In current surveys practice the most common method for handling non-response item is simple imputation. The simple imputation methods are the prediction methods for missing data. The imputed values are treated as if they were observed. This results in under or overestimation of the estimator variance, especially if the missing data mechanism is non-random. Simple imputation is inappropriate when the goal is to construct test statistics and confidence of intervals. The paper shows examples of the impact of imputation on the estimates properties. One solution is to use multiple imputation methods that take into account the so-called imputation error.
2
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
This article presents research designs that employ modern statistical tools to optimize costs and precision of research along with some additional methodological advantages. In planned missing-data designs some parts of information about respondent are purposely not collected. This gives flexibility and opportunity to explore a broad range of solutions with considerably lower cost. Modern statistical tools for coping with missing-data, namely multiple imputation (MI) and maximum likelihood estimation with missing data (ML) are presented. Several missing-data designs are introduced and assessed by Monte Carlo simulation studies. Designs particularly useful in surveys, longitudinal analysis and measurement applications are showed and tested in terms of statistical power and bias reduction. Article shows advantages, opportunities and problems connected with missingdata designs and their application in social science researches.
JavaScript jest wyłączony w Twojej przeglądarce internetowej. Włącz go, a następnie odśwież stronę, aby móc w pełni z niej korzystać.