306 HARRIS, KIRKPATRICK, BLAKESLEE, WARNER AND CARD 



is not so satisfactory as might be desired. Great irregularities are to h 

 expected when a flock of only 415 birds is divided up into such a largi 

 number of classes. The results indicate that if applied to larger flock 

 the prediction equations for the production of a group of remaining month: 

 from three-months recorded production might be made with greate: 

 precision. 



In comparing the error of prediction for a group of subsequent month 1 

 from three-month periods with the errors of prediction for the same perioc 

 when prediction is made from single months we require three sets of equa 

 tions for the prediction of the yields of a group of months from a monthlj 

 record. Two sets of these have been given on page 282 and used in com 

 parison with the results of prediction from bimonthly periods. 



The additional equations required are : 



Month from which predic- T&amp;gt; i t 7 r . 3- / i n i- , 



tion is made Period for which prediction ts made Prediction equation 



November February to October 9 = +134.546 + 1 . 143 d 



December March to October 8 = +122 .721 + 1 . 165 ej 



January April to October E 7 = + 106 . 659 + 1 . 033 e, 



February May to October 6 = +80.512 + 1.342 e 4 



March June to October 5 = +47.994 + 1 .462 e 5 



April July to October 4 = +30. 111 + 1. 2 19* 



May August to October E z = +13.094 + 1 .008 e? 



June September to October &amp;gt; = +3.462 + 0.649 e 8 



July October , = +0.297 + 0.240 e t 



When we compare the results for the prediction of the yield of a grouj 

 of subsequent months from single monthly records and from trimonthh 

 records of production we find that the differences in errors of predictioi 

 are surprisingly small. Specifically we note that in the case of the averagi 

 deviation with regard to sign, shown in the two last columns of table 11 [ 

 the differences in actual errors range from 0.03 to 3.42 eggs while the differ 

 ences in percentage values range from 0.05 to 10.55. In some cases th i 

 three-month period gives a numerically larger error of prediction whil&amp;lt; 

 in other cases the one-month period gives the larger error. 



When the comparison is made on the basis of average deviation withou 

 regard to sign (table 12) the single-month period gives a slightly large 

 average deviation in most cases, 23 out of 27 cases. The differences 

 however, very small, varying from 0.12 to 1.14 eggs. 



Similar results are obtained when the comparisons (between the singl 

 component months and the three-months record as bases of prediction 

 are based upon square root of mean square deviation (table 13). In 2. 

 of the 27 cases prediction from a monthly record gives slightly more variabl 



