But the simpler eastside equations correctly classified 

 a higher percentage of the timber sales. 



It is reassuring to know that as we add more specific 

 information to the process, moving from gate 1 to gate 4, 

 the classification results improve. 



Also, having nonsignificant results with respect to the 

 statistical procedure chosen is reassuring. 



Classification Results and Cutoff 

 Points 



The classification results presented were based on a 

 decision rule that assigns timber sales to the unsold cate- 

 gory if the probability is less than 0.50, and to the sold 

 category if greater than or equal to 0.50. As discussed 

 earlier, one would expect classification results to change 

 as the cutoff points (decision rules) are altered. Figure 1 

 presents the classification results based on the seven 

 cutoff points defined earlier. When examining the 

 eastside discriminant function (fig. 1), the most striking 

 feature of gate 1 is the large fluctuation in the percentage 

 correctly classified as the probability is varied. If the 

 decision rule establishes that a predicted probability of 

 0.20 or greater defines a sold sale, the percentage of the 

 sold sales that are correctly classified by the eastside gate 

 1 discriminant function is 100 percent. The tradeoff, 

 however, is that under that rule, percent of the unsold 

 sales were correctly classified. At the other end of the 

 scale, where a probability of 0.80 or greater defines a sold 

 sale, the percentage of the sold sales correctly classified is 

 7.7; and the percentage of the unsold sales that are cor- 

 rectly classified is 100. The eastside gate 4 discriminant 

 function at a predicted probability of 0.20 correctly classi- 

 fies 64.3 percent of the unsold sales and 96.2 percent of 

 the sold sales. At a probability of 0.80 sold, the classifica- 

 tion results are 100 percent for unsold sales and 65.4 

 percent for sold sales. Figures corresponding to the other 

 equations and gates are presented in appendix A, figures 

 2-4. 



Figure 1 also illustrates how the cutoff points could be 

 varied from gate to gate. As the timber sale nears the 

 auction date, the land manager can become more precise 

 regarding the cutoff point without drastically affecting the 

 percentage correctly classified. At gate 1, which is 5 to 10 

 years from the auction date, the land manager may want 

 to maintain the standard cutoff, 0.50. 



The gate 1 equations produce the largest variation in 

 percentage correctly classified at the various probability 

 levels. The explanation is that gate 1 equations are the 

 least accurate in percentage correctly classified. The gate 

 1 predicted probabilities are clustered around 0.50, mean- 

 ing the equations have difficulty making correct predic- 

 tions. As more and better information is generated (mov- 

 ing from gate 1 to gate 4), the equations are better able 

 to classify sales, and the predicted values do not cluster 

 around 0.50. For the eastside gate 1 discriminant func- 

 tion using a probability of 0.20, the percentage correct 

 varies from to 100; at a probability of 0.80 the percent- 

 age correct ranges from 100 to 7.7 for unsold and sold 

 sales, respectively. The eastside gate 4 discriminant 

 function ranges from 64.3 percent to 96.2 percent for 



GATE 1 



I- 

 U 



LU 



ec 

 c 

 O 

 o 



UJ 



u 



100 r- 

 80 - 

 60 - 

 40 

 20 

 



UNSOLD 

 SALES 



ALL 



SALES 



I 1 / 



I 



I 



I 



SOLD 

 J SALES 



.20 .26 .33 .50 .67 .75 .80 



GATES 2-3 



U 



UJ 



ec 



K 



O 

 u 



Z 



UJ 



u 



K 



UJ 



a 



100 I- 



80 



60 - 



40 



20 - 



UNSOLD 

 ✓ SALES 



_L 



J L 



J 



.20 .25 .33 .50 .67 .75 .80 



GATE 4 



U 



UJ 



ec 

 ec 

 o 

 o 



o 

 ec 



Ui 



a. 



100 t- 



80 - 



60 - 



40 - 



20 - 



_L 



_L 



_L 



J 



.20 .25 .33 .50 .67 .75 .80 

 PROBABILITY 



Figure 1 — Cutoff point analyses for eastside 

 Discriminant model. 



a probability of 0.20, and 100 percent to 65.4 percent for 

 a probability of 0.80. The maximum range for gate 4 is 

 34.6 percent, in comparison to 100 percent for gate 1. 



MANAGEMENT IMPLICATIONS 



The Gates timber sale planning system plays an impor- 

 tant role in timber management. The Gates process can 

 be viewed as adding structure to the designing of timber 

 sales, where a sale must meet certain requirements before 

 moving to the next gate. Additional information can be 

 generated through the Gates process to help timber sale 

 planners to make sound economic decisions about the 



9 



