Page 249 - Special Topic Session (STS) - Volume 4

STS583 Michael W. et al.
depends very much on the quality and, in particular, the timeliness of the
sampling frame, which unfortunately is often not satisfactorily fulfilled.
ii. Sampling from a purely satellite based sampling frame, where the
population values are derived from an index of the built-up area. In this
scenario, we test the possibility of sampling only from a grid that measures
the values of the built-up area within the grid cell. This approach would
allow a sample to be taken entirely without the help of the census-based
sampling frame.
iii. Linking of the satellite-based population data as well as the land coverage
data to the available georeferenced census-based enumeration areas
(hybrid approach). The primary purpose of this approach is to update the
conventional sampling frame to reflect the required timeliness discussed
above, as well as to improve its informativeness by adding the landcover
data. This allows us to conduct a higher degree of stratification, resulting
in a more balanced distribution of the target variable(s) population
variance.

Population Estimates
Population means and totals are estimated from the sample population
with the support of design weights. Design weights are the inverse of the
selection probability of the final sampling unit.
The final estimate is already described above, however the selection
probability in a 2-stage design can be decomposed in 2 components, one for
each sampling stage:
p_design= p_1*p_2=m/M*n/N_M
for random selection at both stages, and:
p_design= p_1+p_2=〖m*MOS〗_m/(∑_(m=1)^M▒〖MOS〗_m )*n/N_M
if 〖MOS〗_m=N_m equation .. becomes:
p_design= p_1+p_2=〖m*N〗_m/(∑_(m=1)^M▒N_m )*n/N_M

for pps selection at the first stage and random selection at the second stage
when household cluster size constitutes the measure of size. A design as this
one is called epsm. Each unit has an equal chance of selection and resulting
population estimates have a lower variance as. However due to non-response
this result hardly holds in practice. For this purpose design weights commonly
undergo some post-survey non-response adjustment. One such approach is
the calibration of weights to some known population totals as described in the
next section.

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