Selective non-response to clinical assessment in the longitudinal study of aging: implications for estimating population levels of cognitive function and dementia.