Expectations and Choice: Evidence from a Home Upgrading Program

In this project we evaluate an existing national program that upgrades the homes of beneficiaries of Progresando con Solidaridad (ProSoli), the conditional cash transfer (CCT) program of the Dominican Republic government. The home improvements include upgrading dirt (and badly damaged cement) floors into new cement floors, as well as painting the exterior walls of the house. Beneficiary families of the program receive a letter that describes the home improvements in general terms. We combine 2 interventions, as follows:

Message specificity
When announcing the program to selected beneficiaries, we provide a detailed description of the home improvements to half of the household beneficiaries ('specific message'), while the remaining treated households are given the status-quo general letter ('general message').

Choice of color:
Half of the treated households in our study are allowed to choose the color that will be used to paint the exterior walls of their homes ('color choice'), while the remaining treated households are not given such choice ('no color choice').

2019-10-01

2020-03-31

Satisfaction and engagement with home improvement program and ProSoli. Satisfaction and perceived value of house. Additional home improvements. Health of the children.

Primary outcomes will be self-reported using household survey instruments. Satisfaction with the house and program will use a likert scale converted into a 0/1 variable. Participants will also report what they expected from the program when it was announced and whether such expectations were met. Engagement with the program while being implemented counts the number of activities the household engaged in with the crew responsible for the upgrades (e.g. moving furniture, painting the walls themselves, fixing the floor, offering beverages or snacks etc). Health of the children is measured by asking about health issues faced by the children of up to 6 years old that live in the household (e.g. diarrhea, worms, respiratory diseases, gastrointestinal disorders, skin issues, among others).

Value of the house and investments in home improvements. Number of school days missed by children. Social economic perception and stress levels of caregivers of children up to 6 years old.

Secondary outcomes are also measured using self-reports from household surveys. Survey participants will report on investments in additional home improvements after the upgrades. Perceived socioeconomic status is measured using a 1-10 ladder where respondents compare their socioeconomic status to that of other households in their region or country. Stress levels are measured using a scale similar to the Perceived Stress Scale developed in Cohen (1994) and will be administered to caregivers of children up to 6 years old.

The two interventions are combined in a 2x2 design, with four treatment and a control group. We group households in clusters, that is, small geographical areas each consisting of 3 to 4 households. Clusters are assigned to one of the four treatment or control group: 10% of the clusters are assigned to the 'control group', where all households in the cluster do not receive the program. The remaining 90% of clusters are divided evenly into the four treatment groups. To assess spillover effects, 20% of households in treated clusters will not be targeted and thus will not receive the program. By design they are located close to the "targeted" households that will receive the program. The experiment thus consists of the following groups of households: (i) control households, (ii) untargeted households, and households that receive (iii) general message and no color choice, (iv) general message and color choice, (v) specific message and no color choice, and (vi) specific message and color choice.

Randomization done in the office by a computer using Stata.

Randomization has two stages. We first randomize clusters into treatment and control. All households in control clusters will not receive the program. Households of treatment clusters will be randomized into targeted households, that will receive the program and untargeted, that will not.

Yes

400 polygons consisting of a small geographical area of 3 to 4 households on average.

1,300 HHs

Following our experimental design, our sample will be divided into the 6 control and treatment arms as follows:
(i) control households -- 130 HHs
(ii) untargeted households -- 234 HHs
(iii) general message and no color choice -- 234 HHs
(iv) general message and color choice -- 234 HHs
(v) specific message and no color choice -- 234 HHs; and
(vi) specific message and color choice -- 234 HHs.

With limited baseline data on two of the main outcomes of interest, we make educated guesses for the means and ICC and compute the MDEs given the sample we have. We focus on two outcomes: satisfaction with the program and the number of activities the household engaged in during the implementation of the program. With the sample size that we have, assuming a mean satisfaction with the program of 60% in the control group and an ICC of 0.05, we can detect an increase in satisfaction among treated households of 0.13. If we assume a more conservative ICC of 0.15, the MDE is 0.14. Assuming that households in the general message (or the no choice) engage in 2 activities on average with a standard deviation of 3, we can detect an increase of 0.79 activities if the ICC is 0.2.