Seven cutoff points (probability levels) were chosen 

 to investigate the changes in-elassificati on results: 0.20, 

 0.25, 0.33, 0.50, 0.67, 0.75, and 0.80. The posterior 

 probabilities were used as the discriminant model's pre- 

 dicted probabilities of a sale selling. The posterior proba- 

 bilities provide a continuous prediction of salability and 

 allow the cutoff points (decision rule) to vary at the levels 

 defined previously. 



As stated previously, the logistic regression predicts a 

 probability, and therefore adapts easily to this analysis. 



CATEGORICAL ANALYSIS OF VARIANCE 



Categorical analysis of variance (Bishop and others 

 1975) was used to analyze the dependent variable, per- 

 centage correct (%c), based on independent variables, 

 classification method (cm), gates (g), and geographical 

 zones (z). The results of this analysis statistically quanti- 

 fied the benefits derived by having employed different 

 classification methods, gates, and geographical zones. 



The mathematical equation is: 



%c... = M + cm. + e. + z. + I. . + e... 



ijk t °j k ijk ijk 



where Lj^^, e.j^, and M are the interaction term, error term, 

 and the overall mean, respectively. 



RESULTS AND DISCUSSION 



In total, 12 classification equations were developed to 

 predict salability — two statistical classification methods 

 for each of three gates in the sale planning process on 

 each of two geographical zones within the Northern 

 Region. Table 1 provides an overall summary of the re- 

 sults, showing that the percent correctly classified ranged 

 fi-om 59 to 90. 



Table 1 suggests that classification success steadily 

 improved with progression from gate 1 to gate 4, that 

 eastside sales are more successfully classified, and that 

 the two analytical procedures produce similar results. 

 In fact, these impressions are correct, as will be shown 

 in later statistical analysis. 



The 22 significant variables used in the equations are 

 defined in table 2. As indicated earlier, measurements 

 on sale variables were made from timber sale records and 

 economic variables from government publications. 



Table 1 — Overall summary of classification success 







Logistic 



Discriminant 



Gate 



Subregion 



regression 



analysis 







Percent correctly classified 



1 



Eastside 



65.0 



72.5 





Westside 



63.3 



59.0 



2-3 



Eastside 



80.0 



77.2 





Westside 



70.2 



68.8 



4 



Eastside 



85.0 



90.0 





Westside 



77.4 



77.9 



Table 2 — Independent variables used in study 



Variable 



Description 



Units 



TOTVOL 



Total sale volume harvested 



M bd. ft. 







(Scribner) 



TOTSALE 



Total sale area 



Acres 



AVGSLOPE 



Average slope 



Percent 



ALPM 



Average logs per thousand 



Number 



TOTROAD 



Total road construction 



Miles 



NEW 



New road construction 



Miles 



RECON 



Old road reconstruction 



Miles 



ACRES 



Acres harvested in sale 



Acres 



DENSE 



Acres harvested divided by 



Number 





total sale area 





VPA 



Volume per acre harvested 



M bd. ft. 







(Scribner) 



DEAD 



Percent volume dead white 



Percent 





pine or lodgepole pine 





%TRAC 



Percent volume tractor yarded 



Percent 



%CABLE 



Percent volume cable yarded 



Percent 



TRACDIST 



Average maximum tractor 



Feet 





yarding distance 





STUMP MILL 



Felling and bucking + skidding 



$/M bd. ft. 





and loading + haul + slash + 







road + advertised rate 





ADVRATE 



Minimum bid price 



$/M bd. ft. 



SPLT 



Selling price, lumber tally 



$/M bd. ft. 



PMETH 



Contract price escalation 



1 = Yes 





clause 



= No 



HAULRAT 



Haul distance to primary 



Number 





appraisal point divided by 







haul distance to secondary 







appraisal point 





UNCUT,_3 



Uncut volume under contract 



Number 





lagged 3 months 





EXCH,_3 



U.S./Canadian exchange 



Number 





rate lagged 3 months 





COMPMILL 



Competing mills at appraisal 



1 = Yes 





point 



= No 



LMBRPROD 



1 2-month percentage change 



Number 





in Inland region lumber 







production 





4 



