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

individuals using ZCTA-level population data. To stay consistent, we also transform our expo- sure variable to trees per 100,000 individuals. As discussed in the data section, mortality data correspond to the address of decedent, not the location of death. Thus, the relationship between exposure to biodiversity and mortality is approximated and based on the assumption that decedents spent some time near their address. Furthermore, people move. Our measure for biodiversity exposure likely reflects acute, recent exposure and may fail to reflect cumulative, chronic exposure. This limitation is common to other researchers who work with cardiovascular mortality rates [42,44]. Urban tree diversity is measured within each ZCTA at the genus-level using the Shannon- Weiner Index, Eq 1. In this expression, the share of trees within the first genus is p1 ¼ n1 N , N is the total number of trees in the ZCTA and n1 is the number of trees in the first genus of K total genera. Shannon Index ¼ ½p1 lnðp1Þ þ p2 lnðp2Þ þ � � � þ pK lnðpK Þ� ð1Þ The Shannon Index measures both evenness and richness across genera. Diversity was mea- sured at the genus-level to capture greater phylogenic diversity as compared to measurement at the species-level. We assess the Shannon Index at the species-level in our sensitivity checks. To account for the possibility that the tree characteristics do not have a linear relationship with heart disease and stroke mortality outcomes, we assigned binary variables for each quan-

tile of the Shannon index. Thresholds are between 0 to 2.491 for the first quantile (N = 1,935), 2.495 to 2.930 for the second quantile (N = 1,926), 2.931 to 3.186 for the third quantile (N = 1,926), and 3.187 to 3.830 for the fourth quantile (N = 1,926). These cut offs are reflected in the heat map in Fig 1 and the distributions are shown in Fig 2. Analysis The pathways through which biodiversity in urban forests contribute to human health appear numerous, nuanced, and likely intertwined in a complex web of other health determinants. Due to the complexity of the connections, causal chains are difficult to establish even for exper- imental studies like those of Li [32] and Thompson et al. [49]. Our approach is to assess macro-level (rather than micro) associations between urban biodiversity and human health outcomes. This approach carries risks: instead of identifying effects from the biodiversity of urban greenspace, there may be confounding unobserved factors correlated with biodiversity of urban greenspace or aggregation bias may lead to spurious correlations (for a humorous dis- cussion of the risk of ecological fallacy within the context of biodiversity and health, see Salkeld and Antonlin [50]. To address this risk, we use a fixed effects panel OLS regression as our main specification, include several sensitivity checks, and make explicit that our analysis is descriptive and not causal. In our most parsimonious specification, we include fixed effects at the three-digit zip

code level which controls for stable unobserved characteristics within the area of a 3-digit zip code level, ϕz. This means that in our main specification, differences in mortality rates are identified off of variation in tree diversity and abundance within a three-digit zip code. yit ¼ a þ b1Shannoni þ b2Treesi þ GXi þ �z þ �t þ �it ð2Þ We also include fixed effects by year, ϕt, and, in sensitivity checks, sociodemographic con- trols Xit. Standard errors are clustered by three-digit zip code to account for unobserved auto- correlation. However, spatial autocorrelation may be more complex and not correspond to three-digit code boundaries. Spatial noise could cause high t-statistics for the Shannon Index and trees per 100,000 individuals measures. To address this, we characterize the importance of PLOS ONE Biodiversity and cardiovascular disease PLOS ONE | https://doi.org/10.1371/journal.pone.0254973 November 3, 2021 7 / 23