XTSET I need help!!

[silvana huanqui]

Hello friends! I've never needed help before as I do now. I have an estimation from panel data, but when setting up the panel, I used xtset time observations (which is supposed to be the other way around), as some reviews are telling me.

The way I did it (reversed) makes perfect sense; the coefficients and everything else come out perfectly. When I change it "as it should be," it gives random results that make no sense at all. But I'm reading the theory, and it says that the panel should be arranged in the following way:
i1t1
i1 t2
i1 t3

However, I have it ordered with time first:
i1 t1
i2 t1
i3 t1
etc.

Im using fixed effects.

Could this help me justify why we set up the panel the other way around and it makes sense?

[Daniel Schaefer]

No, I'm afraid these data structures are identical as far as xt is concerned.

clear
set seed 12345
input int(panel time)
1 1
2 1
3 1
1 2
2 2
3 2
1 3
2 3
3 3
end

gen x = runiform()
gen y = 2 * x + rnormal()

xtset panel time

xtreg y x

sort panel time

xtreg y x

Random-effects GLS regression                   Number of obs     =          9
Group variable: panel                           Number of groups  =          3

R-squared:                                      Obs per group:
     Within  = 0.4030                                         min =          3
     Between = 0.7339                                         avg =        3.0
     Overall = 0.0736                                         max =          3

                                                Wald chi2(1)      =       0.56
corr(u_i, X) = 0 (assumed)                      Prob > chi2       =     0.4560

------------------------------------------------------------------------------
           y | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
           x |   1.521823   2.041412     0.75   0.456    -2.479271    5.522916
       _cons |   .9313119   .9967577     0.93   0.350    -1.022297    2.884921
-------------+----------------------------------------------------------------
     sigma_u |          0
     sigma_e |  1.0363245
         rho |          0   (fraction of variance due to u_i)
------------------------------------------------------------------------------

.
. sort time panel

.
. xtreg y x

Random-effects GLS regression                   Number of obs     =          9
Group variable: panel                           Number of groups  =          3

R-squared:                                      Obs per group:
     Within  = 0.4030                                         min =          3
     Between = 0.7339                                         avg =        3.0
     Overall = 0.0736                                         max =          3

                                                Wald chi2(1)      =       0.56
corr(u_i, X) = 0 (assumed)                      Prob > chi2       =     0.4560

------------------------------------------------------------------------------
           y | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
           x |   1.521823   2.041412     0.75   0.456    -2.479271    5.522916
       _cons |   .9313119   .9967577     0.93   0.350    -1.022297    2.884921
-------------+----------------------------------------------------------------
     sigma_u |          0
     sigma_e |  1.0363245
         rho |          0   (fraction of variance due to u_i)
------------------------------------------------------------------------------

[Carlo Lazzaro]

Silvana:
there is no way that you can justify swapping panelid and timevar. In fact, the only mandatory variable in -xtset- is panelvar. If your results do not make sense (whatever this may mean), double check your model specification.