But, in our specific dataset (the Baby Connectome project), some kids were sampled using an instrument that isn't normed for their age range (they were too old), so we also created a set of models to predict vocabulary score from an undersampled score with a possible ceiling effect) ...