1. Effects of Social Network Structure on Diffusion and Adoption of Agricultural Technology: Evidence from Rural Ethiopia Yasuyuki Todo, Petr Matous (The University of Tokyo) Dagne Mojo Yadate (Ethiopian Institute of Agricultural Research) Tokyo Workshop on International Development October 15, 2012

2. Literature on Adoption of Agricultural Technologies in Less Developed Countries Social networks have a positive effect. – Foster and Rosenzweig (1995) – Conley and Udry (2010) But not always – Munshi (2004) – Duflo, Kremer and Robinson (2004) – Bandiera and Rasul (2006) 2

3. Literature on Social Network Studies The structure of social networks matters to knowledge diffusion. – Watts and Strogatz (1998), Nature Small-world networks  farther and quicker diffusion Similar to role of bridging/external ties (Burt, 1992) – Centola (2010), Science Social experiments creating an online “health forum” Clustered networks  more effective in adoption of behaviors (reinforcement from multiple sources) 3

4. 4 Bridging/external tie with agricultural extension agents Clustered networks in the community Research Questions Which type is more important in diffusion and adoption of agricultural knowledge? Do trust and frequent meetings enhance effects of social networks? (strong ties) New technology innovated outside

5. 5 Highland Altitude: 2500m Midland 2200m Lowland 1800m Data Survey to 297 households in 3 villages in Central Ethiopia: Stratified from high-, mid-, and low- land areas Panel data for 3 years: Dec 2009-Jan 2010, Jan-Feb 2011, Feb 2012 Data for 208 households available Mostly data for latest 2 years are used.

6. 6 Extension center HH who don’t meet extension agent often HH who meet extension agent often Detailed geographic data using GPS devices Pathways of enumerators

7. Social Experiment Donation of mobile phones to winners of a lottery in May 2011 Charge the donated phones every month until Dec 2011 7 All participants Participants in the sample Phone with SMS8761 Phone without SMS8772 Nothing9875 Total272208 Use random distribution of mobile phones as instruments for social networks

8. 8 Teaching how to use mobile phones to winners

9. Innovation and Extension of Agricultural Technologies in Ethiopia 9 Extension agents (DAs) in villages Pilot farmers Research Centers in local districts Knowledge flow of new tech. Farmers Ethiopian Institute of Agricultural Research

10. 2 Agricultural Technologies Examined Row planting – Plant seeds in rows: very simple – Width between rows = 20 cm Compost – More complicated – Requires transfers from a hole to another twice (once every three weeks) – Some inappropriate materials (e.g., leftover foods) Both are encouraged by extension agents. 10

11. Diffusion and Adoption of Knowledge 11 Don’t know Know but don’t use Know and use

12. 12 No Yes Specific Networks with Extension Agents (2012) Trust: Can borrow 500 birr ($30) from the agent

13. Social Networks with Relatives and Friends First-name method (McCarty et al., 1997) 1.Randomly select 39 male and 39 female names from the list of HH heads of the villages 2.From the top of the list of 78 names, ask the respondent whether s/he “knows” a person with the name. 3.Stop asking when the # of known names reaches 14. 4.Ask the relation with, distance to, frequency of meetings with, and religion and ethnicity of each of the 14. 5.Ask if 14 randomly selected pairs of the persons the HH knows know each other (network density) 13

14. Measure of the Size of Personal Social Networks 14 # of names the respondent knows among the first 14 names on the list N = 208 Mean = 8.79 S.D. = 2.26

15. Size of Personal Networks Based on Trust 15 # of names the respondent knows and can borrow 500 birr ($30) from among the first 14 names on the list N = 208 Mean = 3.47 S.D. = 2.94

16. Size of Personal Networks Based on Frequent Meetings 16 # of names the respondent knows and meet at least once 2 weeks among the first 14 names on the list N = 208 Mean = 5.97 S.D. = 2.81

17. (B) Poorly clustered network (A) Highly clustered network Clustered Networks

18. Level of Network Clustering 18 # of pairs who know each other among the randomly chosen 14 pairs N = 208 Mean =11.07 S.D. = 2.63

19. Estimation Procedure Estimate effects of social network variables on knowing and using each technology Linear probability model Identification strategy – Alleviate endogeneity of social networks using random distribution of mobile phones and lagged network variables as instruments in GMM – Distribution of mobile phones: ↑ social networks, but no effect on diffusion/adoption  valid instruments – All other independent variables are 1-year lagged. 19

20. Effect of Mobile Phones 20 (1)(2)(3)(4)(5)(6)(7) Size of networks Clustering level Size of networks based on trust Size of networks based on meetings Knowing an agent Trusting an agent Meeting an agent Mobile-0.0171-0.2920.1680.04910.01580.127*0.0122 (0.314)(0.310)(0.435)(0.393)(0.0517)(0.0653)(0.0772) SMS0.644*0.00293-0.5940.04550.02380.0391-0.0326 (0.341)(0.339)(0.470)(0.423)(0.0557)(0.0709)(0.0837) Observations208 207208 R-squared0.3310.5170.2410.3260.0440.2610.173 (8)(9)(10)(11) Knowing row planting Using row planting Knowing compost Using compost Mobile 0.08460.0559-0.04920.00569 (0.0725)(0.0916)(0.133)(0.116) SMS 0.00447-0.04270.181-0.132 (0.0794)(0.0977)(0.139)(0.125) Observations 15816576111 R-squared 0.0320.0840.1200.096

21. Table 3: Determinants of knowing row planting conditional on not knowing it in the previous year 21 (1)(2)(3)(4)(5)(6) When network variables are … Networks based on knowing Network based on trust Network based on frequent meetings Size of personal networks -0.0281-0.02490.00981-0.05640.02760.0521 (0.0311)(0.0334)(0.0253)(0.0585)(0.0281)(0.0471) --- * network clustering -0.0004310.00613-0.00248 (0.00246)(0.00467)(0.00376) Tie with extension agent 1.289*1.265*0.587*0.610*-0.434-0.389 (0.779)(0.744)(0.305)(0.316)(0.550)(0.524) Years of schooling -0.00287-0.002520.003920.004380.0168*0.0169* (0.0154)(0.0149)(0.00979)(0.00980)(0.00971)(0.00949) Log of cultivated land 0.01910.0198-0.0117-0.0126-0.0226-0.0222 (0.0574)(0.0572)(0.0542)(0.0546)(0.0688)(0.0671) Observations 158 Other controls: distance, dummies for myopia, ethnicity, religion, village ***: p < 1%, **: p < 5%, *: p < 10%

22. Table 3: Determinants of knowing row planting conditional on not knowing it in the previous year 22 (1)(2)(3)(4)(5)(6) When network variables are … Networks based on knowing Network based on trust Network based on frequent meetings Size of personal networks -0.0281-0.02490.00981-0.05640.02760.0521 (0.0311)(0.0334)(0.0253)(0.0585)(0.0281)(0.0471) --- * network clustering -0.0004310.00613-0.00248 (0.00246)(0.00467)(0.00376) Tie with extension agent 1.289*1.265*0.587*0.610*-0.434-0.389 (0.779)(0.744)(0.305)(0.316)(0.550)(0.524) Years of schooling -0.00287-0.002520.003920.004380.0168*0.0169* (0.0154)(0.0149)(0.00979)(0.00980)(0.00971)(0.00949) Log of cultivated land 0.01910.0198-0.0117-0.0126-0.0226-0.0222 (0.0574)(0.0572)(0.0542)(0.0546)(0.0688)(0.0671) Observations 158 Other controls: distance, dummies for myopia, ethnicity, religion, village ***: p < 1%, **: p < 5%, *: p < 10% Knowing an agent is the major determinant.

23. Table 4: Determinants of using row planting conditional on not using it in the previous year and knowing it this year 23 (1)(2)(3)(4)(5)(6) When network variables are … Networks based on knowing Network based on trust Network based on frequent meetings Size of personal networks -0.0449-0.0378-0.01430.0272-0.0289-0.0389 (0.0328)(0.0363)(0.0356)(0.0664)(0.0251)(0.0448) --- * network clustering -0.00118-0.003680.00101 (0.00298)(0.00548)(0.00337) Tie with extension agent -1.193-1.0680.08480.09760.03160.0253 (1.448)(1.351)(0.355)(0.354)(0.382)(0.390) Years of schooling 0.02800.02620.0178*0.0175*0.01650.0164 (0.0186)(0.0174)(0.0104) (0.0101)(0.0100) Log of cultivated land 0.243***0.244***0.203***0.207***0.213***0.211*** (0.0719)(0.0730)(0.0620)(0.0636)(0.0570) Observations 164 165 Other controls: distance, dummies for myopia, ethnicity, religion, village ***: p < 1%, **: p < 5%, *: p < 10%

24. Table 4: Determinants of using row planting conditional on not using it in the previous year and knowing it this year 24 (1)(2)(3)(4)(5)(6) When network variables are … Networks based on knowing Network based on trust Network based on frequent meetings Size of personal networks -0.0449-0.0378-0.01430.0272-0.0289-0.0389 (0.0328)(0.0363)(0.0356)(0.0664)(0.0251)(0.0448) --- * network clustering -0.00118-0.003680.00101 (0.00298)(0.00548)(0.00337) Tie with extension agent -1.193-1.0680.08480.09760.03160.0253 (1.448)(1.351)(0.355)(0.354)(0.382)(0.390) Years of schooling 0.02800.02620.0178*0.0175*0.01650.0164 (0.0186)(0.0174)(0.0104) (0.0101)(0.0100) Log of cultivated land 0.243***0.244***0.203***0.207***0.213***0.211*** (0.0719)(0.0730)(0.0620)(0.0636)(0.0570) Observations 164 165 Other controls: distance, dummies for myopia, ethnicity, religion, village ***: p < 1%, **: p < 5%, *: p < 10% Once farmers know row planting, they use it when their farmland is large.

25. Table 5: Determinants of knowing compost conditional on not knowing it in the previous year 25 (1)(2)(3)(4)(5)(6) When network variables are … Networks based on knowing Network based on trust Network based on frequent meetings Size of personal networks -0.00211-0.0470-0.0394-0.3100.0191-0.143 (0.0627)(0.0533)(0.0553)(0.196)(0.0350)(0.0990) --- * network clustering 0.00947**0.0238*0.0142* (0.00459)(0.0143)(0.00760) Tie with extension agent 0.598-0.1310.678**0.761*0.2030.446 (0.839)(0.931)(0.344)(0.394) (0.370) Years of schooling 0.01880.05670.01610.0446***0.0300*0.0496*** (0.0261)(0.0381)(0.0188)(0.0172) (0.0154) Log of cultivated land -0.00892-0.104-0.0658-0.145-0.0573-0.124 (0.0924)(0.119)(0.104)(0.122)(0.0935)(0.0957) Observations 76 Other controls: distance, dummies for myopia, ethnicity, religion, village ***: p < 1%, **: p < 5%, *: p < 10%

26. Table 5: Determinants of knowing compost conditional on not knowing it in the previous year 26 (1)(2)(3)(4)(5)(6) When network variables are … Networks based on knowing Network based on trust Network based on frequent meetings Size of personal networks -0.00211-0.0470-0.0394-0.3100.0191-0.143 (0.0627)(0.0533)(0.0553)(0.196)(0.0350)(0.0990) --- * network clustering 0.00947**0.0238*0.0142* (0.00459)(0.0143)(0.00760) Tie with extension agent 0.598-0.1310.678**0.761*0.2030.446 (0.839)(0.931)(0.344)(0.394) (0.370) Years of schooling 0.01880.05670.01610.0446***0.0300*0.0496*** (0.0261)(0.0381)(0.0188)(0.0172) (0.0154) Log of cultivated land -0.00892-0.104-0.0658-0.145-0.0573-0.124 (0.0924)(0.119)(0.104)(0.122)(0.0935)(0.0957) Observations 76 Other controls: distance, dummies for myopia, ethnicity, religion, village ***: p < 1%, **: p < 5%, *: p < 10% Clustered networks  knowledge diffusion (reinforcement from multiple sources?)

27. Table 7: Determinants of the Number of Information Sources 27 (1)(2) Row plantingCompost Size of personal network based on knowing * level of network clustering0.03690.0506** (0.0333)(0.0207) Observations119109 Other controls: education, land area, distance, dummies for myopia, ethnicity, religion, village (***: p < 1%, **: p < 5%, *: p < 10%) Clustered networks  ↑ number of information sources for complicated technologies

28. Table 5: Determinants of knowing compost conditional on not knowing it in the previous year 28 (1)(2)(3)(4)(5)(6) When network variables are … Networks based on knowing Network with trust Network based on frequent meetings Size of personal networks -0.00211-0.0470-0.0394-0.3100.0191-0.143 (0.0627)(0.0533)(0.0553)(0.196)(0.0350)(0.0990) --- * network clustering 0.00947**0.0238*0.0142* (0.00459)(0.0143)(0.00760) Tie with extension agent 0.598-0.1310.678**0.761*0.2030.446 (0.839)(0.931)(0.344)(0.394) (0.370) Years of schooling 0.01880.05670.01610.0446***0.0300*0.0496*** (0.0261)(0.0381)(0.0188)(0.0172) (0.0154) Log of cultivated land -0.00892-0.104-0.0658-0.145-0.0573-0.124 (0.0924)(0.119)(0.104)(0.122)(0.0935)(0.0957) Observations 76 Other controls: distance, dummies for myopia, ethnicity, religion, village ***: p < 1%, **: p < 5%, *: p < 10% Networks with an agent based on trust  knowledge diffusion

29. Table 5: Determinants of knowing compost conditional on not knowing it in the previous year 29 (1)(2)(3)(4)(5)(6) When network variables are … Networks based on knowing Network with trust Network based on frequent meetings Size of personal networks -0.00211-0.0470-0.0394-0.3100.0191-0.143 (0.0627)(0.0533)(0.0553)(0.196)(0.0350)(0.0990) --- * network clustering 0.00947**0.0238*0.0142* (0.00459)(0.0143)(0.00760) Tie with extension agent 0.598-0.1310.678**0.761*0.2030.446 (0.839)(0.931)(0.344)(0.394) (0.370) Years of schooling 0.01880.05670.01610.0446***0.0300*0.0496*** (0.0261)(0.0381)(0.0188)(0.0172) (0.0154) Log of cultivated land -0.00892-0.104-0.0658-0.145-0.0573-0.124 (0.0924)(0.119)(0.104)(0.122)(0.0935)(0.0957) Observations 76 Other controls: distance, dummies for myopia, ethnicity, religion, village ***: p < 1%, **: p < 5%, *: p < 10% Education  knowledge diffusion

30. Table 6: Determinants of using compost conditional on not using it in the previous year and knowing it this year 30 (1)(2)(3)(4)(5)(6) When network variables are … Networks based on knowing Network with trust Network based on frequent meetings Extent of networks -0.0540-0.0516-0.0216-0.0291-0.0487-0.104 (0.0419)(0.0500)(0.0426)(0.139)(0.0382)(0.0726) --- * density -0.0002100.0005460.00489 (0.00331)(0.0103)(0.00576) Tie with extension agent 0.6070.593-0.293-0.279-0.521-0.397 (1.215)(1.210)(0.402)(0.444)(0.463)(0.407) Years of schooling 0.002510.002760.02730.02750.01940.0202 (0.0328)(0.0332)(0.0176)(0.0167)(0.0148)(0.0146) Log of cultivated land 0.0511*0.0515*0.0479*0.0476-0.00142-0.00190 (0.0300)(0.0309)(0.0290)(0.0295)(0.0467)(0.0448) Observations 111 Other controls: land area, dummies for myopia, ethnicity, religion, village ***: p < 1%, **: p < 5%, *: p < 10%

31. Table 6: Determinants of using compost conditional on not using it in the previous year and knowing it this year 31 (1)(2)(3)(4)(5)(6) When network variables are … Networks based on knowing Network with trust Network based on frequent meetings Extent of networks -0.0540-0.0516-0.0216-0.0291-0.0487-0.104 (0.0419)(0.0500)(0.0426)(0.139)(0.0382)(0.0726) --- * density -0.0002100.0005460.00489 (0.00331)(0.0103)(0.00576) Tie extension agent 0.6070.593-0.293-0.279-0.521-0.397 (1.215)(1.210)(0.402)(0.444)(0.463)(0.407) Years of schooling 0.002510.002760.02730.02750.01940.0202 (0.0328)(0.0332)(0.0176)(0.0167)(0.0148)(0.0146) Distance from training center 0.0511*0.0515*0.0479*0.0476-0.00142-0.00190 (0.0300)(0.0309)(0.0290)(0.0295)(0.0467)(0.0448) Observations 111 Other controls: land area, dummies for myopia, ethnicity, religion, village ***: p < 1%, **: p < 5%, *: p < 10% Once farmers know compost, they use it when they are remotely located.

32. 32 Networks with extension agents promote knowing simple technologies Conclusions Clustered networks promote knowing complicated technologies Diffusion: Social networks are important. Adoption: Social networks are less important. Information flows from multiple sources are needed (similar to Centola 2010) Trust is needed for complicated technologies