Dear all,
I'm a newbie in Stata (and in this forum) and appreciate to get a short feedback on my current research. I'm working on an unbalanced panel data (time-series cross-section: N(17)<T(about 50)). My deppvar is a continuous variable measuring the amount of change (%) in the partisan composition of governments. Currently, I'm using only two predictors (both continuous variables): enps (effective number of parties) and disprop (disproportionality of the electoral system). I'm testing my hyphothese with 4 models: 1) xtreg changegov enps disprop, re (this one is suggested by the Hausman test); 2) xtreg changegov enps disprop, fe; 3) xtpcse changegov enps disprop v001-v017x, nocons pa c(ar1); 4) xtpcse changegov enps disprop, nocons pa c(ar1) (the third model includes country-dummy variables to an ols panel corrected standard error model). The problem is that I get very different results for the fourth model, where I do not check for the variance within each country. How shoud I interpret these results? How can I explain the huge difference in the coefficients observed in model 4?
1) . xtreg changegv enps disprop if changegv>0, re
Random-effects GLS regression Number of obs = 272
Group variable: v010x Number of groups = 17
R-sq: within = 0.0298 Obs per group: min = 4
between = 0.2281 avg = 16.0
overall = 0.0602 max = 33
Wald chi2(2) = 11.92
corr(u_i, X) = 0 (assumed) Prob > chi2 = 0.0026
------------------------------------------------------------------------------
changegv | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
enps | -3.443985 1.746737 -1.97 0.049 -6.867527 -.020443
disprop | 1.622911 .6206442 2.61 0.009 .4064704 2.839351
_cons | 64.80253 8.612676 7.52 0.000 47.922 81.68307
-------------+----------------------------------------------------------------
sigma_u | 15.294472
sigma_e | 30.698939
rho | .19885385 (fraction of variance due to u_i)
2) . xtreg changegv enps disprop if changegv>0, fe
Fixed-effects (within) regression Number of obs = 272
Group variable: v010x Number of groups = 17
R-sq: within = 0.0317 Obs per group: min = 4
between = 0.1691 avg = 16.0
overall = 0.0460 max = 33
F(2,253) = 4.15
corr(u_i, Xb) = -0.0534 Prob > F = 0.0169
------------------------------------------------------------------------------
changegv | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
enps | -2.127273 1.867207 -1.14 0.256 -5.804523 1.549976
disprop | 1.818288 .6961345 2.61 0.010 .4473312 3.189245
_cons | 54.57076 8.486088 6.43 0.000 37.85839 71.28313
-------------+----------------------------------------------------------------
sigma_u | 19.678583
sigma_e | 30.698939
rho | .29123483 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(16, 253) = 5.68 Prob > F = 0.0000
3) . xtpcse changegv enps disprop v001y-v017y if changegv>0, nocons pa c(ar1)
Number of gaps in sample: 184
(note: computations for rho restarted at each gap)
(note: at least one disturbance covariance assumed 0, no common time periods
between panels)
Prais-Winsten regression, correlated panels corrected standard errors (PCSEs)
Group variable: v010x Number of obs = 272
Time variable: Year Number of groups = 17
Panels: correlated (unbalanced) Obs per group: min = 4
Autocorrelation: common AR(1) avg = 16
Sigma computed by pairwise selection max = 33
Estimated covariances = 153 R-squared = 0.7843
Estimated autocorrelations = 1 Wald chi2(19) = 14384.04
Estimated coefficients = 19 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
| Panel-corrected
changegv | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
enps | -1.919386 1.788073 -1.07 0.283 -5.423945 1.585172
disprop | 1.807869 .761169 2.38 0.018 .3160046 3.299732
v001y | 53.58167 11.95574 4.48 0.000 30.14885 77.01449
v002y | 43.23424 11.21997 3.85 0.000 21.2435 65.22498
v003y | 65.35594 12.72261 5.14 0.000 40.42008 90.2918
v004y | 44.99634 10.54194 4.27 0.000 24.33452 65.65816
v005y | 28.52919 14.81355 1.93 0.054 -.5048315 57.56322
v006y | 47.22787 10.73425 4.40 0.000 26.18913 68.26661
v007y | 64.14213 12.96508 4.95 0.000 38.73103 89.55322
v008y | 52.80235 10.11913 5.22 0.000 32.96922 72.63549
v009y | 77.1562 10.60733 7.27 0.000 56.36622 97.94618
v010y | 31.23241 9.516262 3.28 0.001 12.58088 49.88394
v011y | 53.43765 7.043008 7.59 0.000 39.63361 67.2417
v012y | 47.58272 10.5497 4.51 0.000 26.90568 68.25976
v013y | 96.14394 8.596785 11.18 0.000 79.29455 112.9933
v014y | 77.77919 10.41524 7.47 0.000 57.3657 98.19269
v015y | 92.09427 7.144872 12.89 0.000 78.09058 106.098
v016y | 74.04883 12.67093 5.84 0.000 49.21426 98.8834
v017y | 73.0615 14.3143 5.10 0.000 45.00599 101.117
-------------+----------------------------------------------------------------
rho | .1516119
------------------------------------------------------------------------------
4) . xtpcse changegv enps disprop if changegv>0, nocons pa c(ar1)
Number of gaps in sample: 184
(note: computations for rho restarted at each gap)
(note: at least one disturbance covariance assumed 0, no common time periods
between panels)
Prais-Winsten regression, correlated panels corrected standard errors (PCSEs)
Group variable: v010x Number of obs = 272
Time variable: Year Number of groups = 17
Panels: correlated (unbalanced) Obs per group: min = 4
Autocorrelation: common AR(1) avg = 16
Sigma computed by pairwise selection max = 33
Estimated covariances = 153 R-squared = 0.5996
Estimated autocorrelations = 1 Wald chi2(2) = 477.31
Estimated coefficients = 2 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
| Panel-corrected
changegv | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
enps | 8.66553 .7371991 11.75 0.000 7.220646 10.11041
disprop | 3.336742 .6505748 5.13 0.000 2.061639 4.611845
-------------+----------------------------------------------------------------
rho | .427295
------------------------------------------------------------------------------
Many thanks,
David Romario
I'm a newbie in Stata (and in this forum) and appreciate to get a short feedback on my current research. I'm working on an unbalanced panel data (time-series cross-section: N(17)<T(about 50)). My deppvar is a continuous variable measuring the amount of change (%) in the partisan composition of governments. Currently, I'm using only two predictors (both continuous variables): enps (effective number of parties) and disprop (disproportionality of the electoral system). I'm testing my hyphothese with 4 models: 1) xtreg changegov enps disprop, re (this one is suggested by the Hausman test); 2) xtreg changegov enps disprop, fe; 3) xtpcse changegov enps disprop v001-v017x, nocons pa c(ar1); 4) xtpcse changegov enps disprop, nocons pa c(ar1) (the third model includes country-dummy variables to an ols panel corrected standard error model). The problem is that I get very different results for the fourth model, where I do not check for the variance within each country. How shoud I interpret these results? How can I explain the huge difference in the coefficients observed in model 4?
1) . xtreg changegv enps disprop if changegv>0, re
Random-effects GLS regression Number of obs = 272
Group variable: v010x Number of groups = 17
R-sq: within = 0.0298 Obs per group: min = 4
between = 0.2281 avg = 16.0
overall = 0.0602 max = 33
Wald chi2(2) = 11.92
corr(u_i, X) = 0 (assumed) Prob > chi2 = 0.0026
------------------------------------------------------------------------------
changegv | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
enps | -3.443985 1.746737 -1.97 0.049 -6.867527 -.020443
disprop | 1.622911 .6206442 2.61 0.009 .4064704 2.839351
_cons | 64.80253 8.612676 7.52 0.000 47.922 81.68307
-------------+----------------------------------------------------------------
sigma_u | 15.294472
sigma_e | 30.698939
rho | .19885385 (fraction of variance due to u_i)
2) . xtreg changegv enps disprop if changegv>0, fe
Fixed-effects (within) regression Number of obs = 272
Group variable: v010x Number of groups = 17
R-sq: within = 0.0317 Obs per group: min = 4
between = 0.1691 avg = 16.0
overall = 0.0460 max = 33
F(2,253) = 4.15
corr(u_i, Xb) = -0.0534 Prob > F = 0.0169
------------------------------------------------------------------------------
changegv | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
enps | -2.127273 1.867207 -1.14 0.256 -5.804523 1.549976
disprop | 1.818288 .6961345 2.61 0.010 .4473312 3.189245
_cons | 54.57076 8.486088 6.43 0.000 37.85839 71.28313
-------------+----------------------------------------------------------------
sigma_u | 19.678583
sigma_e | 30.698939
rho | .29123483 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(16, 253) = 5.68 Prob > F = 0.0000
3) . xtpcse changegv enps disprop v001y-v017y if changegv>0, nocons pa c(ar1)
Number of gaps in sample: 184
(note: computations for rho restarted at each gap)
(note: at least one disturbance covariance assumed 0, no common time periods
between panels)
Prais-Winsten regression, correlated panels corrected standard errors (PCSEs)
Group variable: v010x Number of obs = 272
Time variable: Year Number of groups = 17
Panels: correlated (unbalanced) Obs per group: min = 4
Autocorrelation: common AR(1) avg = 16
Sigma computed by pairwise selection max = 33
Estimated covariances = 153 R-squared = 0.7843
Estimated autocorrelations = 1 Wald chi2(19) = 14384.04
Estimated coefficients = 19 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
| Panel-corrected
changegv | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
enps | -1.919386 1.788073 -1.07 0.283 -5.423945 1.585172
disprop | 1.807869 .761169 2.38 0.018 .3160046 3.299732
v001y | 53.58167 11.95574 4.48 0.000 30.14885 77.01449
v002y | 43.23424 11.21997 3.85 0.000 21.2435 65.22498
v003y | 65.35594 12.72261 5.14 0.000 40.42008 90.2918
v004y | 44.99634 10.54194 4.27 0.000 24.33452 65.65816
v005y | 28.52919 14.81355 1.93 0.054 -.5048315 57.56322
v006y | 47.22787 10.73425 4.40 0.000 26.18913 68.26661
v007y | 64.14213 12.96508 4.95 0.000 38.73103 89.55322
v008y | 52.80235 10.11913 5.22 0.000 32.96922 72.63549
v009y | 77.1562 10.60733 7.27 0.000 56.36622 97.94618
v010y | 31.23241 9.516262 3.28 0.001 12.58088 49.88394
v011y | 53.43765 7.043008 7.59 0.000 39.63361 67.2417
v012y | 47.58272 10.5497 4.51 0.000 26.90568 68.25976
v013y | 96.14394 8.596785 11.18 0.000 79.29455 112.9933
v014y | 77.77919 10.41524 7.47 0.000 57.3657 98.19269
v015y | 92.09427 7.144872 12.89 0.000 78.09058 106.098
v016y | 74.04883 12.67093 5.84 0.000 49.21426 98.8834
v017y | 73.0615 14.3143 5.10 0.000 45.00599 101.117
-------------+----------------------------------------------------------------
rho | .1516119
------------------------------------------------------------------------------
4) . xtpcse changegv enps disprop if changegv>0, nocons pa c(ar1)
Number of gaps in sample: 184
(note: computations for rho restarted at each gap)
(note: at least one disturbance covariance assumed 0, no common time periods
between panels)
Prais-Winsten regression, correlated panels corrected standard errors (PCSEs)
Group variable: v010x Number of obs = 272
Time variable: Year Number of groups = 17
Panels: correlated (unbalanced) Obs per group: min = 4
Autocorrelation: common AR(1) avg = 16
Sigma computed by pairwise selection max = 33
Estimated covariances = 153 R-squared = 0.5996
Estimated autocorrelations = 1 Wald chi2(2) = 477.31
Estimated coefficients = 2 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
| Panel-corrected
changegv | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
enps | 8.66553 .7371991 11.75 0.000 7.220646 10.11041
disprop | 3.336742 .6505748 5.13 0.000 2.061639 4.611845
-------------+----------------------------------------------------------------
rho | .427295
------------------------------------------------------------------------------
Many thanks,
David Romario
