Page 206 - Contributed Paper Session (CPS) - Volume 2

CPS1824 Sanggi L.
3. Cross-sectional weighting methodology

3.1. General steps in weighting
There are four general steps in weighting.

① Determine which sample units are in-scope of population
② Calculate the initial weights
③ Adjust for unit non-response
④ Calibrate to known benchmarks

In the KPCLS, the above third step (non-response adjustment) is omitted
due to the use of administrative data. Cross-sectional weights for households
and individuals and longitudinal weight are produced, but due to the
ambiguity of the household concept we do not produce longitudinal weights
for household.

3.2. Cross-sectional weight after wave 2
There have been many discussions on how to assign the initial weight for
new household entrants from the wave 2. We can know the probability of
selection for the PSM in wave 1, but for new entrants it is common to not
know their previous histories. For example, suppose persons A and B each was
in separate households during wave 1. Prior to the second wave A and B move
in together and form a two-person household. Then there are two paths
through which this household may be included in wave 2: by sampling A or B
in wave 1. If we do not calculate the selection probability of this household
properly, we would overstate the number of households with new entrants.

3.3. Approaches : M. Schonlau, M. Kroh and N. Watson (2013)
An estimate for the population total in wave t is given by

= ∑
̂

=1

̂
Taking the expectation yields () = ∑ ( ) . This estimate is

=1
unbiased if the weights are constructed such that () = 1. The standard way
of constructing weights is to choose inverse selection probabilities. The
constituent household refers to entrants that moved together from their old
household to the new household. The probability of selecting household
at wave t is the probability of selecting one or more constituent households.
P() = (ℎ1 ∪ ℎ2 ∪ ⋯ ∪ ℎ) = 1 − (1 − 1)(1 − 2) ⋯ (1 − ) ⋯ ⋯ (1)

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