Enhancing Sampling Weights Using Raking Method

• Activity Number: 334 - New Methods for Survey Weighting and Variance Estimation
• Type: Contributed
• Date/Time: Wednesday, August 5, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Survey Research Methods Section
• Author(s): Yan Wang* and Di Xiong and Kathy Phipps and Steve Lee and Jie Shen and Abrey Danial and Maritza Cabezas and James Crall and Honghu Liu
• Companies: University of California, Los Angles and UCLA and Los Angeles County and UCLA and UCLA and Los Angeles County and Los Angeles County and UCLA and UCLA
• Keywords: sampling weight; survey design; raking; non-response; population; balance

Abstract

Background: Survey samples are usually designed based on available information from population. Designs based on systematic sampling rarely consider variables related to non-response and non-coverage. However, sample weights can be modified when population-level variables are available such as race/ethnicity and socioeconomic status.

Methods: Raking method or sampling balancing is used to adjust design weights according to available marginal distribution of demographic variables at the population level. The weights are modified to estimate ratio variables (e.g. prevalence). We will show that the raking method improved the estimation.

Results: This method is applied to a 2018-2019 children’s oral health needs assessment for kindergarten and third grade students in schools throughout Los Angeles County. The design weight is calculated from systematic sampling. We select variables that are related to non-response and non-coverage to balance the sample. The estimations by both methods are compared. We will report the estimation from both methods.

Conclusion: Raking survey data can modify sampling weights based on the information available at the population level, i.e. marginal population-level distribution. By accounting for non-response and non-coverage, the estimation is more accurate.