Page 286 - Invited Paper Session (IPS) - Volume 1

IPS153 Christine B. et al.
imputation prediction method, also at a unit record level, to produce
migration statistics with associated uncertainties. These new migration
estimates are revised each month to include the latest information.
Estimation at unit record level is beneficial as it allows us to build up any
required aggregations. Importantly, it also allows the model to easily run
across linked administrative datasets, such as those contained within the IDI.
This means we can then estimate migration patterns of a subset of people with
a given set of attributes available in the linked data.

6. Discussion
The 2018 Census dataset consists of people counted through responses to
census forms plus those counted through administrative records. Stats NZ is
now confident that it has compiled a census dataset that will provide census
usually resident population counts and electoral counts of acceptable quality.
The improved accuracy of measures of external migration will provide greater
accuracy for population measurement in between censuses. The value of the
census questionnaire component is evident for some census variables where
no alternative sources are available. These variables consequently have higher
levels of missing data in 2018.
Our experience with the 2018 Census where we faced unacceptably low
response rates has highlighted the value of undertaking the longer-term
research into an admin-based census. Being well-prepared to use
administrative data to supplement responses, if necessary, provides a welcome
risk management strategy.
It has become increasingly clear that a transition towards more effective
use of administrative data in the census is best implemented as a step-wise
approach based on a well-designed programme of research. Gaining the
benefits of that research does not need to wait until conditions are ready to
implement a full admin-based census, but can add considerable value through
supporting the current census.
A statistical approach is essential to effective use of administrative data for
official statistics. Requirements include the ability to safely access linked data
sources, and statistical models developed for various estimation processes. The
New Zealand outcome-based external migration measures are a combination
of deterministic calculated migration estimates and modelled estimates with
associated uncertainty measures. Switching to modelled migration estimates
for the most recent periods (and away from intentions-based measures)
replaces a perceived precise, but potentially biased measure, with a more
variable measure where the uncertainties can be transparently defined. From
an end-user perspective, this is a big change, especially dealing with substantial
revisions for the most recent month when uncertainty is greatest. However,
these approaches prepare us for a world where increasingly administrative

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