Using Principal Component Scores as Stratification Variable: An Alternative to Multiple Frame Sampling Methodology

Type Conference Paper - 59th ISI World Statistics Congress
Title Using Principal Component Scores as Stratification Variable: An Alternative to Multiple Frame Sampling Methodology
Author(s)
Publication (Day/Month/Year) 2013
City Hong Kong
URL http://2013.isiproceedings.org/Files/CPS016-P6-S.pdf
Abstract
In the Philippines, several studies pointed out that samples from Rice and Corn Production
Survey no longer suffice to provide “acceptable” estimates of Livestock and Poultry
statistics. Simulation studies considered the use of inventories of different animals as
stratification variables or size measures. However, the results showed that the use of a
particular animal type would not necessarily yield efficient estimates for other animal
inventories. This leads to the use of multiple frame sampling methodology as an ideal data
collection method which entails high costs. This paper proposes the use Principal
Component (PC)-based scores as the stratification variable as an alternative to multiple
frame sampling to lower the costs because a PC summarizes information contained in a set
of auxiliary variables (e.g. inventory for different animal types); nevertheless, PC’s also
give premium to variables with large variability. Thus, the relatively “rare” animals, i.e.
those present in very few barangays only, can potentially sway the PC to their advantage
at the expense of the more “common” and, perhaps, the more important animal types.
Hence, the authors recommend that only the inventories of the more important animals are
included in the generation of the PC scores that will be used to stratify the population of
interest. Various sampling experiments are performed using barangay level data on
different animal inventories in order to determine the efficiency of the estimates using the
proposed method

Related studies

»