Schooling, Income, and HIV Risk: Some insights from cash transfer experiments

Transcription

1 Schooling, Income, and HIV Risk: Some insights from cash transfer experiments Berk Özler University of Otago and The World Bank Please do not cite without explicit permission from the

2 Outline 1. Income effects on sexual behavior and HIV risk. q Extant literature 2. Effects of CCTs and in-kind transfer programs on sexual behavior and HIV risk. q Extant literature 3. Lessons from Schooling, Income, and HIV Risk (SIHR) project in Malawi q Some new, longer-term findings 4. Thoughts on the role of cash transfers in HIV prevention.

3 Why cash transfers for HIV prevention? Three main approaches to HIV prevention: q Biomedical, q Behavioral, q Structural Cash transfers seem to be slotted into the structural bin, although this categorization is not a clean one

4 1. INCOME EFFECTS ON SEXUAL BEHAVIOR AND HIV

5 Why cash transfers for HIV prevention? Lengthy literature on women s poverty and HIV, which, until recently, was mostly based on observational studies: q Poor women are more likely to: Engage in risky sex; have multiple partners; have riskier types of sex for money Experiencing coerced sex; having partners with multiple partners q But, same literature also found: A positive correlation between household wealth and HIV prevalence Higher income households more likely to suffer an adult death

6 Why cash transfers for HIV prevention? If there are economic reasons behind risky sexual behavior, it stands to reason that there may be income effects on HIV/STI risk. q Long literature on transactional sex q Sex workers accept compensating differentials for unprotected sex There is even a literature on income shocks and onset of marriage in developed countries Two recent papers shed more light on this taking advantage of (quasi-)experimental data q Kohler and Thornton (2011); Robinson and Yeh (2011)

7 Why cash transfers for HIV prevention? Kohler and Thornton (2011) find divergent responses between men and women to a relatively large one-time cash transfer in Malawi (not an income shock as the transfer is, more or less, anticipated): q Within one week of receiving the transfer, men increase sexual activity (and unprotected activity) while women reduce it. Some issues with attrition; self-reported outcomes q Wallet to purse policies (like those in the U.K. back in the day )

8 Why cash transfers for HIV prevention? Robinson and Yeh (2011) found that a sample of commercial and informal sex workers in Western Kenya substantially increased their supply of risky sex to cope with unexpected health shocks, particularly the illness of another HH member. q They interpret their findings to indicate an inability to smooth consumption using other means q They find that these women establish relationships with regular clients, who send them transfers in response to negative income shocks (Robinson and Yeh, 2012) à sexual relationships as insurance to cope with risk.

9 2. EFFECTS OF CCTS AND IN- KIND TRANSFERS ON SEXUAL BEHAVIOR AND HIV

10 Why have CCTs for HIV prevention? If the problem is credit constraints (lack of money), cash transfers (or insurance or access to credit) are likely to be first-best responses. Why would we want to attach conditions to cash transfers for HIV prevention? q The answer, as for the same question in other areas such as schooling or poverty reduction, is when there are market failures or externalities

11 Why have CCTs for HIV prevention? Two kind of inefficiencies can be invoked in the case of HIV: q Private inefficiencies: Hyperbolic discounting; Lack of information; persistent misguided beliefs; failures to update Intra-household bargaining problems q Social inefficiencies Spillover effects

12 Why have CCTs for HIV prevention? In such cases, cash incentives can help more than cash with no strings attached q even though the first-best solution might be an intervention that directly tackles the problem. Provision of information (Dupas, 2011) has highlighted the reality that current behavior change interventions, by themselves, have been limited in their ability to control HIV infection in women and girls in low- and middle-income countries. (McCoy, Kangwende, and Padian, 2009)

13 Why have CCTs for HIV prevention? Again, two recent CCTs (with the conditionality being HIV or STI tests) have been experimentally evaluated: q Malawi Incentive Program (Kohler and Thornton, 2011) Found no effect on any sexual behavior of offering one lumpsum transfer to maintain HIV status for one year. They speculate as to the reasons for lack of impact (I suspect, lack of insurance) q RESPECT (Tanzania; de Walque et al., 2012) Found a 25% reduction on treatable STIs (after 12 months) However, no effect on other STIs, or in earlier periods Presumptive treatment could account for some of the effects

14 Why have CCTs for HIV prevention? Evidence from these CCT experiments is weak at best for the time being. q We can t even talk about cost-effectiveness yet without efficacy first While evidence that economic circumstances (and negative shocks) having something to do with risky sexual behavior seems stronger. It s possible that the infrequent nature of transfers was the culprit for the lack of effect in these two studies.

15 Why have CCTs for HIV prevention? Before I segue into SIHR in Malawi, a quick word about one other intervention: q Duflo, Dupas, Kremer (2012): 2x2 factorial study design: free school uniforms and teacher training on HIV/AIDS curriculum Subsidies reduced early marriage and pregnancy, but no effect on HIV or HSV-2 Teacher training had no effect on any of these outcomes Combined treatment reduced HSV-2 but not HIV q Authors interpret their results to be consistent with a model, where teen pregnancy in committed relationships is desirable for those not staying in school

16 3. ZOMBA CASH TRANSFER PROGRAM: SAMPLING AND SURVEY DESIGN

17 Schooling, Income, and HIV Risk Conditional Cash Transfers for schooling can increase income and school enrollment simultaneously. Zomba Cash Transfer Program (ZCTP) was a two-year randomized intervention that provided cash transfers to schoolgirls (and young women who had recently dropped out of school) to stay in (and return to) school. To our knowledge, this is the first study to assess the impact of a cash transfer program for schooling on the risk of HIV infection.

18 Malawi Zomba, a district in Southern Malawi, is where our study is located. At baseline (2007), education levels were very low and HIV rates very high: q prevalence among old women is 9.1%, compared to 2.1% among males of the same age. q HIV prevalence among women is 24.6%, compared to a national average of 13.3%.

19 Sampling and Survey Design 3,796 young women were sampled from 176 enumeration areas (EAs) in Zomba, a district in Southern Malawi. EAs randomly drawn from three strata: urban, near rural, and far rural. All households in each sampled EA were listed using two forms, then the sample selected from the pool of eligible young women.

20 Sampling and Survey Design Eligibility into the program was defined as follows: q q Baseline dropouts: unmarried girls and young women, aged 13-22, already out of school at baseline (<15% of the target population), AND Baseline schoolgirls: unmarried girls and young women, aged 13-22, who can return to Standard 7-Form 4, enrolled in school at the time of their first interview. Otherwise, there was no targeting of any kind. The surveys employed at baseline and at follow-up are comprised of two parts:

21 Sampling and Survey Design Part I is administered to the HH head, and collects information on the following: q household roster, q dwelling characteristics, q household assets and durables, q consumption (food and non-food), q household access to safety nets & credit, and q shocks (economic, health, and otherwise) experienced by the household q mortality

22 Sampling and Survey Design Part II is administered to the core respondent, who provides further information about her: q family background, q Education, labor market participation, time allocation q health and fertility, q dating patterns, detailed sexual behavior at the partnership level, q knowledge of HIV/AIDS, q social networks, q own consumption of girl-specific goods (soaps, mobile phone airtime, clothing, braids, handbags, etc.).

23 Biomarker data for STIs Self-reported data on sexual behavior may be unreliable: q q Understatement of sexual activity will cause attenuation bias, If correlated with treatment status, misreporting will bias the impact estimates. 18 months after the start of the intervention, VCT teams visited a randomly selected 50% of the panel sample at their homes. (Refusal rate: 3%, attrition rate: 1%) q Why no baseline? q Why only 50%? Rapid tests for HIV, HSV-2, and Syphilis.

24 Timeline Baseline data collection: Sep 2007 Jan Cash Transfers begin: February 2008 Follow-up data collection: Oct 2008 Feb Biomarker data collection: Jun-Sep Cash Transfer Program ends: December 2009.

25 Zomba Cash Transfer Program Design The 176 EAs were evenly split into treatment and control. Treatment EAs were further split to receive conditional (CCT) and unconditional (UCT) treatment. All baseline dropouts received CCTs. Transfers were split between parents and the girls. Parents received $4-10 per month while the girls received $1-5. q The average total transfer of $10/month falls well within the range of CCT programs around the world and is significantly lower than a pilot cash transfer program in Malawi, financed by the Global Fund and supported by UNICEF.

26 Balance in randomization and attrition The treatment and control communities look very similar across a whole host of baseline characteristics, implying that the randomized allocation of treatment units was carried out successfully. Attrition in longitudinal data was low (<7%) and equal across treatment and control areas. See Baird et al. (2012) for more details.

27 18-MONTH IMPACTS ON SEXUALLY TRANSMITTED INFECTIONS

28 Program impacts on STIs HIV and HSV- 2 prevalence among baseline schoolgirls at 18- month Intervention Group Comparison Group Unadjusted odds ratio Adjusted odds ratio HIV 7/490 (1.2%) 17/799 (3.0%) 0.39 ( ) 0.36 ( ) HSV- 2 5/488 (0.7%) 27/796 (3.0%) 0.23 ( ) 0.24 ( ) Syphilis 1/491 (0.8%) 4/800 (0.7%) 1.20 ( ) 0.92 ( )

29 Summary of impacts on HIV and HSV-2 (during the program) 18 months after its inception, the program decreased the prevalence of each of HIV and HSV-2 by 64% and 76%, respectively, among baseline schoolgirls. However, it had no effect on those who had already dropped out of school at baseline. Baseline dropouts account for less than 15% of the target population of year-old never-married females in Zomba.

30 ONE YEAR IMPACTS: SEXUAL BEHAVIOR (PLUS SUB-GROUP ANALYSIS)

31 Program impacts on Sexual Behavior (self-reported) Unadjusted odds ratio Adjusted odds ratio New sexual debut 0.59 ( ) 0.64 ( ) Unprotected sexual intercourse 1.28 ( ) 1.08 ( ) Weekly sexual intercourse 0.44 ( ) 0.46 ( ) Had a sexual partner 25 or older 0.20 ( ) 0.21 ( )

32 One-year impact on sexual activity, baseline schoolgirls Self- reported Sexual Activity Status at 12- month Follow- up Control Treatment ALL Never active Newly active Stops being active Always active ALL

33 HIV Prevalence by sexual activity status at follow-up, baseline schoolgirls HIV Prevalence by Self- reported Sexual Activity Status Control Treatment ALL Never active 2.1% 1.2% 1.7% Newly active 2.2% 1.7% 2.0% Stops being active 1.3% 1.0% 1.2% Always active 12.3% 1.0% 7.1% ALL 3.0% 1.2% 2.2% 33

34 Program impacts on sexual behavior among always-active baseline schoolgirls number of lifetime partners Panel A: Self-reported sexual behavior number of weeks in a relationship (past 12 months) total number of sex acts (past 12 months) total number of unprotected sex acts (past 12 months) Combined intervention ** 5.545* * * (0.113) (3.016) (7.907) (6.640) Change in outcome (control) N.A. N.A. N.A. Mean of outcome (control) Number of observations

35 Figure 6: Number of weeks in a sexual relationship during the past 12 months number of weeks in a relationship in the past 12 months Control Treatment Note: Number of weeks in a relationship in the past 12 months can exceed 52 weeks due to concurrent partnerships. 35

36 Program impacts on partner selection among always-active baseline schoolgirls Panel B: Partner characteristics partner is in school partner is 25 or older partner has been tested for HIV total value of cash and gifts from partner (past 12 months) Combined intervention *** 0.175** (0.121) (0.055) (0.086) (24.717) Change in outcome (control) N.A. N.A. N.A. Mean of outcome (control) Number of observations

37 Age of Partners, by Treatment Status Always Active Schoolgirls Male HIV rate by Age, from DHS Age of male partners Control Male HIV rate (right scale) Treatment 37

38 Summary The program effects were concentrated among a small group, who became sexually active before the program and were active during the program ( always active ). The program impact on HIV was not only due to a reduction in the number of lifetime sexual partners and unprotected sexual activity, but may also owe, in some part to a significant increase in partner s safety. One last question remains: q Is it a schooling effect or an income effect?

39 RESULTS: IMPACTS ON SCHOOLING, MARRIAGE, AND PREGNANCY

40 Program impacts on schooling: Enrollment Panel B: Program impacts on teacher-reported school enrollment Dependent variable: =1 if enrolled in school during the relevant term (1) (2) (3) (4) (5) (6) (7) (8) Year1: 2008 Year2: 2009 Total Year 3: Terms 2010 Term 1, Term1 Term2 Term3 Term1 Term2 Term3 (6 terms) Postprogram Conditional treatment 0.043*** 0.044*** 0.061*** 0.094** 0.132*** 0.113*** 0.535*** 0.058* (0.015) (0.016) (0.018) (0.041) (0.035) (0.039) (0.129) (0.033) Unconditional treatment ** * (0.015) (0.017) (0.023) (0.038) (0.037) (0.039) (0.136) (0.036) Mean in the control group Number of observations 2,023 2,023 2, Prob > F(Conditional=Unconditional) Not for citation without explicit permission from the 40

41 Summary of program impacts on schooling 1. While there was a modest decline in the dropout rate in the UCT arm in comparison to the control group, it was only 43% as large as the impact in the CCT arm. 2. Among those enrolled in school, there is some evidence of higher attendance in the CCT arm. 3. Finally, the CCT arm also outperformed the UCT arm in tests of English reading comprehension. q à It is fair to conclude that CCTs outperformed UCTs in terms of improvements in schooling outcomes. Not for citation without explicit permission from the 41

42 Program impacts on marriage and pregnancy Table VII: Program Impacts on Marriage and Pregnancy Dependent variable: =1 if ever married =1 if ever pregnant (1) (2) (3) (4) Conditional treatment (0.012) (0.024) (0.014) (0.027) Unconditional treatment ** *** *** (0.012) (0.022) (0.017) (0.024) Number of observations 2,087 2,084 2,086 2,087 Mean in control Prob > F(Conditional=Unconditional) Not for citation without explicit permission from the 42

43 How do we reconcile the differential program impacts on schooling, marriage, and pregnancy? Girls in the CCT arm are less likely to drop out of school than those in the UCT arm, but are also more likely to be ever married or pregnant at the end of the two-year intervention. q Bit of a head-scratcher q Existing evidence from sub-saharan Africa suggests that reducing school dropout should lead to declines in teen marriage and pregnancy rates (Duflo, Dupas, and Kremer 2010; Ozier 2010; Ferré 2009, Osili and Long 2008). Not for citation without explicit permission from the 43

44 Marriage and Enrollment at Follow-up Table VIII: Prevalence of Being Ever Married by School Enrollment Status during Term1, 2010 Enrolled Not enrolled Total (1) (2) (3) Control 1.7% 46.9% 19.9% (row %) (59.8%) (40.2%) (100.0%) Conditional treatment 0.5% 50.8% 16.0% (row %) (69.2%) (30.8%) (100.0%) Unconditional treatment 0.3% 25.2% 10.1% (row %) (60.5%) (39.5%) (100.0%) Total 1.1% 44.2% 17.2% (row %) (62.7%) (37.3%) (100.0%) Not for citation without explicit permission from the 44

45 Summary of impacts on schooling vs. marriage and fertility The CCT arm had a significant edge in terms of schooling outcomes over the UCT arm: a large gain in enrollment and a modest yet significant advantage in learning. However, the improvement in the CCT arm was achieved at the cost of denying transfers to non-compliers, who are shown to be particularly at risk for early marriage and teenage pregnancy. The story seems to be the same when it comes to HIV and HSV-2 risk Not for citation without explicit permission from the 45

46 Is the impact through increased schooling or income? Prevalence of HIV by treatment and enrollment status in 2008 Not enrolled Enrolled Total Control Group 8.5% 1.8% 2.8% (row %) UCT treatment 0.0% 1.9% 1.7% (row %) Total 6.1% 1.5% 2.1% (row %)

47 Longer-term impacts on HIV and HSV-2

48 Longer-term impacts on HIV and HSV-2

49 Is the impact through increased schooling or income? The findings are consistent with a significant income effect on sexual behavior and risk of HIV infection during the program. This finding is supported by similar effects on marriage and pregnancy in the UCT arm likely due to effects among those dropping out of school during the program. However, the income effects are fleeting: treatment and control converge within two years of the end of the program. This is not only true for HIV and HSV-2, but also for ever married, ever pregnant, and total number of (live) births. The positive income shock may have empowered young women to make safer choices while they had a steady income q By choosing safer partners and by decreasing the intensity of sexual activity with those partners.

50 Is the impact through increased schooling or income? We rule out increased school enrollment being the main channel for the reduction in sexually transmitted infections. Mainly because the schooling effects were limited (in absolute terms) even in the CCT arm. Not limited to this study, but common across many studies q This does not mean that schooling does not have an effect. In fact, it likely DOES have a substantial effect. q q The question is whether we can (feasibly) target groups among whom the impacts would be very large. Baseline dropouts are one such group, with large and lasting program impacts on schooling, marriage, pregnancy, and total fertility, but with NO effects on HSV-2 or HIV (despite high incidence rates in control)

51 4. IMPLICATIONS FOR THE POTENTIAL OF CASH TRANSFER PROGRAMS FOR HIV PREVENTION

52 New studies are investigating this question UNC study in South Africa (CCTs for schooling with multiple conditions) JHU study in Tanzania (combination intervention with biomedical, behavioral, and structural components) q Incorporation of cash transfers into HIV prevention programs by medical researchers is welcome. q However, the fact that both studies compare a package treatment to control is, IMHO, a significant limitation: We need more studies with factorial designs (say, combining increased non-hiv life expectancy with repeated testing)

53 Cash Transfers for HIV Prevention? It is unlikely that governments or NGOs will run cash transfer programs solely for HIV prevention: q Cost-effectiveness needs to be established compared to not only other HIV prevention efforts, but also to other health interventions. q Perhaps, existing programs can be tweaked with respect to the target populations to leverage greater gains for HIV prevention. We also need to think about whether the effects are fleeting or sustained if programs are short-term

54 Cash Transfers for HIV Prevention? However, it is possible to envision cash transfers being part of larger HIV prevention efforts: q In such efforts, cash transfers can serve not only as insurance by providing a small but steady and reliable source of income, but can also serve as incentive to protect oneself and others from infection. Examples exist in other areas, such as smoking among pregnant women. One could also think of subsidies for VMMC, PrEP, and other medical treatments, for which subsidies may be justified. Such evaluations should exploit factorial designs and incorporate epidemiological models and cost-benefit analysis into RCTs

55 Other breakthroughs in HIV prevention? In this lecture, I tried to fill some of the gaps identified by Heise et al. (JIAS 2013) with respect to the mechanisms through which cash transfers may influence sexual behavior and HIV risk. q The recommendations for future studies are broadly consistent with theirs. Ultimately, efforts in other areas either directly related to HIV prevention (test and treat; vaccines; microbicides) or large increases in QALYs in developing countries may be more effective than what can be done with small cash transfers. q Hopefully, such developments can put us out of this line of work

56

57 Zomba Cash Transfer Research Design Malawi Research Design: Treatment EAs (N=88) Control EAs (N=88) Baseline Dropouts (N=805) Conditional (N=46) T1 CCT T2 Unconditional (N=27) T1 CCT T2a S2 T2b S2 Baseline Schoolgirls Within- Uncon- Within- (N=2,741) CCT village ditional village control CT control T1 Only (N=15) T1 CCT S2 Withinvillage control C1 Pure control C2 Pure control Individual transfer randomized at individual level EA treatment saturation randomized Household transfer randomized at EA level Not for citation without explicit permission from the 57

58 58