Ecological Archives E090-026-A2

Kim Murray and Mary M. Conner. 2009. Methods to quantify variable importance: implications for the analysis of noisy ecological data. Ecology 90:348–355.

Appendix B (Table B1). Performance of Akaike weights with a reduced sample size of 100 to increase model-selection uncertainty.


Four uncorrelated explanatory variables

 

 

 

 

 

Akaike Weight

True Correlation

Variable

Value

Weight

 

Value

Weight

 


 

x1

0.992

25%

0.070

10%

 

 

x2

1.000

25%

0.140

20%

 

 

x3

1.000

25%

0.245

35%

 

 

x4

1.000

25%

0.245

35%

 

Total

3.992

100%

0.700

100%

 

 

 

 

 

 

 

 

 

Two independent and two correlated explanatory variables (rx2x4= 0.40).

x1

0.962

25%

0.070

12%

x2

0.873

23%

0.088

15%

x3

0.999

26%

0.245

41%

x4

0.998

26%

0.194

32%

Total

3.832

100%

0.597

100%

 

 

 

 

 

 

 

 

Four uncorrelated explanatory variables and a spurious variable.

x1

1.000

20%

0.070

10%

x2

1.000

20%

0.140

20%

x3

1.000

20%

0.245

35%

x4

1.000

20%

0.245

35%

x5

1.000

20%

0.000

0%

Total

5.000

1.000

0.700

100%

 

 

 

 

 

 

 

 

Two independent variables, two correlated variables (rx2x4= 0.40) and a spurious variable.

x1

0.988

20%

0.070

12%

x2

0.995

20%

0.088

15%

x3

1.000

20%

0.245

41%

x4

1.000

20%

0.194

32%

x5

0.999

20%

0.000

0%

Total

4.982

1.000

0.597

100%



[Back to E090-026]