Bayesian analysis of Box–Cox transformed linear mixed models with ARMA() dependence

A robust approach tot linear mixed models applied to multiple sclerosis data

“…The dependence structure of C i can be extended to a high order autoregressive moving average (ARMA) dependence as provided by Rochon [7], Lin and Lee [8] and Lee et al [9].…”

Bayesian inference for multivariate meta‐analysis Box–Cox transformation models for individual patient data with applications to evaluation of cholesterol‐lowering drugs

“…For example, trials 15, 17, 20, 22, 23, and 24 only included CHD patients, whereas trial 21 had no CHD patients at all. Also, there was only medium statin potency in trials 13,15,16,17,21,23,24, and 25, whereas there were no low or high statin potencies in some other studies. We further observe that the proportions of DM patients and the distributions of race were quite different across the 26 trials.…”

“…Lee et al . carried out Bayesian analysis of Box–Cox transformed linear mixed models, and Gottado and Raftery developed a Bayesian approach for simultaneous variable and transformation selection. Because of the complexity of Box–Cox transformation models, Bayesian methods may be preferred over the classical methods owing to the recent advance in Bayesian computation and the recent development of Bayesian model comparison criteria.…”

Extending imputation methods for non-normal hierarchical data: an application to a longitudinal survey in the Philippines

“…One method is to normalise these variables by applying a transformation before performing imputation (Goldstein, Carpenter, Kenward, & Levin, 2009;Goldstein, Carpenter, & Browne, 2014). However, studies have shown that this approach may not always yield a valid result and the variable transformation may not always be applicable in some substantive model specification (Box & Cox, 1964;Gurka, Edwards, Muller, & Kupper, 2006;Lee, Lin, Lee, & Hsu, 2005). von Hippel (2012) showed that using variable transformation/back-transformation to fill-in values for non-normal continuous variables can lead to distortion of the true distribution, and may further exacerbate the bias introduced by the missing data.…”

“…Hence, when dealing with nonnormal outcome variables, transformation, such as the logarithmic or the Box-Cox transformation (Box & Cox, 1964), is commonly applied with the goal of correcting the non-normality of the model's error term. However, a number of studies have shown that variable transformation may work in some instances but not for all cases (Box & Cox, 1964;Gurka et al, 2006;Lee et al, 2005;von Hippel, 2012). In addition, there are situations where applying the transformation to correct the non-normality can lead to inaccuracy and introduce bias.…”