My data is stored in a dataframe called 'Gap3',and the structure is like below:
> summary(Gap3)
region time rC2R g
Beijing : 19 1995 : 31 Min. :0.000 Min. : 3.80
Tianjin : 19 1996 : 31 1st Qu.:2.404 1st Qu.: 9.70
Hebei : 19 1997 : 31 Median :2.819 Median :11.50
Shanxi : 19 1998 : 31 Mean :2.898 Mean :11.47
InnerMongolia: 19 1999 : 31 3rd Qu.:3.240 3rd Qu.:12.90
Liaoning : 19 2000 : 31 Max. :5.605 Max. :23.80
(Other) :475 (Other):403 NA's :2
CCover FDI FDS HC
Min. :0.0348 Min. :0.00000 Min. :0.732 Min. : 2.599
1st Qu.:0.2402 1st Qu.:0.01100 1st Qu.:1.884 1st Qu.: 7.040
Median :0.2945 Median :0.02240 Median :2.303 Median : 7.852
Mean :0.3130 Mean :0.03247 Mean :2.438 Mean : 7.805
3rd Qu.:0.3726 3rd Qu.:0.04180 3rd Qu.:2.733 3rd Qu.: 8.546
Max. :0.7852 Max. :0.44940 Max. :7.303 Max. :12.028
NA's :2 NA's :8 NA's :2
I IE MedCResCover MedCWoCover
Min. :0.0000 Min. :0.0000 Min. :0.0000 Min. :0.0000
1st Qu.:0.3332 1st Qu.:0.0792 1st Qu.:0.0536 1st Qu.:0.0518
Median :0.4389 Median :0.1231 Median :0.2201 Median :0.2159
Mean :0.4879 Mean :0.3029 Mean :0.3159 Mean :0.2103
3rd Qu.:0.6251 3rd Qu.:0.3269 3rd Qu.:0.5560 3rd Qu.:0.3000
Max. :1.1126 Max. :2.0513 Max. :1.9183 Max. :0.9507
NA's :3
pGDP rCityRate RCover rFkOut
Min. : 1826 Min. :0.1590 Min. :0.00160 Min. :0.0492
1st Qu.: 6226 1st Qu.:0.3300 1st Qu.:0.04312 1st Qu.:0.1060
Median : 12437 Median :0.4210 Median :0.08665 Median :0.1456
Mean : 19034 Mean :0.4447 Mean :0.21798 Mean :0.1829
3rd Qu.: 26133 3rd Qu.:0.5280 3rd Qu.:0.27025 3rd Qu.:0.2071
Max. :100105 Max. :0.8960 Max. :1.05240 Max. :1.2914
NA's :153 NA's :1
TCover UnCover
Min. :0.0000 Min. :0.0000
1st Qu.:0.0997 1st Qu.:0.1536
Median :0.1591 Median :0.1880
Mean :0.2273 Mean :0.2028
3rd Qu.:0.3090 3rd Qu.:0.2376
And the number of missing values is:
> sum(is.na(Gap3))
[1] 171
the dimension is :
> dim(Gap3)
[1] 589 18
My current purpose is to decide which effect I should use, the Fix Effect or the Random Effect. My model function and what i did is showed below:
> form1
rC2R ~ TCover + MedCResCover + UnCover + pGDP + I(pGDP^2) + g +
FDS + FDI + IE + I + rFkOut + HC + rCityRate
>gap.fe1 <- plm(form1, data=Gap3,model="within")
>gap.rd1 <- plm(form1, data=Gap3,model="random")
>phtest(gap.fe1,gap.rd1)
Error in solve.default(dvcov) :
system is computationally singular: reciprocal condition number = 1.117e-22
Then I multiplied variables who is less than 1 by 100, and did the procedure again, but the singularity sitll exsited. In my 3rd attempt, I deleted the pGDP and I(pGDP^2), this time it worked sucessfully.
phtest(gap.fe3,gap.rd3)
Hausman Test
data: form3
chisq = 94.967, df = 11, p-value = 1.762e-15
alternative hypothesis: one model is inconsistent
Could anybody tell me the reason, why the pGDP and I(pGDP^2) caused singularity?
