understanding of the methodology suggests that relatively large amounts of 

 data gathered at frequent and evenly spread sampling intervals are highly 

 desirable for this methodology to be effective in most instances. 



The general linear model analysis is described at various levels of detail in 

 several statistics books, such as Cochran and Cox (3), Davies (4), Federer (6), 

 Kempthorne (7), Sheffe (10), and Snedecor and Cochran (11). This linear 

 model approach includes analysis of variance and regression analysis. In this 

 approach variations in a response variable measured over time and space are 

 decomposed into assignable sources of variations and these variations are 

 assumed to be additive. Tests of significance, such as F-tests or variance ratio 

 tests to determine the change in the mean value of some variable from several 

 sample events, are based on certain assumptions such as a normal probability 

 density and independent and homogeneous variance (10). Data from samples 

 taken over time frequently do not conform to these assumptions. 

 Nonstationary elements, such as seasonal or diurnal, and tidal components are 

 often present, and the data may be highly correlated in time. 



In the linear model approach time, space and sampling locations, along with 

 replications, become a part of a planned experimental design. As a means for 

 considering spatial and temporal variability in the linear model, the spatial and 

 temporal distributions of biota (i.e., ichthyoplankton) are treated as a sum of 

 responses due to assignable sources of factor levels. In addition, 

 transformations of the response variable are sometimes used to achieve 

 homogeneity of variances. Finally, because one can expect certain physical and 

 biological data to be correlated, these relationships can be effectively utilized 

 by carrying out multivariate analyses of variance and covariance analyses. 



In carrying out the linear model approach to monitoring and impact 

 assessment, the method involves formulating hypotheses or linear contrasts for 

 carrying out the statistical tests. Among these contrasts one tests for main 

 effects due to a defined factor and the interaction of factors of interest. 



SCHEME FOR ESTABLISHING SAMPLING STATIONS 



We will illustrate the use of a simple linear model by an example which 

 relates to deciding where to establish monitoring stations along a 

 cross-sectional area of an estuary which is of interest. For simplicity we will 

 assume that cost constraints limit the number of samples to about 50. Our 

 prior knowledge of the problem suggests that we should be concerned about 

 the depth distribution of the given organism (say winter flounder larvae). 



339 



