MISSING DATA: beyond star trek, the next generation
Drs. Don McLeish and Cyntha Struthers, University of Waterloo
Missing data is ubiquitous. Arguably every dataset from Biostatistics to Banking suffers from missingness, censorship, rounding, coarsening, survivorship bias, non-response bias, or a failure to simultaneously observe all relevant values. This workshop will outline whether and how one should address these incomplete data problems. We begin with an evaluation of the complete case analysis. In this common practice, cases with any missing variables are simply deleted, which may lead to inconsistent estimates. We discuss various ways of dealing with this inconsistency. We examine techniques which adjust for missing data, including reweighting the observed data, and imputing the missing observations under a frequentist and Bayesian paradigm. We will address questions such as the following:
• When do we need to construct a model for missingness?
• Does it matter how we impute missing values? For example, can we just use the last value observed?
• To what extent does our analysis depend on the assumptions made about the missing data mechanism? How sensitive are the results to the model assumptions?
• What are the ramifications for design? When a cheap surrogate variable for an expensive covariate is easily obtained, how can we use it in designing two-stage studies?
• Do we need to adjust for survivorship biases?
• What software is available for the analysis of missing data, and what assumptions are made?
The workshop will include poster presentations by participants; students and post-docs are particularly encouraged to present their research or practicum work, and three poster awards will be given at the closing ceremony. There will be a career panel discussion which is aimed to provide graduate students with career-building advises. The organization of this workshop represents a joint effort between DLSPH and the following organizations: the Southern Ontario Regional Association of SSC, the Southern Ontario Chapter of ASA, the Applied Biostatistics Association, York University and McDougall Scientific.