Work

Using the missing indicator method on a dependent binary variable, a simulation study on bias and equivalence of model parameter estimates

Year: 2016
Type: dissertation
Abstract: When doing a study, researchers are often confronted with missing values after (or during) collecting the data.This can occur because of many reasons, and there are many solutions to handle missing values.One of the solutions is the missing indicator method, in which a new response category is introduced (when applied to nominal data) which indicates whether a value is missing or not (for example: 0 = NO, 1 = YES, 2 = MISSING).This method is infamous for resulting in biased estimates when applied to independent (predictor) variables.Little research is done on how the missing indicator performs when applied to missing values in a dependent (outcome) variable.In this study a Monte Carlo simulation is performed in R, to see if bias is introduced when this missing indicator method is used.Simulations are performed for two designs: cross-sectional and longitudinal.Initial results from a comparison by an ANOVA show there is almost no bias introduced when the missing indicator method is applied to a dependent binary variable in a logistic regression model.However, further analysis using an equivalence-study approach, show that not in all cases the coefficients can be considered to be equal.
Author: Rambhadjan, Jermain
Institution: Leiden University
Language: English
Open Access status: green
OpenAlex ID: W6987842505
Cites: 18
Cited by:
Related to: 9