Semi-parametric selection models for potentially non-ignorable attrition in panel studies with refreshment samples