Table 3 — Variables selected, significant F value, and degrees of freedom (numerator and 

 denominator) produced by stepwise discriminant analysis on coarse-textured 

 parent material data 



Degrees of freedom 



Stratification F Value 



of data Variable Sig. >0.90 Numerator Denominator 



Six habitat types 



Size3 



6.522 



5 



49 





%Gravel3 



3.277 



5 



48 





pH3 



2.902 



5 



47 





Size2 



3.157 



5 



46 





%Cobble2 



2.575 



5 



45 



Overstory series 



pH2 



5.421 



2 



52 



ABGR-THPL-TSHE 











Two overstory series 



pH2 



9.008 



1 



41 



ABGR - TSHE 



Value3 



4.447 



1 



40 





CiltO 



ollto 



A Ajn 



b.£ i o 



1 



on 



oa 



Understory unions 



Size3 



16.187 



1 



53 



CLUN - ASCA 



Chroma2 



7.083 





52 





%Cobble3 



4.589 



j 



51 



ABGR/CLUN - 



%Gravel3 



5.108 



1 



16 



ABGR/ASCA 



%Cobble3 



6.765 



1 



15 





Chroma2 



7.411 



1 



14 





Shape3 



4.973 



1 



13 



TSHE/CLUN - 



Size3 



27.547 



1 



22 



TSHE/ASCA 



%Cobble3 



8.557 



1 



21 



ADO D/r 1 1 MM 



oizeo 



lo.ol O 



Q 



o 



on 



oy 



ABGR/ASCA - 



%Gravel3 



5.534 



3 



38 



TSHE/CLUN - 



Value3 



4.341 



3 



3 



TSHE/ASCA 



pH3 



3.401 



3 



36 





uepirio 



Q Q A 1 



o 



oo 



ABGR/CLUN - 



pH2 



6.429 



1 



12 



TSHE/CLUN 



Clay2 



10.740 





13 





Size3 



15.882 





12 





Shape3 



11.155 





11 



ABGR/ASCA - 



Size2 



15.228 





25 



TSHE/ASCA 



%Gravel3 



6.665 





24 





Shape2 



6.951 





23 



THPL/CLUN - 



Size3 



5.787 



3 



32 



THPL/ASCA - 



%Cobble2 



5.693 



3 



31 



TSHE/CLUN - 

 TSHE/ASCA 



1983). All four of these assumptions were violated to 

 some extent in these analyses, leaving exploratory gener- 

 alizations about both the data structure and discriminant 

 functions as the result, rather than statistically signifi- 

 cant conclusions. 



Using four soil characteristics as variables, the proba- 

 bility of correct classification is equal to or greater than 

 57 percent for the Abies grandis and Tsuga heterophylla 

 series habitat types, with 33 percent or less accuracy for 

 Thuja plicata habitat types (table 4). The probability of 

 simply guessing the correct habitat type is 16.7 percent. 

 Considering the small sample size and the large amount 

 of unexplained variation indicated by principal component 

 analysis, this degree of classification accuracy is quite 



good. Although it is somewhat circular to test results with 

 data used to develop the classification scheme, it does act 

 as an acceptable initial test of classification accuracy. 



In an attempt to increase the sample size per group and 

 reduce apparent variation, the data set was stratified by 

 overstory climax species (that is, Abies grandis, Thuja 

 plicata, Tsuga heterophylla). Table 5 presents the classifi- 

 cation results of discriminant analysis for the three series 

 groups using the same four variables as above. The 

 probability of properly assigning a sample to the A. 

 grandis orT. heterophylla series using the discriminant 

 functions developed is roughly twice the probability of 

 guessing (33.3 percent), whereas for T. plicata it is one- 

 half. Possible reasons for the poor accuracy in T. plicata 



8 



