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

assessing at the species level; ii) source of identifying variation, by estimating without fixed effects at the three-digit zip code level; iii) selection in arborist sampling, by restricting to zip codes with at least 1,000 trees; iv) omitted variables bias, by including socioeconomic and pol- lution burden controls and v) aggregation bias, by estimating separate subsamples by two-digit zip code. Finally, we also estimate our main specification using weighted least squares, weighting by zip code population. Weighting may be justified if sampling is endogenous or the degree of intracluster correlation is low and the variance in the number of observations per cluster (zip code) is high [51]. These conditions are unlikely in our case. We found a high degree of intracluster correlation by Shannon Index, with a Moran statistic of 0.153 (p-value: 0.00) and sampling of tree data by the consortium of private arborists is unlikely to be correlated with heart disease or stroke mortality. However, comparing weighted least squares and ordinary least squares can be a useful diagnostic for model misspecification [52,53]. Results Our results begin with a graphical assessment of how differences in urban forest biodiversity correlate with heart disease and stroke mortality rates. We follow with analytical estimates of linear and non-linear specifications using ordinary least squares. After presenting our main results, we compare the linear estimate across several different specifications to assess the sen- sitivity of our primary estimates.

What is the shape of the raw relationship between urban forest diversity and mortality for heart disease and stroke? Fig 3 shows binned scatter plots of mortality rate per 100,000 people against the zip code’s Shannon Index. As expected from the shape of the histograms of the Shannon Index in Fig 2, the binned scatter points in Fig 3 are concentrated among higher Shannon indices. There are fewer obser- vations with very low Shannon Index values. For both Heart Disease and Stroke, we see a nega- tive linear relationship between the Shannon Index and the mortality rate per hundred thousand people. Next, we more formally estimate the relationship between the Shannon Index and mortality rate with a rich set fixed effects in a panel regression estimated using OLS. Unlike in our scatter plots, our regression controls for shared unobserved characteristics at the three-digit zip code level as well as the overall tree abundance per hundred thousand people within the zip code. Our results, reported in columns one and four of Table 2, have a similar pattern to Fig 3: the coefficient on the Shannon index is negative and precisely estimated for both heart disease and stroke mortality rates. The coefficients are equal to a decrease in the mortality rate per hundred thousand people of -33.5 (se = 11.0) for heart disease and -12.4 (se = 4.95) for stroke for an increase in the Shannon index of one. These coefficients are about half those found in the raw binned scatterplots in Fig 2. The estimate is more precise for Heart Disease, which has fewer

censored zip codes, but the fit is better for Stroke. These estimates imply that increasing the Shannon Index by one standard deviation (0.64) is associated with a decrease in the mortality rate of 21.4 per hundred thousand for heart dis- ease or 13% and 7.7 per hundred thousand for stroke or 16%. The estimate of the coefficient on trees per hundred thousand is unexpectedly positive, however it is imprecisely measured. Changes in an area’s Shannon index in response to tree planting or death will vary based on the underlying composition of the urban forest, thus making a direct and generalizable inter- pretation of the Shannon Index’s coefficient difficult. To aid in interpretation, in the discus- sion section we use our coefficient to estimate potential benefits of a proposed tree planting project in Los Angeles. PLOS ONE Biodiversity and cardiovascular disease PLOS ONE | https://doi.org/10.1371/journal.pone.0254973 November 3, 2021 9 / 23