41 



Partial correlations (Ott 1977) were used to determine if 

 predictive models were statistically free from confounding effects of 

 environmental variables correlated with their dependent variables. 

 I accomplished this by calculating multiple correlation coefficients as 

 the square root of the coefficients of determination for the 

 multivariate models. The multivariate models were then regressed 

 with the covariant dependent variables using the SAS GLM 

 procedure (SAS Inst., Inc. 1985), and I calculated multiple correlation 

 coefficients for these confounded forms of the model. Partial 

 correlation coefficients were calculated using the multiple correlation 

 coefficients to assess whether models with correct dependent 

 variables were significant when the effects of the covariables were 

 held constant. 



Diatom-index values (TROPH 1) (Whitmore 1989) were 

 calculated for the subfossil diatom assemblages in the survey lakes. 

 A predictive model was developed to yield water-column total P 

 estimates from subfossil diatom assemblages for historic WCP 

 (Canfield et al. 1983a) inferences. The 51 lakes used to construct this 

 model were the lakes included in the present study, lakes in 

 Whitmore's (1989) study, and Lake Francis in Highlands County. 

 Water-column total P values for the lakes outside of the present 

 study were median values obtained from the Florida Lakes Data 

 Base. TROPH 1 values for subfossil diatom assemblages were 

 regressed with water-column total P values using the SAS GLM 

 procedure (SAS Inst., Inc. 1985). 



Because of the significant negative correlation between 

 macrophyte presence and water-column nutrient concentrations 



