72
Table 4.5
Variance inflation factors (VIF) of numerical independent variables on the “old”
data.
Variable
VIF
LOAN
1.151227
MORTDUE 2.720736
VALUE
2.817692
YOJ
1.089154
DEROG
1.0712
DELINQ
1.066063
CLAGE
1.174603
NINQ
1.071983
CLNO
1.226312
DEBTINC
1.059467
From Table 4.5, we see there are no large
variance inflation factors, which indicates that
there is no serious problem with collinearity in the “old” data.
Outliers and influential observations in the model are now considered. The following plots
are considered for the presence of outliers and influential observations: half-normal plots of
the
residuals, leverages and Cook’s distance statistics. The half-normal
plot of the residuals is
given in Figure 4.12.
Fig. 4.12
Half-normal plot of residuals for the “old” data.
From Figure 4.12, there does not appear to be any sign of outliers. The half-normal plot of
the leverages is given in Figure 4.13.
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
0
1
2
3
4
Half-normal quantiles
So
rte
d
Da
ta
1565
2715
73
Fig. 4.13
Half-normal plot of leverages for the “old” data.
Figure 4.13 indicates that observations numbered 1877 and 2508 may have the potential to
affect the fit of the model.
Figure 4.14
Half-normal plot of the Cook’s distance statistics for the “old” data.
The half-normal plot of the Cook’s distance statistics is given in Figure 4.14.
This plot
indicates that observations numbered 556 and 1403 may be influential. A logistic regression
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
0.
00
0.
05
0.
10
0.
15
0.
20
0.
25
0.
30
Half-normal quantiles
So
rte
d
Da
ta
1877
2508
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
0.
00
0.
01
0.
02
0.
03
0.
04
0.
05
Half-normal quantiles
So
rte
d
Da
ta
556
1403
74
model was then fitted on the “old” data excluding observations numbered 556, 1403, 1877
and 2508. The estimated parameters of this model were then compared to the model with no
removed observations.
Table 4.6
Comparison of model coefficients when possible
leverage and influential
observations are either included or excluded from the “old” data.
Table 4.6 shows the difference in the parameters when possible leverage and influential
observations are removed. The first column gives the model parameters with all observations
in the “old” data and the second column gives the model parameters when the possible
leverage and influential observations are removed. Looking at Table 4.6, the differences in
the estimated parameters between the models are minimal. Therefore, the possible leverage
and influential points will not be removed from the “old” data.
Do'stlaringiz bilan baham: