LIN and WILLIAMS: INFLUENCE OF SMOLT ON ADULT SALMON 



dence hypothesis can be rejected at a 1% level 

 with a critical value of 9.21 at 2 degrees of free- 

 dom. 



For comparison purposes, the Hokkaido chum 

 salmon data was used to fit the multiplicative-error 

 model (model (2)) by applying the weighted least 

 squares method with the following results: 



ln(A) = -3.27 + 0.583 ln(5) 

 (2.06) (2.06) 



(9) 



R- = 0.19, Durbin-Watson = 1.15, Log-likelihood 

 = -11.22. 



By comparing the values of the log-likelihood func- 

 tion of Equations (7) and (9), it can be concluded that 

 the multiplicative-error model can be rejected a 5% 

 significance level. Even though the multiplicative- 

 error model produces a bigger log-likelihood value 

 than the linear model, the difference between these 

 two log-likelihood values is not statistically signifi- 



(0 



c 

 o 



c 



CD 

 (0 



T3 

 < 



3 - 



1 



140 190 



240 290 340 390 440 

 Fry Released (millions) 



490 540 



Figure 1.— Plot of Hokkaido chum salmon data and estimated relationships: 1950-1969. Dots are actual obser- 

 vations, linear model is Equation (8), multiplicative-error model is Equation (9), and Box-Cox model is Equation (7). 



659 



