I am trying to analyse a panel data of 30 countries and each for 2007-2016. I have gone through the fixed effect models and specifications but still got some questions and will be very thankful if anyone could give some hints.
I am trying several options to come up with correct model. In all cases I have significant coefficient, this is not an issue, but there are still some issues I cannot handle.
First - can you please say what xtreg does when we do not specify re, fe and just leave like that. Is it doing just pooled regression?
Code:
. xtreg growth_rate ln_prod_lagged i.year
Random-effects GLS regression Number of obs = 260
Group variable: id Number of groups = 26
R-sq: Obs per group:
within = 0.3517 min = 10
between = 0.0569 avg = 10.0
overall = 0.3303 max = 10
Wald chi2(10) = 123.74
corr(u_i, X) = 0 (assumed) Prob > chi2 = 0.0000
--------------------------------------------------------------------------------
growth_rate | Coef. Std. Err. z P>|z| [95% Conf. Interval]
---------------+----------------------------------------------------------------
ln_prod_lagged | -.0156277 .0076401 -2.05 0.041 -.030602 -.0006535
|
year |
2008 | -.0651795 .0183597 -3.55 0.000 -.1011638 -.0291951
2009 | -.1140758 .0183588 -6.21 0.000 -.1500583 -.0780932
2010 | .068735 .0183539 3.74 0.000 .0327621 .1047079
2011 | -.0280067 .018377 -1.52 0.128 -.0640249 .0080116
2012 | -.0567071 .0183908 -3.08 0.002 -.0927524 -.0206619
2013 | -.0330568 .0183925 -1.80 0.072 -.0691054 .0029918
2014 | -.0457379 .0184061 -2.48 0.013 -.0818132 -.0096626
2015 | -.0348612 .0184136 -1.89 0.058 -.0709512 .0012288
2016 | -.0325653 .0184291 -1.77 0.077 -.0686857 .0035551
|
_cons | .2311222 .082884 2.79 0.005 .0686725 .3935719
---------------+----------------------------------------------------------------
sigma_u | .00594337
sigma_e | .06168062
rho | .0091993 (fraction of variance due to u_i)
--------------------------------------------------------------------------------
Second-if I have only 30 countries in my database is it valid to use fe. I get good results with it but the coeficient is too big and not intuitive, though significant.
Code:
. xtreg growth_rate ln_prod_lagged, fe
Fixed-effects (within) regression Number of obs = 260
Group variable: id Number of groups = 26
R-sq: Obs per group:
within = 0.1395 min = 10
between = 0.0569 avg = 10.0
overall = 0.0172 max = 10
F(1,233) = 37.78
corr(u_i, Xb) = -0.9701 Prob > F = 0.0000
--------------------------------------------------------------------------------
growth_rate | Coef. Std. Err. t P>|t| [95% Conf. Interval]
---------------+----------------------------------------------------------------
ln_prod_lagged | -.2549382 .0414766 -6.15 0.000 -.3366553 -.173221
_cons | 2.788079 .4491071 6.21 0.000 1.903249 3.672908
---------------+----------------------------------------------------------------
sigma_u | .14074613
sigma_e | .07530423
rho | .77744573 (fraction of variance due to u_i)
--------------------------------------------------------------------------------
F test that all u_i=0: F(25, 233) = 2.06 Prob > F = 0.0030
I also checked testparm i.year and use dummies to control time effects.
Code:
. xtreg growth_rate ln_prod_lagged i.year, fe
Fixed-effects (within) regression Number of obs = 260
Group variable: id Number of groups = 26
R-sq: Obs per group:
within = 0.4450 min = 10
between = 0.0569 avg = 10.0
overall = 0.0622 max = 10
F(10,224) = 17.96
corr(u_i, Xb) = -0.9479 Prob > F = 0.0000
--------------------------------------------------------------------------------
growth_rate | Coef. Std. Err. t P>|t| [95% Conf. Interval]
---------------+----------------------------------------------------------------
ln_prod_lagged | -.2972772 .0458798 -6.48 0.000 -.3876885 -.206866
|
year |
2008 | -.0481278 .0173312 -2.78 0.006 -.0822808 -.0139748
2009 | -.0984127 .0172964 -5.69 0.000 -.1324971 -.0643283
2010 | .0687048 .0171071 4.02 0.000 .0349933 .1024163
2011 | .0059763 .0179805 0.33 0.740 -.0294562 .0414088
2012 | -.0137751 .0184814 -0.75 0.457 -.0501948 .0226445
2013 | .0108456 .0185418 0.58 0.559 -.0256931 .0473843
2014 | .0053554 .0190243 0.28 0.779 -.0321342 .0428449
2015 | .0197774 .019284 1.03 0.306 -.0182238 .0577787
2016 | .0287665 .0198108 1.45 0.148 -.010273 .0678059
|
_cons | 3.248588 .4916846 6.61 0.000 2.279669 4.217507
---------------+----------------------------------------------------------------
sigma_u | .16449166
sigma_e | .06168062
rho | .87672547 (fraction of variance due to u_i)
--------------------------------------------------------------------------------
F test that all u_i=0: F(25, 224) = 2.60 Prob > F = 0.0001
Might it be because of the small database. One of my assumptions is that I have only 10 years periods for 30 countires and it comes 300 datapoints, that is why i cannot properly capture the fixed effects. Please share your thought and maybe there are some suggestions how to deal with small panel datasets?
Thank you very much for the support beforehand!
Sara
0 Response to Panel data analysis using fixed effects models
Post a Comment