On Analytical Interpretation of Straw-Vote Samples
William Leonard Crum · Journal of the American Statistical Association · 1933
It is the purpose of this paper to bring together some of the more important theoretical considerations concerning the validity and significance of straw polls, and to present and discuss selected experimental analyses. From the point of view of the technical statistician, the essential problem involved in the use and interpretation of strawvote data is a problem in sampling. The principal statistical operations useful in the analysis of straw-vote results are accordingly those which look to the criticism and rectification of the sample upon which the straw returns are based. The straw poll inevitably involves sampling, and on three chief counts. Such a poll, to be of practical use, must be held in advance of the actual election; and therefore, even though it succeeded in covering every voter who subsequently participated in the actual election, sampling errors would arise because of changes in voting intentions. Such a poll must be based upon a selection from the qualified voters; and, even though it covered them perfectly, it would be only a sample of those voters who actually participated in the election. Such a poll, for obvious practical reasons, can not be a perfect report of the entire qualified electorate; and therefore sampling difficulties, in the ordinary sense of incomplete reporting, must arise. Statistical theory suggests that the best sample to use in a case of this sort is a random sample-the term random being understood in the technical sense, and not in the loose sense frequently adopted by lazy or careless investigators who ignore or are unaware of the imperfections of the sampling processes which they use. The straw-vote problem is, unfortunately, not a simple case of random sampling; for the total is not a homogeneous body, in that there are within the population several fairly distinct groups which presumably have different voting tendencies. In this respect, the problem is somewhat similar to that of commodity prices: in this problem, as in that problem, the sampling procedure must cover adequately each of the more important groups within the total population. Any such group may be fairly homogeneous within itself, and the ordinary method of random sampling can be used in making selections from that group. So far