Power analysis for a linear regression model when regressors are matrix sampled

TL;DR

This paper develops a method to perform power analysis for linear regression models using data obtained through multiple matrix sampling, addressing issues of missing data and estimation precision.

Contribution

It derives asymptotic variances for regression estimates under matrix sampling and demonstrates their application with psychological survey data.

Findings

Matrix sampling causes loss of precision in regression estimates.

Semi-parametric multiple imputation improves estimation with non-normal data.

Sample size requirements are highly variable depending on the parameter space.

Abstract

Multiple matrix sampling is a survey methodology technique that randomly chooses a relatively small subset of items to be presented to survey respondents for the purpose of reducing respondent burden. The data produced are missing completely at random (MCAR), and special missing data techniques should be used in linear regression and other multivariate statistical analysis. We derive asymptotic variances of regression parameter estimates that allow us to conduct power analysis for linear regression models fit to the data obtained via a multiple matrix sampling design. The ideas are demonstrated with a variation of the Big Five Inventory of psychological traits. An exploration of the regression parameter space demonstrates instability of the sample size requirements, and substantial losses of precision with matrix-sampled regressors. A simulation with non-normal data demonstrates the advantages of a semi-parametric multiple imputation scheme.