direction and extent of development. This is especially 

 important today when government spending is a major 

 issue. Using classification procedures, like those devel- 

 oped in this pai>er, coupled with the Gates process, the 

 timber sale planner can generate a critical piece of 

 information — ^likely salability. The planner must be 

 aware of how his planning decisions affect salability. 



The sale planning process starts with a timber manage- 

 ment specialist inspecting a proposed sale area. Based on 

 initial site and volume estimates, the forester determines 

 if the area can support a timber sale. From a modeling 

 standpoint, the volume estimates and site information can 

 be used to assess salability. 



At gates 2 £ind 3, the sale takes on more definite form, 

 and more specific information is known about the prospec- 

 tive sale. It is now possible to more reliably evaluate 

 salability. Using the more specific information, prediction 

 capability of classification procedures has been improved. 

 The ability to quantify sale design decisions, such as add- 

 ing another mile of road or decreasing the volume per acre 

 harvested, is essential to predicting salability. The pre- 

 diction of salability is most important at this point. If 

 salability can be accurately predicted at this point, the 

 timber sale planner can implement the necessary changes 

 to produce a salable offering. 



Gate 4 offers little time to make necessary changes 

 to produce a viable timber sale, with the auction only 

 3 months away. If selling the offering is questionable, 

 the sale could be delayed and the necessary changes im- 

 plemented to produce a salable timber offering. 



The logistic regression and discriminant function were 

 evaluated as salability classification tools. The categori- 

 cal analysis of variance results verified that there was no 

 statistically significant difference in prediction results 

 with respect to the classification procedure. 



But the logistic regression approach may have an 

 added advantage over discriminant analysis. The logistic 

 regression produces an estimated probability of a sale 

 selling. This prediction supplies more information to the 

 user about the degree of salability. For example, if the 

 logistic model predicts a 10 percent chance of selling, this 

 should indicate to the planner that major renovations are 

 needed. The above sale is accurately classified as an 

 unsold sale, but it is very different from a sale that might 

 have an estimated probability of 0.45 (45 percent chance 

 of selling). Given our 50-50 decision rule, they would both 

 be classified as unsalable sales, even though the sale with 

 an estimated probability of 0.10 should probably not be 

 offered, while the other could be offered under favorable 

 market conditions. The predicted probability provides the 

 planner with a flexible decision rule. This flexibility could 

 lead to a rule that allows for a zone of indecision. The 

 zone could be defined as any sale having a probability less 

 than 0.30 will be deferred, greater than 0.30 will be 



revised, and any sale having a probability greater than 

 0.70 will be advertised for sale. Any sale having an esti- 

 mated probability falling in the 0.30 to 0.70 range of inde- 

 cision will be withheld, revised, or advertised based on the 

 professional judgment of the planning staff. 



The equations described here were based on a sample of 

 sold and unsold timber sales in the Northern Region of 

 the Forest Service, during 1980 to 1985. The results 

 should not be used to predict sold and unsold sales in any 

 other region of the Forest Service nor for any other seller 

 of stumpage. 



REFERENCES 



Artley, D. 1986. DLOGPRICE economic model. Unpub- 

 lished drafl supplied to author by D. Artley. 



Bishop, Y. M. M.; Fienberg, S. E.; Holland, P. W. 1975. 

 Discrete multivariate analysis: theory and practice. 

 Cambridge, MA: MIT Press. 348 p. 



Dixon, W. J., ed. 1981. BMDP statistical sofi;ware. 

 Berkeley, CA: University of California Press. 725 p. 



Johnson, R. A.; Wichern, D. W. 1982. Applied multivari- 

 ate statistical analysis. Englewood Cliffs, NJ: Prentice- 

 Hall. 594 p. 



Johnson, R. N. 1979. Oral auction versus sealed bids: an 

 empirical investigation. Natural Research Journal. 

 19(1): 315-335. 



Lachenbruch, P. A. 1975. Discriminant analysis. New 

 York: Hafner. 128 p. 



Maddala, G. S. 1983. Limited-dependent and qualitative 

 variables in econometrics. Cambridge, MA: Cambridge 

 University Press. 401 p. 



Merzenich, J. P. 1985. Transaction evidence timber ap- 

 praisals in the Northern Rockies. Missoula, MT: U.S. 

 Department of Agriculture, Forest Service, Northern 

 Region. 19 p. [Mimeo]. 



Morrison, D. F. 1976. Multivariate statistical methods. 

 New York: McGraw-Hill. 415 p. 



Nie, N. H. 1983. SPSSX statistical software. New York: 

 McGraw-Hill. 806 p. 



Peterson, T. 1980. Timber sale feasibility analysis. 

 Unpublished paper on file at: U.S. Department of 

 Agriculture, Forest Service, Bitterroot National Forest, 

 Hamilton, MT. 15 p. 



Pindyck, R. S.; Rubinfeld, D. L. 1981. Econometric models 

 and economic forecasts. New York: McGraw-Hill. 630 p. 



Press, S. J.; Wilson, S. 1978. Choosing between logistic 

 regression and discriminant analysis. Journal of Ameri- 

 can Statistical Association. 73(364): 669-704. 



U.S. Department of Agriculture, Forest Service. 1985. 

 Forest Service Manual 2400 Timber Management. 

 Washington, DC: U.S. Government Printing Office: 

 2414.27, 2431.2. 



10 



