Economical data collected by Statistics Netherlands usually contains missing items. Various imputation methods are available to fill in these gaps, so that completed datasets can be analyzed using...Show moreEconomical data collected by Statistics Netherlands usually contains missing items. Various imputation methods are available to fill in these gaps, so that completed datasets can be analyzed using standard statistical tools. One of the methods often used, the ratio imputation method, appears not to perform very well if we want the completed data to satisfy certain restrictions. This is our motivation to investigate other imputation methods. We look at several methods that we subdivide over two groups. The first group consists of methods based on models that assume a joint distribution for all variables for an individual, and that these variables are all independent. Here we will discuss methods that assumes the data are truncated normally distributed, or exponentially distributed. We propose the proportional variance method, and investigate various possible underlying models. The second group is made up of methods that only specify certain conditional distributions. Here we will investigate the commonly used ratio imputation method and both the classical and the Bayesian variants of sequential regression imputation methods. After we have discussed these methods, we repeatedly apply them to a dataset provided by Statistics Netherlands in which we make a missing pattern ourselves. We use the results of these simulations to assess the performance of the methods on several criteria.Show less
