On the influence of bias and of personal equation in statistics of Ill-defined qualities: An experimental study
George Udny Yule · Proceedings of the Royal Society of London Series A Containing Papers of a Mathematical and Physical Character · 1906
Abstract I. This experiment was undertaken to elucidate the real character of such statistics as those of eye-colour, hair-colour, temper, health, etc., which have been given, e. g., by Mr. Galton and by Professor Pearson. The statistics are, it should be noted, not merely statistics of qualities, but of ill-defined qualities, the only guidance to the use of the terms of classification being—with some exceptions—common usage. Strictly speaking we must remember that data so collected are statistics, not of qualities themselves, but of names assigned thereto. It was desired to determine how far the distinction is of importance (1) as regards the naming of single samples: (2) as regards the naming of pairs, two samples of a quality being named more or less together, by themselves, for forming a contingency-table. A matt-surfaced photographic paper was printed by successive exposure to 16 depths of tint, from a slightly impure white to a deep blackish-brown. Small scraps of about ⅜-inch square were cut from each tint, and mounted on cards, two scraps being placed on each card, combined in such a way that every possible combination occurred, making 16 x 16 = 256 cards. Observers were then asked to name the tints on each card under one or more of the following schemes of classification, each observer naming the whole pack:—