effect on a timber sale selling. Of the variables in the 

 discriminant function, %TRAC is the most important 

 discriminator (standardized coeflBcient = 0.778). 



Comparing westside and eastside results, one can con- 

 clude that the westside timber offerings are more difficult 

 to explain. Given the complexity, a more complicated 

 equation is needed to produce the observed classification 

 results. Therefore, one should not expect an identical 

 equation structure between geographical zones. 



The overall classification results for the two approaches 

 are also quite similar (see table 6b). The logistic regres- 

 sion correctly classifies 70.2 percent of the sales, while the 

 discriminant function correctly classifies 68.8 percent. 

 The differences exist in the equation's ability to classify 

 the individual categories. The discriminant function cor- 

 rectly classifies a higher percentage of the unsold sales 

 (69.7 vs. 57.9 percent), the logistic regression correctly 

 classifies more of the sold sales (78.9 vs. 68.1 percent). 

 The holdout method indicates that prediction accuracy for 

 the discriminant model is 66.5 percent, a slight decrease. 

 Also, there is a slight decrease in the individual group 

 classification results. 



The gate 2-3 equations have changed in comparison to 

 gate 1 to reflect the better information available. Afi^r 

 adding the information from gates 2 and 3, the overall 

 classification results increased fi-om 221 to 245 (10.9 per- 

 cent) for the logistic regression and from 206 to 240 (16.5 

 percent) for the discriminant equation. 



Gate 4 



At the final gate before the offering, appraisal and eco- 

 nomic information is added to site and sale characteris- 

 tics. This point in the gates process can be viewed as one 

 that converts sale characteristics information into dollars 

 per thousand board feet. This gate is approximately 2 to 

 3 months before the initial offering. 



EASTSroE EQUATIONS 



In general, the gate 4 eastside equations indicate 

 that the stump-to-mill cost (STUMPMILL), the size 

 (TOTVOL), and the price of final product derived from 

 the logs (SPLT) are the important factors affecting sala- 

 bility (table 7). The discriminant function contains three 

 additional variables, the natural logarithm of average logs 

 per thousand, In(ALPM), the natural logarithm of percent 

 volume dead white pine or dead lodgepole, In(DEAD), and 

 the price escalation clause used in the timber sale con- 

 tract (PMETH). These characteristics indicate the num- 

 ber and quality of the logs, and the contractual agreement 

 of the sale. The standardized discriminant coefficients 

 conclude that the STUMPMILL variable is the most im- 

 portant factor in determining sold and unsold sales, with 

 SPLT second. 



Table 7b presents the classification results for the 

 eastside equations. The discriminant function correctly 

 classified 90.0 percent of the timber sales, 92.9 percent 

 of the unsold sales, and 88.5 percent of the sold sales. 

 The holdout method indicates that the above classification 

 results are unstable. The overall correct classification 



Table 7 — Gate 4 eastside equations and classification results 



A. Equations 



Logistic regression 



Discriminant analysis 



Variabie Coefficient (Std Err) 



Coefficient 



(Std Coeff) 



STUMPMILL -0.078 (0.030) 







Ln(STUMPMILL) 





-6.102 



(-1.132) 



TOTVOL 



.0012 (.0004) 







(TOTVOL)"^ 





.037 



(.783) 



SPLT 



.088 (.038) 







Ln(SPLT) 





6.656 



(.832) 



Ln(ALPM) 





-.615 



(-.169) 



Ln(DEAD) 





.287 



(.351) 



PMETH 





-.667 



(-.274) 



Constant 



2.620 (5.488) 



-2.622 





B. Classification results 





Logistic regression 



Discriminant analysis 



Actual Total 



Correct Percent 



Correct 



Percent 



group sales 



predict correct 



predict 



correct 



Unsold 14 



10 71.4 



13 



92.9 (71.4)' 



Sold 26 



24 92.3 



23 



88.5 (73.1)' 



All sales 40 



34 85.0 



36 



90.0 (72.5)' 



'Indicates percent correctly classified using the holdout method. 



decreases to 72.5 percent, with 71.4 percent for the unsold 

 sales and 73.1 percent for the sold sales. The logistic 

 regression correctly classified 85.0 percent of the timber 

 sales, with 71.4 percent correct classification for the un- 

 sold sales and 92.3 percent for the sold sales. 



With the introduction of the gate 4 information, the 

 equations have improved the overall classification results 

 by 6.3 percent for the logistic regression and 16.1 percent 

 for the discriminant equation. 



WESTSroE EQUATIONS 



At gate 4, all information fi-om the Gates timber sale 

 planning process plus market information found in other 

 sources is used to develop the equations. Given this fact, 

 these equations are the most complicated. This has led to 

 an equation that produces the most accurate prediction of 

 salability. 



The gate 4 equations are displayed in table 8. One 

 should examine these equations fi-om the standpoint of 

 which variables have a positive effect and which have a 

 negative effect on salability. The following variables have 

 a positive effect on salability: the selling price (SPLT), the 

 ratio of haul distances to primary and secondary ap- 

 praisal points (HAULRAT), whether the contract follows 

 the WWPA price index or remains fixed (PMETH), 

 whether the mill site is competitive (COMPMILL), the 

 12-month percentage change in lumber production in the 

 Intermountain zone (LMBRPROD), the volume-per-acre 

 harvested (VPA), and the ratio of harvested acres to sale 

 acres (DENSE). In general, the above characteristics 

 indicate that a sale is likely to be sold: when composed 



7 



