Use of non-parametric methods for the imputation of missing data. A comparison based on extensive Montecarlo simulations