1# RECORD b3ec523ce2f4119ea59d806bdaeb62c8215fe9a82# TITLE: The implicit value of tree cover in the U.S.: A meta-analysis of hedonic property value studies3# YEAR: 2016.0 VENUE: nan DOI: 10.1016/j.ecolecon.2016.04.0164# AUTHORS: S. D. Siriwardena; K. Boyle; T. Holmes; P. Wiseman5# PAGES: 9 CANDIDATE TABLE PAGES (1-based): [5, 7, 9]6# ABSTRACT: Trees in residential neighborhoods and communities provide benefits for homeowners that are capitalized into residential property values. In this paper, we collected data from hedonic property value studies and merged these data with ancillary spatial data describing forest and socio-economic characteristics surrounding each study area to conduct a meta-analysis of the impact of tree canopy cover on the value of residential properties. The meta-analysis suggests that property-level tree cover of about 30% and county-level tree cover of about 38% maximize the implicit price of tree cover in property values. Currently, tree cover in the original study areas was about 14%, on average, around or near study properties. The empirical results, therefore suggest under investment of tree cover on private property from the perspective of individual property owners and from a societal perspective. The findings also have implications for community forest programs regarding planting trees and protection of mature trees to address potential changes in tree abundance, species diversity and stand age due to development and climate change.789===== PAGE 1 (score 17, numeric-paren 0, table-words 0, var-words 17) =====1011Analysis12The implicit value of tree cover in the U.S.: A meta-analysis of hedonic13property value studies☆14Shyamani D. Siriwardena a,⁎,1, Kevin J. Boyle b,1, Thomas P. Holmes c,1, P. Eric Wiseman d,115a Department of Forest Resources and Environmental Conservation, 313 Cheatham Hall (0324), Virginia Tech, Blacksburg, VA 24061, USA16b Virginia Tech Program in Real Estate, 420 Bishop-Favrao Hall (0715), Virginia Tech, Blacksburg, USA17c USDA Forest Service, Southern Research Station, Research Triangle Park, NC, USA18d Department of Forest Resources and Environmental Conservation, Virginia Tech, Blacksburg, USA19a b s t r a c t20a r t i c l e21i n f o22Article history:23Received 18 March 201524Received in revised form 1 March 201625Accepted 3 April 201626Available online 4 May 201627Trees in residential neighborhoods and communities provide benefits for homeowners that are capitalized into28residential property values. In this paper, we collected data from hedonic property value studies and merged29these data with ancillary spatial data describing forest and socio-economic characteristics surrounding each30study area to conduct a meta-analysis of the impact of tree canopy cover on the value of residential properties.31The meta-analysis suggests that property-level tree cover of about 30% and county-level tree cover of about3238% maximize the implicit price of tree cover in property values. Currently, tree cover in the original study33areas was about 14%, on average, around or near study properties. The empirical results, therefore suggest34under investment of tree cover on private property from the perspective of individual property owners and35from a societal perspective. The findings also have implications for community forest programs regarding plant-36ing trees and protection of mature trees to address potential changes in tree abundance, species diversity and37stand age due to development and climate change.38© 2016 Elsevier B.V. All rights reserved.39Keywords:40Hedonic valuation41Tree canopy cover42Community forestry43Forest matrix44Climate change451. Introduction46Tree cover, the land area covered by tree crowns, in residential47neighborhoods is an environmental attribute that provides a suite of48ecosystem services (e.g., landscape esthetics, wildlife habitat, energy49conservation, stormwater control) to homeowners and communities50(Nowak et al., 2010). Most trees in urban and suburban neighborhoods51are located on private property where the costs of tree planting and52maintenance are borne by homeowners, while the benefits of trees53can be enjoyed by the greater community. The disparity between who54pays the costs (homeowners) and who receives the benefits from55trees on privately-owned property (homeowners and the community)56suggests that the level of tree cover on privately-owned land may be so-57cially sub-optimal. A better understanding of the costs and benefits of58trees, within the context of multiple social-ecological settings (Pickett59et al., 2011), can help community planners design tree planting and60protection programs to improve the quality of life for people and justify61the costs of such programs.62Although the economic benefits of tree canopy cover (tree cover63hereafter) are not directly priced, one component of their economic64value, the implicit value to property owners, can be estimated using65hedonic property value models. Hedonic models decompose the total66price paid for a property into the “implicit” prices of property character-67istics. The implicit price of a characteristic, such as the amount of tree68cover, is the amount home buyers pay for a small change in the level69of the characteristic, holding all other property characteristics constant70(Taylor, 2003).71Previous hedonic studies have found that the value of forested areas72varies with the degree of forest fragmentation and urbanization. For73example, larger forest patches are valued more in the urban core,74while a more fragmented and diverse landscape is preferred in rural–75urban interfaces (Cho et al., 2008). Trees are valued more in areas76where they are scarce and less so in heavily forested areas (Netusil77et al., 2010). Large trees with full canopies have been shown to enhance78retail settings (Wolf, 2004) and large shade trees are preferred by79residents in areas with hot summers (Schroeder et al., 2006). Temporal80variations of the value of trees has also been observed by Cho et al.81(2009) where the amenity value of trees increased over time with82declining forest patch size and density.83Forested areas can also present real or perceived risks that lower84property values. Locations that have experienced or where there are85Ecological Economics 128 (2016) 68–7686☆This research was supported by the Southern Research Station of the U.S. Forest87Service (12-JV-11330143-106) and Virginia Agricultural Experimental Station.88⁎ Corresponding author.89E-mail addresses: shyadil@vt.edu (S.D. Siriwardena), kjboyle@vt.edu (K.J. Boyle),90tholmes@fs.fed.us (T.P. Holmes), pwiseman@vt.edu (P.E. Wiseman).911 Siriwardena, Boyle and Wiseman are PhD candidate, Professor and Associate92Professor, respectively, at Virginia Tech. Holmes is an Economist with the U.S. Forest93Service.94http://dx.doi.org/10.1016/j.ecolecon.2016.04.016950921-8009/© 2016 Elsevier B.V. All rights reserved.96Contents lists available at ScienceDirect97Ecological Economics98journal homepage: www.elsevier.com/locate/ecolecon99100101===== PAGE 5 (score 36, numeric-paren 0, table-words 9, var-words 14) =====102103purchasers place on tree cover. These socioeconomic variables and104the other ancillary variables discussed above are likely to be fixed105for any given hedonic study, but vary spatially across studies and106provide the potential for a meta-analysis to provide insights on the107value of tree cover beyond what can be observed in any single108hedonic study.1094. Model Specification110The first model to be estimated is the selection model (Eqs. (4a) and111(4b)). The selection Eq. (4a) is specified as:112h Z; γ113ð114Þ ¼ ϒ0 þ ϒ1onprop þ ϒ2onprop medtrees þ ϒ3onprop oldtrees115þ ϒ4onprop invasive þ ϒ5onprop temp þ ϒ6popdensity116þ ϒ7medianincome þ ω117ð5aÞ118where Z is a vector of study and context characteristics, and γ is a vector119of coefficients to be estimated. Each of the variables listed in Eq. (4a) is120defined in Table 2. The dependent variable is binary and indicates if an121implicit price is positive (=1) or negative (=0), and the selection equa-122tion is specified as a probit. The selection equation includes an explana-123tory variable (onprop) delineating whether the tree cover measure is on124a property (=1) or includes tree cover near a property (=0), and this125variable is interacted with other variables that may influence whether126tree cover is viewed as an amenity or disamenity (medtrees, oldtrees,127invasive, and temp). As noted previously, older trees may provide128shade, but also potentially pose a greater risk of damage or injury. Inva-129sive tree pests may affect the amenity value of trees and may increase130the risk of tree breakage. Higher temperatures may create a desire for131more tree shade for outdoor comfort and air conditioning savings,132while lower temperatures may create a desire for less tree shade to133allow winter sunlight to heat the home and hasten melting of ice and134snow. Population density and median income are included to consider135if these demographic characteristics affect whether tree cover is viewed136as an amenity.137The meta-Eq. (4b) is specified as:138gþ X; θþ139140141¼ θþ1420 þ θþ1431 treecover þ θþ1442 treecover2 þ θþ1453 ctreecover146þθþ1474 ctreecover2 þ θþ1485 houseprice þ θþ1496 eastern þ θþ1507 pacificnw151þ θþ1528 medtrees þ θþ1530 oldtrees þ θþ15410signif þ θþ15511popdensity156þ θþ15712medianincome þ εþ158ð5bÞ159where all variables are defined in Table 2 and the θi160+ are coefficients to161be estimated. The dependent variable is the consistent measure of the162implicit price (effect size) explained in the previous section. The163treecover variable represents the tree cover variable from the original164hedonic models, and ctreecover is included to consider if tree cover in165the greater study area where people work, shop, and recreate also166Table 2167Definitions and summary statistics of variables used to estimate selection equation and meta-equation.168Variable169Description170Positive observations (IP N 0)171Mean172(min, max)173Negative observations (IP b 0)174Mean175(min, max)176Dependent variable in selection equation177amenity1781 if implicit price (IP) is positive and 0 if IP is negative1790.64180(0.00, 1.00)181NA182Dependent variable in meta-equation183IP184Implicit price of tree cover for a 1% increase ($2013)185$239.41186($0.32, $2696.15)187-$156.26188(−$895.94, −$0.04)189Independent variables190onprop1911 if treecover on property, 0 otherwise1920.35193(0.00, 1.00)1940.40195(0.00, 1.00)196medtreesa197% of county forests aged 40 to 119 years19869199(8, 91)20064201(45, 91)202oldtreesa203% of county forests aged 120 years or older20411205(0.00, 44)20616207(0.00, 44)208invasive2091 if invasive species or pathogen, 0 otherwise2100.80211(0.00, 1.00)2120.70213(0.00, 1.00)214temp215Annual days with temperature above 90 °F21615217(0.00, 76)21814219(3, 45)220popdensity221County population density (people/mile2)222886223(7, 2715)2241245225(7, 2715)226medianincome227County median income228$56,084229($34,000, $81,330)230$53,911231($35,613, $81,330)232treecover233% tree cover on or near property23414235(0.10, 61)23612237(0.00, 40)238ctreecover239% county-level tree cover24044241(13,66)24248243(13, 66)244houseprice245Mean sale price of properties ($2013)246$285,138247($24,576, $494,894)248$249,200249($24,576, $494,894)250easternb2511 if study area is east of Mississippi River, 0 otherwise2520.69253(0.00, 1.00)2540.47255(0.00, 1.00)256pacificnwb2571 if study area is in Oregon or Washington, 0 otherwise2580.21259(0.00, 1.00)2600.42261(0.00, 1.00)262signif2631 if tree cover significant at 10% level, 0 otherwise2640.66265(0.00, 1.00)2660.47267(0.00, 1.00)268Minimum and maximum of the variables are given in parenthesis.269a Percent of county forest below 40 years of age is the omitted category.270b Omitted category includes the following study locations in the mountain west: Greater Flagstaff area, Arizona; Grand County, Colorado; and Flathead, Lake, Sanders, Lincoln271and Missoula Counties, Montana.27272273S.D. Siriwardena et al. / Ecological Economics 128 (2016) 68–76274275276===== PAGE 6 (score 26, numeric-paren 0, table-words 8, var-words 10) =====277278affects the implicit value of tree cover on or near a property. Both of279these variables are included as linear and squared terms (treecover2280and ctreecover2) to allow for nonlinear relationships between the im-281plicit value and the tree cover measures.282The mean value of house prices (houseprice) from the original283hedonic studies is included to see if implicit values of tree cover are284affected by selling prices of properties. Binary variables for studies285conducted in the eastern U.S. and Pacific Northwest are included to286control for spatial fixed effects in different areas of the country287(e.g., the risk of forest fires). The omitted category includes studies288conducted in the mountain west. Tree age is included because the289age of trees affects the size of the crown and could affect the implicit290price of tree cover. The omitted category is young trees (b40 years of291age). Finally, a variable is included to see if the significance of292implicit price estimates in the original hedonic studies affects the293magnitudes of the estimates in meta-equation; all implicit price294estimates, statistically significant and insignificant, from the hedonic295studies are included in the meta-data.296The selection model estimates Eqs. (5a) and (5b) jointly297(Wooldridge, 2002), and we conducted a test for the presence of298selection from only using studies with positive implicit prices in299the estimation of Eq. (5b):300Ho : ρ ¼ 0 vs Ha : ρ ≠0301ð6Þ302where ρ is the correlation between the error terms in Eqs. (5a) and303(5b) (ω and ε+, respectively). If the null hypothesis cannot be304rejected, then the meta-equation (5b) can be estimated solely,305without the selection equation.306A number of robustness analyses are conducted to provide insight307into how much confidence can be placed in the meta-equation coeffi-308cient estimates and thus conclusions drawn from the meta-analysis.309Given that most studies provided multiple implicit-price estimates310and these multiple estimates from individual studies cannot be as-311sumed to be independent observations, the meta-equation is estimated312using OLS, study fixed effects, random effects and multilevel mixed-313effects. In addition, horizontal robustness checks, excluding individual314observations or studies (groups of observations) are also conducted315(Boyle et al., 2013). The primary focus of these robustness analyses is316to consider if the exclusions of observations from the data affects the317estimates of the coefficients on treecover and ctreecover. The results of318these robustness analyses are reported below.3195. Results320Summary statistics for the variables used in the estimation are321reported in Table 2. These data indicate that 64% of the observations322have positive implicit prices. Of the studies with positive implicit323prices, the average implicit price for a 1% change in tree cover is324$239, and the comparable figure for negative implicit prices is −325$156. Values of the variable treecover for the positive implicit prices326range from 0.1% to 61% and the comparable range for negative327implicit prices is 0% to 40%.3285.1. Selection Equation329Estimation of the Heckman sample selection model (Eqs. (5a) and330(5b)) resulted in the conclusion that the null hypothesis of no correla-331tion (Eq. (6)) could not be rejected (p = 0.93). As there is no sample332selection, the primary focus of the analyses reported in this paper is333the meta-equation. The results from the estimation of the Heckman334selection equations are reported in the Appendix A.3355.2. Meta-equation Estimation Results and Discussion336The meta-equation is estimated with White's consistent standard337errors (Nelson and Kennedy, 2009).6 In this estimation, we found338that popdensity and medianincome were insignificant. This is driven339in large part by the study-specific auxiliary data being correlated,340which reduces the efficiency of the estimation of all coefficient esti-341mates. Thus, we test the null hypothesis that the coefficients on342popdensity and medianincome are jointly insignificant in the meta-343equation:344Ho : θþ34511 ¼ θþ34612 ¼ 0 vs Ho : θþ34711 ≠0; θþ34812 ≠0349ð7Þ350We could not reject this null hypothesis (F = 2.11, p = 0.13). Given351this result, the estimation results reported here are for Eq. (5b) with the352popdensity and medianincome variables excluded.7353The treecover, treecover2, ctreecover and ctreecover2 variables all have354significant coefficient estimates (Table 3). The linear terms of these var-355iables have positive coefficient estimates and the squared terms have356negative coefficients. This supports the intuition from previous research357that people value tree cover up to a point, but beyond this point tree358cover declines in value. This could occur for a variety of reasons,359including:360• some shade is good, but too much shade blocks sunlight for lawns and361passive solar heating;362• tree cover provides privacy, but too much cover may hamper home363security or obscure vistas;364• trees make a home more attractive, but too much tree cover may lead365to buildup of mildew and algae on exterior structures and excessive366cleanup of leaves and litter;367• trees attract watchable wildlife, but too much tree cover may attract368nuisance wildlife;369• trees provide cover from harsh wind and rain, but may break apart in370inclement weather or create a wildfire hazard;371There could be a variety of other explanations, but these considerations372are consistent with what other investigators have found.373It is interesting to note that the implicit prices of tree cover are374higher for eastern study sites (eastern) than for mountain west study375sites (the omitted category in the estimation), and implicit prices for376Pacific Northwest studies are not significantly different from mountain377west studies. This difference could be due to a higher risk of forest378fires in the mountain west and Pacific Northwest, a preference for379more vistas in the west, a preference for more privacy in the east, or380other explanations.381The coefficient for older forests (oldtrees) is positive and significant,382indicating significantly higher implicit prices for study sites with older383versus younger forests (the omitted category) in the county where the384studies were conducted. This suggests people value living in areas of385mature forests that provide more shade and may be visually appealing.83866 To account for multiple observations from most studies, we estimated the meta-387equation using fixed-effects, random-effects, and mixed-effects estimation. The fixed-388effects estimation did not work because of collinearity issues. The random-effects and389mixed-effects estimations did not change the quantitative and qualitative results for the390meta-equation coefficient estimates.3917 We estimated the meta-equation using negative implicit prices as the dependent var-392iable and none of the explanatory variables were significant.3938 This is an interesting outcome when compared to the result from the selection equa-394tion reported in the Appendix. Older trees increase the probability of a negative implicit395price, but also increase the magnitude of positive implicit prices. These results suggest396the concurrent amenity/disamenity of older trees; more shade and a desirable esthetic,397but higher perceived risks with more maintenance costs.39873399S.D. Siriwardena et al. / Ecological Economics 128 (2016) 68–76400401402===== PAGE 7 (score 58, numeric-paren 11, table-words 21, var-words 3) =====4034045.3. Tree Cover than Maximizes Property Values405Using the results shown in Table 3, the levels of treecover (tc) and406ctreecover (ctc) that maximize implicit prices were computed as:407∂IPþ408∂tc ¼ 0 and ∂IPþ409∂ctc410ð8Þ411The level of treecover that maximizes implicit prices is 30% with a412standard error of 15% and the comparable level for ctreecover is 38%413with a standard error of 3%.9414It is worth noting that treecover ranges from 0.1% to 61% in the data415used to estimate the meta-equation and the comparable range for416ctreecover is 13% to 66% (see Table 2). Thus, the levels of treecover and417ctreecover that maximize implicit prices are within the observed data418on tree cover. Increasing tree cover beyond these levels does not result419in a negative implicit price; it is just that the implicit prices decline from420their maximum values.421An additional consideration is that the mean value of treecover is42214% (Table 2), while the value that maximizes implicit price for this423variable is 30% (more than one standard deviation below the value424that maximizes implicit prices). Further, 14% tree cover on or near425a property is much less for tree cover in the county where the prop-426erties are located (44%, Table 2). These relationships support the no-427tion that individual property owners underinvest in trees on their428property from both the perspective of individual property owners429and a societal perspective.4305.4. Robustness of Meta-equation Coefficient Estimates431Conducting robustness checks in any empirical analysis is always432important, but it is particularly important for this analysis when estima-433tion is done with only 68 observations. With a small sample size, a single434observation or group of observations (study) has the potential to influ-435ence estimation results.436Removal of single observations and re-estimation of the meta-437equation did not induce major changes in the estimates of the key coef-438ficients for treecover, treecoversq, ctreecover and ctreecoversq. None of439these coefficient estimates changed sign, the average change in the440magnitude of the estimated coefficients for these variables is only 4%,441and there were three occurrences where treecover and treecoversq442became insignificant (Table 4). The only variable that exhibited443substantial sensitivity to removing observations is houseprice with an444average effect of 19%, but this variable is insignificant in the original445meta-equation reported in Table 3 and only becomes significant in446two instances in the observation-removal robustness analysis.447In contrast, all coefficient estimates for treecover, treecoversq,448ctreecover and ctreecoversq exhibited more variability when studies449(sets of observations) are removed from the estimation. Recall,450there are thirteen studies with positive implicit price estimates and451the number of observations range from one to ten per study452(Table 1). Here the coefficient estimates for treecover, treecoversq,453ctreecover and ctreecoversq sometimes changed significance or sign454up to two times; the average change in coefficient estimates ranges45535% (treecover) to 56% (ctreecover) (Table 5).456Given these study sensitivity results, it is logical to ask which studies457are affecting the estimation. This analysis is conducted by looking at the458changes in the level of tree cover that maximizes implicit prices459(Table 6). Recall the level of treecover that maximizes implicit prices,460based on the full estimates with all observations, is 30% and the compa-461rable value for ctreecover is 38%. The good news is that only three studies462resulted in changes of the level of treecover by more than 10%, but the463bad news is that the effects were quite large: 51% (Holmes et al.,4642006), −26% (Netusil et al., 2010) and −11% (Drake-McLaughlin and465Netusil, 2011). The effects on the level of ctreecover were generally466smaller; again, two studies resulted in greater than 10% changes in the467maximum implicit price(−13% — Holmes et al., 2006 and 15% —468Holmes et al., 2010). It is important to note that studies with the largest469percentage deviations in the levels of treecover that maximize implicit470prices, when their observations are removed from the estimation,471have some of the largest numbers of observations in the data (9 —472Holmes et al., 2006; 10 — Holmes et al., 2010; 7 — Netusil et al., 2010;4737 — Drake-McLaughlin and Netusil, 2011). The key insight from the474study robustness checks is that while coefficient estimates varied, in475most cases the linear and square term coefficient estimates moved in476a relationship that does not change the levels of treecover and ctreecover477that maximize implicit prices.478The violations of robustness do not imply that the influential479studies should be removed from the analysis, nor do they imply480that the meta-analysis results are invalid, but they do suggest that481the findings reported here should be interpreted as preliminary482and with caution. More studies are needed to add spatial breadth483and location-specific depth of implicit price estimates to confirm484the preliminary findings presented here. It is crucial that the future485studies focus on percent tree cover as the key variable so that the486new empirical estimates can be merged with the empirical estimates487used here to estimate a new meta-equation.4886. Conclusions489This study finds a nonlinear relationship between the implicit490value of tree cover and two types of tree canopy cover: tree cover491on or near a property, and tree cover in the county where a property492is located. This nonlinear relationship allows the computation of the4939 The standard errors were approximated using the Marcov Chain Monte-Carlo494(MCMC) approach using the Gibbs sampler (Plassmann and Khanna, 2007). This method495generates a large number of samples from posterior distributions of coefficient estimates496for treecover and ctreecover and posterior distributions of turning points of treecover and497ctreecover that maximize the implicit prices. Posterior means and standard deviations498are calculated for these finite sample approximations of the sampling distribution of the499levels of treecover and ctreecover that maximize implicit prices.500Table 3501Meta-equation coefficient estimates (positive implicit prices, IP+).502Variables503Coefficient estimates504treecover50522.32⁎506(12.64)507treecover2508−0.37⁎⁎509(0.18)510ctreecover51184.10⁎⁎512(35.94)513ctreecover2514−1.11⁎⁎515(0.47)516houseprice517−0.0002518(0.0007)519eastern5201393.97⁎⁎521(681.76)522pacificnw523351.75524(239.73)525oldtrees52641.07⁎⁎527(17.66)528medtrees5295.08530(4.63)531signif532100.54533(98.15)534intercept535−2971.9⁎⁎536(1176.81)537n53868539R 25400.27541Robust standard errors are given in parentheses.542⁎ Indicates significance at the 10% level.543⁎⁎ Indicates significance at the 5% level.54474545S.D. Siriwardena et al. / Ecological Economics 128 (2016) 68–76546547548===== PAGE 8 (score 18, numeric-paren 0, table-words 3, var-words 12) =====549550amount of property-level and county-level tree cover that maxi-551mizes the implicit value of tree canopy cover: 30% for tree cover on552or near a property, and 38% at the county level. This pattern of results553suggests that property owners prefer more tree cover in the general554area where they work, shop and recreate, as a public good, than555on their own property, as a private good. This discrepancy may556be due to the fact that homeowners generally pay all of the557planting and maintenance costs and bear most of the perceived558risks (i.e., wildfire and storm damage) for trees located on their559property. When these costs and risks are either borne by other560private landowners, or are shared by all taxpayers for trees lo-561cated on public land, higher levels of county-level tree cover562are preferred.563The finding that the implicit value of tree cover is maximized at564about 38% for county-level cover is consistent with the often-cited56540% tree cover espoused by American Forests as an ecological goal for566communities on the East Coast and Pacific Northwest of the United567States. The average urban tree cover in the U.S. is currently about 35%568(Nowak et al., 2010). Therefore, in many areas, overall tree cover569could be rightly increased with a concomitant positive effect on residen-570tial property values.571The preference for old trees (more than 120 years old) found in572this meta-analysis is another interesting result. A simplistic view of573land development may be to cut down old trees and then plant new574ones upon completion of construction rather than invest in protec-575tion and restorative care of old trees; here again, this may be indic-576ative of underinvestment in community tree cover. These veteran577trees are much more ecologically valuable than young trees for car-578bon storage, air pollution abatement, stormwater interception, and579energy conservation, and these results provide evidence that they580have more amenity value too. Thus, communities might wish to581consider public outreach programs, zoning policies, and tax incen-582tives to protect and maintain older tress. Although such endeavors583come at a cost for the community, these costs might be recouped in584part through additional real estate tax revenues on higher-585appraising residential property.586While the findings of this study may be helpful to local officials587involved in planning community forestry programs, we expect that588decision-making by both community planners and households589will become more complex in the future due to various outcomes590anticipated from a changing climate. On the one hand, a hotter cli-591mate may increase the value of tree cover for its ability to shade592and cool homes. This preference may be most acute in urban593areas that are vulnerable to “heat island” effects. Further, tree can-594opy and forested riparian buffers intercept stormwater, helping to595mitigate the flooding and water quality effects of runoff. On the596other hand, climate change may increase the frequency and sever-597ity of extreme weather events, which could lead to an increase in598storm damage to trees, causing increased power outages and debris599cleanup costs. This may push communities to think twice about600maintaining large, mature trees in residential areas. Consequently,601the future may bring complex challenges to residential property602owners and community forest planners alike. Because tree-cover603considerations may vary by geographic location, preferences for604property-level and county-level tree cover may become even605more nuanced and, over time, local planners may find it useful to606engage residents to discuss and refine the goals of community for-607estry programs.608Table 4609Observation robustness (removing one observation at a time from estimation).610Variables611Number of612times sign613changed614Number of times615significance changed616(at 10%)617Absolute, percent changes in618coefficient magnitudes619Average620(min, max)621treecover622062336244%625(0, 52)626treecover2627062836294%630(0, 39)631ctreecover632063306344%635(0, 21)636ctreecover2637063806394%640(0, 19)641houseprice6422643064419%645(0, 177)646eastern647064816495%650(0, 52)651pacificnw652065366545%655(0, 40)656oldtrees657065816594%660(0, 31)661medtrees662066326647%665(0, 179)666signif667066806697%670(0, 55)671Minimum and maximum of the variables are given in parentheses.672Table 5673Study robustness (removing one study at a time from estimation).674Variable675Number of676times sign677changed678Number of times679significance changed680(at 10%)681Absolute, percent changes in682coefficient magnitudes683Average684(min, max)685treecover6860687368835%689(1, 97)690treecover26912692269346%694(3, 167)695ctreecover6961697169856%699(1, 291)700ctreecover27011702170355%704(0, 294)705houseprice70657071708225%709(4, 930)710eastern7112712271365%714(6, 272)715pacificnw7161717271866%719(2, 315)720oldtrees7211722272359%724(7, 244)725medtrees7262727372888%729(0, 443)730signif7310732073319%734(0, 95)735Minimum and maximum of the variables are given in parentheses.736Table 6737Changes in implicit prices by study for study robustness.738Study739Percent changes740treecover741ctreecover742Coley (2005)7434%744−2%745Dimke (2008)746NA747−2%748Drake-McLaughlin and Netusil (2011)749−11%7501%751Holmes et al. (2006)75251%753−13%754Holmes et al. (2010)7552%75615%757Kim and Wells (2005)7582%759−1%760Mansfield et al. (2005)7610%7625%763Netusil et al. (2010)764−26%7650%766Paterson and Boyle (2002)767NA768−3%769Price et al. (2010)770−1%7711%772Sander et al. (2010)7732%774−2%775Sander and Haight (2012)7761%777−1%778Stetler et al. (2010)7797%780−4%781NA indicates coefficient estimate of treecoversq and ctreecoversq have the wrong signs and782are insignificant so tree cover that maximizes implicit prices could not be calculated.78375784S.D. Siriwardena et al. / Ecological Economics 128 (2016) 68–76785786787===== PAGE 9 (score 46, numeric-paren 22, table-words 1, var-words 18) =====788789Appendix A. Heckman Selection Equation Estimation790Table A1791Joint estimation of selection equation and meta-equation.792Variable793Coefficient794Selection equation795onprop7963.27797(4.03)798onprop⁎medtrees7990.003800(0.06)801onprop⁎oldtrees802−0.19⁎⁎803(0.07)804onprop⁎invasive805−2.87806(2.69)807onprop⁎temp8080.01809(0.05)810popdensity8110.0004⁎⁎812(0.0002)813medianincome8140.00002⁎815(0.00001)816intercept817−1.03818(0.71)819Meta-equation820treecover82119.09822(11.92)823treecover2824−0.33⁎825(0.17)826ctreecover82749.23828(40.06)829ctreecover2830−0.70831(0.51)832houseprice833−0.002834(0.002)835eastern836735.87837(1132.8)838pacificnw839397.96840(667.9)841medtrees84211.72843(8.55)844oldtrees84519.53846(23.83)847signif848112.53849(94.06)850popdensity851−0.21852(0.15)853medianincome8540.005855(0.02)856intercept857−1696.32858(1151.4)859ρ860−0.05861(0.52)862Robust standard errors are given in parentheses.863⁎ Indicates significance at the 10% level.864⁎⁎ Indicates significance at the 5% level.865References866Bentz, B., 2008. Western U.S. bark beetles and climate change. U.S. Department of Agricul-867ture, Forest Service, Climate Change Resource Center (http://www.fs.fed.us/ccrc/868topics/bark-beetles.shtml, accessed 11.02.2014).869Boyle, K.J., Parmeter, C.F., Boehlert, B.B., Paterson, R.W., 2013. Due diligence in meta-870analyses to support benefit transfers. Environ. Resour. Econ. 55 (23), 357–386.871Cho, S.-H., Poudyal, N.C., Roberts, R.K., 2008. Spatial analysis of the amenity value of green872open space. Ecol. Econ. 66 (2), 403–416.873Cho, S.-H., Kim, S.G., Roberts, R.K., Jung, S., 2009. Amenity values of spatial configurations874of forest landscapes over space and time in the Southern Appalachian Highlands.875Ecol. Econ. 68 (10), 2646–2657.876Cho, S.H., Lambert, D., Kim, S., Roberts, R., Park, W., 2011. Relationship between value of877open space and distance from housing locations within a community. J. Geogr. Syst.87813 (4), 393–414.879Coley, M.C., 2005. House and Landscape Value: An Application of Hedonic Pricing Tech-880nique Investigating Effects of Lawn Area on House Selling Price (MS Thesis) The Uni-881versity of Georgia.882Dale, V.H., Joyce, L.A., McNulty, S., Neilson, R.P., Ayres, M.P., Flannigan, M.D., Hanson, P.J.,883Irland, L.C., Lugo, A.E., Peterson, C.J., Simberloff, D., Swanson, F.J., Stocks, B.J.,884Wotton, B.M., 2001. Climate change and forest disturbances. Bioscience 51 (9),885723–734.886Dimke, K.C., 2008. Valuation of Tree Canopy on Property Values of Six Communities in887Cincinnati, Ohio (PhD Dissertation) The Ohio State University.888Drake-McLaughlin, N., Netusil, N.R., 2011. Valuing walkability and vegetation in Portland,889Oregon. In: von Haefen, R.H. (Ed.), Twenty-second Interim Report and Proceedings890from the Annual Meeting, W2133, pp. 173–201.891Frankel, S.J., 2008. Forest Plant Diseases and Climate Change. U.S. Department of Agricul-892ture, Forest Service, Climate Change Resource Center (http://www.fs.usda.gov/ccrc/893topics/plant-diseases.shtml, accessed 11.02.2014).894Ghermandi, A., Nunes, P.A., 2013. A global map of coastal recreation values: results from a895spatially explicit meta-analysis. Ecol. Econ. 86, 1–15.896Heckman, J.J., 1979. Sample selection bias as a specification error. Econometrica 47 (1),897153–161.898Holmes, T.P., Murphy, E.A., Bell, K.P., 2006. Exotic forest insects and residential property899values. Agric. Resour. Econ. Rev. 35 (1), 155–166.900Holmes, T.P., Murphy, E.A., Bell, K.P., Royle, D.D., 2010. Property value impacts of hemlock901woolly adelgid in residential forests. For. Sci. 56 (6), 529–540.902Homer, C., Dewitz, J., Fry, J., Coan, M., Hossain, N., Larson, C., Herold, N., McKerrow, A.,903VanDriel, J.N., Wickham, J., 2007. Completion of the 2001 national land cover database904for the conterminous United States. Photogramm. Eng. Remote Sens. 73 (4), 337–341.905Hope, D., Gries, C., Zhu, W., Fagan, W.F., Redman, C.L., Grimm, N.B., Nelson, A.L., Martin, C.,906Kinzig, A., 2003. Socioeconomics drive urban plant diversity. Proc. Natl. Acad. Sci. 100907(15), 8788–8792.908Huggett Jr., R.J., 2003. Fire in the Wildland-Urban Interface: An Examination of the Effects of909Wildfire on Residential Property Markets (PhD Dissertation) North Carolina State910University.911Iverson, L.R., Prasad, A.M., 2001. Potential changes in tree species richness and forest com-912munity types following climate change. Ecosystems 4 (3), 186–199.913Johnston, R.J., Besedin, E.Y., Stapler, R., 2014. Enhanced geospatial data for meta-analysis914and environmental benefit transfer: an application to water quality improvements.915Presented at the Meta-analysis of Economics Research Network, MAER-Net 2014916Athens Colloquium, Athens, Greece.917Kim, Y.S., Wells, A., 2005. The impact of forest density on property values. J. For. 103 (3),918146–151.919Mansfield, C., Pattanayak, S.K., McDow, W., McDonald, R., Halpin, P., 2005. Shades of920green: measuring the value of urban forests in the housing market. J. For. Econ. 11921(3), 177–199.922Nelson, J.P., Kennedy, P.E., 2009. The use (and abuse) of meta-analysis in environmental and923natural resource economics: an assessment. Environ. Resour. Econ. 42 (3), 345–377.924Netusil, N.R., Chattopadhyay, S., Kovacs, K.F., 2010. Estimating the demand for tree cano-925py: a second-stage hedonic price analysis in Portland, Oregon. Land Econ. 86 (2),926281–293.927Nowak, D.J., Greenfield, E.J., 2012. Tree and impervious cover change in US cities. Urban928For. Urban Green. 11 (1), 21–30.929Nowak, D.J., Stein, S.M., Randler, P.B., Greenfield, E.J., Comas, S.J., Carr, M.A., Alig, R.J., 2010.930Sustaining America's urban trees and forests: a forests on the Edge report. Gen. Tech.931Rep. NRS-62. U.S. Department of Agriculture, Forest Service, Northern Research932Station.933Paterson, R.W., Boyle, K.J., 2002. Out of sight, out of mind? Using GIS to incorporate visi-934bility in hedonic property value models. Land Econ. 78 (3), 417–425.935Pickett, S.T., Cadenasso, M.L., Grove, J.M., Booned, C.G., Groffman, P.M., Irwin, E., Kaushal,936S.S., Marshall, V., McGrath, B.P., Nilon, C.H., Pouyat, R.V., Szlavecz, K., Troy, A., Warren,937P., 2011. Urban ecological systems: scientific foundations and a decade of progress.938J. Environ. Manag. 92 (3), 331–362.939Plassmann, F., Khanna, N., 2007. Assessing the precision of turning point estimates in940polynomial regression functions. Econ. Rev. 26 (5), 503–528.941Prasad, A., Iverson, L.S., Matthews, S., Peters, M., 2009. Atlases of tree and bird species hab-942itats for current and future climates. Ecol. Restor. 27 (3), 260–263.943Price, J.I., McCollum, D.W., Berrens, R.P., 2010. Insect infestation and residential property944values: a hedonic analysis of the mountain pine beetle epidemic. Forest Policy Econ.94512 (6), 415–422.946Sander, H., Haight, R.G., 2012. Estimating the economic value of cultural ecosystem947services in an urbanizing area using hedonic pricing. J. Environ. Manag. 113,948194–205.949Sander, H., Polasky, S., Haight, R.G., 2010. The value of urban tree cover: a hedonic property950price model in Ramsey and Dakota Counties, Minnesota, USA. Ecol. Econ. 69 (8),9511646–1656.952Schroeder, H., Flannigan, G., Coles, R., 2006. Residents' attitudes toward street trees in the953UK and US communities. Arboricult. Urban For. 32 (5), 236–246.954Stetler, K.M., Venn, T.J., Calkin, D.E., 2010. The effects of wildfire and environmental955amenities on property values in northwest Montana, USA. Ecol. Econ. 69 (11),9562233–2243.957Taylor, L.O., 2003. The hedonic method. In: Champ, P., Boyle, K., Brown, T. (Eds.), A primer958on nonmarket valuation. Kluwer Academic Publishers, New York.959Troy, A.R., Grove, J.M., O'Neil-Dunne, J.P., Pickett, S.T., Cadenasso, M.L., 2007. Predicting960opportunities for greening and patterns of vegetation on private urban lands. Envi-961ron. Manag. 40 (3), 394–412.962U.S. Department of Labor, Bureau of Labor Statistics, 2015. Consumer Price Index Data963from 1913 to 2016. (http://www.usinflationcalculator.com/inflation/consumer-964price-index-and-annual-percent-changes-from-1913-to-2008/, accessed July9652013).966Wolf, K.L., 2004. Trees and business district preferences: a case study of Athens, Georgia,967US. J. Arboric. 30 (6), 336–346.968Woodall, C.W., Oswalt, C.M., Westfall, J.A., Perry, C.H., Nelson, M.D., Finley, A.O., 2009. An969indicator of tree migration in forests of the Eastern United States. For. Ecol. Manag.970257 (5), 1434–1444.971Wooldridge, J., 2002. Econometric Analysis of Cross Section and Panel Data. MIT Press,972Cambridge.97376974S.D. Siriwardena et al. / Ecological Economics 128 (2016) 68–76975