Well-being, Preferences and Basic Income Support: Experimental Evidence from HIV/AIDS Patients
Last registered on December 24, 2018


Trial Information
General Information
Well-being, Preferences and Basic Income Support: Experimental Evidence from HIV/AIDS Patients
Initial registration date
December 21, 2018
Last updated
December 24, 2018 4:34 AM EST

This section is unavailable to the public. Use the button below to request access to this information.

Request Information
Primary Investigator
Erasmus University Rotterdam
Other Primary Investigator(s)
PI Affiliation
Erasmus University Rotterdam
Additional Trial Information
On going
Start date
End date
Secondary IDs
Background: In the context of a 3ie funded impact study assessing the use of SMS reminders in improving retention and adherence to antiretroviral therapy, we descriptively examined the stability of preferences (time, risk, altruism) and subjective wellbeing among people living with HIV in Burkina Faso. More specifically, from the larger impact study we randomly selected a sub-sample of 400 individuals to participate in a phone survey measuring preferences (incentivized with telephone credit). Of the 400 initially sampled individuals, 336 completed four rounds of choice experiments covering altruism, risk and time preferences along with further survey questions.

Research Aim: We are registering a trial as a follow-up to the above, descriptive study. We study the impact of cash transfers on preferences as well as subjective well-being. Some patients were randomly chosen to receive unconditional cash transfers over a 6 month period (treatment group 1 receives monthly transfers, treatment group 2 receives one equivalent large transfer at the end). We will follow up with two surveys to assess impacts. The baseline survey is the last round of the previous descriptive study. We register the intervention and study before the collection of follow-up data. Our main research question is: What are the impacts of unconditional cash transfers on preferences and well-being indicators? Relatedly, are a regular or a one-time transfer equally impactful?
External Link(s)
Registration Citation
Rieger, Matthias and Natascha Wagner. 2018. "Well-being, Preferences and Basic Income Support: Experimental Evidence from HIV/AIDS Patients." AEA RCT Registry. December 24. https://www.socialscienceregistry.org/trials/3634/history/39555
Experimental Details
We randomly split our sample of 336 individuals (who previously completed four rounds of phone surveys measuring preferences, December 2016 to December 2017) into three groups of individuals:

T1 (80 individuals): Subjects receive a basic unconditional income of 4000 FCFA (~19.04 PPP US$) every month between September 2018 to February 2019 for an entire period of 6 months. Based on the 2017 GNI per capita of Burkina Faso of 1,650.10 PPP$, this corresponds to roughly 13.85% of monthly GNI per capita. Since most of our study participants reported little income in previous surveys (65% did not report any salaried work), the effective income support will be perceived as even higher.

T2 (80 individuals): Subjects receive one payment of 24,000 FCFA in January 2019, equivalent in size to the 6 monthly disbursements of 4000 FCFA reiceved by individuals in Group 1.

Control (176 individuals): Subjects do not receive any payment

All payments are disbursed by phone as mobile money.

The intervention follows a larger impact study that has been registered under ISRCTN16558614. However the unconditional cash transfer aspect is a new intervention and the 2 rounds of related follow-up data have not yet been collected.
Intervention Start Date
Intervention End Date
Primary Outcomes
Primary Outcomes (end points)
Hypothesized impacts of the cash transfers are noted in brackets.

Primary Outcome Family 1 (Preferences): We will estimate effects on four preference indicators and also report average standardized effects.

a) Risk taking (+)

b) Patience (+)

c) Dictator game: Anonymous donations to other, undisclosed HIV patients (in % of total) (+)

d) Dictator game: Anonymous donations to HIV support program (in % of total) (+)

e) Average standardized effect a) - d)

Primary Outcome Family 2 (Well-Being): We will estimate effects on two well-being outcomes.

a) Subjective health (+)

b) Happiness (+)

c) Average standardized effect a) - b)
Primary Outcomes (explanation)
Our primary outcomes are informed by our main research question: What are the impacts of unconditional cash transfers on preferences and well-being? Relatedly, are a regular or a one-time transfer equally impactful?

Hypothesized impacts of the cash transfers are noted in brackets above.

Variable definitions:

Primary Outcome Family 1:

All measures are incentivized with cell phone credit in local money (FCFA). Money is paid out after the survey and transferred directly to the mobile phone (except if opted for a late payment in b)).

a) The participant can invest an endowment (500 FCFA) into a risky option and a safe option. The money in the risky option is tripled or lost depending on the flip of a coin. Our outcome is the fraction invested into the risky option.

b) The participant decides between 500 FCFA today and some increased amount in one week from now. Six such choices are made (with increasing pay-outs in a week’s time). Pay-out was determined by chance with the throw of a dice. As our primary outcome we pick the choice in row 6 (500 FCFA today vs. 800 FCFA in one week) and create a binary measure of patience.

c) The participant receives 500 FCFA and can decide to donate anonymously to another, undisclosed HIV patient. The variable used is the % donated of the original 500 FCFA.

f) The participant receives 500 FCFA and can decide to donate to an HIV support program. The variable used is the % of the total endowment that is donated.

Primary Outcome Family 2 (both used as running variables in the analysis):

a) How do you judge your health today on a scale from 1 to 5? (5=Very Good…1=Very Bad).

b) How often do you feel happy in your daily life? (5=Very often…1=Never)
Secondary Outcomes
Secondary Outcomes (end points)
We will also estimate effects on secondary outcomes:
a) Liquidity (2 measures)
b) Current working status
c) Cognition (2 measures)
d) Real effort time allocation
e) Survey-based measures of preferences
Secondary Outcomes (explanation)
The secondary outcomes are meant to uncover some of the mechanisms underlying impacts (if any) on the primary outcomes.
a) We will look at impacts on two questions (used as running variables in the analysis):
i. Do you currently have more cash than usual? (5=Much more than usual….1=Much less than usual).
ii. Do you expect to have more money than usual in the next 7 days? (5=Much more than usual….1=Much less than usual).
These question are meant to test the immediate impact of the treatment itself (akin to a ‘first stage’). Average impacts and q-values will also be provided (see below).

b) Binary question on working status: Do you currently work? (yes=1, no=0)

c) We will test if the income shock has a positive impact on two simple cognitive tasks.
i. Forward digit span test (up to 8 numbers). The outcome will be coded as the highest number of recalled digits.
ii. 10 simple +/- calculations. Total number of problems solved.
Average impacts and q-values will also be provided (see below).

d) At the end of the survey, the participants are told that 10 out of 300 respondents will be asked at random to answer some more qualitative questions (without pay or incentive money). If picked they can respond to a shorter qualitative survey of 15 min tomorrow or 30 min in two weeks. Respondents are asked what they prefer (binary outcome).

e) We will contrast impacts on preferences using experimental measures against survey-based measures of risk, time and altruism. Average impacts and q-values are also provided for this group (see below). These questions are defined as follows:

i. Please tell me, in general, how willing or unwilling you are to take risks. (0 =completely unwilling to take risks…. 10=very willing to take risks).

ii. How willing are you to give up something that is beneficial for you today in order to benefit more from it in the future? (0=Completely Unwilling - 10=Very Willing to Do So)

iii. I have the habit to postpone tasks even if I know that it would be better to do them now (0= not at all….10= absolutely yes)

iv. Are you willing to donate to good causes (church, mosque, community) without anything in return? (0= completely unwilling…10= very willing)
Experimental Design
Experimental Design
Individuals were randomized into one of two treatments or the control group. Baseline data have been collected for all indicators in the Primary Outcome Family 1 and two indicators in Family 2. Two rounds of data collection are planned for January and July 2019.

Modelling Strategy:

We will measure linear intent-to-treat models (i denotes individuals and t time where 0 is the baseline survey prior to the intervention and 1,2 correspond to the planned survey rounds 1 and 2).

Model 1 – For outcomes for which baseline and two follow-up data are available (t=0,1,2)
a) Y_it = c + aT1_it + bT2_it + error_it
b) Y_it = c + aT1_it + bT2_it + round_t + error_it
c) Y_it = c + aT1_it + bT2_it + round_t + c’ X_i + error_it
d) Y_it = c + aT1_it + bT2_it + round_t + k_i + error_it

where T1 and T2 are time-varying treatment status indicators, round_t are a set of survey round dummies, X_i are baseline covariates (see list in section randomization method and associated balancing table), k_i individual fixed effects. The sensitivity of point estimates will be gauged as the model is build up from a) to d).

Extensions: We will also provide combined impacts of T1 and T2 (collapsing both indicators). And in a dynamic analysis we will provide specific impacts by follow-up round 1 and round 2 by adding an interaction term of treatment with survey round dummies (in model 1a). Standard errors will be clustered at the individual level. Another extension considered is to regress changes between follow-up (1 or 2) and baseline on treatment dummies, with and without baseline covariates.

Model 2 – For outcomes for which only follow-up data are available (t=1,2), i.e. all expect one of the secondary outcomes:
a) Y_it = c + aT1_i + bT2_i + error_it
b) Y_it = c + aT1_i + bT2_i + round_t + c’ X_i + error_it

where X_i are baseline covariates (see list in section randomization method).

Extensions: We will also provide combined impacts of T1 and T2 (collapsing both indicators). And in a dynamic analysis we will provide specific impacts by round. Standard errors will be clustered at the individual level when both rounds are pooled, and otherwise robust.

No heterogeneity analysis is envisaged.

We will check for attrition by regressing drop-out from the study on the treatment status, as well as baseline co-variates (see list below used for covariate balancing) on attrition. In case attrition is indeed systematic and substantial (>10%), we employ attrition-robust bounds (for instance Lee bounds).

Multiple hypothesis testing:
We will give p-values and false discovery rate corrected q-values (Benjamini & Hochberg,1995) within each family of indicators.

Average standardized effects:
We will present average effects on standardized indicators by indicator family to avoid cherry picking (Kling, Liebman, and Katz 2007).
Experimental Design Details
Not available
Randomization Method
Covariate-based randomization using a computer was employed based on: age, marital status, being a member of the main ethnicity Mossi, being Muslim, not having any schooling, being the head of the household, household size, gender, the length under ARV at baseline, adherence, allocation to the original SMS treatment, and the CD4 count (at overall study entry) as sorting criteria.

Following this randomization, we compared the average characteristics in the three groups (treatment 1 and 2 and the control). Moreover, we compared the average outcome indicators from the baseline primary outcomes in the three groups (treatment 1 and 2 and the control). Differences were insignificant at the 5% level. Also changes in primary outcomes between experimental survey rounds 1 and 4 were well balanced (note that round 4 is the baseline of the RCT registry).

Note: All balancing tests above are uploaded as part of the RCT registry. Both p-values and q-values (Benjamini & Hochberg, 1995) are presented for balancing tests.
Randomization Unit
Patients living with HIV/Aids
Was the treatment clustered?
Experiment Characteristics
Sample size: planned number of clusters
Not applicable.
Sample size: planned number of observations
336 individuals at baseline and surveyed twice for follow-up: 1008 observations.
Sample size (or number of clusters) by treatment arms
Treatment 1: 240
Treatment 2: 240
Control: 528

Sample sizes for three rounds of data collection. Sample sizes are lower for outcomes only collected at follow-up 1 and 2.
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
With two repeated observations per treatment arm (80 subjects) and control (176 subjects), power 0.8, alpha 0.05, correlation of repeated measures 0.4 (found in baseline data), we are powered to detect an estimated effect size of 0.18 (estimate using power in STATA).
IRB Name
IRB Approval Date
IRB Approval Number
IRB Name
Research Ethics Committee at the International Institute of Social Studies of Erasmus University Rotterdam
IRB Approval Date
IRB Approval Number
Ethics 2018-04