Introduction Sample frame Sample size Stratification

268 ANNEX IV SAMPLE DESIGN AND IMPLEMENTATION

IV.1 Introduction

The sampling design of the ALCS 2013-14 was developed to produce results that are statistically reliable for most of the indicators at national and provincial level. In addition, the aim of the sampling design was to have representative estimates by season according to the Shamsi calendar used in Afghanistan, 65 in order to capture seasonal fluctuations in a number of key indicators. The design developed for the 2013-14 survey round was a stratified, two-stage cluster approach. The sample distribution is sufficiently close to the national urban-rural distribution that separate analysis for these populations is justified.

IV.2 Sample frame

The pre-census household listing that was conducted by CSO in 2003-05, updated in 2009 was used as the sampling frame. For three provinces, the sampling frame consisted of the Socio-Demographic and Economic Survey SDES household listings: Bamyan data collected in 2010, Ghor and Daykundi both with data collected in 2012. Prior to the fieldwork, the selected EAs – urban and rural – were visited for a mapping update of the households, on the basis of which the second sampling stage was implemented. The sampling frame that was used for the Kuchi population was the 2003-04 National Multi-sectoral Assessment of Kuchi NMAK-2004. Although far from perfect given the rate of settlement of Kuchis in recent years and ongoing discussion about the definition of Kuchi, this is the best frame available for this part of Afghanistan’s population.

IV.3 Sample size

Analysis of previous NRVA rounds showed that a sample size of around 21 thousand households with a cluster size of ten households would produce sufficiently reliable estimates for most variables. Consequently, this sample size was maintained as the standard the 2013-14 ALCS.

IV.4 Stratification

The sample was stratified into 35 analytical domains: one for each of the 34 provinces of Afghanistan and one for the Kuchi population. For an optimal sample allocation across the provinces, a balance was obtained between proportional allocation and equal-size allocation with a Kish power allocation of I = 0.25. This assured sufficient weight for provinces with small populations, while improving the comparability of results across the provinces. Since the provincial sub-samples were equally distributed over 12 survey months, the four seasons also become separate analytical domains. The Kuchi stratum was only divided into the 2013-14 Shamsi calendar 1392 winter and 2014 1393 summer season in view of the practical difficulty of locating migrating communities in spring and autumn. 65 For conversion of Shamsi season dates, see Annex VII. 269

IV.5 Cluster size and number of clusters