Data were divided into two parts: data screening and parameter estimation were 

 done with data collected in the first measurement period; models were verified by means 

 of data collected in the second measurement period. 



Age, diameter breast high, height, crown class, basal area per acre of the stand, 

 percent defect, and diameter growth rate prior to mortality were the variables in the 

 data set which were considered potential predictors of mortality. Instructions for 

 estimating percent defect as specified in "Instructions for Forest Inventory of 

 Potlatch Forests, Inc. Logging Units" as revised July 8, 1959, are reproduced in 

 the appendix. SCREEN was run independently for each of the 10 north Idaho species 

 included in this study. 



Output from SCREEN consists of a tree diagram showing characteristics that best 

 predict mortality. SCREEN also reports how mortality rate changes over the range of 

 these predictors. 



We have considered three levels of significance: a high level (95 percent), a 

 moderate level (75 percent), and a low level (50 percent). The objective of data 

 screening is to identify combinations of independent variables that are related to 

 the probability of mortality. By including the 75 and 50 percent significance 

 levels, we were able to gain a better understanding of the relationships that exist 

 between the probability of mortality and the set of independent variables. However, 

 only independent variables significant at the 95 percent level were included in the 

 models developed in this study. 



An example of the use of SCREEN to develop tree mortality predictors for grand 

 fir [Abies gvandis) follows. The SCREEN output for this example is displayed in 

 figure 1. At the 95 percent significance level two tree characteristics, percent 

 defect and crown class, are significant predictors of tree mortality. The output 

 indicates that grand fir trees with defect percent from to 90 have a 5-year mortality 

 rate of 81/1523 = 0.053, while trees with 91 to 100 percent defect have a mortality 

 rate of 36/259 = 0.139. Additionally, among trees with percent defect of to 90 

 percent, those with crown class of 1, 2, or 3 (dominant, codominant, intermediate) 

 have a mortality rate of 58/1310 - 0.044, while trees with crown class 4 (suppressed) 

 have a mortality rate of 25/213 = 0.108. 



At the 75 percent significance level, the significant predictors are percent 

 defect, crown class, diameter, and age. At the 50 percent significance level, the 

 significant predictors are the same as the predictors at the 75 percent level of 

 significance. This implies that only a negligible amount of the unexplained variation 

 can be explained by any of the unused independent variables. 



Once the "optimal" set of independent variables has been selected, it is necessary 

 to estimate the parameters of the mortality model. The general form of the logistic 

 mortality model is 



Parameter Estimation 



P = {1 + exp [-(Bo + Bixi + ^2^2 + •••)]} 



-1 



where 



P 



probability of mortality in a specific time interval 



tth nonlinear regression coefficient (model parameters), i = 0,1,... 



X . 



3 



j'th independent variable, j = 1,2,... 



5 



