Freshwater fisheries conservation can increase biodiversity

Constructing the control group ComFish was not allocated randomly. We construct the control group by estimating the prob- ability of participating in ComFish, given what we know about how the program was allocated, as described above. In the absence of a baseline survey, we use the Agricultural Census 2011/ 12 as the source of information for the estimation of the propensity score and the construction of the control group. Importantly, the Census contains information about a wide variety of vil- lage characteristics that might have conditioned the decision to promote ComFishIII, includ- ing the importance of fishing for local livelihoods. The process of constructing the control group followed four steps. Step 1: Identification of participating villages. We focused on four provinces in Central and Southern Laos (Bolikhamxay, Khammouanne, Savannakhet and Champasak). Because we relied on the Agricultural Census 2011/12 as a source of pre-intervention information, we excluded those villages where implementation of the program started in 2011/12. In addition, several ComFishIII villages had to be dropped from the analysis, as they could not be identified in the Agricultural Census. Step 2: Identification of potential control villages. The identification of a set of villages that are statistically identical to those that were included in the ComFishIII was based on the estimation of the propensity score. To satisfy the assumptions underlying this approach (see [26, 28]), in particular the CIA (Eq 4), we relied on the information collected through the Agricultural Cen-

sus 2011/12. We had access to information on: 1) Importance of fishing and other livelihood activities; 2) Access to roads (and, with them, outside markets) and public services; 3) Impor- tance of ethnic heterogeneity, as measured by the number of ethnic groups with more than 5% and 10% of the village population, as well as the fraction of the population that belongs to the first and second most important ethnic groups in the village; 4) Village history, notably their experience with resettlement; and 5) Location of administrative headquarters (District and/or Province). In the absence of detailed geographical information regarding access to freshwater, we limited the estimation of the propensity score to those villages that had a percentage of households who fish that was greater or equal to the minimum of this variable in ComFishIII villages (25%), which we took as a proxy for this determinant of program placement. The estimates of the probability of participating in ComFish (ie, the propensity score), pre- sented in S1 Appendix of Table A1, were obtained using the command –pscore– for Stata, described in [31]. The common support option was imposed. Once balance on covariates was achieved, and based on these estimates, we chose as potential control for each ComFishIII vil- lage the three non-participating villages with the closest value of the propensity score (ie, the three nearest neighbors). Step 3: Local validation of the list of matches. The control group built in the previous step

would be valid under the hypothesis that the Agricultural Census contains all information that might impact on this decision. Although the Agricultural Census contains a rich variety of information, this seems unlikely: for example, the Census does not contain information about the nearby existence of deep pools, on which ComFishIII conservation activities are based. To overcome this limitation, the list of potential matches was presented and discussed with Pro- vincial Agricultural and Forestry Offices (PAFO), with the objective of collecting additional information (existence of deep pool, presence of other fishery development projects) that might impact on its inclusion as a possible control village. As a result of these discussions, we dropped from the control group those villages that had no deep pool close by or where other fisheries development projects were present. An important concern when evaluating conservation programs is the potential importance of spillover effects [32]. This is particularly important when their success hinges on the PLOS ONE FCZ and biodiversity PLOS ONE | https://doi.org/10.1371/journal.pone.0233775 May 27, 2020 5 / 14