We used a method of cross validation to select the number of latent variables in the PLS. A detailed evaluation of the particular cross validation we used reported elsewhere (Huang and Harrington, 2004), and is summarized here. In the semiparametric AFT, the intercept is a nuisance parameter that is absorbed into the error distribution and not directly estimated. The estimated model provides information only on how subjects differ from the overall mean response. Our main objective for the analysis of the ACTG 333 HIV data is to rank the subjects according to their reduction of HIV-1 RNA levels from baseline to week 8, which can be achieved by estimating or predicting the difference (Y — Yj) between two subjects. Specifically, suppose 3 is an estimate of the model coefficients and temporarily assume that the nuisance intercept ao is known. The error in estimating the response difference is
Was this article helpful?