9-33 



APPENDIX 



The purpose of this statistical appendix is to clarify the 

 basic assumptions and constraints relevant to the tests of specific 

 hypotheses carried out on the data. The 3-way G test used to evaluate 

 hypothesis (1) had no constraints other than those already described in 

 the Methods Section. 



The statistical analysis used to test hypothesis (2) con- 

 tained some bias. The assiomptions underlying a covariance analysis 

 include the usual ones of normality and homogeneity of variance as well 

 as additional assumptions concerning the linearity of dependent and 

 concomitant variables. The very important assumption of homoscedasticity 

 (homogeneity of variance) was not met by the data set even after the 

 application of various transformations. Differences in variances were 

 severe between years and less severe but usually significant between 

 stations in the same year. There is a strong possibility that oysters 

 each year were not sampled from the same population, i.e., the stocks 

 from which the experimental oysters were acquired each year had suffi- 

 ciently different genomes that the environmental conditions to which 

 they were exposed resulted in different phenotypic expressions of the 

 parameters measured. 



The covariance analysis adjusts differences among treatments 

 for the effects of the covariate. In this analysis differences in final 

 length due to differences in initial length were removed so that only 

 variation due to station, year and "error" were assessed. The Scheffe 

 method is generally used to test contrasts among means in a covariance 

 analysis. This is a very conservative technique when used in pairwise 

 comparisons and when data are heterogeneous the procedure maintains an 

 "average" alpha level over the series of all possible linear compari- 

 sons. The Scheffe method for pairwise comparisons was employed in this 

 analysis. A completely unbiased parametric test for this hypothesis is 

 not currently available. 



Hypothesis (3) , length changes at the two experimental stations 

 occur in the same time periods for a given year, was tested using a Geisser- 

 Greenhouse conservative F-test (Kirk, 1968) for fixed-effects in a mixed- 

 model, randomized-block design with repeated measures ANOVA. Such a design 

 requires the sringent assumption of symmetry of the variance-covariance 

 matrix; however, with the conservative procedure used there is no need to 

 test for or meet this requirement if treatment effects are significant. 



