452 



Fishery Bulletin 97(3), 1999 



as maturity. The latter is often (Efron, 1975), but not 

 always (Prager and Fabrizio, 1990), the more effec- 

 tive technique. However, it is less robust to viola- 

 tions of the assumption of identical covariance ma- 

 trices. Multivariate normality is also assumed if lin- 

 ear discriminant functions are fitted parametrically 

 (Press and Wilson, 1978), an assumption we avoided 

 in this study (see below). 



For combined gross-histological classifications of 

 maturity, we used predictive discriminant analysis 

 (Huberty, 1994) to distinguish between mature (ripe, 

 ripening or resting) and immature fish. Nonparamet- 

 ric discriminant functions were derived by using the 

 uniform-kernel method with Euclidean distances 

 (Huberty, 1994). We used logistic regression (Hosmer 

 and Lemeshow, 1989) to evaluate whether maturity 

 classifications based on histology could be predicted 

 by the relationships among body size, OW, and OV. 

 Logistic regression coefficients were estimated by 

 maximum likelihood methods. Using each technique, 

 we evaluated the full (three-variable) model first, 

 then a two-variable model without OV. Because 

 sample sizes were small for each species, all sample 

 fish of each species were evaluated histologically. 



Once all individuals had been assigned to matu- 

 rity classes based on histological and gross morpho- 

 metric criteria, we assigned specimens to 2-cm length 

 classes and calculated the percentage mature for each 

 length class using each set of criteria. Percent matu- 

 rity (0-100%) was then related to length class by 

 using nonlinear regression weighted by the square 

 root of the numbers offish in each length class. The 

 logistic model, 



P, = 100/(l+e-'°"''^^'), 



where P = percentage mature in each length class; 

 a and b = fitted parameters; and 



L^Q - (-a/6), was fitted to the data by using 

 maximum likelihood. 



Data were analyzed using STATISTIX ( v 4. 1 ; Ana- 

 lytical Software, 1994: logistic regression) and SAS 

 (V 6.03; SAS Institute, Inc., 1988: PROC REG, GLM, 

 andDISCRIM). 



in Table 2 and used in all subsequent analyses. OFBW 

 was significantly related to FL for each species (ehu: 

 logOFSW=-1.720 + 2.9711ogFL, r-'=0.979, n = 172, 

 P<0.001; kalekale: logOFBW=-1.785 + 3.0101ogPL, 

 r-'=0.966, n=75, P<0.001). 



Effect of ovarian stage on relation 

 of ovary welgfit to body length 



The OW-to-body length relation marginally differed 

 with ovarian developmental stage for ehu and 

 kalekale (FL x stage interaction in ANCOVA; 

 P., j,5=3.19 and ^2,74=2-69, P=0.04 and 0.06, respec- 

 tively; reject Hq. slopes equal). Thus, ovary weights 

 could not be adjusted by body lengths and subse- 

 quently used to distinguish among maturation states 

 for either species. Body length was a poor predictor 

 of ovary weight for both ehu and kalekale (r-=0.58 

 and 0.60, respectively; Table 2). 



Effect of egg size 



FL and OV together, however, accounted for 91% 

 (ehu) and 82.5% (kalekale) of the variation in OW 

 (Table 2). Colinearity between FL and OV (or OFBW 

 and OV) was unimportant. For example, FL ex- 

 plained only 12% of the variation in OV for ehu. 



Histological evidence of maturity 



Most specimens of ehu and kalekale were readily 

 classified as either immature or mature according 

 to histological criteria. Kalekale included fish that 

 were immature, ripening mature, and ripe mature 

 (none were resting mature). Unlike kalekale, a mi- 

 nority of the large (>40 cm FL) ehu were resting 

 mature, with ovaries undergoing substantial atre- 

 sia. The ovaries of kalekale usually contained HYD 

 oocytes if POFs were present (19 of 21 cases). Ehu 

 included specimens with HYD oocytes (7 cases) or 

 POFs ( 98 cases ), but POFs were not observed if HYDs 

 were present (0 of 7 cases). Ehu and kalekale clearly 

 differed in the co-occurrence of HYD oocytes and 

 POFs (Fisher exact test; P<0.001), suggesting that 

 the average spawning intervals of individual females 

 probably differ between these two species. 



Results 



Body size interrelations 



After log-transformation, the influences of body 

 length (FL) and ovary-free body weight (OFBW) on 

 ovary weight (OW) were virtually identical; hence, 

 FL (the easier variable to measure) only is reported 



Statistical classifications of maturity 



For both species, maturity classifications based on 

 discriminant analysis of the three gross metrics (body 

 length, ovary weight, and oocyte volume) generally 

 agreed (=98% correct) with those where histological 

 criteria were used (Table 3). Maturity classes deter- 

 mined by histology also were accurately predicted 



