Page 383 - Invited Paper Session (IPS) - Volume 2

IPS280 Erniel B. Barrios et al.
where E= total number of PSUs; e – number of sample PSUs; - total
number of HH in PSU; ℎ – total number of sample households in sample
ℎ
ℎ
PSU, and; – measurement of indicator for sample household in sample
ℎ
ℎ
PSU.
Modelling
We used the estimates obtained from the matched and unmatched
samples to generate estimates of annual totals for the indicators. The annual
estimate of the mean and the corresponding variance is

Assessment
The reliability of estimates per domain and per disaggregation is evaluated
using the Coefficient of Variation (CV). Efficiency of an estimator is considered
acceptable if CV is lower than 10%.
Simulation exercise
We simulate various rotation patterns and assess its effect on CV of
provincial level estimates of different indicators. We vary the composition
share of matched and unmatched samples for the following cases: 40%
matched and 60% unmatched, 75% matched and 25% unmatched and 90%
matched and 10% unmatched.

3. Results
Provincial Estimates of Total/Mean Income and Total/Mean Expenditure
by Income Class, Expenditure Class, Family Size
The use of the matched and unmatched samples generally yield reliable
estimates reliable for smaller areas (85 provinces and highly urbanized cities)
while retaining the sample size of 45,000 per semester visit of the respondents.
At the national level, for overall mean or total income and expenditure
estimates, all provinces exhibited estimates with CV not exceeding 10%.
Disaggregation income class and expenditure class, estimates in all provinces
in each class also posted reliable CVs. When disaggregated by family size,
some provinces posted CVs more than 10% CV but not more than 20% CV.
For one-person family group, 23 provinces had estimates with CV higher than
10%. For two-person family, only 11 provinces yield estimates with CV higher
than 10%. Among three-person and four-person family, only 10% of provinces
have CV exceeding 10%.

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