Page 18 - Invited Paper Session (IPS) - Volume 1

IPS30 Stefan D. et al.
of the model together. The results of the different surveys and census
components will therefore be linked in a central data stock instead of storing
them separately as in 2011. In doing so, data can be cross-checked and
validated at an early stage of data processing. Inconsistencies and
implausibilities can be removed by rules or even by manual checks. This helps
to improve data quality and helps reducing efforts to link the data with each
other consecutively.
• The use of paper questionnaires has to be reduced as much as possible by an
“online-first” strategy. This is an important part in being on time in the next
census round, which can not only contribute to an important cost reduction,
but also decrease response burden because respondents are specifically
guided through the questionnaire.
• Building up the address register has to start earlier and one of the data sources
will not be used any more. In 2011, three main sources were leveraged to
collect addresses: the Population Registers, data of the Federal Mapping
Agency and data of the Federal Employment Agency. The latter will not be
used in 2021 anymore as there were no further addresses added by this
source, but many cross-checks were necessary due to different spelling of
cities, streets and house numbers.
• In 2011, data of the Federal Employment Agency were furthermore used to
generate data on employment: The data were of high quality but users
complained the complexity of analyses, since different employment figures
were released depending on whether they were based on the combined
model or the household survey only. Looking at employment in a broader
sense, this source had to be analyzed in combination with the household
survey to cover self-employed or unemployed as well.
• The weighting scheme of the supplementary household survey was targeted
primarily at a highly precise number of inhabitants. The production of results
for census variables that were not available from registers was only considered
as a second priority in the development of the estimators. The weighting
procedure needs to be optimized in order to minimize any risk of bias in case
of the census variables not available from the registers.

4. Future perspectives towards a register-based census
Although the combined census model was implemented successfully,
already before conducting 2021 census work, preparations for a change-over
to a fully register-based census following the 2021 census round have already
started. These developments are driven by several considerations, which are
presented in the following paragraphs.
As demonstrated by the German Statistical Council and recent discussions
at the European level, users require census data more frequently, more timely,
and in more differentiated regional breakdown (Eurostat (2017, 2018)). The

7 | I S I W S C 2 0 1 9