8. Temperature (TEMP). --The highest daily temperature (°F) taken while the yard- 

 ing crew was working. 



9. Wind velocity (WINVEL) . --The highest daily wind velocity (mi/h) taken while 

 the yarding crew was working. 



Measurements were also taken on precipitation but most of the measurements were 

 zero or trace levels so this provided little useful information. 



The coded independent variables were defined as follows: 



1. Surface type. --Coded from one (little surface obstruction) to three (severe 

 surface obstruction) . 



2. Surface condition . --Coded from one (dry, firm soil) to three (wet, muddy 

 soil) . 



3. Operator . --Coded from one (below average) to three (above average). 



4. Landing . --Coded from one (spacious landing) to three (limited landing). 



5. Deck. --Coded from one (easy to land logs and unhook chokers) to three (diffi- 

 cult to land logs and unhook chokers) . A four indicates this variable was not applicable 



A more complete description of the criteria for the classification system is 

 contained in Gibson and Rodenberg (1975) . 



ANALYSIS 



Model building was partitioned into four classes so that easy comparisons could 

 be made. The first class was composed of the independent variables, except TEMP and 

 WINVEL. The second class also was composed of the independent variables, except TEMP 

 and WINVEL, but interactions were allowed. The third was composed of all the indepen- 

 dent variables. The fourth class was composed of all the independent variables plus 

 interactions. These four classes were then used to compare the simple models (no 

 interaction terms) and the interaction models since the physical constraints of the 

 system implied that most of the independent variables were linear. The effect of TEMP 

 and WINVEL was analyzed separately since there were a few levels (many repeated values) 

 of these two variables; estimates were based on only a few unique observations. TEMP 

 and WINVEL were included in the final analysis, but their contribution to the regres- 

 sions was weak since there were only a few levels of each variable. 



The coded independent variables were eliminated from the final model building 

 because no significant correlations with the dependent variables were found. Analysis 

 using graphs and dummy variables indicated there was not enough range in the observa- 

 tions at each level to meaningfully measure the association between the coded independ- 

 ent and dependent variables. The analysis and field observations also indicated a 

 need for a more quantitative method of rating the coded variables. The authors defined 

 the coded variables in this report because they could be important on other logging 

 studies if the variable levels were better defined and the range of conditions was 

 sufficiently large. 



The four classes of models were used to form predictive equations for TT and for 

 the components TU, TLO, HC, TL, and UC. A statistical program, BMD02R from the U.C.L.A. 

 Biomedical package, was used to determine the most important variables within each 

 class during model selection. 



6 



