Does the oil palm certification create trade-offs between environment and development in Indonesia?

***This article has been accepted in Environmental Research Letters. Please refer to the DOI for the accepted manuscript: https://doi.org/10.1088/1748-9326/abc279.*** Environmental and social problems triggered by rapid palm oil expansion in the tropics have spurred the proliferation of sustainability certification systems such as the Roundtable on Sustainable Palm Oil (RSPO). While the RSPO aims to improve the impact of oil palm production on people and environments, its effect on local development, environmental quality, and, especially, potential trade-offs between these outcomes remain unclear. Here, we evaluate whether RSPO certification of large-scale industrial concessions has promoted village development and supported environmental quality in Indonesia, the top global palm oil producer. Using a panel dataset with observations from 11,000 villages in Kalimantan and Sumatra from 2003-2014, we apply rigorous quasi-experimental methods to quantify the RSPO's impacts on village development and environmental outcomes. In the short-run, RSPO contributed to environmental conservation, but had limited development outcomes. On average, relative to villages with non-certified concessions, RSPO certification reduced deforestation and protected primary forests in Sumatra and lowered the incidence of village-reported land pollution in Kalimantan. RSPO certification also increased the number of private educational facilities in Kalimantan, but had no statistically significant impacts on other development indicators. However, the trade-offs and complementarities between conservation and development vary by slope, a proxy for ecosystem fragility and oil palm profitability. On gentler slopes, we generally find complementarities between conservation and development outcomes. In Kalimantan, certification increased the number of private educational facilities and reduced deforestation and the incidence of land pollution on slopes <2°. In Sumatra, certification increased primary forests, decreased deforestation and the incidence of water pollution on slopes <1°, along with a decrease in population density. Higher slopes in both locations were associated with environment and development trade-offs. We highlight the need to better understand the mechanisms behind the impacts of RSPO and emphasized how the outcomes of certification depend on the communities' bargaining power and the profitability of the land for oil palm production. Thus, we provide insights into understanding these mechanisms behind the impacts of RSPO, which is a prerequisite for improving the design of certification systems and their impacts on the ground.


Introduction
Oil palm, along with beef, soy, and timber, has been identified as a primary driver of deforestation and degradation especially in the tropics (Pendrill et al., 2019). Globally, about a third of forest disturbance across the world from 2000-2015 was associated with commodity-driven deforestation (Curtis et al., 2018). The rapid large-scale expansion of these commodities in developing tropical regions has spurred concerns related to increased land grabbing, loss of ecosystem services and traditional livelihoods, worker safety, as well as poverty and increased migration (Kessler et al., 2007;Obidzinski et al., 2012;Pye et al., 2012;Zoomers et al., 2016).
In response to these environmental and social problems, there has been an increase in voluntary marketbased sustainability certification programs in recent years (Lambin et al., 2014). These provide incentives to commodity producers trading with developed countries or part of a global supply chain to adopt environmentally appropriate and socially beneficial practices (Auld et al., 2008;Cashore et al., 2004). However, despite their recent proliferation (e.g., Milder et al., 2015), little is known about their effectiveness on the ground (Blackman and Rivera, 2011;Evidensia, 2020;Oya et al., 2017).
Given the recent proliferation of oil palm, The Roundtable on Sustainable Palm Oil (RSPO) certification program originated in 2004 as a way to incentivize sustainable production of palm oil (RSPO, 2019). Initiated by non-government organizations and palm oil producers, it is based on a set of Principles and Criteria (P&C), which oil palm producers adopt to minimize the negative impacts of oil palm cultivation on the environment and local communities. The RSPO certification is granted to mills and concessions. Certified oil palm producers are evaluated by independent third-party auditors who verify compliance with the RSPO P&C; if a producer fails to comply with the requirements, certification is revoked. As of November 2019, the volume of RSPO certified palm oil produced globally was ~14.6 M tonnes or 19% of global annual palm oil production (RSPO, 2019). Since 2009 when the first RSPO certificate was issued in Indonesia, the country's share of certified oil palm production has skyrocketed: Currently, approximately 51% of certified oil palm production globally comes from Indonesia, which is also the world's largest producer of palm oil (RSPO, 2019). Because of the significance of oil palm production, the prevalence of poverty especially in rural areas, and the country's importance for conservation and climate change mitigation, we evaluate the impact of RSPO certification on ecosystems and village development in Indonesia.
The evidence on RSPO's impact on ecosystems and communities is still very limited. An emerging body of literature has focused on quantifying the impact of RSPO certification using quasi-experimental and other research designs. Because of Indonesia's significance for conservation and climate change mitigation and the extent of oil palm development as well as data availability, most existing research has focused on the country. For example, using concession holders as the unit of analysis,  find that in Sumatra and Kalimantan RSPO significantly decreased fire activity in certified concessions on lands other than peatlands during wet years, but had no statistically significant impact during dry years, when the likelihood of fires was high.  attempt to quantify the difference in economic profits, poverty, health centers, orangutan habitat and fire activity at the plantation level in Kalimantan. For the most part, they find no statistically significant differences between the certified and non-certified concessions except for increased economic returns. Noojipady et al. (2017) find certification is correlated with a decrease in deforestation and fire-driven deforestation, but did not eliminate them. Based on plantations as the unit of analysis,  demonstrate that RSPO certification significantly lowered deforestation rates in Sumatra and Kalimantan, although the real impact on forest conservation was relatively small. Using villages as the unit of analysis, Santika et al. (2020) examine the impact of certification on poverty and village development in Sumatra, Kalimantan, and Papua and suggest heterogeneity in the outcomes correlated with the pre-existing village level of development.
Building upon existing studies, we provide evidence on the impact of RSPO certification on a set of village development and environmental indicators in Sumatra and Kalimantan, the two major oil palm producing regions in Indonesia (Fig. 1). We complement existing studies by using a detailed theory of change framework, rigorous quasi-experimental and panel data methods, and a rich village-level panel dataset from 2003 to 2014. In contrast to previous studies, we examine the heterogeneity of impacts across a proxy for ecological fragility--specifically, slope, which affects the suitability of land for agricultural production as well as the provision of ecosystem services. We highlight where trade-offs between RSPO's environmental and development goals take place. We find that, relative to noncertified concessions, RSPO contributed to environmental conservation (specifically reducing deforestation and pollution and protecting remaining primary forests), had a limited impact on village infrastructure, and often generated trade-offs between the environment and development goals. These appear to be correlated by changes in the population density. By illustrating the heterogeneity of the RSPO impacts, our results have important implications for understanding the mechanisms behind RSPO's impacts and improving its design.

Average impacts of RSPO in Sumatra and Kalimantan
In terms of the environmental indicators, we find small impacts of certification on average between 2003 and 2014 (Table 1, Fig. 2a). Specifically, relative to traditional concessions, we find certification reduced deforestation by ~0.05% and 1% in Sumatra and Kalimantan, respectively. These results are consistent with previous studies that find small or statistically significant impacts on ecosystems (e.g., ; ). However, we also find that deforestation increased with the village slopes on both islands, ceteris paribus. RSPO certification also conserved more remaining primary forest in Sumatran villages and decreased the incidence of village land pollution in Kalimantan by 21%.
In terms of the village development indicators, we find, on average, RSPO certification increased the number of private educational facilities in Kalimantan, but had no statistically significant effect on the rest of the village development indicators on either island (Table 2). However, the average effect becomes insignificant, when the changes in the population density are considered (SI Table 4). In Sumatra we find that the certification scheme had a negative impact on the number of private educational facilities when the slope increases, ceteris paribus. In Kalimantan we observe that the probability of a village having a health center increases with slope. While in Kalimantan RSPO had no statistically significant impact, it decreased population density in Sumatra (Table 2).

Heterogeneity of RSPO impacts
Using the marginal effects from the regression models, we find that the impact of certification on environmental outcomes varied with the village slope ( Fig. 3a & b). For example, we find that, relative to traditional concessions, in both Kalimantan and Sumatra RSPO certification decreased deforestation on small slopes (<2 degrees, impacts significant at the 10% level), but had the opposite effect on slopes >3 degrees. RSPO also resulted in more remaining primary forest and reduced the incidence of water pollution in Sumatran villages on slopes <3 degrees (impact significant at the 10%); in Kalimantan, it reduced the incidence of land pollution on slopes <2 degrees (impact significant at the 10%).
We find some heterogeneity in the impact on village development as well: we find RSPO was associated with a higher number of private educational facilities on slopes <3 degrees in Kalimantan and fewer such facilities on slopes >1 degrees in Sumatra (Fig. 4). However, these results seem to be driven by changes in population density. When changes in the latter are accounted for, the slope of the relationship for the number of private educational facilities and slope becomes positive for both islands (SI Fig. 8), with a statistically significant (at the 10%) increase in the number of facilities in villages with average slope ~4 degrees in Kalimantan. We did not detect a statistically significant impact of RSPO on the number of households with non-state sources of electricity even after accounting for changes in the population density in Kalimantan (Fig. 4, SI Fig 8). In Sumatra, there was a decrease in the number of households with nonstate sources of electricity on slopes between 1 and 3 degrees (Fig. 4); the patterns are consistent when we account for changes in the population density as well (SI Fig. 8). While in Sumatra RSPO did not have a statistically significant impact on the probability of a village having a health center, it increased the probability in villages on slopes < 2 in Kalimantan (Fig. 4). Finally, we observe a statistically significant reduction in the number of people in treated villages on small slopes (<2 degrees) in Sumatra between 2003 and 2011 relative to observationally similar control villages.
Our results indicate that while the impacts of RSPO on promoting village development via increased infrastructure were limited, it had a positive impact on protecting ecosystems. We also find evidence of trade-offs in the environmental and development goals along our proxy for ecological fragility. For example, in Kalimantan the statistically significant increase in the number of private educational facilities and in the probability of a village having a health center coincides with increased deforestation in treatment villages; the villages with slopes <2 that saw decreases in deforestation and land pollution did not have any statistically significant increases in village infrastructure. In Sumatra, the trade-offs between environmental and development goals occur on slopes <3 degrees: In these villages RSPO decreased deforestation and water pollution and protected remaining primary forests, but reduced household access to non-state sources of electricity. These coincide with a decrease in the number of people.
Our results are robust across multiple specifications (SI Tables 4-5). We also do not find evidence of substitution between state and non-state sources of electricity. That is, we do not find a statistically significant increase in the household access to state electricity or educational facilities (SI Tables 4-5).

Fig. 3a.
Heterogeneity in the impacts on the proxies for forest health for Kalimantan (left) and Sumatra (right) based on marginal effects from the panel data regression models on the matched sample (Eq. 1). An impact is statistically significant if the 95% CI do not cross the 0 line. Fig. 3b. Heterogeneity in the impacts on the pollution proxies for Kalimantan (left) and Sumatra (right) based on marginal effects from the panel data regression models on the matched sample (Eq. 1). An impact is statistically significant if the 95% CI do not cross the 0 line. Fig. 4. Heterogeneity in the village infrastructure and population impacts for Kalimantan (left) and Sumatra (right)based on marginal effects from the panel data regression models on the matched sample (Eq. 1). An impact is statistically significant if the 95% CI do not cross the 0 line.

Sumatra Kalimantan
Discussion Using a novel village-level dataset spanning 2003-2014 and rigorous quasi-experimental methods, we demonstrate that RSPO certification resulted in small, often heterogeneous and geographically limited environmental and village infrastructure impacts relative to traditional oil palm concessions. Between environmental and development goals, we identify trade-offs on both islands. While in Kalimantan the impact on population was statistically insignificant, in Sumatra the trade-offs are correlated with a statistically significant decrease in the number of people in the treated villages. By highlighting the heterogeneity of the RSPO impacts across space and types of outcomes, we provide the first step needed to understand the mechanisms through which certification schemes effect change on the ground.
The trade-offs in environmental and development outcomes seem to coincide with decreases, especially on small slopes, in population density in the RSPO villages. While our analysis involving changes in population was limited by the unavailability of data in 2014, we still detect statistically significant impacts on outmigration only 2 years after the first villages were certified. The decrease in population density raises questions whether certification improved practices to promote environmental conservation or whether the observed impacts are driven by decreased development and natural resource pressures.
It is worth re-emphasizing that our analysis considers the additional impact of RSPO certification relative to traditional oil palm concessions. A potential explanation for RSPO's limited impact on village development is that the contributions to village infrastructure took place when the oil palm concessions were established or developed. Since technically all land in Indonesia falls under the boundaries of villages, previous studies (Budidarsono et al. 2013 in the context of oil palm; Engel et al. 2006 in the context of commercial logging) have suggested that industrial actors may compensate villages, in order to be allowed to operate on areas that overlap village land, regardless of its government designation. Thus, the additional impact of certification that takes place after the concession is established may be small. None of the concessions in our sample were created or developed during the study period. Further contributing to the lack of a statistically significant additional impact are the concomitant policies of the Indonesian government that aims to develop the oil palm producing regions, regardless of certification status, in Sumatra and Kalimantan (Budidarsono et al., 2013).
Our study is subject to a few caveats. First, we do not control for the spatial dependence of many of the outcomes (e.g., distinguish between upstream and downstream villages for water pollution, account for wind patterns for air pollution and fire incidence); we also consider average annual impacts and do not allow for seasonality of the impacts (e.g., differences between the wet and dry seasons). We also do not attempt to quantify the impact of RSPO on emissions from oil palm production. The reason is that in contrast to timber harvesting etc, major emissions from oil palm production occur during land clearing and in ponds containing oil palm detritus from around mills; the latter are said to be a major source of methane emissions (Taylor et al., 2014). While the RSPO P&C includes provisions for minimizing waste and GHG emissions, we do not have data on emissions from non-certified concessions. Our analysis focuses on quantitative changes in the village infrastructure; because of data unavailability, we do not consider the quality and size of facilities and the services they provide. For example, despite the lack of statistically significant positive impacts, for the most part, of RSPO on the number of private educational facilities, incidence of health centers, or household access to non-state electricity, it may be the case the certification scheme improved the quality of these facilities or the services they provide. Lastly, our analysis focuses on large-scale industrial concessions only; even though RSPO certification exists for small holders, we exclude the latter from our analysis due to data unavailability. Previous studies have shown that, on average, oil palm production is beneficial to smallholders in Indonesia (Lee et al., 2013), but the impacts vary by access to land and labor, and tenure security (Krishna et al., 2017). Thus, the exclusion of smallholders from our analysis is likely to introduce a downward bias of the impact of RSPO, making our estimates more conservative.

Conclusion
Drawing on a rich dataset and rigorous impact evaluation methods, ours is the first attempt to evaluate trade-offs between development and environmental impacts of RSPO on local communities. We demonstrate that, in the short-run, while RSPO can contribute to environmental conservation, its impact on rural development may be limited and correlated with decreases in the population density. It remains to be seen what the longer term impacts of certification are on the local environment and communities.
Our work addresses previous calls for Conservation Evaluation 2.0 to consider the heterogeneity of impacts and helps form hypotheses about the mechanisms through which certification effects change on the ground (Miteva et al., 2012). Understanding the mechanisms in a particular context will help improve the design of the intervention. Similar to previous studies (Miteva et al., 2012), we also highlight the need to integrate impact evaluation into the design of interventions like RSPO and collect data that would allow researchers to evaluate the causal impact of the P&C on rural development and the provision of key ecosystem services on the ground. Rigorous impact evaluations that are aligned with the P&C can then be used for credible communication to oil palm consumers to further increase the demand for and compliance with certification.

Methods
Theory of change framework. We first developed a theory of change framework that synthesizes causal pathways between RSPO certification of oil palm plantations and subsequent change in environmental and development outcomes on the ground due to certification. Following Blackman et al. (2017), our framework is based on a literature review on the effectiveness of RSPO as well as a review of the Principles and Criteria from the RSPO (SI Fig 1-2, Table 1). It comprises inputs into the intervention, the intervention itself, intermediate outcomes and impact of the intervention.
Inputs. The process of RSPO certification begins when the oil palm company submits a letter of intent (LOI) to the RSPO Secretariat, notifying stakeholders of the intent to pursue RSPO certification.
Interventions. The LOI defines the scope of the certification and planned assessment dates and invites interested parties to submit comments. In preparation for certification, plantation managers may undertake measures to comply with the certification requirements. For this reason, the LOI better represents the beginning of the certification process ('certification initiation') than the date of granting certification . Upon compliance with the certification standards, an audit takes place. If the audit shows that the plantation adheres to the Principles and Criteria of the RSPO, the plantation is awarded a RSPO certificate.

Intermediate outcomes & Impact indicators.
RSPO-certified plantations must comply with all of RSPO's principles and criteria. We assume that all certified RSPO companies adhere to the Principles and Criteria laid out in the RSPO. We note that cases of non-compliance with RSPO criteria do take place and have been recorded in audit reports (Bishop, 2017). A recent review of audit reports show that noncompliances most frequently reported were related to issues on health and safety, waste, and smallholder training (Bishop, 2017). If producers seeking certification were initially not in compliance with the standard, implementing these practices to achieve certification could result in changes as compared to noncertified oil palm concessions. Based on our review of the RSPO Principles and Criteria (SI Table 1), we categorized how RSPO criteria contribute to social well-being outcomes and considered 14 categories which ranged from improving land transaction transparency to improving worker safety and reducing discrimination in the plantation's workforce. We were not able to evaluate the impact of the RSPO for all 14 categories due to the lack of data for most categories (e.g., reduce child labour, improve rights for employees) as information for these criteria are difficult to obtain especially for noncertified oil palm companies. We focused our analysis on evaluating two categories of outcomes: village development proxied by village infrastructure and ecosystem health.
Improve village development and infrastructure: Under RSPO criteria 6.5 and 6.11, RSPO-certified plantations are expected to provide adequate housing, water supplies, medical, educational and welfare amenities to national standards or above, especially where no such public facilities are available or accessible. Certified plantations are also encouraged to contribute to local sustainable development where appropriate. These contributions towards local development can be interpreted widely by certified palm oil companies, and they could be in the form of Community Development or Corporate Social Responsibility programs provided by the palm oil companies. Some common activities include provision of healthcare facilities and services, as well as participation in school construction.
We used the following impact indicators to evaluate the impact the RSPO has on village development and infrastructure: the number of private educational facilities, the number of households with access to non-state sources of electricity, and the presence of health centers in the village (SI Table 2). These outcomes have been suggested by previous studies (Budidarsono et al., 2013) as being impacted by oil palm development.
Improve ecosystem health: Under RSPO criteria 4.3, 4.4, 5.1, 5.2, 5.3 and 5.6, certified plantations are expected to reduce soil erosion, maintain water quality, practice responsible waste disposal, reduce air pollution and preserve High Conservation Value areas. In addition, RSPO-certified plantations are expected under RSPO criteria 5.5 to avoid the use of fires for waste disposal or any form of land preparation. While we expect that these criteria will be applied within RSPO plantation concessions, the impacts of these criteria could have direct or indirect effects on villages surrounding the RSPO plantation concession.

13
We used the following indicators to evaluate the impact the RSPO has on ecosystem health: the presence of air, land or water pollution, the amount of forest loss relative to forest area since 2000, the proportion of village under primary forests, and the number of fire events in the village.
Estimation. For each outcome we estimate an equation of the kind: where is the outcome for village i in district d at time t, is a set of individual village fixed effects,tyear fixed effects.
is a set of exogenous time-varying covariates (rainfall) assumed to affect the outcome; 0 is a baseline covariate (average village slope) likely to modify the impact of the treatment, indicated by . The treatment variable is a vector of continuous variables indicating the village area under RSPO in a given year. The control group comprises villages under traditional oil palm concessions that have never been certified. We estimate Eq. 1 via a fixed effects regression, clustering the standard errors at the district level to account for the hierarchical nature of the model (Angrist and Pischke, 2009).
Note that the location of RSPO is endogenous: The placement of certification depends on the location and characteristics of the villages it spans. For example, villages with fewer residual forests could be less likely to contribute to recent deforestation and increase the eligibility for RSPO certification (See Covariates for more details). Thus, the estimates can be unstable and depend on the specification if there is limited overlap in the covariate distributions of the treatment and control groups (Imbens and Wooldridge, 2009). To address this issue, we follow recent recommendations (e.g., Ferraro and Miranda, 2017) and employ matching techniques, specifically, nearest neighbour covariate matching with a Mahalanobis distance metric and trimming on the propensity score, to pre-process the sample to ensure sufficient covariate overlap. The covariates used in the matching include population density in 2003, road density, proportion of the village area under peat, oil palm in 2000, forest and primary forest, proportion of the village under non-forest, production forest and conservation forest land use, dependence on fuelwood, the proximity to cities, mills, and ports, the village area, the length of the river network within a village, slope, elevation, temperature, precipitation, the number of bird species expected in a village, as well as the baseline values of the village development proxies (the presence of Tables S3 and S4). We then apply fixed effects panel data estimators on the matched sample, using frequency weights to account for some observations in the control group being used more than once. The fixed effects model further helps address potential endogeneity by removing linear time-invariant unobserved characteristics that may have not been balanced by the matching exercise.
Because of the differences in the history of oil palm development between Sumatra and Kalimantan, we perform the estimation for the two islands separately.

Robustness checks to rule out alternative explanations
We conduct a series of robustness checks to alternative specifications regarding the role of matching to preprocess the sample, functional form of the equation, and outcomes (See Supplementary Information). In addition, we test whether the results are driven by changes in population density and changes in government-funded infrastructure. Because data on population numbers were unavailable for 2014, we restrict the analysis on population to 2011.

Study area
The rise of Indonesia as the world's leading palm oil producer is a result of a phenomenal expansion of its oil palm plantation sector. Between 1970 and 2017, the total area of oil palm plantations in Indonesia increased from 100,000 ha to 12.3 million ha (BPS, 2017;Hasegawa and Matsuoka, 2015). This rapid expansion was fueled by state policies to use smallholder tree crop planting as a form of socioeconomic development in the outer islands (e.g., Sumatra, Kalimantan) and to attract foreign private investment to boost the nation's oil palm industry (McCarthy, 2010).
We focus on nine Indonesian provinces (four in Sumatra, five in Kalimantan) which represent 83% (26.7 M tonnes) of Indonesia's palm oil production and 90% (1.5 M ha) of RSPO certified concessions in 2017 (BPS, 2017).

Data
Combining village-level census data and fine resolution geospatial information, our dataset includes 5 time points (2003, 2005, 2008, 2011, and 2014) across 11 years, and includes 7,983 and 3,545 villages in Sumatra and Kalimantan, respectively. To evaluate the impact of certification on the outcomes of interest, we use the village as our unit of analysis. An Indonesian village or "Desa" includes human settlements and adjacent land as mapped by the Indonesian Bureau of Statistics (BPS). Because many of the community development programs and corporate social responsibility programs from RSPO certified oil palm companies are implemented at a village-level, we expect any impacts from RSPO to be detectable at this administrative level.

Village Development and Environmental Indicators.
We use the Indonesian Village Potential Statistics (PODES) datasets that are based on information from village heads and purport to be the overall socioeconomic conditions of the village. The village head (kepala desa) is the elected government official in the village and there is one elected head per village under the Indonesian village administration. PODES data have been used widely to evaluate the effect of land-use policies such as REDD+ (Jagger and Rana, 2017), forest certification schemes (Miteva et al., 2015), social forestry programs (Santika et al., 2019a), and oil palm development (Santika et al., 2019b) on social and environmental outcomes. In constructing the panel dataset, we only retained villages whose village names were consistent in our PODES dataset for 2003, 2005, 2008, 2011, and 2014 and removed any villages with incomplete data. This process reduced our original dataset of 11,874 villages to 7,983 villages in Sumatra, and 6,524 villages to 3,545 villages in Kalimantan. These include villages that were not spanned by any oil palm concessions; for the analysis, we retained the villages that partially or fully under an industrial oil palm concession. We obtained six village infrastructure outcomes and three environmental outcomes from PODES (SI Table 2) and derived three additional environmental outcomes using spatial data on forest cover and fires (SI Table 2).

Treatment definition.
The treatment is based on the timing of the letter of intent (LOI) submission. Villages that overlap partially or fully by concessions that submitted such letters are considered treated in LOI year ("certified villages" or "treated villages" for short). The control group comprises villages that have at least a fraction of their area that overlaps with a traditional (noncertified) concession ("noncertified villages" for short). Villages which overlap with both RSPO and non-RSPO concessions were considered part of the treatment group. We exclude from the analysis for villages with letter of intent (LOI) in 2014 and 2015, as we do not have data post 2014. Our sample includes 569 and 149 treated and 1,779 and 1,607 control villages for Sumatra and Kalimantan, respectively (SI Table 6).
Covariates. The covariates that affect the spatial distribution of RSPO certified concessions and affect the outcomes are presented in SI Table 3. RSPO oil palm concessions tend to have a longer history of oil palm establishment and have less forests to begin with . These oil palm concessions contain fewer residual forests, likely reducing the opportunity cost imposed through participation in the RSPO. State regulations on land-use zones defined by the Indonesian Ministry of Environment and Forestry would also influence a company's decision to certify its oil palm plantations (Giessen et al., 2016). RSPO concessions have to abide by national laws and follow state regulations on land-use that allow for the development of oil palm plantations over areas zoned for non-forestry uses (areal pengunaan lain; APL). It is prohibited under the RSPO for the concession to be under any of the forest estate (kawasan hutan) land zones which are set aside for production including: limited production forests (hutan produksi terbatas; HPT), permanent production forests (hutan produksi; HP) and convertible production forests (hutan produksi yang dapat di konversi; HPK). To account for this, we included village-level extent of overlap with land-use zones defined by the Indonesian government (e.g., non-forest estates (APL) grouped as APL, forest-estates set aside for production including HPT, HP and HPK grouped as Production Forests, and forest-estates set aside for conservation and protection grouped as Conservation Forests). Areas important for biodiversity conservation or climate change mitigation are also less likely to be targeted by RSPO as companies are able to avoid costs related to biodiversity and carbon assessments of these areas. To account for this nonrandom placement of RSPO concessions, we include covariates on the extent of different types of landcover in a village such as primary forest cover in 2000, planted oil palm extent in 2000, and peatland extent as well as the number of bird species. To control for distance to markets and proxy for transportation costs, we use the proximity to oil palm mills, ports, and cities, the length of the river network and the road density within a village. In addition, we include covariates to control for different oil palm suitability (slope, elevation, temperature, precipitation).
We use these covariates in the matching to preprocess the sample and ensure covariate overlap (SI Tables  7-9).
Heterogeneity of the treatment. We proxy ecological fragility by slope. Because of the importance of slope for determining agricultural suitability and the profitability of logging, we examine how slope modifies the impact of RSPO certification.  Tables 1-9 Legends for Datasets S1-2 SI References

Other supplementary materials for this manuscript include the following:
Datasets S1-2

Supplementary Text
Evidence base on the effectiveness of RSPO. We conducted a literature search on the Web of Science using search terms "Roundtable of Sustainable Palm Oil" AND "RSPO" and obtained 30 publications from Sept 2017 to July 2018 that evaluated the effectiveness of the RSPO across environmental, social and economic issues (SI Fig. 1; Dataset S1). We categorized the studies into qualitative and quantitative studies. Qualitative studies were either in the form of a review about the effectiveness of the RSPO, or they were studies from the field of political economy or sociology which analyzed decision-making processes in the RSPO and how these processes may exclude smallholders and indigenous communities (SI Fig 1). Quantitative studies derived social and environmental variables and assessed how variables associated with the RSPO showed an increase or decrease as compared to the same set of variables associated with noncertified oil palm plantations (SI Fig. 1).
Of the 30 publications, three studies used a counterfactual approach to determine the effect of the RSPO on social and environmental variables .  used annual remotely sensed metrics of tree cover loss and fire occurrence to evaluate the impact of certification on deforestation and fire occurrence from 2001 to 2015 in Sumatra and Kalimantan. They combined matching and panel methods and found that certification lowered deforestation by 33% from a counterfactual of 9.8 to 6.6% y -1 but had no causal impact on forest loss in peatlands and active fire detection rates.  used environmental (biodiversity, fire) and socioeconomic metrics (poverty, health services, yield, profits) to evaluate the impact of certification in Kalimantan. They used a before and after control impact analysis for each metric and found that certification was associated with a greater increase in profit and yields but had no effect on environmental (orangutan presence, fire incidents) and social outcomes (households receiving government assistance, availability of rural health services).  compared fire activity on RSPO-certified and non-certified oil palm plantations using nonparametric matching methods to control for factors such as the area of the concession and the mean road density, which may bias the outcome of fire activity. They found that fire activity from 2012-2015, was significantly lower on RSPO certified concessions than non-certified concessions over non-peatlands in wet years but did not find any difference in fire activity over peatlands in all years or non-peatlands in dry years between RSPO and non-certified oil palm plantations.
We note that there is a body of literature that is critical of the way the RSPO has formulated its sustainability standards to address social issues related to oil palm expansion (Cheyns, 2014;Oosterveer et al., 2014). Some concerns include the dispossession of smallholders and communities of their way of life and a reliance on companies or non-governmental organizations for certification (Cheyns and Riisgaard, 2014), the focus on economic principles and efficiency practices instead of local concerns of land sovereignty (Silva-Castañeda, 2012), as well as the skewed processes of inclusion and poor participation of local communities (Ponte and Cheyns, 2013). We acknowledge these broader societal problems related to the RSPO but limit our study to focus only on specific social wellbeing outcomes derived from the sustainability standards under the RSPO.  principles (no. 2,4,5,6) which were related to improving social well-being outcomes for estate workers and/or local communities (SI Table 1). We did not include criteria no. 7 on New Plantings since many of the criteria are a repeat of the criteria under no. 1-6 but applied in the context of developing new plantations. These criteria and indicators are used during certification audits to evaluate whether the oil palm company undergoing RSPO certification contribute to any demonstrable improvements in social wellbeing indicators of their employees and local communities (Bishop, 2017 6. Reduce conflicts with communities (Criteria 6.1, 6.2, 6.3) 7. Improve working conditions for employees (Criteria 6.5) 8. Improve village infrastructure (Criteria 6.5) 9. Improve rights for employees (Criteria 6.6) 10. Reduce child labour (Criteria 6.7) 11. Reduce discrimination in workforce (Criteria 6.8) 12. Prevent sexual harassment in workforce (Criteria 6.9) 13. Improve business transparency for smallholders and local businesses (Criteria 6.10) 14. Improve village development (Criteria 6.11) Benefits from improved oil palm management in RSPO certified oil palm plantations are likely to accrue at the village level as companies tend to engage with villages within and surrounding their oil palm concession. The purpose of engagement could include ensuring that the company's practices are not harming local environmental conditions, or to resolve any disputes that may arise from the company's practices such as pollution or land conflicts. Often, these engagements are conducted via the company's public relations division or HUMAS (Hubungan Masyarakat), which is the organization in the company that corresponds and engages with village leaders on any matters that involve the practices of the oil palm company. RSPO auditors also conduct interviews with members of villages in and surrounding the oil palm concession to assess the company's compliance to principles and criteria related to social well-being. Village units as defined and mapped under the Indonesian census (BPS, 2016) combine both human settlements and adjacent land use, and their boundaries overlap with the concession boundaries in a way that any management practices implemented in the oil palm concession has a direct and possibly cumulative effect on the village.

DATA SOURCES Oil palm concession boundary dataset in Indonesia.
Oil palm concessions boundaries were obtained from the RSPO secretariat and additional digitization of polygons from maps made available from audit reports hosted on the RSPO website, and supplemented with plantation boundaries from annual communications of progress (ACOP) reports. For more details on this dataset, please refer to .

Village boundary dataset in Indonesia.
Spatially explicit village-level boundaries were obtained from the Indonesian Bureau of Statistics (Badan Pusat Statistika, BPS) and the National Planning Agency (BAPPEDA) for four provinces in Sumatra, (North Sumatra, Riau, Jambi and South Sumatra) and three provinces in Kalimantan (North Kalimantan, East Kalimantan and Central Kalimantan). Village-level boundaries for West and South Kalimantan were obtained from the Database of Global Administrative Areas (https://gadm.org/data.html). Village-level boundaries for Riau were dated at 2010, while that of Jambi, South Sumatra and North Sumatra at 2003. Village boundaries for Riau were assumed to be updated to present, but some names with obvious spelling mistakes were corrected to match those recorded in PODES 2014. Village boundaries for Jambi and South Sumatra were entirely re-drawn according to village boundaries seen on Google maps, resulting in the creation of villages which were not present in 2003. This creation of new villages is often due to the splitting or combining of villages over the years, a common practice in Indonesia which also presents a challenge to reliability of counterfactual longitudinal studies. By updating the village boundaries using more recent data we have attempted to improve alignment between our PODES dataset and contemporarily available geospatial data. However, Google maps is not entirely up to date. For example, some regency boundaries were changed in 2013 but were not reflected on Google maps. Where there were mismatches, the more recent regency boundaries were adhered to. This is particularly in Banyu Asin and Muara Enim regencies of South Sumatra. Village-level boundaries for North and East Kalimantan were dated at 2014, while that of Central Kalimantan was dated at 2003. Village-level boundaries for West and South Kalimantan were undated yet the village boundaries from this data set (GADM) were found to be sufficiently similar when compared with Google 2018 boundaries. Other than the quality checks using Google Earth as a reference, village boundary names were checked for changes and consistency using the Indonesian Ministry of Home Affairs website (https://www.kemendagri.go.id/). Village boundaries were updated to agree consistently with the names in our PODES dataset across the years 2003, 2005, 2008, 2011 and 2014. Villages that were missing or were not reported consistently in our PODES dataset were removed. Our final database consists of 7,983 villages in Sumatra and 3,545 villages in Kalimantan. These include villages that were not spanned by any oil palm concessions; for the analysis, we retained the villages that partially or fully under an industrial oil palm concession.
Forest and Primary Forest variables. We defined 'forest' using Hansen et al. (2013) percentage tree cover maps derived from Landsat and only used pixels with >90% tree cover . We excluded any pixels that were rubber, fiber or oil palm plantations based on the Carlson et al. (2013) and Gunarso et al. (2013) dataset. To define 'primary forest', we restricted forest pixels that were labeled as primary forest according to Margono et al. (2014) dataset, which defined primary forest as mature natural forest of ≥5 ha retaining natural composition and structure that has not been completely cleared and replanted in recent history. This includes primary degraded forest that has suffered partial canopy loss (e.g., via logging), and which therefore has altered composition and structure. This dataset is available for all years from 2000 to 2018.
We use the proportion of primary forest remaining in 2000 to assess whether the village is under an active oil palm concession. None of the villages in our sample were entirely under primary or other forest: In Sumatra, about 3% of the area of the treated villages was under primary forest (10, max=68%); on average, 4.5% of the area of the control villages was under primary forest (12.7, max=82%). In Kalimantan, the values for the treated and control villages are 8.7% (13.8, max=70%) and 11.5% (17.7, max=75%), respectively. Fire hotspot variable. We used Moderate Resolution Imaging Spectroradiometer (MODIS) global monthly fire location product (MOD14 v6) to identify fire hotspots Giglio and Justice, 2015). We obtained the number of hotspot pixels in each village for years 2003, 2005, 2008, 2011, and 2014. Our village environmental indicator for fires is the number of fire events within a given year for the village.

PODES Data. The Indonesian Potential Statistics (PODES) dataset is managed by the Indonesain Bureau of Statistics (BPS or Badan Pusat Statistika)
and is an extensive survey conducted once every 3-4 years since 1976. The surveys collect data at the lowest administrative level of the local government, which is the village or desa. Sub-district level BPS agents typically interview the village or urban neighbourhood head to gather information on variables that cover various aspects of village life (e.g., education, health, land use, economy, cultural activities). PODES reports the presence or absence of a number of health facilities in the village including polyclinics, public health centers, subsidiary health centers, maternity posts and integrated health posts. We pooled all the health facilities into one variable and indicated presence of health facilities when the village reported presence of any of the above listed health facilities (Presence of health facilities). PODES reports on the number of education facilities which include state and private education facilities such as kindergartens, elementary schools, junior schools, and senior schools. We pooled all state education facilities into one variable (No. state education facilities) and all private education facilities into one variable (No. private education facilities). We also obtained the total number of education facilities in the village (No. education facilities) by summing state and private education facilities. PODES also reports on the number of families with access to electricity from state owned company PLN or Perusahaan Listrik Negara (No. families with access to state electricity) and nonstate owned company sources (No. families with access to nonstate electricity) for each village. We used these six indicators to evaluate the effect of RSPO on improving village infrastructure. This dataset is available for all years.
We also pooled together data related to population for each village based on the summary of number of men and number of women reported at the village level in PODES. This data is available for all years except 2014. PODES reports on the presence or absence of land, water and air pollution in the village. We used these variables (Presence of land pollution, Presence of water pollution, Presence of air pollution) to evaluate the effect of RSPO on reducing the incidents of land, water, and air pollution which indicate a healthier ecosystem for the well-being of people. We used these three indicators from PODES to evaluate the effect of RSPO on improving ecosystem health. This dataset is available for all years. In addition to the PODES derived environmental indicators, we also obtained three other environmental indicators as proxy for improving ecosystem health for villages. These indicators were derived using spatial data on forests and fires (see Forest and Primary Forest variables and Fire hotspot variable).

Robustness checks
We perform a series of robustness checks to test the following hypotheses:  The results are sensitive to the functional form chosen. We test this hypothesis by employing multiple distributions of outcome variables (Poisson, logistic distributions and dependent variable transformations as well as specifications without matching).  Changes in population density drive the results. To test the hypotheses, we standardize the non-binary village development indicators for a given year by the population in that year. Because of data unavailability for the population in 2014, we limit the analysis to 2011.  The state takes up the functions of the private sector, when the latter underperforms. To test the hypothesis, we repeat the analysis using the number of state educational facilities and the number of households with state electricity to test for substitution of private facilitis with state.
The results from the robustness checks are summarized in SI Tables 4-5. We find that the results are robust to multiple specifications. We find that the impacts of the RSPO on the number of educational facilities in Sumatra and Kalimantan as well as on the household access to electricity in Kalimantan appear driven by changes in population density. We discuss the results on the impact of population density in greater detail in the main text. We find no statistically significant impacts of the RSPO on state-funded infrastructure.
12 SI Fig. 7. Trend plots for incidence of water, land and air pollution in Sumatra and Kalimantan for treated (RSPO) and control (noncertified) villages using the raw (unmatched) sample. SI Table 1. List of RSPO criteria related to improving social well-being outcomes for estate workers or local communities. We identified 22 criteria under four RSPO principles which include no. 2) Compliance with applicable laws and regulations, no. 4) Use of appropriate best practices by growers and millers; no. 5) Environmental responsibility and conservation of natural resources and biodiversity; and no. 6) Responsible consideration of employees, and of individuals and communities affected by growers and mills.

Description of RSPO Criteria
Description of RSPO Indicators Contribution to social wellbeing 2.2 The right to use the land can be demonstrated, and is not legitimately contested by local communities with demonstrable rights.
 Documents showing legal ownership or lease, history of land tenure and the actual legal use of the land.  Evidence that legal boundaries are clearly demarcated and visibly maintained.  Where there are, or have been, disputes, additional proof of legal acquisition of title and that fair compensation has been made to previous owners and occupants; and that these have been accepted with free prior and informed consent.  Absence of significant land conflict, unless requirements for acceptable conflict resolution processes (criteria 6.3 and 6.4) are implemented and accepted by the parties involved. Practices minimise and control erosion and degradation of soils.

Improve land transaction
 Maps of fragile soils must be available.  A management strategy should exist for plantings on slopes above a certain limit (needs to be soil and climate specific).  Presence of road maintenance programme.

Improve ecosystem health:
Companies are obliged to minimise soil degradation and erosion which will reduce  Subsidence of peat soils should be minimised under an effective and documented water management programme.  A management strategy should be in place for other fragile and problem soils (e.g. sandy, low organic matter, acid sulfate soils) surface runoff and water pollution.

4.4
Practices maintain the quality and availability of surface and ground water.
 An implemented water management plan.  Protection of water courses and wetlands, including maintaining and restoring appropriate riparian buffer zones.  Monitoring of effluent BOD.  Monitoring of mill water use per tonne of FFB .

Improve ecosystem health:
Companies are obliged to address the effects of their water use and effects of their activities on local water resources.  Documented impact assessment.  Where the identification of impacts requires changes in current practices, in order to mitigate negative effects, a timetable for change should be developed.

Improve ecosystem health:
Companies are obliged to evaluate all aspects of their operations and document positive and negative impacts they may have to the environment.

5.2
The status of rare, threatened or endangered species and high conservation value habitats, if any, that exist in the plantation or that could be affected by plantation or mill management, shall be identified and their conservation taken into account in management plans and operations.
Information should be collated that includes both the planted area itself and relevant wider landscape-level considerations (such as wildlife corridors). This information should cover:  Presence of protected areas that could be significantly affected by the grower or miller.  Conservation status (e.g. IUCN status), legal protection, population status and habitat requirements of rare, threatened, or endangered species, that could be significantly affected by the grower or miller.  Identification of high conservation value habitats, such as rare and threatened ecosystems, that could be significantly affected by the grower or miller. If rare, threatened or endangered species, or high conservation value habitats, are present, appropriate measures for management planning and operations will include:  Ensuring that any legal requirements relating to the protection of the species or habitat are met.

Improve ecosystem health & Improve community access to sites with cultural values:
Protection of high conservation value habitats include sites which hold significant value for biodiversity and ecosystem services, as well as sites that hold important cultural and traditional values for communities.
 Avoiding damage to and deterioration of applicable habitats.  Controlling any illegal or inappropriate hunting, fishing or collecting activities; and developing responsible measures to resolve human-wildlife conflicts (e.g., incursions by elephants).

5.3
Waste is reduced, recycled, re-used and disposed of in an environmentally and socially responsible manner.
 Documented identification of all waste products and sources of pollution  Safe disposal of pesticide containers.  Having identified wastes, a waste management and disposal plan must be developed and implemented, to avoid or reduce pollution.

Improve ecosystem health:
Companies are obliged to dispose waste responsibly which has direct implications for how the environment is polluted.

5.5
Use of fire for waste disposal and for preparing land for replanting is avoided except in specific situations, as identified in the ASEAN guidelines or other regional best practice.
 Documented assessment where fire has been used for preparing land for replanting.

Improve ecosystem health:
Companies are encouraged not to use fires for waste disposal and land clearing which has direct implications on air pollution and health effects for surrounding communities.

5.6
Plans to reduce pollution and emissions, including greenhouse gases, are developed, implemented and monitored.
 An assessment of all polluting activities must be conducted, including gaseous emissions, particulate/soot emissions and effluent (see also criterion 4.4). Significant pollutants and emissions must be identified and plans to reduce them implemented.  A monitoring system must be in place for these significant pollutants which goes beyond national compliance.  The treatment methodology for POME is recorded.

Improve ecosystem health:
Companies are obliged to assess all polluting activities such as greenhouse gas emissions and particular/soot emissions. This direct implications on air pollution and health effects for surrounding communities. 6.1 Aspects of plantation and mill management that have social impacts are identified in a participatory way, and plans to mitigate the negative impacts and promote the positive ones are made, implemented and monitored,  A documented social impact assessment including records of meetings.  Evidence that the assessment has been done with the participation of affected parties. Participation in this context means that affected parties are able to express their views through their own representative institutions, or freely chosen spokespersons, during the identification of impacts, reviewing findings and plans for mitigation, and monitoring the success of implemented plans.

Reduce conflicts with communities:
This criterion obliges companies to perform a social impact assessment for their plantation and mill operations and provides a platform for communities to raise their opinion regarding any social impacts they face from to demonstrate continuous improvement.
 A timetable with responsibilities for mitigation and monitoring, reviewed and updated as necessary, in those cases where the assessment has concluded that changes should be made to current practices.  Particular attention paid to the impacts of outgrower schemes (where the plantation includes such a scheme).
the companies. Companies are obliged to have a time-bound plan to mitigate the problem raised.

6.2
There are open and transparent methods for communication and consultation between growers and/or millers, local communities and other affected or interested parties.
 Documented consultation and communication procedures.  A nominated management official responsible for these issues.  Maintenance of a list of stakeholders, records of all communication and records of actions taken in response to input from stakeholders.

Reduce conflicts with communities:
This criteria oblige companies to provide and document procedures for communities to raise any social impacts they face.

6.3
There is a mutually agreed and documented system for dealing with complaints and grievances, which is implemented and accepted by all parties.
 The system resolves disputes in an effective, timely and appropriate manner.  Documentation of both the process by which a dispute was resolved and the outcome.  The system is open to any affected parties.

Reduce conflicts with communities:
This criteria oblige companies to out in place a grievance process for resolving disputes. Companies must document the process how disputes are resolved and ensure that the system is open to any affected parties. 6.4 Any negotiations concerning compensation for loss of legal or customary rights are dealt with through a documented system that enables indigenous peoples, local communities and other stakeholders to express their views through their own representative institutions.
 Establishment of a procedure for identifying legal and customary rights and a procedure for identifying people entitled to compensation.  A procedure for calculating and distributing fair compensation (monetary or otherwise) is established and implemented. This takes into account gender differences in the power to claim rights, ownership and access to land; differences of transmigrants and longestablished communities; differences in ethnic groups' proof of legal versus communal ownership of land.  The process and outcome of any negotiated agreements and compensation claims is documented and made publicly available.

Improve land transaction transparency:
This criterion obliges companies to establish a procedure for allowing indigenous communities to be compensated for loss of any legal or customary rights. The process and outcome of any negotiated agreements have to be documented and made publicly available.
6.5 Pay and conditions for employees and for employees of contractors  Documentation of pay and conditions.  Labour laws, union agreements or direct contracts of employment detailing payments and conditions of Improve working conditions for employees & Improve village infrastructure: always meet at least legal or industry minimum standards and are sufficient to provide decent living wages employment (e.g., working hours, deductions, overtime, sickness, holiday entitlement, maternity leave, reasons for dismissal, period of notice, etc) are available in the languages understood by the workers or explained carefully to them by a management official.  Growers and millers provide adequate housing, water supplies, medical, educational and welfare amenities to national standard or above, where no such public facilities are available or accessible (not applicable to smallholders).
Companies are obliged to provide at least legal or industry minimum standards for pay and working conditions. Companies are obliged to provide adequate housing, education, and welfare amenities where no such public facilities are available or accessible.

6.6
The employer respects the right of all personnel to form and join trade unions of their choice and to bargain collectively. Where the right to freedom of association and collective bargaining are restricted under law, the employer facilitates parallel means of independent and free association and bargaining for all such personnel.
 A published statement in local languages recognizing freedom of association  Documented minutes of meetings with main trade unions or workers representatives.

Improve rights for employees:
Companies are obliged to allow employees to form and join unions to improve their working conditions. 6.7 Children are not employed or exploited. Work by children is acceptable on family farms, under adult supervision, and when not interfering with education programmes. Children are not exposed to hazardous working conditions.  Documentary evidence that minimum age requirement is met.

Reduce child labour:
Child labour is prohibited in companies.
6.8 Any form of discrimination based on race, caste, national origin, religion, disability, gender, sexual orientation, union membership, political  A publicly available equal opportunities policy including identification of relevant/affected groups in the local environment.  Evidence that employees and groups including migrant workers have not been discriminated against.

Reduce discrimination in workforce:
Companies are obliged to have an equal opportunity policy for employment.
affiliation, or age, is prohibited. 6.9 A policy to prevent sexual harassment and all other forms of violence against women and to protect their reproductive rights is developed and applied.
 A policy on sexual harassment and violence in the workplace and records of implementation.  A specific grievance mechanism is established.

Prevent sexual harassment in workforce:
Companies are obliged to develop a policy on sexual harassment and record the implementation of this policy. Companies are obliged to develop a grievance mechanism to allow reporting of any sexual harassment cases. 6.10 Growers and mills deal fairly and transparently with smallholders and other local businesses.
 Current and past prices paid for FFB shall be publicly available.  Pricing mechanisms for FFB and inputs/services shall be documented (where these are under the control of the mill or plantation).  Evidence shall be available that all parties understand the contractual agreements they enter into, and that contracts are fair, legal and transparent.  Agreed payments shall be made in a timely manner.

Improve business transparency for smallholders and local businesses:
Companies are obliged to treat their smallholders and local businesses in a fair and transparent manner.

6.11
Growers and millers contribute to local sustainable development wherever appropriate.
 Demonstrable contributions to local development that are based on the results of consultation with local communities.

Improve village development:
There is a wide interpretation of this criteria from oil palm companies and this criterion fall under many of the existing Community Development or Corporate Social Responsibility activities by companies. Some common activities include road maintenance, providing healthcare facilities or services and participation in school construction. Job vacancies are offered to local people directly or indirectly. Fraction of village area overlapped with oil palm extent in 2000. (Austin et al., 2017) Peat area Fraction of village area overlapped with peatland extent (Wahyunto et al., 2005(Wahyunto et al., , 2003 Distance to city km Distance of village to city. (Minnemeyer et al., 2009)