Effectiveness of Different Missing Data Treatments in Surveys with Likert-Type Data: Introducing the Relative Mean Substitution Approach

Abstract

This article introduces a new approach to the substitution of missing values in surveys with Likert-type scales: relative mean substitution. The effectiveness of this method is demonstrated in comparison with three other commonly used methods for dealing with missing values, making use of actual field data. The emphasis is on two aspects of global effectiveness: (a) the accuracy in estimating various parameters at the same time and (b) the accuracy in estimating, for Likert-type scales with different psychometric characteristics, these various parameters under different conditions, such as different numbers of respondents (1,674; 400; and 100) and different distributions of missing values (two random and three nonrandom situations). The results indicated that this new relative mean substitution approach globally produced the most accurate estimates, mainly because of the more accurate estimation of the variances and the sensitivity to items with deviating means, provided that the Likert-type scales are sufficiently homogeneous.