Nearest neighbour in least squares data imputation algorithms for marketing data

Marketing research operates with multivariate data for solving such problems as market segmentation, estimating purchasing power of a market sector, modeling attrition. In many cases, the data collected or supplied for these purposes may have a number of missing entries.The paper is devoted to an empirical evaluation of method for imputation of missing data in the so-called nearest neighbour of least-squares approximation approach, a non-parametric computationally efficient multidimensional technique. We make contributions to each of the two components of the experiment setting: (a) An empirical evaluation of the nearest neighbour in least-squares data imputation algorithm for marketing research (b) experimental comparisons with expectation–maximization (EM) algorithm and multiple imputation (MI) using real marketing data sets. Specifically, we review “global” methods for least-squares data imputation and propose extensions to them based on the nearest neighbours (NN) approach. It appears that NN in the least-squares data imputation algorithm almost always outperforms EM algorithm and is comparable to the multiple imputation approach.