This comparison is made only for the past 

 4 years, a period when forecasts have been 

 most successful. To go back further is of 

 little use, because methods have changed 

 as new data became available and old data 

 became unsatisfactory. 



It is possible to calculate an empirical stand- 

 ard error for the three methods that have 

 several comparisons, that is, the square root 

 of the average of the squares of deviations 

 of forecast from actual value. Because we are 

 evaluating the method, it is appropriate to use 

 the figure (12.0) for the 1968 forecast based 

 on the yearling estimate. The standard errors 

 are: 



Regression on the kill at 



age 3 and n-iean date 2.6 thousand 



Regression on temperature . . . 7.9 thousand 

 Based on yearling estimate . . . 0.3 thousand 



The first two standard errors are in agree- 

 ment with the theoretical values noted in recent 

 forecasts; the last is much smaller than sug- 

 gested by theory. Two points may be noted. 

 First, because the standard error of 0.3 

 thousand is based on only three comparisons, 

 sampling error may be large, and an extension 

 of the killing season in 1965 and a reduc- 

 tion in 1968 reduced the error in this fore- 

 cast. Second, even if we allowed for departures 

 from routine, the standard error of 0.3 thou- 

 sand as derived fronn the empirical com- 

 parisons would be increased to only 1.3 thou- 

 sand. 



Table 28 shows a comparison of recent 

 forecasts with the actual kill, by method, for 

 3-year-olds. 



FORECAST OF THE KILL OF 4-YEAR-OLD 



MALE SEALS 



Four nnethods used to forecast the kill of 4- 

 year-old males are:(l)Regressionofthekillof 

 4-year-old male seals on the kill of 3-year- 

 old male seals and the mean round of the 

 kill of 3-year-old male seals; (2) regression 

 of the kill at ages 3 and 4 on mean air tenn- 

 perature; (3) regression of the kill at ages 

 3 and 4 on the mean weight of pups and the 

 count of dead pups; and (4) an estimate of the 

 number of yearling male seals. 



Regression of the Kill of 4-year-old Male 

 Seals on the Kill of 3 -Year -Old Male Seals 

 and the Mean Round of the Kill of 3 -Year -Old 

 Male Seals 



This regression uses data from the 1953 

 and subsequent year classes; 5 August was con- 

 sidered the end of the male kill, and adjust- 

 ments were made where necessary. For ex- 

 ample, 18,706 3-year-olds and 13,279 4-year- 

 olds were killed in 1968; an additional 2,000 

 3-year-olds and 680 4-year-olds would have 

 been taken had the kill been extended fronn 2 

 to 5 August, the "standard" termination date 

 of recent years. These estimates are based 

 on an average daily kill of 981 3-year-old 



Table 28 Comparisons of forecasti/ of kill of 3 -year -old male seals, by 



method, St. Paul Island, 1965-68 



Method 



Year 



1965 



1966 



1967 



1968 



-Thousands • 



Regression on mean air 

 temperature 



Regression on pup weight 



24. 0(++) 31.0(+ + ) 21. 2(--) 18.9 



19. 7(--) 24. 7(-) 26. 5(++) 



Regression on the count of 

 dead pups 



19. 2(+) 



Forecast derived from 

 yearling estimate 



Actual kill-'' 



28.6(++) 



19.0 25.5 



27. 



21.7(+) 

 18.7 



1/ Erred positively by a moderate (+) or a large (++) amount or negatively 

 by a moderate (-) or a large ( — ) amount. 



2/ The season for killing males ended 9 August in 1965, 5 August in 1966 

 and 1967, and 2 August in 1968. 



23 



