Interpretation of Correlations.-Godfrey H. Thomson in three papers (49,50,51) has made a fundamental contribution to the interpretation of correlations from the psychological point of view. The fallacy of reasoning from hierarchies of coefficients to the necessity of Spearman's general common factor has led him to elaborate his earlier suggestion of group factors into a wider theory for the explanation of the correlations of mental activities. "The mind, in carrying out any activity, such as a mental test, has two levels at which it can operate. The elements of activity at the lower level are entirely specific, but those at the higher level are such that they may come into play in different activities. Any activity is a sample of these elements. The elements are assumed to be additive like dice, and each to act on the 'all or none' principle, not being in fact further divisible" (51)-On account of the presence of group factors in mental activities a table of intercorrelations tends to take the form of hierarchy as naturally as a frequency distribution of observations takes the form of the normal probability curve. The hierarchy may be due to the sampling of the elements which make up the variates. In mental testing these are not chance samples of the elements but samples chosen to measure different kinds of activities (49). Contrary to Spearman's contention, a hierarchy is proved not even to indicate that a general common factor is the most probable explanation of the relations found.Attention is called by Thomson in a note (50) to the fact that correlations may be produced otherwise than by overlap. For example, in a hand at whist the number of hearts in my hand correlates positively with the number of spades in my neighbor's hand, although my hand may contain no spades and his no hearts. This needs to be harmonized with the frequent interpretation accepted by Otis (31) that "a coefficient of correlation between two series of values is a measure of the percentage of elemental causes common to both." Common causes seems to be a different concept from overlapping observable factors. A common fallacy in the interpretation of partial correlations is discussed by Thomson (50) at some length. The difference between a total correlation between x and y, of .50 and the partial coefficient of .30 between x and y when % is kept constant, does not tell the extent to which the