Boosting patience for inter-generational poverty reduction

Investments in children are low in developing countries, in spite of high returns. I propose an RCT to test whether training designed to increase mothers' future-orientation can be an effective way to boost investments in children. In addition, I study whether the focus of those trainings matter depending on the shape of the mother's baseline time-preferences.

Indeed, a companion paper (Lichand and Thibaud, 2018) has documented that a large fraction of parents do not discount their future consumption to the same extent as that of their children. Parents who are more patient for their children than for themselves tend to plan to be more generous towards their children in the future and to reallocate this planned consumption towards themselves when the time comes to execute those plans. We call those respondents parent-biased. In theory, the most effective way to increase investments in children for those parent-biased subjects is to increase the discount factor they apply to their own future consumption. Increasing the discount factor they apply to their child's consumption would lead them to make more ambitious plans for their children's future but would not help them stick to those plans.

I present here an RCT designed to test the following research questions:
1- Can training interventions aiming at developing future-orientation in mothers be effective to boost investments in children?
2- If mothers are parent-biased are the training interventions more effective when they focus on increasing the mothers' own discount factor?

In collaboration with a Malawian NGO, Student Driven Solutions (SDS), I have developed two sets of trainings that were randomly assigned to villages. The first set of workshops aimed at increasing patience and decreasing present-bias when making decisions involving one's own consumption(``Treatment 1''), the second one at increasing patience and decreasing present-bias when making decisions involving one's child consumption (``Treatment 2''). Both sets of trainings are inspired by the protocol of previously evaluated interventions who have been successful in shaping intertemporal preferences and by SDS' female empowerment workshops. SDS translated the material, recruited and trained the trainers. The intervention was monitored through random spot checks.

A subset of ``Control'' and ``Treatment 1'' villages received leaflets containing information about returns to investments in children, to control for the potential priming effect of ``Treatment 2''. Each treatment was composed of a set of eight two-hours trainings. In treatment villages, 30 participants were invited to attend the trainings, which were held approximately twice a week.

2019-03-25

2019-04-26

Time-preferences:
- Incentivized measure of delta_a (discount factor that the mother applies to her own consumption)
- Incentivized measure of parent-bias

Investments in children:
-Willingness-to-pay for a school textbook,
-Index of investments in children,
- School attendance as reported by school teachers.

-Willingness-to-pay for commitment: willingness-to-pay to open a saving account in the child's name

Parent-bias:
I define parent-biased respondents as respondents who discount their own consumption to a larger extent than that of their children.

To elicit parent-bias, we ask parents to allocate five packs of peanuts between themselves and their child to be consumed two days later
and a month later. To help with this decision, the parents are invited to share 5 packets of peanuts between two plates, one entitled "you, in two days", the other one ``Your child in two days''. The enumerator records this decision. Then the parents are invited to do the same thing for the next allocation.
To ensure that all decisions are consequential, the parents are informed that a randomly drawn subset of the respondents will see their decision implemented.
I define parent-biased respondents as those deciding to allocate a larger share of peanuts to their child at t=3 than t=2.

Delta_a: The respondents split the consumption of three packages of peanuts, for their own consumption, between t=2 and t=3. For each package not consumed at t=2, they received r additional packages at t=3 The respondents were asked to make this decision for three interest rates between 0.5 and 1.5 For each interest rate, I impute the value of delta_a associated with the respondents' decision. I use their average as the value of delta_a in my analysis. This will be my preferred measure of delta_a

Willingness-to-pay for a textbook: At endline, I offer respondents the possibility to enter a lottery, in which they can earn a certain amount of money (between 0 and 2,000 kwachas) in the future. For all amounts larger than zero, they can choose to either receive the money or a school textbook, appropriate for their child's school level. This elicitation game follows the design of a similar task used in Dizon (2018) in Malawi.

The participants make this decision for different amounts of money. At the end of the survey, the participants learn which amount they have earned in the lottery and their decision for that amount is executed. This design is a version of the Becker-DeGroot-Marschak mechanism and ensures that all questions are incentive-compatible.

I measure the parents' willingness-to-pay for investments in children through a series of three to four interdependent hypothetical binary choices between receiving money or the investment in the child, following a ``staircase'' procedure. The sequence of interdependent questions I ask and the inputed willingness-to-pay for investments in children is detailed in the attached document

Index of investment in children:

At baseline and endline, I surveyed parents about actual investments in their children's education and health in the recent past. I restricted my attention to the child involved in the experiment.

To control for family-wise error rate in the context of multiple hypotheses testing, I build two separate summary measures of investments in children, depending on whether the child is between three and five or six and twelve years old. Each summary index measure is the equally weighted average of z-scores of its components. For each variable, I build z-scores by substracting its mean among respondents in the control group and dividing by its standard deviation among parents in the control group. Among those respondents, each component of the summary index has therefore mean zero and standard deviation one.

I build this index separately for children younger and older than 6 years old. It is composed of the following:

Mean expenses on preventative health-care for children aged 0-12 years old in the 4 weeks before the experiment,
Has the child been vaccinated during the measles and rubella immunization campaign in July 2017? (not asked at endline),
Was the child given any drug for intestinal worms in the 6 months before the experiment?,
Was the child given Multiple Micronutrient powder in the 7 days before the experiment?,
Was the child given iron supplements in the 7 days before the experiment?,
Was the child given therapeutic food in the 7 days before the experiment?,
Was the child given supplementary food in the 7 days before the experiment?,
Was the child given a vitamin A dose in the 3 months before the experiment?,
Has the child been taken to a well-baby or under-5 clinic for a health check up in the 3 months before the experiment? (only for children younger than 6 years old),
Has the child been taken to a well-baby or under-5 clinic for a growth check up in the 3 months before the experiment?
(only for children younger than 6 years old
Mean expenses on education for children aged 2-12 years old,
Attendance to Early Childhood Development Programmes for children under 6,
For children aged 6-18: numbers of days the child attended school in the month before the experiment,
Educational support index: how often do you:
• Help your child with homework or schoolwork.
• Ask your child if s/he did his/her homework or schoolwork
• Help your child to organize the school material, such as books, notebooks, and backpack.
• encourage your child to not miss class or be late for school.
• Ask about her/his grades in tests, activities and classes
• Incentivize your child to study or read.
• Ask your child about his/her day in school
• Go to school parent meetings
• Talk to your child's teachers

School attendance: Because my two previous measures of investments in children may suffer from experimenter demand effect, I further designed a school questionnaire to ask to the teachers of the primary-school-age children in my sample. The measures of school attendance are collected using the school ledgers
I will build an index of school attendance based on the following components:
- Days the child was present in school from june 1st-21st
- Frequency of parental interactions with the teacher over the previous month

-WTP for a savings' account
The mothers enter a lotery in which they can earn 0 or 10,000 kwachas. Before learning the lotery outcome, they can choose between 2 options:
1- Receiving the whole money in cash;
2- Opening a savings account at the National bank in their child's name and depositing 5,000 kwachas. Our team will accompany the respondent and the child at the bank and help them with the paper work. The respondent will receive the remaining money in cash.

The respondent are asked this question, with a different “price” associated with each option. If the respondent earns 10,000 kwachas in the lotery, a price will be randomly chosen at the end of the interview and the respondent's decision at that price will be executed.

I measure the parents' willingness-to-pay for the savings' account through a series of three interdependent binary choices. The sequence of interdependent questions I ask and the inputed willingness-to-pay for the savings device is shown in the attached document

Educational achievement: I use a measure of the children's progress in maths and english as reported by the teacher as a measure of educational achievement.

Two stage randomization:
At the village level, villages are allocated to three treatment arms:
-Control
-Treatment 1: received 8 2-hour long workshops focusing on inter-temporal decisions involving the mother's own consumption
-Treatment 2: received 8 2-hour long workshops focusing on inter-temporal decisions involving the child's consumption

At the individual level, respondents were then assigned to either receive additional information about returns to investments in children or no additional information.

This experiment was conducted in 80 villages of Dedza district in Malawi, with 2,400 participants. Households were eligible if both parents lived at home, if they had at least one child aged between 3 and 12 years old and if no one in the household was allergic to peanuts.

At baseline, if multiple children were present, one child was randomly selected to be the focus of the data collection.

At endline, if multiple children enrolled in primary school were present, one child was randomly selected to participate in the data collection. If no children enrolled in primary school was present in the household, the questionnaire focused on the child selected at baseline.

Randomization done in office by a computer

-The trainings were randomly assigned at the village level

-The additional information leaflets were randomly assigned at the individual level

Yes

80 villages

30 per cluster=2400

16 to control
32 to treatment 1
32 to treatment 2

-I observe an ICC of less than 0.001 for my main outcomes variables at baseline. -General impact of the trainings: I will pool treatment 1 and 2 in this specification. With a sample size of 80 clusters, 30 households per cluster, an ICC of 0.001 and power of 0.8, I am powered to detect a 0.145 standard deviation difference between the control and the treatment means. -Restricting my sample to parent-biased respondents who are either in treatment 1 or treatment 2, I study whether they benefit disproportionately more training focusing on their own discount factor (treatment 1) With a sample size of 64 clusters, 21 parent-biased households per cluster, an ICC of 0.001 and power of 0.8, I am powered to detect a 0.155 standard deviation additional benefits of the Treatment 1 treatment for parent-biased respondents.