Online Program

Much Ado About Almost Nothing: How to Deal with Limited Data (302943)

Keywords: messy data, incomplete data, missing data, exact methods

In this era of Big Data, applied statisticians are still faced with situations in which there appear to be no data, or data of only limited usefulness. For example, when attempting to find a confidence interval for a binomial proportion, the sample may contain no successes. Such a scenario could be encountered, for example, when attempting to estimate the incidence of an extremely rare side effect associated with the administration of a newly developed drug. Other statistical inference situations in which there may be no or only limited data include estimating an odds ratio when one of the cells in a 2x2 table is empty, estimating Cohen’s measure of agreement—kappa—when one of the cells (or perhaps a row or column total) is zero, and incorporating observations below the limit of detection of a measuring device into a statistical analysis. In this presentation, the author illustrates each of these scenarios and others with real data he has encountered in his consulting practice and describes statistical methods that can be used to deal with them.