288 HARRIS, KIRKPATRICK, BLAKESLEE, WARNER AND CARD 



therefore, that the absolute error of prediction will be smaller than when 

 the prediction is made for a longer period. The relative errors of prediction 

 are conspicuously larger than those found when the prediction is made 

 for the year as a whole. Furthermore, these relative (percentage) errors 

 increase as the period for which prediction is made becomes shorter. The 

 test shows clearly that prediction of the results of short remaining periods 

 cannot be made, at least by means of the linear equations for prediction 

 from one month s record tested in this paper, with a satisfactory degree 

 of accuracy. 



When prediction is made for the period subsequent to the (p + l)th 

 month the average deviations with regard to sign vary from 0.56 to 3.32 

 or from 0.50 to 18.74 percent of the actual production for the period. The 

 average deviations without regard to sign vary from 28.47 eggs for the 

 prediction of January-to-October production from November production 

 to 5.67 eggs for the prediction of October production from August record. 

 The percentage values range from 19.38 to 96.59 percent. Similar results 

 are found in the case of the square root of mean square deviation which 

 ranges from 37.05 eggs for the prediction of January-to-October production 

 from November record to 6.92 eggs for the prediction of October produc 

 tion from August record. The percentage values range from 25.2 to 117.8 

 percent of the actual records. 



The graphic representation of the errors of the prediction of the remaining 

 months of the year is made in diagrams 3 and 4. 



The slope of the lines and the moderate narrowness of the shaded areas 

 as well as the fair agreement of the empirical and the predicted means for 

 the remaining periods, evidence for fairly satisfactory prediction for the 

 first six months of the year. As the end of the year is approached anc 

 as the period of remaining months becomes shorter the slopes of the lines 

 are more moderate. The narrowness of the shaded areas, representing 

 the difference between the averages of the errors of over-prediction anc 

 under-prediction, does not indicate great accuracy of prediction as com 

 pared with that attainable in the earlier months, but merely that (because 

 of the smaller egg record made by birds in the latter months of the year] 

 great deviations from prediction are improbable. It is evident, therefore 

 that for the prediction of the record of the later months of the year froir 

 the record of immediately preceding months the equations have relativel} 

 little value. 



It is quite clear that while the prediction of a group of remaining month: 

 may be made with a relatively high degree of accuracy early in the year 

 the predictions are relatively poor toward the end of the year. 



