Factors in the Variability of Response in Enumerative Studies
Gladys L. Palmer · Journal of the American Statistical Association · 1943
THERE ARE SEVERAL sources of error in enumerative studies. The error that arises in sampling is well recognized, but other sources, which may be more important, are often overlooked, even though they are present as much in complete coverages as in samples. For instance, there is a variability of response to be reckoned with even within the framework of well defined schedule concepts and well trained enumerators. Poorly worded schedule questions and inadequate training of enumerators not only intensify this variability, but are the cause of unreliability in the sense of biases. However, in the absence of a standard operation, which by definition would yield no error and would enable one to differentiate correct from incorrect results, and thus to recognize biases, it is perhaps better to speak of variability of response merely as a source of difference between the results of two or more coverages of the same universe, rather than to speak of a bias in one or both coverages. The purpose of this paper is to present some results on variability of response.' Relatively little is known about the degree of variability of response for different types of information or its relationship to other conditions of enumeration because the opportunity to measure it is limited to studies in which two sets of interviews requesting the same information for the same persons for the same period of time have been made. Relatively high costs have discouraged this kind of enumerative study while unfortunately, at the same time, differences in the results from applications of parallel techniques to the same universe for measuring the same phenomenon, without adequate explanation of the causes of these differences, have led to public confusion. As a consequence, doubt has been cast on the results of enumerative studies, yet these are the