Urban forest biodiversity and cardiovascular disease: Potential health benefits from California’s street trees

a) Health outcomes. We use California Department of Public Health’s Death Profile’s by zip code and year from 2010 to 2018. The data is a tabulation of the number of deaths to regis- tered California residents, organized by mortality-type. Because the data are prepared for pub- lic use–aggregated, deidentified, and censored–informed consent was not required. We extracted deaths from Disease of the Heart (Heart Disease) and Cerebrovascular disease (Stroke). Place of death is unobserved: deaths of California residents that occurred outside the boundaries of the zip code of their legal address were still counted toward the total for their address zip code. Non-California residents who died inside of California are not reflected in zip code totals. The California Department of Public Health censors mortality counts below 11 deaths per zip code. Zip codes with small populations had many years of censored data. Like Khatama et al.’s [47] study of how Medicaid access affects cardiovascular mortality, we restrict our analysis to uncensored units. Among 1,010 potential zip codes, we restricted to the 857 zip codes that were never censored between 2010–2018 for Heart Disease and the 241 zip codes that were never censored between 2010–2018 for Stroke. b) Urban forest characteristics. To calculate biodiversity, we use a unique dataset from several arborist companies who collectively make up the majority of arborist services in Cali- fornia. The shared tree inventory data was collected between 2014 and 2019 and includes one

observation per tree sampled. The data come from a convenience sampling frame; the likeli- hood of sampling was not randomly assigned. The likelihood of sampling is the product of interactions between municipalities, private companies, individuals and the arborist compa- nies servicing their trees. In terms of the sampling rate, we estimate it is around 3.6% based on data for Los Angeles: our data include about 25.5 thousand trees whereas the city of Los Ange- les is estimated to have about 700 thousand trees (LA Bureau of Street Services 2020). Non- random sampling brings a risk of selection bias in the types of trees included in our sample. We address this concern with robustness checks that i) limit the sample to zip codes with many trees and ii) include socioeconomic controls. We also plot correlations between our bio- diversity measure and household income and pollution in Supporting Information. Fig 1 shows the extent of our data in California by zip code tabulation area (ZCTA), a geos- patial unit used by the US Census Bureau to facilitate relating census data to zip codes. We measure biodiversity using the Shannon Index, described in detail in the Measures section, for each ZCTA. We also control for exposure to trees using the number of trees sampled per 100,000 individuals by ZCTA. We merged these data with mortality rates calculated using the California Department of Public Health’s data. Connecting health data with our urban forest characteristics comes with two empirical challenges. First, the urban forest biodiversity measure is based on sampling between 2014–

2019, whereas the mortality data range from 2010–2018. This introduces measurement error that could bias our estimates downward. Second, the geographical boundaries of each dataset are not identical: the arborist data and population estimates are at the ZCTA level and mortal- ity data is at the zip code level. Although U.S. Census-defined ZCTA boundaries are designed to correspond to US Postal Service zip-code boundaries, they often differ spatiotemporally as zip codes boundaries are moved, removed, or added without correspondence with ZCTA boundaries [48]. Some degree of mismatch joining across zip code and ZCTA is unavoidable but is assumed to be manageably small given that ZCTA is intended to be most compatible with zip code geographic boundaries. We begin our analysis in 2010 in order to maintain 2010 Census-defined ZCTA boundaries across observations. To facilitate exposition, we use the term zip code when referring to our geographic unit of analysis in the results section of the paper. c) Sociodemographic and pollution characteristics. To calculate mortality and tree exposure rates per individual, we use two sources for zip code population. For 2010 we used PLOS ONE Biodiversity and cardiovascular disease PLOS ONE | https://doi.org/10.1371/journal.pone.0254973 November 3, 2021 5 / 23