250 likes | 398 Views
Research Ethics. Kenny Ajayi October 6, 2008. Global Poverty and Impact Evaluation. Econometric Resources. Duflo, Esther, Rachel Glennerster, and Michael Kremer. Using Randomization in Development Economics Research : A Toolkit. Poverty Action Lab White Paper, MIT Wooldridge, Jeffrey.
E N D
ResearchEthics Kenny Ajayi October 6, 2008 Global Poverty and Impact Evaluation
Econometric Resources • Duflo, Esther, Rachel Glennerster, and Michael Kremer. • Using Randomization in Development Economics Research: A Toolkit. Poverty Action Lab White Paper, MIT • Wooldridge, Jeffrey. • Introductory Econometrics: A Modern Approach
Research Ethics • Ethics • Group Projects • Guest Presentation: Robert Van Buskirk
Ethical Concerns • Fairness (in resource allocation) • Dangers (of unsuccessful programs)
Dupas, 2006 • HIV information for Kenyan teenagers • How effective is the current government program? • How effective would additional information be? • How responsive are teenagers to information on relative risks?
Dupas, 2006 • HIV information for Kenyan teenagers • National HIV Prevention Curriculum: • Abstinence • Average HIV prevalence in the population • Teacher reinforcement training • NGO Relative Risks Campaign: • Information on HIV prevalence disaggregated by gender and age group • Presentation by trained NGO officer
Dupas, 2006 • Statistics Provided by the NGO:
Dupas, 2006 • Statistics Provided by the NGO:
Dupas, 2006 • Statistics Provided by the NGO:
Model: Risk Information • Risks • HIV infection for those who are not infected • Pregnancy for girls • Reduction Strategies • Abstinence • Condom use
Model: Partner Choice • Adult Men differ from Teenage Boys • Higher HIV Prevalence Rate • Greater Financial Resources • Derive relatively less utility from condom-protected sex • Girls choose sexual partners based on: • Perceived riskiness of partnership • Size of compensation transfer
Information Experiment • Provide Relative Risks Information in a randomly selected subset of schools • Treatment Cohort: 13,000 • 8th graders at time of campaign (2004) • Treatment Students: 2,500 • Students enrolled in treatment schools during campaign • Comparison Cohort: • Students a year ahead of or below treatment cohort
Empirical Strategy • Compare outcomes (Y) one year later • Self-reported sexual behavior • Childbearing and marital status • Characteristics of sexual partners of girls who had started childbearing • Unobserved (Z):
Empirical Strategy • Compare outcomes (Y) one year later • Self-reported sexual behavior • Childbearing and marital status • Characteristics of sexual partners of girls who had started childbearing • Unobserved (Z): HIV status of teenagers in sample
Empirical Strategy • Compare outcomes (Y) one year later • Self-reported sexual behavior • Childbearing and marital status • Characteristics of sexual partners of girls who had started childbearing • Unobserved (Z): HIV status of teenagers in sample • How reliable are these outcome measures?
Ethical Concerns • If we give people information on HIV, they respond with their behavior • Unethical to withhold information that we know is right, if it can be implemented. but • Unethical to give out information that we are not sure is right, or may be misleading
Ethics of NOT Evaluating • Girls prefer to have sexual relationships with older men because these men can provide greater resources. • Girls might substitute away from sexual partnerships with older men if they knew about disaggregated HIV prevalence instead of just average prevalence rates.
Results • Significant Impacts • Increase in reported sexual activity • 65% decrease in incidence of pregnancies by adult partners • Decrease in HIV infection risk • Larger behavioral changes on the intensive margin (choice of partner and protection level) than on the extensive margin (choice of abstinence or not)
Results • Spillover Effects • (using variation in density of treated students across secondary schools)
Ethics of Randomization • Deny control group the intervention, but: • Don’t actively hurt them • Don’t give wrong information • Don’t make them worse off than they would otherwise be • Can give a token gift/compensation (but small enough not to make it a treatment in itself)
Ethics of Randomization • Ethics are less contested if: • Budget constraints would have prevented everyone from receiving the intervention anyway • Everyone eventually receives the intervention (i.e. phased-in rollout so that the control group is only left out initially) • Program provides basic treatment (or current best practices) for everybody
Solutions to Ethical Concerns • Study Design • Phase-in (everybody eventually gets treatment) • Lottery (fair means to address budget constraints) • Encouragement (not a direct intervention) • Cluster (often less controversial than randomizing within clusters) • Human Subjects Protocol
Conclusion • Ethics are important to consider even in information campaigns (and especially in HIV studies) • There are ways to address ethical concerns creatively