Hello.
I am working with a fixed effects regression model on panel data for redistributive effects on democratic institutions amongst and a few other control variables. I am running my model with clustered standard errors. My baseline model had an extremely low-within R-squared, but a between R-square value of approx .25.
My questions is now, how come my r-squared between panel variables declines disproportionately when including time fixed effects and control variables whilst my within r-squared increases? Is the answer, that the collinearity with my explanatory variables and the time variables was so strong. That is, that the global time trends were causing the high r-squared between countries at first, and after including time fixed effects and other predictors, the within explanatory power of my model has increased to a reasonable level?
Or have I grossly misunderstood or overlooked something obvious?
My results are posted below:
*Regression with clustered SE and control variables*
. xtreg deltaG vdem gni trade unemployment, fe vce(cluster countrycode)
Fixed-effects (within) regression Number of obs = 1,140
Group variable: countrycode Number of groups = 54
R-squared: Obs per group:
Within = 0.1525 min = 1
Between = 0.3166 avg = 21.1
Overall = 0.2969 max = 22
F(4,53) = 8.01
corr(u_i, Xb) = -0.3574 Prob > F = 0.0000
(Std. err. adjusted for 54 clusters in countrycode)
------------------------------------------------------------------------------
| Robust
deltaG | Coefficient std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
vdem | -.0302197 .0232621 -1.30 0.200 -.0768775 .0164381
gni | 1.55e-06 3.43e-07 4.50 0.000 8.57e-07 2.23e-06
trade | -.0001599 .0000609 -2.63 0.011 -.000282 -.0000378
unemployment | -.0005061 .0003052 -1.66 0.103 -.0011183 .0001061
_cons | .0891974 .0185867 4.80 0.000 .0519171 .1264777
-------------+----------------------------------------------------------------
sigma_u | .02902027
sigma_e | .01100382
rho | .87429766 (fraction of variance due to u_i)
------------------------------------------------------------------------------
. *Regression with clustered SE, control variables and time fixed effects*
. xtreg deltaG vdem gni trade i.year, fe vce(cluster countrycode)
Fixed-effects (within) regression Number of obs = 1,140
Group variable: countrycode Number of groups = 54
R-squared: Obs per group:
Within = 0.2741 min = 1
Between = 0.0047 avg = 21.1
Overall = 0.0169 max = 22
F(24,53) = 7.41
corr(u_i, Xb) = -0.2665 Prob > F = 0.0000
(Std. err. adjusted for 54 clusters in countrycode)
------------------------------------------------------------------------------
| Robust
deltaG | Coefficient std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
vdem | -.0209061 .0228226 -0.92 0.364 -.0666824 .0248702
gni | 5.03e-07 3.44e-07 1.46 0.150 -1.87e-07 1.19e-06
trade | -.0002384 .0000864 -2.76 0.008 -.0004117 -.000065
|
_cons | .0952987 .0186797 5.10 0.000 .0578319 .1327655
-------------+----------------------------------------------------------------
sigma_u | .03387181
sigma_e | .01027886
rho | .9156753 (fraction of variance due to u_i)
------------------------------------------------------------------------------
Thank you very much in advance!
I am working with a fixed effects regression model on panel data for redistributive effects on democratic institutions amongst and a few other control variables. I am running my model with clustered standard errors. My baseline model had an extremely low-within R-squared, but a between R-square value of approx .25.
My questions is now, how come my r-squared between panel variables declines disproportionately when including time fixed effects and control variables whilst my within r-squared increases? Is the answer, that the collinearity with my explanatory variables and the time variables was so strong. That is, that the global time trends were causing the high r-squared between countries at first, and after including time fixed effects and other predictors, the within explanatory power of my model has increased to a reasonable level?
Or have I grossly misunderstood or overlooked something obvious?
My results are posted below:
*Regression with clustered SE and control variables*
. xtreg deltaG vdem gni trade unemployment, fe vce(cluster countrycode)
Fixed-effects (within) regression Number of obs = 1,140
Group variable: countrycode Number of groups = 54
R-squared: Obs per group:
Within = 0.1525 min = 1
Between = 0.3166 avg = 21.1
Overall = 0.2969 max = 22
F(4,53) = 8.01
corr(u_i, Xb) = -0.3574 Prob > F = 0.0000
(Std. err. adjusted for 54 clusters in countrycode)
------------------------------------------------------------------------------
| Robust
deltaG | Coefficient std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
vdem | -.0302197 .0232621 -1.30 0.200 -.0768775 .0164381
gni | 1.55e-06 3.43e-07 4.50 0.000 8.57e-07 2.23e-06
trade | -.0001599 .0000609 -2.63 0.011 -.000282 -.0000378
unemployment | -.0005061 .0003052 -1.66 0.103 -.0011183 .0001061
_cons | .0891974 .0185867 4.80 0.000 .0519171 .1264777
-------------+----------------------------------------------------------------
sigma_u | .02902027
sigma_e | .01100382
rho | .87429766 (fraction of variance due to u_i)
------------------------------------------------------------------------------
. *Regression with clustered SE, control variables and time fixed effects*
. xtreg deltaG vdem gni trade i.year, fe vce(cluster countrycode)
Fixed-effects (within) regression Number of obs = 1,140
Group variable: countrycode Number of groups = 54
R-squared: Obs per group:
Within = 0.2741 min = 1
Between = 0.0047 avg = 21.1
Overall = 0.0169 max = 22
F(24,53) = 7.41
corr(u_i, Xb) = -0.2665 Prob > F = 0.0000
(Std. err. adjusted for 54 clusters in countrycode)
------------------------------------------------------------------------------
| Robust
deltaG | Coefficient std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
vdem | -.0209061 .0228226 -0.92 0.364 -.0666824 .0248702
gni | 5.03e-07 3.44e-07 1.46 0.150 -1.87e-07 1.19e-06
trade | -.0002384 .0000864 -2.76 0.008 -.0004117 -.000065
|
_cons | .0952987 .0186797 5.10 0.000 .0578319 .1327655
-------------+----------------------------------------------------------------
sigma_u | .03387181
sigma_e | .01027886
rho | .9156753 (fraction of variance due to u_i)
------------------------------------------------------------------------------
Thank you very much in advance!
