Yasumnshi et at: Effect of population abundance and climate on 2 populations of Oncorhynchus keta 
209 
GLS regression modeling, and VAR modeling (Simetar, 
Simetar Inc., College Station, TX; Excel 2013, Microsoft 
Corp., Redmond, WA). 
Model specification and selection was based on a 
Bonferonni correction factor and the lowest Schwarz 
information criterion (SIC) for each of the 12 growth 
variables. The SIC was applied to add a penalty for the 
number of predictor variables in the model. Abundance 
indices were added separately to GLS models because 
of multicollinearity. A Bonferroni correction factor was 
applied to the P-value of the coefficients for abundance 
to account for the number of abundance indices used in 
each model (a=0. 05/number of possible predictors vari- 
ables). The final GLS/VAR models were tested for het- 
eroscedasticity of the residuals (Goldfeld-Quandt test, 
absolute residuals vs. fitted values plot), normality of 
the residuals (qq plots, studentized residuals vs. fitted 
values, and Shapiro test), and serial correlation of the 
residuals (autocorrelation test). Selection of the final 
GLS model and a comparison of the GLS and the GLS/ 
VAR model were based on the SIC. 
To test model performance, an F-test was used to 
determine whether the mean squared error was sig- 
nificantly greater for reserved observations and esti- 
mates than for in-sample observations and estimates. 
If the test statistic exceeded the critical value, model 
performance was judged to have deteriorated over the 
reserved observations, indicative of model misspecifi- 
cation. Lehmann’s correlation was used to determine 
whether the mean squared error was reduced by in- 
tegrating the climate and residuals using the VAR 
models and for determining model performance 
(Lehmann, 1959). 
To compare our findings with other observa- 
tional data, we examined the mean body length 
of juvenile chum salmon and the surface trawl 
catch of juvenile pink and chum salmon from 
continental shelf waters off Icy Point in the 
GOA from 1997 to 2011 (courtesy of the SE 
Coastal Monitoring Program, Alaska Fisheries 
Sciences Center, Juneau, Alaska). 
Results 
Trends in body size and growth 
Trends in annual mean body length at maturity 
for age-0.3 male chum salmon were similar for 
the Fish Creek and Quilcene River stocks but on 
different scales (Fig. 2). From the 1970s to the 
early 1990s, mean length decreased by about 
11.6% (99 mm) for Fish Creek chum salmon 
(from 815 to 720 mm) and by 8.6% (60 mm) for 
Quilcene River chum salmon (from 780 to 720 
mm). The decline was followed by an increase 
in body length in both populations, but the in- 
crease from the early 1990s to 2000 was much 
larger for Fish Creek chum salmon (9.7%) than 
for Quilcene River chum salmon (2.8%). How- 
ever, neither stock had regained body lengths observed 
in the 1970s. 
Trends in SW2 and SW3 incurred offshore were simi- 
lar for Fish Creek and Quilcene River chum salmon 
(Fig. 3). SW2 and SW3 declined from the mid-1970s to 
the mid-1990s and increased from the mid-1990s to the 
mid-2000s. Declines in growth from the mid-1970s to 
the mid-1990s also occurred in SWla, SWlb, and SW4 
for Fish Creek chum salmon but remained low from the 
mid-1990s to the mid-2000s. No trends were observed in 
SWlc for Fish Creek chum salmon and in SWla, SWlb, 
SWlc, and SW4 for Quilcene River chum salmon. 
Growth models 
Growth was linearly and inversely related to both 
chum and pink salmon abundance indices. Growth was 
inversely related to chum or pink salmon abundance 
(or both abundances) for SWlb, SW2, SW3, and SW4 
for chum salmon from Fish Creek and for SW2, SW3, 
and SW4 for chum salmon from Quilcene River. Only 
one growth variable was more strongly correlated with 
climate than chum and pink salmon abundance. SWlc 
for Fish Creek chum salmon was inversely related to 
the magnitude of fall wind speed in the northern GOA 
(Table 2). 
Models for Fish Creek chum salmon 
SWlb was significantly and negatively correlated with 
the estimated abundance of juvenile pink salmon from 
