236 



Fishery Bulletin 100(2) 



August, August-September, and September-October, re- 

 sulting high correlations between adjoining months, and 

 large covariances between distant months (Table 4). 



The prime advantage of a longitudinal study is its ef- 

 fectiveness for studying change, and a repeated measures 

 analysis of variance can be applied to a complete data set 

 with a constant correlation (Diggle et al., 1994). However, 

 our data set was incomplete because the number of anglers 

 who fish in each month changed (see n in Fig. 4) and had a 

 different correlation structure among month (Table 4). We 

 tested the differences between successive monthly catch 

 estimates of season-permit anglers by using a parametric 

 bootstrapping method. In the central limit theorem, the 

 sample distribution of a monthly total catch estimate 

 can be regarded as a normal with the mean C^'-^'and the 

 variance ViCj."). Based on the two point estimates, vari- 

 ance estimates and the correlation coefficient between suc- 

 cessive two months, we generated 10,000 bivariate normal 

 random variables (Gentle, 1998). The means and 95% con- 

 fidence intervals of the differences between two monthly 

 total catch estimates were -226,561 1-650,404-203,080) 



for June and July, 870,720 1455,091-1,277,402] for July 

 and August, 470,594 1161,013-783,1681 for August and 

 September, and 537,488 1290,905-782, 727| for September 

 and October. Significance levels were corrected for multi- 

 ple testing by using the Bonferroni ajustment factor (So- 

 kal and Rohlf, 1995). The confidence interval for June and 

 July straddled 0, showing no significant difference. On the 

 other hand, three other confidence intervals did not in- 

 clude 0; therefore the monthly differences were statisti- 

 cally significant (Fig. 8). 



Discussion 



Bias and source of variation 



The estimate of the total annual catch of ayu by the rec- 

 reational fishery was the first obtained in Japan and was 

 much larger than expected. The total number of day-per- 

 mits sold was 9066, and was quite small (1.9%) compared 

 with the estimated total number of anglers (477,520). 



