The usage of data from the second classification period on plot 61 does pose an addi- 

 tional problem. During the second classification period, only trees 7.6 inches and larger 

 were measured. All stand density variables, therefore, must be expressed in trees 7.6 inches 

 and larger. In addition, the usage of predicted basal area growth is precluded because of the 

 equation's dependency upon knowing stand basal area of all trees 3.6 inches and larger. 



The following is a summary of the process used to develop the final conversion model. 

 Further details are in appendix F. 



I anticipated the factors influencing the conversion rate would be the same as those that 

 might influence tree vigor, as expressed by growth. Previously it was shown these factors 

 include measures of productivity, diameter class size, and stand density. Keeping in mind 

 both these factors and the restriction that only stand attributes for trees 7.6 inches and 

 larger could be used, a number of variables were defined for use as independent variables. 



As with mortality, I decided to express the rate as a proportion of the trees in a diameter 

 class converting from blackjack to yellow pine and to model this rate using the nonlinear 

 logistic function. Therefore, I assigned the dependent variable a value of one if the black- 

 jack tree converted to yellow pine, and a value of zero if it did not. 



Using these variables, a number of nonlinear regression runs were made, and chi-square 

 "goodness-of -f it" values were examined. Predictions from all models displayed significant 

 differences from the actual values. Examination of the models and data revealed that the 

 greatest misfit was due to data from a single subplot. Therefore, I decided to strengthen 

 the data base by adding to it the data from those subplots originally eliminated because of 

 the paved highway. A second set of nonlinear regression runs was made, but again the resulting 

 models displayed poor fit in the same diameter class. 



Because tree vigor is such a subjective classification, the apparently atypical subplot 

 was eliminated and new equations determined. The resulting model's predictions were not 

 significantly different from the actual values (see table 16) . To test this model further, 

 chi-square values were also computed for the data reserved for validation. The result indicated 

 a significant difference between the model predictions and the actual data, and the cause was 

 traced to another subplot with an unusual conversion rate in the same diameter class. I could 

 not determine why the two subplots exhibited such high conversion in the same diameter class, 

 but the elimination of the second subplot from the data did produce an insignificant chi- 

 square value for the validation data. 



The final conversion model is: 



-X -1 



PrCON = (1.0 + e ) 



where 



2 



X = -3.777269 + 0.5012605D - 6 . 597465E-03D - 0.0982176S 



PrCON = proportion of trees in a diameter class converting from blackjack pine to yellow 

 pine 



D = diameter class size 

 S = Minor's site index 



A plot of this equation is in figure 4. To use this model, each independent variable is 

 assigned the value existing for the diameter class at the start of the growth period prior to 

 reclassification. Therefore, at the start of every fourth growth period, a prediction is made 

 as to how many trees will convert by the start of the fifth growth period. 



21 



