Fixed-Effect Regression - Model w/ Specific Years

[Sally Silk]

Dear community,

I am currently working on running a fixed regression and my outputs are as follows thus far:

xtreg gdpg epu polyarchy2 polyepu prienr secenr trade gcf mort infl leab i.year, fe

Fixed-effects (within) regression Number of obs = 361
Group variable: con_cod Number of groups = 21

R-squared: Obs per group:
Within = 0.6076 min = 5
Between = 0.6961 avg = 17.2
Overall = 0.5081 max = 25

F(34,306) = 13.94
corr(u_i, Xb) = -0.7789 Prob > F = 0.0000

------------------------------------------------------------------------------
gdpg | Coefficient Std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
epu | -.0195063 .0065379 -2.98 0.003 -.0323712 -.0066414
polyarchy2 | -5.74489 2.823997 -2.03 0.043 -11.3018 -.1879795
polyepu | .0217498 .0081547 2.67 0.008 .0057035 .0377961
prienr | .0005445 .032489 0.02 0.987 -.0633856 .0644746
secenr | -.0323323 .0174088 -1.86 0.064 -.0665885 .0019239
trade | .0361706 .013165 2.75 0.006 .0102653 .0620759
gcf | .3896079 .0531247 7.33 0.000 .285072 .4941438
mort | .064435 .0450484 1.43 0.154 -.0242089 .1530788
infl | -.2670408 .1809947 -1.48 0.141 -.6231926 .089111
leab | -.1593164 .1956075 -0.81 0.416 -.5442225 .2255897
|
year |
1998 | -2.040763 .8565902 -2.38 0.018 -3.726316 -.3552108
1999 | -.145276 .8528607 -0.17 0.865 -1.82349 1.532938
2000 | .68755 .8604055 0.80 0.425 -1.00551 2.38061
2001 | -1.370337 .8860054 -1.55 0.123 -3.113771 .3730976
2002 | -1.626749 .8789538 -1.85 0.065 -3.356308 .1028092
2003 | -1.334191 .8848809 -1.51 0.133 -3.075413 .4070303
2004 | -.2823252 .929538 -0.30 0.762 -2.111421 1.54677
2005 | -.9311301 .8976217 -1.04 0.300 -2.697422 .835162
2006 | .2996599 .9222856 0.32 0.745 -1.515165 2.114484
2007 | -.5428204 .9231149 -0.59 0.557 -2.359277 1.273636
2008 | -3.309176 .9445857 -3.50 0.001 -5.167881 -1.45047
2009 | -6.626927 1.038822 -6.38 0.000 -8.671065 -4.582788
2010 | .2988279 1.017783 0.29 0.769 -1.70391 2.301566
2011 | -1.138885 1.00184 -1.14 0.257 -3.110253 .8324817
2012 | -2.412823 1.00586 -2.40 0.017 -4.3921 -.4335463
2013 | -1.566465 1.092598 -1.43 0.153 -3.71642 .5834908
2014 | -1.203544 1.116053 -1.08 0.282 -3.399655 .9925657
2015 | -1.809251 1.10572 -1.64 0.103 -3.985028 .3665264
2016 | -1.23565 1.140929 -1.08 0.280 -3.48071 1.00941
2017 | -.6799038 1.125669 -0.60 0.546 -2.894935 1.535128
2018 | -.4203813 1.09799 -0.38 0.702 -2.580947 1.740184
2019 | -1.113978 1.130128 -0.99 0.325 -3.337784 1.109828
2020 | -8.227144 1.058676 -7.77 0.000 -10.31035 -6.143938
2021 | 2.954225 1.04375 2.83 0.005 .9003882 5.008061
|
_cons | 11.14383 14.50456 0.77 0.443 -17.39747 39.68513
-------------+----------------------------------------------------------------
sigma_u | 3.0034887
sigma_e | 2.0802914
rho | .67579923 (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(20, 306) = 3.56 Prob > F = 0.0000

I am now looking to run the same model, but rather than including time-fixed effects for each year, I want to use the dummy variables for the years that are significant as a part of my equation. I am unsure how to do this and would appreciate any help. Thank you!

Sally Silk

[Carlo Lazzaro]

Sally:
what you have in mind does not seem right to me.
The results of your outcome table are adjusted for the predictors that you included in.
Changing the number of predictors (including significant years only, in your case) will effect your results.
That said, you may create a two-level categorical variable (1=years that turned out statistically significant in your previous regression; 0=the opposite) and see what happens.
Last but not least, as you are probaly dealing with a T>N panel dataset, your should take a look at -xtregar-.

[Sally Silk]

Thank you for your suggestions Carlo! I will give them a try.