1. Sahan Dissanayake, Colby College and Portland State University Abebe Damte Beyene, EfD Ethiopia and EEPFE Randall Bluffstone, Portland State University Zenebe Gebreegziabher, EfD Ethiopia and EEPFE Peter Martinsson, University of Gothenburg Alemu Mekonnen, EfD Ethiopia and EEPFE Michael Toman, World Bank The Opportunity Cost of Carbon Sequestration in Low- Income Countries and Preferences for REDD+ Contract Attributes: A Choice Experiment in Ethiopia

2. Motivation 2 Forests and Deforestation and Degradation Sub-Saharan Africa has the highest deforestation rate in the world 0.8% or about 4 million acreas Deforestation and degradation are responsible for 12% – 20% of emissions (more than all transport) Compared with other abatement options, reduced deforestation and degradation may be cheap

3. Maine Economics Conference 3 Saturday, April 26, 2014 Funded by Bates, Bowdoin, Colby, and the University of Maine Hosted by Colby College in Waterville, Maine If you are interested in presenting a paper and/or serving as a discussant, please e-mail maine.economics.conference@gmail.com by Friday, March 14. Please include the paper abstract for a paper submission. You need not submit a paper in order to serve as a discussant. maine.economics.conference@gmail.com Organizing Committee: Tim Hubbard, Colby College, timothy.hubbard@colby.edu Jim McConnon, University of Maine, mcconnon@maine.edu Paul Shea, Bates College, pshea@bates.edu Dan Stone, Bowdoin College, dstone@bowdoin.edu

4. Outline 4 Forests in the world 5 countries (Brazil, China, Russia, Canada, USA) have 53% of world forests Annual net deforestation about 5.2 million hectares or ≈ 0.14%/year. 13 million hectares deforested per year, almost all in tropical regions.

5. Motivation 5 Forests in the world 5 countries (Brazil, China, Russia, Canada, USA) have 53% of world forests Annual net deforestation about 5.2 million hectares or ≈ 0.14%/year. 13 million hectares deforested per year, almost all in tropical regions. Sub-Saharan Africa has the highest deforestation rate 0.8% per year (about 4 million hectares annually.) About 30% of developing country forests are CCFs.

6. Motivation 6 Forests in the world 5 countries (Brazil, China, Russia, Canada, USA) have 53% of world forests Annual net deforestation about 5.2 million hectares or ≈ 0.14%/year. 13 million hectares deforested per year, almost all in tropical regions.

7. Motivation 7 Forests in the world Sub-Saharan Africa has the highest deforestation rate 0.8% per year (about 4 million hectares annually.) About 30% of developing country forests are CCFs.

8. Motivation 8 Deforestation and GHGs Deforestation and degradation are responsible for 12% – 20% of emissions (more than all transport). Compared with other abatement options, reducing deforestation and degradation may be cheap.

9. Forests and GHGs Motivation 9 McKinsey and Co. (2009)

10. Motivation 10 Community Controlled Forests and REDD+ The UN FCCC program Reduced Emissions from Deforestation and Degradation (REDD+) may provide an opportunity to improve livelihoods in low-income countries. Community controlled forests (CCFs) make up about 18% of world forests, 25% to 30% of developing country forests.

11. Motivation 11 Forests and GHGs McKinsey and Co. (2009)

12. Motivation 12 Community Controlled Forests and REDD+ Often forests are de jure government owned, but community controlled forests (CCFs) make up about 18% of world forests, 25% to 30% of developing country forests and much more in low-income countries. Black carbon (second to CO2), partly emitted by small biomass stoves mainly in Asia and Africa may be responsible for as much as 10% of anthropogenic climate change. Much of this wood is collected in CCFs. The FCCC program Reduced Emissions from Deforestation and Degradation (REDD+) may provide an opportunity to improve livelihoods in low-income countries.

13. Motivation 13 Community Controlled Forests and REDD+ Community controlled forests (CCFs) make up about 18% of world forests, 25% to 30% of developing country forests. The UN FCCC program Reduced Emissions from Deforestation and Degradation (REDD+) may provide an opportunity to improve livelihoods in low-income countries.

14. Motivation 14 REDD REDD stands for: Reducing Emissions from Deforestation and forest Degradation REDD+ (or REDD plus or REDD-plus): Extends REDD by Sustainable Forest Management Conservation of Forests Enhancement of carbon sinks REDD++: Extends REDD+ by Low-carbon but high biodiversity lands

15. Motivation 15 REDD+ Contracts Pilot REDD+ projects starting throughout the world. The preferences and attitudes of local community participants need to be better understood to create effective contracts. The opportunity cost of contracts is not known. Not much work done on REDD+ contracts for community managed forests.

16. Research Questions 16 We focus on REDD+ contracts for CCF What are the respondent’s preferences towards the institutional structure of REDD+ contracts and contract attributes? What is the opportunity of cost of REDD+ contracts? Analyze the tradeoffs between REDD+ contract attributes. Understand the heterogeneity in preferences across socio-demographic variables.

17. NON-MARKET METHODS Valuation of non-market goods and services Methods: Revealed Preference Uses existing data from a related market for values No direct link to market, might not capture full value Stated Preference Elicit values from the public using surveys Will people actually make the payments? Memory is faulty, we don’t remember what we think we do? Does Non-Market valuation consider the functional values (?)

18. Why do we need to estimate ecosystem services benefits? 18 To support decision on public spending on conservation To compare and prioritize different projects To consider the public's values and preferences To maximize benefits per dollar spent

19. Valuation of non-market goods and services 19 Methods: Revealed Preference Uses existing data from a related market for values No direct link to market, might not capture full value Cannot value new and hypothetical goods/services Stated Preference Elicit values from the public using surveys Will people actually make the payments? Memory is faulty, we don’t remember what we think we do?

20. Valuation of non-market goods and services 20 Methods: Revealed Preference Travel Cost Method Uses cost incurred in travels as measure of value Price Hedonics Method Assumes that the value of non-market goods are capitalized in to the market goods Stated Preference Contingent Valuation Method Asks questions about value Choice Experiment Asks respondents to make choice amongst bundles

21. Several methodologies: all based on attempting to draw a demand for a good or service that does not have a market. Valuation of non-market goods and services P Q(0,0) Demand = WTP (WTP = Willingness to Pay) D

22. Methods – Choice Experiment Survey 22 CE Surveys are a stated preference survey method used to assess individuals’ preferences for specific goods, services or policies. Why a CE Survey? CE surveys allow partworth utilities and tradeoffs between attributes to be calculated. Based on Lancaster’s (1966) consumer theory: Consumers derive utility not from goods themselves but rather from the attributes or characteristics that the goods possess.

23. Methods – Choice Experiment Survey 23 Choice Experiment Surveys A CE survey typically consists of a set of choice questions Individuals choose between two or more hypothetical bundles representing the good or policy. The characteristics (or attributes) of each bundle change A sample survey from Jan et. al (2000)

24. Methods – Survey Design Identifying Attributes Informal focus groups Discussion with researchers Discussions with community members Formal focus groups (15 sets) Final list of attributes – REDD + payments (per household per month) – Portion of the REDD+ payment going to the community. – Term of REDD+ commitment – Reduction in amount of fuel wood collected – Grazing land reduced Measure of opportunity cost 24

25. Methods – Experimental Design 25 The Experimental Design With 4 attributes with 3 levels and a cost with 6 levels 3 4 *3 4 *6*6= 236,196 possible choice combinations An orthogonal fractional factorial design Includes the interaction of attributes 84 unique choice sets Sample of Survey Design Full factorial design Fractional factorial design that is balanced and orthogonal 3 attribute, 2 level example (2 3 =8 possibilities) Profile Number Attribute Levels Attributes Full factorial design Main effects for attribute A Interaction effects for attribute AB

26. Methods – Experimental Design 26 The Experimental Design With 4 attributes with 3 levels and a cost with 6 levels 3 4 *3 4 *6*6= 236,196 possible choice combinations An orthogonal fractional factorial design Includes the interaction of attributes 84 unique choice sets Sample of Survey Design Profile Number Attribute Levels Attributes

27. Methods – Survey Design 27 The Experimental Design An orthogonal fractional factorial design Includes the interaction of attributes 84 unique choice sets The Survey Instrument A block design blocks of 6 choice profiles 14 unique surveys The final survey instrument 7 sets of binary choice questions Attribute non-attendance questions a demographic questionnaire

28. Methods – Survey Design 28 The Survey Instrument

29. Methods – Data Collection 29 Data Collection (In-Person Survey) Sampling procedure (for main survey) 110 CPR community survey sites where EEPFE collected data in 2012. We removed 15 of these sites (covered during the REDD+ FGD) and all CPR community survey sites from Tigrai. So we have 84 CPR community survey sites in our sampling frame.

30. Methods – Data Collection 30 Data Collection (In-Person Survey) Sampling procedure (for main survey) 110 community survey sites where EEPFE collected data in 2012. We removed 15 of these sites (covered during the REDD+ FGD) from Tigrai. So we had 84 community survey sites in our sampling frame.

31. Methods – Data Collection 31 Data Collection (In-Person Survey) Sampling procedure (for main survey) 110 community survey sites where EEPFE collected data in 2012. We removed 15 of these sites (covered during the REDD+ FGD) from Tigrai. So we had 84 community survey sites in our sampling frame from Amhara, Oromia and SNNP. We picked 36 sites using stratified-proportionate random sampling. Sampling frame: Master list of households in a Got obtained from the Kebele/PA administration. Systematic random sampling applied in selection of sample households.

32. Sampling 3 Randomization of sample sites into 6 treatment groups We used variables/indicators wealth variance in FUG, existence of forest rules and regulations, and % biomass change over 5 years Not much variation among the sample villages considered especially in relation to the forest related indicators. So, we considered mainly distance to market and % of HHs with access to piped water as criteria for the grouping.

33. Methods – Data Collection 33 Data Collection (In-Person Survey) Sampling procedure (for main survey) From these remaining 84 sites, we picked 36 at random using stratified- proportionate random sampling technique. We proportionately distributed the 36 sites to the three regions involved: Amhara ≈20%, Oromia ≈50%, and SNNP ≈30% (based on forest cover).

34. Methods – Survey Design 34 Data Collection (In-Person Survey) Sampling procedure (for main survey) We carried mean comparison tests (Kruskal Wallis Test) No significant difference across the groups for the most relevant variables. Only the variable % of HHs with access to piped water significant for some of the groups

35. Sampling 5 We discovered injera is not common in Borena So, we had to do re-grouping of sample sites into six groups after replacing the three sample sites from Borena. We used same lottery method to replace these sites from FUGs in Oromia and tested. Still no significant difference across the groups especially for the most relevant variables.

36. Methods – Data Collection 36 Sampling procedure Selection of sample households From these remaining 84 sites, we picked 36 in Amhara, Oromia and SNNP using stratified-proportionate random sampling. Got (local name for sub-Kebele/PA) represents a site where sample households are selected from. Sampling frame: Master list of households in a Got obtained from the Kebele/PA administration. Systematic random sampling applied in selection of sample households.

37. Methods – Model Estimation 37 The respondent chooses between two options, A and B and get the following utility, U, where V is the indirect utility and the error terms are drawn independently from a fixed distribution Respondent choosing A over B implies We estimate the probability of Assuming an Independent and Identically distributed (IID) extreme type I distribution, Gumble distribution, we use a logit model for the estimation

38. Methods – Model Estimation 38 The respondent chooses between two options, A and B and get the following utility, U, where V is the indirect utility and the error terms are drawn independently from a fixed distribution Respondent choosing A over B implies We estimate the probability of Assuming an Independent and Identically distributed (IID) extreme type I distribution (Gumble distribution) we can use a logit model for the estimation

39. Methods –Model- Estimation 39 Homogeneous utility for alternative j with k attributes Can calculate the marginal WTP/WTA for attributes (  k /  p ) Utility for individual q choosing alternative j with k attributes Models preference heterogeneity; deals well with repeated choices CONDITIONAL LOGITMIXED MULTINOMIAL LOGIT

40. Methods –Model- Estimation 40 Homogeneous utility for alternative j with k attributes The marginal value of attribute k is the ratio between the parameter  k and -  p. Utility for individual q choosing alternative j with k attributes Models preference heterogeneity; deals well with repeated choices CONDITIONAL LOGITMIXED MULTINOMIAL LOGIT

41. Methods – Model Estimation 41 Model Specifications (10)(11)(12) Attribute interaction terms Regional interaction terms Maine Effects

42. Results 42 Maine Effects

43. Results 43 Marginal Willingness to Accept Respondents are willing to accept ≈160 birr ($8) to reduce grazing by 10% ≈180 birr ($9) to give 10% more to the community

44. Results – Interaction Effects 44 There is a non-linear relationship between grazing restriction and the payment variable. Compared to respondents from Oromia, respondents from Amhara require a higher payment when a larger portion of the payment is given to the community. Compared to respondents from Oromia, respondents from SNNP require a higher payment to reduce the amount of grazing.

45. Results and Discussion 45 We find that the opportunity cost of the REDD+ contracts are low Firewood reduction is not significant Enforcement not likely? Heterogeneity due to availability of substitutes? Latent class models?

46. Results and Discussion 46 We find that the opportunity cost of the REDD+ contracts are low Firewood reduction is not significant Enforcement not likely? Heterogeneity due to availability of substitutes? Need to analyze individual level differences in preferences

47. THE END 47 Acknowledgements World Bank, EfD- Ethiopia, EEPFE, and EfD Comments from seminar participants at University of New Hampshire Thank you for listening! Questions ? If you have questions or suggestions Email:sdissan2@gmail.com Web: http://sahan.org/sdissan2@gmail.comhttp://sahan.org/