Abstract:
In the customary ‘circular systematic sampling’ (CSS) scheme with selection ‘probabilities proportional to sizes’ (PPS), with (i) a single random start, (ii) a pre-assigned number of draws and (iii) a preassigned ‘sampling interval’, for certain pairs of units the joint inclusion-probabilities may be zero leading to non-availability of an ‘unbiased’ variance estimator for a linear estimator of a finite population total. To get over this problem a well-known ‘convention’ is to apply the CSSPPS scheme with 2 ‘independent’ random starts. An alternative approach has been taken by the authors with a ‘single random start’ but with a ‘random’, instead of a ‘fixed’ “sampling interval” for which ‘modified CSSPPS’ scheme every pair of units has a positive inclusion-probability admitting thereby unbiased variance estimation. Numerical evidences are presented to examine relative efficacies of some of the competing procedures relevant to CSSPPS suggesting possibilities of improvements over the conventional one by certain alternatives.
Page(s):
259-271
DOI:
DOI not available
Published:
Journal: Pakistan Journal of Statistics, Volume: 19, Issue: 2, Year: 2003