Several continuous covariates show potential heteroskedasticity issues. However, this variability is not consistent when segregated by Koala Management Regions (KMRs).
Below are histograms of each continuous covariates in respective KMRs.
I extracted the estimated transformation parameter, lambda values, from the boxcox transformation for each covariate within each KMR and the combined KMRs.
Based on a typical transformation approach suggested using a set of "standard" transformations instead of using exact lambda value for transformation , below is the suggested transformation for each covariates.
As each KMR has independent risk model, the covariate transformations could be conducted at KMR level. However, transformation (and reverse transformation when interpreting data) could increase the complexity of the analysis.
I am not sure how sensitive are the models constructed with integrated nested Laplace approximation (INLA) to heteroskedascity, and therefore if should we conduct customised transformation to each KMR, uniform transformation across all KMRs or limited transformation to highly skewed covariates.
Several continuous covariates show potential heteroskedasticity issues. However, this variability is not consistent when segregated by Koala Management Regions (KMRs).
Below are histograms of each continuous covariates in respective KMRs.
I extracted the estimated transformation parameter, lambda values, from the boxcox transformation for each covariate within each KMR and the combined KMRs.
Based on a typical transformation approach suggested using a set of "standard" transformations instead of using exact lambda value for transformation , below is the suggested transformation for each covariates.
As each KMR has independent risk model, the covariate transformations could be conducted at KMR level. However, transformation (and reverse transformation when interpreting data) could increase the complexity of the analysis.
I am not sure how sensitive are the models constructed with integrated nested Laplace approximation (INLA) to heteroskedascity, and therefore if should we conduct customised transformation to each KMR, uniform transformation across all KMRs or limited transformation to highly skewed covariates.