Estimating the expected effects of large genetic deletions or duplications on a trait such as IQ can be challenging since most large copy number variants (CNVs) are extremely rare or often unique. However, this question is of great interest to clinicians in pediatric settings. Large CNVs can contain many genes, and the severity of the impact of a CNV is well-known to be associated with the number of genes affected, and the size of the CNV. Using several different kinds of annotation information for the genes in CNVs, we have implemented hierarchical Bayesian models to try and improve the estimates of the effect of a CNV on performance IQ and verbal IQ, and to find association genomic regions. We explore a variety of models and transformations of the annotation information, and illustrate results in population-based cohorts: Imagen and the Saguenay Youth Study.