Slow escape or trapped in poverty? Evidence from a randomized control trial in Bangladesh

The Transfer Modality Research Initiative (TMRI) was designed as a cluster-randomized controlled trial in two zones of Bangladesh (north and south). WFP-Bangladesh implemented the intervention, and the International Food Policy Research Institute (IFPRI) conducted the impact evaluation research. Intervention arms were assigned at the village level, stratified by zone, where arms were as follows: (1) cash transfers [north and south]; (2) cash transfers + nutrition behavior communication change (BCC) [north only]; (3) food transfers [north and south]; (4) food transfers + nutrition BCC [south only]; (5) food-cash split [north and south]; and (6) control [north and south]. Within treatment villages, women living in very poor households were targeted to receive benefits for two years, from May 2012-April 2014. For the original impact evaluation, longitudinal data on 5000 households were collected from 2012-2015.

The objective of this study is to assess, 12 years after a pilot safety net intervention ended in April 2014, the sustainability of the intervention's impact on household consumption and poverty. Given available funding for re-survey work, the sample is based on the 2000 households that were included in the following arms in the north: (1) cash transfers; (2) cash transfers + nutrition BCC; (3) food transfers; and (4) control. Subject to attrition being uncorrelated with the intervention arms, the resulting sample will allow unbiased estimates of the impacts of various treatments as of 2026, 12 years after completion of the program.

2012-05-01

2014-04-30

We have two primary outcomes: log real per capita consumption; and whether the household is below the extreme poverty line. Because headcount poverty alone cannot distinguish households trapped in poverty from households converging slowly toward the poverty line, we additionally designate the poverty gap (P1) and the squared poverty gap (P2) as co-primary adjudication outcomes, measured at the extreme poverty line and at a line 20 percent above it.

Based on power calculations, in the North, five upazilas (sub-districts) were selected using simple random sampling from a list of upazilas where in 2010 the proportion of households living below Bangladesh’s lower poverty line was 25% or higher. All villages within these five upazilas were listed. Villages classified as urban or villages with fewer than 125 households were dropped. Using a random number generator, each village was assigned a random number. Villages were then sorted in ascending numerical order with the first 275 retained. The first 50 villages were assigned to treatment group 1, the second 50 to treatment group 2, the third 50 villages to treatment group 3, the fourth 50 villages to treatment group 4, and the fifth 50 villages to the control group. The remaining 25 villages were held as a reserve. A complete village census was carried out in each of the 250 selected villages, collecting information on household demographics, a set of poverty indicators, and whether households participate in safety nets and other targeted interventions. Using these data, a list was compiled of households that: (1) were considered poor (i.e. based on the poverty indicators collected, they were estimated to have consumption below Bangladesh’s lower poverty line); (2) would have at least one child aged 0-24 months when the intervention began; and (3) were not receiving benefits from other safety net interventions. These households were eligible to participate in the study. Using simple random sampling, 10 eligible households were selected from each village. The total sample in the North in the 2012-14 study thus included 250 clusters and 2,500 households; note, however, that for this work we only include households that were in the cash, the food, the cash and BCC, and control arms (200 clusters and 2,000 households at baseline).

Not available

Based on power calculations, in the North, five upazilas (sub-districts) were selected using simple random sampling from a list of upazilas where in 2010 the proportion of households living below Bangladesh’s lower poverty line was 25% or higher. All villages within these five upazilas were listed. Villages classified as urban or villages with fewer than 125 households were dropped. Using a random number generator, each village was assigned a random number. Villages were then sorted in ascending numerical order with the first 275 retained. The first 50 villages were assigned to treatment group 1, the second 50 to treatment group 2, the third 50 villages to treatment group 3, the fourth 50 villages to treatment group 4, and the fifth 50 villages to the control group. The remaining 25 villages were held as a reserve. A complete village census was carried out in each of the 250 selected villages, collecting information on household demographics, a set of poverty indicators, and whether households participate in safety nets and other targeted interventions. Using these data, a list was compiled of households that: (1) were considered poor (i.e. based on the poverty indicators collected, they were estimated to have consumption below Bangladesh’s lower poverty line); (2) would have at least one child aged 0-24 months when the intervention began; and (3) were not receiving benefits from other safety net interventions. These households were eligible to participate in the study. Using simple random sampling, 10 eligible households were selected from each village. The total sample in the North in the 2012-14 study thus included 250 clusters and 2,500 households; note, however, that for this work we only include households that were in the cash, the food, the cash and BCC, and control arms (200 clusters and 2,000 households at baseline)

Village

Yes

Planned Number of Clusters: 200 villages

Planned Number of Observations 2000

Approximately 500 per treatment arm

Slow escape or trapped in poverty: Pre-analysis plan