DAILY VAEIATION— MAY, JUNE AND JULY. 



67 



In these circumstances one would not expect any consistent daily variation of the 

 temperature, and a first glance at the curves for these months given in figure 20, page 54, 

 seems to confirm this opinion. 



A closer investigation, however, shows that the curves are not entirely irregular. By 

 taking the mean of the three months, May, June and July together, we obtain curve 1 in 

 figure 24. This curve has been derived from observations extending over four years, it is 

 therefore the mean curve of 368 days ; 

 hence it should be free from accidental 

 irrf^gularities. This curve shows two remark- 

 able features : 



(a) The temperature is above the aver- 



age from 9 A.M. to 7 p.m. and 

 below for the remainder of the 

 day except for 



(b) a sudden and pronounced rise at 



4 A.M. • 



The question at once arises — -are these 

 features real ? 



The usual method for testing the reality 

 of a feature which is shown in the average 

 of a large number of separate events, is to 

 divide the data up into different blocks and 

 investigate each one separately. In the 

 present case our data are taken from three 

 months' observations. May, June and July, 

 in each of four separate years. A conve- 

 nient method of dividing them up is to 

 combine the three months together for each 

 year giving four separate blocks ; again the 

 data for May from the four years may be 

 combined, and similarly those for June and 

 July giving another three separate blocks. 

 By this method we can see whether the feature looked for is shown in each year and also 

 whether it appears separately in each of the three months. The number of observations 

 which falls in each sub-division is small, therefore the irregularities will not neutralise one 

 another so well as when all the observations are included in one block. We must therefore 

 expect the curves for each sub-division to be more irrcnilar than that for the whole data 

 shown in curve I. But if the features are real they should show up against the irregu- 

 larities. 



We are looking for two features : if (a) is present, i.e., if the temperature is above the 

 average from 9 a.m. to 7 P.M., then in each of the divisions the sum of the depar- 

 tures from the mean between these times should be positive ; if (b) is present and 

 there is a sudden and pronounced rise at 4 a.m., the departure at 4 a.m. should be 

 greater than the average departure at 2 a.m. and 6 a.m.; in other words. 



Fig. 24. Daily variation of temperature. McMiudo 

 Sound. May, June and July. 



departure at 4 a.m. — 



departure at 2 a.m. + departure at 6 a.m. 



should be positive. Thus we have two numerical relationships to look for. 



The data have been divided as described above and the resulting departures for every two 

 hours from the mean of the day calculated. The curves for the mean of each month, i.e., 



