An Additive Model for a Heavily Right-Skewed Outcome
Allshouse, A. A., Wang, J., Mathias, J., & Irish, W. D. (2007, August). An Additive Model for a Heavily Right-Skewed Outcome. Presented at ASA JSM 2007, Salt Lake City, UT.
When estimating and modeling healthcare costs, challenges arise rooted in the heavily right-skewed nature of the distribution of non-zero annual costs. No consensus currently exists regarding the most appropriate method for analyzing heavily right-skewed data bounded at zero. Traditional approaches to this problem include fitting the logarithmic-transformed data with an ordinary least squares regression, resulting in a multiplicative model when an additive interpretation is desired. An additive model to accommodate skewness and heterogeneity has been developed. Heavily right-skewed non-zero data were generated to emulate health-care costs, and the statistical properties of the additive model compared to the traditional approach were evaluated using simulation techniques.