FISHERY BULLETIN: VOL. 80, NO. 3 



DISCUSSION 



A classification system should have two major 

 attributes: It should be objective and easy to em- 

 ploy, and it should be biologically meaningful. 

 The major impediment to satisfying both con- 

 cerns appears to be the high degree of variability 

 which exists within and between populations of 

 L. pealei. This species has a wide geographic 

 range, from the coast of South America to Nova 

 Scotia (Cohen 1976), and thus populations living 

 in different parts of the range are exposed to dif- 

 ferent and varying environmental conditions 

 (temperature, salinity, photoperiod, and food 

 availability) throughout their respective annual 

 cycles. Such environmental variation is mani- 

 fested by both spatially and temporally varying 

 growth rates, which result in the presence of 

 multiple cohorts within a year class (Summers 

 1968, 1971; Mesnil 1977; Lange and Johnson 

 1979), and by differences in the timing of inshore 

 movement and gonad maturation (Hixon 1980; 

 Macy 1980). Thus it is evident that samples used 

 to construct a sexual development classification 

 should at least reflect both the inshore and off- 

 shore portions of the range. This was done (Fig. 

 1). The Vovk (1972) classification, however, was 

 based only on offshore collections, and as a result 

 probably included relatively few spawning indi- 

 viduals and young of the year in early develop- 

 ment. Summers (1968, 1969, 1971), on the other 

 hand, sampled both inshore and offshore, but dis- 

 tinguished only between mature and immature 

 individuals. 



Objectivity and Utility of 

 the Multivariate Approach 



The multivariate approach to the classifica- 

 tion problem is appropriate and objective when 

 geographic (environmental) variability of un- 

 known magnitude is superimposed on the usual 

 random variation among individuals, producing 

 the observed age or size and reproductive struc- 

 ture of the population or species. This is true 

 because the analytical approach used here can 

 effectively integrate the information and pro- 

 vide a simple numeric classification rule. 

 Growth rates for L. pealei have only been esti- 

 mated from modal size progressions (Summers 

 1971; Mesnil 1977), and hence the age of an indi- 

 vidual or cohort can only be roughly estimated. 

 Since multiple cohorts occur even within a small 

 area (Narragansett Bay), mean and standard de- 



viation values of morphometric characters, such 

 as of DML or GW, have limited value for discrim- 

 ination because of the large variability range of 

 individuals (i.e., multimodality). It is known, for 

 example, that considerable variation exists in 

 the age or size of L. pealei at spawning (Haefner 

 1964; Summers 1971; Macy 1980). Standardiza- 

 tion by the use of ratio parameters or indices may 

 provide a partial solution to the variability prob- 

 lem. 



In the interest of speed and ease of measure- 

 ment, a set of nominal variables to assess the 

 relative abundance of eggs or spermatophores 

 (ASN, ASP, AEO, AEOV; Table 1) was used in 

 addition to the ratio or interval variables (MWI, 

 TLI, SPI, NGI). The "none," "some," "many" 

 rating scale is not strictly objective, but such 

 coarse evaluations can be done reliably with 

 little training and have proved to be valuable dis- 

 criminators. Theoretically it is possible to de- 

 velop an entirely objective classification system. 

 In practice, a somewhat more subjective ap- 

 proach usually proves necessary. What other in- 

 vestigators have found important in distinguish- 

 ing different maturity groups, such as gonad to 

 body weight ratios (Hayashi 1970), must certain- 

 ly influence the initial selection of variables to be 

 measured. The investigator must also decide how 

 detailed a classification is desired and what addi- 

 tional parameters may be required. Repeated 

 use of the exploratory technique of cluster analy- 

 sis followed by stepwise discriminant analysis, 

 as employed here, provides a way of learning 

 how many groups may be present and how to 

 "best" identify them using only those variables 

 which can be shown to significantly aid discrimi- 

 nation. But, at this stage too, the researcher must 

 at least roughly determine the basis for case clus- 

 tering (by size, color, sexual development, etc.). 



Biological Relevance and 

 Accuracy 



Statistically accurate and precise results may 

 prove meaningless in reality. Thus the most im- 

 portant verification of the classification scheme 

 is the demonstration of biological relevance in 

 the appropriate context. In each of the 2 yr when 

 the system was routinely employed, a predictable 

 and logical progression of sexual development 

 from hatching to spawning was observed. More- 

 over, the findings which resulted from this appli- 

 cation of the method are reasonable and have 

 significant, broader implications. 



456 



