IMPORTANCE OF VARIABILITY 91 



three months from October to December. The little figures in 

 the body of the diagram near the zero lines show the number of 

 days on which each point in the various lines is based. Thus, to 

 begin in the upper left-hand corner, during the months of October 

 to December in the eight years under discussion there were only 

 four days when the mean temperature was more than 25° lower 

 than on the preceding day. On those four days the deaths 

 increased by about 3.5 per cent over what they had been on the 

 preceding days. So extreme a drop in temperature is not good. 

 During the same period two days showed a drop of from 21° to 

 25°, and their deathrate averaged 12 per cent less than on the 

 preceding days. Next come nine days with a drop of 18° to 20° 

 and a decline of more than 5 per cent in the deathrate; then 

 nineteen days with a drop of 15° to 17° in temperature and of 

 over 8 per cent in the deaths. Notice that when the temperature 

 do€s not change, there is almost no change in the deathrate, while 

 when it rises, the deaths also increase. 



Figure 20 is perhaps the most significant feature of this whole 

 book. In spite of some irregularity due to epidemics and other 

 accidental causes, each of the four curves has a marked down- 

 ward tendency from left to right. The tendency is strongest in 

 summer when the three days with a drop of 15° to 17° in temper- 

 ature showed a decline of about 40 per cent in the deathrate, 

 while the eight days with a rise of 9° to 10° suffered an increase 

 of 17 per cent in the deaths. This means that there were only 

 sixty deaths when the temperature fell most rapidly, while there 

 were 117, or nearly twice as many, when it rose most rapidly. 

 In spring the difference between a rise and fall of temperature 

 is not quite so great, but if we take the nine days with a drop of 

 13° to 14°, we find an average of only eighty-one deaths (—19 

 per cent) against 114 on the nine days with a similar rise of 

 temperature. 



In general the average slope of the four lines diminishes in 

 proportion to the temperature. That is, changes of temperature 



