A series of multi-national surveys based on complex geo-sampling probability-based designs were conducted with the goals of ensuring adequate coverage within each country and facilitating meta-analysis across countries. Such a complex design, where researchers are trying to balance between standardization across countries with different administrative structure and adaptation within each country, is susceptible to increased design effects.
This paper discusses the use of sampling design weights, nonresponse weights, and post-stratification weights, in addition to imputation, trimming, calibration, and raking techniques towards a final combined weights which can be used for analysis within each country and comparisons across countries.
Different weighting techniques are explored and assess in order to identify the optimum weights. Although the design effects for these studies were not unacceptably high before weight-trimming, but modest use of weight trimming was able to lower the unequal weighting effects and help improve the effective sample size.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.