Please join us for the upcoming Harvard Catalyst Biostatistics Journal Club. Alan Fossa (BIDMC) will lead the discussion entitled “Examining the conventional wisdom of modeling correlated data: An example studying the effect of shift hour on the likelihood of hospital admission from the BIDMC Emergency Room.”
The ability to model cluster correlated data using generalized estimating equations with working correlation structures is a useful tool for appropriate error estimation. Conventional wisdom says that this practice should yield similar effect estimates while capturing the appropriate amount of statistical information present in our data. In practice, effect estimates obtained through modeling data as correlated are rarely identical to those obtained by traditional methods. What should be done when small changes in effect estimates are meaningful and how do we reconcile these differences with our need for appropriate error estimation? In this journal club we will discuss the merits and limitations of traditional estimation, correlation structures, and stratification in the context of an exemplary analysis investigating the likelihood of hospital admission from the Beth Israel Deaconess Medical Center Emergency Department.
To facilitate the upcoming discussion, please read the following paper:
Hanley JA, Negassa A, Forrester JE. Statistical analysis of correlated data using generalized estimating equations: an orientation. American journal of epidemiology. 2003 Feb 15;157(4):364-75.