Journal of Toxicology and Environmental Health, Part B

Critical Reviews

ISSN: 1093-7404 (Print) 1521-6950 (Online) Journal homepage: https://www.tandfonline.com/loi/uteb20

Attention Restoration Theory II: a systematic

review to clarify attention processes affected by
exposure to natural environments

Matt P. Stevenson, Theresa Schilhab & Peter Bentsen

To cite this article: Matt P. Stevenson, Theresa Schilhab & Peter Bentsen (2018) Attention
Restoration Theory II: a systematic review to clarify attention processes affected by exposure to
natural environments, Journal of Toxicology and Environmental Health, Part B, 21:4, 227-268, DOI:
10.1080/10937404.2018.1505571

To link to this article: https://doi.org/10.1080/10937404.2018.1505571

View supplementary material Published online: 21 Aug 2018.

Submit your article to this journal Article views: 3099

View related articles View Crossmark data

Citing articles: 55 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=uteb20
JOURNAL OF TOXICOLOGY AND ENVIRONMENTAL HEALTH, PART B
2018, VOL. 21, NO. 4, 227–268
https://doi.org/10.1080/10937404.2018.1505571

Attention Restoration Theory II: a systematic review to clarify attention

processes affected by exposure to natural environments
Matt P. Stevensona, Theresa Schilhabb, and Peter Bentsena,c
a
Centre for Outdoor Recreation and Education, University of Copenhagen, Fredensborg, Denmark; bFuture Technology, Culture, and
Learning, Department of Education, University of Aarhus, Copenhagen, NV, Denmark; cHealth Promotion Research, Steno Diabetes Center
Copenhagen, Gentofte, Denmark

ABSTRACT KEYWORDS
Attention Restoration Theory (ART) predicts exposure to natural environments may lead to Executive functions;
improved cognitive performance through restoration of a limited cognitive resource, directed greenspace; mental fatigue;
attention. A recent review by Ohly and colleagues (2016) uncovered substantial ambiguity sur- restorative environments
rounding details of directed attention and how cognitive restoration was tested. Therefore, an
updated systematic review was conducted to identify relevant cognitive domains from which to
describe elements of directed attention sensitive to the restoration effect. Forty-two articles that
tested natural environments or stimuli against a suitable control, and included an objective measure
of cognitive performance, had been published since July 2013. Articles were subjected to screening
procedures and quality appraisal. Random effects meta-analyses were performed to calculate
pooled effect sizes across 8 cognitive domains using data from 49 individual outcome measures.
Results showed that working memory, cognitive flexibility, and to a less-reliable degree, attentional
control, are improved after exposure to natural environments, with low to moderate effect sizes.
Moderator analyses revealed that actual exposures to real environments may enhance the restora-
tion effect within these three domains, relative to virtual exposures; however, this may also be due
to differences in the typical lengths of exposure. The effect of a participants’ restoration potential,
based upon diagnosis or fatigue-induction, was less clear. A new framework is presented to qualify
the involvement of directed attention-related processes, using examples of tasks from the three
cognitive domains found to be sensitive to the restoration effect. The review clarifies the description
of cognitive processes sensitive to natural environments, using current evidence, while exploring
aspects of protocol that appear influential to the strength of the restoration effect.

Introduction environments has rapidly expanded counter-

movements that promote regular use of natural
From a flashing advertisement that reminds us we
environments to maintain a healthy lifestyle. The
are hungry, to a beeping phone that keeps us
expansion of nature-based programs in education
updated; modern urban environments contain dis-
(Barfod et al. 2016) and the suggestion that health-
tractors that are designed to be distracting. To
care providers should be offering prescriptions for
navigate such environments, the brain must rely
exposure to natural environments as a viable inter-
heavily on cognitive resources that direct attention
vention for a range of conditions (Zarr, Cottrell,
towards a desired goal while simultaneously ignor-
and Merrill 2017), created an urgency to under-
ing distraction. However, the capacity to direct
stand the mechanisms underlying the often
attention in such a way cannot be sustained inde-
reported positive effects on brain activity.
finitely, which contributes to the waxing and wan-
In 2016, Ohly and colleagues conducted a sys-
ing of our ability to concentrate throughout the
tematic review and meta-analysis of the objective
day. Prolonged use of this cognitive resource leads
evidence available on the effects natural environ-
to a state of mental fatigue that needs to be over-
ments have on cognitive processes, following the
come. The increasing cognitive demand of modern
leading theory in the area, Attention Restoration

CONTACT Matt P. Stevenson [email protected] Centre for Outdoor Recreation and Education, University of Copenhagen, Noedebovej 77A, 3480,
Fredensborg, Denmark
Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/uteb.
Supplemental data for this article can be accessed here.
© 2018 Taylor & Francis
 228

Table 1. Characteristic of studies identified by systematic literature search, ordered by: randomized crossover trials, randomized controlled trials, quasi-experimental studies, and
natural experiments.
Author, Year, (sub- Study characteristics: population; Intervention and control characteristics: exposure type;
study) Study design Country, setting n gender; age activity; subgroup/environments; duration Cognitive performance measure
Geniole et al. (2016) RCT (w) Canada; naturalised 31 University students; male only; Actual; walking; naturalised landfill vs. business/ Stroop Task
landfill, urban area m = 24.61 years commercial urban area (some natural stimuli); 15ʹ00
Jung et al. (2017) RTC (w) United States; 40 Heart failure patients, healthy Virtual; viewing images; nature scenery vs. urban views; Multi-Source Interference Task;
laboratory matched adults; 47.5% female; 07ʹ00 Digit Span Forwards; Digit Span
m = 59.2 years Backwards; Trail Making Test A,
B; StroopTest
M. P. STEVENSON ET AL.

Rider and Bodner RCT (w) Canada; university 24 University students; 83.3% Actual; walking; natural vs. urban vs. indoor; 10ʹ00 Free Recall Memory (words);
(2016) (1) campus female; m = 22 years Forced-Choice Recognition
Memory
Rider and Bodner RCT (w) Canada; university 24 University students; 79.2% Actual; walking; natural vs. urban vs. indoor; 10ʹ00 Free Recall Memory (words);
(2016) (2) campus female; m = 22 years Forced-Choice Recognition
Memory
Rogerson et al. (2016) RCT (w) United Kingdom; 24 General population (n = 13), Actual; cycling; view of sportsfield, tress vs. indoor view; Digit Span Backwards
universty campus university staff (n = 1), university 15ʹ00
students (n = 10); 79.2% female;
m = 35.1 years
Sahlin et al. (2016) RCT (w) Sweden; urban 51 General population; 76.5% Actual; guided relaxation; outdoor natural vs. Indoor; Necker Cube Pattern Control
areas, nature female; m = 45 years 30ʹ00
reserve
Schutte, Torquati, and RCT (w) United States; urban 67 Children (four age groups); 52.2% Actual; walking; natural vs. urban; 20ʹ00 Spatial Working Memory Task;
Beattie (2017) areas female; m = 4.53, m = 5.48, Go/NoGo Task; Continuous
m = 7.4 years, m = 8.5 years Performance Task; Digit Span
Backwards
Sonntag-Öström et al., RCT (w) Sweden; city and 20 Patients with Exhaustion Actual; walking, relaxing; forest by lake vs. rock outcrop, Necker Cube Pattern Control
(2014) forest areas Disorder; female only; vs. spruce forest vs. city; 10ʹ00, 40ʹ00
m = 41.6 years
Triguero-Mas et al. RCT (w) Spain; Collserola NP, 26 General population adults, lowest Actual; national park, beach vs. neighbourhood Digit Span Backwards
(2017) Castelldefels Beach, 50th percentile scores on MHI-5; environment; 30ʹ00, 180ʹ00
Eixample 57.69% female;
(neighbourhood) median = 44.32 years
Berry et al. (2014) RCT (b) United States; 185 University students; 58% female; Virtual; viewing images; natural vs. urban vs. geometric; Delayed Discounting Task
laboratory m = 20.88 years EDTP
Berry et al. (2015) RCT (b) United States; 43 University students; 60.5% Virtual; viewing images; natural vs. urban; EDTP Delayed Discounting Task;
laboratory female; m = 22.5 years Interval Bisection Task
Beute and De Kort RCT (b) The Netherlands; 50a University students; 43.3% Virtual; viewing images; natural vs. urban; 03ʹ00 Stroop Task; Operation Span
(2014), (1) laboratory female; m = 22.2 yearsa Task;
Beute and De Kort RCT (b) The Netherlands; 61b University students; 42.1% Virtual; viewing images; natural vs. urban; 03ʹ00 2-Back Task
(2014), (2) laboratory female; m = 21.1 yearsb
Bratman et al. (2015) RCT (b) United States; 60 General population; 55% female; Actual; walking, taking photos; urban park vs. busy Operation Span Task; Change
university surrounds m = 22.9 years street; 50ʹ00 Detection Task; Attention
Network Task; Digit Span
Backwards
(Continued )
Table 1. (Continued).
Author, Year, (sub- Study characteristics: population; Intervention and control characteristics: exposure type;
study) Study design Country, setting n gender; age activity; subgroup/environments; duration Cognitive performance measure
Chaw and Lau (2015), RCT (b) Hong Kong, 42 University students; 66.7% Virtual; viewing images, imagination; natural vs. non- Unsolvable Anagrams
(1) laboratory female; m = 20.81 years viewing (resting); 06ʹ00
Chaw and Lau (2015), RCT (b) Hong Kong; 58 University students; 62.1% Virtual; viewing images, imagination; natural vs. urban; Logical Reasoning - Graduate
(2) laboratory female; m = 19.79 years 04ʹ00 Exam Record
Chaw and Lau (2015) RCT (b) United States; 185b General population; 66% female; Virtual; viewing images, clicking on attractive areas; Unsolvable Anagrams
(3) laboratory m = 38.58 yearsb natural vs. urban; 01ʹ30
Craig et al. (2015) RCT (b) United States; 48 University studentsc; 45.8% Virtual; viewing images; natural vs. urban; 04ʹ10 Sustained Attention to
laboratory female; m = 20.1 years Response Task
Emfield and Neider RCT (b) United States; 192 University students; 66.1% Virtual; viewing images/listening to sounds/both; Digit Span Backwards; Attention
(2014) laboratory female; m = 19.8 years natural vs. Urban; 07ʹ00 - 10ʹ00 Network Task; Functional Field
of View
Evensen et al. (2015) RTC (b) Norway; laboratory 85 University students; 67.1% Actual; viewing; window + plants vs. window Reading Span Task
female; m = 24.9 years + inanimate objects vs. window only vs. no window
+ plants vs. no window + inaminate objects vs. No
window only; EDTP
Gamble, Howard, and RCT (b) United States; 56 University students, general Virtual; viewing images; natural vs. urban; 05ʹ50 Attention Network Task; Digit
Howard (2014) laboratory population; NR; m = 20.54 years, Span Backwards
m = 69.1 years
Greenwood and RTC (b) United Kingdom; 120 High school students; 55% Actual; relaxing vs. talking with peer vs. Playing game Necker Cube Pattern Control*
Gatersleben (2016) high school, South female; 16–18 years on mobile phone; Natural outdoor vs. small indoor
West London room; 20ʹ00
Haga et al. (2016) RTC (b) Sweden; university 90 University students; 68% female; Virtual; pink noise stimuli - attributed to natural Attention Network Task
campus m = 24.76 years (waterfall) or non-natural (machinery) source; 03ʹ00
Han (2017) RCT (b) Taiwan; university 116 University students; 55.2% Actual; walking - ‘high natural’ vs. ‘low natural’; jogging Forward Spatial Span, Digit
campus female; m = 20.85 years - ‘high natural’ vs. ‘low natural’; 15ʹ00 Span Backwards
Jenkin et al. (2017) RCT (b) England; 75 Children; 50.6% female; Virtual; viewing video of natural scenes vs. Viewing Delay of Gratification; Stroop
8–11 years video of urban scenes; 03ʹ00
Johansson et al. (2015) RTC (b) Sweden; 34 Aquired brain injury patients, Actual; walking; Nature walks in groups vs. face-to-face Digit Symbol Coding;
Gothenburg reporting mental fatigue; 82.3% MBSRd vs. internet MBSRd; 90ʹ00/week (x 8) Attentional Blink Task
Botanical Gardens female; m = 48.0, 46.6, 51.2 years
Lee et al. (2015) RCT (b) Australia; laboratory 150 University students; 71% female; Virtual; Viewing city with green roof vs. Viewing city Sustained Attention to
m = 20 years without green roof; 00ʹ40 Response Task
Li and Sullivan (2016) RCT (b) United States; high 94 High school students; 56.4% Actual; Greenspace window view vs. builtspace window Digit Span Fowards + Digit
schools female; NR view vs. no window view; EDTP (10ʹ00 rest before post Span Backwards (summary
test) score)
Lin et al. (2014) RCT (b) Taiwan; laboratory 138 University students; 52.9% Virtual; viewing images; urban street with simulated Digit Span Backwards
(classroom) female; NR natural stimuli (minimal, moderate, or heightend
awareness towards natural stimuli) vs. urban street
without natural stimuli; 01ʹ40
Lobo, Ramachandran, RCT (b) India; laboratory 40 Students; 50% female; Virtual; viewing images + imagination; natural Navon Global-Local Task
JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B

and Tiwari (2015) m = 22.95 years (pleasant, neutral) vs. urban (unpleasant vs. neutral);
00ʹ50
(Continued )
229
 230

Table 1. (Continued).
Author, Year, (sub- Study characteristics: population; Intervention and control characteristics: exposure type;
study) Study design Country, setting n gender; age activity; subgroup/environments; duration Cognitive performance measure
Lymeus, Lundgren, and RCT (b) Sweden; small 51 University students; 72.5% Virtual; mindfulness meditation with nature images vs. Letter-Digit Substitution Test
Hartig (2017) classroom female; m = 25 years Conventional mindfulness meditation vs. Rest with
nature images; 15ʹ00
Pilotti et al. (2014) RCT (b) United States; 63 University student advisors; Virtual; viewing video; natural vs. urban; 15ʹ00 Auditory Oddball Task
laboratory 61.9% female; m = 31.79 years
Studente, Seppala, and RCT (b) United Kingdom; 108 Universty students; NR; NR Actual; viewing; window + plants vs. no window + no Alternative Uses Test; 30 Circles
M. P. STEVENSON ET AL.

Sadowska (2016) laboratory plants + green paper vs. No window + no plants; EDTP Test
Tanaka et al. (2013) RCT (b) Japan; housing 16 General population; 100% female; Actual; viewing; nature vs. no nature; 30ʹ00 Advanced Trail Making (A, B, C)
median = 43.5 years
Van Rompay and Jol RCT (b) The Netherlands; 120 High school students; 50% Virtual; viewing images + imagination; natural vs. Tests for Creative Thinking -
(2016) laboratory female; m = 13.8 years urban; EDTP Drawing Production
Wang et al. (2016) RCT (b) China; laboratory 140 University students; 50% female; Virtual; viewing videos; natural vs. urban; 08ʹ00 Digit Span Backwards
m = 22.38 years
Wang et al. (2017) RCT (b) China; laboratory 118 University students; 78.8% Virtual; viewing videos; natural vs. urban; 10ʹ00 Taylor’s Aggression Paradigm
female; m = 21.23 years (Reaction Time Competition)
Zhang, Kang, and Kang RCT (b) China; urban park 70 University students; 51.4% Actual; resting; urban park enivronment (within- Complement Numbers Test
(2017) female; m = subjects) with primarily bird and insect sounds vs. (simple addition fluency)
Primarily periodic machinery sound vs. Primarily
continous transport sound; 40ʹ00
Ferraro (2015) Quasi-experiment United States; 25 University students; NR; Actual; hiking; wilderness hike vs. indoor control; six Remote Associations Test
Canoe trip in m = 18.71, 18.38 years days
Minnesota
Kelz, Evans, and Quasi-experiment Austria; high 133 High school students; 48.5% Actual; Green school yard design vs. control school yard Attention Network Task
Röderer (2015) schools female; 13–15 years n; 6–7 weeks
Van den Berg et al. Quasi-experiment The Netherlands; 170 School children; 42.9% female; Actual; viewing; “green wall” with living plants vs. no Digit Letter Substitution Test;
(2016) primary schools m = 9 years green wall; 4 months Sky Search Task
Dadvand et al. (2015) Natural Spain; school and 2623 School children; 50% female; Actual; Exposure to greenspace (residential, commute n-Back Test; Attention Network
experiment residential areas in m = 8.5 years to school, school); Cognitive testing progress over Task
Barcelona 12 months
Dadvand et al. (2016) Natural Spain; residential 888; Children; 48.8% female (K-CPT & Actual; Residential exposure to greenspace; K-CPT at Conner’s Kiddie Continuous
experiment areas in Sabadell 978 ANT); birth-7 years 4–5 years, ANT at 7 years Performance Test; Attention
and Valencia Network Test
Ulset et al. (2017) Natural Norway; daycare 562 Young children; 53% female; Actual; time spent outdoors during daycare hours; Digit-Span Forwards Test
experiment centres m = 52.45 months 4 years
Ward et al. (2016) Natural New Zealand; 72 Intermediate (middle) school Actual; Exposure to greenspace measured using CNS Vital Signs Neurocognitive
experiment intermediate students; 56.9% female; portable GPS; 7 days Test Battery
(middle) schools m = 12.66
Zijlema et al. (2017) Natural Spain, The 1493- General population adults; 54,1% Actual; Residential distance to greenspace, residential Color Trails Test
experiment Netherlands, United 1602m female; m = 48 years surrounding greenness; n/a
Kingdom
Notes: RCT = randomized controlled trial; w = within-participants; b = between-participants; MHI = Mental Health Index; EDTP = exposure during task performance; MBSR = mindfulness-based stress
reduction; NR = not reported; K-CPT = Kiddies Continuous Performance Task; ANT = Attention Network Task.
 JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B 231

Theory (ART). According to ART (Kaplan 1995; significantly affected by natural environments
Kaplan and Kaplan 1989), the mental fatigue that was the Trail Making Test B, which is generally
is associated with a depleted capacity to direct considered a measure of cognitive flexibility
attention may be overcome by spending time in (Kortte, Horner, and Windham 2002) or atten-
environments rich in natural stimuli. A central tional control that allows switching between two
premise of ART suggests that natural stimuli are rule sets (Arbuthnott and Frank 2000). Thus, tasks
intrinsically fascinating in a way that evokes a type in which a restoration effect is most reliably
of attention that does not require effortful fixation detected seem to rely on directed attention capa-
or cognitive effort (Kaplan 1995). As the brain city in different ways and to varying degrees. In
engages externally driven and effortless attention, this sense, the exact aspects of directed attention
the capacity to direct attention in an effortful which is influenced by natural environments
manner is restored. The process of restoration is remain obscured. One task that isolates the ability
not necessarily exclusive to natural environments; to direct attention while simultaneously ignoring
however, these environments are nominated as distraction is the Attention Network Task (Fan
being particularly suitable for covering the 4 char- et al. 2005). Unfortunately, when Ohly and collea-
acteristics of a restorative environment (Kaplan gues (2016) performed their literature search in
1995). According to the theory, environments 2013, there had been only one study using this
that contain so-called softly fascinating stimuli refined measure of directed attention that isolates
should be optimally restorative when they also the ability from other cognitive processes.
evoke a feeling of being away or separated from Therefore, there is not currently sufficient evi-
cognitively fatiguing situations; have a sense of dence to determine whether natural environments
extent̶ which can be literal in the sense of physical only restore directed attention indirectly through
size or conceptual by evoking the feeling of being affecting additional related cognitive processes
connected to something larger; and be compatible activated during task performance or whether
with the individual’s desires and goals so that it restoration is also possible when the ability to
does not impede but supports the realization of direct attention is isolated by a task’s
these desires (Kaplan 1995). characteristics.
Ohly and colleagues (2016) concluded that the
field would benefit from identifying measures of
Ambiguity surrounding directed attention in directed attention that are most likely to be sensi-
ART tive to natural environments and by then consid-
The phenomenon of attention restoration is ering the specific characteristics of the tasks and
believed to be captured by studies reporting associated attentional processes that appear to be
improvements on cognitive tasks after exposure driving the effect. They argued that there should
to a natural environment or natural stimuli, rela- be a goal to reach an agreement on what should
tive to a suitable control. Ohly and colleagues’ constitute a measure of directed attention and to
(2016) systematic review was timely in pointing establish an optimal testing protocol, so that the
out the ambiguity in the field surrounding the restoration effect can be tested consistently across
construct of directed attention, reflected in the multiple studies, leading to a more reliable synth-
diversity of tasks used to measure the effect. The esis of data in the future.
Digit Span Forwards (DSF) and Digit Span
Backwards (DSB) Tasks were two of the three
Challenges faced by Ohly and colleagues
outcome measures that showed significant positive
(2016)
effects in their meta-analysis that pooled data
across the literature base. These tasks are tradi- In a letter to the editor in response to the review,
tionally considered working memory tasks with Hartig and Jahncke (2017) echoed the concerns
varied loads of cognitive effort or demands on raised by the article, calling for a tightening
other processes, such as attention (Conway et al. between theory and measurement of effects.
2005). The remaining task found to be Their commentary was positive and praised the
232 M. P. STEVENSON ET AL.

substantial effort and contribution by the authors. related to change scores is challenging for meta-
They did, however, also indicate additional con- analysis, due to the difficulty of obtaining the
cerns over the meta-analysis itself and warned standard deviations (SD) of the standardized
against putting too much emphasis on such pub- mean difference. To calculate effect sizes based
lications without considering the finer details. The on change scores, Ohly et al. (2016) would have
problems arising from Ohly and colleagues (2016) had to obtain the Pearson’s r correlation between
meta-analysis can be attributed to the diversity of the pre-test scores for each study, which is seldom
methods they uncovered, which may have influ- calculated or reported. Imputation of r for calcu-
enced their conclusions. lating SD in this manner is possible, though not
Firstly, the need to build links between protocol recommended (Cochrane Collaboration 2008).
and theory is exemplified by the inclusion or Therefore, Ohly and et al (2016) conducted their
exclusion of a cognitive loading task in attention meta-analysis using only post-test attention scores
restoration studies. A cognitive loading task is following the environmental interventions. Even
sometimes included prior to baseline attention with randomized within-subject designs, the base-
measures or environmental treatment to induce a line is necessary to control for intra-individual
state a mental fatigue. Because the effect of a differences that occur between sessions. In atten-
natural environment rests on the premise that an tion restoration research, many studies use sepa-
individual experiences restoration from a fatigued rate days for each environmental treatment.
state, the inclusion of such a task seems critical to Therefore, without baseline measurements, it is
the existence or at least the strength of the effect unclear whether post-test measures are related to
being studied. ART itself predicts that participants the treatment of the session or an unforeseen
who are cognitively fatigued prior to an environ- spurious variable that differed between the testing
mental treatment will have a higher restoration days, such as amount or quality of sleep the night
potential and are therefore more likely to demon- before.
strate greater improvement in task performance
than those who are not. Thus, on a theoretical
level, combining effect sizes using fatigued and Clarifying directed attention and exploring
moderator variables
non-fatigued participants, as Ohly et al. (2016)
did, seems problematic. Incidentally, one of An updated systematic literature review and meta-
Hartig’s own studies (Hartig et al. 2003) in which analysis was conducted to supplement the investi-
fatigue induction was used as a between-subjects gation by Ohly and colleagues (2016). The primary
factor, showed no marked effect of the fatigue- aim was to expand the current understanding of
inducing task. Despite this result going against directed attention within the field to remove ambi-
the predictions of ART, it seems important to guity surrounding the cognitive processes sensitive
consider the influence of fatigue induction when to the restoration effect associated with exposure
calculating effect sizes at a meta-analytic level. to natural environments. The secondary aim was
Secondly, Ohly and colleagues (2016) were to explore the influence of protocol characteristics,
faced with the difficulty of baseline imbalances through moderator analysis, to begin a dialogue
and studies that did not include baseline measures towards optimal protocols for attention restoration
from which to measure changes in cognitive per- research such that greater homogeneity might be
formance related to the environmental treatments. achieved across future studies.
They noted that when baseline measures were To address these aims, a subgroup of studies
included, the vast majority of authors in the field was selected from the original (Ohly et al. 2016)
preferred to analyze change scores (post-test and updated literature searches where baseline
minus pre-test) to measure treatment effect, rather measures prior to environmental exposure had
than using ANCOVA on post-test scores where been included. The following were recorded: 1)
the baseline is included as a covariate, as is pre- timing of baseline measurement (Figure 1); 2) the
ferred for randomized treatment studies (Van average baseline score; 3) whether restoration
Breukelen 2013). The calculation of effect sizes potential of participants was heightened, either
 JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B 233

Figure 1. Timeline of a typical randomized controlled trial in attention restoration research. T1 and T2 represent commonly used
baseline measurements, occurring before or after the fatigue induction phase. T3 may represent the timing of post-test or single
measurement of cognitive performance when no baseline measurement is taken.

by an experimental fatigue-induction phase or Preferred Reporting Items for Systematic Reviews

through participant characteristics; and 4) type of and Meta-analyses (PRISMA) statement, which
environmental exposure (virtual stimuli presented aimed to optimize systematic reviewing of scientific
in highly controlled laboratory settings or actual research (Moher et al. 2009). Details of the search
exposures taking place in real environments). protocol were published on the PROSPERO data-
For initial analysis, outcome measures were base (Reference: CRD42017057718).
grouped based on relevant cognitive domains, in
order to provide pooled mean effect sizes based on
how directed attention is recruited. This enabled Literature search
us to discuss the findings of our analyses in rela-
A systematic search strategy designed to review the
tion to underlying attentional processes by consid-
objective evidence surrounding ART had been
ering the specific characteristics of the cognitive
developed independently, prior to the publication
domains and outcome measures shown to be most
of Ohly et al. (2016) meta-analysis, with a high
sensitive to natural environments. For moderator
level of similarity. The original search strategy was
analysis, the extent to which the pooled mean
created to capture existing research that contained
effect estimates for each cognitive domain could
points of interest within all three key concept
be explained by a balance of baseline scores,
domains. Firstly, all research that investigated cog-
restoration potential of participants, and type of
nition after exposure to the natural environment
environmental exposure was explored. The results
or natural stimuli was included. The term ‘natural
are discussed by presenting a three-axis framework
that may be useful in differentiating and qualifying environment’ is broad and in psychological
research may be simply used to describe a non-
the role of directed attention in each task or cog-
laboratory setting. Therefore, many relevant search
nitive domain. The proposed framework may be
terms and phrases that were inspired by prelimin-
used to guide theory development and future
ary searches were included. Large variation was
research, particularly in terms of task selection.
found in the way authors chose to describe the
use of natural environments in their research. For
example, Taylor, Kuo, and Sullivan (2002) used 6
Methods
different ways in their abstract to express the con-
The search, collection, and review of literature was cept of nature (i.e., “views of nature,” “contact
conducted following the guidelines put forth by the with nature,” “near-home nature,” “nearby
234 M. P. STEVENSON ET AL.

nature,” “naturalness,” and “green space”). Our criteria were put in place in order to capture litera-
definition of exposure to the natural environment ture relevant to our purpose. Limits were set during
was broad and therefore included anything from the literature search in order to only include peer-
passively viewing nature scenes on a computer reviewed original research articles published after
screen to active activities (e.g., running) in real July 2013, the month in which Ohly et al. (2016)
environments. Secondly, only studies that used conducted their search. The articles were screened by
measures of cognitive function were considered. title and abstract in order to identify potentially
Therefore, search terms related to cognitive abil- relevant studies. Full-text screening took place for
ities and attention restoration were included. all articles that appeared relevant after the first two
Finally, only studies that measured cognitive abil- screenings. Full texts were excluded from the final
ity using objective standardized cognitive tasks review if they fell under any of the exclusion criteria
were included. Thus, a final key concept related that were defined during screening. Specifically, stu-
to tasks, along with additional relevant terms was dies were excluded that (a) were not experimental
included in the search strategy. studies; (b) did not use a natural environment or
Based upon the almost complete overlap of natural stimuli; (c) did not include a suitable control
search terms and studies uncovered in our original group or comparison environment (e.g., natural vs
search strategy and those reported by Ohly et al. urban); or (d) did not include objective outcome
(2016), we were confident in using our search measures derived from standardized cognitive
string to follow-up their work. However, to tasks. The search was not limited based on partici-
address discrepancies, two search terms (“garden*” pant characteristics such as gender, age, or diagnostic
and “attention restorat*”) used by Ohly and col- status of participants. Nor was the search limited by
leagues (2016) were added that were not included country, culture, or presence of water in the natural
in our original search string. The following search environment. Research published in languages other
string was applied to systematically screen titles, than English was not included.
abstracts, and keywords from 6 databases
(PsycINFO, SPORTDiscus, Medline, Embase,
Data extraction
Scopus, and Web of Science) in order to cover a
broad range of scientific disciplines: The same study characteristics were selected as
(“natural environment*” OR “natural area*” OR presented by Ohly et al. (2016) review.
“natural space*” OR “green area*” OR “green Specifically, study design, participant details, set-
space*” OR greenspace* OR forest OR woodland* ting details for natural and control conditions,
OR “nature exposure*” OR “exposure* to nature” type of exposure (virtual, physical), engagement
OR “nature experience*” OR “experience* in nat- with the environment, duration of exposure, and
ure” OR “contact with nature” OR “nature contact” cognitive measures were recorded. Data based on
OR “interact* with nature” OR “restorative envir- pre- and post-test means and standard deviations
onment*” OR “natur* view*” OR garden* OR were extracted. Supplementary information was
“attention restorat*”) added regarding the restoration potential of parti-
AND cipants. This included details of any reported fati-
(cognit* OR attention* OR memory OR concen- gue induction using a stressor task or cognitive
trat* OR “executive function*” OR “problem sol- loading task, the duration of the induction phase,
ving” OR inhibition OR impulsiv* OR “mental and whether participant characteristics might con-
fatigue” OR self-regulation”) tribute to restoration potential, for example, by
AND having a relevant diagnosis.
(test* OR task OR battery OR exam*)
Quality appraisal
Study selection and eligibility criteria
To further the continuity with Ohly et al. (2016)
The final systematic search of databases took place investigation, the same quality appraisal system
during November 2017. A number of eligibility was used to rate the included studies. Their quality
 JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B 235

appraisal scale contained a combination of indica- moderator analyses were performed to determine
tors that were derived from three rating systems the influence of three methodological variables on
previously developed to identify potential bias and the pooled effect size. Within each cognitive domain,
quantify the robustness of evidence. For specific studies were grouped by baseline balance (yes, par-
descriptions of each individual indicator, see tial, unsure, no); restoration potential (heightened,
Table 2 in Ohly and colleagues (2016). A two- normal); and exposure type (actual, virtual). Meta-
factor checklist was added to the quality appraisal and moderator analyses were conducted using
system to identify studies that were suitable for Comprehensive Meta-Analysis (CMA 3.3.070).
inclusion in our meta-analysis.
It is understood that not all information could
be provided in published manuscripts, so in agree- Results
ment with Ohly et al. (2016), authors were directly Part one - the systematic literature search
contacted for clarifications when an initial assess-
ment was rated “unclear” (n = 41). The ratings After searching 6 databases, 3997 results were
were changed based on this practice for 31 studies identified that were screened for inclusion in our
(75.6% response rate). review. From this initial number, a total of 42
publications were included in the final review
(Figure 2). Three publications reported results
Data synthesis from multiple studies that were suitable for our
A series of random effects meta-analyses were analysis. In total, there were a total of 46 separate
performed using data from a subset of studies studies included in our final review.
identified by the literature search performed by It is worth noting the increasing productivity in
ourselves and Ohly and colleagues (2016). A com- the area of attention restoration research.
mon variance component was not assumed across Although our systematic literature search covered
studies or subgroups. For studies with more than only 52 months between July 2013 and November
one outcome measure, it was noted, where possi- 2017, there was a considerably larger number of
ble, which study was performed first at post-test, relevant studies published during this time than in
as these outcomes will be more sensitive to the the years prior to July 2013. Figure 3 represents
restoration effect. For studies with multiple out- this increase in productivity graphically, where the
come measures from the same participant in the red line marks the time at which Ohly et al. (2016)
same meta-analysis, the variance was adjusted conducted their systematic search. All studies
assuming non-independence and was represented represented to the left of the line (n = 24) were
as combined effects. Studies were selected based included in Ohly and colleagues (2016) review,
on the main inclusion criterion, namely, a mea- and all studies represented to the right of the line
surement of baseline performance. (n = 42) were included in the current review.
To begin, Ohly et al. (2016) method of comparing
post-test outcome data between natural environ-
Study characteristics
ments or stimuli (experimental) with non-natural
environments or stimuli (control) was followed. Table presents the characteristics of included stu-
Here, pooled mean differences were calculated by dies and is segmented by type of study. The major-
extracting means, SD, and sample sizes for each ity (82.6%) of the 46 included studies utilized a
environmental condition and each outcome mea- randomized-controlled trial design. Of these, 29
sure. Outcomes that measured conceptually similar studies used a between-subjects design where
cognitive constructs across different tasks were then each subject was exposed only one environmental
grouped. This was done to obtain standardized mean treatment and 9 studies used a within-subjects
effect size estimates (Hedges’ g) for separate cogni- crossover design where each participant was
tive domains, as a platform for discussing the role of exposed to both or all environmental treatments.
directed attention within each domain. Finally, to There were 5 natural experiments that tested
address the challenges noted by Ohly et al. (2016), whether exposure to natural environments
Table 2. Quality appraisal of included studies.
236

Berry Berry Beute and De Beute and De Bratman Chow and Chow and Chow and Craig Dadvand Dadvand Emfield and Evensen Gamble, Howard,
et al. et al. Kort (2014), Kort (2014), et al. Lau (2015), Lau (2015), Lau (2015), et al. et al. et al. Neider et al. Ferraro and Howard
(2014) (2015) (1) (2) (2015) (1) (2) (3) (2015) (2015) (2017) (2014) (2015) (2015) (2014)
Quality indicators
Study design
Power calculation reported No No No No No No No No No No No No No No No
Inclusion/exclusion criteria reported No Yes No No Yes No No No No Yes No Yes No No No
Individual level allocation Yes Yes Yes Yes Yes Yes Yes Yes Yes n/a n/a Yes Yes No Yes
Random allocation to groups/ Yes Yes Yes Yes Yes Yes Yes Yes Yes n/a n/a Yes Yes No Yes
condition/order
Randomization procedure appropriate Yes Yes Yes Yes Yes Yes Yes Yes Yes n/a n/a Yes Yes No Yes
M. P. STEVENSON ET AL.

Confounders
Groups similar (sociodemographic) Yes Yes Yes Yes Yes Un. Un. Un. Yes n/a n/a Yes Un. Yes Pa.
Groups balanced at baseline Un. Un. Un. Un. Yes Un. Un. Un. Yes n/a n/a Yes Yes Yes Pa.
(cognitive scores)
Participants blind to research Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
question
Intervention integrity
Demonstrated need for attention No No Yes Yes No Yes Yes Yes No n/a n/a No Yes No No
restoration
Clear description of intervention and Yes Yes Yes Yes Yes Yes Yes Yes Yes n/a n/a Yes Yes No Yes
control
Consistency of intervention (within Yes Yes Yes Yes Pa.p No Yes Yes Yes n/a n/a Yes Yes No Yes
and between groups)
Data collection methods
Outcome assessors blind to group No No No No Un. Un. Un. Un. No n/a n/a Yes Un. Yes No
allocation
Baseline attention measures taken No No No No Yes No Yes No Yes n/a n/a Yes No No Yes
before the exposure
Consistency of data collection Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Analyses
All attention outcomes reported Yes Yes Yes No No Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
(means and SD/SE)
All participants accounted for (i.e., Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
losses/exlusions)
ITT analysis conducted (all data No No No No Yes Yes Yes Yes Yes n/a n/a No Yes Yes Yes
included after allocation)
Individual level analysis Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Statistical analysis methods Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
appropriate for study design
External validity
Sample representative of target Noa Noa Noa Noa Un. Noa Noa Un. Noa Yes Yes Noa Noa Noa Pa.
population
Overall quality score
Total number of points 24/40 26/40 26/40 24/40 29/40 24/40 28/40 26/40 30/40 16/18 14/18 32/40 28/40 20/40 29/40
Quality rating as % 60 65 65 60 72.5 60 70 65 75 88.9 77.8 80 70 50 72.5
Responded to query about Yes Yes Yes Yes No No No No Yes n/a n/a Yes No Yes Yes
“uncertain” ratings
Criteria for meta-analysis
Fatigue task No No Yes Yes No Yes Yes Yes No n/a n/a No Yes No No
Baseline No No No No Yes No Yes No Yes n/a n/a Yes Yesb No Yes
 Geniole Greenwood and Haga Jenkin Jung Lee Li and Lin Lobo Rider and Rider and
et al. Gatersleben et al. Han et al. Johansson et al. Kelz, Evans, and et al. Sullivan et al. et al., Lymeus, Lundgren, Bodner (2016) Bodner
(2016) (2016) (2016) (2017) (2017) et al. (2015) (2017) Röderer (2015) (2015) (2016) (2014) (2015) and Hartig (2017) (1) (2016)(2)
Quality indicators
Study design
Power calculation reported No No No No No No Yes No No No No No Yes No No
Inclusion/exclusion criteria Pa. No No Yes No Yes Yes No No Yes Pa. No Yes No No
reported
Individual level allocation Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Random allocation to groups/ Yes Yes Yes Yes Yes Pa. Yes No Yes Yes Yes Yes Yes Yes Yes
condition/order
Randomization procedure Yes Yes Yes Yes Yes Yes Yes n/a Yes Yes Yes Yes Yes Yes Yes
appropriate
Confounders
Groups similar Yes Yes Un. Yes Un. Yes Pa. Yesd Yes Yes Yes Un. Yes Pa. Pa.
(sociodemographic)
Groups balanced at baseline Un. Un. Un. Yes Yes° Yes Pa. No Yes Yes Yes Un. Yes Un. Un.
(cognitive scores)
Participants blind to research Yes Yes Un. Yes No Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
question
Intervention integrity
Demonstrated need for attention No Yes Yes No Nom Pa.c Pa.n n/a No Yes No Yes Pa.f No No
restoration
Clear description of intervention Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
and control
Consistency of intervention Yes Yes Yes Yes Yes No Yes Yes Yes Yes Yes Yes Yes Yes Yes
(within and between groups)
Data collection methods
Outcome assessors blind to No Yes Un. No Yes No Yes No No Un. No Yes No No No
group allocation
Baseline attention measures Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes No Yes No No
taken before the exposure
Consistency of data collection Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Analyses
All attention outcomes reported No Yes Yes No Yes Yes Yes Yes Yes No Yes No No Yes Yes
(means and SD/SE)
All participants accounted for (i. Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
e., losses/exlusions)
ITT analysis conducted (all data Yes Yes Yes No No No Yes No Yes Yes Yes Yes No Yes Yes
included after allocation)
Individual level analysis Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Statistical analysis methods Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
appropriate for study design
External validity
Sample representative of target Noa Yes Noa Noa No Yes Yes Yes Noa Un. Noa Noa Noa Noa Noa
population
Overall quality score
JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B

Total number of points 27/40 34/40 26/40 28/40 26/40 30/40 37/40 24/36 30/40 32/40 31/40 26/40 31/40 25/40 25/40
Quality rating as % 67.5 85 65 70 65 75 92.5 66.7 75 80 77.5 65 77.5 62.5 62.5
(Continued )
237
 238

Table 2. (Continued).
Geniole Greenwood and Haga Jenkin Jung Lee Li and Lin Lobo Rider and Rider and
et al. Gatersleben et al. Han et al. Johansson et al. Kelz, Evans, and et al. Sullivan et al. et al., Lymeus, Lundgren, Bodner (2016) Bodner
(2016) (2016) (2016) (2017) (2017) et al. (2015) (2017) Röderer (2015) (2015) (2016) (2014) (2015) and Hartig (2017) (1) (2016)(2)
Responded to query about Yes Yes No Yes Yes Yes Yes Yes Yes No Yes Yes Yes Yes Yes
“uncertain” ratings
Criteria for meta-analysis
Fatigue task No Yes Yes No No Pac Pa.n n/a No Yes No Yes Pa.f No No
Baseline Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes No Yes No No
Rogerson Sahlin Sonntag- Studente, Seppala, Tanaka Triguero- Ulset Van Den Van Rompay Wang Wang Ward Zhang, Kang, Zijlema
M. P. STEVENSON ET AL.

et al. et al., )Schutte, Torquati, Öström et al., and Sadowska et al., Mas et al. et al. Berg et al. and Jol et al., et al. et al. and Kang et al.
(2016) (2016) and Beattie (2017) (2014) (2016) (2013) (2017) (2017) (2016) (2016) (2016) (2017) (2016) (2017) (2017)
Quality indicators
Study design
Power calculation reported No No No Yes No No Yes No No No No No No No No
Inclusion/exclusion criteria No No Yes Yes No Yes Yes No Yes No Yes No Yes No No
reported
Individual level allocation Yes Yes Yes Yes Yes Yes Yes n/a No Yes Yes Yes n/a Yes n/a
Random allocation to Yes Yes Yes Yes Yes Yes Yes n/a No Yes Yes Yes n/a No n/a
groups/condition/order
Randomization procedure Yes Yes Yes Yes Yes Yes Yes n/a n/a Yes Yes Yes n/a Nol n/a
appropriate
Confounders
Groups similar Yes Yes Yes Yes Yes Un. Yes n/a Yes Yes Yes Yes n/a Un. n/a
(sociodemographic)
Groups balanced at baseline Yes Un. Un. Un. Yes Yes Yes° n/a Yes° No Yes Yes n/a Un. n/a
(cognitive scores)
Participants blind to research Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Un. Yes
question
Intervention integrity
Demonstrated need for No No Yes Yesg No Yes No n/a n/a No Yesj Yesk n/a Yes n/a
attention restoration
Clear description of Yes Yes Yes Yes Yes Yes Yes n/a Yes Yes Yes No n/a Yes n/a
intervention and control
Consistency of intervention Yes No Yes Yes Yes Yes Yes n/a Yes Yes Yes Yes n/a Pa. n/a
(within and between
groups)
Data collection methods
Outcome assessors blind to No Un. Pa. No Yesh Un. No n/a Un. Yesh Yes No n/a Un. n/a
group allocation
Baseline attention measures Yes Yes No No No Yes Yes n/a Yes No Yes Yes n/a Yes n/a
taken before the exposure
Consistency of data Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
collection
Analyses
All attention outcomes Yes Yes Yes Yes No No No Yes Yes Pa. Yes Pa. No Pa. Yes
reported (means and SD/
SE)
(Continued )
Table 2. (Continued).
Rogerson Sahlin Sonntag- Studente, Seppala, Tanaka Triguero- Ulset Van Den Van Rompay Wang Wang Ward Zhang, Kang, Zijlema
et al. et al., )Schutte, Torquati, Öström et al., and Sadowska et al., Mas et al. et al. Berg et al. and Jol et al., et al. et al. and Kang et al.
(2016) (2016) and Beattie (2017) (2014) (2016) (2013) (2017) (2017) (2016) (2016) (2016) (2017) (2016) (2017) (2017)
All participants accounted Yes Yes Yes Yes No Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
for (i.e., losses/exlusions)
ITT analysis conducted (all Yes Yes Yes No No Yes Yes n/a No Yes Yes Yes n/a No n/a
data included after
allocation)
Individual level analysis Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Statistical analysis methods Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
appropriate for study
design
External validity
Sample representative of Un. Un. Yes Yes Noa Un. Yes Yes Un. Yes Noa Noa Un. Noa Yes
target population
Overall quality score
Total number of points 30/40 26/40 33/40 32/40 24/40 30/40 34/40 14/18 24/36 29/40 36/40 29/40 12/18 18/40 14/18
Quality rating as % 75 65 82.5 80 60 75 85 77.8 66.7 72.5 90 72.5 66.7 45 77.8
Responded to query about Yes Yes Yes Yes Yes No Yes n/a No Yes Yes Yes Yes No n/a
“uncertain” ratings
Criteria for meta-analysis
Fatigue task No No Yes Yesg No Yes No n/a n/a No Yesj Yesk n/a Yes n/a
Baseline Yes Yes No No No Yes Yes n/a Yes No Yes Yes n/a Yes n/a
Notes. Total number of possible points: 40 for RCT design; 36 for quasi-experimental design; 18 for natural experiment design.
a
No target population per se, however, participants were a convenience sample of university students.
b
Baseline measures, fatigue task, and post measures all completed during exposure of natural/non-natural stimuli.
c
Not experimentally induced. Participants reported mental fatigue.
d
Education level of parents differed between groups. This was controlled for in the statistical analysis, with no influence on effect.
e
Single outcome measure derived from two tasks (DSF + DSB).
f
Not experimentally induced. Study took place at the end of a workday.
g
Not experimentally induced. Participants recruited based on diagnosis of Exhaustion Disorder.
h
Particularly important for the subjective nature of outcome measures.
i
ATMT contains subtasks with varying difficulty level. Combined scores analyzed and reported.
j
Participants gave a speech in English. Used to evoke stress response rather than fatigue attention.
k
Depletion used as between-subjects factor.
l
Participants not randomized, but matched on gender and grade of a recent test.
m
An attempt was made to induce ‘ego depletion’ but that task was not suitable for the participants, likely due to age.
n
Participants were patients who had suffered heart failure.
° Baseline used as covariate in the linear regression model.
JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B
239
240 M. P. STEVENSON ET AL.

(commonly referred to as greenspace in these stu- environments as they normally would (Triguero-
dies) influenced cognitive development over Mas et al. 2017). More active engagement was
longer time periods. Three quasi-experiments employed in some studies, including hiking
were included where assignment to environmental (Ferraro 2015) and cycling (Rogerson et al. 2016).
exposure could not be randomized. In two cases Passive viewing of actual natural environments or
(Kelz, Evans, and Röderer 2015; Van Den Berg stimuli was included in randomized-controlled
et al. 2016) these studies took place in school trials (Evensen et al. 2015; Greenwood and
settings, while the remaining study examined the Gatersleben 2016; Li and Sullivan 2016; Studente,
effects of a wilderness hike (Ferraro 2015). Seppala, and Sadowska 2016) and quasi-experi-
The included studies were conducted largely in mental studies (Van Den Berg et al. 2016). While
western countries. Twenty studies took place in natural experiments explored exposure to green
Europe (43.5%); 15 in North America (32.6%); space in residential areas (Dadvand et al. 2015,
one in Australia (2.2%); one in New Zealand 2017) and/or school environments (Dadvand
(2.2%); and 9 in Asia (19.5%). One of the natural et al. 2015; Ward et al. 2016) Virtual exposure to
experiments involved data from multiple countries environments was used in 55.3% of all rando-
(Zijlema et al. 2017). mized-controlled trials. Virtual exposure consis-
A range of age groups, from young children tent almost exclusively of image or video viewing;
(m = 4.53 years; Schutte, Torquati, and Beattie however, sound was also investigated by Emfield
2017) to elderly (m = 69.1 years; Gamble, Howard, and Neider (2014) as a between-subjects factor.
and Howard 2014), were represented in the sample Haga and colleagues (2016) explored expectation
populations across the included studies. However, bias and stimuli-source attribution by telling par-
young adults, represented mostly by university stu- ticipants who listened to white noise that they
dents, made up the largest proportion (52.2%). Most were hearing either a natural environment with a
investigators recruited participants with typical cog- waterfall or an industrial environment with
nitive function; however, several studies included machinery. In some studies, virtual environments
participants where restoration potential was were supplemented with other engagement activ-
expected to be heightened. In three cases this was ities such as imagining presence in the environ-
based on a specific diagnosis, including heart failure ment (Chow and Lau 2015; Van Rompay and Jol
(Jung et al. 2017), acquired brain injury (Johansson 2016) or mindfulness meditation (Lymeus,
et al. 2015), and exhaustion disorder (Sonntag- Lundgren, and Hartig 2017).
Ostrom et al. 2014). Other investigations included The duration of exposure to environmental
participants with presumed higher restoration treatments ranged considerably between the stu-
potential due to prior lectures (Lin et al. 2014) or dies. Randomized-controlled trials ranged from
end-of-day testing sessions (Lymeus, Lundgren, and ‘micro-breaks’ of 40 s (Lee et al. 2015) to three-
Hartig 2017). The final study to consider restoration hr exposures (Triguero-Mas et al. 2017), while
potential based on participant characteristics was some included a course of exposures lasting sev-
Craig and colleagues (2015), who grouped university eral weeks (Johansson et al. 2015; Lymeus,
students based on typical or heightened depressive Lundgren, and Hartig 2017). In the quasi-experi-
symptom ratings. mental studies and natural experiments, where
The included studies used actual physical expo- exposure to natural environments or stimuli was
sure (54.3%) or virtual exposure (45.7%) as the of interest, durations were considerably longer,
environmental treatments. Walking was the most ranging from 6 days (Ferraro 2015) to several
frequently recorded type of engagement in studies years (Dadvand et al. 2017; Ulset et al. 2017).
that involved actual physical exposure. In some As reported in Table 2, the quality ratings of the
cases, walking was used in combination with included studies all fell within the moderate (34–
another activity, such as taking photos (Bratman 66%) and high (67–100%) classifications used by
et al. 2015) and relaxing (Sonntag-Ostrom et al. Ohly et al. (2016). As Ohly et al. (2016) noted, the
2014; Zhang, Kang, and Kang 2017); while one quality ratings judged publications on overall
study instructed the participants to act in the experimental quality as well as the extent to
 JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B 241

Figure 2. PRISMA flowchart of the screening procedure.

Figure 3. Timeline of attention restoration studies by year. The dashed line represents the time of Ohly et al.’s (2016) systematic
literature search.

which related to the reviews on attention restora- included experiments. Trends revealed by the
tion research, which was not always the aim of the quality assessment include a lack of power
242 M. P. STEVENSON ET AL.

calculations, variety in the extent to which inclu- working memory domain of cognition (Table 3).
sion and exclusion criteria were reported, whether These measures were grouped together as perfor-
participants were blind to research hypotheses, mance relied on the ability to hold and manipulate
and whether outcome assessors were blinded to to-be-remembered information in the short-term
group allocation during the assessment. An analy- (Baddeley 2003). The outcome measures can be
sis of whether participants were representative of differentiated by aspects such as their demand on
the target population was rarely analyzed or other cognitive processes (e.g., Digit Span Forward
reported. In many cases, the target population vs. Digit Span Backward; Jung et al. 2017) or
was not explicit, though this is likely due to the whether performance is based on verbal (e.g.,
frequency of university students recruited as Digit Span Backwards) or visuospatial memory
participants. processes (Forward Spatial Span; Han 2017).
As illustrated in Figure 4 and Table 4, natural
environments or stimuli were found to have a sig-
Part two - meta-analysis: the effect of natural
nificantly positive effect on working memory per-
environments on cognitive performance
formance when effect sizes were pooled across all
To synthesize the most relevant data, only studies levels of moderators (n = 22), with a pooled esti-
that reported baseline measures of cognitive per- mated mean effect size (Hedges’ g) of 0.162. As seen
formance were included. Our results were supple- in Figure 4B (supplementary material), the positive
mented with data from the studies identified by effect remains (Hedges’ g = 0.172) when only
Ohly and colleagues (2016) that also included including studies with reported baseline balance
baseline measures of cognitive performance in (n = 10). A significant difference was found
order to conduct our meta-analyses. Across both between studies in which participants’ restoration
reviews, data from a total of 19 different cognitive potential was heightened and studies where restora-
tasks were included, spanning 7 domains of cogni- tion potential was normal. As Figure 4C (supple-
tion. A total of 49 separate outcome measures were mentary material) demonstrates, studies that
derived from the cognitive tasks. One conglomer- included participants with heightened restoration
ate score of attention performance was not potential reported a higher pooled effect size
included (Cimprich and Ronis 2003) as the results (Hedges’ g = 0.307) than those with normal restora-
from its subtasks were included separately. tion potential (Hedges’ g = 0.052). Studies that used
Results are presented in separate sections based actual exposure to real environments showed sig-
on each of the 8 general cognitive domains nificant positive effects (Hedges’ g = 0.156) in favor
assessed by the outcome measures identified in of natural environments, (n = 13) as presented in
the review (i.e., working memory, attentional con- Figure 4D (supplementary material). However,
trol, visual attention processes, vigilance, cognitive there was no significant difference in effect sizes
flexibility, impulse control, processing speed, and between actual and virtual exposures.
other emerging domains). First, general character-
istics of the studies are presented, including
Attentional control
means, SD and sample sizes, and characteristics
of the moderators to be considered for analysis. A total of 10 outcome measures derived from 4
The meta-analysis based on post-test scores is then cognitive tasks were categorized as being relevant
reported as well as additional moderator analyses, to attentional control (see Table 5). This cognitive
where possible. For cognitive domain sections domain is sometimes referred to as executive
where meta-analysis was not appropriate, a narra- attention, focused attention, selective attention,
tive analysis is provided. interference control, and inhibition at the level of
attention (Diamond 2013). A key component of
these tasks is the need to direct attention towards a
Working memory
target while simultaneously ignoring distractors.
A total of 7 outcome measures derived from 4 Therefore, these measures may be most directly
tasks were categorized as being relevant to the related to the effortful type of attention described
Table 3. Data and moderator details from included studies using outcome measures classified in the working memory domain.
Control environment/
Natural environment/stimuli stimuli Moderators
Pre Post Pre Post Restoration potential Baseline Exposure
Cognitive task Author, Year, (sub- Source of Induction Baseline Baseline Baseline Exposure
(outcome measure) study) sub-group n mean (SD) mean (SD) n mean (SD) mean (SD) RP length time balance average type
DSB (span score) Berman, Jonides, and - 37 7.90 (2.28) 9.40 (2.49) 37 7.90 (1.85) 8.4 (2.01) task 35ʹ00” T1 yes 7.90 actual
Kaplan (2008) (1)
Berman, Jonides, and - 12 7.92 (3.33) 9.33 (2.98) 12 7.83 (3.60) 8.83 (3.12) none - T2† partial 7.88 virtual
Kaplan (2008) (2)
Berman et al. (2012) - 19 7.42 (3.00 8.63 (2.87) 19 8.26 (2.51) 7.84 (2.24) participant NR T1 no 7.84 actual
+ task
Cimprich and Ronis - 83 4.99 (1.37) 5.20 (1.28) 74 4.51 (1.12) 4.58 (1.20) participant - T2* no 4.75 actual
(2003)
Emfield and Neider sound 28 5.3 (1.44) 6.11 (1.42) 28 5.75 (1.04) 5.79 (1.23) none - T2† yes 5.53 virtual
(2014)
pictures 27 5.28 (0.98) 5.52 (1.33) 27 5.64 (1.22) 5.60 (1.23 none - T2† yes 5.46 virtual
sound + pictures 28 5.40 (1.16) 5.88 (1.20) 27 5.52 (1.16) 5.44 (1.25) none - T2† yes 5.46 virtual
Gamble, Howard, and older adults 15 5.78 (1.72) 7.11 (1.69) 15 6.09 (2.02) 7.00 (2.61) none - T2† partial 5.94 virtual
Howard (2014)
younger adults 13 7.91 (2.43) 8.27 (2.61) 13 7.56 (2.60) 9.33 (2.65) none - T2† partial 7.74 virtual
Han (2017) collapsed over activity 58 9.84 (2.85) 10.52 58 9.83 (3.11) 10.67 none - T2† yes 9.84 actual
conditions (2.62) (2.50)
Jung et al. (2017) healthy adults 20 5.05 (0.95) 5.40 (1.57) 20 4.55 (0.89) 5.00 (1.45) none - T2† partial 4.80 virtual
heart failure patients 20 4.40 (1.54) 4.55 (1.79) 20 4.30 (1.75) 4.85 (1.42) participant - T2* partial 4.35 virtual
Lin et al. (2014) collapsed over 104 6,43 (1.53) 7,31 (1,47) 34 7.12 (1.45) 6.53 (1.38) participant - T2* yes 6.78 virtual
awareness conditions
Stark (2003) - 29 5.1 (1.3) 5.4 (1.6) 28 4.7 (1.1) 5.0 (1.6) participant - T2* partial 4.90 actual
DSB (total correct/
14)
Triguero-Mas et al. after 30” exposure 52 4,54 (2,5) 5,25 (2,5) 26 5.21 (2.80) 5.65 (2.66) none - T2† yes 4.88 actual
(2017)
after 180” exposure 52 4,54 (2,5) 5,37 (2,68) 26 5.21 (2.80) 5.88 (2.83) none - T2† yes 4.88 actual
DSB (total correct/9) Rogerson et al. (2016) - 24 3.2 (1.7) 3.5 (1.7) 24 3.5 (1.4) 3.0 (1.6) none T2† yes 3.35 actual
DSF (span score) Cimprich and Ronis - 83 6.71 (1.26) 6.98 (1.37) 74 6.54 (1.12) 6.53 (1.38) participant - T2* no 6.63 actual
(2003)
Jung et al. (2017) healthy adults 20 6.85 (1.18) 6.90 (1.41) 20 6.80 (1.15) 7.25 (1.21) none - T2† partial 6.83 virtual
heart failure patients 20 6.80 (1.44) 6.55 (1.19) 20 6.70 (1.22) 6.55 (1.15) participant - T2* partial 6.75 virtual
Stark (2003) - 29 6.8 (1.2) 7.1 (1.4) 28 6.6 (1.0) 6.7 (1.1) participant - T2* partial 6.70 actual
FSS (span score) Han (2017) collapsed over activity 58 9.26 (1.77) 9.97 (1.94) 58 8.78 (1.83) 9.47 (1.40) none - T2† yes 9.02 actual
conditions
RST Evensen et al. (2015) window view, T1 31 0.65 (0.11) 0.71 (0.10) 19 0.65 (0.11) 0.71 (0.10) task 15ʹ00” T1 yes 0.65 actual
baseline
JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B

window view, T2 31 0.69 (0.09) 0.71 (0.10) 19 0.69 (0.09) 0.71 (0.10) task 15ʹ00” T2* yes 0.69 actual
baseline
(Continued )
243
244 M. P. STEVENSON ET AL.

in ART. Three of these tasks (Stroop, Attention

Notes. DSB = Digit Span Backwards; DSF = Digit Span Forwards; FSS = Forward Spatial Span; RST = Reading Span Task; *Participants with heightened restoration potential at baseline measurement;
Exposure
Exposure

type
Network Task (ANT), and Multi-Source

actual

actual

actual

actual
Interference Task (MSIT) involve two trial types,
congruent and incongruent. Incongruent trials

Baseline
average
0.64

0.67

11.38

11.17
involve targets and distractors designed to evoke
cognitive conflict that must be overcome using
attentional control processes. Recruitment of
Baseline
Induction Baseline Baseline
balance

partial

partial
yes these processes is reflected in response times to

yes
such trials being typically slower. Conversely, con-
Moderators

gruent trials involve harmonious targets and dis-

time

T2† tractors where responses reflect cognitive

T2†
T2*
15ʹ00” T1

processes without this conflict resolution.

Restoration potential

15ʹ00”
length

Optimally, outcome measures take both trial

-

types into account when calculating attentional

control scores to adjust for general processing
Source of

speed.
RP

none

none
31 0.64 (0.10) 0.69 (0.07) 19 0.64 (0.07) 0.72 (0.09) task

31 0.66 (0.09) 0.69 (0.07) 19 0.68 (0.09) 0.72 (0.09) task

As illustrated by Figure 5 and Table 6, natural

environments or stimuli were found to exert a sig-
n mean (SD) mean (SD) n mean (SD) mean (SD)
Control environment/

Post

nificant positive effect on attentional control when

(2.89)

(3.22)
12.17

11.25

effect sizes were pooled across all levels of modera-

stimuli

tors (n = 17), with a pooled estimated mean effect

size (Hedges’ g) of −0.156. As Figure 5B (supple-
Pre

(2.76)

(2.89)
6 11.92

6 10.67

mentary material) shows, the significant effect was

lost when only including studies with reported
baseline balance (n = 7). Figure 5C (supplementary
Natural environment/stimuli
Post

material) illustrates that no significant differences

(2.31)

(3.68)
11.92

10.58

were found in relation to restoration potential. As

noted in Figure 5D, investigations that used actual
exposure to real environments exhibited signifi-
Pre

(2.32)

(3.82)
6 10.83

6 11.67

†Participants without heightened restoration potential at baseline measurement.

cantly positive effects (Hedges’ g = −2.612) in

favor of natural environments (n = 7); however,
there was no significant difference in effect sizes
no window view, T1

no window view, T2

between actual and virtual exposures.

sub-group

baseline

baseline

women

Visual attention
Bodin and Hartig (2003) men

Seven outcome measures derived from two cogni-

Author, Year, (sub-

tive tasks were categorized as being related to

visual processing (Table 7). Six of the outcome
study)

measures were derived from the Functional Field

of View Task (Emfield and Neider 2014), which
measures how distance and placement of a target
Table 3. (Continued).

stimulus from a fixation point affect speed and

(outcome measure)

accuracy of short-term spatial recall. Thus, the

Cognitive task

(composite)

task is believed to indicate the spatial area in

DSF + DSB

which stimuli can be attended to (Ball et al.

1988). The final outcome measure was the orient-
ing score from the ANT, which assesses the ability
 JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B 245

Figure 4. Overall forest plot showing pooled effect sizes (Hedges’ g) for outcome measures related to working memory. No
moderator variables are included.

Table 4. Summary of results from the working memory meta-analysis of post-test scores and moderator analyses exploring the
influence of baseline balance, restoration potential, and exposure type.
Working Memory
n Hedge’s G (95% CI) I2 Z-value p-value Test of heterogeneity between subgroups
Overall 22 0.162 (0.053 – 0.270) 0.00 2.924 0.003 -
Moderators
Baseline balance Q = 4.535, DF = 2, p = 0.104
yes 10 0.172 (0.005 – 0.339) 26.37 2.020 0.043
partial 10 0.010 (−0.200 – 0.219) 0.00 0.090 0.928
no 2 0.389 (0.107 – 0.670) 0.00 2.705 0.007
unsure - - - - -
Restoration potential Q = 5.211, DF = 1, p = 0.022
heightened 8 0.307 (0.142 – 0.472) 0.00 3.641 0.000
normal 14 0.052 (−0.092 – 0.195) 0.00 0.710 0.478
Exposure type Q = 0.018, DF = 1, p = 0.894
actual 13 0.156 (0.024 – 0.289) 0.00 2.310 0.021
virtual 9 0.172 (−0.016 – 0.360) 0.00 1.797 0.072

to select visual information from sensory input to vigilance (Table 9). This cognitive domain is
(Fan et al. 2002). sometimes referred to as sustained attention and
As illustrated in Figure 6 and Table 8, natural describes tasks where participants need to main-
environments or stimuli were found to exert no tain alertness over longer periods of time in order
marked effect on visual attention measures when to detect irregular or unpredictable events (Sarter,
effect sizes were pooled across all levels of moderators. Givens, and Bruno 2001). Outcome measures
Baseline balance was the only moderator that had derived from performance requiring a simple
sufficient data across at least two levels. However, as response over an extended time were included in
Figure 6B (supplementary material) demonstrates, no this cognitive domain. Most outcome measures
marked effect of environment for studies with full were taken from the Sustained Attention to
(n = 3) or partial (n = 3) baseline balance was found. Response Task, where participants need to
respond to all stimuli except a target; however
congruent trials of the MSIT (Jung et al. 2017)
Vigilance and the alerting score of the ANT (Berman,
A total of 7 outcome measures derived from three Jonides, and Kaplan 2008; Gamble, Howard, and
cognitive tasks were categorized as being relevant Howard 2014) were also considered relevant. Data
 246

Table 5. Data and moderator details from included studies using outcome measures classified in the attentional control domain.
Natural environment/stimuli Control environment/stimuli Moderators
Pre Post Pre Post Restoration potential Baseline Exposure
Cognitive task
(outcome Author, Year, Source of Induction Baseline Baseline Baseline Exposure
measure) (sub-study) sub-group n mean (SD) mean (SD) n mean (SD) mean (SD) RP length time balance average type
NCPC (number of Greenwood and collapsed over 60 5.76 (1.99) 3.84 (1.69) 60 5.32 (2.32) 4.59 (2.46) task 17”00 T2 unsure 5.54 Actual
reversalsa) Gatersleben context
(2016) conditions
Hartig et al. post test during 28 4.17 (1.92) 4.06 (1.90) 28 3.70 (1.35) 4.43 (1.67) none - T2† yes 3.94 Actual
M. P. STEVENSON ET AL.

(2003) exposure, no
fatigue
post test after 28 4.17 (1.92) 3.88 (2.21) 28 3.70 (1.35) 4.09 (1.72) none - T2† yes 3.94 Actual
exposure, no
fatigue
post test during 28 4.37 (1.92) 3.96 (1.93) 28 3.87 (1.85) 4.76 (2.19) task 40”00 T1 yes 4.12 Actual
exposure, fatigue
post test after 28 4.37 (1.92) 3.98 (2.44) 28 3.87 (1.85) 4.67 (2.63) task 40”00 T1 yes 4.12 Actual
exposure, fatigue
NCPC (number of Sahlin et al. (2016) - 51 5.27 (2.25) 4.78 (2.74) 51 5.73 (2.82) 5.61 (3.63) none - T2† unsure 5.50 Actual
reversalsb)
NCPC (% Cimprich and - 83 10.29 (60.22) 19.48 (37.81) 74 17.87 (66.50) 13.87 (62.88) participants - T2* no 14.08 Actual
reductionc) Ronis (2003)
ANT (executive Berman, Jonides, - 12 86 (39.14) 67 (29.27) 12 81 (53.69) 93 (62.22) none - T2† partial 83.50 Virtual
score) and Kaplan (2008)
(2)
Kelz, Evans, and - 72 137.25 (42.25) 104.79 (34.61) 61 165.14 (65.74) 114.00 (33.58) none - T2† no 151.20 Actual
Röderer (2015)
Gamble, Howard, older adults 15 146.77 (59.53) 101.65 (51.74) 15 103.07 (47.45) 95.22 (41.87) none - T2† partial 124.92 Virtual
and Howard
(2014)
younger adults 13 130.40 (37.33) 104.05 (37.93) 13 106.07 (54.88) 104.03 (54.61) none - T2† partial 118.24 Virtual
Emfield and sound 28 105 (36.0) 102 (35.7) 28 118 (36.8) 107 (35.5) none - T2† yes 111.50 Virtual
Neider (2014)
pictures 27 110 (52.6) 101 (52.2) 27 113 (32.9) 106 (34.7) none - T2† yes 111.50 Virtual
sound + pictures 28 117 (54.7) 124 (49.9) 27 114 (41.5) 110 (30.4) none - T2† yes 115.50 Virtual
Stroop Task (% Jung et al. (2017) healthy adults 20 11.70 (9.82) 4.24 (4.38) 20 10.82 (10.47) 5.70 (7.26) none - T2† partial 11.26 Virtual
error,
incongruent)
heart failure 20 23.61 (18.28) 18.19 (18.30) 20 19.72 (15.81) 14.31 (16.03) participants - T2* partial 21.67 Virtual
patients
Stroop Task (RT, Jung et al. (2017) healthy adults 20 1630 (510) 1445 (467) 20 1595 (522) 1407 (423) none - T2† partial 1612.50 Virtual
incongruent)
heart failure 20 1641 (541) 1616 (513) 20 1598 (353) 1659 (387) participants - T2* partial 1619.50 Virtual
patients
(Continued )
Table 5. (Continued).
Natural environment/stimuli Control environment/stimuli Moderators
Pre Post Pre Post Restoration potential Baseline Exposure
Cognitive task
(outcome Author, Year, Source of Induction Baseline Baseline Baseline Exposure
measure) (sub-study) sub-group n mean (SD) mean (SD) n mean (SD) mean (SD) RP length time balance average type
Stroop Task (total Jenkin et al. - 25 45.68 (3.21) 47.44 (2.52) 25 44.76 (3.64) 46.20 (3.38) none - T2† yes 45.22 Virtual
correct, (2017) (1)
incongruent)
Jenkin et al. - 26 40.12 (11.48) 42.65 (9.58) 28 37.43 (11.34) 39.82 (10.92) none - T2† yes 38.78 Virtual
(2017) (2)
Stroop Task Geniole et al. - 31 −51.31 (80.67) −24.45 (71.42) 31 −44.50 (70.43) −29.95 (57.88) none - T2† unsure −47.91 Actual
(interference (2016)
score)
MSIT (% error, Jung et al. (2017) healthy adults 20 4.13 (4.00) 1.77 (1.99) 20 3.67 (3.93) 2.13 (2.20) none - T2† partial 3.90 Virtual
incongruent)
heart failure 20 4.79 (3.40) 4.59 (4.42) 20 4.67 (2.42) 4.27 (3.92) participants - T2* partial 4.73 Virtual
patients
MSIT (RT, Jung et al. (2017) healthy adults 20 921 (148) 890 (129) 20 921 (152) 886 (154) none - T2† partial 921 Virtual
incongruent)
heart failure 20 1047 (150) 1019 (145) 20 1043 (123) 1015 (128) participants - T2* partial 1045 Virtual
patients
Notes. NCPC = Necker Cube Pattern Control; ANT = Attention Network Task; MSIT = Multi-Source Interference Task. *Participants with heightened restoration potential at baseline measurement;
†Participants without heightened restoration potential at baseline measurement.
a
Average of 2x30s periods.
b
One 30 s period.
c
From baseline to holding conditions.
JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B
247
248 M. P. STEVENSON ET AL.

Figure 5. Overall forest plot showing pooled effect sizes (Hedges’ g) for outcome measures related to attentional control. No
moderator variables are included.

Table 6. Summary of results from the attentional control meta-analysis of post-test scores and moderator analyses exploring the
influence of baseline balance, restoration potential, and exposure type.
Attentional Control
n Hedge’s G (95% CI) I2 Z-value p-value Test of heterogeneity between subgroups
Overall 17 −0.156 (−0.272 – 0.039) 0 −2.618 0.009 -
Moderators
Baseline balance Q = 0.848, DF = 3, p = 0.838
yes 7 −0.132 (−0.331 – 0.066) 0 −1.306 0.191
partial 5 −0.052 (−0.355 – 0.251) 0 −0.337 0.736
no 2 −0.182 (−0.412 – 0.048) 0 −1.548 0.122
unsure 3 −0.221 (−0.454 – 0.013) 1.33 −1.852 0.064
Restoration potential Q = 0.191, DF = 1, p = 0.662
heightened 4 −0.192 (−0.395 – 0.010) 0 −1.866 0.062
normal 13 −0.137 (−0.280 – 0.005) 0 −1.888 0.059
Exposure type
actual 7 −0.199 (−0.348 – −0.050) 0 −2.612 0.009 Q = 0.825, DF = 1, p = 0.364
virtual 10 −0.088 (−0.275 – 0.098) 0 −0.926 0.355

for a well-known outcome measure in vigilance impulse control (Table 11). This cognitive
research, d-prime, which measures the ability to domain is related to self-control and response
discriminate targets from other stimuli, was only inhibition, sometimes termed behavioral inhibi-
available from one publication (Berto 2005). tion, and refers to the ability to control behavior
As shown in Figure 7 and Table 10, natural rather than thought or attentional processes
environments or stimuli were found to exert no (Diamond 2013). Outcome measures were cate-
marked effect on vigilance when effect sizes were gorized as assessing impulse control if they assess
pooled across all levels of moderators (n = 8). Due the ability to exert inhibition of a pre-potent
to a lack of sufficient data across more than two response. The measures identified include errors
levels of baseline balance and restoration potential, of commission on the SART where the partici-
moderator analyses were not performed. The pant failed to override the pre-potent response
influence of exposure type could not be explored tendency to respond to non-target stimuli and
as all studies used virtual exposures. mistakenly responded to a target, for which the
correct response is to withhold. The other out-
come measures identified is the reverse score on
Impulse control the same task, where correct responses to targets
Two outcome measures derived from one cogni- were tallied so that a higher count reflects better
tive task were categorized as being relevant to impulse control.
Table 7. Data and moderator details from included studies using outcome measures classified in the visual attention domain.
Natural environment/stimuli Control environment/stimuli Moderators
Restoration
Pre Post Pre Post potential Baseline Exposure
Cognitive task Author, Year, (sub- Source Induction Baseline Baseline Baseline Exposure
(outcome measure) study) sub-group n mean (SD) mean (SD) n mean (SD) mean (SD) of RP length time balance average type
ANT (orienting Berman, Jonides, and - 12 47 (22.38) 55 (25.39) 12 46 (34.68) 43 (16.39) none - T2† partial 46.50 virtual
score) Kaplan (2008) (2)
Gamble, Howard, and older adults 15 41.59 (32.43) 71.72 (38.70) 15 67.80 (53.88) 91.54 (123.77) none - T2† partial 54.70 virtual
Howard (2014)
younger adults 13 64.12 (29.72) 50.67 (33.55) 13 35.78 (29.30) 45.18 (35.81) none - T2† partial 49.95 virtual
FFOV (RT, 10⁰) Emfield and Neider sound 28 935 (253) 811 (232) 28 951 (240) 779 (204) none - T2† yes 943.00 virtual
(2014)
pictures 27 933 (250) 739 (200) 27 823 (246) 817 (288) none - T2† yes 878.00 virtual
sound 28 961 (304) 817 (288) 27 909 (278) 740 (200) none - T2† yes 935.00 virtual
+ pictures
FFOV (accuracy, Emfield and Neider sound 28 0.768 (0.309) 0.848 (0.151) 28 0.791 (0.204) 0.912 (0.113) none - T2† yes 0.780 virtual
10⁰) (2014)
pictures 27 0.766 (0.232) 0.864 (0.151) 27 0.729 (0.316) 0.814 (0.249) none - T2† yes 0.748 virtual
sound 28 0.746 (0.320) 0.846 (0.266) 27 0.828 (0.231) 0.896 (0.126) none - T2† yes 0.787 virtual
+ pictures
FFOV (RT, 20⁰) Emfield and Neider sound 28 1018 (294) 834 (236) 28 941 (224) 790 (208) none - T2† yes 979.50 virtual
(2014)
pictures 27 959 (269) 743 (210) 27 949 (292) 804 (229) none - T2† yes 954.00 virtual
sound 28 979 (300) 816 (279) 27 914 (263) 756 (219) none - T2† yes 946.50 virtual
+ pictures
FFOV (accuracy, Emfield and Neider sound 28 0.651 (0.283) 0.759 (0.277) 28 0.703 (0.243) 0.849 (0.144) none - T2† yes 0.677 virtual
20⁰) (2014)
pictures 27 0.681 (0.293) 0.824 (0.171) 27 0.659 (0.321) 0.754 (0.284) none - T2† yes 0.670 virtual
sound 28 0.698 (0.311) 0.800 (0.275) 27 0.800 (0.225) 0.861 (0.148) none - T2† yes 0.749 virtual
+ pictures
FFOV (RT, 30⁰) Emfield and Neider sound 28 1064 (381) 851 (258) 28 989 (249) 804 (210) none - T2† yes 1026.50 virtual
(2014)
pictures 27 989 (263) 795 (237) 27 999 (297) 834 (252) none - T2† yes 994.00 virtual
sound 28 983 (317) 820 (308) 27 934 (274) 767 (249) none - T2† yes 958.50 virtual
+ pictures
FFOV (accuracy, Emfield and Neider sound 28 0.543 (0.238) 0.662 (0.200) 28 0.535 (0.217) 0.627 (0.174) none - T2† yes 0.539 virtual
30⁰) (2014)
pictures 27 0.511 (0.268) 0.619 (0.247) 27 0.524 (0.268) 0.608 (0.256) none - T2† yes 0.518 virtual
sound 28 0.564 (0.306) 0.634 (0.284) 27 0.659 (0.149) 0.710 (0.147) none - T2† yes 0.612 virtual
+ pictures
JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B

Notes. ANT = Attention Network Task; FFOV = Functional Field of View. *Participants with heightened restoration potential at baseline measurement; †Participants without heightened restoration potential
at baseline measurement.
249
250 M. P. STEVENSON ET AL.

Figure 6. Overall forest plot showing pooled effect sizes (Hedges’ g) for outcome measures related to visual attention. No moderator
variables are included.

Table 8. Summary of results from the visual attention meta-analysis of post-test scores and moderator analyses exploring the
influence of baseline balance, restoration potential, and exposure type.
Visual Attention
n Hedge’s G (95% CI) I2 Z-value p-value Test of heterogeneity between subgroups
Overall 6 0.100 (−0.147 – 0.347) 0 0.793 0.427
Moderators
Baseline balance Q = 0.031, DF = 1, p = 0.859
yes 3 0.083 (−0.218 – 0.386) 0 0.546 0.585
partial 3 0.132(−0.296 – 0.559) 0 0.602 0.547
no - - - - - -
unsure - - - - - -

Results of the meta-analysis pooled across all require participants to fill-in as many letters or
levels of the moderator variables (Figure 8 and digits as possible in a given time period, based on
Table 12) show that natural stimuli were found a letter-digit- or symbol-key provided. There were 4
to exert no marked effect on impulse control, versions of these tasks used in the meta-analysis.
(n = 5). As illustrated in Figure 8B (supplementary Results of the meta-analysis pooled across all levels
material), there was also no effect of the environ- of the moderator variables (Table 14 and Figure 9)
ment when studies were analyzed based upon full demonstrate that natural environments or stimuli
(n = 3), or partial (n = 2), baseline balance. Due to exerted no effect on processing speed, (n = 10). As
a lack of sufficient data across more than two observed in Figure 9B (supplementary material), there
levels of restoration potential exposure type, was also no significant effect of the environment when
further moderator analyses were not performed. studies were analyzed based on full (n = 2), partial
(n = 5), or no (n = 3) baseline balance. As Figure 9C
and 9D (supplementary material) illustrate, there
Processing speed were no significant effects found when analyzing stu-
A total of 8 outcome measures derived from 7 dies based on restoration potential or exposure type.
cognitive tasks were categorized as being relevant
to processing speed (Table 13). Outcome measures
Cognitive flexibility
were categorized in this cognitive domain if they
involved timed performance on tasks that were Three outcome measures derived from two cognitive
relatively short in duration, that is, not tasks mea- tasks were categorized as being relevant to processing
suring vigilance or sustained attention. The speed speed (Table 15). Outcome measures were included in
by which the tasks were performed may depend on this domain if performance involved switching
mechanisms closely related to cognitive processing between two cognitive sets, or sets of rules, within a
(e.g., Stroop Task, congruent trials) or occur some- task. This ability is sometimes called set-shifting and is
where in conjunction of cognitive and motor pro- classically assessed using the Wisconsin Card Sorting
cessing (e.g., substitution tasks). Substitution tasks Task (Milner 1963). The Trail Making Task, version B,
Table 9. Data and moderator details from included studies using outcome measures classified in the vigilance domain.
Natural environment/stimuli Control environment/stimuli Moderators
Pre Post Pre Post Restoration potential Baseline Exposure
Cognitive task
(outcome Author, Year, (sub- Source of Induction Baseline Baseline Baseline Exposure
measure) study) sub-group n mean (SD) mean (SD) n mean (SD) mean (SD) RP length time balance average type
ANT (alerting Berman, Jonides, and - 12 32 (23.76) 31 (18.12) 12 36 (22.59) 46 (38.18) none - T2† partial 34.00 virtual
score) Kaplan (2008) (2)
Gamble, Howard, older 13 25.81 (55.46) 45.87 (38.32) 13 49.94 (40.36) 49.17 (34.99) none - T2† partial 37.88 virtual
and Howard (2014) adults
younger 15 34.97 (26.26) 34.08 (32.60) 15 8.79 (48.16) 15.25 (30.47) none - T2† partial 21.88 virtual
adults
SART (d-prime) Berto (2005) (1) - 16 1.40 (0.71) 1.86 (0.89) 16 1.97 (0.96) 2.00 (0.95) none - T2† partial 1.69 virtual
Berto (2005) (3) - 16 2.12 (1.08) 2.47 (1.04) 16 1.79 (1.24) 2.03 (0.93) none - T2† partial 1.96 virtual
SART (RT) Berto (2005) (1) - 16 313.17 (38.36) 267.38 (73.78) 16 319.59 (70.98) 299.61 (41.43) none - T2† partial 316.38 virtual
Berto (2005) (3) - 16 311.27 (35.6) 302.22 (32.09) 16 306.21 (78.79) 297.91 (52.98) none - T2† partial 308.74 virtual
SART (RT, first Lee et al. (2015) - 75 551.64 (149.91) 526.73 (159.35) 75 541.64 (139.69) 525.01 (172.34) none - T2† yes 546.64 virtual
half)
SART (RT, second Lee et al. (2015) - 75 526.35 (157.01) 519.29 (168.79) 75 525.92 (160.56) 501.56 (158.05) none - T2† yes 526.14 virtual
half)
SART (SD of RT, Lee et al. (2015) - 75 97.30 (36.37) 90.67 (31.09) 75 95.37 (34.29) 102.27 (42.26) none - T2† yes 96.34 virtual
first half)
SART (SD of RT, Lee et al. (2015) - 75 101.24 (43.73) 96.41 (35.07) 75 107.38 (41.14) 116.46 (49.10) none - T2† yes 104.31 virtual
second half)
SART (omission Lee et al. (2015) - 75 0.49 (1.30) 0.37 (0.95) 75 0.65 (2.60) 0.63 (1.65) none - T2† yes 0.57 virtual
errors, first half)
SART (omission Lee et al. (2015) - 75 0.63 (1.73) 0.68 (1.39) 75 0.50 (1.13) 1.43 (1.47) none - T2† yes 0.57 virtual
errors, second
half)
MSIT (% error, Jung et al. (2017) healthy 20 0.36 (1.60) 0.00 (0.00) 20 0.05 (0.24) 0.05 (0.21) none - T2† partial 0.21 virtual
congruent) adults
heart 20 0.56 (1.18) 1.24 (4.42) 20 0.48 (0.88) 1.08 (1.96) participants - T2* partial 0.52 virtual
failure
patients
MSIT (RT, Jung et al. (2017) healthy 20 661 (93) 629 (79) 20 662 (122) 636 (127) none - T2† partial 661.50 virtual
congruent) adults
heart 20 755 (143) 737 (123) 20 741 (116) 721 (114) participants - T2* partial 748.00 virtual
failure
patients
Notes. ANT = Attention Network Task; SART = Sustained Attention to Response Task; MSIT = Multi-Source Interference Task. *Participants with heightened restoration potential at baseline measurement;
JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B

†Participants without heightened restoration potential at baseline measurement

251
252 M. P. STEVENSON ET AL.

Figure 7. Overall forest plot showing pooled effect sizes (Hedges’ g) for outcome measures related to vigilance. No moderator
variables are included.

Table 10. Summary of results from the vigilance meta-analysis of post-test scores and moderator analyses exploring the influence of
baseline balance, restoration potential, and exposure type.
Vigilance
2
n Hedge’s G (95% CI) I Z-value p-value Test of heterogeneity between subgroups
Overall 8 −0.124 (−0.325̶ 0.077) 0 −1,214 0.225 -
Moderators Insufficient data

involves switching between letter- or number-based Therefore, it was not possible to conduct meta-
cognitive sets, whereas the Switching Stroop Task, analyses using their outcome measures. Instead,
requires participants to switch between words and these represent potential new directions in atten-
colors. tion restoration research where further studies need
As shown in Figure 10 and Table 16, natural envir- to be conducted before assessing their suitability.
onments or stimuli were found to produce a signifi- Table 17 summarizes the basic statistics and mod-
cantly positive effect on cognitive flexibility when erator characteristics of the three tasks.
effect sizes were pooled across all levels of moderators The Delay of Gratification Task (Jenkin et al.
(n = 6), with a pooled estimated mean effect size 2017) measures the ability to effortfully control
(Hedges’ g) of −0.317. As Figure 10B (supplementary thoughts and behavior regarding a temporally sali-
material) shows, the positive effect remains (Hedges’ ent reward in order to receive a greater reward at a
g = −0.610) when only including studies with reported later time. It is classified within tasks related to self-
baseline balance (n = 2). Figure 10C (supplementary regulation and was shown to predict positive out-
material) demonstrates no significant effect found comes later in life (Mischel, Shoda, and Rodriguez
when only including participants with heightened 1989). Jenkin and colleagues (2017) reported a sig-
restoration; however, significance was reached for nificant improvement in this ability after viewing
studies with participants of normal restoration poten- videos of natural environments when compared to
tial, (n = 3). Studies that used actual exposure to real video of urban environments containing little traf-
environments reported significant positive effects fic and few people (study 1), but not videos of urban
(Hedges’ g = −0.417) in favor of natural environments environments with higher traffic flow and more
(n = 4), as seen in Figure 10D (supplementary mate- people to create more noise (study 2).
rial). However, the subgroup difference between Taylor’s Aggression Paradigm was used by Wang
actual and virtual exposure types did not reach and colleagues (2017) to measure provoked and
significance. unprovoked aggression after watching videos of nat-
ural and urban environments. Using ego-depletion
(or task-induced cognitive fatigue) as a between-
Emerging cognitive domains
subjects factor, they found that participants with
Three cognitive tasks that fulfilled the inclusion heightened restoration potential displayed lower
criteria for the meta-analysis were only used once. levels of provoked aggression after watching natural
Table 11. Data and moderator details from included studies using outcome measures classified in the impulse control domain.
Natural environment/stimuli Control environment/stimuli Moderators
Pre Post Pre Post Restoration potential Baseline Exposure
Cognitive task Author, Year, (sub- Source of Induction Baseline Baseline Baseline Exposure
(outcome measure) study) sub-group n mean (SD) mean (SD) n mean (SD) mean (SD) RP length time balance average type
SART (commission Berto (2005) (1) - 16 1.81 (3.83) 2.06 (4.79) 16 3.25 (6.22) 1.62 (4.96) none - T2† partial 2.53 virtual
errorsa)
Berto (2005) (3) - 16 1.62 (2.5) 0.81 (1.42) 16 1.68 (2.62) 0.75 (1.52) none - T2† partial 1.65 virtual
SART (commission Lee et al. (2015) - 75 1.97 (2.17) 2.76 (2.68) 75 2.54 (2.68) 3.42 (3.20) none - T2† yes 2.26 virtual
errorsa, first half)
SART (commision Lee et al. (2015) - 75 2.52 (2.42) 2.83 (2.68) 75 2.86 (2.60) 3.32 (2.68) none - T2† yes 2.69 virtual
errorsa, second
half)
SART (total correct Berto (2005) (1) - 16 11.68 (5.28) 13.62 (5.37) 16 13.25 (5.09) 13.00 (5.4) none - T2† partial 12.47 virtual
response to
target)
Berto (2005) (3) - 16 14.81 (5.55) 17.12 (4.09) 16 12.50 (6.36) 14.56 (5.95) none - T2† partial 13.66 virtual
SART (% total Craig et al. (2015) High depression 12 63.19 (26.76) 63.54 (29.52) 12 66.32 (20.06) 62.85 (26.32) participant - T2* yes 64.76 virtual
correct response symptom rating
to target)
Typical 12 63.54 (13.19) 76.74 (18.76) 12 64.58 (17.81) 47.57 (17.27) none - T2* yes 64.06 virtual
depression
symptom rating
Notes. SART = Sustained Attention to Response Task. *Participants with heightened restoration potential at baseline measurement; †Participants without heightened restoration potential at baseline
measurement.
a
Incorrect response to target (unsuccessfully withheld response).
JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B
253
254 M. P. STEVENSON ET AL.

Figure 8. Overall forest plot showing pooled effect sizes (Hedges’ g) for outcome measures related to impulse control. No moderator
variables are included.

Table 12. Summary of results from the impulse control meta-analysis of post-test scores and moderator analyses exploring the
influence of baseline balance, restoration potential, and exposure type.
Impulse Control
n Hedge’s G (95% CI) I2 Z-value p-value Test of heterogeneity between subgroups
Overall 5 0.333 (−0.760 – 0.094) 56.68 −1.530 0.216
Moderators
Baseline balance Q = 0.778, DF = 1, p = 0.378
yes 3 −0.528 (−1.302 – 0.247) 76.65 −1.335 0.182
partial 2 −0.118 (−0.579 – 0.362) 0 −0.480 0.631
no - - - - - -
unsure - - - - - -

environments compared to urban environments. For effect and consider features of associated outcome
participants with normal restoration potential, pro- measures to better describe processes related to
voked aggression was higher after watching natural directed attention, the resource purported to be
environments. No significant effects were found for restored. Our search revealed a substantial increase
the unprovoked aggression measure. in interest and productivity within the area. This
Finally, Chow and Lau (2015) used the Graduate suggests an urgency to address the ambiguity sur-
Record Exam (GRE) to measure logical reasoning rounding directed attention as the cognitive
abilities before and after viewing digital photographs resource underlying the restoration effect, and the
of natural and urban scenes. The GRE is used to diversity of protocols used to measure it. A meta-
screen applicants for graduate studies in the United analysis was conducted on all outcome measures
States and requires responders to apply mental mod- from studies identified by ourselves and Ohly and
els, make inferences, and draw logical conclusions to colleagues (2016) that included baseline measure-
solve the problems (Schmeichel, Vohs, and ment. When analyzing across all levels of baseline
Baumeister 2003). By analyzing change scores, Chow balance, improved performance was found for the
and Lau (2015) reported a significant positive effect domains of working memory, attentional control,
on logical reasoning after viewing natural scenes com- and cognitive flexibility, with low to moderate effect
pared to urban scenes. sizes. However, improvements in attentional con-
trol were not detected when only examining studies
Discussion with fully balanced baseline measures. Moderator
analyses revealed that actual exposures to real envir-
Key findings onments are likely to enhance the restoration effect
Our systematic search of literature published after within these three domains. However, it was typical
July 2013 uncovered 42 articles where the restora- for actual exposures to be longer than virtual expo-
tive effect of natural environments was investigated sures, suggesting exposure length might also explain
using objective outcome measures. The aim was to the result. The effect of restoration potential among
identify cognitive domains that are sensitive to this participants was less clear. These findings suggest
Table 13. Data and moderator details from included studies using outcome measures classified in the processing speed domain.
Natural environment/stimuli
Control environment/stimuli Moderators
Pre Post Pre Post Restoration potential Baseline Exposure
Cognitive task Author, Year, (sub- mean mean mean mean Source of Induction Baseline Baseline Exposure
(outcome measure) study) sub-group n (SD) (SD) n (SD) (SD) RP length time balance Baseline average type
DLST (total correct, Van den Berg et al., 2-month follow up 84 31.68 37.37 86 32.13 37.94 none - T2† yes 31.91 actual
90s) 2017 (8.99) (9.39) (6.67) (8.47)
4-month follow up 84 31.68 40.68 86 32.13 40.52 none - T2† yes 31.91 actual
(8.99) (10.09) (6.67) (8.62)

LDST (total correct, Lymeus et al., 2017 prior to 9 42.00 41.89 7 42.43 44.71 participant T2* yes 42.22 virtual
60s) mindfulness (6.75) (5.56) (2.99) (2.63)
training
2 weeks 9 42.33 42.56 7 42.43 45.14 participant - T2* yes 42.38 virtual
mindfulness (7.53) (4.50) (5.44) (3.81)
training
4 weeks 9 43.44 43.00 7 43.14 41.29 participant - T2* yes 43.29 virtual
mindfulness (8.28) (6.14) (1.35) (7.16)
training
6 weeks 9 45.67 43.56 7 46.14 46.14 participant - T2* yes 45.91 virtual
mindfulness (9.78) (7.78) (4.53) (4.18)
training
8 weeks 9 43.78 45.44 7 45.71 44.43 participant - T2* yes 44.75 virtual
mindfulness (7.48) (7.86) (4.72) (5.22)
training

SDMT (total correct, Bodin & Hartig, 2003 men 6 51.17 48.83 6 51.50 49.17 none - T2† partial 51.34 actual
90s) (7.17) (4.22) (3.67) (5.38)
women 6 56.33 52.50 6 56.00 52.83 none - T2† partial 56.17 actual
(11.33) (10.04) (9.36) (9.56)

SST (total correct, 60s) Johansson, Hartig, and alone 20 38.65 37.85 20 37.85 37.80 none - T2† no 38.25 actual
Staats. 2011 (5.28) (5.10) (5.21) (4.87)
with friend 20 40.00 36.35 20 37.70 36.85 none - T2† no 38.85 actual
(6.78) (5.09) (4.78) (4.79)

Sky Search Task (total Van den Berg et al., 2016 2-month follow up 84 9.64 10.73 86 9.19 10.28 none - T2† yes 9.42 actual
correct) (3.18) (3.33) (3.04) (3.48)
4-month follow up 84 9.64 13.05 86 9.19 12.03 none - T2† yes 9.42 actual
(3.18) (3.10) (3.04) (3.11)

TMTA (completion Cimprich & Ronis, 2003 - 83 30.23 25.65 74 37.08 31.21 participants - T2* no 33.66 actual
time) (11.84) (9.38) (19.79) (11.53)
Jung et al., 2017 healthy adults 20 27.44 24.68 20 25.03 24.95 none - T2† partial 26.24 virtual
JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B

(9.67) (6.92) (7.28) (7.57)

heart failure 20 41.90 41.80 20 41.92 44.38 participants T2* partial 41.91 virtual
patients (14.87) (21.67) (17.99) (28.18)
255

(Continued )
 256

Table 13. (Continued).

Natural environment/stimuli
Control environment/stimuli Moderators
M. P. STEVENSON ET AL.

Pre Post Pre Post Restoration potential Baseline Exposure

Cognitive task Author, Year, (sub- mean mean mean mean Source of Induction Baseline Baseline Exposure
(outcome measure) study) sub-group n (SD) (SD) n (SD) (SD) RP length time balance Baseline average type
Stark, 2003 - 29 22.06 19.84 28 21.67 19.60 participants - T2* partial 21.87 actual
(7.09) (4.79) (5.46) (5.33)

Stroop Task (% error, Jung et al., 2017 healthy adults 20 0.29 1.17 20 1.17 0.44 none - T2† partial 0.73 virtual
congruent) (0.88) (2.67) (3.38) (1.04)
heart failure 20 4.17 2.5 20 3.47 2.92 participants - T2* partial 3.82 virtual
patients (8.14) (3.24) (4.84) (3.55)

Stroop Task (RT, Jung et al., 2017 healthy adults 20 1282 1178 20 1242 1180 none - T2† partial 1262.00 virtual
congruent) (300) (275) (332) (352)
heart failure 20 1432 1352 20 1285 1281 participants - T2* partial 1358.50 virtual
patients (476) (338) (304) (296)

Stroop Task (% error, Jung et al., 2017 healthy adults 20 8.89 4.01 20 6.76 4.54 none - T2† partial 7.83 virtual
nonswitched) (16.40) (5.93) (10.33) (7.42)
heart failure 20 10.49 10.24 20 11.12 8.66 participants - T2* partial 10.81 virtual
patients (15.69) (14.28) (17.21) (12.78)

Stroop Task (RT, Jung et al., 2017 healthy adults 20 1578 1333 20 1408 1341 none - T2† partial 1493.00 virtual
nonswitched) (479) (540) (441) (484)
heart failure 20 1485 1370 20 1509 1492 participants - T2* partial 1497.00 virtual
patients (405) (416) (519) (447)
Notes. DLST = Digit-Letter Substitution Task; LDST = Letter-Digit Substitution Task; SDMT = Symbol-Digit Modalities Task; TMTA = Trail Making Task A *Participants with heightened restoration potential at
baseline measurement; †Participants without heightened restoration potential at baseline measurement.
 JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B 257

Table 14. Summary of results from the processing speed meta-analysis of post-test scores and moderator analyses exploring the
influence of baseline balance, restoration potential, and exposure type.
Processing Speed
n Hedge’s G (95% CI) I2 Z-value p-value Test of heterogeneity between subgroups
Overall 10 −0.138 (−0.299 – 0.022) 0 −1.687 0.092 -
Moderators
Baseline balance Q = 0.997, DF = 2, p = 0.607
yes 2 −0.072 (−0.357 – 0.214) 0 −0.490 0.624
partial 5 0.049 (−0.252 – 0.350) 0 0.322 0.748
no 3 −0.215 (−0.648 – 0.218) 55.59 −0.972 0.331
unsure - - - -
Restoration potential Q = 0.369, DF = 1, p = 0.544
heightened 4 −0.161 (−0.541 – 0.219) 49.32 −0.832 0.406
normal 6 −0.025 (−0.243 – 0.192) 0 −0.229 0.819
Exposure type Q = 1.200, DF = 1, p = 0.273
actual 7 0.163 (−0.368 – 0.042) 16.05 −1.557 0.119
virtual 3 0.084 (−0.308 – 0.476) 0 0.422 0.673

Figure 9. Overall forest plot showing pooled effect sizes (Hedges’ g) for outcome measures related to processing speed. No
moderator variables are included.

the construct of directed attention, and restoration James (1892) suggested the central component
effect of exposure to natural environments needs to to voluntary attention was a cognitive effort. In
be updated using existing evidence to guide future order to obtain a desired goal or outcome, a degree
research. of voluntary effort is required to focus on certain
stimuli. The cognitive effort required to success-
fully perform a task can be determined by an
Conceptualizing directed attention in interaction of perceptual, sensory, and cognitive
attention restoration theory loads (Sörqvist et al. 2016). Perceptual load is
determined by the ratio of task-relevant to task-
Kaplan and Kaplan (1989) adopted the term ‘direc- irrelevant stimuli present; sensory load is deter-
ted attention’ to help distinguish from William mined by the ease or difficulty of which task-
James’ (1892) concept of ‘voluntary attention’ from relevant stimuli may be processed; and cognitive
which it was inspired. This enabled the term to be load is determined by the extent to which a task
used as a distinct construct under the scope of ART. recruits cognitive resources. Cognitive effort is
In the original conceptualizations of directed atten- often conceptualized through these three factors
tion within ART (Kaplan 1995; Kaplan and Berman in terms of resistance to distraction during task
2010; Kaplan and Kaplan 1989), there appear to be performance. For example, when a task has a high
three characteristics central to its nature. perceptual load (through a large number of task-
I. Directed attention is a finite capacity that relevant stimuli) or high sensory load (through
requires voluntary cognitive effort more salient task-related stimuli), it is easier to
 258
M. P. STEVENSON ET AL.

Table 15. Data and moderator variables from included studies using outcome measures classified in the cognitive flexibility domain.
Natural environment/stimuli Control environment/stimuli Moderators
Pre Post Pre Post Restoration potential Baseline Exposure
Cognitive task Author, Year, Source of Induction Baseline Baseline Baseline Exposure
(outcome measure) (sub-study) sub-group n mean (SD) mean (SD) n mean (SD) mean (SD) RP length time balance average type
TMTB (completion Cimprich and - 83 65.01 (30.98) 56.51 (26.60) 74 77.09 (42.15) 68.01 (34.93) participants - T2* no 71.05 actual
time) Ronis (2003)
Jung et al. healthy 20 58.10 (22.37) 53.03 (22.32) 20 56.88 (20.66) 55.25 (21.84) none - T2† partial 57.49 virtual
(2017) adults
heart failure 20 97.78 (47.52) 87.37 (45.35) 20 98.29 (52.80) 88.04 (44.39) participants - T2* partial 98.04 virtual
patients
Shin et al. (2011) first walk 30 37.03 (6.81) 29.48 (6.82) 30 37.03 (6.81) 39.24 (21.23) none - T2† yes 37.03 actual
second walk 30 37.04 (6.90) 29.45 (6.72) 30 37.04 (6.90) 39.17 (21.23) none - T2† yes 37.04 actual
Stark (2003) - 29 46.55 (11.92) 38.89 (10.75) 28 48.15 (10.42) 40.40 (9.30) participants - T2* partial 47.35 actual
Stroop Task (% error, Jung et al. healthy 20 6.04 (7.13) 1.05 (3.15) 20 5.88 (10.56) 1.40 (4.35) none - T2† partial 5.96 virtual
switched) (2017) participants
heart failure 20 12.75 (14.99) 7.16 (8.71) 20 9.63 (13.80) 6.55 (8.19) participants - T2* partial 11.19 virtual
patients
Stroop Task (RT, Jung et al. healthy 20 1547 (515) 1311 (410) 20 1515 (536) 1283 (417) none - T2† partial 1531.00 virtual
switched) (2017) participants
heart failure 20 1619 (693) 1654 (547) 20 1373 (376) 1475 (464) participants - T2* partial 1496.00 virtual
patients
Notes. TMTB = Trail Making Task B. *Participants with heightened restoration potential at baseline measurement; †Participants without heightened restoration potential at baseline measurement.
 JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B 259

Figure 10. Overall forest plot showing pooled effect sizes (Hedges’ g) for outcome measures related to cognitive flexibility. No
moderator variables are included.

Table 16. Summary of results from the cognitive flexibility meta-analysis of post-test scores and moderator analyses exploring the
influence of baseline balance, restoration potential, and exposure type.
Cognitive Flexibility
n Hedge’s G (95% CI) I2 Z-value p-value Test of heterogeneity between subgroups
Overall 6 −0.317 (‘-0.532 – ‘0.101) 16.04 −2.877 0.004 -
Moderators
Baseline balance Q = 5.470, DF = 2, p = 0.065
yes 2 −0.610 (−0.972 – −0.249) −3.308 0.001
partial 3 −0.034 (‘-0.364 – 0.296) −0.201 0.840
no 1 −0.372 (−0.686 – −0.057) −2.315 0.021
unsure - - - - -
Restoration potential Q = 1.055, DF = 1, p = 0.304
heightened 3 −0.224 (−0.493 – 0.044) 11.09 −1.639 0.101
normal 3 −0.454 (−0.802 – −0.107) 19.44 −2.565 0.010
Exposure type Q = 3.560, DF = 1, p = 0.059
actual 4 −0.417 (−0.632 – −0.201) 0 −3.793 0.000
virtual 2 0.046 (−0.384 – 0.477) 0 0.211 0.833

ignore distracting task-irrelevant stimuli (Lavie in order to maintain focus (Kaplan 1995). This
et al. 2004). In this case, it is believed that when skill is sometimes referred to broadly as interfer-
a task demands high cognitive effort due to per- ence control, where distractors may be internal or
ceptual or sensory processes, the necessary external (Diamond 2013). Internal distractors,
increase in attentional control facilitates the inhi- such as unwanted thoughts or memories, are
bition of distractors. Conversely, when a task ignored through the process of cognitive inhibi-
demands high cognitive effort due to high cogni- tion. Whereas inhibition of external distractors
tive load, the necessary increase in attentional con- takes place at the perceptual level where target
trol places higher demands on cognitive resources stimuli are attended to while others are ignored
that are limited, which causes an attenuation of the (Diamond 2013). In earlier definitions of directed
ability to ignore distractors (Lavie 2005). attention in the ART literature, it was not explicit
II. Directed attention requires focus on task- whether cognitive inhibition, or the suppression of
relevant stimuli and simultaneous inhibi- internal distractors, was included. However,
tion of task-irrelevant distractors Kaplan and Berman (2010) suggested that this
process might also be sensitive to restorative
Directed attention has been described loosely as
environments.
a “. . . means of achieving focus in a confusing
environment” (Kaplan and Kaplan 1989, p. 188) III. Directed attention is a resource recruited
and more succinctly as the ability to “fix attention by other cognitive processes
on certain stimuli while suppressing other stimuli” Kaplan and Berman’s (2010) extension of the
(Kaplan and Berman 2010, p. 44). These defini- directed attention construct adopted by ART
tions highlight another central process required appears to have done service to the field by
for directed attention, the inhibition of distraction bringing the theory to a wider readership with
 260
M. P. STEVENSON ET AL.

Table 17. Data from included studies using outcome measures classified as emerging cognitive domain.
Natural environment/stimuli Control environment/stimuli Moderators
Pre Post Pre Post Restoration potential Baseline Exposure
Cognitive task Author, Year, Source Induction Baseline Baseline Baseline Exposure
(outcome measure) (sub-study) sub-group n mean (SD) mean (SD) n mean (SD) mean (SD) of RP length time balance average type
DoG (total delayed Jenkin et al. - 25 3.12 (1.51) 3.44 (1.36) 25 3.20 (1.87) 2.52 (1.61) none - T2† yes 3.16 virtual
choices) (2017) (1)
Jenkin et al. - 26 3.04 (1.78) 3.27 (1.25) 28 3.46 (1.04) 2.96 (1.07) none - T2† yes 3.25 virtual
(2017) (2)
TAP (unprovoked) Wang et al. non-depletion 29 0.05 (0.89) 0.02 (0.86) 29 −0.02 (0.74) −0.09 (0.87) none - T2† yes 0.02 virtual
(2017)
depletion 31 −0.17 (0.78) −0.11 (0.79) 29 0.14 (0.82) 0.20 (0.99) task NR T1 yes −0.02 virtual
TAP (provoked) Wang et al. non-depletion 29 −0.17 (0.87) −0.07 (0.75) 29 −0.07 (0.77) −0.23 (0.86) none - T2† yes −0.12 virtual
(2017)
depletion 31 −0.05 (0.85) −0.08 (0.91) 29 0.30 (0.95) 0.38 (0.83) task NR T1 yes 0.13 virtual
GRE (total correct) Chow and Lau - 29 6.00 (2.51) 7.31 (2.49) 29 7.28 (3.07) 7.21 (2.47) task 6”00’ T1 unsure 6.64 virtual
(2015) (2)
Notes. DoG = Delay of Gratification Task; TAP = Taylor’s Aggression Paradigm; GRE = Graduate Record Exam. *Participants with heightened restoration potential at baseline measurement; †Participants
without heightened restoration potential at baseline measurement.
 JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B 261

the inclusion of executive functioning and self- induction as ego-depletion (Chow and Lau
regulation. The extension is perhaps responsible 2015).
for the diversity in outcome measures identi-
fied. Kaplan and Berman (2010) suggested that
other cognitive processes required to perform Expanding the definition of directed
tasks associated with these outcome measures attention
may also be sensitive to the restoration effect. Based upon the characteristics of directed atten-
Indeed, models exploring the interrelatedness of tion described above, a framework is proposed in
executive functions do not necessarily place which to categorize the cognitive domains and
primacy on directed attention, but rather con- outcome measures found to be sensitive to the
siders it as one type of inhibitory control that restorative effect of natural environments. It is
interacts with other executive abilities to sub- conceivable that the framework provides a useful
serve goal-directed behavior (Diamond 2013). base from which to unpack our findings so that
The involvement of directed attention in cog- directed attention can be better described based
nitive tasks cannot be quantified; however, it is upon current evidence.
possible to examine aspects of these tasks to
ascertain a better description of directed atten-
tion, as a cognitive resource sensitive to the The Three D’s of Directed Attention
restoration effect. Stimuli contained in cogni-
tive tasks can be categorized into those to be It is proposed that outcome measures used within
attended (targets) and those to be suppressed ART research may be categorized along three axes
(distractors). This is obvious in certain tasks or continuums: 1) cognitive Demand (high–low),
measuring the ability to suppress external dis- based on levels of perceptual, sensory, and cogni-
traction, such as the Flanker Task (Eriksen tive loads; 2) Direction of attentional focus (inter-
nal–external), whether goal-relevant task
1995) where the center arrow acts as the target
components are stimuli that are perceived exter-
to be attended to while the flanking arrows act
nally or information that is represented internally;
as distractors. However, Kaplan and Berman
(2010) explained that tasks measuring aspects and 3) locus of Distraction (internal–external),
of self-regulation rely on a comparable ability whether goal-irrelevant task components are sti-
of suppressing internal distractors. An example muli that are perceived externally or information
of this was demonstrated by Baumeister et al. that is represented internally. Each axis is
(1998) findings on self-control where partici- described in more detail below and then illustrated
pants forced themselves to eat radishes rather by using representative examples from the three
than chocolate by overriding a preferred cognitive domains found to be positively influ-
response and suppressing internal cues asso- enced by natural environments during our meta-
ciated with the more favored food choice. The analysis.
investigators manipulated the situation by ask- 1) Cognitive Demand
ing participants to skip one meal before the The ability to direct attention requires voluntary
session and by having the experimental area cognitive effort. Tasks purported to measure this
smell of fresh chocolate and baking. Thus, the ability vary in the amount of effort that is required
demand for suppression of internal cues (hun- to perform them successfully. This variation may
ger) was heightened through the senses. Those occur between task types but also within different
who exerted the necessary self-control were less manipulations of the same task. As described ear-
likely to persist on a subsequent unsolvable lier, the cognitive effort required to perform a task
puzzle task, suggesting that suppression of is determined by the associated perceptual, sen-
internal (and external) distraction depleted a sory, and cognitive loads (Lavie et al. 2004). For
cognitive resource under a process they called example, Chow and Lau (2015) enhanced cogni-
ego-depletion (Baumeister et al. 1998). Indeed, tive demand through the high sensory load by
several studies in our review referred to fatigue increasing the salience of distracting thoughts
262 M. P. STEVENSON ET AL.

using hunger and smell. Loadings may be com- ignored as part of the same identification and
bined to give an overall cognitive demand rating. selection processes. Therefore, tasks measuring
Tasks with the highest cognitive demand, deemed directed attention may be associated with either
most difficult to perform, are predicted to have: or a combination of external and internal distrac-
low perceptual load of targets, where the number tors that are manipulated purposefully by the
of task-relevant stimuli is lower than the number tasks’ demands. The use of a continuum, rather
of task-irrelevant stimuli; low sensory load, where than categorical ratings, seems most appropriate as
the salience of task-related stimuli is low; and high there is likely to be an interplay between how
cognitive load, where task performance requires external distractors influence internal distractors,
the simultaneous use of other cognitive resources. and vice versa.
Conversely, tasks with the lowest cognitive Figure 11 presents an example of hypothetical
demand, deemed easiest to perform, are predicted ratings using the 3Ds of Directed Attention frame-
to have: high perceptual load, where the ratio of work with three different outcome measures. For the
task-relevant to task-irrelevant stimuli is high; executive score of the ANT from the attentional
high sensory load, where task-relevant stimuli are control domain: Cognitive demand is relatively
salient; and low cognitive load, where the task is high for this task due to moderately low target to
performed using directed attention unencumbered distractor ratio (perceptual load), non-salient stimuli
by other cognitive capacities. (sensory load), and added conflict processing from
2) Direction of Attentional Focus congruent versus incongruent trials (cognitive load).
Other versions of the task might heighten cognitive
This component of directed attention tasks deals
demand by including more distractor stimuli when
with the target stimuli. Target stimuli are identified
the target is displayed. The direction of the atten-
as those to which attention needs to be directed in
tional focus is purely external as the target stimuli are
order to successfully perform the task. Following an
present only on the screen. Similarly, the locus of
earlier taxonomy of attention (Chun, Golomb, and
distraction is purely external as distractor stimuli are
Turk-Browne 2011), target stimuli were categorized
present only on the screen.
as either external or internal. External stimuli are
For the DSB Task from the working memory
those that are processed using sensory modalities –
domain: Cognitive demand is high for this task
vision, hearing, touch, smell, and taste. Chun and
due to high cognitive load where non-salient target
colleagues (2011) explained that attention may be
stimuli must be integrated into the internal mental
directed towards each separately or in combination
workspace (Hasher and Zacks 1988) as well as
in order to achieve a required goal. Attention may
manipulated with other mental representations of
also be directed towards spatial and temporal char-
target stimuli to be remembered. The direction of
acteristics of the target stimuli, features of the stimuli
attentional focus is balanced between internal and
or the stimuli as complete objects (Chun, Golomb,
external. In this task, goal-relevant stimuli are
and Turk-Browne 2011). Internal target stimuli refer
perceived externally, for example, audially, and
to goal-related representations that exist internally
then needs to be held as internal representations.
among goal-unrelated representations held in work-
The locus of distraction is purely internal as dis-
ing memory and long-term memory, which may
tractors include only thoughts, memories, and
relate to task rules, decisions, and responses (Chun,
representations needing to be suppressed to keep
Golomb, and Turk-Browne 2011).
the mental workspace clear. All sensory stimuli are
3) Locus of Distraction goal-related in this task.
The locus of distraction refers to the way in For the TMTB from the cognitive flexibility
which goal-unrelated stimuli are presented and domain: Cognitive demand is high because of the
processed during task performance. While target need to switch between competing rule sets while
stimuli may be categorized as being identified also holding internal representations of them in
through sensory modalities or selected from inter- order to do so. Further, there is always only one
nal representations (Chun, Golomb, and Turk- non-salient target stimulus in the array which
Browne 2011), distractors need to be successfully changes after each is located. The direction of the
 JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B 263

Figure 11. Hypothetical application of the 3Ds of Directed Attention framework using examples from the review. Using the three
axes (demand, direction, distraction) to describe and compare outcome measures used in attention restoration studies allows
researchers to consider in more detail the processes being influenced by the environment under their experimental conditions.

attentional focus is both external and internal. baseline measures was of primary importance;
Goal-relevant stimuli are found visually, while however, there were limitations in how this infor-
internal representations of rule sets must also be mation might be utilized. It is recommended that
attended to in order to perform the task. The locus future studies using group randomization adjust
of distraction is primarily external where all but for baseline using ANCOVA, while studies where
one visual stimulus are distractors. However, it is group allocation is not randomized depend on
also possible that perceiving these goal-irrelevant assumptions of expected group differences (Van
stimuli may evoke internal representations that are Breukelen 2013).
needed to be suppressed in order to successfully Our analysis revealed that actual exposure to
attend to the goal-relevant representation. natural environments significantly influenced
effect sizes of working memory, attentional con-
trol, and cognitive flexibility, compared to virtual
Study designs and future research exposures. ART predicts that actual experiences in
nature will restore cognitive abilities through indu-
By including moderator analyses during our meta- cing a psychological state of restoration. If virtual
analyses it was possible to explore whether varia- nature is sufficient to restore cognitive abilities,
tions in study design affected the effect sizes of this suggests that a restorative state may be
pooled standardized means. The inclusion of induced by visual (and sometimes auditory)
264 M. P. STEVENSON ET AL.

representations alone and that these are sufficient. widely in intervention, participant groups, and
While this may be the case in several individual ART-related protocols. As also noted by Ohly
studies, it was found that exposure to actual and colleagues (2016), it highlights the need to
restorative environments provides a stronger, synthesize a more homogeneous set of studies to
more reliable effect and is therefore recommended ensure a more reliable synthesis. However, a more
for future researchers, where possible. homogeneous set of experiments cannot occur
With that in mind, an alternative interpretation without clarification of the directed attention con-
of the results, concerning the length of exposure, struct and selection of appropriate outcome mea-
should also be addressed. Across the studies, it was sures. It is hoped our review and proposed
typical for actual exposures to be longer than vir- framework might help guide future studies.
tual exposures. Therefore, the moderating effect of The inclusion of baseline measurement was
exposure type may also be explained by exposure of primary importance to our synthesis; how-
length. It is recommended future investigators ever, dealing with baseline in meta-analysis is
consider both type and length of exposure when often not straightforward. Testing for baseline
designing restoration studies. differences within a study is considered mis-
Finally, heightened restoration potential, through leading and is not recommended (Moher et al.
fatigue induction or participant characteristics, 2012). Further, it is often not possible to use
strengthened the restoration effect on working the preferred method of adjustment using
memory, but not attentional control or cognitive ANCOVA during meta-analysis, unless all par-
flexibility. This finding goes directly against ART, in ticipant-level data are available. For study-level
that heightened restoration potential should pro- analysis, it is recommended to use only trials
duce a more reliable restoration effect, if the with good to moderate baseline balance to cal-
mechanism underlying the effect is indeed restora- culate treatment effects (Riley et al. 2013). If
tion of a depleted cognitive resource. Essentially, significant effects based on post-test data are
this means there was mixed evidence for a restora- lost, then one might conclude that individual-
tion mechanism during our meta-analyses. If this level data are needed to adequately adjust for
observation was to be taken at face value, then one baseline. This was the case for the attentional
would need an additional explanation of how nat- control cognitive domain, suggesting any con-
ural environments improve cognitive performance, clusions drawn from post-test-only data for
which may act directly or indirectly with the this domain should be considered with caution.
restoration effect, such as the immune system It is also difficult to apply the stringent mea-
boost hypothesized by Kuo (2015). However, it sures of study quality relevant in other fields to
may also suggest that binary categorization of the context of attention restoration research.
restoration potential is not appropriate and restora- One obvious problem is that environmental
tion potential is something that needs to be interventions cannot be blinded. It is therefore
addressed at an individual level. For researchers, it responsible to acknowledge the potential for
should be encouraged to include an assessment of something akin to a placebo or expectation
the success of fatigue induction as insufficiently effect when discussing how being in a natural
fatigued participants may obscure results. A mea- environment influences brain function. This is
sure of fatigue, physiological or cognitive, before especially true when one considers that most
and after the fatigue induction period might serve studies uncovered in the literature search used
as an optimal manipulation check. The time points adult populations from developed countries
presented in Figure 1 offers a suggestion of where where there likely exists cultural narratives
these may be placed a restoration protocol. around the idea that contact with nature is
good for human health. Whether cognitive
restoration from exposure to natural environ-
Limitations
ments is a phenomenon relevant only to socie-
A key limitation for the current review was synthe- ties with such a cultural narrative is yet to be
sizing post-test data from studies that varied explored.
 JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B 265

Conclusions Funding
Our updated literature review and meta-analysis This work was supported by Trygfonden, Denmark [ID#109942].
showed that working memory, cognitive flexibility
and to a less-reliable degree, attentional control,
may be improved after exposure to nature through
References
restoration of cognitive processes related to direct-
ing attention. When considering the role of direc- Arbuthnott, K., and J. Frank. 2000. Trail making test, part B
ted attention within these cognitive domains, it is as a measure of executive control: Validation using a set-
not possible to quantify how much is recruited. switching paradigm. Journal of Clinical and Experimental
Neuropsychology 22:518–28.
However, one can qualify how directed attention is
Baddeley, A. 2003. Working memory: Looking back and
recruited during tasks within these cognitive looking forward. Nature Reviews Neuroscience 4:829.
domains. It is suggested that the Three D’s of Ball, K. K., B. L. Beard, D. L. Roenker, R. L. Miller, and D. S.
Directed Attention framework may be useful for Griggs. 1988. Age and visual search: Expanding the useful
this purpose. Within the framework, it is proposed field of view. Journal of the Optical Society of America A
cognitive domains and their related outcome mea- 5:2210–19.
Barfod, K., N. Ejbye-Ernst, L. Mygind, and P. Bentsen. 2016.
sures may vary along axes of the Demand imposed
Increased provision of udeskole in Danish schools: An
by simultaneous cognitive processes; Direction of updated national population survey. Urban Forestry and
attentional focus towards targets or goal-relevant Urban Greening 20:277–81.
stimuli; and locus of Distraction, produced by dis- Baumeister, R. F., E. Bratslavsky, M. Muraven, and D. M.
tractors or goal-irrelevant stimuli. When observing Tice. 1998. Ego depletion: Is the active self a limited
the current evidence within this framework, data resource? Journal of Personality and Social Psychology
74:1252–65.
indicate that cognitive processes found to be sen-
Berman, M. G., E. Kross, K. M. Krpan, M. K. Askren, A.
sitive to the restoration effect of natural environ- Burson, P. J. Deldin, S. Kaplan, L. Sherdell, I.H. Gotlib and
ments are: 1) more likely to be high in cognitive J. Jonides. 2012. Interacting with nature improves cogni-
demand, requiring the maintenance of several pro- tion and affect for individuals with depression. Journal of
cesses simultaneously; 2) related to the control of affective disorders, 140:300–305.
attention towards both internal and external goal- Berman, M. G., J. Jonides, and S. Kaplan. 2008. The cognitive
benefits of interacting with nature. Psychological Science
relevant stimuli; and 3) related to the suppression
19:1207–12.
of internal distractors, such as goal-irrelevant Berry, M. S., M. A. Repke, N. P. Nickerson, L. G. Conway, A.
thoughts, as well as external distractors, such as L. Odum, and K. E. Jordan. 2015. Making time for nature:
goal-irrelevant stimuli. It is hoped this information Visual exposure to natural environments lengthens sub-
may guide future studies that are needed to better jective time perception and reduces impulsivity. PLoS One
understand the actual process of attention restora- 10:e0141030.
Berry, M. S., M. M. Sweeney, J. Morath, A. L. Odum, and K.
tion and how it relates to an individual’s need for
E. Jordan. 2014. The nature of impulsivity: Visual expo-
restoration or restoration potential. sure to natural environments decreases impulsive deci-
sion-making in a delay discounting task. PLoS One 9:
e97915.
Acknowledgments Berto, R. 2005. Exposure to restorative environments helps
restore attentional capacity. Journal of Environmental
We would like to thank Heather Ohly and her co-authors for Psychology 25:249–59.
their contact and support during the completion of this Beute, F., and Y. A. W. De Kort. 2014. Natural resistance:
article. We would also like to thank Terry Hartig for his Exposure to nature and self-regulation, mood, and phy-
valuable input on addressing aspects of research designs siology after ego-depletion. Journal of Environmental
and analysis of data. Finally, we would like to thank all Psychology 40:167–78.
authors for their cooperation and patience during the review, Bodin, M., and T. Hartig. 2003. Does the outdoor environ-
quality assessment, and data synthesis. MPS’s research is ment matter for psychological restoration gained through
funded by a grant from Trygfonden, Denmark (ID#109942). running?. Psychology of Sport and Exercise, 4:141–53.
The opinions expressed are the authors only and are not Bratman, G. N., G. C. Daily, B. J. Levy, and J. J. Gross. 2015.
necessarily those of Trygfonden, Denmark. No potential con- The benefits of nature experience: Improved affect and
flicts of interest were reported by the authors. cognition. Landscape and Urban Planning 138:41–50.
266 M. P. STEVENSON ET AL.

Chow, J. T., and S. Lau. 2015. Nature gives us strength: Gamble, K. R., J. H. Howard Jr., and D. V. Howard. 2014.
Exposure to nature counteracts ego-depletion. Journal of Not just scenery: Viewing nature pictures improves execu-
Social Psychology 155:70–85. tive attention in older adults. Experimental Aging Research
Chun, M. M., J. D. Golomb, and N. B. Turk-Browne. 2011. A 40:513–30.
taxonomy of external and internal attention. Annual Geniole, S. N., J. P. F. David, R. F. R. Euzébio, B. Z. S.
Review of Psychology 62:73–101. Toledo, A. I. M. Neves, and C. M. McCormick. 2016.
Cimprich, B., and D. L. Ronis. 2003. An environmental Restoring land and mind: The benefits of an outdoor
intervention to restore attention in women with newly walk on mood are enhanced in a naturalized landfill
diagnosed breast cancer. Cancer Nursing 26:284–92. area relative to its neighboring urban area.
Collaboration, C. 2008. Cochrane handbook for systematic Ecopsychology 8:107–20.
reviews of interventions. Cochrane Collaboration. Greenwood, A., and B. Gatersleben. 2016. Let’s go outside!
Chichester, UK: John Wiley & Sons, Ltd. Environmental restoration amongst adolescents and the
Conway, A. R., M. J. Kane, M. F. Bunting, D. Z. Hambrick, impact of friends and phones. Journal of Environmental
O. Wilhelm, and R. W. Engle. 2005. Working memory Psychology 48:131–39.
span tasks: A methodological review and user’s guide. Haga, A., N. Halin, M. Holmgren, and P. Sörqvist. 2016.
Psychonomic Bulletin and Review 12:769–86. Psychological restoration can depend on stimulus-source
Craig, C., M. I. Klein, C. V. Menon, and S. B. Rinaldo. 2015. attribution: A challenge for the evolutionary account?.
Digital nature benefits typical individuals but not indivi- Frontiers in Psychology 7:1831.
duals with depressive symptoms. Ecopsychology 7:53–58. Han, K.-T. 2017. The effect of nature and physical activity on
Dadvand, P., M. J. Nieuwenhuijsen, M. Esnaola, J. Forns, X. emotions and attention while engaging in green exercise.
Basagana, M. Alvarez-Pedrerol, I. Rivas, M. Lòpez-Vicente, Urban Forestry and Urban Greening 24:5–13.
M. De Castro Pascual, J. Su, M. Jerrett, X. Querol, and J. Hartig, T., G. W. Evans, L. D. Jamner, D. S. Davis, and T.
Sunyer. 2015. Green spaces and cognitive development in Gärling. 2003. Tracking restoration in natural and urban
primary schoolchildren. Proceedings of the National field settings. Journal of Environmental Psychology 23:109–23.
Academy of Sciences of the United States of America Hartig, T., and H. Jahncke. 2017. Letter to the editor:
112:7937–42. Attention restoration in natural environments: Mixed
Dadvand, P., C. Tischer, M. Estarlich, S. Llop, A. Dalmau mythical metaphors for meta-analysis. Journal of
Bueno, M. López Vicente, A. Valentìn, C. De Keijzer, A. Toxicology and Environmental Health B 20:305–15.
Fernández-Somoano, N. Lertxundi, C. Rodriguez-Dehli, Hasher, L., and R. T. Zacks. 1988. Working memory, com-
M. Gascon, M. Guxens, D. Zugna, X. Basagaña, M. J. prehension, and aging: A review and a new view.
Nieuwenhuijsen, J. Ibarluzea, F. Ballester, and J. Sunyer. Psychology of Learning and Motivation 22:193–225.
2017. Lifelong residential exposure to green space and James, W. 1892. Psychology. The briefer course. New York:
attention: A population-based prospective study. Henry Holt and Co.
Environmental Health Perspectives 125:097016. Jenkin, R., I. Frampton, M. P. White, and S. Pahl. 2017. The
Diamond, A. 2013. Executive functions. Annual Review of relationship between exposure to natural and urban envir-
Psychology 64:135–68. onments and children’s self-regulation. Landscape
Emfield, A. G., and M. B. Neider. 2014. Evaluating visual and Research 43:1–14.
auditory contributions to the cognitive restoration effect. Johansson, B., H. Bjuhr, M. Karlsson, J.-O. Karlsson, and L.
Frontiers in Psychology 5:548. Rönnbäck. 2015. Mindfulness-based stress reduction (MBSR)
Eriksen, C. W. 1995. The flankers task and response compe- delivered live on the internet to individuals suffering from
tition: A useful tool for investigating a variety of cognitive mental fatigue after an acquired brain injury. Mindfulness
problems. Visual Cognition 2:101–18. 6:1356–65.
Evensen, K. H., R. K. Raanaas, C. M. Hagerhall, M. Johansson, M., T. Hartig, and H. Staats. 2011. Psychological
Johansson, and G. G. Patil. 2015. Restorative elements at benefits of walking: Moderation by company and outdoor
the computer workstation: A comparison of live plants and environment. Applied Psychology: Health and Well-Being,
inanimate objects with and without window view. 3:261–80.
Environment and Behavior 47:288–303. Jung, M., J. Jonides, L. Northouse, M. G. Berman, T. M.
Fan, J., B. D. McCandliss, J. Fossella, J. I. Flombaum, and M. Koelling, and S. J. Pressler. 2017. Randomized crossover
I. Posner. 2005. The activation of attentional networks. study of the natural restorative environment intervention
Neuroimage 26:471–79. to improve attention and mood in heart failure. Journal of
Fan, J., B. D. McCandliss, T. Sommer, A. Raz, and M. I. Cardiovascular Nursing 32:464–79.
Posner. 2002. Testing the efficiency and independence of Kaplan, R., and S. Kaplan. 1989. The experience of nature: A
attentional networks. Journal of Cognitive Neuroscience psychological perspective. New York, NY: Cambridge
14:340–47. University Press.
Ferraro, F. M. III. 2015. Enhancement of convergent creativ- Kaplan, S. 1995. The restorative benefits of nature: Toward
ity following a multiday wilderness experience. an integrative framework. Journal of Environmental
Ecopsychology 7:7–11. Psychology 15:169–82.
 JOURNAL OF TOXICOLOGY & ENVIRONMENTAL HEALTH, PART B 267

Kaplan, S., and M. G. Berman. 2010. Directed attention as Pilotti, M., E. Klein, D. Golem, E. Piepenbrink, and K. Kaplan.
a common resource for executive functioning and self- 2015. Is viewing a nature video after work restorative?
regulation. Perspectives on Psychological Science 5:43– Effects on blood pressure, task performance, and long-
57. term memory. Environment and Behavior, 47:947–69.
Kelz, C., G. W. Evans, and K. Röderer. 2015. The restorative Rider, N. D., and G. E. Bodner. 2016. Does taking a walk in
effects of redesigning the schoolyard: A multi-methodolo- nature enhance long-term memory? Ecopsychology 8:27–34.
gical, quasi-experimental study in rural Austrian middle Riley, R. D., I. Kauser, M. Bland, L. Thijs, J. A. Staessen, J.
schools. Environment and Behavior 47:119–39. Wang, F. Gueyffier, and J. J. Deeks. 2013. Meta-analysis of
Kortte, K. B., M. D. Horner, and W. K. Windham. 2002. The randomised trials with a continuous outcome according to
trail making test, part B: Cognitive flexibility or ability to baseline imbalance and availability of individual partici-
maintain set? Applied Neuropsychology 9:106–09. pant data. Statistics in Medicine 32:2747–66.
Kuo, M. 2015. How might contact with nature promote Rogerson, M., V. F. Gladwell, D. J. Gallagher, and J. L.
human health? Promising mechanisms and a possible cen- Barton. 2016. Influences of green outdoors versus indoors
tral pathway. Frontiers in Psychology 6:1093. environmental settings on psychological and social out-
Lavie, N. 2005. Distracted and confused?: Selective attention comes of controlled exercise. International Journal of
under load. Trends in Cognitive Sciences 9:75–82. Environmental Research and Public Health 13:363.
Lavie, N., A. Hirst, J. W. De Fockert, and E. Viding. 2004. Sahlin, E., A. Lindegård, E. Hadzibajramovic, P. Grahn, J.
Load theory of selective attention and cognitive control. Vega Matuszczyk, and G. Ahlborg Jr. 2016. The influence
Journal of Experimental Psychology 133:339–54. of the environment on directed attention, blood pressure
Lee, K. E., K. J. H. Williams, L. D. Sargent, N. S. G. Williams, and heart rate—An experimental study using a relaxation
and K. A. Johnson. 2015. 40-second green roof views sustain intervention. Landscape research, 41:7–25.
attention: The role of micro-breaks in attention restoration. Sarter, M., B. Givens, and J. P. Bruno. 2001. The cognitive
Journal of Environmental Psychology 42:182–89. neuroscience of sustained attention: Where top-down
Li, D. Y., and W. C. Sullivan. 2016. Impact of views to school meets bottom-up. Brain Research Reviews 35:146–60.
landscapes on recovery from stress and mental fatigue. Schmeichel, B. J., K. D. Vohs, and R. F. Baumeister. 2003.
Landscape and Urban Planning 148:149–58. Intellectual performance and ego depletion: Role of the
Lin, Y.-H., -C.-C. Tsai, W. C. Sullivan, P.-J. Chang, and C.-Y. self in logical reasoning and other information proces-
Chang. 2014. Does awareness effect the restorative function sing. Journal of Personality and Social Psychology 85:33–
and perception of street trees?. Frontiers in Psychology 5:906. 46.
Lobo, A., H. Ramachandran, and A. Tiwari. 2015. The Effect Schutte, A. R., J. C. Torquati, and H. L. Beattie. 2017. Impact
of Viewing Nature and Urban Pictures on Affect and of urban nature on executive functioning in early and
Cognition. Journal of the Indian Academy of Applied middle childhood. Environment and Behavior 49:3–30.
Psychology, 41:162. Shin, W. S., C. S. Shin, P. S. Yeoun, and J. J. Kim. 2011. The
Lymeus, F., T. Lundgren, and T. Hartig. 2017. Attentional influence of interaction with forest on cognitive function.
effort of beginning mindfulness training is offset with Scandinavian Journal of Forest Research, 26:595–598.
practice directed toward images of natural scenery. Sonntag-Ostrom, E., M. Nordin, Y. Lundell, A. Dolling, U.
Environment and Behavior 49:536–59. Wiklund, M. Karlsson, B. Calrberg, and L. S. Järvholm.
Milner, B. 1963. Effects of different brain lesions on card 2014. Restorative effects of visits to urban and forest envir-
sorting: The role of the frontal lobes. Archives of onments in patients with exhaustion disorder. Urban
Neurology 9:90–100. Forestry and Urban Greening 13:344–54.
Mischel, W., Y. Shoda, and M. I. Rodriguez. 1989. Delay of Sörqvist, P., Ö. Dahlström, T. Karlsson, and J. Rönnberg.
gratification in children. Science 244:933–38. 2016. Concentration: The neural underpinnings of how
Moher, D., S. Hopewell, K. F. Schulz, V. Montori, P. C. cognitive load shields against distraction. Frontiers in
Gøtzsche, P. J. Devereaux, D. Elbourne, M. Egger, and D. Human Neuroscience 10:221.
G. Altman. 2012. CONSORT 2010 explanation and elabora- Stark, M. A. 2003. Restoring attention in pregnancy: the nat-
tion: Updated guidelines for reporting parallel group rando- ural environment. Clinical Nursing Research, 12:246–65.
mised trials. International Journal of Surgery 10:28–55. Studente, S., N. Seppala, and N. Sadowska. 2016. Facilitating
Moher, D., A. Liberati, J. Tetzlaff, D. G. Altman, and P. creative thinking in the classroom: Investigating the effects
Group. 2009. Preferred reporting items for systematic of plants and the colour green on visual and verbal crea-
reviews and meta-analyses: The PRISMA statement. PLoS tivity. Thinking Skills and Creativity 19:1–8.
Medicine 6:e1000097. Tanaka, M., H. Yamada, T. Nakamura, A. Ishii, and Y.
Ohly, H., M. P. White, B. W. Wheeler, A. Bethel, O. C. Watanabe. 2013. Fatigue-Recovering Effect of a House
Ukoumunne, V. Nikolaou, and R. Garside. 2016. Designed With Open Space. Explore: The Journal of
Attention restoration theory: A systematic review of the Science and Healing, 9:82–86.
attention restoration potential of exposure to natural Taylor, A. F., F. E. Kuo, and W. C. Sullivan. 2002. Views
environments. Journal of Toxicology and Environmental of nature and self-discipline: Evidence from inner city
Health B 19:305–43. children. Journal of Environmental Psychology 22:49–63.
268 M. P. STEVENSON ET AL.

Triguero-Mas, M., C. J. Gidlow, D. Martínez, J. De Bont, park scenes in Shanghai, China. Urban forestry & urban
G. Carrasco-Turigas, T. Martínez-Íñiguez, G. Hurst, D. greening, 15:112–22.
Masterton, D. Donaire-Gonzalez, E. Seto, M. V. Jones, Wang, Y., Y. She, S. M. Colarelli, Y. Fang, H. Meng, Q. Chen, X.
and M. J. Nieuwenhuijsen. 2017. The effect of rando- Zhang, and H. Zhu. 2017. Exposure to nature counteracts
mised exposure to different types of natural outdoor aggression after depletion. Aggressive Behavior 2017:1–9.
environments compared to exposure to an urban Ward, J. S., J. S. Duncan, A. Jarden, and T. Stewart. 2016.
environment on people with indications of psycholo- The impact of children’s exposure to greenspace on
gical distress in Catalonia. PLoS One 12:e0172200. physical activity, cognitive development, emotional well-
Ulset, V., F. Vitaro, M. Brendgen, M. Bekkhus, and A. I. being, and ability to appraise risk. Health and Place
Borge. 2017. Time spent outdoors during preschool: 40:44–50.
Links with children’s cognitive and behavioral develop- Zarr, R., L. Cottrell, and C. Merrill. 2017. Park prescription
ment. Journal of Environmental Psychology 52:69–80. (DC Park Rx): A new strategy to combat chronic disease in
Van Breukelen, G. J. 2013. ANCOVA versus CHANGE from children. Journal of Physical Activity and Health 14:1–2.
baseline in nonrandomized studies: The difference. Zhang, Y., J. Kang, and J. Kang. 2017. Effects of soundscape
Multivariate Behavioral Research 48:895–922. on the environmental restoration in urban natural envir-
Van Den Berg, A. E., J. E. Wesselius, J. Maas, and K. Tanja- onments. Noise and Health 19:65–72.
Dijkstra. 2016. Green walls for a restorative classroom Zijlema, W. L., M. Triguero-Mas, G. Smith, M. Cirach, D.
environment: A controlled evaluation study. Environment Martinez, P. Dadvand, M. Gascon, M. Jones, C. Gidlow, G.
and Behavior 49:791–813. Hurst, D. Masterton, N. Ellis, M. Van Den Berg, J. Maas, I.
Van Rompay, T. J., and T. Jol. 2016. Wild and free: Van Kamp, P. Van Den Hazel, H. Kuize, M. J.
Unpredictability and spaciousness as predictors of creative Nieuwenhuijsen, and J. Julvez. 2017. The relationship
performance. Journal of Environmental Psychology 48:140–48. between natural outdoor environments and cognitive
Wang, X., S. Rodiek, C. Wu, Y. Chen, and Y. Li. 2016. Stress functioning and its mediators. Environmental Research
recovery and restorative effects of viewing different urban 155:268–75.