Page 325 - Contributed Paper Session (CPS) - Volume 6

CPS1942 Daniel D. M. P.
(5% and 10%), SRSWOR appear to generate less bias across variations of
auxiliary variable, particularly at small values of sd(X). This is not surprising
since SRSWOR estimates work best under homogenous populations.

Table 3.4 shows the average absolute percentage of SRSWOR, PPAS, and
PPSS estimates across different error multiplier (k). It is given that as (k)
increases the model fit suffers and so does modelassisted estimation
techniques. At 1% sampling rate, PPAS produces comparable results with PPSS
and superior results when compared to SRSWOR estimates. At 5% and 10%
sampling rates, PPAS estimates generate more bias as misspecification
increases. Conversely, the relative bias of SRSWOR estimates remain roughly
the same. This is expected since auxiliary information does not affect selection
of SRSWOR samples.

Table 3.5 summarizes the average absolute percentage of SRSWOR,
PPAS, and PPSS estimates across varying covariate effect. A high covariate
effect (b) increases the magnitude of linear association between the auxiliary
and target variable. For higher covariate effect (b=5), PPAS estimates generate
lower Average . This is because the increase in magnitude of Y dominated
the error settings. Also, it was noted previously in Table 3.2 that increasing the
covariate effect improved the efficiency of PPAS estimates, under similar
model fit. In other words, an increase in covariate effect decreased bias and
improved precision of the estimate, particularly under good model fit.

4. Conclusions
The findings of the study are summarized as follows:
1. PPAS estimates are more precise than SRSWOR and PPSS, particularly
in populations with less variability.

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