Changes in Census Methods

Abstract

Oriqin of Methodological development. The methodological developments we will describe are those that have taken place in the Census Bureau during approximately the last thirty years. This period has been characterized by substantial changes in the methods of census taking, by the application of survey methods to new subject-matter areas and by an extensive program of research and development. Most of the changes in data-collection methods developed from intensive work on the theory and practical applications of finite-population sampling. Indeed, many of the developments can be viewed as extensions of the principles and methods of sampling theory. Sampling theory provided both a philosophy and a set of criteria for application to many of the problems of survey and census methodology. Sampling provided in addition both models and techniques. All of this has had a profound influence on current methods of census taking. The sampling philosophy and criteria first produced methods of optimizing sample designs. Later came a generalization of this approach, which led to attempts to optimize survey designs in a much broader context. Considerations of optimization created a profound change in attitude regarding both the role of sampling and the relationship of sampling errors to other errors in the census. Thus, for example, starting in 1940 and with increasing emphasis in subsequent censuses, the entire range of population census activities was examined to determine on a logical basis, what were the sources of errors or other inaccuracies in the census and how, for a given total budget, the total impact of all errors combined could be minimized. Before 1940, the emphasis in census taking had been primarily on producing the most accurate census possible, however, with being defined in a rather limited sense. In the main, the problems of accuracy were viewed as those of data processing. Inordinate amounts of money and effort were spent to reduce clerical errors and errors of tabulation. These were the visible errors; the errors whose impact could most easily be seen. Starting with the 1940 Census, this view began to change. Two other aspects of the effect of errors came under the scrutiny. The first was the question of the required accuracy and the extent its improvement is worth additional cost. Attempts to evaluate the cost of achieving different levels of accuracy in comparison with the possible effect of wrong decisions resulting, in turn, from errors in statistics were stimulated by the fact that this type of thinking was fundamental to modern sampling methods. The second was the increased knawledge of the relationship of data-processing errors to other sources of inaccuracy in the censuses, based on the view that the proper methodology was that which