Content uploaded by Demetrios Gatziolis
Author content
All content in this area was uploaded by Demetrios Gatziolis on Mar 29, 2019
Content may be subject to copyright.
The Relationship Between Trees and
Human Health
Evidence from the Spread of the Emerald Ash Borer
Geoffrey H. Donovan, PhD, David T. Butry, PhD, Yvonne L. Michael, ScD,
Jeffrey P. Prestemon, PhD, Andrew M. Liebhold, PhD,
Demetrios Gatziolis, PhD, Megan Y. Mao
Background: Several recent studies have identifıed a relationship between the natural environment
and improved health outcomes. However, for practical reasons, most have been observational,
cross-sectional studies.
Purpose: A natural experiment, which provides stronger evidence of causality, was used to test
whether a major change to the natural environment—the loss of 100 million trees to the emerald ash
borer, an invasive forest pest—has influenced mortality related to cardiovascular and lower-
respiratory diseases.
Methods: Two fıxed-effects regression models were used to estimate the relationship between
emerald ash borer presence and county-level mortality from 1990 to 2007 in 15 U.S. states, while
controlling for a wide range of demographic covariates. Data were collected from 1990 to 2007, and
the analyses were conducted in 2011 and 2012.
Results: There was an increase in mortality related to cardiovascular and lower-respiratory-tract
illness in counties infested with the emerald ash borer. The magnitude of this effect was greater as
infestation progressed and in counties with above-average median household income. Across the 15
states in the study area, the borer was associated with an additional 6113 deaths related to illness of the
lower respiratory system, and 15,080 cardiovascular-related deaths.
Conclusions: Results suggest that loss of trees to the emerald ash borer increased mortality related
to cardiovascular and lower-respiratory-tract illness. This fınding adds to the growing evidence that
the natural environment provides major public health benefıts.
(Am J Prev Med 2013;44(2):139–145) Published by Elsevier Inc. on behalf of American Journal of Preventive
Medicine
Introduction
There is increasing evidence from multiple scien-
tifıc fıelds that exposure to the natural environ-
ment can improve human health.
1⫺4
However,
existing research has often been hampered by cross-
sectional study design and a failure to adequately ad-
dress confounding factors.
5
Quasi-experimental de-
signs, such as the pioneering work by Ulrich,
6
who
showed that patients recovered faster from surgery in a
room with a view of a natural scene than those without
such a view, are needed to provide stronger evidence of
a causal link between the natural environment and
health.
To address this gap in the literature, a longitudinal
study design was used to quantify the public health effects
of an introduced forest pest, the emerald ash borer, which
has killed tens of millions of ash trees since it was fırst
detected in the U.S. in 2002. The spread of the borer is a
unique natural experiment allowing the evaluation of the
effect of changes in the natural environment on public
health. The goal of the study was not to track the borer
per se, but to use it as a proxy for tree loss.
From the Pacifıc Northwest Research Station, (Donovan, Gatziolis, Mao),
U.S. Department of Agriculture Forest Service, Pacifıc Northwest Research
Station, Portland, Oregon; the National Institute of Standards and Tech-
nology, (Butry), Gaithersburg, Maryland; the Department of Epidemiology
and Biostatistics, (Michael), Drexel University, Philadelphia, Pennsylvania;
the U.S. Department of Agriculture Forest Service (Prestemon), Southern
Research Station, Research Triangle Park, North Carolina; and the North-
ern Research Station, (Liebhold), U.S. Department of Agriculture Forest
Service, Morgantown, West Virginia
Address correspondence to: Geoffrey H. Donovan, PhD, U.S. Depart-
ment of Agriculture Forest Service, PNW Research Station, 620 SW Main,
Suite 400, Portland OR 97205. E-mail: gdonovan@fs.fed.us.
0749-3797/$36.00
http://dx.doi.org/10.1016/j.amepre.2012.09.066
Published by Elsevier Inc. on behalf of American Journal of Preventive Medicine Am J Prev Med 2013;44(2):139–145 139
Natural experiments ap-
proximate RCTs, as the
mechanism determining
exposure is independent of
the outcome, and therefore,
common prior causes of ex-
posure and outcome are
equally distributed between
those exposed and those un-
exposed.
7
The borer spreads
directly from county to
county, but it is spread also
by accidental transport—
typically on fırewood—
which results in satellite
populations (Figure 1). This
accidental spread adds an
important random element
to the current natural exper-
iment. Nonetheless, natural
experiments remain obser-
vational studies and cannot
prove causality.
This study examined
whether the spread of the
emerald ash borer is asso-
ciated with increased
mortality related to car-
diovascular and lower-respiratory-tract illness. These
two types of health issues were chosen because they are
the fırst and third most common causes of
death in the U.S.,
8
and there are plausible
mechanisms linking these types of deaths
with trees. Specifıcally, the natural environ-
ment has been shown to decrease stress,
9
increase physical activity,
10
and improve air
quality.
11
In turn, stress,
12,13
lack of physical
activity,
14,15
and poor air quality
16
have
been linked with cardiovascular and lower-
respiratory-tract disease.
The pioneering work in the fıeld by Ulrich
6
found that
patients recovering from gall bladder⫺removal surgery
in a room with a view of a natural scene recovered faster
and took fewer pain medications than patients in a room
with a view of a brick wall. However, extending Ulrich’s
work has been problematic, because most health out-
comes of interest have causes that long precede the short
surgical recovery period, and most people spend little
time in environments as controlled as a hospital room.
Observational studies of the relationship between the
natural environment and health have examined a range
of health outcomes. Mitchell and Popham
17
found that,
after controlling for SES, “greenness” was negatively as-
sociated with overall mortality in England. This relation-
ship was particularly strong for cardiovascular-related
mortality. Takona and colleagues
18
stud-
ied the 5-year survival rate of 3144 senior
citizens living in Tokyo. They found a
positive association between survival
rate and access to walkable green space.
In Holland, Maas et al.
19
reported a
positive association between greenness
and self-reported health. In a later study,
Maas et al.
20
analyzed the health records
of 345,000 people. They found that those
living in greener areas were less likely to be diagnosed
with 15 of the 24 health outcomes examined. Results were
particularly strong for anxiety and depression and for
children and those with lower SES. Park and colleagues
9
showed that walking in a forest reduced heart rate and
cortisol levels. Finally, in New York City, Lovasi et al.
21
found that children who lived in areas with more street
trees were less likely to have asthma.
Two studies have examined the relationship between
the natural environment and birth outcomes. Donovan
et al.
2
found that mothers living in Portland OR with
more tree canopy within 50 meters of their homes, or who
lived closer to open space, were less likely to have a baby
Extent
of
infestation
-
2002
-
2007
-
2010
0 125 250 500 750 1000
km
Figure 1. Counties where the emerald ash borer had been detected in 2002, 2007, and 2010
See
related
Commentary by
Frumkin in this
issue.
140 Donovan et al / Am J Prev Med 2013;44(2):139–145
www.ajpmonline.org
that was small for gestational age. Dadvand and col-
leagues
1
conducted a similar study in Spain. They found
that women with more greenness within 100 meters of
their homes, or who lived within 500 meters of a major
green space, gave birth to heavier babies, although results
only held for women with the lowest level of education.
Emerald Ash Borer
The emerald ash borer, Agrilus planipennis, is a phloem-
feeding borer native to East Asia. It was discovered in
North America in 2002, when it was identifıed as the
cause of widespread ash mortality (Fraxinus spp.) in
Detroit MI and nearby Windsor, Ontario.
22
By 2012, this
borer had killed approximately 100 million trees in the
U.S. (D. McCullough, Michigan State University, per-
sonal communication, 2012). However, its potential im-
pact is much larger, as there are 7.5 billion ash trees in the
country.
22
In addition, the borer kills all 22 species of
North American ash and virtually all infested trees, so it is
a good proxy for ash tree death. For more information
about this borer, see the video in Appendix A (available
online at www.ajpmonline.org).
Methods
Study Area and Data
Data were collected from 1990 to 2007, and the analyses were
conducted in 2011 and 2012. The study sample consists of the 15
states that had at least one confırmed case of the borer in 2010. Data
were observed annually at the county level (1296 counties), from
1990 through 2007 (maximum number of observations⫽22,032,
but because of missing data, the actual number of observations⫽
21,080).
This sample allowed observation of mortality before and after
2002, when the emerald ash borer was initially discovered in the
U.S. By 2007, the U.S. Department of Agriculture (USDA) Animal
and Plant Inspection Service had detected the borer in 244 counties
(once a county is infected, it remains infected; Figure 1). Two
variables were used to describe the presence of the borer: a dummy
variable, which takes on a value of 1 in infested counties, and a
continuous variable (0⫺6) that denotes the number of years it has
been present in a county.
Mortality data were obtained from the National Center for
Health Statistics
23,24
and stratifıed by age (⬍18 years and ⱖ18
years) and cause of death (major cardiovascular disease [ICD-10
Codes I00-I78]; chronic lower-respiratory-tract disease [ICD-10
codes J40-47]; accidental death [ICD-10 Codes V01-X59, Y85-
86]). Demographic covariates were chosen based on neighborhood
determinants of cardiovascular and lower-respiratory-tract mor-
tality.
25
Demographic data were obtained from the 1990 and 2000
censuses and the 2009 American Community Survey. The authors
estimated census variables for all other years by interpolation.
The impact of the borer—on tree mortality and public health—
depends on the number and distribution of ash trees in a county.
Unfortunately, comprehensive data on ash abundance are not
available. Therefore, a two-stage process was used to estimate ash-
canopy cover, which is the area occupied by a tree’s crown when
viewed from above.
First, total tree canopy was estimated in a county using National
Land Cover Data (NLCD) raster maps from 1992, 2001, and 2006.
Tree canopy for all other years was estimated by linear interpola-
tion or extrapolation from these 3 years. The NLCD maps were
generated by processing 30-meter-resolution satellite imagery us-
ing class defınitions that have remained consistent through time.
26
Second, total tree canopy in a county was weighted by the pro-
portion of ash in a state (ash as a percentage of total tree canopy
varied from a low of 1.5% in Virginia to a high of 7.9% in New
York). For example, if a county had 40% tree canopy and 5% of tree
canopy was ash, then ash canopy was 2%. Data from the USDA
Forest Service’s Forest Inventory and Analysis (FIA) program were
used to estimate statewide ash abundance. State-level data, as op-
posed to county-level, were used to estimate ash-canopy cover, as
some counties have little forestland and therefore have few plots.
Data Analysis
Two regression models were estimated relating the presence of the
borer with rates of adult mortality related to cardiovascular and
lower-respiratory-tract illness. Models of the following general
form can be fıt to longitudinal data (where idenotes county and t
denotes time):
Yi,t ⫽

⫹

Xi,t ⫹
i⫹
i,t,
where Y
i,t
is the mortality rate (per 100,000 adults); X
i,t
is a vector of
independent variables;
i,t
is an i.i.d. error term uncorrelated with
the county-specifıc residual
i
; and

s are coeffıcients that are
estimated in the regression step. Typically, linear models of this
form are estimated using either fıxed-effects or random-effects
estimators. Fixed-effect estimators were used, as a Hausman spec-
ifıcation test showed that the assumptions underlying the random-
effects estimators were not met.
27
Heteroskedasticity is a common problem in panel-data analysis.
It can arise when observational units vary greatly in scale. In this
analysis, mortality rates were used rather than number of deaths,
which addresses the issue of scaling. However, in less-populous
counties, mortality counts are low, which means mortality rates are
sensitive to small changes in mortality counts (counties with low
mortality counts were not dropped from the analysis). Therefore, a
priori, error-term variance is expected to be higher in low-population
counties. For this reason, model coeffıcients were estimated with het-
eroskedasticity-robust fıxed-effects estimators.
28
Variables were selected for inclusion in the fınal model using
iterative backwards selection. Progressively lower signifıcance
thresholds were used with a fınal threshold of 0.1. A variance⫺
covariance matrix was used to avoid including highly collinear
combinations of variables in the model. When similar demo-
graphic variables were collinear (those describing income, for ex-
ample), the variable from a group that had the lowest p-value when
individually regressed against mortality was used.
Each variable that was dropped from the model was checked to
determine whether it varied signifıcantly between counties that
were infested and those that were not. If a variable did vary, then it
was reintroduced and retained, if it caused the coeffıcients on the
two borer variables to change by more than 10%.
29
None of the
reintroduced variables met this threshold.
In addition to controlling for potential confounders, all models
included a linear time-trend variable (1⫺18 years) to account for
Donovan et al / Am J Prev Med 2013;44(2):139–145 141
February 2013
broad trends in mortality—improved medical technology, for
example—that would not be captured by demographic covariates.
In addition, a 1-year lag of mortality rate was included to address
temporal autocorrelation (AR(1) correction). Finally, a variable
denoting the amount of ash-canopy cover in a county was in-
cluded, because if the borer does have a negative public health
effect, then one would expect ash to have a positive effect,
especially in counties not yet infected.
Interaction terms between demographic covariates and both
presence of the borer in a county and amount of ash also were
included. This was done because past research has shown that
access to greenness varies among demographic groups.
30
Thus, the
borer’s impact would be expected to depend on the demographic
makeup of a county.
Natural experiments provide stronger evidence of causation
than observational studies, but it is still possible that the results
were influenced by an omitted variable that is correlated with the
spread of the emerald ash borer. Therefore, to provide an addi-
tional safeguard, a model was estimated with accidental death as
the dependent variable, because this is a type of death that the borer
could not plausibly affect (the same model selection criteria were
used as those used for the cardiovascular and lower-respiratory-
tract models).
Results
Respiratory-Related Mortality
Regression results for the respiratory-related mortality
model are shown in Table 1. The negative coeffıcient on
the time trend confırms that overall respiratory-related
mortality declined over the 18-year study period.
31
The
presence of the borer was signifıcant by itself and in
interaction with years of infestation and median income
(dichotomized at the median split).
The positive coeffıcient on the income interaction
term suggests that the borer has a bigger effect on
mortality in wealthier counties. This is consistent with
previous research showing a positive correlation be-
tween tree cover and income in urban areas.
32
This
result was mirrored by the effect of ash on mortality
related to lower-respiratory-tract illness, as counties
with more ash trees had lower rates of this type of
mortality. In addition, the effect of the interaction
term between ash trees and median income is negative,
which suggests that the health benefıts of ash are
greater in wealthier counties.
Given the signifıcance of the interaction terms, the
coeffıcients on these terms cannot be interpreted in
isolation. Therefore, the delta method
33
was used to
calculate the net marginal effect of the borer on respi-
ratory-related mortality taking into account the direct
effect and the effect through interaction terms. The
presence of the borer in a county is associated with 6.8
additional deaths per year per 100,000 adults (95%
CI⫽4.8, 8.7).
To determine how the effect changes over time, sepa-
rate marginal effects were calculated for each year of
infestation (Table 2). Results show that as infestation in a
county progresses, the magnitude of the marginal effect
also increases. Applying these marginal effects to the
appropriate infested counties shows that the emerald ash
borer was associated with 6113 excess deaths between
2002 and 2007.
The delayed effect of the borer may be due to the 2⫺5 years
it takes an ash tree to die after initial infestation. In addition,
any effect on human mortality would be expected to lag
behind the borer’s effect on tree mortality. Indeed, it may be
surprising that the borer has any effect on human mortality
in the fırst year of infestation. However, once the borer is
detected in a county, often many healthy trees are cut down
Table 1. Longitudinal regression model of adult lower-
respiratory-tract disease–related mortality, adjusting for
covariates, U.S., 1990 –2007
Variable
Beta coefficient
a
(95% CI) p-value
Time trend ⫺2.98 (⫺3.23, ⫺2.72) ⬍0.001
1-year mortality-rate lag 0.31 (0.303, 0.310) ⬍0.001
Percentage non-
Hispanic white
9.40 (6.40, 12.40) ⬍0.001
Percentage Native
Hawaiian and other
Pacific Islander
2.14 (0.32, 3.97) 0.022
High median income 13.95 (6.50, 21.39) ⬍0.001
Aged ⬎25 years with
no high school
diploma, %
1.22 (0.92, 1.52) ⬍0.001
Aged ⬎25 years with
college degree, %
⫺0.33 (⫺0.70, 0.03) 0.077
Population below 100%
of poverty line, %
2.24 (1.89, 2.58) ⬍0.001
Percentage of county
covered by ash
canopy
⫺5.22 (⫺7.79, ⫺2.64) ⬍0.001
Emerald ash borer ⫺4.24 (⫺8.10, ⫺0.39) 0.031
Emerald ash borer X
high median
income
6.23 (2.23, 10.22) 0.002
Years of infestation 1.44 (0.95, 1.92) ⬍0.001
Ash canopy X high
median income
⫺0.85 (⫺1.30, ⫺0.41) ⬍0.001
R
2
Within counties 0.609
Between counties 0.187
Overall 0.352
a
Mortality rate per 100,000 adults
142 Donovan et al / Am J Prev Med 2013;44(2):139–145
www.ajpmonline.org
to prevent its spread. This practice was particularly common
during the early years of its spread.
In addition, there is often extensive media coverage
when the borer is fırst detected in a county, which may
cause some of the same stressful responses as tree mortal-
ity. For example, within the fırst year of discovery in
Michigan, the Detroit Free Press ran 39 stories on the
borer including four on the front page. Similarly, the
Chicago Tribune printed 53 stories relating to the borer in
the fırst year of infestation in Illinois, including 16 on the
front page. For the current paper, stories were identifıed
using emerald ash borer as a search term. Each story was
checked to make sure the emerald ash borer was in fact
the subject.
Cardiovascular-Related Mortality
Results for the cardiovascular model are shown in Table 3.
The signifıcance and magnitude of the coeffıcients on the
borer and borer interaction terms are similar to those for
the respiratory model. The effects of other covariates are
generally consistent, although ash is not associated with
cardiovascular-related mortality except in interaction
with median income.
The marginal effect of the borer on cardiovascular-
related mortality is 16.7 additional deaths per year per
100,000 adults (95% CI⫽5.7, 27.7) for a total of 15,080
excess deaths from 2002 to 2007 (Table 4).
In the accidental-mortality model, no effect of the
borer was found. Specifıcally, using a Wald test, the null
hypothesis that the coeffıcients on the borer, the interac-
tion of the borer with median income, and years of infes-
tation were jointly zero was not rejected (p⫽0.22). Al-
though the accidental-death model is not a control in a
formal sense, it is encouraging that the model found the
borer to be uncorrelated with a cause of death it could not
plausibly affect.
Discussion
Results suggest that the widespread death of ash trees
from the emerald ash borer lead to an increase in mortal-
ity related to cardiovascular and lower-respiratory-tract
illness. These results are consistent with previous re-
search that has identifıed a correlation between the natu-
ral environment and health. They also provide stronger
support for a causal relationship.
The borer had a greater effect in counties whose me-
dian household income was above average. There are a
number of possible interpretations for these results. Peo-
ple in wealthier counties may have greater access to ash
trees, so the death of these trees has a greater impact on
Table 2. Estimated marginal effects of the emerald ash
borer on lower-respiratory-tract disease⫺related
mortality by years of infestation
Years of
infestation
Number of
observations
Marginal effect
(95% CI) p-value
1 126 3.80 (2.08, 5.52) ⬍0.001
2 102 5.96 (4.13, 7.79) ⬍0.001
3 72 8.09 (5.92, 10.27) ⬍0.001
4 50 10.09 (7.41, 12.77) ⬍0.001
5 22 13.24 (9.96, 16.53) ⬍0.001
6 6 15.32 (12.78, 17.86) ⬍0.001
Across all
years
6.77 (4.84, 8.69) ⬍0.001
Wald test: all variables
a
⫽0
a
Emerald ash borer, years of infestation, and interaction terms
Table 3. Longitudinal regression model of adult
cardiovascular-related mortality adjusting for covariates
in the U.S., 1990⫺2007
Variable
Beta coefficient
a
(95% CI) p-value
Time trend ⫺6.49 (⫺7.45, ⫺5.54) ⬍0.001
1-year mortality-rate lag 0.45 (0.43, 0.47) ⬍0.001
High median income 11.03 (5.71, 16.34) ⬍0.001
Native Hawaiian and
other Pacific
Islander, %
30.07 (2.44, 57.71) 0.033
Aged ⬎25 years with
no high school
diploma, %
5.80 (4.67, 6.92) ⬍0.001
Aged ⬎25 years with
college degree, %
⫺1.92 (⫺3.26, ⫺0.57) 0.005
Population below
poverty line, %
⫺8.99 (⫺10.33, ⫺7.64) ⬍0.001
Percentage of county
covered by ash
canopy
⫺1.80 (⫺9.51, 5.91) 0.648
Emerald ash borer ⫺13.51 (⫺25.38, ⫺1.64) 0.026
Ash canopy X high
median income
18.24 (5.45, 31.02) 0.0005
Years of infestation 2.77 (1.05, 4.48) 0.002
Emerald ash borer X
high median
income
⫺3.42 (⫺4.71, ⫺2.13) ⬍0.001
R
2
Within counties 0.753
Between counties 0.298
Overall 0.488
a
Mortality rate per 100,000 adults
Donovan et al / Am J Prev Med 2013;44(2):139–145 143
February 2013
them. In particular, urban areas within wealthier counties
may have more trees, or better maintain them. Indeed, past
studies have found that within a city, wealthier neighbor-
hoods have more tree-canopy cover.
32
It also is possible that trees provide different benefıts in
wealthier areas. For example, Troy and Grove
34
found
that proximity to urban parks increased the sales price of
homes in wealthier neighborhoods, whereas, in poor
neighborhoods, houses close to parks sold for less. The
authors suggest that parks may attract criminal behavior
in poorer neighborhoods, so residents are not able to
benefıt from the park as much as people living in a wealth-
ier neighborhood. In addition, risk factors such as air
quality, which trees can mediate, may be different in
wealthier counties.
Results do not provide any direct insight into how trees
might improve mortality rates related to cardiovascular
and lower-respiratory-tract illness. However, there are
several plausible mechanisms including improving air
quality,
11,35
reducing stress,
13
increasing physical activ-
ity,
14
moderating temperature,
36
and buffering stressful
life events.
37
Future research could fruitfully investigate
the possible mechanisms linking the natural environ-
ment and health.
Limitations
This study has several limitations. It is possible that de-
spite the natural-experiment design, the results are an
artifact of an omitted risk factor that is correlated with the
borer or residual confounding. The authors believe that
this possibility is unlikely for three reasons. First, a wide
range of covariates that have been shown to influence
mortality related to cardiovascular and lower-respiratory
system illness were included in the model. Second, a
confounder would have to be strongly correlated with the
borer across both space and time. Third, an omitted vari-
able would need to influence these types of mortality but
not accidental death.
Nonetheless, it is re-emphasized that this is an obser-
vational study, and the results await confırmation. In
addition, this is an ecologic study, so the overall results do
not necessarily apply to a particular county or group of
counties. Finally, even well-controlled ecologic studies
can be subject to ecologic bias.
The variables used to denote the borer and ash are
another potential source of error. Specifıcally, data avail-
ability forced use of a simple dummy variable to denote
the presence of the borer, and past research has shown
that modeling a continuous process with a binary variable
can result in coeffıcients that are biased upward.
38
In
contrast, ash abundance was measured continuously.
However, the ash coverage variable is a composite of
county-level canopy-cover data and state-level data on
ash abundance and is, therefore, a coarse approximation
of the true amount of ash in a county. However, when
models were estimated in which ash-canopy cover was
replaced with canopy cover from all tree species, there
was little change in the coeffıcients of borer-related vari-
ables. Therefore, the choice of ash variable does not affect
the conclusions about the relationship between the borer
and mortality.
Conclusion
Tree loss from the spread of the emerald ash borer is
associated with increased mortality related to the cardio-
vascular and lower-respiratory systems. This relationship
is particularly strong in counties with above-average me-
dian household income.
No fınancial disclosures were reported by the authors of this
paper.
References
1. Dadvand P, Nazelle Ad, Figuearas F, et al. Green space, health inequal-
ity and pregnancy. Environ Int 2012;44(1):3–30.
2. Donovan GH, Michael YL, Butry DT, Sullivan AD, Chase JM. Urban
trees and the risk of poor birth outcomes. Health Place 2011;17:390 –3.
3. Frumkin H. Beyond toxicity: human health and the natural environ-
ment. Am J Prev Med 2001;20(3):234–40.
4. Tzoulas K, Korpela K, Venn S, et al. Promoting ecosystem and human
health in urban areas using green infrastructure: a literature review.
Landsc Urban Plann 2007;81:167–78.
5. Lee AC, Maheswaran R. The health benefıts of urban green spaces: a
review of the evidence. J Public Health (Oxf) 2011;33(2):212–22.
6. Ulrich RS. View through a window may influence recovery from sur-
gery. Science 1984;224:420–1.
Table 4. Estimated marginal effects of the emerald ash
borer on cardiovascular-related mortality by years of
infestation
Years of
infestation
Number of
observations
Marginal effect
(95% CI) p-value
1 126 9.68 (1.79, 17.57) 0.006
2 102 14.68 (7.09, 22.28) ⬍0.001
3 72 19.90 (11.4, 28.4) ⬍0.001
4 50 24.61 (14.27, 34.96) ⬍0.001
5 22 33.58 (21.14, 46.02) ⬍0.001
6 6 38.59 (23.62, 53.56) ⬍0.001
Across all
years
16.70 (5.73, 27.67) 0.001
Wald test: all variables
a
⫽0⬍0.001
a
Emerald ash borer, years of infestation, and interaction terms
144 Donovan et al / Am J Prev Med 2013;44(2):139–145
www.ajpmonline.org
7. Glymour MM. Natural experiments and instrumental variable analy-
ses in social epidemiology. In: Oakes JM, Kaufman JS, eds. Methods in
social epidemiology. San Francisco CA: Jossey-Bass, 2006:429–60.
8. CDC. Leading causes of death, 2009. www.cdc.gov/nchs/fastats/
lcod.htm.
9. Park B-J, Tsunetsugu Y, Ishii H, et al. Physiological effects of shinrin-
yoku (taking in the atmosphere of the forest) in a mixed forest in
Shinano Town, Japan. Scand J Forest Res 2008;23:278–83.
10. Kaczynski AT, Henderson KA. Environmental correlates of physical
activity: a review of evidence about parks and recreation. Leisure Sci
2007;29(4):315–54.
11. Nowak DJ, Crane DE, Stevens JC. Air pollution removal by urban trees
and shrubs in the U.S. Urban Forest Urban Green 2006;4:115–23.
12. Wright RJ, Rodriguez M, Cohen S. Review of psychosocial stress and
asthma: an integrated biopsychosocial approach. Thorax 1998;
53(12):1066–74.
13. Everson-Rose SA, Lewis TT. Psychosocial factors and cardiovascular
disease. Annu Rev Public Health 2004;26(1):469–500.
14. Oguma Y, Shinoda-Tagawa T. Physical activity decreases cardiovascu-
lar disease risk in women: review and meta-analysis. Am J Prev Med
2004;26(5):407–18.
15. Lucas SR, Platts-Mills TAE. Physical activity and exercise in asthma:
relevance to etiology and treatment. J Allergy Clin Immunol 2005;
115(5):928–34.
16. Pope CA, Burnett RT, Thurston GD, et al. Cardiovascular mortality
and long-term exposure to particulate air pollution. Circulation
2004;109:71–7.
17. Mitchell R, Popham F. Effects of exposure to natural environment on
health inequalities: an observational population study. Lancet 2008;
372:1655–60.
18. Takona T, Nakamura K, Wattamabe M. Urban residential environ-
ments and senior citizens’ longevity in megacity areas: the importance
of walkable green spaces. J Epidemiol Community Health 2002;
56(12):913–8.
19. Maas J, Verheij RA, Groenewegen PP, de Vries S, Spreeuwenberg.
Green space, urbanity, and health: how strong is the relation? J Epide-
miol Community Health 2006;60:587–92.
20. Maas J, Verheij RA, de Vries S, Spreeuwenberg P, Schellevis FG, Groe-
newegen PP. Morbidity is related to a green living environment. J
Epidemiol Community Health 2009;63(12):967–73.
21. Lovasi GS, Quinn JW, Neckerman KM, Perzanowski MS, Rundle A.
Children living in areas with more street trees have lower prevalence of
asthma. J Epidemiol Community Health 2008;62(7):647–9.
22. Poland TM, McCullough DG. Emerald ash borer: invasion of the urban
forest and the threat to North America’s ash resource. J Forestry
2006;104:118–24.
23. NCHS. Compressed mortality fıle, 1989-1998. Washington DC:
National Center for Health Statistics, 2007.
24. NCHS. Compressed mortality fıle, 1999-2007. Washington DC:
National Center for Health Statistics, 2010.
25. Krieger N, Chen JT, Waterman PD, Mah-Jabeen S, Subramanian SV,
Carson R. Geocoding and monitoring of U.S. socioeconomic inequal-
ities in mortality and cancer incidence: does the choice of area-based
measure and geographical level matter? Am J Epidemiol
2002;156(5):471–82.
26. Fry J, Xian G, Jin S, et al. Completion of the 2006 National Land Cover
Database for the conterminous U.S. J Photogramm Eng Remote Sens
2011;77(9):858–64.
27. Baum CF. Residual diagnostics for cross-section rime series regression
models. Stata J 2001;1:101–4.
28. Stock JH, Watson MH. Heteroskedasticity-robust standard errors for
fıxed effects panel data regression. Econometrica 2008;40:155–74.
29. Rothman KJ, Greenland S, Lash TL. Modern epidemiology. 3rd ed.
Philadelphia PA: Wolters Kluwer, 2008.
30. Mitchell R, Popham F. Effect of exposure to natural environment on
health inequalities: an observational population study. Lancet
2008;372(9650):1655–60.
31. National Heart, Lung, and Blood Institute. Morbidity and mortality:
2007 chart book on cardiovascular, lung, and blood diseases. Bethesda
MD: NHLBI, 2007.
32. Donovan GH, Butry DT. Trees in the city: valuing street trees in
Portland, Oregon. Landsc Urban Plann 2010;94:77–83.
33. Greene WH. Econometric analysis. 4th ed. Upper Saddle River NJ:
Prentice Hall, 2000.
34. Troy A, Grove JM. Property values, parks, and crime: a hedonic anal-
ysis in Baltimore, MD. Landsc Urban Plann 2008;87(3):233–45.
35. Feng J, Yang W. Effects of particulate air pollution on cardiovascular
health: a population health risk assessment. PLoS One 2012;7(3):
e33385.
36. Donovan GH, Butry DT. The value of shade: estimating the effect of
urban trees on summertime electricity use. Energ Build 2009;41:662– 8.
37. van den Berg A, Maas J, verheij RA, Groenewegen PP. Green space as a
buffer between stressful life events and health. Soc Sci Med
2010;70(8):1203–10.
38. Angrist JD, Imbens GW. Two-stage least squares estimation of average
causal effects in models with variable treatment intensity. J Am Stat
Assoc 1995;90(430):431–42.
Appendix
Supplementary data
Supplementary data associated with this article can be found, in the
online version, at http://dx.doi.org/10.1016/j.amepre.2012.09.066.
Donovan et al / Am J Prev Med 2013;44(2):139–145 145
February 2013