| Abstract: | Survey researchers can choose (1) to maximize their accuracy in estimating the relations among a few abstract variables or (2) to maximize the number of variables covered. In the former case, many indicators per variable are required; in the latter case, few or one. Lazarsfeld's "doctrine of interchangeable indices" offers some support for the second approach, but this doctrine is shown to neglect or perpetuate three kinds of measurement problems: (1) attenuation of the relations between unmeasured abstract variables, (2) correlated measurement errors, and (3) low data quality and imprecision. An alternative form of the doctrine is presented which is more compatible with the implications of causal models. |
