4.4 EVALUATING THE GOODNESS OF FIT WITH THE ROC METHOD
The ROC characteristic is a measure for the goodness of fit of a logistic regression
model similar to the R2
statistic in Ordinary Least Square regression (Pontius and
Schneider, 2001). A completely random model gives a ROC value of 0.5; a perfect fit
results in a ROC value of 1.0.
Calculate the ROC value for your regression model.
Click: Graphs | ROC Curve. Select the observed land use as ‘State variable’ with value 1
for occurrence of this land use type and the ‘Predicted probability’ as ‘Test variable’.
Click ‘OK’.
The results will show a ROC curve and the ‘area under the curve’ that is the test result
(Figure 20).
46
Figure 20. ROC curve of a logistic regression with SPSS
USING THE REGRESSION RESULTS AS INPUT TO THE CLUE-S MODEL
The regression results are used by the CLUE-S model to determine the suitability of the
locations for the different land use types. For each land use type considered a separate
regression model is used. The result of the analysis in SPSS should therefore be
translated to input for the CLUE-S model. The input file for the regression equations is
called ‘alloc1.reg’ and is located in the installation directory and can be edited through
the user interface of CLUE-S (click ‘Edit Input’).
This file is structures as follows (Figure 21):
Line 1: Number code for land use type (e.g. forest).
Line 2: Constant of regression equation for land use type (ß0).
Line 3: Number of explanatory factors (sc1gr#.fil files) in the regression
equation for that land use type.
Line 4 and
further:
On each line the beta coefficients (ß1, ß2, etc.) for the explanatory factor
and the number code of the explanatory factor.