optimal model was selected using the p value (with a threshold of 0.15) and Akaike’s
information criterion (AIC) value using the stepwise method.
3
Using studentized residuals for the selected models,
4 we found that Tono City in
Iwate Prefecture and Kikugawa City in Shizuoka Prefecture are outliers. When
abnormal values are measured, Tono City, Iwate Prefecture and Edogawa Ward,
Tokyo, show strong influence. When these abnormal values and strong influence
values were removed and regression analysis performed, an improved model was
obtained:
3.2
0e+00 1e+05 2e+05
Certified solar power generation facilities (kW)
3e+05 4e+05 5e+05 6e+05 7e+05
3.4
3.6
3.8
4.0
Solar radiation (kWh/m 2 /day)
4.2
Fig. 21.3 Relationship between solar radiation and certified solar power generation facilities using
local regression smoothing (loess) (horizontal axis: kWh/m
2
/day, vertical axis: kW)
3 Using the step AIC function in the car package in R.
4 Outliers and strong influence points were analyzed using the outlier test function and influence plot
function, respectively, in the car package in R.
394
W. Zhou et al.