122 ANNUAL REPORT SMITHSONIAN INSTITUTION, 19 3 5 



pressure with a positive fluctuation is greater as far east as Vienna 

 and as far west as the Bermudas than it is at the Azores. There 

 is also to be seen in the second chart conspicuous warmth in the 

 east of the United States as well as in northwest Europe, and 

 marked cold to the southeast of the Mediterranean as well as along 

 the northeast of North America. On rainfall, in the lowest chart, 

 the influence is less wide-spread. The small amount of persistency 

 is shown in figure 3. The first of its three graphs shows how close 

 are the relationships of pressure in December with the figures ex- 

 pressing the fluctuations of the North Atlantic in that month; the 

 second and third, which give the relationships of pressure and tem- 

 perature in January with the fluctuations of the oscillation of the 

 December before, show that little effect of the December conditions 

 continues after a month. 



The more critical in my audience may object that if you are suffi- 

 ciently astute in choosing your successive numbers for the fluctua- 

 tion you can make a certain amount of agreement with any system 

 of pressures and temperatures ; and to this the reply is that the fit is 

 very much closer than can be explained in this vfny. Others may 

 urge that all these arguments are merely numerical, and quote the 

 jibe that by statistics you can prove anything. But if you wish to 

 understand phenomena you must collect the facts, and if they are 

 nimierical it is only in the very simplest of cases that you can see 

 relationships by merely plotting curves and comparing them. Statis- 

 tical methods are inevitably forced on us by common sense when we 

 want accurate and reliable inferences from series of data, just as a 

 sextant is forced on a sailor when he wants to determiue accurately 

 the altitude of the sun. One who has lost an important lawsuit, 

 owing to the ingenious argument of the opposing counsel, may object 

 that by logic you can prove anything; but that is an indequate de- 

 fense for being illogical on all occasions. As a matter of fact, when 

 studying relations of cause and effect statistical methods show us 

 what quantities vary together, but strictly by themselves they tell us 

 nothing as to causation. If we compare heights of fathers and sons, 

 we learn that tall sons have tall fathers; but in spite of that fact 

 we are not convinced that the child is literally father to the man. 



Let us consider an example from data published in 1906 regarding 

 unemployment and illiteracy as measured by the percentage of per- 

 sons who could not sign their name in the marriage register (fig. 4). 

 Clearly the correlation coefficient between these two factors might 

 lead to most undesirable inferences regarding the usefulness of edu- 

 cation. But we could not expect to arrive at the truth if we ignored 

 such an important fact as the amount of trade, and on admitting the 

 data of this factor we gee at once that faith in the value of our ele- 



