Skip to Main Content
 

Global Search Box

 
 
 
 

ETD Abstract Container

Abstract Header

Linear Regression Analysis Using Survey Sample Data: An Evaluation Of Diagnostic Tests For The Use Of Weights

Abstract Details

2020, Master of Science, Ohio State University, Public Health.
Every day, surveys are used to gather information about a population to better understand and improve the world we live in. Many times, though, the sample obtained does not mirror the population and adjustments must be made so that the sample is a good representation of the whole. One of these adjustments occurs through the use of weighting. Though several tests have previously been proposed to determine whether or not weights are informative for estimating a linear regression model for a given sample, there are gaps in the literature as to the performances of these tests. We have designed a Monte Carlo simulation study to assess the methods proposed by DuMouchel & Duncan (1983), Pfeffermann (1993), Pfeffermann & Sverchkov (1999), and Pfeffermann & Sverchkov (2007) in terms of Type 1 error and power. The degree of correlation between the predictor and outcome and the amount of bias in the selected sample’s intercept and slope was varied to evaluate the performances in several scenarios. We then examined the performances of these tests in a real-world application examining the association between media exposure and mental distress among U.S. adults at the beginning of the COVID-19 pandemic using data from the Understanding America Study. Among the four overall tests, DuMouchel & Duncan (1983) and Pfeffermann & Sverchkov (1999) were found superior to Pfeffermann (1993) and Pfeffermann & Sverchkov (2007) with low Type 1 error and high power. The overall tests of DuMouchel & Duncan (1983) and Pfeffermann (1993) were then modified to assess tests for the slope only and the intercept only. The DuMouchel & Duncan (1983) test was found superior with both close-to-nominal Type 1 error rates and as high or higher power than that of Pfeffermann (1993) regardless of the bias in selected sample coefficients or correlation between outcome and predictor. We conclude that DuMouchel & Duncan (1983) and Pfeffermann & Sverchkov (1999) are sensitive enough to truthfully indicate whether or not weighting is appropriate in a variety of settings and are the preferred tests.
Rebecca Andridge (Advisor)
Kellie Archer (Committee Member)
54 p.

Recommended Citations

Citations

  • Keiter, A. (2020). Linear Regression Analysis Using Survey Sample Data: An Evaluation Of Diagnostic Tests For The Use Of Weights [Master's thesis, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu1603969812675187

    APA Style (7th edition)

  • Keiter, Ashleigh. Linear Regression Analysis Using Survey Sample Data: An Evaluation Of Diagnostic Tests For The Use Of Weights. 2020. Ohio State University, Master's thesis. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=osu1603969812675187.

    MLA Style (8th edition)

  • Keiter, Ashleigh. "Linear Regression Analysis Using Survey Sample Data: An Evaluation Of Diagnostic Tests For The Use Of Weights." Master's thesis, Ohio State University, 2020. http://rave.ohiolink.edu/etdc/view?acc_num=osu1603969812675187

    Chicago Manual of Style (17th edition)