INTRODUCTORY NOTE 



Furthermore, Mr. Clayton's early studies proved that the sign of 

 correlation between solar changes and the temperature changes at 

 Buenos Aires alternates between plus and minus from summer to 

 winter. There are transition months, May and October, when it is 

 sometimes indeterminate. Besides this, the months December. Janu- 

 ary, and February have always proved the most unfavorable of the 

 year for the Smithsonian solar radiation work, not only in Chile 

 but in Arizona, on account of disturbed sky conditions. 



It occurred to me, therefore, to analyse Air. Hoxmark's 131 weekly 

 correlation coefficients with reference to the time of the vear. Also, 

 I noted in a cursory inspection that there were few small correlation 

 coefficients, but nearly all were large, either positive or negative. 

 Hence, I rated them according to magnitudes, and kept separated the 

 weeks in which his forecasts correlated with the right sign from the 



Table B. — Siivimary of Tabic A 



weeks in which he erred as to sign. The result seems to me very 

 illuminating-, and is given in table A. Some of the outstanding 

 features of table A are gathered in table B. 

 From this analysis there stand out clearly : 



1. The months December, January and February are the poorest, 

 as was to be expected, because the radiation data are then poorest. 



2. The months May and October, although yielding a large per- 

 centage of high correlations, are unsatisfactory because so many of 

 these large correlations are negative to the event. 



3. All remaining months of the year yield much better results. 



4. It is remarkable that over 60 per cent of the weeks forecasted 

 yield correlation coefficients exceeding 30 per cent. 



5. It is unfortunate that of the large correlation coefficients nearly 

 one third are of the wrong sign. This, of course, brings down the 

 average positive correlation to the low figures of 16 per cent for 

 all months and 21 per cent for the best 7 months, as shown in table A. 



