Three additional diameter classes were also created by determining the 1-inch diameter 

 class in which the mean stand diameter exists and then adding to it the surrounding one, two, 

 or three 1-inch diameter classes. The number of trees and the basal area in each of these 

 three "mean stand diameter classes" were then determined. 



Because of insufficient data, the time since the last good seedling year was not incor- 

 porated into this study. 



Other independent variables thought to be important for predicting ingrowth were site 

 index, mean stand diameter, and quadratic mean stand diameter. Moser's (1972) work suggested 

 the latter two independent variables. The definitions of all variables used to predict total 

 ingrowth are found in table 33. 



In the first phase of the regression analysis, a set of screening runs was made on the 

 virgin, uneven-aged data and on the managed, uneven-aged data using program REX. The variables 

 were divided into sets and groups as defined in table 34, and the combinations selected were 

 made by picking, at most, one set from each group. This resulted in 6,334 regressions being 

 examined for each data set. 



From these screening runs, seven of the most promising models were selected, their 

 regression coefficients were determined, and then examined for reasonableness of behavior 

 between data sets. A preliminary analysis was also made concerning normality of residuals. 

 The skeuTiess and kurtosis statistics indicated that the residuals were not badly "nonnormal," 

 and therefore "t" tests could be used to roughly check the significance of the regression 

 coefficients . 



Based on the reasonableness and significance criterion, the six original basal area 

 classes (BCj, BC2, BC3, BCi^, BC5, BC5) were collapsed, first to the three classes (BCi ; BC2 

 + BC3; BCi, + BC5 + BCg) and ultimately to two classes (BCj; BC2 + BC3 + BC^ + BC5 + BC5) . 

 The latter two classes were redefined as BACL;^ and BACL2 . 



The next step was to combine the virgin and managed data sets. The two best models from 

 the previous set of runs were chosen, and two new screening runs were made. These runs "forced" 

 the basic models upon the combined data sets and then allowed adjustment for time-since- last- 

 cutting by screening all combinations of the products of the independent variables with the 

 three time-since-last-cutting variables (A^ , A2, and A3). (The definitions of groups, sets, 

 and variables are found in table 35.) 



An examination of the runs disclosed that the signs of the coefficients on both in (site 

 index) and In(4-inch diameter class basal area growth) were negative. This was not considered 

 reasonable. The desire to have site index as a component of the total ingrowth model dictated 

 that a reasonable function of site index be "forced" upon the model by fitting in (Total 

 Ingrowth/Site Index) as the dependent variable. This approach was the same as used earlier in 

 developing the basal area growth equations. Another screening run was performed using the 

 same strategy as for model 1 of table 35; except the in (site index) independent variable was 

 eliminated. From this run, I concluded that the basal area classes could be further collapsed 

 to the final two classes BACLi ^""^ BACL2 . ^ final screening run was then made to provide the 

 following log model: 



79 



