Figure 3 • "Catch-per-hundred-hooks (in number of fish) computed as an 

 average-of-rat ios statistic and a rat io-of-averages statistic for each 

 montin in Marsden square 130 (lat. 30°-'4O° N., long. \kO°-\50° E.), 

 1949-61. 



Thus we examined the trend in CPUE over 

 the years for each January , each February, etc. 

 Examination of this trend was accomplished by 

 computing for each month linear and quadratic 

 regression of average CPUE on year for each 

 Marsden square that had observations on a suf- 

 ficient series of years to warrant study. The 

 Marsden squares that contained a sufficient 

 series of data define the albacore grounds and 

 are bounded by lat. 20° to 40° N. and long. 130° 

 E. to 170° W. The quadratic regressions exhib- 

 ited significant reduction (at the 5-percent 

 level) in sum of squares in 23 of the 112 re- 

 gressions. Owing to the small number of sig- 

 nificant quadratic regressions and the close- 

 ness of the average values of the quadratic re- 

 gressions to the average values of the corre- 

 sponding linear regressions we decided, for 

 simplicity, to consider only the linear regres- 

 sions in our analyses. Thus each month's 

 linear regression line was used to obtain an 

 estimate of the average value of CPUE at the 

 time of initial fishing in 1949 and for 1961, the 

 last year for which we have data. This average 

 value for apparent abundance was greater in 

 1949 than 1961 with only a few exceptions (fig. 

 4): these exceptions were for nonpeak months. 

 The fact that the estimated apparent abundance 

 in 1949 was almost always greater than that for 



1961 indicates that except for the few noted ex- 

 ceptions all of the slopes were negative and 

 that, on the average, apparent abundance de- 

 clined throughout the study period. 



We also considered the problem of evaluating 

 the variability associated with each regression 

 relative to the magnitude of the slope for each 

 regression. We realized a priori that in in- 

 stances many of the regression slopes would 

 not be significantly different from zero owing 

 to the small sample sizes (a maximum of 14 

 data points for each regression). Significant 

 differences from the hypothesis of zero slope 

 were determined by the usual t-test at the 5- 

 percent level. We note--with respect to the 

 usual assumptions involved in these tests of 

 significance--that each datum is based on an 

 average so the probability distribution of the 

 data must tend toward a normal distribution; 

 that since the sample sizes within each month 

 tend to be approximately equal, the assumption 

 of homogeneity of variance appears to be fairly 

 well approximated; and that consideration of 

 each month separately tends to favor indepen- 

 dence of errors among the data. We conclude 

 from the consistently negative slopes and the 

 small sample sizes that the percentage of sig- 

 nificant slopes (29 percent) would increase if 

 sample sizes could have been larger, suggest- 



