2 - Sample Selectivity and Generalizability of the Results of the Berlin Aging Study

Summary

In epidemiological investigations, one common but rarely analyzed threat to generalizability is sample selectivity or nonrandom sample attrition. In this chapter, we describe our approach to the study of selectivity and provide indepth analyses of the magnitude of sample selectivity in the Berlin Aging Study. Of all individuals eligible for participation (the verified parent sample, N = 1,908), 27% reached the highest level of participation (the Intensive Protocol, N = 516). With respect to levels of performance, projection of selectivity observed on lower levels of participation onto Intensive Protocol constructs indicates that the Intensive Protocol sample was, indeed, positively selected on medical, social, and psychological dimensions. However, the magnitude of observed selectivity effects did not exceed 0.5 standard deviations for any construct. In addition, variances and covariance relations observed in the Intensive Protocol sample were not markedly different from those found at lower levels of participation. We conclude that the degree of selectivity in BASE fell within the usual range and did not result in a decrease of sample heterogeneity. Given the magnitude of sample attrition and the high mean age of the sample, this is a satisfactory result.

Introduction

A major goal in science is to ensure that the validity of empirical patterns does not remain restricted to the observed events, but can be generalized to a larger space of potential measurements.