302 HARRIS, KIRKPATRICK, BLAKESLEE, WARNER AND CARD 



Prediction of the production of a subsequent period from the sum of thre( 



monthly records 



The equations required are the following: 



Months from which prediction Period for which predic- 



is made lion is made Prediction equation 



November, Dec. and Jan. February to October E 9 = +126.742 + 0.770 (a + e + ,) 



Dec., Jan. and Feb. March to October 8 = +112.051 + 0.806 (e z + e 3 + e t ) 



Jan., Feb. and March April to October E 7 = +81 .464 + 0.935 (e 3 + e 4 + 6 ) 



February, March and April May to October 6 = + 49.753 + 0.955 fa + e 6 + ,) 



March, April and May June to October E& = +25 .210 + 0.850 (e & + e 6 + 7 ) 



April, May and June July to October E^ = +7 .063 + 0. 770 (e 6 + e 7 + e t ) 



May, June and July August to October E 3 = 1.975 + 0.594 (e 7 + e 8 + e,) 



June, July and August September to October EZ = 4.701 + 0.377 (e + e 9 + e 10 ) 



July, August and September October EI = -3.343 + 0. 172 (e 9 + do + n) 



Table 11 contains the average deviations with regard to sign, of the 

 predicted yield of remaining months, from the actual productions, when 

 prediction is made from the total yield of three consecutive months. 



The deviations range from 0.20 to 3.30 eggs or from 0.27 to 18.22 per 

 cent of the actually observed yield. As far as this criterion shows, predic 

 tions are excellent for all periods from that including February to October 

 to that for August to October. The September-to-October record and the 

 October record, however, cannot be predicted with a high degree of accuracy, 

 the errors being over 17 percent of the mean value for these months. 



The average deviations without regard to sign, shown in table 12, range 

 from 5.24 to 25.93 eggs, the values decreasing as the length of the period 

 for which prediction is made becomes smaller. The reverse is true of the 

 percentage values which increase from 18.66 percent for the period February 

 to October to 89.23 percent of the actual yield for the month of October. 



Similar results are obtained when the formulae are judged by the square 

 root of mean square deviation of the predicted from the actually observed 

 egg record as shown in table 13. These root mean square deviations range 

 from 33.84 for February to October to 6.44 for the month of October alone, 

 or from 24.30 percent for the group of 8 remaining months of the year to 

 109.67 percent for the last (single) month. 



The results for the prediction of two of the groups of remaining months 

 from the combined records of three-months production are represented 

 graphically for the three months November to January in diagram 9 and 

 for the three months March to May in diagram 10. It is the lower figure 

 which is to be consulted in each case. 



The gentle slope of the lines and the considerable irregularities of the 

 means show that prediction of the record of a period of remaining mont 



