Estimating student learning value-added models from repeated cross-sections