Dear all, I am running a fixed effects model with the the following outcome:
The variables NOMCOMM until ATT are independent variables, the rest are controls . I also run separate models for each "set" of independent variables (e.g., one model only with NOMCOMM and NOMCOMM_IND as indep. var. and so on). Now the significances of the different variables are usually the same between the "joint" model and the separate model. However, AUDCOMM is insignificant in the joint model (as shown above) but significant at the 5% level in the model where only AUDCOMM and AUDCOMM_IND are included as independent variables next to the controls. Now my question is:
What is the most likely reason for this? Is it just that AUDCOMM might be impacted by some of the other independent variables that are included in the joint model? Or is there something else that could be the reason? I have found out through a VIF and correlation matrix analysis that multicollinearity is not a problem in the dataset (VIF < 2 for all variables in the joint model).
Thanks a lot in advance!
Code:
xtreg EMTOTAL NOMCOMM NOMCOMM_IND COMPCOMM COMPCOMM_IND AUDCOMM AUDCOMM_IND ATT SUSCOMM BSIZE BGD INC INDEP DUAL ROA LEV FSIZE MULT SKILLS i.YEAR, fe vce(cluster ID) // NOMCOMM_IND*; COMPCOMM_IND**; AUDCOMM_IND**; FSIZE***; MULT**
Fixed-effects (within) regression Number of obs = 2,197
Group variable: ID Number of groups = 374
R-squared: Obs per group:
Within = 0.2327 min = 1
Between = 0.3455 avg = 5.9
Overall = 0.3084 max = 12
F(28, 373) = .
corr(u_i, Xb) = 0.3376 Prob > F = .
(Std. err. adjusted for 374 clusters in ID)
------------------------------------------------------------------------------
| Robust
EMTOTAL | Coefficient std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
NOMCOMM | .2009752 .1856693 1.08 0.280 -.1641146 .566065
NOMCOMM_IND | -.0047824 .0028462 -1.68 0.094 -.010379 .0008141
COMPCOMM | .1019664 .1088173 0.94 0.349 -.1120058 .3159386
COMPCOMM_IND | .0129957 .0054742 2.37 0.018 .0022316 .0237599
AUDCOMM | .2679236 .3238402 0.83 0.409 -.3688578 .904705
AUDCOMM_IND | -.0148516 .0058586 -2.54 0.012 -.0263715 -.0033316
ATT | .0000887 .001503 0.06 0.953 -.0028667 .0030441
SUSCOMM | .0202437 .046531 0.44 0.664 -.0712522 .1117396
BSIZE | -.1299499 .1058398 -1.23 0.220 -.3380674 .0781677
BGD | .0000569 .0023852 0.02 0.981 -.0046332 .004747
INC | .0149433 .026777 0.56 0.577 -.0377096 .0675962
INDEP | -.003699 .0035347 -1.05 0.296 -.0106493 .0032514
DUAL | -.0500078 .0593761 -0.84 0.400 -.1667617 .0667461
ROA | -.0103688 .0201865 -0.51 0.608 -.0500624 .0293249
LEV | -.3416139 .214623 -1.59 0.112 -.7636366 .0804088
FSIZE | .4993285 .0971064 5.14 0.000 .3083838 .6902731
MULT | .2298811 .091051 2.52 0.012 .0508436 .4089186
SKILLS | .0003486 .001168 0.30 0.765 -.001948 .0026453
|
YEAR |
2012 | -.138169 .0659451 -2.10 0.037 -.2678397 -.0084983
2013 | -.2258837 .0604424 -3.74 0.000 -.3447343 -.1070331
2014 | -.266024 .059275 -4.49 0.000 -.3825791 -.1494689
2015 | -.2482914 .057194 -4.34 0.000 -.3607545 -.1358284
2016 | -.2634871 .0606245 -4.35 0.000 -.3826957 -.1442784
2017 | -.3663103 .0613012 -5.98 0.000 -.4868496 -.245771
2018 | -.3977307 .0642845 -6.19 0.000 -.5241363 -.2713251
2019 | -.4206851 .0652578 -6.45 0.000 -.5490044 -.2923657
2020 | -.4750202 .0725768 -6.55 0.000 -.6177312 -.3323092
2021 | -.5931662 .07366 -8.05 0.000 -.738007 -.4483253
2022 | -.6212537 .0779522 -7.97 0.000 -.7745346 -.4679729
2023 | -.6250788 .0816359 -7.66 0.000 -.7856032 -.4645545
|
_cons | 2.565551 2.273944 1.13 0.260 -1.905806 7.036907
-------------+----------------------------------------------------------------
sigma_u | 2.0473429
sigma_e | .31648166
rho | .97666218 (fraction of variance due to u_i)
------------------------------------------------------------------------------
What is the most likely reason for this? Is it just that AUDCOMM might be impacted by some of the other independent variables that are included in the joint model? Or is there something else that could be the reason? I have found out through a VIF and correlation matrix analysis that multicollinearity is not a problem in the dataset (VIF < 2 for all variables in the joint model).
Thanks a lot in advance!
