Freshwater fisheries conservation can increase biodiversity

29 species. This effect is different from zero (with a p-value<0.05) and robust to the choice of matching algorithm. Sensitivity to unobserved confounders These estimates can be interpreted as causal when the CIA holds. There is no way to directly test this assumption, and instead we present an analysis of the sensitivity of the estimates to the importance of potential unmeasured confounders trough the estimation of Rosenbaum bounds, implemented using the command –rbounds– described in [37]. The results are pre- sented in Table 3. Because we mostly worry about positive selection bias (ie, the possibility that the program may have been allocated to areas where it is mostly expected to be beneficial), we mostly care about the possibility that we are over-estimating the ATT. For that reason, we focus on the lower values of statistical significance of our estimates (sig+) and the lower bound of ATT (t- hat+) under varying importance of potential confounders, as measured by eγ. At the usual cut- off of eγ = 2 (to be interpreted as “unmeasured confounders importance in explaining treat- ment status is the double of the importance of measured confounders”), our estimates are still statistically different from 0 at the usual levels of significance (p-value = 0.034) and the estimate of ATT is still important, with implementation of ComFishIII leading to an average increase in species richness of 19 species. Table 2. Impact of ComFishIII on number of species caught. Matching NT NC ATT s.e. t 95% CI Nearest neighbor 16 13 29.000 9.220 3.145 9.92 − 46

Kernel 16 16 32.724 8.526 3.838 13.05 − 47.48 Radius (r = 0.05) 16 16 31.058 9.014 3.446 11.62 − 47.60 NT: number of treated (ComFishIII) villages. NC: number of control villages. ATT: Average Treatment on the Treated. Bootstrapped standard errors with 1299 repetitions. The Confidence Intervals presented are the 95% Bias-Corrected CI, presenting more conservative estimates of the effect of the program. https://doi.org/10.1371/journal.pone.0233775.t002 Table 3. Impact of ComFishIII on biodiversity: Rosenbaum bounds. eγ sig+ sig- t-hat+ t-hat- 1 .002 .002 29 29 1.2 .005 .001 26.5 32 1.4 .009 .000 23.5 35 1.6 .016 .000 22.5 36.5 1.8 .024 .000 21 37.5 2 .034 .000 19 38 eγ: log odds of differential assignment due to unobserved factors sig+: upper bound significance level sig-: lower bound significance level t-hat+: upper bound Hodges-Lehmann estimate t-hat–: lower bound Hodges-Lehmann estimate https://doi.org/10.1371/journal.pone.0233775.t003 PLOS ONE FCZ and biodiversity PLOS ONE | https://doi.org/10.1371/journal.pone.0233775 May 27, 2020 9 / 14