Within R-squared in a panel with fixed effects
Posted: Thu Aug 04, 2011 3:27 am
Hi there,
I am using eviews version 7 and I am struggling with the r-squared interpretation in eviews and the within Rsquared calcualation. I know there has been two posts on that topic already, but I was not sure how to calculate the within R-squared manually. Even more so, my research in on web regarding fixed effect panel estimation with regard to within R-squared was very poor.
Here is what my econometric specification is about: I have a fixed effect panel regression model with a cross-section of 19 countries. I regress log(spread) on various country-specific and global macrofactors.
For the coefficient of determination eviews describes in its manual: "R-squared and F-statistics are based on the difference between the residuals sums of squares from the estimated model, and the sums of squares from a single constant-only specification, not from a fixed-effect-only specification"
In order to get the within R2 I did the following (but I am not really sure whether this is correct):
First I calculated a constant only model (with no fixed effects) where I regress log(spread) against a constant and take the reported sum of squared residuals (SSQ1) (is this by the way the total sum of squared?)
In a second step I estimate the complete model, but with no fixed effects (is this the commonly termed restricted model? it seems there is no unanimous terminology.) and then take the reported sum of squared residuals (SSQ2).
Eventually, I calculate the R-squared within as: R2_within=(SSQ1-SSQ2)/SSQ1=1-(SSQ2/SSQ1)
I am really not sure whether this is the right way to do it and would therefore really appreciate if you could comment on it.
Lastly, I am interested in knowing whether there is an easy method or a function in eviews that yields the percentage variance explained for each regressor.
I would really appreciate if you could answer my questions. I would help me a lot. Thank you very much in advance.
best,
Simon
I am using eviews version 7 and I am struggling with the r-squared interpretation in eviews and the within Rsquared calcualation. I know there has been two posts on that topic already, but I was not sure how to calculate the within R-squared manually. Even more so, my research in on web regarding fixed effect panel estimation with regard to within R-squared was very poor.
Here is what my econometric specification is about: I have a fixed effect panel regression model with a cross-section of 19 countries. I regress log(spread) on various country-specific and global macrofactors.
For the coefficient of determination eviews describes in its manual: "R-squared and F-statistics are based on the difference between the residuals sums of squares from the estimated model, and the sums of squares from a single constant-only specification, not from a fixed-effect-only specification"
In order to get the within R2 I did the following (but I am not really sure whether this is correct):
First I calculated a constant only model (with no fixed effects) where I regress log(spread) against a constant and take the reported sum of squared residuals (SSQ1) (is this by the way the total sum of squared?)
In a second step I estimate the complete model, but with no fixed effects (is this the commonly termed restricted model? it seems there is no unanimous terminology.) and then take the reported sum of squared residuals (SSQ2).
Eventually, I calculate the R-squared within as: R2_within=(SSQ1-SSQ2)/SSQ1=1-(SSQ2/SSQ1)
I am really not sure whether this is the right way to do it and would therefore really appreciate if you could comment on it.
Lastly, I am interested in knowing whether there is an easy method or a function in eviews that yields the percentage variance explained for each regressor.
I would really appreciate if you could answer my questions. I would help me a lot. Thank you very much in advance.
best,
Simon