Page 374 - Invited Paper Session (IPS) - Volume 2

IPS280 VEUN, Thy et al.
survey in Cambodia so far. This frame contains the auxiliary information such
as the names and codes of provinces down to villages, the location codes
whether the village is in the urban or rural area within the province, the total
number of households and population in each village, and the number of EAs
(enumeration area) in each village. Based on the frame, there are total of
14,340 villages and 2,983,350 households across the country of 24 provinces
and one capital. The average number of households (considered as the size
measure in this study) per village is 208 with the minimum of 10 and the
maximum of 4,158 households. The result from the study on the frame shows
that 24 out of 25 provinces and capital have its skewness statistic greater than
1 (1.4 to 7.8) which implies that the number of households in each village of
all provinces and capital, except Kep province with the skewness statistics of -
0.04, are highly and positively skewed distributed. This indicates that the size
measure characterizes a skewed distribution.
The alternative sampling design for the household survey to be proposed
in the study is the stratified three-stage sampling design the same as the
design in the existing survey. The difference between this proposal and the
existing one is that this proposal introduced a further stratification within each
province instead of individual province or grouped province. In this design,
each province frame will be stratified into five strata using the number of
households in each village as the stratification variable. The reason behind the
decision to use five strata in each province came from the experiment on the
trial uses of three, four, and five strata in each province. The results of these
trials showed that five strata gain the highest precision level among the three
options,and can be controllable. As a result, five strata were chosen to be
further stratified in each province in this study. The strata boundaries within
each province were determined using three methods as follows:
1) The generalized Lavallée and Hidiroglou method with Kozak
algorithm (LHK): This iterative method was used to determine the optimal
strata boundaries of the stratifying variable X, the total number of
households in each village, by accounting for the stratum’s relative weight,
mean, and variance. In this method, the minimum sample sizes per stratum
were calculated under three precision levels of 1%, 3%, and 5% CV
(coefficient of variation), and under Neyman Allocation rule. To determine
the stratum boundaries, the population in each province was sorted with
respect to the values of stratifying variable in the ascending order, , … ,
0

where is the minimum value and is the maximum value of stratifying

0
variable. The following steps were followed:
Step 1: Start with some arbitrary boundaries such that < ′ < ⋯ <
0
1
′
′ −1 < , with the constraint ≥ 2; ℎ = 1, … , ; and 2 ≤ ≤ , ℎ =
ℎ

ℎ
ℎ
361 | I S I W S C 2 0 1 9