Grant: CIHR Fellowship

Missing data can create significant problems for researchers throughout the health sciences. Missing data arise when participants drop out of a study before completing it, as in longitudinal studies when people move away or become too busy to continue in a study, or when participants skip questions on a survey or parts of a test, because they are bored, or tired, or do not want to answer sensitive or difficult questions. Traditional statistical analysis methods are designed to be used with complete data, so researchers often delete whole cases (participants) who have missing data points. This practice, however, frequently leads to biased results. As such, unplanned missingness can be a serious problem for research generalization. Planned missingness, on the other hand, has tremendous (and for the most part unrealized) potential. With new statistical methods that produce unbiased results, planned missing data designs are an efficient way to manage cost, improve data quality, and reduce participant fatigue and practice effects. To date, however, there is little guidance for researchers on how to design such studies: how to choose the variables which will have missing values, what percent of data should be missing, how large a sample is needed, and to what extent the model will be able to detect meaningful effects. The present proposal aims to provide comprehensive guidance to researchers for using planned missing research designs in longitudinal research. The results will educate researchers on when, how, and why to use planned missing elements in their research designs. Planned missing data designs will increase the validity, generalizability, and cost-effectiveness of applied health research.