[Carlo Lazzaro]

Alex:
- one reason amongst others for different outputs is due to fact that, while -xtreg, fe- gets rid of time-invariant predictors, -regress- does not;
- considering -Age- as continuous is correct;
- including a squared term without the underlying linear one is not correct. Moreover, in your model -c.age##c.age- makes sense when compared with range for -Age-;
- I woudl not care at all about -vif-.

[Alex Mueller]

Carlo,

thanks for sharing your thoughts about this.

Is it after all reasonable to do this comparison between xtreg and reg , given they cannot really deliver the same output (as you stated in your last post)?

So would you simply look at a correlation matrix between the independent variables? Or skip any possible multicollinearity issue completely?

Cheers, Alex

[Carlo Lazzaro]

Alex:
1) imperfect as it may be, you're some times induced to make such comparison just to have a rough idea of wha's going on, because some post estimation commands which are available for -regress- are, in fact, unavailable as Stata official commands, for -xtreg- (it may however imply that -vif-, say, is possibly overvalued and that you are better off with investigating the -vce- matrix after -xtreg);
2) exception made for rare instances when a sky-rocketing -vif- biases the regression model) I would not pay much attention to -vif-.

[Alex Mueller]

Carlo,

thanks for your opinion on this topic.

This implies:

 xtreg performance promoted_in_pbservation_period (...controlvariables...) i.year, fe robust
estat vce, correlation

right?

I will give it a go with both approaches (vce and vif) and see if my model is ok.


Thanks for all your advices during the last days, Carlo!

[Carlo Lazzaro]

Alex:
your code is correct.
I would stop at -estat vce-, though (ie, without considering -vif-).

[Alex Mueller]

Thanks for your help, Carlo!

[Alex Mueller]

Carlo, (or anyone else please feel free to answer)

One more question that is maybe a little more related to statistics in general rather to stata specifically:

does

 reg c.dependent_variable c./i.independent_variable c./i.controlvariables

calculate the least squared dummy variable (LSDV) estimator of the regression and is therefor equivalent to

 xtreg dependent_variable Independent_variable controlvariables , fe

?

Best, A.

[Carlo Lazzaro]

Alex:
LSDV estimator, that relates to panel data analysis, implies the creation of a dummy for each -panelid-, N-1 of them to be incuded as predictors in the right-hand side of the regression equation.
LSDV estimator cam be fit via -areg-; the results are the same as the ones obtained via -xtreg, fe-, as you can see from the following toy-example:

use "http://www.stata-press.com/data/r15/nlswork.dta"
. areg ln_wage age, absorb(idcode)

Linear regression, absorbing indicators         Number of obs     =     28,510
Absorbed variable: idcode                       No. of categories =      4,710
                                                F(   1,  23799)   =    2720.20
                                                Prob > F          =     0.0000
                                                R-squared         =     0.6636
                                                Adj R-squared     =     0.5970
                                                Root MSE          =     0.3035

------------------------------------------------------------------------------
     ln_wage |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         age |   .0181349   .0003477    52.16   0.000     .0174534    .0188164
       _cons |   1.148214   .0102579   111.93   0.000     1.128107     1.16832
------------------------------------------------------------------------------
F test of absorbed indicators: F(4709, 23799) = 8.808         Prob > F = 0.000

. xtreg ln_wage age, fe

Fixed-effects (within) regression               Number of obs     =     28,510
Group variable: idcode                          Number of groups  =      4,710

R-sq:                                           Obs per group:
     within  = 0.1026                                         min =          1
     between = 0.0877                                         avg =        6.1
     overall = 0.0774                                         max =         15

                                                F(1,23799)        =    2720.20
corr(u_i, Xb)  = 0.0314                         Prob > F          =     0.0000

------------------------------------------------------------------------------
     ln_wage |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         age |   .0181349   .0003477    52.16   0.000     .0174534    .0188164
       _cons |   1.148214   .0102579   111.93   0.000     1.128107     1.16832
-------------+----------------------------------------------------------------
     sigma_u |  .40635023
     sigma_e |  .30349389
         rho |  .64192015   (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(4709, 23799) = 8.81                 Prob > F = 0.0000

.

[Alex Mueller]

Carlo, thanks for that clarification!

I will try to implement your hints in my model!

Best wishes for you! Alex

[Noor Jabeen]

Hello all, kindly help me with this issue i am working on a panel data. Even i have tried using sigmamore and sigmaless. Should i still go for fixed effect?


hausman fe re

---- Coefficients ----
| (b) (B) (b-B) sqrt(diag(V_b-V_B))
| fe re Difference S.E.
-------------+----------------------------------------------------------------
lpop | .2540629 .4703511 -.2162882 .0572511
lpop_hos | -.1704653 .4568731 -.6273384 .0510274
year |
2005 | .2221213 .1722483 .049873 .0046667
2010 | .369298 .2665685 .1027295 .010209
2015 | .4890097 .3392333 .1497764 .0153708
2017 | .5225391 .3538961 .168643 .0174148
------------------------------------------------------------------------------
b = consistent under Ho and Ha; obtained from xtreg
B = inconsistent under Ha, efficient under Ho; obtained from xtreg

Test: Ho: difference in coefficients not systematic

chi2(6) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= 184.50
Prob>chi2 = 0.0000
(V_b-V_B is not positive definite)

[Carlo Lazzaro]

Noor:
I would go -fe-.