Multiple imputation of missing data for survey data analysis

Typical management of missing data consists in excluding incomplete observations from the statistical analysis -listwise deletion- which limits its statistical power. A selection bias can also occur if the sample analyzed is no longer representative of the initial sample. Another approach to handling missing data is reviewed. Multiple imputation of missing data allows to keep all data information and to conduct the analysis without loss of power or introduction of bias. An example of application of this method is given with the analysis of risk markers for chicken carcass condemnation at the slaughterhouse.