Problem of Missing Data in Questionnaire Surveys[Problém chybějících dat v dotazníkových šetřeních]

Almost any data set can be encountered to the problem of missing data; it is well known in the phenomena relating to people populations and researched in sample surveys. In recent decades, the issue of missing data received considerable attention, because the simple omission of units, for which data are lacking, from the analysis may lead to erroneous conclusions. The approach that accepts the existence of missing data through the modification of the probabilities of units selection with probabilities of obtaining data on them, leads to the construction and use of the weights. Different solution lies in filling in missing data. Using the arithmetic mean or a regression function, recommended for this purpose before, leads at the relevant variables at least to an underestimation of variability; furthermore, it is applicable only for measurable variables. Alternative approaches to missing data are based on the likelihood of collected data assuming some model. Two directions of their development can be distinguished again, estimating population parameters without imputation of missing data on the one hand (EM algorithm) and multiple imputation methods on the other.