rather than all of them. The obvious
advantage in this approach is that it greatly reduces
the likelihood of halo effects in the ratings. The third
major difference in our approach follows directly from
the first two. With each participant responsible for a
single construct and a single item per construct, we were
able to include far more settings than in previous studies
in this area. The maximum number of settings rated was
eight in the Hartig et al. studies and five in the Laumann
et al. study. In contrast, we had 70 settings, 35 each from
urban and natural environments. With so many settings,
there is very little likelihood that any of our results
might be influenced by the peculiarities of individual
settings.
The fourth major difference in our approach is that
we used settings as the units of analysis rather than
raters. Consider what this means for a significant
positive correlation between extent and relaxation. With
raters as units of analysis and results presented
separately for each setting, as was the case in the earlier
measurement studies, the positive correlation means
that raters tend to provide similar ratings for both
variables. One might infer that a setting high in extent
tends also to be seen as high in relaxation, but the
inference is indirect because the covariation is across
raters within settings and not across settings. With
settings as units of analysis, the same positive correla-
tion tells us directly that settings seen as high in extent
tend also to be seen as high in relaxation. Given that the
predictions of ART seem to be about the features of
settings that are important for restoration, there would
seem to be an advantage in looking directly at
covariation across settings. In any event, comparison
of results with settings as units to results with raters as
units seems likely to be insightful.
The last major difference in our approach is that we
obtained a direct measure of the perceived restorative-
ness of the settings. No one has yet attempted to do this.
There have been measures of a host of related constructs
(preference, relaxation, etc.), but none of the previous
studies attempting to measure the four components of
restoration also attempted to measure directly the
perception of restorative potential. We must concede
at once that our measure might not be equivalent to
actual restorativeness. Our measure was based on the
scenario approach of Herzog et al. (1997) which
produced results in agreement with the predictions of
ART, thus providing some confidence in the validity of
the measure. In that approach, the research participant
is invited to recall an occasion when he or she had
worked hard on a project that required intense and
prolonged effort, finally reaching the point where the
ability to work effectively had started to decline and a
break was needed. The participant is then instructed to
rate each setting on how good a place it would be for
taking that break and restoring the ability to work
effectively. With this approach, the measure of perceived
restorative potential, or PRP, is treated as a criterion
variable. Thus, part of the process of validating the
measures of the predicted components of restorativeness
is to see if they have positive relations with the PRP
criterion measure.
Finally, in addition to the measures of the compo-
nents of restoration and of PRP, we also measured five
other variables. One was preference, and the purpose
was to see how it related to our PRP criterion measure.
On theoretical grounds, one would expect a substantial
(but ideally not excessive) positive correlation. There are
two reasons for this expectation. First, ART emphasizes
the importance of soft fascination in the restorative
process. The aesthetic-beauty component of soft fasci-
nation should produce a positive correlation between
preference and restoration. Second, it seems reasonable
that preference, although conceptually distinct from
restoration, might play a role in attracting people to
restorative environments and in keeping them in such
environments for a longer time than would otherwise be
the case.
The other four measured variables may be viewed as
exploratory. Visual and locomotor access to a setting
were assessed by the variables openness (how wide open
the space in the setting appears to be), visual access (how
well one can see into all parts of the setting without
having one’s view blocked), and movement ease (how
easy it would be to move within or through the setting).
Variables like these have been found to be effective
predictors of preference and danger reactions in
T.R. Herzog et al. / Journal of Environmental Psychology 23 (2003) 159–170 161
previous research (e.g. Herzog & Miller, 1998;Herzog &
Chernick, 2000). The final exploratory variable was
perceived setting care. Both theory (e.g. Nassauer, 1995)
and previous research (e.g. Herzog & Miller, 1998;
Herzog & Chernick, 2000) have implicated this kind of
variable in accounting for preference and security
reactions.
For all four of these exploratory variables, we wanted
to know whether the predicted components of restora-
tion operate differently after these variables have been
added to the analysis as compared to before. If the
influence of the restorative components is substantially
weakened in the presence of the exploratory variables,
then it is possible that the relation between the
restorative components and either target variable (PRP
or preference) is spurious. That is, the relation between
the restorative components and either target variable
occurs only because the exploratory variables are related
to both the restorative components and the target
variables. On the other hand, if the influence of the
restorative components is essentially unchanged by the
presence of the exploratory variables, then any effects of
the restorative components cannot simply be reduced to
the influence of perceived access or care. It is also
possible that these exploratory variables might have a
different pattern of relations to the two criterion
variables, PRP and preference. If so, the distinctiveness
of the two constructs is supported. If not, one would
have to look elsewhere for evidence to support the
discriminability of the two constructs.
2. Method
2.1. Participants
The sample consisted of 512 undergraduate students
(166 males, 346 females) at a university in the
Midwestern United States. Participation fulfilled a
course requirement for introductory psychology.
Thirty-two sessions consisting of from 8 to 25 partici-
pants were run.
2.2. Stimuli
The settings consisted of 70 color slides, 35 from each
of two a priori categories: urban and field/forest natural
environments.
1
We sampled broadly with respect to
such basic features as openness, smoothness of ground
texture, and setting care but avoided extremes of social
desirability such as back alleys, churches, and water-
scapes. Some sample scenes from each of the two
categories are presented in Figs. 1–4. No settings
contained people. All were photographed in summer
or early fall. All slides were oriented horizontally.
2.3. Measures
All participants in each session rated each of the 70
settings on only one of the 10 measured variables. All
ratings used a five-point rating scale ranging from A
(‘‘very high (highest possible rating)’’) to E (‘‘not at all
(lowest possible rating)’’). The letters A–E were later
converted to the numbers 5–1, respectively, for analysis.
The major criterion variable was PRP, defined as
follows: ‘‘Recall one of those times when you worked
hard on a project that required intense and prolonged
effort. Remember how it felt. You probably reached a
point where you could tell that your ability to work
effectively had started to decline and that you needed a
break. You needed to do something during the break
that would restore your ability to work effectively on the
project. Put yourself in that mind set now and then
please rate each of the settings you will be shown on how
good a place you think it would be to take a break and
restore your ability to work effectively on the project.’’
For comparison, the common criterion variable Pre-
ference was also included and defined as follows: ‘‘How
much do you like the setting? This is your own personal
degree of liking for the setting, and you do not have to
worry about whether you are right or wrong or whether
you agree with anybody else.’’ The primary set of
predictor variables consisted of the four components of
a restorative setting as specified by ART. Being Away
was defined as follows: ‘‘Sometimes even when you are
very near home it can feel like you are far away from
everyday thoughts and concerns. How much does the
setting have that feeling of being away?’’ The definition
of Extent was ‘‘Sometimes even a small setting can feel
like a whole world of its own. It can seem like there is
enough room to get completely involved in the setting
and not even think about anything else. How much does
the setting seem like such a ‘whole other world’?’’
Fascination was ‘‘How much does the setting draw your
attention without any effort on your part? How much
does it easily and effortlessly engage your interest?’’ Our
broad generality approach to Compatibility yielded the
following definition: ‘‘Settings can either help you feel
comfortable and at ease or they can make it hard to do
so. How much does it seem like the setting would make
it easy for you to feel comfortable and at ease?’’ We also
had a secondary set of four predictors variables.
Openness was defined as ‘‘How wide-open is the
space in the setting?’’ Visual Access was ‘‘How easy is
it to see into this setting? How well can you see all parts
1
We factor analysed the raw preference ratings (principal-axis
factoring, varimax rotation). With a two-factor solution and a cutoff
for factor loadings of |0.40| on one factor only, we obtained a perfect
recovery of the 35 settings in the urban a priori category. For the
nature a priori category, we recovered 33 of the 35 settings. The
remaining two settings, although they failed to meet our cutoff, had
their highest loadings, 0.34 and 0.37, on the nature factor.
T.R. Herzog et al. / Journal of Environmental Psychology 23 (2003) 159–170162
of this setting without having your view blocked or
interfered with?’’ Movement Ease was ‘‘How easy would
it be to move within or through this setting?’’ Finally,
Setting Care was ‘‘How well cared-for does the setting
seem to be? Is it in good condition and well main-
tained?’’
2.4. Procedure
Sessions proceeded as follows. After explaining the
task and obtaining informed consent, four practice
slides were presented to help participants get used to the
task and their instructions for responding. Then
Fig. 1. Urban settings with the highest mean ratings for PRP. Mean ratings are 3.43 (upper left), 3.15 (upper right), 2.64 (lower left),
and 2.61 (lower right).
Fig. 2. Urban settings with the lowest mean ratings for PRP. Mean ratings are 1.29 (upper left), 1.53 (upper right), 1.56 (lower left),
and 1.57 (lower right).
T.R. Herzog et al. / Journal of Environmental Psychology 23 (2003) 159–170 163
participants rated 74 slides, presented in two sets of 37
slides each with a 2-min rest between sets. The first and
last slide within each set were fillers. The remaining 70
slides yielded the data for analysis. These slides were
presented in one of two orders. The first order was used
for the first 16 sessions, the second order for the last 16
sessions. Within each block of sessions using a given
slide order, there were four sessions devoted to PRP and
to preference and one session devoted to each of the
other eight rated variables. Aside from the constraints
on the ordering of sessions just noted, the ordering of
sessions was haphazard. One of the slide presentation
orders was generated randomly with the constraint that
no more than three consecutive settings from either a
Fig. 3. Nature settings with the highest mean ratings for PRP. Mean ratings are 4.34 (upper left), 4.33 (upper right), 4.28 (lower left),
and 4.24 (lower right).
Fig. 4. Nature settings with the lowest mean ratings for PRP. Mean ratings are 2.50 (upper left), 2.65 (upper right), 2.71 (lower left),
and 2.74 (lower right).
T.R. Herzog et al. / Journal of Environmental Psychology 23 (2003) 159–170164
priori category were allowed. The second presentation
order was derived by interchanging the halves of the first
order. Viewing time was 15 s/slide in all sessions. Final
sample sizes were 144 for PRP and preference, 33 for
fascination, 30 for openness, 28 for movement ease, 27
for being away, visual access, and setting care, and 26
for extent and compatibility.
3. Results
Unless noted otherwise, all analyses were based on
settings as the units of analysis and setting scores as raw
scores. A setting score is the mean score for each setting
based on all participants who completed one of the
rating tasks. Thus, for each rated variable, every setting
had a setting score.
Internal-consistency reliability was assessed for each
rating variable by computing Cronbach’s alpha, based
on settings as cases and participants as items. We thus
have a form of inter-rater reliability, measuring the
tendency of the raters to agree on their ranking of
the settings. The reliability coefficients for nine of the
variables exceeded 0.90, the sole exception being extent
with a coefficient alpha of 0.85. For both criterion
variables, PRP and preference, coefficient alpha was
0.99. For the other three restorativeness components,
the coefficients were 0.97, 0.91, and 0.92 for being away,
fascination, and compatibility, respectively.
To help the reader form an impression of the setting
features that affect PRP, Figs. 1 and 2 contain the urban
settings with the four highest and lowest mean ratings
for PRP, respectively. Figs. 3 and 4 present the same
information for the nature settings. For the urban
settings, overall signs of setting care and a smooth
ground surface composed of varied textures seem to
enhance PRP, while clutter, automobiles, lack of
visibility, and blocked views seem to detract from
PRP. For the natural settings, open views, smooth
ground surfaces, and signs of setting care seem to be
positive features, while within-forest views high in visual
obstruction as well as open views with a poorly
maintained ground surface seem to be negatively
perceived. Thus, for both categories, maintenance and
visibility seem to be implicated in PRP.
Table 1 presents mean ratings for each rating variable
and each setting category. The table also contains p
values for a test of the significance of the difference
between the two category means and a correlation
measure of the effect size. With a Bonferroni adjusted
alpha of 0.05/10=0.005, the setting category effect was
significant for both criterion variables and all of the
restorative components except fascination. Each of these
effects was in the expected direction, with the nature
category having the higher mean. The other noteworthy
feature of the table is the extremely high correlation
(0.93) between being away and setting category. Given
the redundancy of these two variables plus the fact that
being away is the variable in which we had greater
interest, in subsequent regression analyses we did not
include setting category as a predictor.
2
Table 2 contains correlations among the rated
variables for all 70 settings and also separately within
each setting category. Several points are of interest.
First, the two criterion variables are very highly
correlated. Second, there are some very high correla-
tions among the four restorative components, which
bodes ill for any attempt to include them all as
predictors in the same regression analysis. The most
alarming correlation is between extent and fascination.
Third, although there are substantial correlations
among the four additional predictors, correlations
between those predictors and all of the other variables
are generally fairly modest. The one exception to this
trend occurs with compatibility, which tends to have
positive correlations with the additional predictors,
especially within the nature category. Fourth, the four
restorative components seem to be the best predictors of
both PRP and preference.
The next step was to see how the predictors worked
together in regression analyses. Having already decided
not to include setting category because of its redundancy
with being away, our strategy for the regression analyses
was to proceed in two steps, first entering the restorative
components as predictors and then entering the four
additional predictors. We first checked the collinearity
diagnostics for this two-step approach to see if there
Table 1
Mean ratings (M) and standard deviations (s.d.) for all rated variables
as a function of setting category
Variables Setting category
Urban Nature pr
Ms.d.Ms.d.
PRP 2.01 0.46 3.45 0.53 o0.001 0.83
Preference 2.36 0.56 3.43 0.49 o0.001 0.72
Being away 1.93 0.40 3.72 0.29 o0.001 0.93
Extent 2.66 0.44 3.38 0.37 o0.001 0.66
Fascination 2.71 0.64 2.98 0.45 0.048 0.24
Compatibility 2.88 0.51 3.66 0.56 o0.001 0.60
Openness 2.50 0.70 3.01 1.04 0.019 0.28
Visual access 3.32 0.68 3.06 0.95 0.187 0.16
Movement ease 2.44 0.53 3.14 0.87 0.085 0.21
Setting care 3.18 0.58 2.99 0.71 0.214 0.15
Note: PRP is ‘‘perceived restorative potential’’; pis the probability of
the difference between the two category means; ris a correlation
measure of the effect size.
2
We checked the collinearity diagnostics for a regression analysis
that included all eight rated predictors plus setting category. As might
be expected from Table 1, one of the sets of variables with
multicollinearity problems was being away and setting category.
T.R. Herzog et al. / Journal of Environmental Psychology 23 (2003) 159–170 165
were problems with multicollinearity among the pre-
dictor variables. According to Tabachnick and Fidell
(1996, p. 87), ‘‘Criteria for multicollinearity are a
conditioning index >30 and at least two variance
proportions >0.50 for a given root number.’’ On the
first step, the diagnostics indicated multicollinearity
involving extent and fascination. Dropping either
predictor eliminated the problem. However, because
we could not decide which predictor to eliminate, we
elected to do parallel sets of analyses, one without extent
and one without fascination. On the second step, the
diagnostics revealed two more sets of variables with
multicollinearity problems. One set consisted of com-
patibility and setting care, the other set of being away,
openness, and visual access. Always favoring the
retention of the restorative components because our
major purpose was to investigate them, we eliminated
setting care from the first set. For the second set, the
choice was between openness and visual access. Because
the simple correlations with the criterion variables for
the entire set of 70 settings seemed more promising for
openness (Table 2, top section), we eliminated visual
access. The final set of five predictors consisted of three
restorative components (with either extent or fascination
eliminated) and the additional predictors openness and
movement ease. Collinearity diagnostics indicated no
problems with this set of predictors.
Tables 3 and 4 contain the results for the regression
analyses with PRP and preference, respectively, as
criterion variables. The top panel of each table contains
the results with extent as a predictor; the bottom
panel contains the results with fascination as a
predictor. Within each panel, the left half contains the
results from the first step of the regression analysis in
which only the restorative components were entered as
predictors. The right half contains the results from the
second step of the analysis in which the two additional
predictors were also entered. With five predictors in each
analysis, alpha was set at the adjusted value of 0.05/
5=0.01.
Table 2
Correlations among the rating variables for all settings and within each setting category
Variables 1 234567 8 9 10
All settings (N¼70)
1. PRP — 0.95** 0.90** 0.81** 0.57** 0.86** 0.33* 0.00 0.00 0.24
2. Preference — 0.84** 0.88** 0.75** 0.86** 0.31* 0.06 0.06 0.26
3. Being away — 0.78** 0.46** 0.64** 0.25 0.14 0.22 0.10
4. Extent — 0.80** 0.70** 0.36* 0.09 0.06 0.02
5. Fascination — 0.58** 0.17 0.18 0.08 0.18
6. Compatibility — 0.49** 0.29 0.21 0.50**
7. Openness — 0.79** 0.63** 0.16
8. Visual access — 0.64** 0.32*
9. Movement ease — 0.41**
10. Setting care —
Urban settings (N¼35)
1. PRP — 0.88** 0.74** 0.58** 0.67** 0.69** 0.13 0.07 0.03 0.56**
2. Preference — 0.66** 0.77** 0.88** 0.73** 0.01 0.10 0.04 0.50**
3. Being away — 0.48* 0.59** 0.30 0.04 0.04 0.16 0.21
4. Extent — 0.90** 0.47* 0.34 0.30 0.15 0.01
5. Fascination — 0.55** 0.13 0.22 0.09 0.19
6. Compatibility — 0.10 0.21 0.04 0.69**
7. Openness — 0.78** 0.71** 0.37
8. Visual access — 0.53** 0.10
9. Movement ease — 0.20
10. Setting care —
Natural settings (N¼35)
1. PRP — 0.94** 0.52** 0.70** 0.75** 0.91** 0.36 0.44* 0.47* 0.71**
2. Preference — 0.69** 0.78** 0.86** 0.82** 0.28 0.40* 0.27 0.60**
3. Being away — 0.77** 0.86** 0.32 0.03 0.11 0.26 0.04
4. Extent — 0.85** 0.54** 0.18 0.27 0.10 0.31
5. Fascination — 0.60** 0.11 0.27 0.02 0.28
6. Compatibility — 0.61** 0.68** 0.66** 0.79**
7. Openness — 0.94** 0.74** 0.53**
8. Visual access — 0.68** 0.54**
9. Movement ease — 0.70**
10. Setting care —
Note: PRP is ‘‘perceived restorative potential.’’ *po0:01:**po0:001:
T.R. Herzog et al. / Journal of Environmental Psychology 23 (2003) 159–170166
Several points are worth noting. First, extent and
fascination produce identical results, as might be
expected given their intercorrelation. Neither predictor
is effective with PRP as the criterion variable, but either
variable is a positive predictor with preference as the
criterion variable. Second, the other two restorative
components, being away and compatibility, are positive
predictors in all analyses. Third, the partial correlations
suggest compatibility is an equally strong predictor of
both criterion variables, but the other restorative
components seem to differ in their predictive power as
a function of the criterion variable. Being away appears
to be a stronger predictor of PRP than of preference
while the reverse holds true for extent/fascination.
Fourth, the additional predictors, openness and move-
ment ease, are effective predictors of PRP, openness a
negative predictor and movement ease a positive
predictor. With preference as the criterion variable,
openness yields mixed results, while movement ease is
not a significant predictor. Fifth, even though the two
criterion variables are very strongly correlated with each
other, it is clear from the preceding points that the
pattern of effective predictors is not the same for PRP
and preference.
4. Discussion
Because we were attempting a new approach to the
measurement of the restorative components of environ-
ments, our major concern has to be with the reliability
and validity of our measures. As for reliability, with the
lowest internal-consistency coefficient at 0.85 and all but
one of the coefficients exceeding 0.90, there is no cause
for concern. Most of the rest of this discussion speaks in
one way or another to the issue of validity.
Several aspects of our results deserve comment in this
regard. First, the correlations among the criterion
variables and the restorative-component measures were
generally very strong, so much so that we had to struggle
with redundancy issues at several points in our data
analyses. For comparison, the reader is referred to the
Table 3
Regression of PRP on the predictor set including extent (top panel) or fascination (bottom panel)
Predictor Bprp Bprp
Being away 0.50 0.82 o0.001 0.55 0.86 o0.001
Extent 0.10 0.15 0.227 0.15 0.25 0.041
Compatibility 0.61 0.80 o0.001 0.61 0.81 o0.001
Openness 0.17 0.49 o0.001
Movement ease 0.18 0.44 o0.001
Adjusted R2¼0:94;po0:001 Adjusted R2¼0:95;po0:001
Being away 0.53 0.88 o0.001 0.59 0.90 o0.001
Fascination 0.06 0.13 0.304 0.07 0.17 0.169
Compatibility 0.62 0.80 o0.001 0.61 0.79 o0.001
Openness 0.16 0.46 o0.001
Movement ease 0.18 0.42 o0.001
Adjusted R2¼0:94;po0:001 Adjusted R2¼0:95;po0:001
Note:Bis the raw-score regression weight and pr is the partial correlation.
Table 4
Regression of preference on the predictor set including extent (top panel) or fascination (bottom panel)
Predictor Bprp Bprp
Being away 0.21 0.52 o0.001 0.19 0.49 o0.001
Extent 0.51 0.61 o0.001 0.54 0.67 o0.001
Compatibility 0.48 0.73 o0.001 0.56 0.78 o0.001
Openness 0.11 0.34 0.005
Movement ease 0.00 0.00 0.991
Adjusted R2¼0:92;po0:001 Adjusted R2¼0:93;po0:001
Being away 0.34 0.83 o0.001 0.34 0.80 o0.001
Fascination 0.44 0.77 o0.001 0.43 0.76 o0.001
Compatibility 0.44 0.87 o0.001 0.48 0.76 o0.001
Openness 0.06 0.23 0.062
Movement ease 0.02 0.06 0.606
Adjusted R2¼0:95;po0:001 Adjusted R2¼0:95;po0:001
Note:Bis the raw-score regression weight and pr is the partial correlation.
T.R. Herzog et al. / Journal of Environmental Psychology 23 (2003) 159–170 167
tables of correlations in Laumann et al. (2001) in which
the correlations are generally more modest. We cannot
be sure why this happened; there were probably a
number of contributing factors. We deal with defini-
tional issues below. One factor that we suspect is
operating is our use of settings as the units of analysis
as compared to raters in previous studies. We suspect
that the range of true-score variation is greater with
settings as units than with raters as units. If so, then the
potential for stronger covariation would exist.
It has become standard in developing measures of the
restorative components of environments to validate
one’s measures by comparing means scores for urban
and natural settings. We followed this emerging tradi-
tion and obtained generally positive results. For both
criterion variables and all four measures of the
restorative components, the means were in the predicted
direction (nature higher), and all of the comparisons
were significant except the one for fascination (Table 1).
In the case of fascination, we confess that in retrospect
we think we went after the wrong construct. Our
definition in terms of ‘‘How much does the setting draw
your attention without any effort on your part’’ seems
aimed at fascination in general, whereas the kind of
fascination that is supposed to be especially beneficial,
according to ART, is soft fascination. This may not be
the only problem, but we strongly suspect that a
definition worded in terms of ‘‘gentle’’ or ‘‘peaceful’’
attraction of interest would yield the predicted signifi-
cant difference between urban and natural settings. We
see this as matter for future research.
Returning to the redundancies among our measures,
the most troublesome one among the measures of the
restorative components was the very high correlation
between extent and fascination. This correlation was
strong enough to exceed the criterion for multicollinear-
ity and prevent us from using both variables as
predictors in the same analysis. Not surprisingly, in
separate analyses, both variables yielded the same
pattern of results (Tables 3 and 4). Here again we think
the main problem was definitional. The definitions for
both variables were worded in such a way that they
could have been interpreted as ‘‘How interesting is this
setting?’’ We feel a need to maintain user friendliness in
our definitions while at the same time emphasizing the
distinctive features of each construct. We have com-
mented above on the direction in which we need to take
the definition of fascination. For extent, we need to find
a way to emphasize the scope and connectedness aspects
of the construct rather than the potential for getting
‘‘completely involved in the setting.’’ Particulars aside, it
is clear that our present results emphasize the need for
more distinctive definitions of extent and fascination.
A related problem is that neither fascination nor
extent was effective as a predictor of PRP even though
the two predictors were never included together in the
same analysis. If this pattern continued to hold with
improved definitions of the predictors and over several
studies using settings as units of analysis, it would be
damaging to ART, which clearly specifies four distinct
and effective component predictors.
It should be noted that the issue of positive
correlations among the restorative components becomes
problematic only when they are very high. Modest
positive correlations among the components are to be
expected on theoretical grounds. The Kaplans provide
several examples (Kaplan & Kaplan, 1989, pp. 182–
195). To cite one such example, fascination is not likely
to be sustained very long by ‘‘random sequences of
interesting objects’’ (p. 185). Connection to a larger
framework is helpful. ‘‘Thus, fascination and extent are
mutually supportive’’ (p. 185). The implication of their
entire discussion is that the restorative components are
not orthogonal predictors but rather are positively
correlated, yet distinct, constructs.
Another troubling redundancy in our results was the
one between the two criterion variables. One obvious
interpretation is that when raters are asked to evaluate a
setting in terms of how good a place it would be for
taking a break from an attentionally demanding task,
they have a strong tendency to evaluate the setting in
terms of how much they like it. If so, this would call into
question the discriminant validity of our PRP measure.
There should have been substantial overlap between the
two measures but not almost complete overlap. On this
view, one must despair of finding a distinctive definition
of PRP for rating purposes because it is hard to imagine,
on grounds of face validity, how we could have devised a
more distinctive measure than the one we used. A
second interpretation is that perhaps the high correla-
tion is domain specific and would not appear with
different setting categories. Further research would be
necessary to address this possibility. A third interpreta-
tion is that the construct we measured, PRP, is not the
same as the actual restorative potential of a setting. The
two constructs might be positively related and yet
sufficiently distinct that their relations with other
variables might be different. Thus, it could be the case
that preference and actual restorative potential might
not have an excessively high correlation with each other.
If this line of thought has merit, it highlights the urgency
of finding out what measures do indeed correlate with
actual restorative potential in future research.
Even though the two criterion variables were largely
redundant, there may still be some basis for distinguish-
ing between them in terms of their pattern of relations
with the restorative components. The partial correla-
tions in Tables 3 and 4 suggest that being away may be a
more potent predictor of PRP than of preference and
that the reverse may be true for extent/fascination.
Thus, even if ratings of PRP are largely a matter of
preference, as suggested in the previous paragraph, the
T.R. Herzog et al. / Journal of Environmental Psychology 23 (2003) 159–170168
relative potency of the restorative components as
predictors suggests that the raters like the settings for
different reasons when given an attentional-fatigue set
versus a preference set. If the focus is on recovering from
fatigue, being away becomes strongly salient; if the focus
is on preference in general, extent/fascination is more
important. The possibility that the restorative compo-
nents may differ in the relative potency of their
predictive power in the presence of different rated
criterion variables is very intriguing and worthy of
further investigation. That different outcomes may be
associated with different rating-instruction sets is also
compatible with the findings of Herzog et al. (1997) and
of Staats, Kieviet, and Hartig (2003).
Two final issues concerning our predictor variables
deserve comment. First, as noted earlier, compatibility
was defined at a very general level. We could be justly
criticized for making no explicit reference in our
definition to the fit between an individual’s purposes
or inclinations and the kinds of activities supported or
demanded by a setting. We can only acknowledge this
criticism and observe that compatibility is a sufficiently
complicated construct that it deserves a research
program of its own.
Second, the conceptual and empirical roles of the
additional predictor variables (openness, visual access,
movement ease, and setting care) in our analytic
approach might be viewed as less than satisfying.
Conceptually, we treated these predictors as exploratory
and made sure that the role of the ART components was
assessed before these additional predictors were entered
into the analysis. We did this because we could find no
clear mandate in ART for what role these predictors
might play and yet they seemed relevant because of their
past effectiveness in predicting preference. Given the
weakness of our conceptual rationale for these pre-
dictors, one could entertain alternative roles for these
kinds of predictors (landscape qualities) in future
research on ART. For example, one might explore
whether the effect of a specific landscape quality on
restoration is mediated by one or more of the ART
components.
Empirically, the exploratory variables had only
limited impact. This was necessarily the case because
they were entered last in the regression analysis, and the
restorative components had already accounted for at
least 92% of the variance of the criterion variables. The
exploratory variables were able to account for only an
additional 1% at best. Given our analytic strategy, the
most important observation about the exploratory
variables is that their presence did not alter the pattern
of relations between the restorative components and the
criterion variables. Thus, the restoration components
are related to the criterion variables independently of the
exploratory variables, and there is no evidence that the
associations between the restoration components and
the criterion variables are spurious. Meanwhile, given
the miniscule amount of additional variance accounted
for by the exploratory predictors, we are disinclined to
make much of the pattern of their partial correlations.
This is especially true in the case of the negative partial
correlation between openness and the criterion vari-
ables.
In conclusion, we view the results of this first study
using the single-item rating approach to measure the
restorative components of environments as a ‘‘mixed
bag.’’ The results were generally supportive, and the
convenience of the measures, along with the potential to
evaluate a large number of settings in a reasonable time,
justifies further development work. Furthermore, the
focus on covariation across settings seems closer to the
spirit of the predictions of ART than a focus on
covariation across raters. We did encounter some
serious problems with redundancies among our mea-
sures. In the case of extent and fascination, we have
suggested some approaches for alternative measures of
the constructs. In the case of the primary criterion
variable, it seems apparent that there is a crucial need
for a comparison of PRP with a direct measure of actual
restorative potential, presumably embodied in a mea-
sure of actual restorative outcome. Only then can the
validity of rating approaches to perceived environmental
restorativeness be properly evaluated.
Acknowledgements
We thank Stephen and Rachel Kaplan for their
helpful suggestions about wording definitions. The
responsibility for the final products is, however, entirely
ours.
References
Hartig, T., Kaiser, F. G., & Bowler, P. A. (1997). Further development
of a measure of perceived environmental restorativeness. Working
Paper No. 5, Uppsala University, Institute for Housing Research,
G.
avle, Sweden.
Hartig, T., Korpela, K., Evans, G. W., & G .
arling, T. (1997). A
measure of restorative quality in environments. Scandinavian
Housing and Planning Research,14, 175–194.
Herzog, T. R., Black, A. M., Fountaine, K. A., & Knotts, D. J. (1997).
Reflection and attentional recovery as distinctive benefits of
restorative environments. Journal of Environmental Psychology,
17, 165–170.
Herzog, T. R., & Chernick, K. K. (2000). Tranquility and danger in
urban and natural environments. Journal of Environmental
Psychology,20, 29–39.
Herzog, T. R., & Miller, E. J. (1998). The role of mystery in perceived
danger and environmental preference. Environment and Behavior,
30, 429–449.
Kaplan, R. (2001). The nature of the view from home: Psychological
benefits. Environment and Behavior,33, 507–542.
T.R. Herzog et al. / Journal of Environmental Psychology 23 (2003) 159–170 169
Kaplan, S. (1995). The restorative benefits of nature: Toward an
integrative framework. Journal of Environmental Psychology,15,
169–182.
Kaplan, S. (2001). Meditation, restoration, and the management of
mental fatigue. Environment and Behavior,33, 480–506.
Kaplan, R., & Kaplan, S. (1989). The experience of nature: A
psychological perspective. New York: Cambridge University Press.
Kaplan, R., Kaplan, S., & Ryan, R. L. (1998). With people in mind:
Design and management of everyday nature. Washington, DC:
Island Press.
Knopf, R. (1987). Human behavior, cognition, and affect in the natural
environment. In D. Stokols, & I. Altman (Eds.), Handbook of
environmental psychology, Vol. 1. (pp. 787–825). New York: Wiley.
Korpela, K., & Hartig, T. (1996). Restorative qualities of favorite
places. Journal of Environmental Psychology,16, 221–233.
Korpela, K. M., Hartig, T., Kaiser, F. G., & Fuhrer, U. (2001).
Restorative experience and self-regulation in favorite places.
Environment and Behavior,33, 572–589.
Kuo, F. E. (2001). Coping with poverty: Impacts of environment and
attention in the inner city. Environment and Behavior,33, 5–34.
Kuo, F. E., & Sullivan, W. C. (2001). Environment and crime in the
inner city: Does vegetation reduce crime? Environment and
Behavior,33, 343–367.
Kuo, F. E., Bacaicoa, M., & Sullivan, W. C. (1998,). Transforming
inner-city landscapes: Trees, sense of safety, and preference.
Environment and Behavior,30, 28–59.
Laumann, K., G.
arling, T., & Stormark, K. M. (2001,). Rating scale
measures of restorative components of environments. Journal of
Environmental Psychology,21, 31–44.
Nassauer, J. I. (1995). Messy ecosystems, orderly frames. Landscape
Journal,14, 161–170.
Staats, H., Kieviet, A., & Hartig, T. (2003). Where to recover
from attentional fatigue: An expectancy-value analysis of environ-
mental preference. Journal of Environmental Psychology,23,
147–157.
Tabachnick, B. G., & Fidell, L. S. (1996). Using multivariate statistics
(3rd ed.). New York: HarperCollins.
Taylor, A. F., Kuo, F. E., & Sullivan, W. C. (2001). Coping with
ADD: The surprising connection to green play settings. Environ-
ment and Behavior,33, 54–77.
Taylor, A. F., Wiley, A., Kuo, F. E., & Sullivan, W. C. (1998).
Growing up in the inner city: Green spaces as places to grow.
Environment and Behavior,30, 3–27.
Wells, N. M. (2000). At home with nature: Effects of ‘‘greenness’’ on
children’s cognitive functioning. Environment and Behavior,32,
775–795.
T.R. Herzog et al. / Journal of Environmental Psychology 23 (2003) 159–170170
