57544 GEF Annual Impact Report 2008 OctOber 2009 Global Environment Facility Evaluation Office GEF Annual Impact Report 2008 October 2009 (The main findings and recommendations of this evaluation were presented to the GEF Council in November 2008.) Evaluation Report No. 48 © 2009 Global Environment Facility Evaluation Office 1818 H Street, NW Washington, DC 20433 Internet: www.gefeo.org Email: gefevaluation@thegef.org All rights reserved. The findings, interpretations, and conclusions expressed herein are those of the authors and do not necessarily reflect the views of the GEF Council or the governments they represent. The GEF Evaluation Office does not guarantee the accuracy of the data included in this work. The boundaries, colors, denomi- nations, and other information shown on any map in this work do not imply any judgment on the part of the GEF concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Rights and Permissions The material in this work is copyrighted. Copying and/or transmitting portions or all of this work without permission may be a violation of applicable law. The GEF encourages dissemination of its work and will normally grant permission promptly. ISBN-10: 1-933992-19-0 ISBN-13: 978-1-933992-19-8 Credits Director of the GEF Evaluation Office: Robert D. van den Berg Task Manager: David Todd, Senior Evaluation Officer, GEF Evaluation Office Editing and design: Nita Congress Cover design: Jean Wegimont, Atelier2 Cover photos: © Tran Thi Hoa/World Bank (Vietnam); Forwardcom (bridge); World Bank (Latin America). Evaluation Report No. 48 A FREE PUBLICATION Contents Foreword ........................................................................................................................................ v Acknowledgments ...................................................................................................................... vi Main Report ................................................................................................................................... 1 1 Overview of Impact Evaluation Work in 2008 ........................................................................................ 1 2 Case Study of the Social Impacts of Protected Areas: North and Northeast Thailand................. 3 3 Case Study on the Social Impacts of Protected Areas: Costa Rica ..................................................... 4 4 Conclusions on Impact Evaluation Work in 2008 .................................................................................. 5 Annex: Selected Social Impacts of Protected Areas in Thailand and Costa Rica................... 7 iii Foreword This is the second annual impact report pro- and evaluate the assumptions of why project duced by the Evaluation Office of the Global interventions are supposed to achieve impact on Environment Facility (GEF). The Office presents the global environment. its impact work in a consolidated annual report, which covers work that has produced final prod- At the time of this report's publication, prepara- ucts during the period covered, as well as ongoing tions were under way for a theory-based impact work. The Office continues to explore the range evaluation of GEF support in countries with econ- of methodologies that can be usefully applied to omies in transition to reduce production and con- evaluate the impacts achieved by GEF-supported sumption of ozone-depleting substances. activities. Because no decisions were proposed on the basis The Annual Impact Report 2008 brings together of this report, it was presented to the GEF Coun- the results of two interrelated approaches that cil in November 2008 for information purposes explore the value of using quasi-experimental only. The documents on which this annual impact methods to construct accurate counterfactuals for report is based are available on the GEF Evalua- project intervention areas. In the current cases, tion Office Web site (www.gefeo.org). these approaches establish the outcomes had protected area systems not been created. While such quasi-experimental studies are interesting and provide insights into what is happening in a well-structured and rigorous way, the Office's main methodology for tackling impact questions Rob van den Berg is a theory-based approach, which aims to unwrap Director, Evaluation Office v Acknowledgments The Evaluation Office impact evaluation work coauthors were Paul Ferraro, Department of Eco- was coordinated by task manager David Todd, nomics, Andrew Young School of Policy Studies, Senior Evaluation Officer, who has been respon- Georgia State University; and Margaret Holland, sible for further developing the Office's approach Gaylord Nelson Institute for Environmental Stud- to impacts, building on the material reported in ies, University of Wisconsin­Madison. the Annual Impact Report 2007. A separate but related quasi-experimental study in Thailand was conducted by Katharine Sims of For the current report, quasi-experimental work Amherst College. in Costa Rica was conducted by principal author Kwaw Andam, International Food Policy Research The Evaluation Office remains fully responsible Institute, Addis Ababa, Ethiopia. The report's for the contents of this report. vi Main Report 1 Overview of Impact Evaluation and on the Coordinating Committee of the United Work in 2008 Nations Evaluation Group (UNEG); he is also co- chair of the UNEG Task Force on Impact Evalua- The GEF Annual Impact Report 2007 concluded tion. Thus, the Evaluation Office is fully informed that, in its impact evaluation work, the Evalua- regarding current best practice trends in impact tion Office of the Global Environment Facility evaluation, and its experience and products are (GEF) would pursue "a mixed-method approach widely known. that includes macro-level statistical analysis...as well as case studies of projects." This approach has Following up on the initial quasi-experimental been pursued in 2008 through a number of inter- evaluation of the impacts of protected areas on related activities. deforestation in Costa Rica--an effort conducted in collaboration with the GEF's Scientific and Impact evaluation has become a high-profile topic Technical Advisory Panel (STAP)--which was in the international development arena and one reported on in GEF Annual Impact Report 2007, subject to considerable debate. Much of the discus- two new quasi-experimental impact evaluations sion has revolved around the efficacy and accept- have been performed, also in collaboration with ability of different methodological approaches. the STAP. These were conducted by providing The GEF Evaluation Office has been actively limited funds to specialist researchers, enabling engaged in this international debate and is collab- them to analyze existing data sets to explore top- orating in numerous initiatives both to remain on ics relevant to the GEF. In 2008, the following two the cutting edge of the discussion and to share its studies were commissioned and managed by the growing expertise.1 A senior evaluation specialist GEF Evaluation Office; both are now completed: in the Office serves on the Steering Committee of the Network of Networks on Impact Evaluation z Evaluating the Local Socioeconomic Impacts of Protected Areas: A System-Level Compar- ison Group Approach. This study focused on 1 One relevant partner with which intended collab- the protected area system of Thailand, which is oration has not yet been initiated is the United Nations about to receive GEF support. Environment Programme's Early Warning group, which has access to substantial data sets, notably of z Measuring the Social Impacts of Protected satellite imagery and aerial photography, on environ- Areas: An Impact Evaluation Approach. mental change. An agreement to collaborate in devel- oping impact evaluation uses of these data sets will be This study focused on the Costa Rica protected developed during 2009. area system (which has received GEF support) 1 and enabled a comparison with the Evaluation studies of protected area systems noted above. Office 2007 study of avoided deforestation of This third study looks at a completed GEF the same system. project--the Regional Integrated Silvopastoral Approaches to Ecosystem Management Proj- The two studies have provided important insights ect (GEF ID 947)--which was conducted by the on the impacts of protected areas, an area of inves- World Bank in Colombia, Costa Rica, and Nicara- tigation first undertaken by the Office with its gua and had an experimental design featuring par- efforts in East Africa in 2007. One reported find- ticipant and control groups of farmers The case ing of that body of work was the negative socio- study draws on existing research, compiled by a economic impacts experienced by one subgroup doctoral researcher formerly associated with the (the Batwa) in the locality neighboring one pro- GEF Evaluation Office. Limited follow-up field- tected area. This raised the broader issue of the work was conducted to evaluate the strengths and socioeconomic impacts of protected area systems, weaknesses of the project's experimental design a major component of GEF-supported activities and impacts at the field level, including the adop- in the biodiversity focal area. The two studies con- tion of improved silvopastoral practices, environ- ducted in 2008 found the following: mental benefits, socioeconomic benefits, and the z Districts surrounding protected areas in Costa sustainability of land use changes. The case study Rica and Thailand experienced less poverty was completed in November 2008 and will be than carefully controlled counterfactual dis- reported on in the next annual impact report. tricts not adjacent to protected areas with simi- The Office's major impact evaluation activity in lar geographic and physical characteristics. 2008 entailed the methodological development z When these districts were compared with for and initial implementation of an impact evalu- counterfactual districts in similar locations but ation of GEF activities concerning the reduction not adjacent to protected areas, the latter were of ozone-depleting substances. This evaluation even poorer. Thus, proximity to a protected uses a theory-based approach and will include area in fact emerged as having a positive effect extensive statistical analysis of the impacts of GEF on income. activities as compared with those of the Multi- z Income inequality increased near protected lateral Fund of the Montreal Protocol. It entails a areas in Thailand (data on this factor were detailed desk review of all relevant GEF activities not available for Costa Rica), so an aggregate (focusing particularly on terminal evaluations) income improvement may disguise pockets of and the scientific literature, and an analysis of sta- worsening poverty. tistical data available from the Montreal Protocol. Detailed fieldwork will be conducted in Kazakh- Because Costa Rica and Thailand have relatively stan, Russia, Ukraine, and Uzbekistan. During high income levels as compared with most devel- the study preparation, discussions were held with oping countries, and since both already have well- STAP members and with the Evaluation Offices of developed tourism industries, the specific national- GEF partners, as a result of which the evaluation is level findings should not be assumed to apply to being conducted in collaboration with the United protected area systems in other countries. Nations Environment Programme and the United The Evaluation Office is conducting a third case Nations Industrial Development Organization. study in addition to the two quasi-experimental The initial findings of this work will be incorpo- 2 GEF Annual Impact Report 2008 rated into the final report of the GEF Fourth Over- selection of protected area locations, which com- all Performance Study (OPS4) and included in the plicates the construction of a useful comparison GEF Annual Impact Report 2009. group. The extensive development of impact evalua- The approach presented here analyzes a protected tion approaches has been fed into the design of area system across a national or subnational area the methodology for the evaluation of results in with respect to socioeconomic and environmen- OPS4. In addition to including the findings of the tal impacts at the community level. It is applied in protected area and ozone-depleting substances the context of Thailand's national protected area evaluations in the OPS4 results analysis, the system, using data at the subdistrict level from the Office's theory-based approach is being adapted north and northeast regions of the country. To so that it can enable an improved understanding measure socioeconomic outcomes, data are used and reporting of results throughout the GEF port- from new poverty mapping techniques that esti- folio. Theories of change are being developed for mate community-level incomes and poverty rates. all major areas of GEF activity; early testing has To assess impacts, the approach relies on evalu- shown that they facilitate an improved under- ating differences between communities with pro- standing of the sustainability and catalytic effects tected land and comparison communities in the of GEF support after formal project closure. same province or district with a similar likelihood of protection and similar preprotection develop- 2 Case Study of the Social ment potential. The comparison group was con- Impacts of Protected Areas: structed on the basis of an analysis of the history North and Northeast Thailand of protected area designation in Thailand, in order to account for the chief factors that determined This evaluation approach develops and applies a protection and might also influence outcomes. new comparison group­based method for evalu- ating the socioeconomic effects of protected areas The method applied here could productively be on local communities across a protected area sys- used to evaluate protected areas in other countries tem. The approach was designed to extend and or to evaluate impacts of other large-scale envi- complement program evaluation methods previ- ronmental projects supported by the GEF. Ideally, ously developed by the GEF Evaluation Office. it would complement existing studies, including case comparisons or household survey work, by Protected areas, including those supported by providing a broader overview of impacts across a the GEF, now cover a significant fraction of the larger number of sites. global land area. However, little is known about their net effects on local incomes or poverty rates. The results of this case study indicate that pro- Community-level economic development could tected forest areas in north and northeast Thai- be reduced by restrictions on land use or resource land have prevented forest clearing that otherwise extraction activities, but could also be supple- would have occurred and thus have imposed a mented by a new tourism sector or increased constraint on land available for agricultural use. environmental benefits. Empirical work on the Subdistricts with more land in protected areas actual impacts of protected areas has been limited had significantly more forest cover by 2000 than to date by the lack of data on poverty outcomes at did appropriate comparison subdistricts (9 to the appropriate spatial scale and the nonrandom 25 percentage points more for national parks, Main Report 3 and 11 to 32 percentage points more for wildlife 3 Case Study on the Social sanctuaries). Impacts of Protected Areas: The study found that national parks and wildlife Costa Rica sanctuaries, even though they reduce the land This case study evaluated the socioeconomic available for agricultural production, did not harm impacts of Costa Rica's protected area network, average consumption levels or increase poverty a network in which the GEF has invested for rates. Looking only at correlations, subdistricts many years. The study used a quasi-experimen- with more land in protected areas were indeed tal approach to provide estimates of the aggre- substantially poorer than the province averages. gate social impacts of protected areas. It looked However, after controlling for geographic char- to answer the question "What is the effect of this acteristics and preprotection development poten- protected area on economic outcomes within tial, the analysis indicates that this poverty is not neighboring communities?" the result of the protected areas. Subdistricts with To answer this question, the effects of other vari- more land in wildlife sanctuaries did not have sig- ables on the economic outcomes in local com- nificantly different consumption levels or poverty munities affected by protected areas must be headcounts than counterpart comparison sub- isolated. This in turn requires the establishment districts. Subdistricts with more land in national of a counterfactual: "What would have happened parks had significantly higher consumption levels if this protected area had not been established?" (2 to 7 percent) and lower poverty rates (4 to 12 Matching methods provide a way to find suitable percent) than comparison subdistricts. Inequality comparisons for communities affected by protec- measures are higher on average for communities tion, thus establishing the counterfactual. near the national parks, which indicates that a dis- proportionate share of these gains went to higher The study measured the impacts of Costa Rican income households. protected areas established before 1980 on changes in socioeconomic outcomes between The results suggest that, on average, at the com- 1973 and 2000. It used matching methods to iden- munity level, the gains from protection have been tify suitable counterfactuals for protected census sufficient to offset the costs of land use constraints. segments to control for the overt bias from non- The most probable mechanism for the positive eco- random placement of protection. It matched each nomic effect of national parks is increased income area affected by protection with similar unpro- from tourist visits in and near the parks. The Thai tected areas based on relevant preprotection vari- government has actively promoted national parks ables that affect the likelihood of protection as well as tourist destinations, and official statistics cite as changes in socioeconomic outcomes. It also over 10 million tourist visits to national parks in estimated the spatial spillover effects of protection 2000. Consumption levels are positively associated on unprotected areas located near the protected with the popularity of parks as measured by tourist areas and assessed the sensitivity of the results to visits; a higher flow of tourists is a likely explanation various changes in the sample or matching speci- for the stronger positive effects for national parks as fication (see tables A.4 and A.5 in the annex). compared with wildlife sanctuaries, where tourism opportunities are limited. See tables A-1 through The study found no evidence that protected areas A-3 in the annex for a summary of key results. in Costa Rica have had harmful impacts on the 4 GEF Annual Impact Report 2008 aggregate livelihoods of local communities. On the 4 Conclusions on Impact contrary, it found that protection has had positive Evaluation Work in 2008 effects on socioeconomic outcomes. For example, The establishment of protected areas is associated The Evaluation Office has made substantial with a lower poverty index in local communities progress in developing and implementing a affected by protection. It also found that protec- variety of approaches for assessing the impacts tion led to better outcomes in terms of the con- of GEF interventions. The Office is now evalu- dition of housing and access to water supply, but ating the impacts of the GEF program aimed at found no significant differences in other (slightly assisting in the elimination of ozone-depleting higher income level) indicators such as measures substances in countries with economies in tran- of access to electricity or telephones. sition; the findings from this evaluation will be incorporated into OPS4. Additionally, two quasi- Conventional statistical evaluation techniques experimental evaluations have been completed, (such as a difference in means test or ordinary least both addressing an issue of great significance to squares regression) produced biased estimates GEF policy and practice--the socioeconomic when applied to the study sample. In contrast to impacts of protected area projects. The con- the results indicated above, those conventional methods erroneously implied that protection had clusions from these two analyses show that the negative impacts on the livelihoods of local com- most effective evaluative perspective is gained by munities. These findings suggest that conventional combining methodological approaches to ensure methods that fail to control for confounding fac- that both macro- and local-level impacts are tors or outcome baselines can lead to inaccurate accurately assessed. The impact evaluation work estimates. The case study demonstrates the spe- of the Evaluation Office has contributed to and cific value delivered by applying an impact evalu- benefited from substantive engagement in key ation approach, which carefully identifies suitable international forums that are leading the further counterfactuals for measuring the social impacts development and implementation of approaches of protected areas. in the field. Main Report 5 Annex: Selected Social Impacts of Protected Areas in Thailand and Costa Rica Table A.1 Consumption/Poverty Headcount Ratio and Protected Areas: Thailand No Province fixed Slope/elevation Geographic Full Parameter controls effects only controls controls controls Dependent variable: log mean consumption National park (percent) -0.191*** -0.170*** 0.061 0.133*** 0.133*** Standard error (0.044) (0.045) (0.040) (0.037) (0.037) Wildlife sanctuary (percent) -0.278*** -0.217*** -0.000 0.098* 0.106* Standard error (0.069) (0.075) (0.055) (0.055) (0.055) Northeast dummy Yes Yes Yes Yes Yes Province fixed effects No Yes Yes Yes Yes Slope and elevation controls No No Yes Yes Yes Geographic controls No No No Yes Yes Historical forest cover No No No No Yes Adjusted r2 0.143 0.417 0.466 0.570 0.574 N 4113 4113 4113 4113 4113 Dependent variable: log poverty headcount ratio National park (percent) 0.576*** 0.458*** -0.110 -0.251*** -0.251*** Standard error (0.125) (0.099) (0.067) (0.061) (0.062) Wildlife sanctuary (percent) 1.006*** 0.595*** 0.057 -0.124 -0.142 Standard error (0.232) (0.168) (0.129) (0.129) (0.128) Northeast dummy Yes Yes Yes Yes Yes Province fixed effects No Yes Yes Yes Yes Slope and elevation controls No No Yes Yes Yes Geographic controls No No No Yes Yes Historical forest cover No No No No Yes Adjusted r2 0.265 0.616 0.655 0.709 0.711 N 4113 4113 4113 4113 4113 Note: Probability: *** p < 0.01; ** p < 0.05; * p < 0.10. Standard errors are robust, clustered at the district level. Slope and elevation controls = (log of) average slope, average elevation. Geographic controls = (log of) distance to major city, distance to rail line, distance to mineral depos- its, distance to any roads (1962), distance to major roads (1962), max elevation, max slope, distance to national boundary, distance to navigable river; average temperature, average rainfall, ecoregion 2, ecoregion 3, near watershed. Historical forest cover = forest cover in 1973. 7 Table A.2 Additional Socioeconomic Outcomes and Protected Areas: Thailand Province Province Province Province fixed fixed fixed fixed effects Full effects Full effects Full effects Full Parameter only controls only controls only controls only controls Log squared Dependent variable Log poverty gap poverty gap Log gini coefficient Population density National park (%) 0.359*** -0.245*** 0.246*** -0.185*** 0.007 0.060* -170.556*** 15.953 Standard error (0.093) (0.061) (0.078) (0.053) (0.022) (0.033) (33.007) (15.045) Wildlife sanctuary (%) 0.528*** -0.112 0.390** -0.073 -0.023 0.040 -139.317*** 33.692** Standard error (0.167) (0.125) (0.150) (0.117) (0.046) (0.051) (30.673) (15.786) Northeast dummy Yes Yes Yes Yes Yes Yes Yes Yes Province fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Slope and elevation No Yes No Yes No Yes No Yes controls Geographic controls No Yes No Yes No Yes No Yes Historical forest cover No Yes No Yes No Yes No Yes Adjusted r2 0.609 0.684 0.586 0.644 0.455 0.477 0.140 0.346 N 4113 4113 4113 4113 4113 4113 4113 4113 Note: Probability: *** p < 0.01; ** p < 0.05; * p < 0.10. Standard errors are robust, clustered at the district level. Slope and elevation controls = (log of) average slope, average elevation. Geographic controls = (log of) distance to major city, distance to rail line, distance to mineral depos- its, distance to any roads (1962), distance to major roads (1962), max elevation, max slope, distance to national boundary, distance to navigable river; average temperature, average rainfall, ecoregion 2, ecoregion 3, near watershed. Historical forest cover = forest cover in 1973. 8 GEF Annual Impact Report 2008 Table A.3 Forest Cover and Protected Areas: Thailand Ordinary least squares regressions: whole sample Province fixed Slope/elevation Geographic Parameter No controls effects only controls controls Full controls Dependent variable: forest cover, 2000 (percent) National park (percent) 0.805*** 0.667*** 0.197*** 0.171*** 0.171*** Standard error (0.047) (0.043) (0.047) (0.048) (0.042) Wildlife sanctuary (percent) 0.857*** 0.681*** 0.262*** 0.233*** 0.215*** Standard error (0.054) (0.099) (0.062) (0.062) (0.052) Northeast dummy yes yes yes yes yes Province fixed effects no yes yes yes yes Slope and elevation controls no no yes yes yes Geographic controls no no no yes yes Historical forest cover no no no no yes Adjusted r2 0.452 0.636 0.835 0.845 0.866 N 4113 4113 4113 4113 4113 Panel approach Subdistrict fixed effects Ordinary least Subdistrict w/ common Parameter squares (2000) fixed effects First differences Random effects support Dependent variable: forest cover, by year (percent) National park (percent) 0.101** 0.115*** 0.082 0.121*** 0.122*** Standard error (0.050) (0.039) (0.063) (0.038) (0.043) Wildlife sanctuary (percent) 0.114 0.143*** 0.174** 0.130** 0.142*** Standard error (0.094) (0.051) (0.066) (0.052) (0.052) Province fixed effects yes -- -- -- -- Geographic controls yes no no yes no Subdistrict fixed effects no yes no yes yes Period fixed effects no yes yes yes yes Adjusted r2 0.768 0.351 0.132 -- 0.316 N 1386 5473 4089 5473 3677 Note: Probability: *** p < 0.01; ** p < 0.05; * p < 0.10. Standard errors are robust, clustered at the district level. Slope and elevation controls = (log of) average slope, average elevation. Geographic controls = (log of) distance to major city, distance to rail line, distance to mineral deposits, distance to any roads (1962), distance to major roads (1962), max elevation, max slope, distance to national boundary, distance to navigable river; average temperature, average rainfall, ecoregion 2, ecoregion 3, near watershed. Historical forest cover = forest cover in 1973. Panel approach limits observations to those with more than 10 percent of forest cover in 1973, less than 20 percent cloud cover, and less than 20 percent land area in water. Ordinary least squares (2000) repeats the ordinary least squares cross-section specification in the full controls column. Subdistrict fixed effects includes subdistrict and period fixed effects. First differences regresses changes in forest cover on changes in percent protected. Random effects uses random effects estimation including the same additional fixed covariates as the full controls column. Subdistrict fixed effects w/ common support repeats the subdistrict fixed effects specification for the sample with common support (propen- sity score between 0.01 and 0.7). Annex: Selected Social Impacts of Protected Areas in Thailand and Costa Rica 9 Table A.4 Effects of Protected Areas on Socioeconomic Outcomes: Costa Rica Percentage of houses Percentage of households Poverty In bad In Without Without Without Outcome index condition slums telephone electricity water supply Matching estimates (effect of protection on change in outcome 1973­2000) covariate matching ­ Mahalanobis -3.251*** -6.429*** -2.142** -1.032 -1.731 -5.856*** Standard error (0.973) (2.189) (1.064) (2.051) (3.697) (1.652) covariate matching ­ Mahalanobis w/ calipers -1.941*** -4.714** -1.976** -1.782 2.155 -4.201*** Standard error (0.543) (1.489) (0.795) (1.709) (2.772) (1.212) N outside calipers [65] [72] [63] [57] [60] [63] Replicating conventional methods (effect of protection on change in outcome 1973­2000) Ordinary least squares 2.068*** 2.364*** 0.621* 11.243*** 7.354*** -2.622** Standard error (0.403) (0.818) (0.347) (1.462) (2.347) (1.022) Replicating conventional methods (effect of protection on post-protection outcome measured in 2000) Difference in means 9.170*** 6.114*** 0.695** 29.085*** 19.270*** 4.352*** N treated 399 399 399 399 399 399 N available controls (15988) (15988) (15988) (15988) (15988) (15988) Note: Significance: *** at 1 percent; ** at 5 percent; * at 10 percent. Calipers restrict matches to units within 1 standard deviation of each covari- ate. Ordinary least squares model regresses the outcome on protection while controlling for key covariates. Difference in means: A t-test is applied to evaluate the difference in means of post-protection outcomes between treated and control segments. Table A.5 Estimates of the Spillover Effect of Protected Areas on Socioeconomic Outcomes in Neighboring Unprotected Areas: Costa Rica Percentage of houses Percentage of households Poverty In bad In Without Without Without Outcome index condition slums telephone electricity water supply Matching estimates (effect of protection on change in outcome 1973­2000) covariate matching ­ Mahalanobis 0.134 -1.241* -0.282 -0.621 10.071*** -0.725* Standard error (0.258) (0.673) (0.257) (1.165) (1.903) (0.416) covariate matching ­ Mahalanobis w/ calipers 0.147 -1.373** -0.223 -0.654 10.101*** -0.589 Standard error (0.252) (0.665) (0.252) (1.161) (1.894) (0.390) N outside calipers [5] [8] [7] [10] [5] [5] N treated 786 786 786 786 786 786 N available controls (11782) (11782) (11782) (11782) (11782) (11782) Note: Data are matching estimates for the effect of protection on changes in outcome over the period 1973­2000. Significance: *** at 1 per- cent; ** at 5 percent; * at 10 percent. Calipers restrict matches to units within 1 standard deviation of each covariate. 10 GEF Annual Impact Report 2008 GEF Evaluation Office Publications Number Title Year Evaluation Reports 47 Midterm review of the resource Allocation Framework 2009 46 GeF Annual report on Impact 2007 2009 45 GeF country Portfolio evaluation: cameroon (1992­2007) 2009 44 GeF Annual country Portfolio evaluation report 2008 2008 43 GeF country Portfolio evaluation: South Africa (1994­2007) 2008 42 GeF country Portfolio evaluation: Madagascar (1994­2007) 2008 41 GeF country Portfolio evaluation: benin (1991­2007) 2008 40 GeF Annual Performance report 2007 2008 39 Joint evaluation of the GeF Small Grants Programme 2008 38 GeF Annual Performance report 2006 2008 37 GeF country Portfolio evaluation: Samoa (1992­2007) 2008 36 GeF country Portfolio evaluation: the Philippines (1992­2007) 2008 35 evaluation of the experience of executing Agencies under expanded Opportunities in the GeF 2007 34 evaluation of Incremental cost Assessment 2007 33 Joint evaluation of the GeF Activity cycle and Modalities 2007 32 GeF country Portfolio evaluation: costa rica (1992­2005) 2007 31 GeF Annual Performance report 2005 2006 30 the role of Local benefits in Global environmental Programs 2006 29 GeF Annual Performance report 2004 2005 28 evaluation of GeF Support for biosafety 2006 third Overall Performance Study 2005 GeF Integrated ecosystem Management Program Study 2005 biodiversity Program Study 2004 climate change Program Study 2004 International Waters Program Study 2004 Evaluation Documents eD-3 Guidelines for GeF Agencies in conducting terminal evaluations 2008 eD-2 GeF evaluation Office ethical Guidelines 2008 eD-1 the GeF evaluation and Monitoring Policy 2006 Global Environment Facility Evaluation Office 1818 H Street, NW Washington, DC 20433 USA www.gefeo.org