Rebutting Existing Misconceptions About Multiple Imputation as a Method for Handling Missing Data

Abstract

Missing data is a problem that occurs frequently in many scientific areas. The most sophisticated method for dealing with this problem is multiple imputation. Contrary to other methods, like listwise deletion, this method does not throw away information, and partly repairs the problem of systematic dropout. Although from a theoretical point of view multiple imputation is considered to be the optimal method, many applied researchers are reluctant to use it because of persistent misconceptions about this method. Instead of providing an(other) overview of missing data methods, or extensively explaining how multiple imputation works, this article aims specifically at rebutting these misconceptions, and provides applied researchers with practical arguments supporting them in the use of multiple imputation.

Routine Cognitive Behavioural Therapy for Anxiety and Depression is More Effective at Repairing Symptoms of Psychopathology than Enhancing Wellbeing

“…Partially offsetting this concern, it is reassuring that an identical pattern of findings emerged when analysing the imputed dataset (effectively an intention-to-treat sample). There is ongoing debate about the reliability of multiple imputation with a high degree of missing data, when the data may be MNAR and when only a limited set of auxiliary variables are available to predict missing values (van Ginkel et al 2019;Jakobsen et al 2017;Madley-Dowd et al 2019). Nevertheless, multiple imputation may provide less biased results than list wise deletion when data are MNAR.…”

“…There was a relatively high degree of missing WEM-WBS data at the final session assessment, so analyses were run on both a complete case basis and a multiple imputation basis (to simulate missing values). This is because there is ongoing debate in the literature about how best to analyse data were there is a relatively high proportion of missing data, where there is reason to think data may be 'missing not at random' (MNAR), and where there is a limited pool of auxiliary variables to use to predict missing values (Jakobsen et al 2017;Madley-Dowd et al 2019;van Ginkel et al 2019). If an identical pattern of findings emerges with both analytic approaches, this suggests that the bias inherent in either method is unlikely to have substantially contaminated the results.…”

Mood Symptoms and Other Maternal Factors in Pregnancy and Adverse Neonate Outcomes

“…Multiple imputation was necessary because ve variables (SGA, Apgar 1 minute, Apgar 5 minute, LBW, pregnancy stress, and maternal age category) had missing values that ranged from less than 1 to 17 percent. Multiple imputation is a principled way of avoiding biased estimates that can result from non-random missingness and overly narrow con dence intervals (55).…”

Missing data imputation by the aid of features similarities

“…Data pattern describes the location of the 'holes' in the data and does not explain why the data are missing. Although the missing data mechanisms do not offer a causal explanation for the missing data, they do represent generic mathematical relationships between the data and missingness (Kang, 2013;Silva and Zárate, 2014;Hamidzadeh and Moradi, 2019;Schmitt et al, 2015;Madley-Dowd et al, 2019;Choi et al, 2019;Aleryani et al, 2018;Perkins et al, 2018;van Ginkel et al, 2019;Wei et al, 2018;Simpson et al, 2019). Three broad types of missingness mechanisms are:…”