Unpacking a Multi-Faceted Program to Build Sustainable Income for the Very Poor

For the multi-faceted program in Ghana, Graduating from Ultra Poverty ("GUP"), implementers first identified poor communities in poor regions of the country. In each identified community, staff members then facilitated a Participatory Wealth Ranking (PWR), in which members of the community worked together to rank households by economic status. Finally, staff members returned for a verification of the households judged to be the poorest. The program was implemented by Presbyterian Agricultural Services, a local nongovernmental organization, in coordination with Innovations for Poverty Action, a non-profit research organization.

The basic GUP program involved the transfer of a productive asset; skills training for the management of the asset as well as life skills training, a weekly cash stipend for consumption support, worth between $6 and $9 PPP depending on family size, lasting for 3-10 months; access to a savings account at a local bank (details below in the experimental design section, as this is one of the components unpacked); and some basic health services and health education. The productive asset was provided at the beginning of the program, and households were permitted to choose a package of assets from a set list. The rest of these services were delivered over two years via regular visits (typically weekly) by a field officer from the implementing organization.

2011-07-08

2013-07-07

asset value index, consumption index, financial inclusion index, food security index, income index

For each variables within a given index, we take the control mean and standard deviation and and compute the z-score. Then we average all the z-scores in the index, and again take the control mean and standard deviation to compute our final indexed outcome. Our asset value index contains just one variable: the total asset value, where each asset is priced in goats using price data from other countries where we have price data (under the assumption that relative prices are constant across countries). Our consumption index is total consumption. Our financial inclusion index includes total savings balances and total loans received in the last year. Our food security index includes three variables. The first two variables equal 0 if the household answered "all year" or "during the lean season only" to the following questions, about adults and kids, respectively: "Did adults/kids ever reduce number of meals per day or reduce portions over the past year?" The third variable equals 0 if the household answered "all year" or "during the lean season only" to the question "Did adults ever skip entire days without eating?" Our income index includes monthly business income, monthly crop income, monthly wage income, and monthly animal revenue.

business income, crop income, animal revenue, wage income, savings deposits, livestock value (goat, fowl, pig, sheep, cow)

Savings deposits come directly from administrative data collected at the individual-week level. To compute livestock value, we take the median buy price and the median sell price across villages, and then take the average of the two. We then multiple this price by the number of livestock owned.

The program included several experimental arms designed to unpack whether specific components were sufficient on their own. First, we introduced variation in the core program at the household level: 50% of sample households within GUP villages were randomly assigned to the graduation program without savings ("GUP without savings"), and 50% received the standard program with collection of
savings for deposit into a local bank by the field agent ("GUP with savings").

Second, we introduced two additional treatment groups at the village-level. For each of these two additional treatment groups, a two-level design was maintained, thus creating treatment households in treatment villages, control households in treatment villages, and control households in control villages.

In Asset-Only villages, 50% of sample households were assigned to treatment, and received only a productive asset, without skills training on how to use it, or any of the other GUP components. These households were simply given four goats, since this was the most popular asset in GUP (71% of households chose a package of assets that included four goats). Goats were chosen because most households have had or have some goats, and thus would typically be content with receiving more, and we did not want to interact with the households and give them choices for fear that such engagement itself could change behavior.

In Saving Out of Ultra Poverty ("SOUP") villages, 59% of sample households were assigned to the SOUP treatment, and received a visit from the field agent to collect savings, just as in the GUP with savings group, but did not receive any other components of the program. Of this 59%, half received savings accounts and deposit collection without a match (“SOUP without match”) and half received savings accounts and deposit collection with a 50% match ("SOUP with match"). Specifically, for every GHC 1 deposited, households in this group received a matching contribution of GHC 0.50. The remaining households in SOUP villages were assigned to the SOUP control group.

In office by computer

First, treatment was assigned at the village level (control, GUP, SOUP, or asset-only). Within control villages, all households were assigned control. Within asset-only villages, households were then assigned to either control or treatment. Within GUP and SOUP villages, households were then assigned to control or to one of the two sub-treatments.

Yes

276 villages.

4177 households.

76 villages and 1299 households were assigned pure control. 78 villages were assigned GUP, with 642 households assigned control, 333 households assigned GUP, and 333 households assigned GUP with savings. 77 villages were assigned SOUP, with 510 households assigned control, 371 households assigned SOUP, and 362 households assigned SOUP with match. 45 villages were assigned asset-only, with 163 households assigned control and 164 households assigned asset-only.

The minimum detectable effect size for our indexed outcomes at the individual level is 0.127 standard deviations.