Page 65 - Special Topic Session (STS) - Volume 4

STS563 Pete Jones
an integrated LMS/PCS was worth pursuing for more detailed statistical
testing.
A key milestone for ONS social survey and population statistics
transformation is to deliver a large scale LMS/PCS test in 2020. The aim of this
test is to produce population estimates and labour market statistics for
comparison with official estimates at national level. Early simulations have
identified a potential approach based on a two-stage stratified clustered
8
design. Under this design, small areas (ONS output areas) are selected from
strata at random as the primary sampling unit. Within the output areas
selected, a fixed number of addresses will be randomly selected to take part
in the survey. The stratification is based on assigning each output area into 1
of 6 strata for over-coverage and under-coverage using the minimum variance
stratification method. The design variable for over-coverage is based on
measuring numbers of people moving into and out of addresses in the area.
For under-coverage we assessed a number of predictors, including migration
within the area, ethnicity, and other evidence of administrative data activity
not captured on the SPD. Using principle components analysis, we have
developed a composite design variable to stratify for under-coverage derived
from these variables. An initial assessment of model fit gives an R2 value of
0.44 for the over-coverage measure and 0.48 for under-coverage. These are
broadly comparable with the model used to predict Census non-response as
part of the 2011 hard-to-count index (0.47) . Sample sizes for the 2020 test
9
will be calculated once the impact of a stratified design is fully understood for
the precision of labour market statistics.

4. Discussion and Conclusion
Initial operational tests for an integrated LMS / PCS survey have
demonstrated potential for combining survey questions on labour market and
population coverage. Further research is needed before concluding that the
proposed design will produce robust statistical outputs. From the PCS
perspective, an understanding of how response rates to a voluntary survey will
impact the quality coverage assessment is needed. While the CCS, which is
used to support the traditional Census estimation framework is also a
voluntary survey, it achieves high response rates (90%+) on the basis that
respondents associate the collection with the mandatory Census that is
conducted a few weeks prior to the CCS. The proposed integrated LMS/PCS
survey will undoubtedly achieve lower response rates, posing an increased risk
of bias in the resulting estimates. Understanding the relationship between
survey non-response and registration on administrative data is of important
consideration when evaluating the impact of lower response rates, and further
work will be undertaken in this area. In addition, a method for adjusting over-
coverage within the dual-system-estimation framework is also needed in
preparation for the 2020 test.

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