Data ALCS 2013 14 Main Report English 20151222

281 ANNEX VI TECHNICAL NOTE ON SURVEY TO SURVEY IMPUTATION: POVERTY PROJECTION FOR AFGHANISTAN The ‘survey-to-survey imputation’ technique uses a model built from existing data from one survey to estimate missing data in another survey. Using the NRVA 2011-12 survey, which has consumption data, we used the survey-to-survey imputation method to estimate household consumption expenditure and poverty rates for the ALCS 2013-14, which does not have consumption data. This annex describes in detail the survey-to-survey imputation methodology. 68

VI.1 Data

Data used for the survey-to-survey exercise come from three rounds of national household surveys, NRVA 2007-08, NRVA 2011-12, and the 2013-14 ALCS survey. All three surveys are multi-topic surveys that collected a wide range of individual, household, and community-level socio-economic information, each over a one-year period to capture seasonal variations. The NRVA surveys’ sampling frame from the CSO’s 2003-05 pre-census household listing are representative at the national and the province level. For the ALCS 2013-14 the sampling frame stems from the 2003-05 household listing, subsequently updated in 2009. Comparability between surveys are maintained as much as possible by using similar questionnaires, training and data collection. ALCS 2013-14 collects many of the same variables as the 2007-08 and 2011-12 NRVA surveys, except for the food consumption. Besides the household surveys, we included conflict data from the UN Security Information and Operation Center SIOC. UN SIOC collects district-level daily conflict data. Total number of casualties in each district represents a proxy for level of conflict and insecurity. For each household, we calculated the total casualties for the district where the household resides for 1 month, 2 months, and 3 months prior to the household surveys. We used NRVA 2011-12 data and NRVA 2007-08 and ALCS 2013-14 data to create, validate, and ‘impute’ our final household consumption estimates. The final data used for survey-to-survey imputation excludes households from Helmand and Khost provinces from all three surveys because consumption aggregates for these two provinces for NRVA 2011-12, the base year for the consumption model, are not reliable. 68 We applied the survey-to-survey imputation method in Yoshida et al. 2015, the basis for a new household survey instrument called SWIFT Survey of Well-being via Instant, Frequent Tracking. The SWIFT approach used recent innovations in survey-to-survey imputation techniques developed by Harvard University, Stanford University and the World Bank’s research department. 282

VI.2 Model development