Table 4— Results of classifying six habitat types by four soil characteristics (Size2, Size3, %Cobble2, %Cobble3) using discrim- 

 inant analysis. Probability of guessing correct classification group is 16.7 percent 



Habitat 



type Sample 

 -Phase size 



ABGR/CLUN 

 -CLUN 



Predicted group membership 



ABGR/CLUN ABGR/ASCA THPL/CLUN THPL/ASCA TSHE/CLUN TSHE/ASCA 

 -CLUN -ASCA -CLUN -ASCA -CLUN -ASCA 



57.1 



• Percent — 







28.6 



14.6 



ABGR/ASCA 

 -ASCA 



12 



16.7 



66.7 



8.3 



8.3 



THPL/CLUN 

 -CLUN 



16.7 



16.7 



33.3 



33.3 



THPL/ASCA 

 -ASCA 



33.3 



3.3 



16.7 



16.7 



TSHE/CLUN 

 -CLUN 



TSHE/ASCA 



15 



11.1 



11.1 



13.3 



77.8 





 86.7 



Table 5 — Results of classifying three overstory series by four soil 

 characteristics using discriminant analysis 



Predicted group membership 



Sample Group 



series size ABGR THPL TSHE 



Percent 



ABGR 



19 



63.2 



5.3 



31.5 



THPL 



12 



33.3 



I6.7 



50.0 



TSHE 



24 



33.3 







66.7 



Table 6 — Results of classifying two understory unions by four soil 

 characteristics using discriminant analysis 



Predicted group membership 



Sample Group 



union size CLUN ASCA 



Percent 



CLUN 22 77.3 22.7 



ASCA 33 18.2 81.8 



classification may be that a different set of variables is re- 

 quired as discriminators for this climax tree species, or 

 there simply is too much noise (for example, small data 

 set) in this midground portion of what appears to be a 

 relatively narrow environmental continuum. This prob- 

 lem also occurred in the stepwise discriminant analysis 

 (table 3), wherein no significant variables could be found 



for habitat type groupings of T. plicata by itself or when 

 combined with samples from the A grandis series. 



A much greater accuracy of classification is achieved by 

 stratifying the data based on two understory unions of 

 Clintonia uniflora (Schult.) Kunth. and Asarum caudatum 

 Lindl. Table 6 presents the results of this discriminant 

 classification showing approximately 77 percent and 82 

 percent proper classification, respectively. Stratification of 

 the data into subsets of a single overstory species and two 

 different understory unions should further increase clas- 

 sification accuracy. 



The analysis conducted with only 55 samples may have 

 produced results that reflect a simple random structure in 

 the data set. If so, statisticians refer to this model as "over- 

 fitting the data" and not a true response to the system 

 being modeled. Therefore, stratification of these data be- 

 yond the present level precludes further meaningful analy- 

 sis. 



Tables 7, 8, and 9 present the discriminant score formu- 

 las produced for classification of unknown samples into one 

 of six habitat types, one of three overstory climax series, or 

 one of two understory unions. Appendix B defines values 

 for field quantification of structural ped size and percent- 

 age of cobbles. 



Using four soil characteristics, the formulas calculate a 

 discriminant score for each vegetation unit within a strati- 

 fication group. The formula that produces the highest 

 discriminating score (DS) has the highest probability of 

 being classified correctly. As an example, one of the origi- 

 nal sample plots, assigned by vegetation analysis to the 

 ABGR/CLUN-CLUN habitat type, has the following values 

 for the four discriminating soil characteristics: 



Size2 = 4 %Cobble2 = 10 



Size3 = 4 %Cobble3 = 20 



9 



