Overestimation of mean squares by the method of expected numbers

• 成果类型: Article
• 署名作者: Comstock, RE
• 署名单位: University of Minnesota System; University of Minnesota Twin Cities
• 刊物名称: JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION
• ISSN/ISSBN: 0162-1459
• DOI: 10.2307/2279784
• 发表日期: 1943
• 页码: 335-340

摘要

The actual subclass numbers in the data concerning [sex difference in growth rate of swine] did not deviate significantly, as judged by chi- square, from the expected numbers. However, the method of expected numbers by comparison with that of fitting constants overestimated the mean square for the sex-litter interaction by enough to have a bearing on interpretation of the results and the mean squares for litters by a slight amount. The differences in results of the two methods were interpreted to be the consequence of shifts in weighting squared devations. associated with the shift from observed to expected nuimbers This led to expected results in reasonable agreement with those observed. The overestimation of mean squares reported should not be interpreted as a serious defect of the method of expected numbers. When the condition laid down by [G. W.] Snedecor for the fit of expected numbers to proportionality is met, the maximum overestimation of class and interaction mean squares is not large and the type of data for which a very close approach to this maximum occurs should not be overly frequent. Certainly the advantages cited by Snedecor when contrasted with the defect discussed command the method for wide usage. What has been presented should actually assist in judging in critical cases whether bias from the source considered is important, thereby permitting increased rather than decreased confidence in the usefulness of the method..