Page 284 - Invited Paper Session (IPS) - Volume 1

IPS153 Christine B. et al.
the M* census subset should all have been linked to the IDI spine, and those
who have not been linked are missed matches. We obtain an estimate of 1.4
percent of the M* subset who were incorrectly not linked to the IDI spine, with
rates available by strata (age, sex, geographic areas, and ethnic groups). We then
assume that same rate of missed matches applies to the remainder of the census
file. The appropriate number of people are removed through random selection
within strata. This adjustment does not remove exactly the people whose census
record should have matched to the IDI. Rather it removes a random selection of
the right kind of people to a fine demographic breakdown.
Accounting for the quality of admin location data: After these
adjustments we can be confident that the remaining eligible IDI-ERP people
should have been counted by the census. However, we also consider the
accuracy of admin location data, which decreases as geographies get smaller.
To limit the errors for small sub-national geographies, we remove people who
have a low probability of a correct meshblock location. The trade-off here is
between including more individuals in the census dataset, and protecting the
integrity of small area geographies. We include people for whom there is at least
a 50 percent probability that we have their correct meshblock. This meshblock
cut-off is the main driver of which admin people are included or excluded from
the census file.
Previous censuses have applied statistical imputation methods to count
some of the missed respondents in the final census dataset. The benefits of
admin enumerations over imputation methods are evident in the 2018 results,
and in comparisons with the 2013 Census. The IDI-ERP admin population does
include people who are traditionally hard to count in a census. Including high
quality admin enumerations does better at improving the census distributions
than imputation methods that relied on local area missing at random
assumptions, given the disproportionate nature of census non-respondents.
In addition, the admin enumerations are real people for whom we may have
associated characteristics from alternative sources.

5. New outcome-based measures of external migration
In New Zealand, the largest component of population change since the
2013 Census has been external migration. A reliable measure of migration is
critical in ensuring a consistent and credible resident population estimate. A
unit record level estimation of migration allows for the removal of people
present within the administrative sources who have departed New Zealand,
and addition of those who arrived into New Zealand as migrants. Therefore,
with an individual level measure of migration, the largest component of
population change can be coupled directly to the source data of population
estimation.

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