Outliers in Exponential Family of Distribution: A Case of Extreme Values in Health Care Expenditure

• Activity Number: 135
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Consulting
• Author(s): Gandhi R. Bhattarai*+
• Companies: OptumInsight
• Keywords: Outliers ; Healthcare Expenditure ; Gamma Distribution ; Program Saving ; Case Management ; Extreme Observations

Abstract

Many outlier detection methods abound for normally distributed data ranging from simple interquartile ranges, standard deviation, etc., to complex multivariate techniques including robust regression or distance matrixes. Often log-linear transformation is used when raw values have non-normal distributions. Healthcare expenditure data generally have skewed distributions and include many extreme values. Traditional approaches of removing outliers in such distributions often suffer from masking effects or remove large proportion of data. There has been some progress in identifying outliers in the exponential family of distribution including the Gamma distribution. This study applies the concepts of Tk statistics proposed by Nooghabi et al. (2010) to identify outliers in real healthcare expenditure data. A two-time period panel data model is developed to predict the savings from a case management program. The standard deviation and root mean square of estimates are derived from bootstrapping with 1,000 resampling and the model performances with and without outliers removed samples are compared.