The Design of Social Protection for the Poor: Asset for Cash Transfers

The anti-poverty interventions we study take two forms. The first offered poor households productive assets in-kind. To determine the menu of assets to offer, in each village we first conducted an assessment of those assets most likely to provide high market returns. The menu of assets offered typically covered livestock, those enabling households to start a small retail business (e.g. grocery shop, fruit stall), crop farming, and other forms of self-employment (e.g. tailoring). In conjunction with these in-kind asset transfers, households were also offered training. Each training course divided into two parts: the first provided general training on establishing and running small-scale enterprises, and the second provided technical training specific to the chosen asset(s). The value of training was fixed at PKR12K. Hence the total value of transfers and training offered was PKR62K (around 620USD). We refer to this as treatment T1.

The second intervention is identical to the first but with one more listed option on the menu: to take an one-off unconditional cash transfer of PKR62K. To mimic the timing of transfers and training in T1, the delivery of cash transfers was staggered as an up-front payment of PKR50K followed by PKR12K a month later. We refer to this as treatment T2.

2014-01-01

2014-04-01

Labor activities of households. Economic preferences related to redistribution and pro-market beliefs.

Labor activities on the extensive margin will be the primary focus: labor force participation, engagement in casual wage labor, versus forms of self-employment. This will be so for male and female heads of household. Economic preferences will be measured by indices in relation to preferences for redistribution, pro-market beliefs and beliefs over the pro-sociality of others.

Our evaluation follows a two-stage randomization design. In the first, we randomly assign villages to T1, T2 or control. Randomization is stratified by district. At a second stage, within treated villages, we randomly assign the actual offer of treatment among eligible households. Half of those eligible are actually offered treatment. Among the poor in treated villages, we thus distinguish between the treated poor (TP) and the not treated poor (NTP).

Randomizztion in office by a computer.

First stage randomization is at the village level. Second is at the household level (within eligibles in treated villages).

Yes

We sample all eligible households in treated and control villages, namely all households with a poverty score of 18 or below. This covers 6237 poor households: 3052 reside in control villages, 1598 are in T1 villages (of which 854 are treated), and 1587 are in T2 villages (of which 942 are treated). We use our census to draw a random sample of non poor (hence never eligible) households from across all deciles of poverty scores. We denote non poor households as NP. We survey 9435 non poor households in total, 3130 reside in controls, 3306 in T1 villages, and 2999 in T2 villages.

16,000 households tracked over time (although not all households are surveyed at each survey wave).

16,000 households tracked over time in 88 villages (29 are assigned to T1, 29 to T2 and 30 to C). A further 15 C villages are available at baseline but these were then treated as part of another projects intervention later in our study.