Each week, we publish layman’s abstracts of new articles from our prestigious portfolio of journals in statistics. The aim is to highlight the latest research to a broader audience in an accessible format.

Manuel, C., Sinha, S. and Wang, S. (2022), Reducing bias due to misclassified exposures using instrumental variables. Can J Statistics. https://doi.org/10.1002/cjs.11705

The article featured today is from the Canadian Journal of Statistics with the full article now available to read here.

Reported values of variables are often erroneous- measured with uncertainties. For a discrete variable, this uncertainty may show up as misclassification. In studying the association between the variable that is measured with errors and a response (outcome) variable through a regression, one must make proper adjustments of the misclassification; otherwise, results will be biased. This paper contains sufficient conditions for identifying the parametric regression model for a binary/dichotomous response variable when a set of instrumental variables are observed along with the response and erroneous measurements of the variable of interest, and other covariates that are assumed to be measured without any errors. The instrumental variables must be related to the variable of interest. This paper contains two Bayesian approaches of analyzing such data with a misclassified exposure using the MCMC method and
the variational inference technique. To illustrate, the proposed method was applied to the SEER breast cancer data to study the association between 5-year survival and the treatment therapy after adjusting the effect of observable confounding variables.
The reported treatment therapy in the SEER database was subject to misclassification, and there was validation data to assess the degree of misclassification. In this analysis, insurance status was used as an instrumental variable for the treatment therapy. For practitioners’ convenience, the paper is also accompanied with R code so that anyone can apply the proposed analysis techniques on their own data.