Example of “Random Promotion” or

“Encouragement Design”

Rebecca ThorntonUniversity of Michigan

Randomized Evaluation of HIV Testing

Recall – Promotion Design1. Promotion was random: those who get the

promotion are just like those who didn’t2. Promotion worked: those who get the

promotion actually act on it3. Promotion doesn’t affect outcome

behavior except through the behavior itself

HIV Testing“Taking an HIV test is the single most

influential driver for behavior change” (Global Business Coalition; Richard Holbrooke)

“Making people aware of their HIV status is the only way to make them change their sexual behavior” (Bill Clinton)

55% of the total HIV/AIDS expenditure spent on testing (Mozambique)

Can we justify these financial resources being spent on HIV testing?

Conventional View #1“Once people learn their HIV status, they

will engage in safer sexual behavior”

Why might this be ambiguous for HIV positives?Altruistic towards partner …but safe sex is costly and there is no longer a

return to use protectionWhy might this be ambiguous for HIV

negatives?Increased return to safe sex …but no longer a need to act altruisticallyIndividuals may also believe that they face lower risk

Conventional View #2“It is difficult to get people to learn their HIV

status”

“I cannot accept to be tested. It is better to just live without knowing my status and see if I get sick than to know that I have AIDS in advance because then I cannot be living happily.”

“The cost of stigma is quite high, more than the bus fare to town.”

Research QuestionsHow does the demand for learning HIV

results respond to small changes in costs and benefits?

What is the impact of learning HIV results on sexual behavior?

Research DesignLimitations with existing studies and

challengesSelf-reported behavior

How might this bias the results?Selection

Who chooses to test or learn resultsWhere testing centers are locatedHow might self-selection bias the results?

We want to measure the effect of learning HIV results

Think through your exact question and plan for analysis

Among HIV positives: compare those who know their results with those who don’t

Among HIV negatives: compare those who know their results with those who don’t

Potential for selection bias?What would we randomize?

Research Design

1Got Resultsi i iY

Research DesignHow can we randomize getting HIV

results?

Incentives to learn results (subsidies of price)

Distance to results centers (subsidies of time/transport)

Malawi Panel StudyPanel study (1998, 2001, 2004)

125 rural villages in three districts – why important?

Randomly selected households (one in four)

Women and their husbands, adolescents

2004: Offered free HIV and STI tests93% accepted (2,859) 6.7% HIV Positive

Experimental DesignRandomly assigned voucher to learn HIV

resultsBetween $0-$3.00 (average=$1.04)20% received no incentive

How much were the incentives?Average daily wage: ~$1.00Smallest amount: $0.10

Why randomize by individual?

Experimental DesignResults centers

Randomly placed based on household GPS coordinates and clustered by village

Average distance = 2 km

Experimental Design

Experimental Design

Experimental Design

Percent Learning HIV Results

ReceivedNo Incentive

80%

60%

40%

20%

0Received

Some Incentive

Percent Learning HIV Results

0 0.1 0.3 0.6 1.1 1.6 2.1 2.6

Amount of Incentive (Dollars)

80%

60%

40%

20%

0

-0.2 -0.5 -1.0 -1.5 -2.0 -2.5 -3.0

Percent Learning HIV Results

Distance from the results center

90%

80%

70%

60%

0km 1 km 2 km 3 km 4km

.6.7

.8.9

1

0 1 2 3 4Distance to VCT(KM)

Fan Regression 95% upper band 95% lower band

Experimental DesignHow do we measure the effect of learning HIV

results (going to the tent)?Need a follow-up with outcome variables – like

what?Follow-up survey (2 districts)

Gave $0.30 – why do this?Sold condoms at half retail priceOnly allowed individuals to purchase condoms from the

money they were givenDesign issues

Why have respondents purchase condoms?Why give them any money?Why cap the number of condoms allowed?

Impact of Learning HIV StatusTwo months after HIV results were available

62% reported having sex8% reported purchasing condoms

Sold condoms24% purchased condoms3.7 condoms on average

We want to measure the demand for safe sex. Why might using condom sales be problematic for this?One idea: Distinguish between those that

were sexually active and those that were not

Percent Purchasing Condoms

• HIV Negatives who had sex at Baseline

Did Not Get Results

Got Results

50%

40%

30%

20%

10%

0

Percent Purchasing Condoms

• HIV Positives who had sex at Baseline

Did Not Get Results

Got Results

50%

40%

30%

20%

10%

0

Effects after Learning HIV ResultsHIV positives with a sexual partner:

Significant increase in the likelihood of purchasing subsidized condom

But the number of condoms purchased was small – only 2 additional condoms

No effects among HIV negativesNo effects among those not having sexNo effect on likelihood of sexual activity

How could this apply to MC?Prices of MCPrizesLocation of mobile clinicsInformation – counselors, media

Other ways?

Distribution of Incentives – Day 1

Theoretical and Actual Total Distribution

0.0000

0.0500

0.1000

0.1500

0.2000

0.2500

0 30 60 90 120

150

180

210

240

270

300

Theoretical

Actual

Distribution of Incentives - overall

0.2

.4.6

.8K

ern

al D

ensi

ty

0 1 2 3Total Amount of Incentive (Dollars)

Actual Distribution Theoretical Distribution