54
4.3.2. Significance Test of Model and Predictors
And if an acceptable model is found, the statistical significance of each of the coefficient is evaluated
.
Table 4.3 . Variables in the Equation of null model
β S.E.
Wald df
Sig. ExpB
Step 0 Constant
-.272 .019
200.330 1
.000 .762
The initial test for the null model which the coefficients for all independent variables are 0. The finding of significance in Table 4.3
indicates this null model should be rejected, because of Sig. = 0 or 0.05 and
β ≠ 0.
Table 4.4.
Omnibus Tests of Model Coefficients
Chi-square df
Sig. Step
94.011 1
.000 Block
94.011 1
.000 Step 1
Model 94.011
1 .000
Step 57.422
1 .000
Block 151.433
2 .000
Step 2 Model
151.433 2
.000 Step
52.756 1
.000 Block
204.189 3
.000 Step 3
Model 204.189
3 .000
Step 18.473
1 .000
Block 222.662
4 .000
Step 4 Model
222.662 4
.000 Step
5.379 1
.020 Block
228.041 5
.000 Step 5
Model 228.041
5 .000
Step 3.967
1 .046
Block 232.009
6 .000
Step 6 Model
232.009 6
.000
55
Table 4.5. Hosmer and Lemeshow Test
Step Chi-square
df Sig.
1 .000
. 2
.000 .
3 28.804
2 .000
4 26.016
3 .000
5 24.907
4 .000
6 44.944
6 .000
Table 4.6. Contingency Table for Hosmer and Lemeshow Test
LU_90_97 = 0 LU_90_97 = 1
Observed Expected
Observed Expected
Total Step 1
1 6252
6252.000 4762
4762.000 11014
1 2596
2570.541 1549
1574.459 4145
Step 2 2
3656 3681.459
3213 3187.541
6869 1
2091 2014.536
1063 1139.464
3154 2
2655 2731.464
2149 2072.536
4804 3
505 556.129
486 434.871
991 Step 3
4 1001
949.871 1064
1115.129 2065
1 2091
2016.910 1063
1137.090 3154
2 2354
2401.158 1787
1739.842 4141
3 505
549.390 486
441.610 991
4 301
327.932 362
335.068 663
Step 4
5 1001
956.610 1064
1108.390 2065
1 1828
1779.887 941
989.113 2769
2 263
236.411 122
148.589 385
3 1991
2064.714 1542
1468.286 3533
4 789
775.948 601
614.052 1390
5 779
751.959 712
739.041 1491
Step 5
6 602
643.082 844
802.918 1446
1 1240
1166.615 554
627.385 1794
2 588
615.602 387
359.398 975
3 1093
1048.434 656
700.566 1749
4 1161
1249.071 1008
919.929 2169
5 608
594.415 447
460.585 1055
6 642
668.763 639
612.237 1281
7 607
574.291 558
590.709 1165
Step 6
8 313
334.809 513
491.191 826
56 The chi-square goodness-of-fit test tests the null hypothesis that
the step is justified. Here the step is from the constant only model null model to the all-independents model. When as here the step was to
add a variable, the inclusion is justified if the significance of the step is less than 0.05. Table 4.4 Step 6, is showing value of Sig. is 0 or less
than 0.05 so means that the model is fit to predict deforestation The Hosmer and Lemeshow Goodness-of-fit Test computes a
chi-square from observed and expected frequencies. The p-value Sig. = 0, here is computed from the chi-square distribution with 6 degree of
freedom df and indicates that the logistic model is a good fit. That is, if the Hosmer and Lemeshow Goodness-of-Fit test statistic is 0.05 or
less, it reject the null hypothesis that there is no difference between the observe and predicted value of deforestation dependent variable.
Table 4.7. Classification Table
Observed Predicted
LU_90_97 1
Percentage Correct
LU_90_97 6165
87 98.6
1 4552
210 4.4
Step 1 Overall Percentage
57.9 6165
87 98.6
LU_90_97 1
4552 210
4.4 Step 2
Overall Percentage 57.9
5251 1001
84.0 LU_90_97
1 3698
1064 22.3
Step 3 Overall Percentage
57.3 4950
1302 79.2
LU_90_97 1
3336 1426
29.9 Step 4
Overall Percentage 57.9
5650 602
90.4 LU_90_97
1 3918
844 17.7
Step 5 Overall Percentage
59.0 5381
871 86.1
LU_90_97 1
3884 878
18.4 Step 6
Overall Percentage 56.8
a The cut value is .500
57 The classification table above Table 4.7 is a 2x2 table which
tells correct and incorrect estimates for the full model with the independents as well as the constant. The effectiveness of the model
is prediction of the phenomenon can be summarized in a simple data Table 4.7 from the total 11,014 sample pixels, 4762 were changed
deforested and 6252 pixels were unchanged forest and non forest stable. In every step of the regression, all pixels are evaluated by
the model and a value is predicted for each pixel. The value of both correctly and wrongly predicted pixels for each step is shown in Table
4.7. In other words, in this table, the group of changed forest and unchanged area pixels are compared with what is predicted for them
by the model. From the sixth step last step in Table 4.7 is clear that 18.4 of deforested pixels correctly classified by the model, and
86.1 of stable pixels are classified correctly. In this step also is showed overall percentage correct 56.8, means a total prediction
accuracy is 56.8. The accuracy of prediction is 56.8 is actually too small for
being used the model for predicting deforestation rate at the same location and different time 1997 – 2001 according to available data.
This less overall accuracy was caused by there are many independent variables contributed deforestation driving force that being assumed
are able to increase deforestation occurrence. For instants, center of population LogR_CP contributed only 7.4 of value 1
deforestation, shore line LogR_SL less than 1 km is only 25, distance from river less than 1.000 meters contributed just 38.9
percents from total area, and distance from existing road less than 1 km contributed the worst value only 2.7.
At least, all independent variables except altitude can be used to predict deforestation, and good-fit for the model of logistic
regression and its equation. This reason is based on the statistical significance testing.
The Wald test is a way of testing the significance of particular explanatory variables in a statistical model. In logistic regression we
have a binary outcome variable and one or more explanatory variables. For each explanatory variable in the model there will be an
associated parameter.
Table 4.8. Variables in the Equation and Wald test
95.0 C.I.for EXPB β
S.E. Wald
df Sig.
ExpB Lower
Upper Step 6f
LOGR_RIV -.232
.042 30.593
1 .000
.793 .730
.861 LOGR_ROAD
1.130 .135
70.293 1
.000 3.097
2.378 4.033
LOGR_SL .348
.046 57.956
1 .000
1.416 1.294
1.548 LOGR_CP
.354 .080
19.584 1
.000 1.425
1.218 1.667
LOGR_COM .082
.041 3.968
1 .046
1.086 1.001
1.177 LOGR_SLP
-.150 .055
7.506 1
.006 .861
.773 .958
Constant -.238
.055 18.903
1 .000
.788
The Wald test, described by Polit 1996 and Agresti 1990 in Blackwell 2006, is one of a number of ways of testing whether the
parameters associated with a group of explanatory variables are zero. If for a particular explanatory variable, or group of explanatory
variables, the Wald test is significant, then we would conclude that the parameters associated with these variables are not zero, so that the
variables should be included in the model. If the Wald test is not significant then these explanatory variables can be omitted from the
model. In Table 4.8 it can be seen that none of the explanatory
variable or predictor is zero, it means that Wald test is failed to reject null hypothesis and It means that all variables of road, river, distance
from population center, distance form coast line, slope, and aspect can be used to predict deforestation.
LogR_COM or independent variables of aspect and LogR_Slp Slope have higher Sig. value. In Appendix IV, it can be seen that,
58
59 aspect and slope have contributed more cells that being assumed for
deforestation occurrence. Slope has contributed 83.8 of value 1 as indicated tends to
be deforested, and aspect contributed 43.4.
4.3.3. Logistic coefficients and correlation