This paper presents the design and validation of an econometric model that uses routinely collected administrative data to target unconditional cash and in-kind assistance to Syrian refugees in Lebanon. The authors compare the prediction accuracy of the proposed model to a traditional Proxy Means Test (PMT) approach. A traditional PMT approach draws on national household survey data to identify the household characteristics that best predict household poverty. These characteristics are used to develop criteria for program eligibility as well as a short-form survey (or ‘scorecard’) that is administered to the entire target population to identify eligible households.
The analysis relies on: (1) nationally representative survey data from the 2018 Vulnerability Assessment of Syrian Refugees in Lebanon (VASyR), which includes detailed information on households and expenditure patterns; (2) UNHCR administrative data as of June 2018; and (3) data as of June 2018 from the Refugee Assistance Information System (RAIS), which includes information on all refugee families who were receiving assistance in Lebanon from any of the major international organizations or their partners. The authors use VASyR data to calculate expenditure per capita for each household, which is then linked (using unique household and individual identifiers) to household demographic data drawn from UNHCR administrative data.
Key findings:
- Overall, households that are predicted to be poor tend to be larger, have a higher share of disabled members, are substantially more likely to be female-headed, are less likely to have a working-age male, and have a higher share of dependents. Additionally, the model is more likely to identify eligible households with lower education and with a larger share of members who had no previous occupation before their arrival in Lebanon.
- The use of basic demographic information from typical administrative records held by aid organizations and governments is approximately as accurate in targeting the poor compared to traditional PMT approach. There is no substantive difference in the capacity of administrative data—which does not include any information on assets—to predict poverty, relative to traditional survey-based methods. The survey-based approach decreases the likelihood of in inclusion and exclusion errors by about two percentage points, but these differences are not statistically significant.
- A small number of fields in the survey data provide additional predictive power. A small number of basic household furniture questions provide modest improvements. Consequently, adding a housing question to the administrative database would improve targeting accuracy by around two percentage points in overall error.
The authors conclude that routinely collected administrative data on refugees can potentially offer an equally reliable and less costly alternative to existing PMT approaches to targeting social or aid programs. The proposed model also avoids several problems associated with the PMT approach, in particular: the fact that some of the targeted households cannot be reached or do not respond to the short-form survey; or they do not provide accurate answers on important survey questions, such as the household structure or the availability of assets.