SAS routines for variance estimation of poverty measures based on sample cumulated over waves of a panel

The main issue of the present work is to pool data with objective to enhance the sample size, in particular with reference to subnational (regional) estimates for which sample sizes are usually too small. This introduces an additional issue of dealing with correlations between samples from consecutive waves of a rotational panel such as EUSILC survey. The poverty status (and related indicators) of an individual are defined independently for each cross-sectional sample; hence the direct way to estimate variance of a cumulative sample is to pool the cross-sectional samples and apply a suitably adapted standard variance estimation procedure. Such direct estimation requires full information on the sample structure. At a minimum this includes specification at the micro level of: sample weights, stratum, primary sampling unit (PSU), permitting linking of data across waves. The SAS routines presented in this work have been developed for application when full information on the sample structure is available. Our empirical application is based on micro-data for the survey of Spain for year 2009, 2010 and 2011, to which we have had a privileged access through a project with OECD. Unfortunately, EU-SILC micro data available to researcher generally lack full information on sample structure. In general, the variance estimation procedure would need adaptation (and some additional assumptions) to deal with the situation when full information on sample structure is lacking. While this work does not address alternative procedures for the purpose, we have developed and applied those in previous research. For completeness, the technical steps involved have been outlined in the concluding section.