IBUC 2006 Abstract

Work Flow for the Projection of the German Microcensus Data Using Blaise Bascula
Kirsten Iversen ; Statistisches Bundesamt, Germany

The German Microcensus originally focussed on a specified reference week in spring. In January 2005 it started as
a continuous survey in order to obtain estimates of quarterly and annual means. The levy of the data is carried out
decentral, that means the statistical offices of the 16 regions in Germany are doing the field work. The survey is
compulsory.

For every quarter the data is weighted in two steps: First there is compensation that adjusts the known nonresponses.
In the second step the sample distribution of some auxiliary variables (for example age, sex and
nationality) is adjusted to the known population totals. The intention of this bounded second step weighting is to
reduce the bias of estimates due to unrecognised non-responses. The variance for variables which are high
correlated with the auxiliary variables is also minimised. Once a year the data of the four quarters are put together
for a deep regionalised weighting.

The compensation model and the weighting model are implemented as batch processes in Blaise Bascula. Besides
this some Blaise programs are preparing the data sets going into the different batch processes (for example
samples corresponding to households or individuals). The statistical offices of the 16 regions are applying the
program; the programming work is done by the federal statistical office.

This paper will demonstrate how the data is manipulated in the different steps on the way to calculate the final
weights for the quarterly and the yearly results of the German Microcensus.

Contact: kirsten.iversen@destatis.de