Abstract
A PAPER on “The Degree of Accuracy of Statistical Data,” by Mr. Carl C. Engberg, has been published by the University of Nebraska. “This paper,” Mr. Engberg tells us, “is written as a protest against the unnecessary refinement of statistical computations as carried out by the biometricians of to-day.” Mr. Engberg complains that the more “prominent biometricians” have worked with five or six figures when they might have worked with three or four with equally good results. He illustrates this by comparing Prof. Pearson's work on enteric fever, published in 1894, with a revision of it by himself using only three places of decimals. He considers that the one is as good as the other. He does not, however, apply the test for relative goodness of fit of observation to theory—Phil. Mag., July, 1900—but discards it without examining the analysis by which it is reached, on the basis of a paradox that he has not been able to see through. He appears to dislike the test because if 16,000 observations are distributed in the same proportions in n groups as 1000 observations the former distribution shows a lower probability for the fit than the latter, if the same curve be used in both cases. This, however, must be right. 16,000 observations should give a result nearer a smooth curve than 1000. The percentage error has been discarded for years by trained biometricians; it was merely a temporary modus vivendi.
Article PDF
Rights and permissions
About this article
Cite this article
Probable Error in Vital Statistics . Nature 69, 93 (1903). https://doi.org/10.1038/069093a0
Issue Date:
DOI: https://doi.org/10.1038/069093a0