I have a df based on a national survey conducted every two years; the time period is 2010-14 and I filtered the df in order to have only person that appears al least two times. In this way I have a panel df but unbalanced.
I run a regression to study which variables influence the participation in complementary pension (it is voluntary in my country). I run a one-side fixed effect regression and now I want to run a two side fixed effect regression (both individual and time).
The individual variable is uid and time variable is year. I used the plm package in r:
df.p <- plm.data(df, c("uid", "year")
and run the regression:
reg1 <- plm(pens ~ woman + age + I(age^2/100) + high + medium + nord + centre, model="within", effect="twoways", data=df.p)
where high and medium are dummies regarding the education level and nord and centre regard geographic location. For the sake of simplicity I omitted other variables that are present in the original model (20 variables).
After at least 1 hour of working I run the summary command:
summary(reg1)
after another hour of working I got the error:
Error in crossprod(t(X), beta) : non-conformable arguments
so I supposed there was a multicollinearity problem. So I check the multicollinearity with the correlation matrix:
p1 <- with(df, data.frame(woman=woman, age=age, high=high, medium=medium, nord=nord, centre=centre))
round(cor(p1),3)
Consider that I created the matrix using all the variables (here omitted for the sake of simplicity, as I wrote). I didn't find any relevant value. I also check for the variance inflation factor:
vif(p1)
and I got:
No variable from the 20 input variables has collinearity problem.
At this point I suppose the the collinearity problem could be determined by the fact that I run a two side regression but I don't know how to manage the problem.
Thanks in advance.
