The identification, impact and management of missing values and outlier data in nutritional epidemiology.
Rosa M. Abellana, Andreu Farrán-Codina · PubMed · 2015
When performing nutritional epidemiology studies, missing values and outliers inevitably appear. Missing values appear, for example, because of the difficulty in collecting data in dietary surveys, leading to a lack of data on the amounts of foods consumed or a poor description of these foods. Inadequate treatment during the data processing stage can create biases and loss of accuracy and, consequently, misinterpretation of the results. The objective of this article is to provide some recommendations about the treatment of missing and outlier data, and orientation regarding existing software for the determination of sample sizes and for performing statistical analysis. Some recommendations about data collection are provided as an important previous step in any nutritional research. We discuss methods used for dealing with missing values, especially the case deletion method, simple imputation and multiple imputation, with indications and examples. Identification, impact on statistical analysis and options available for adequate treatment of outlier values are explained, including some illustrative examples. Finally, the current software that totally or partially addresses the questions treated is mentioned, especially the free software available.