Missing data in multiple correspondence analysis under the available data principle of the nipals algorithm

Abstract

Multiple correspondence analysis (MCA) in the presence of missing data is usually performed by removing the records that have missing or not available (NA) data; sometimes, an entire row or column of a data matrix is removed, which is not ideal because relevant information on an individual or variable of the study is lost. In some cases, it is assumed that the missing data are a category of the qualitative variable, resulting in a greater variance dispersion in the new axes. Possible solutions to this problem can be the imputation of the missing data or using an algorithm suited to the presence of this type of data. This work is focused on performing the MCA method in the presence of missing data, without using imputation techniques, by using the available data principle of the nonlinear estimation by iterative partial least squares (NIPALS) algorithm [25].