Cost-effective climate mitigation via conservation incentives targeting poverty: Bolsa Verde's impact in Brazilian Amazonia settlements

doi:10.21203/rs.3.rs-2122112/v1

Download PDF

Article

Cost-effective climate mitigation via conservation incentives targeting poverty: Bolsa Verde's impact in Brazilian Amazonia settlements

https://doi.org/10.21203/rs.3.rs-2122112/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Conservation incentives have been implemented globally to support forests and the ecosystem services they provide (e.g., greenhouse-gas sequestration). Paying forest managers can increase local acceptance of conservation, while also targeting local poverty. Evidence shows that the impacts of incentives vary across contexts and scheme designs, from zero to high. We assess whether deforestation is avoided in sites where households enrolled in Bolsa Verde, a Brazilian governmental conservation program implemented between 2011 and 2018. For enrolled settlements in Amazonia, we estimate that Bolsa Verde reduced deforestation by a total of ~80’000 ha, implying 35 megatons fewer CO₂ emissions over the program’s duration and with no evidence of a rebound in forest loss after the program. We find no evidence of avoided deforestation in other program sites – either settlements outside Amazonia or protected areas inside or outside Amazonia. Still, the monetized value of the climate-mitigation benefits for settlements in Amazonia only may well exceed the program’s cost across all its regions.

climate mitigation

benefit-cost analysis

deforestation

impact evaluation

conservation incentives

Incentives for conservation (often called ‘Payments for Ecosystem Services’ or ‘PES’) are monetary or in-kind transfers to land managers in exchange for agreed land management ^1–3. PES have been increasingly implemented both in developing and developed countries ^4,5, often complementing land-use and rural-development policies. Some PES target people in ‘multiple-use’ protected areas (e.g., in IUCN Category V-VI) and/or indigenous or local communities, to incentivize conservation of forests while alleviating poverty ^6–9. PES are also one key pillar of many REDD+ pilot initiatives ¹⁰.

If designed for local acceptance and impact (targeting avoided-deforestation potential, aligning with local opportunity costs while respecting social norms, and enforcing conditionality), PES can be a cost-effective tool to reduce forest loss, and thus carbon emissions, while protecting biodiversity and water quality ^2,3,11. Emerging rigorous evidence concerning their impacts – using losses in similar locations to estimate what would have occurred in PES areas without PES − reveals that that impacts of PES vary from zero to high, also reflecting heterogeneity across contexts and designs ^3,12–15. On average, there are modest gains in tree cover from modest (conditional) transfers to poor rural households ^7,13,16,17. The rigorous selection of similar control sites, in order to generate better “apples-to-apples” comparisons, has been facilitated by the improved availability of causal inference software and remotely-sensed data ^18–23.

Worldwide, many PES schemes have been pilot projects, or programs that experienced budget reductions or were defunded entirely ^24–29. Yet some evidence suggests potential for lasting impact if programs not only restricted land use but also built capacity to manage natural resources or boosted pro-conservation social norms ^25,28,30,31. Such durable impacts can be hard to detect when a program ends while deforestation pressures are rising, such that post-program forest loss is higher than during the program. Yet they exist if loss would have been even higher in the absence of the program– another ‘counterfactual’ to be estimated using appropriate controls. For example, if the path of post-program losses only slowly catches up over time with the loss that would have occurred without PES – as opposed to being completely and immediately undone due to a large post-PES “rebound effect”-, associated CO₂emissions can be actually “delayed” (as suggested in Figure 1, which is adapted from one of ²⁴’s post-PES scenarios). These useful delays can still provide important climate-mitigation benefits, as compared to a situation without PES²⁴.

Examining both during and after the program, then, we evaluate the impacts of the Bolsa Verde (“Green Grant”, hereafter “BV”) program on tree cover and avoided CO₂ emissions, in Brazil. BV was a federal cash-transfer program targeting households in extreme poverty who were living in various types of eligible conservation sites across different biomes. It was implemented during 2011-2018 and offers an excellent case to test whether conservation incentives targeting poverty can still avoid forest loss − even within sites already subject to environmental regulation. For targeted rural settlements in the Amazon biome, we estimate that Bolsa Verde reduced tree cover losses by ~ 80,000 hectares, implying ~ 35 megatons fewer CO₂ emissions at a program cost of about USD 1.73 per ton of avoided CO₂. We also find no evidence of a ‘rebound’ in forest loss after BV, i.e., some deforestation and CO₂ emissions may have been permanently and cost-effectively delayed. We see no evidence that BV avoided deforestation in protected areas (PAs) in Amazonia, which on average already had low deforestation rate. We also find no evidence of avoided deforestation in rural settlements or protected areas for non-Amazon biomes (i.e., Caatinga, Cerrado, Mata Atlântica) which offered less potential to reduce deforestation via modest incentives.

Amazonia Settlements: avoided deforestation during BV, without a post-program ‘loss rebound’

We start by plotting average annual tree-cover loss as a percent of 2000 tree cover, in each site, for the 299 enrolled sites in Amazonia Settlements and matched unenrolled control Settlements. These sites were particularly targeted, as reflected with the largest number of enrolled households (Figure S2), consistent with the social and environmental eligibility criteria of the program (See Methods). Average annual tree-cover loss (in % of forest cover area in 2000) in BV sites was between 0.25% and 0.5% for 2011-2018, though higher before 2007 and in 2020 (Figure 2, top-left panel). Their matched, unenrolled controls often had higher average annual tree cover loss, particularly after 2013 (while enrolled and unenrolled sample had on average a similar forest cover loss in percentage during 2006-2009).

Our econometric analysis confirms a statistically significant negative Average Treatment Effect on annual forest cover loss for each year between 2015 and 2017 (i.e., less loss in enrolled as compared to similar unenrolled sample), while the annual effect is not statistically significant between 2011 and 2014, and even positive but non-significant in 2012 (Figure 2, bottom-left panel). The sum of the annual effects confirms that the Cumulative Average Treatment Effect (in % of forest cover in 2000) is statistically significant. We can then estimate the overall BV impact on the average annual forest cover change rate over the period 2011-2017. The observed average annual forest cover change rate 2011-2017 in enrolled sites was r_a= -0.40 (% per year), while the estimated counterfactual average annual forest cover change rate was r_c= -0.55, implying a standardized forest cover effect size r_a-r_c= 0.15, SE= 0.07, 95% CI [0.02, 0.28]) (Table 1, Enrolled Settlements in Amazonia and left second row of Figure S5), i.e., BV has contributed to decrease the average annual deforestation rate by 0.15 percentage point in enrolled settlements of the Amazonia between 2011 and 2017. This forest cover effect size is of similar order of magnitudes than the effect sizes found in similar contexts¹². Without BV, the total forest cover loss in 2011-2017 would have been about 37% (95% CI [10%, 63%]) higher in the enrolled sites. We estimate that BV avoided a total of ~ 79,897 ha (95% CI [12,7799 ha, 147,015 ha]) of forest loss in enrolled settlements of Amazonia, an area about half the size of the city of São Paulo (but it represents only about 1.86 % of the total forest cover lost in Brazilian Amazonia between 2011 and 2017).

Overlapping carbon biomass density maps ³² with the extend of forest cover at baseline for enrolled sites, we find that those sites had on average 120 (SE= 2) tons of aboveground biomass carbon per hectare in 2010. We use the standard conversion factor of 3.67 to calculate a potential of 440 [SE= 8] tons of sequestered CO₂ per hectare aboveground. Combining our estimated effect size on avoided tree cover loss with this CO₂ sequestration potential, we find that BV avoided 79,897* 440 ~ 35.13 megatons (propagated SE combining the SE of the effect size and the SE of CO₂ sequestration potential = 15.68, 95% CI [4.39, 65.86]) of CO₂ emissions within those enrolled Amazonia Settlements, between 2011 and 2018. This cumulative 7-year avoided emissions amount is roughly equivalent to the annual carbon footprint of the city of São Paulo during the single year 2015 ³³.

Post-intervention, we observe that tree cover loss (in % of tree cover at baseline, Figure 2) increased between 2018 and 2020. Yet, tree-cover loss was even higher in the matched, unenrolled controls during the same period (Figure 2). The econometric analysis further confirms that there is a statistically significant difference between enrolled and unenrolled sample on years 2018 and 2020, but not in 2019 (Figure 2, bottom-left panel). Thus, deforestation avoided during BV implementation appears to have been delayed (and eventually permanently delayed but confirming it would require more post-intervention time periods).¹

Non-Amazonia Settlements & All PAs: no evidence of avoided deforestation in different contexts

In the 51 enrolled PAs in Amazonia, average annual tree-cover loss was systematically well under 0.5% (of 2000 tree cover) during both 2001-2010 and 2011-2017 (Figure 2, third panel at the top). Tree-cover loss was slightly higher in enrolled sites than in matched, unenrolled controls in 2015-2017 although the gap is small for the entire enrollment period. While our sample size may not be statistically powered enough to precisely detect an eventual small effect, our finding is consistent with the fact that low pressures in Amazonia PAs limit the potential for BV to additionally avoid deforestation in these sites (Fig. S4, S5, S12 and Table S2). We note however that the difference in forest cover loss between enrolled and non-enrolled sites is statistically significant in 2018 and 2019 (Figure 2, third panel at the bottom), although we lack field evidence to causally attribute this post-BV difference to a legacy effect of BV instead of other causes not related to BV.²

In the 9 enrolled PAs outside Amazonia, average annual tree-cover loss was between 0.45% and 0.75% during 2011-2017. While that is below the preceding years (Figure 2, fourth panel at the top), it is higher for enrolled sites than observed for matched unenrolled controls during the entire enrollment period (although the matching in this case is insufficient for rigorous “apple-to-apple” comparisons, see also Table S4). The econometric analysis also shows a lack of statistically significant impact (Figure 2, fourth panel at the bottom, see also Figs. S4, S5, S13, and Table S3). Thus, we do not find that BV avoided deforestation in PAs outside Amazonia.

From similar analyses for the 544 enrolled Settlements outside Amazonia (Figure 2, second panels top and bottom), again we cannot reject the null hypothesis of zero BV impact on deforestation (see also Figs S4, S5, S13 and Table S4). Our findings suggest that BV in settlements outside Amazonia may not have offered incentives high enough to induce a detectable impact in these sites, noting that there was proportionally little deforestation area to avoid in absolute terms as compared to settlements in Amazonia.

Benefit-Cost Analysis: climate-mitigation benefits versus program costs

Amazonia Settlements Alone

We now want to compare the estimated costs of the program and its monetized climate mitigation benefits, both expressed in USD per ton of avoided CO_2.As total BV expenditures were ~ USD 135 million (in constant USD 2016) (Table S5), for Amazonia Settlements we assume 45% of this cost, given the share of total enrolled households. We conservatively assume that the program did not induce a positive spillover after its end, i.e., we do not consider eventual positive post-program impacts on forest cover, so our analysis only takes into account the avoided CO2 emissions that occurred during the program implementation, and where supposedly not immediately voided by a post-program rebound effect. This yields an estimated average cost per ton of avoided aboveground CO₂of ~ USD 1.73 (95% CI [0.22, 3.24]), about two times higher than the cost per ton of CO₂ for a REDD+ pilot project for 350 households in the State of Amazonas ³⁴, suggesting either decreasing returns to scale in such incentive programs, but also potentially differences in implementation context. This cost is almost 4 times the cost found for a pilot PES project in Uganda ¹⁶, albeit a project which was implemented at a far smaller scale and within a country where the cost of living is lower than in Brazil.

Using an estimate of the Social Cost of Carbon (monetary economic damages from emitting one additional ton of CO₂ into the atmosphere, ^35,36), we can approximately monetize the climate-mitigation benefits from BV delaying CO₂ emissions resulting from avoided aboveground carbon biomass loss during program implementation (see Methods). For calculus simplification purposes, we first need to make assumptions to determine the year since when deforestation has started being delayed. As BV’s impact started to be statistically detectable in 2015 in this region (Fig. 2), with cumulative effect still growing in 2017, we proceed as if the avoided deforestation occurred uniformly on average on 2016, which also is the enrollment year for most enrolled households.³ Given no evidence that formerly enrolled settlements “caught up” with foregone deforestation during 2018-2020, we then make the assumption that deforestation was delayed at least 4 years (and likely longer given no evidence of post-intervention “rebound”). Using these assumptions, we can finally roughly compare the estimated costs per avoided ton of CO₂with the estimated monetized benefits of the program, corresponding to the value of different years of avoided deforestation.

Figure 3 shows benefits per ton of avoided CO₂, as a function of years of delay in deforestation, juxtaposed with program costs(i.e., governmental expenditures, including not only operating costs but also incentive transferred to households) in USD per ton of avoided CO₂(red bar). The cost bar, (expressed in constant USD 2016) corresponds to the costs that have occurred between 2011 and 2018, and thus does not increase, i.e., we proceed as if all the costs of the program have been spent in 2016. If deforestation were delayed only 4 years (i.e., if the foregone deforestation were reversed immediately after 2020), the program benefits (USD 1.26, 95% CI [0.0.92 USD, 1.61 USD]) would be below program costs. Using our central estimate, once deforestation has been delayed for at least 6 years, benefits are above costs − and of course the benefit-cost ratio keeps rising with greater years of delayed deforestation. For instance, 10 years of delay - i.e., if it will take 10 years to catch foregone deforestation in formerly enrolled sites - yields a benefit-to-cost ratio of 1.78.

Comparing Amazon Settlement Benefits with Total Program Cost

As just under half of all enrolled households were in Amazonia Settlements, while we detected no impacts on deforestation elsewhere -likely an underestimation of the true impact of the program-, here we retain the benefit per ton of avoided CO₂ from above – again essentially a function of years of delayed deforestation − but compare that with total program costs (versus only 45% of costs for Amazonia Settlements above). By construction, the average cost per avoided CO₂ more than doubles (USD 3.84 (95% CI [0.48, 7.20]).⁴ In this case, using our central estimate, the delay in deforestation must be at least 13 years for BV’s benefits to outweigh its costs, a scenario that is not implausible given the absence of immediate post-PES rebound effect.

¹ A back-of-the envelope calculation indicates that it would take an equivalent of 7 years of post-BV deforestation higher by 0.15 percentage point in formerly enrolled sites (as compared to never enrolled sites) to wipe out avoided deforestation accumulated during BV implementation (2011-2017).

² Note that if we only had taken into account the ATT without statistical significance, we will conclude that BV in Amazonian PAs has contributed to avoid about 8,459 ha of forest. Without BV, the total tree-cover loss in 2011-2017 would have been about 7% higher in these sites, which would correspond to about 3.7 megaton of CO₂ emissions.

³ Choosing 2014 (mid-program year) as the reference year for our calculation only marginally affect our estimation of benefits, i.e., by construction it approximately corresponds to moving the benefits curve two years earlier without affecting the slope of this curve.

⁴ If we also include the estimated 3.7 megatons of CO₂avoided in protected areas in Amazonia, the average estimated cost of the program is of USD 3.48 [-0.80, 7.75] per ton of avoided CO₂.

Our findings show that while impacts from incentives for forest conservation depend upon the context, incentives can be effective in avoiding deforestation in priority regions. We extend prior studies focused on early BV cohorts^37,38, notably in Amazonia³⁹ through further analysis of PAs and settlements outside Amazonia, finding no evidence of impact for those sites. Where we do find impacts − as expected, more likely in frontier settlements containing extensive forest area facing a moderate to high risk of deforestation − we add a critical element, an analysis of post-intervention tree cover loss. While post-intervention loss increased, we find no evidence, for the years we can study (up to 2020), that formerly enrolled BV households “catch up” with foregone deforestation. Formerly enrolled sites are still losing forests, but at least a proportion of deforestation might have been “permanently” delayed ^24,31, which still implies important climate mitigation benefits.

Noting that loss rates are on average quite stable before and during BV in enrolled settlements in Amazonia, it seems impact in this priority region is not primarily due to reducing absolute loss, but instead may result from avoiding the rise in losses experienced in similar never-enrolled comparison sites. In keeping with our emphasis on the importance of contexts that vary over both space and time, it is also important to note that tree-cover loss was rising in Amazonia since 2011 due to weakened governmental environmental regulations ^40–42.

Surprisingly, loss rates, in relative terms, were generally higher outside of Amazonia (the sites are smaller and closer to roads and have higher agricultural potential). There, however, the incentives within the BV program presumably were too low for impact, and the enrolled forest areas outside Amazonia remained low anyway. On the contrary, PAs in Amazonia faced low deforestation risk, even without the BV program. PAs tended to be in areas less likely to be converted to agriculture, with lower population density − limiting the possibilities, on average, to further reduce deforestation through modest incentives. While there is still a substantial volume of forest loss in Amazonian PAs, this represents a tiny fraction of their forest cover. Our varied findings confirm the need to pay attention to the heterogeneous impacts of incentives programs, not only due to differences in site location but also because sites may be affected differentially by policy and economic changes ^8,34,43 (see also ^44,45 for systematic reviews of how contexts help explain impacts of various types of forest conservation).

We might expect that incentives targeting poverty will sometimes but not always overlap sites facing deforestation risk, in which they might actually avoid deforestation ^46–48. For BV, which most targeted settlements in Amazonia (with almost half of enrolled households), fortuitously sites containing many poor households often faced a moderate to high risk of being deforested – i.e., neither too low nor too high for impact. Whether BV could have raised its forest impact and cost-effectiveness at scale by prioritizing households facing higher risks of loss across these distinct regions ⁴⁹ is an open question. Yet prioritizing only sites containing important forest areas facing risk of deforestation, i.e., enrolling more households from Amazonia as compared to other biomes, might well have diminished the initial, national-level political support for this type of publicly funded program.

The program was nevertheless gradually defunded after 2015 and terminated at the beginning of 2018, in a context of diminishing political support for Amazonian conservation. Our findings suggest that the program may still have induced some lasting impacts beyond the duration of the program in Amazonian settlements. While increased forest cover loss may be expected after the end of the program (as compared to the loss during the program), especially in a context of rising pressures in deforestation, we have shown that the increase is not necessarily higher than what has happened in control sites (i.e., no evidence of a “rebound effect” compared to the counterfactual scenario). Future research is needed to better quantify and understanding the drivers and the duration of post-program impacts^{24–26,29,31}.

In conclusion, conservation incentives clearly have a cost-effective and benefit-to-cost potential to contribute to stabilizing deforestation in frontier regions, notably via increasing compliance with site natural resource management regulations ³, even programs that are short-lived and have dual conservation and poverty alleviation objectives. Yet incentives are not a “one-size-fits-all silver bullet” that will straightforwardly stop deforestation in all contexts, notably when they are implemented at large scale¹³, or when there are reversals in the attitude of government regarding forest conservation⁴¹. Careful diagnosis of the ability of conservation incentives to address the key drivers of deforestation and CO₂ emissions, within different contexts and across long time periods ^49,50, could increase the performance, and the lasting impacts, of such interventions − alongside other regulatory, capacity-building, and supply-chain efforts to reduce critical losses of forest ^51,52.

Study site and context

The BV program was created in 2011. Its objectives were to increase and diversify the incomes of poor households in eligible conservation sites, while incentivizing conservation and sustainable management of ecosystems ^53–56. Eligible sites consisted of i) inhabited federal protected allowing the sustainable use of natural resources, ii) settlements allocated by agrarian reforms to traditional populations involved in agro-extractive activities and sustainable forest management, and iii) recognized traditional communities living on Brazilian public lands (Fig. S1).⁵ The three different tenure systems are respectively regulated by distinct Brazilian federal agencies, and property rights are subjected to legal restrictions and an approved management plan on the use of natural resources.⁶ Within eligible sites, eligible households are those with livelihoods reliant on natural resources, and a monthly income per capita under 85 Brazilian reais (~ 24 USD in 2016). Households interested in enrolling on the program also had to be included in the governmental list of households eligible to receive governmental social subsidies (Bolsa Familia program). At the beginning of the program (October 2011), BV initially enrolled households from eligible sites in the Legal Amazon, corresponding to Amazonia and some parts of the Cerrado biome. BV then expanded, in 2012, to other biomes in Brazil (Caatinga, Cerrado, Marinho Costeiro, and the Mata Atlântica) while continuing to impose similar social criteria in terms of eligibility.

Between 2011 and 2018, almost 80,000 households enrolled, though some for less than the full duration of the program (Fig. S2). They had to comply with a list of activities that foster sustainable management and conservation, in exchange for quarterly payments of 300 Brazilian reais (about USD 86 in 2016). Complementary funding supported training in diversifying sources of income and complying with local environmental regulations, which was monitored by officers ^37–39. The number of enrolled households started to fall after 2016, notably following the non-renewal of many contracts because the implemented agency argued that many households were no longer meeting the extreme poverty criteria. BV ended in early 2018. Many sites experienced some deforestation during the implementation of the program (Fig. S3) and particularly after 2015, given a gradual weakening of environmental law enforcement in all Brazil ^40–42.

Reports based on data from BV sites note positive socioeconomic impacts (via cash transfers, training, income diversification) as well as environmental impacts (via contracts’ incentives, clarification of rules governing the management of natural resources, and better reporting of illegal activities) ^37,38. One rigorous impact evaluation concluded that BV reduced forest loss by about half of the counterfactual for Settlements and PAs in Amazonia during 2011 and 2015 ³⁹, while causally attributing most of the impact not to the enrolled households’ behaviors but instead better monitoring of and increased fines for illegal deforestation by outsiders. An evaluation using a socio-economic survey in 2015, for all biomes, also concluded BV did not impact enrolled households’ logging practices but, rather, raised their familiarity with environmental regulations and, for supporting local livelihoods, facilitated their acquisition of new productive equipment ³⁷. Our analysis complements these studies by considering BV’s entire duration and distinguishing its impact by biome, while also separating the sample of settlements from protected areas. We extend all prior studies by considering the possibility of post-program impacts on forests.

Evaluating BV Impacts

We evaluate impact at the site level, given BV’s individual and site-level elements (e.g., both payment transfers and training to increase the compliance with site management plans). Examining the polygons for inhabited federal protected, and those for settlements allocated to traditional populations, then comparing them with respective control sites, is consistent with BV having impact via these conjectured causal mechanisms: a) reduction of deforestation in enrolled households’ properties resulting from a compensation for opportunity costs; b) diversification of income that reduces extensive land-use change; c) better awareness and compliance with local site management plans; and d) better reporting and sanctioning of illegal deforestation activities that are carried out by trespassers. As BV did not include all the households in each site, impact may involve raising compliance with site regulations regarding resources management − especially through increased reporting of illegal deforestation by non-BV actors.

Our key assumption is that, without BV, BV’s enrolled sites would have had on average the same loss as “the most similar” unenrolled sites. The validity of this assumption depends upon how similar the controls are, i.e., whether they represent the counterfactual situation of what would have happened in the absence of BV. Unenrolled sites are selected from a sample of sites of the same tenure system and located in the same biome that each sample of enrolled sited. We further select sites with the most similar values of a summary statistic for relevant characteristics, the estimated probability (or ‘propensity’) of being enrolled in BV, knowing that the program targeted poor households whose livelihoods depend on natural resources. Propensity scores are one way to process the sample selection for “apples-to-apples” comparisons of enrolled with unenrolled sites, such that post-processing differences in group outcomes are reasonable estimates of BV impact.

BV has been targeting distinct tenure systems and has been implemented in several biomes of Brazil (see S4 for covariate distribution across tenure systems and biomes). Rigorous studies of the impacts of conservation interventions have shown that the impacts of a single intervention are heterogeneous across regions, notably because the magnitude of deforestation threats is not the same across regions and sites^9,44,58. Measuring the average impact of the whole program, as often measured in impact evaluation of small-scale conservation programs, would not be informative in our case given this regional heterogeneity.⁷ Thus, in our analysis we distinguish Amazonian sites from sites in other biomes (Caatinga, Cerrado, and Mata Atlântica) not only because of the distinctive drivers of deforestation and regulations in the former but also due to the largest number of enrolled households and area included in the program, as well as earlier time of initial enrolment in the program. We also distinguish the impact of the program on protected areas from the impacts of the program on settlements, as these two distinct tenure systems are likely to differ in terms of magnitude of the drivers of deforestation and regulation (Figs. S3 and S4). We therefore evaluate BV’s impacts by biome and tenure system: Amazonia PAs; non-Amazonian PAs, Amazonia Settlements; and non-Amazonian Settlements.

In each region, we evaluate the cumulative BV impact (i.e., the sum of annually avoided tree-cover loss during program implementation 2011-2017) by computing a generalized Difference-in-Differences, i.e., comparing changes over time in tree-cover loss (expressed in percentage of site forest cover in 2000) of in enrolled sites with those in similar unenrolled sites. We take into account that enrolled sites started in different years and that BV’s impacts could vary across years, e.g., due to shifts in local deforestation pressures. We also evaluate the post-program (2018-2020) differences in annual forest cover loss.

Unit of analysis

BV targeted households below the poverty line defined by the program and located in eligible sites. Individual recipients’ plots were not publicly available, but the Ministry of Environment’s publicly available database provided the name of the sites containing enrolled households. We identify 70 PAs and 902 settlements containing beneficiary households. We match this database with the shapefile of all federal PA recognized by the Instituto Chico Mendes de Conservação da Biodiversidade, ICMBio, which is the Brazilian Ministry of the Environment’s administrative agency (version of September 2018) as well as the shapefile of all the settlements created by the Instituto Nacional de Colonização e Reforma Agrária, INCRA, which is the Brazilian agency administering land reforms (version of September 2018). We remove all sites that were only legally recognized by the government after 2011 as well as State PA and federal "strict" PA because these two types were not eligible to the program. We also filter data on Settlements to only keep the management categories eligible to the program (“environmental distinctive settlements). Our working sample contains 113 PAs and 6048 settlements, respectively 60 enrolled PA and 53 non-enrolled PA as well as 843 enrolled settlements and 5205 non-enrolled settlements. In Amazonia, our working sample of enrolled sites includes 51 PA and 299 settlements.

We use a sample of never-enrolled sites that correspond to sites belonging to same types of sustainable use federal PAs and environmentally distinctive settlements eligible to the program, but where no households have been enrolled in the program. Explanation about why no household has been enrolled in the program in these sites includes the fact that no household met the poverty eligibility criteria, that the sites were not properly registered in the social program database, or that households were not interested in joining the program. Never-enrolled sites faced in principle the same obligation to comply with sites’ management plans than enrolled sites, even if compliance and/or enforcement may vary across time and between sites, independently of the presence or absence of BV.

Our analysis at site level allows us to consider eventual internal leakages, i.e., displacement of deforestation from one plot enrolled in the program to another plot within the same site. However, our unit of analysis may in some cases underestimate impact if only a small proportion of each site has been affected by the program.

Treatment status

The pre-matched sample of non-enrolled sites corresponds to sites where no households have been enrolled in the program but belonging to categories eligible to the program. Within each enrolled site, we compute the number of enrolled households each year. We do not distinguish the month of enrollment within each year but we “group” sites according to the year of first enrollment in the program. We also do not distinguish impacts as a function of the evolving number of enrolled households within each site.

Outcome variables

We used MapBiomas version 6.0 as a data source for computing tree cover loss. MapBiomas provides high resolution (30*30m pixels) annual raster built from Landsat satellite images. MapBiomas covers the six biomes of continental Brazil (Amazônia, Cerrado, Caatinga, Mata Atlântica, Pantanal and Pampa). We extracted pixels corresponding to natural forests (Forest, Savanna, Mangrove, and “restinga” coastal forests) and excluded forest plantations and other non-forest land covers. We excluded reforested pixels (i.e., any pixel that is classified as natural forest in a year while it was not classified as natural forest the previous years) to focus only on tree cover loss. We removed site that contained less than 10 ha of forests in 2010. We only kept annual raster corresponding to the years between 2000 and 2020 and calculate the outcome variable as the annual tree cover loss divided by the forest cover in 2000 in each site expressed in percentage for all the (enrolled and not enrolled) PA and settlements in our database. The program legally ended at the beginning of 2018, when the last enrolled households received their last quarterly payments. Therefore, our main econometric specifications are performed using the year 2017 as endline, while the years 2018 to 2020 corresponds to post-intervention years.

Confounding variables

We calculate the travel time to nearest city of 50 000 or more people in 2000 (Joint Research Centre of the European Commission‘s global map of Accessibility). We also measure the linear distance between each conservation site‘s centroid and the nearest road (using a database of road infrastructure in 2005, produced by the Brazilian Institute for Geography and Statistics (IBGE), as well as to the nearest river World Rivers from Harvard’s WorldMap. We add a discrete map of topographical regions with 5 classes: flat (less than 3% slope), undulated (3 to 20 % slope), very undulated (20 to 45%), mountainous (45 to 75%), and rugged (more than 75%) derived from IBGE’s Digital Elevation Model. We calculate the total forested area in 2010 in hectare for each site. To reflect the temporal variations of deforestation, we also calculate pre-program tree cover loss split into several periods in our base models: 2000-2004, 2004-2007, and 2007-2010, and for each of these periods we divided the total loss by the forest cover in 2000, expressed in percentage. We also used the 2010 IBGE census to obtain the Human Development Index, although this metric is at the municipal rather than site level. In the case of settlements, we also use household density (number of registered households on site divided by forest cover in 2010). The number of registered households per site is not available systematically for PA, so we do not use this confounding variable for PA systems. Finally, due to the small sample size for PAs outside Amazonia, the matching variables are limited to the distance to roads, and the 3 periods of past tree cover loss.

Comparing annual tree cover loss in enrolled and never enrolled sites

To adjust for factors affecting the probability of deforestation, we compute characteristics for each conservation site polygon in our database, and these variables are used to estimate a Propensity Score, that is the probability that a site would contain households enrolled in the program given the set of observed characteristics. Similar but never-enrolled enrolled sites were identified by performing an Inverse Probability Weighting based on the estimated Propensity Score in respective samples. Tables S1 to S4 provide covariate balance tables before and after weighting in each of the region. In Fig. 2, we plot the annual average tree cover loss in enrolled sites and the annual weighted average in similar but never-enrolled sites to comparatively visualize tree cover loss before, during and after the intervention. The plots are simply illustrating annual tree cover loss of the enrolled and matched controls, so eventual post-2011 differences between the curves cannot be directly interpreted as an indication of a causal impact because the observed difference is not adjusted for the staggered enrollment into the program, or the eventual differences in tree cover loss at baseline, as well as do not display the statistical uncertainty around the differences between enrolled and never-enrolled sites.

Econometric identification strategy

Our main econometric estimates are based on a generalized Difference-in-Differences estimation (Doubly Robust Difference-in-Differences with multiple time periods) ^59,60. Doubly Robust estimators combines a Difference-in-Differences design (to model the evolution of the outcome) with an Inverse Probability Weighting based on the estimated Propensity Score. This estimator is less susceptible of bias than canonical Difference-in-Differences or matching procedures because the Doubly Robust property ensure that the models are correctly specified even if either (but not both) the Propensity Score and the Difference-in-Differences outcome regression models are miss-specified.

Our generalized DID first dis-aggregates treatment effect into group-time Average Treatment Effects (ATT), i.e., the ATT of each group at each time period, where a group is defined by the year when each site is first enrolled in the program (Fig. S4) ^61,62. Group-time ATT are obtained by performing a simple DID for each group (defined by year of first enrollment) and each time periods, and these group-time ATT can later be aggregated into a more precisely estimated cumulative ATT, corresponding to the sum of calendar year effects during program duration. The cumulative ATT of the intervention across all groups until the end of the program is the sum of the annual impacts of the program on tree cover loss (Fig. S7). Under the Parallel Trend Assumption that both enrolled and non-enrolled sites would have followed the same tree cover loss trend, conditional on observed confounders, we obtain an unbiased estimation of the impact of the program on the average accumulated avoided tree cover loss.

All our estimations were performed with R version 4.1.2 (https://cran.r-project.org/). We implemented our econometric estimates using the R package did, version 2.1.1 (https://bcallaway11.github.io/did/index.html), and R package DRDID version 1.0.3 (https://psantanna.com/DRDID/index.html). Robust analytical standard errors are clustered at the individual level in each specification in the main text.

We provide disaggregated figures of Average Treatment Effects in SI for our key region of interest (Settlements in Amazonia), namely a) Fig.S6: the complete group-time ATT (annual ATT for each pre and post initial enrollment effect, with respect to different year of first enrollment), b) Fig. S7 event-study graph (the annual impact of the program expressed for different duration before and after the first year of enrollment into the program, c) Fig. S8: Group ATT ( the aggregation of ATT with respect to groups defined as sites having the same year of initial enrollment, and d) Fig. S10: Calendar ATT, (the aggregation of ATT with respect to calendar year) (See also Supplementary Text). InS11 to S14, we also provide specification charts testing the sensitivity of our main estimates in all regions to variations in econometric specifications and subgroups obtained from the spec_chart R package (https://github.com/ArielOrtizBobea/spec_chart) ^63,64. In Table S1 to Table S4, we also provide covariate balance tables (obtained from cobalt 4.3.2 R package https://ngreifer.github.io/cobalt) to visualize the performance of our initial Inverse Probability Weighting.

In Supplementary Text, Figs. S6 to S11, and table S1, we show complementary tests of the robustness of the estimated cumulative Average Treatment Effect. We conclude that our econometric specifications are robust to various set of covariates, but that variables characterizing pre-program forest cover loss play a preeminent role in our main estimate. We also observe important heterogeneity across years of initial enrollment into the program, as the average effect seems to be mostly driven by the cohort that started enrollment in 2012 (Figures S6, S8, and S9). We also assess that the program is not necessarily more effective in areas closer to roads or in sites where a higher proportion of households has been enrolled (Figure S11), although rigorously isolating the moderating effect of each variable on the effectiveness of BV would require further analysis.

Converting Average Treatment Effect into Standardized Forest cover effect sizes

Our econometric estimates correspond to the accumulated impact of the program on tree cover loss, expressed in percentage points of tree cover in 2000. To enable comparisons with evaluation performed in other contexts, we want to transform our statistical estimate into an interpretable Standardized Forest cover effect size (12). To do so, we multiply the cumulative ATT by the sum of tree cover in 2000 to obtain an estimate of the total avoided tree cover loss in hectare attributable to the program, under the assumption of homogeneous effect of the program across years and sites. Adapting the formula from ¹², we note F_{counterfactual, 2017} = F_{observed, 2017} + (ATT * F_{observed, 2000}) the counterfactual total forest cover at the end of the program in enrolled sites, and F_{observed, 2017}andF_{observed, 2000}respectively the total observed forest cover at the end of the program and in 2000 in these sites. ATT is negative when the program did avoid deforestation compared to the counterfactual scenario. The standardized forest cover effect size (i.e. the annualized effect of the program on forest cover change) corresponds to the quantity r_a-r_c = r_observed-r_{counterfactual} = ln (F_{observed, 2017} / F_{counterfactual, 2017}) / (2017-2010), as we use 2010 as pre-intervention baseline year, with approximate Standard Error = SE(ATT) / [F_{counterfactual, 2017}/(2017-2010)]. r_a= r_observed=ln (F_{observed, 2017} / F_{observed , 2010}) / (2017-2010) and r_c=r_{counterfactual =} ln (F_{counterfactual, 2017} / F_{observed , 2010}) / (2017-2010) are respectively the annual observed deforestation and the counterfactual deforestation rate in enrolled sites, and F_{observed , 2010}is the total observed forest cover in 2010 in this sites.

Benefit-cost analysis

Our estimate of the avoided CO2 emissions is based on BV’s average deforestation impact. It does not take into account any avoided degradation, avoided emissions in non-forest woody vegetation, shifts in either belowground or soil carbon biomass, or new CO2 storage in tree regrowth. It assumes that when a tree is cut, all the CO2 eventually will be released to the atmosphere. We assume that CO2 emissions due to the program (induced tree cover loss or deforestation spatial spillovers or from gasoline for vehicles during program implementation) are negligible.

For cost data, we are using data on the annual expenditure of BV (not just the payment to enrolled households but the total program expenditures, including operational costs but not including non-budgeted costs eventually supported by enrollees and other stakeholders) in the Non-contributory Social Protection Programmes Database of the Economic Commission for Latin America and the Caribbean (https://dds.cepal.org/bpsnc/programme?id=60).

The estimation of the climate mitigation benefits of the program is obtained using the following formula from ¹⁶: Benefits per tons of avoided CO₂ emissions = SCC* [1/ (1 + r^S)] * [1 - [1 / ( (1 +r) ^D) ]]. Our estimate of climate-mitigation benefits is essentially a linear function of the number of years of delay for deforestation, and the slope is determined by various parameters linking forest cover to CO2 sequestration. Extrapolating from the Technical Support Document released by the US Interagency Working Group on Social Cost of Greenhouse Gases (February 2021), the Social Cost of Carbon (SCC) in 2016 with a 3% discount rate corresponds to USD 40.64 (SE=5.65) in constant US dollars 2016 (SCC= 61.88 with SE=7.77 when discount rate is 2.5%). We choose a discount rate of 3%, consistent with the suggestions of the United States Office of Management and Budget circular A-4 (2003), but also provide estimates for a discount rate of 2.5% (Figs. S15 and S16). We extrapolate the annual growth rate of the SCC from the evolution of this SCC between 2020 and 2050 and estimate the annual rate of growth to be about 2.19%. The effective discount rate r is then equal to (1.03/1.022)-1=0.079. Finally, we follow ¹⁶ in assuming that there is a D=10 years gap between tree-clearing and the complete release of CO₂ into the atmosphere (45% of the biomass released immediately, 45% decomposing on average over 15 years, and 10% used as lumber allowing carbon to be stored for 30 years.). We have not tried to monetize other environmental (e.g., carbon stored in ground and belowground biomass but also other biodiversity and hydrological benefits) and social benefits (e.g., poverty alleviation) of the program, which could have further increased the cost-effectiveness of BV.

We use 2016 as the reference year in our calculation, i.e., all monetary values are expressed in constant 2016 USD dollars while the post-2016 annual costs and benefits of the program are discounted (using a 3% discount rate).

⁵ We did not include recognized traditional communities living on Brazilian public lands in our analysis because we were not able to find georeferenced information about these sites. Enrolled households living in these sites represented about 7% of the total number of households enrolled in the program.

⁶ In each tenure system, management plans notably state the rules, decision-making, and benefit-sharing processes related to collective and individual natural resources management. The compliance with the management plan and the degree of collective action can vary from one site to another ⁵⁷.

⁷ The first row of Fig. S5 presents (statistically non-significant) findings at program scale, respectively in Settlements and Protected Areas, without taking into account systematic differences between Amazonian and non-Amazonian biomes.

Our evaluation of the impacts of the program on tree cover still has some limitations. Tree cover at site level can be influenced by many factors beyond the observed characteristics that we use in our econometric specifications (such as the internal governance of each site, the relative poverty level across sites, the systematic differences in governmental law enforcement efforts, and different level of access to commodities supply chains), so our analysis is still susceptible of bias in the estimation of the impact of the program because we did not use a unit of analysis catching the effect of the program at household scale.

We also did not adjust for eventual heterogeneous time-varying confounders associated with evolving drivers of deforestation, so we assume that all time-varying characteristics affect similarly enrolled sites and the counterfactual, conditional on observed time-invariant characteristics in our weighted regressions. Besides, our econometric estimator does not allow to measure the dose-response effect of changes in the number of enrolled households across time. We detected most of BV’s impact after 2015, which could be due to some years being required to reach a critical enrollment level, or to a time delay between the start of implementation and effects we could identify.

Finally, our climate mitigation benefit-cost analysis also based on several modelling assumptions (regarding the relevant discount rate and Social Cost of Carbon, but also about the conversion of econometric estimates into avoided deforestation and CO₂ emissions), so our main findings reflect our best approximation of the true impact of the program.

Acknowledgments

We thank the staff of Conservation International office in Brazil for the assembling of the data on the number of enrolled households and sites. SC and MM acknowledges the support received from Betty and Gordon Moore, and John and Jody Arnhold. We are grateful for the feedback received from L. Villalobos-Fiat, W. Turner, J. Busch, S. Desbureaux, A. Roopsind, and B. Griscom on earlier version of this manuscript as well as the questions and comments received during the two seminars at Conservation International, a session at the annual conference of the American Association of Geographers in 2019 in Washington, DC, the online annual conference of the European Association of Environmental Economists in 2021, the international BIOECON Conference in 2021, the Environment for Development Initiative annual meeting in 2021, and the International Congress for Conservation Biology in 2021.

Contributions

B.C and S.C. assembled data; S.C., A.P., B.C. and M.B. designed research; S.C. analyzed data; and S.C. and A.P. wrote the paper.

Jack, B. K., Kousky, C. & Sims, K. R. E. Designing payments for ecosystem services: Lessons from previous experience with incentive-based mechanisms. Proceedings of the National Academy of Sciences of the United States of America 105, 9465–9470 (2008).
Ferraro, P. J. & Kiss, A. Direct Payments to Conserve Biodiversity. Science 298, 1718–1719 (2002).
Wunder, S., Börner, J., Ezzine-de-Blas, D., Feder, S. & Pagiola, S. Payments for Environmental Services: Past Performance and Pending Potentials. Annual Review of Resource Economics 12, 209–234 (2020).
Landell-Mills, N. Developing markets for forest environmental services: an opportunity for promoting equity while securing efficiency? Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences 360, 1817–1825 (2002).
Ezzine-de-Blas, D., Wunder, S., Ruiz-Pérez, M. & Moreno-Sanchez, R. del P. Global Patterns in the Implementation of Payments for Environmental Services. PLOS ONE 11, e0149847 (2016).
Giudice, R., Boerner, J., Wunder, S. & Cisneros, E. Selection biases and spillovers from collective conservation incentives in the Peruvian Amazon. Environ. Res. Lett. (2019) doi:10.1088/1748-9326/aafc83.
Alix-Garcia, J., Sims, K. R. E. & Yañez-Pagans, P. Only One Tree from Each Seed? Environmental Effectiveness and Poverty Alleviation in Mexico’s Payments for Ecosystem Services Program. American Economic Journal: Economic Policy 7, 1–40 (2015).
Cisneros, E., Börner, J., Pagiola, S. & Wunder, S. Impacts of conservation incentives in protected areas: The case of Bolsa Floresta, Brazil. Journal of Environmental Economics and Management 102572 (2021) doi:10.1016/j.jeem.2021.102572.
Chervier, C. & Costedoat, S. Heterogeneous Impact of a Collective Payment for Environmental Services Scheme on Reducing Deforestation in Cambodia. World Development (2017) doi:10.1016/j.worlddev.2017.04.014.
Transforming REDD+: lessons and new directions. (Center for International Forestry Research, 2018).
Engel, S. The Devil in the Detail: A Practical Guide on Designing Payments for Environmental Services. IRERE 9, 131–177 (2016).
Samii, C., Lisiecki, M., Kulkarni, P., Paler, L. & Chavis, L. Effects of payment for environmental services (PES) on deforestation and poverty in low and middle income countries: a systematic review. Campbell Systematic Reviews 10, (2014).
Snilstveit, B. et al. Incentives for climate mitigation in the land use sector: a mixed-methods systematic review of the effectiveness of payment for environment services (PES) on environmental and socio-economic outcomes in low- and middle-income countries. CAMPBELL SYSTEMATIC REVIEWS 14, (2018).
Miteva, D. A., Pattanayak, S. K. & Ferraro, P. J. Evaluation of biodiversity policy instruments: What works and what doesn’t? Oxford Review of Economic Policy 28, 69–92 (2012).
Börner, J. et al. The Effectiveness of Payments for Environmental Services. World Development 96, 359–374 (2017).
Jayachandran, S. et al. Cash for carbon: A randomized trial of payments for ecosystem services to reduce deforestation. Science 357, 267–273 (2017).
Costedoat, S. et al. How Effective Are Biodiversity Conservation Payments in Mexico? PLoS ONE 10, e0119881 (2015).
Greenstone, M. & Gayer, T. Quasi-experimental and experimental approaches to environmental economics. Journal of Environmental Economics and Management 57, 21–44 (2009).
Ferraro, P. J. Counterfactual thinking and impact evaluation in environmental policy. New Directions for Evaluation 2009, 75–84 (2009).
Baylis, K. et al. Mainstreaming Impact Evaluation in Nature Conservation. CONSERVATION LETTERS n/a-n/a (2015) doi:10.1111/conl.12180.
Christie, A. P. et al. Quantifying and addressing the prevalence and bias of study designs in the environmental and social sciences. Nature Communications 11, 6377 (2020).
Blackman, A. Evaluating forest conservation policies in developing countries using remote sensing data: An introduction and practical guide. Forest Policy and Economics 34, 1–16 (2013).
Sills, E. O. & Jones, K. Chapter 9 - Causal inference in environmental conservation: The role of institutions. in Handbook of Environmental Economics (eds. Dasgupta, P., Pattanayak, S. K. & Smith, V. K.) vol. 4 395–437 (Elsevier, 2018).
World Bank. Evaluating the permanence of forest conservation following the end of payments for environmental services in Uganda. (2018).
Pagiola, S., Honey-Rosés, J. & Freire-González, J. Evaluation of the permanence of land use change induced by payments for environmental services in Quindío, Colombia. PLoS ONE 11, (2016).
Roopsind, A., Sohngen, B. & Brandt, J. Evidence that a national REDD+ program reduces tree cover loss and carbon emissions in a high forest cover, low deforestation country. PNAS (2019) doi:10/ggdhqn.
Hayes, T., Murtinho, F., Wolff, H., López-Sandoval, M. F. & Salazar, J. Effectiveness of payment for ecosystem services after loss and uncertainty of compensation. Nat Sustain 1–8 (2021) doi:10.1038/s41893-021-00804-5.
Rasch, S., Wünscher, T., Casasola, F., Ibrahim, M. & Storm, H. Permanence of PES and the role of social context in the Regional Integrated Silvo-pastoral Ecosystem Management Project in Costa Rica. Ecological Economics 185, 107027 (2021).
Erbaugh, J. T. Impermanence and failure: The legacy of conservation-based payments in Sumatra, Indonesia. Environ. Res. Lett. (2022) doi:10.1088/1748-9326/ac6437.
Wunder, S. Payments for environmental services: some nuts and bolts. Center for International Forestry Research (2005) doi:10.17528/cifor/001760.
Carrilho, C. D., Demarchi, G., Duchelle, A. E., Wunder, S. & Morsello, C. Permanence of avoided deforestation in a Transamazon REDD+ project (Pará, Brazil). Ecological Economics 201, 107568 (2022).
Spawn, S. A., Sullivan, C. C., Lark, T. J. & Gibbs, H. K. Harmonized global maps of above and belowground biomass carbon density in the year 2010. Scientific Data 7, 112 (2020).
Moran, D. et al. Carbon footprints of 13 000 cities. Environ. Res. Lett. 13, 064041 (2018).
Simonet, G., Subervie, J., Ezzine-de-Blas, D., Cromberg, M. & Duchelle, A. E. Effectiveness of a REDD+ Project in Reducing Deforestation in the Brazilian Amazon. Am J Agric Econ 101, 211–229 (2019).
National Academies of Sciences, Engineering, and Medicine. Valuing Climate Changes: Updating Estimation of the Social Cost of Carbon Dioxide. 24651 (National Academies Press, 2017). doi:10.17226/24651.
Interagency Working Group on Social Cost of Greenhouse Gases, United States Government. Technical Support Document: Social Cost of Carbon, Methane,and Nitrous Oxide Interim Estimates under Executive Order 13990. 48 (2021).
Andrade, B. T. & Coria, J. Has the Green Grant Program Had an Impact on Conservation in Brazil? (School of Business, Economics and Law. Göteborg University, 2018).
McCoshan, K. Protecting the People and the Forest: Bolsa Verde, Brazil. https://oxfamilibrary.openrepository.com/bitstream/handle/10546/621099/cs-bolsa-verde-protecting-people-planet-131120-en.pdf?sequence=4 (2020).
Anderson, L. O., Harding, T., Kuralbayeva, K., Pessoa, A. M. & Wong, P. Y. Pay For Performance and Deforestation: Evidence from Brazil. 69 https://drive.google.com/file/d/1Sn6fR3CAlCl31Eq3l6X5NmSVuicGmVI9/view (2021).
West, T. A. P. & Fearnside, P. M. Brazil’s conservation reform and the reduction of deforestation in Amazonia. Land Use Policy 100, 105072 (2021).
Burgess, R., Costa, F. J. M. & Olken, B. The Brazilian Amazon’s Double Reversal of Fortune. https://osf.io/67xg5 (2019).
Azevedo, A. A. et al. Limits of Brazil’s Forest Code as a means to end illegal deforestation. PNAS 114, 7653–7658 (2017).
West, T. A. P., Börner, J., Sills, E. O. & Kontoleon, A. Overstated carbon emission reductions from voluntary REDD+ projects in the Brazilian Amazon. Proc Natl Acad Sci USA 202004334 (2020) doi:10.1073/pnas.2004334117.
Börner, J., Schulz, D., Wunder, S. & Pfaff, A. The Effectiveness of Forest Conservation Policies and Programs. Annual Review of Resource Economics 12, 45–64 (2020).
Borner, J. et al. The Effectiveness of Payments for Environmental Services. WORLD DEVELOPMENT 96, 359–374 (2017).
Ferraro, P. J. & Simorangkir, R. Conditional cash transfers to alleviate poverty also reduced deforestation in Indonesia. Science Advances 6, eaaz1298 (2020).
Alix-Garcia, J., McIntosh, C., Sims, K. R. E. & Welch, J. R. The Ecological Footprint of Poverty Alleviation: Evidence from Mexico’s Oportunidades Program. The Review of Economics and Statistics 95, 417–435 (2013).
Delacote, P., Le Velly, G. & Simonet, G. Revisiting the location bias and additionality of REDD+ projects: the role of project proponents status and certification. Resource and Energy Economics 67, 101277 (2022).
Robalino, J., Pfaff, A., Sandoval, C. & Sanchez-Azofeifa, G. A. Can we increase the impacts from payments for ecosystem services? Impact rose over time in Costa Rica, yet spatial variation indicates more potential. Forest Policy and Economics 132, 102577 (2021).
Sims, K. R. E. et al. Improving Environmental and Social Targeting through Adaptive Management in Mexico’s Payments for Hydrological Services Program. Conservation Biology n/a-n/a (2014) doi:10.1111/cobi.12318.
Lambin, E. F. et al. Effectiveness and synergies of policy instruments for land use governance in tropical regions. Global Environmental Change 28, 129–140 (2014).
Börner, J., Schulz, D., Wunder, S. & Pfaff, A. The Effectiveness of Forest Conservation Policies and Programs. Annu. Rev. Resour. Econ. (2020) doi:10.1146/annurev-resource-110119-025703.
Presidência da República, Decreto n^o 7572 (2011) (28 de setembro de 2011).
MINISTÉRIO DO DESENVOLVIMENTO SOCIAL E COMBATE À FOME, LEI N^o 12512 (2011) (14 de outubro de 2011).
Bergallo, A. C. A saída da situação de extrema pobreza: uma análise do critério de exclusão do Programa Bolsa Verde entre os anos de 2014 e 2017. IDeAS 11, 131–174 (2017).
Maciel, H. M. & Khan, A. S. Política ambiental no Brasil: a implantação do programa bolsa verde. RICA 6, 172–186 (2015).
Tourneau, F.-M. L. & Beaufort, B. Exploring the boundaries of individual and collective land use management: institutional arrangements in the PAE Chico Mendes (Acre, Brazil). International Journal of the Commons 11, 70–96 (2017).
Pfaff, A., Robalino, J., Lima, E., Sandoval, C. & Herrera, L. D. Governance, Location and Avoided Deforestation from Protected Areas: Greater Restrictions Can Have Lower Impact, Due to Differences in Location. World Development 55, 7–20 (2014).
Callaway, B. & Sant’Anna, P. H. C. Difference-in-Differences with multiple time periods. Journal of Econometrics (2020) doi:10.1016/j.jeconom.2020.12.001.
Sant’Anna, P. H. C. & Zhao, J. Doubly robust difference-in-differences estimators. Journal of Econometrics 219, 101–122 (2020).
de Chaisemartin, C. & D’Haultfœuille, X. Two-Way Fixed Effects Estimators with Heterogeneous Treatment Effects. American Economic Review 110, 2964–2996 (2020).
Goodman-Bacon, A. Difference-in-differences with variation in treatment timing. Journal of Econometrics 225, 254–277 (2021).
Desbureaux, S. Subjective modeling choices and the robustness of impact evaluations in conservation science. Conservation Biology 35, 1621–1626 (2021).
Ortiz-Bobea, A., Ault, T. R., Carrillo, C. M., Chambers, R. G. & Lobell, D. B. Anthropogenic climate change has slowed global agricultural productivity growth. Nat. Clim. Chang. 11, 306–312 (2021).

Table 1. Standardized effect of BV in enrolled regions

		Total number of enrolled sites	Total forest cover in Mha (2010)	Total forest cover in Mha (2017)	Cumulative Average Treatment Effect (2011-2017) in % of tree cover in 2000 [SE]	Observed average annual tree cover change rate (2011-2017) in % r_a	Counterfactual average annual tree cover change rate (2011-2017) in % r_c	Standardized effect on forest cover change rate in p.p. [SE] r_a- r_c
Settlements	Amazonia	299	7.79	7.57	-0.99 [0.42] **	-0.40	-0.55	0.15 [0.07] **
Settlements	Outside Amazonia	544	0.76	0.70	0.29 [0.53]	-1.24	-1.18	-0.05 [0.10]
Protected Areas	Amazonia	51	16.79	16.67	-0.05 [0.12]	-0.11	-0.11	0.00 [0.02]
Protected Areas	Outside Amazonia	9	0.06	0.06	-1.57 [2.19]	-0.49	-0.75	0.26 [0.37]

Mha : millions of hectares. SE : Standard Error. pp.: percentage point. ** corresponds to a 95% confidence interval. Shaded Areas corresponds to econometric estimates rather than observation. Negative r_aor r_c indicate tree cover loss. r_a- r_c is positive if the observed tree cover change rate is inferior to the counterfactual tree cover change rate, and negative otherwise.

There is NO Competing Interest.

natCCbolsaverdeSI20220821.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Cost-effective climate mitigation via conservation incentives targeting poverty: Bolsa Verde's impact in Brazilian Amazonia settlements

Status:

Version 1

Abstract

Figures

Introduction

Results

Discussion

Methods

Limitations

Declarations

References

Table

Additional Declarations

Supplementary Files

Status:

Version 1