The Illusion of Sustainability Michael Kremer Edward Miguel

The Illusion of Sustainability Michael Kremer Edward Miguel Atheendar Venkataramani – ECON 730 a

Background Recent interest in sustainable development projects Creation/promotion of projects with one time/limited subsidies that can continue without external support n Community “ownership” Does it work? Is this possible? n n n Microfinance Cost-sharing in public health Other examples

Objectives of this paper Study sustainability in the context of deworming project Strategies to promote sustainability n n User fees Health education – move beyond focus on drugs to changing behaviors Diffusion of worm prevention information and behaviors through peer effects/social networks* Verbal commitment intervention

Relevant Background – Primary School Deworming Project conducted in Busia district, Kenya. n Recall from Miguel and Kremer [2004] – sizeable social benefits, ¾ of which were due to externalities Randomized phase in: Group 1 schools treated 1998 -2001, Group 2 schools - 1999 -2001 and Group 3 – starting in 2001. Aspects of program relevant for this study: n n n Health education efforts at treatment schools – preventative behaviors Of the 50 Group 1 and 2 schools, half were randomized to cost-sharing program in 2001. Randomizing schools to treatment generates variation in knowledge of treatment within an individual’s given social network; knowledge prior to inception of program in 1998 likely minimal.

Peer Effects and Technology Adoption: Theory Model of information spread and take up of a new technology in a social network Basic idea: people adopt deworming if expected private benefits exceed expected cost. People have differing priors over the effect of deworming drugs and are heterogeneous with respect to tastes and preferences for the regimen. Model includes four types of peer effects: n n Pure imitation Information on how to use technology effectively Provide information on benefits of technology Modification of disease environment (treatment externalities)

Peer Effects and Technology Adoption: Theory The first RHS term incorporates individual’s beliefs about drug effectiveness, level of infection, and specific taste for deworming. Second RHS term reflects financial, time, or utility costs of treatment. Final term is the desire to imitate social context*share of social contacts who took up drug in previous period. Peer effects influence beliefs about effectiveness, infection burden, costs, and utility from imitation.

Peer Effects and Technology Adoption: Theory Before technology is introduced, individuals have priors over effectiveness. Heterogeneity – people can be optimistic or pessimistic relative to actual effectiveness. This may vary with level of schooling of individual, “traditional” thinking Updating priors and learning: n n Own experience – individuals learn something through experience. Other shocks to health noisy signals Social networks - the network is infinite and single path connects any two nodes. individuals have m direct links, and each of those links have n direct links, etc. Probability of information transfer is p each period. People send information at beginning of period and receive messages at end of period from their contacts.

Peer Effects and Technology Adoption: Theory Steady State n n n As long as some fraction of individuals adopt technology, information will eventually diffuse through the network. Thus, in the steady state beliefs about effectiveness and costs faced will converge to “true” values. Two cases: No imitation effects: unique equilibrium exists Imitation effects: multiple steady states possible if imitation contribution to utility sufficiently large. n Implications for subsidies Subsidizing a small number of people will learn about returns as well as how to best use technology enough to ensure widespread long-run adoption of technologies with positive private returns.

Peer Effects and Technology Adoption: Theory Information transfer: probability that signal is transmitted from j th order link: So, the direct impact of an additional signal acquired by a jth order link on take-up is: The second piece above can be nonzero depending on how the signal impacts beliefs, cost of take up, infection status, and so forth.

Peer Effects and Technology Adoption: Theory Beliefs Use of technology n C(. ) is a decreasing function of total signals ever received about technology (i. e. C’(. ) < 0, C’’(. ) > 0, C(0) > 0) Treatment externalities from social contacts Impact of early treatment links on expected private benefits to adoption:

Peer Effects: Empirical Analysis Basic idea: are households with more social links to randomly chosen treatment schools more likely to take deworming drugs, conditional on total social contacts? Data: n n Surveyed representative subsample of parents with children in Group 2 and 3 schools. Information on five friends, five relatives with whom they speak frequently, as additional social contacts whose kids go to primary school and individuals with whom they talk about child health issues. Information on deworming status of social links’ children as well as health effects, frequency of interaction with links, respondent knowledge of public health practices. Second order link measures from 40 parents in each Group 2 and Group 3 school. Construction of average number of links parents have to early (Group 1 and 2) and late (Group 3) treatment schools measurement error?

Peer Effects: Empirical Analysis Probit Model First term is vector of social links to early treatment schools. Includes both first order and higher order links. Second term is total number of social links. Third term is randomized cost-sharing project. Fourth term is a vector of household socioeconomic characteristics (education, assets), demographic characteristics (fertility), group and community fixed effects. Additional specifications include interactions with main treatment variable. Checking identification: direct contacts with treatment schools, cost sharing variables, membership in group 2 generally uncorrelated with vector of observables.

Peer Effects: Empirical Analysis

Peer Effects: Empirical Analysis Main results: n n n Each additional direct link is associated with 3. 1 percentage point decline in probability of receiving deworming regimen. Implies that majority of effect due to treatment externalities or learning about effectiveness. Social effects are stronger for respondents with more education more educated were more optimistic about benefits prior to program? Social contacts of children also appear to be salient. Stronger for adolescents? Higher order contacts appear to have as strong an impact as direct contacts, though third order contacts do not.

Peer Effects: Further Issues Are negative social network effects due to treatment externalities or learning about drug? Latter seems to be dominant: n n n Having direct social links to early treatment schools is associated with lower rates of moderate-heavy infection, but the effect is not statistically significant Prior infection status does not seem to be associated with take-up (not necessary causal, however) At best, the externality pathway would account for 15% of reduction in take-up due to an additional social contact in early treatment school. Endogeneity of social networks: Were networks measured in 2001 affected by program? n Bias would likely be positive (health conscious parents would find health conscious friends) Direct evidence of learning about effectiveness:

Peer Effects: Simulation Use school to school connection matrix and estimated first order effects to simulate take-up gains along transition path to steady state. Unlike deworming, the technology under consideration provides true private benefits which are larger than most individual’s priors diffusion of knowledge speeds adoption. Assume: n Health benefits * idiosyncratic utility uniformly distributed n Everyone in a given school has same prior; priors differ across schools n All social effects from learning about benefits of technology. Find: n Beliefs about technology and take-up rates converge quickly, even when signals have high variance n Optimal seeding does not yield additional benefits n Large additional subsidies provide little additional benefit to take-up; so in cases where difference in (expected) social benefits is negative, sustainability might not be possible

Cost Sharing

Impact of Health Education

Verbal Commitment Strategy

Findings User fees, even very small ones, lead to sizeable drops in treatment rates. Additional social links to individuals in treatment schools (random assignment of treatment schools source of identification here) associated with decreased rates of take-up. n Peer effects of learning about the benefits of technology outweigh those of imitation and learning how to use it. Suggest that large ongoing subsidies might be required to sustain high-take up for technologies with large positive externalities. Broader lesson: “It may be difficult for external interventions to promote sustainable voluntary local public good provision. ”