Need Help, ols Prob > chibar2 = 1.0000; but fe vs re Prob>chi2 = 0.5277, which model should i choose?

[Rama Dhani]

Dear friends, I need help with a problem that I counter, I try to choose which method is suitable for my analysis whether fe, re or OLS. But due to my lack of understanding in interpreting data I keep circling which method should I use
herewith what I've done so far
much appreciate any clue that anyone could pointed out

. egen Object=group(Country)

. xtset Object Year,yearly
panel variable: Object (strongly balanced)
time variable: Year, 1990 to 2019
delta: 1 year

. xtreg MfgGrowth CP mfglaborgrowth, fe

Fixed-effects (within) regression Number of obs = 60
Group variable: Object Number of groups = 2

R-sq: Obs per group:
within = 0.2131 min = 30
between = 1.0000 avg = 30.0
overall = 0.3531 max = 30

F(2,56) = 7.58
corr(u_i, Xb) = 0.5641 Prob > F = 0.0012

--------------------------------------------------------------------------------
MfgGrowth | Coef. Std. Err. t P>|t| [95% Conf. Interval]
---------------+----------------------------------------------------------------
CP | .0675011 .0286445 2.36 0.022 .0101193 .1248828
mfglaborgrowth | .9766219 .2798877 3.49 0.001 .4159394 1.537304
_cons | .02143 .0276787 0.77 0.442 -.0340171 .0768771
---------------+----------------------------------------------------------------
sigma_u | .02553164
sigma_e | .08792706
rho | .07775999 (fraction of variance due to u_i)
--------------------------------------------------------------------------------
F test that all u_i=0: F(1, 56) = 1.27 Prob > F = 0.2641

. estimate store fe

. xtreg MfgGrowth CP mfglaborgrowth, re

Random-effects GLS regression Number of obs = 60
Group variable: Object Number of groups = 2

R-sq: Obs per group:
within = 0.2095 min = 30
between = 1.0000 avg = 30.0
overall = 0.3581 max = 30

Wald chi2(2) = 31.80
corr(u_i, X) = 0 (assumed) Prob > chi2 = 0.0000

--------------------------------------------------------------------------------
MfgGrowth | Coef. Std. Err. z P>|z| [95% Conf. Interval]
---------------+----------------------------------------------------------------
CP | .0901471 .0204808 4.40 0.000 .0500055 .1302887
mfglaborgrowth | 1.041938 .2744863 3.80 0.000 .5039551 1.579922
_cons | .0014943 .0213528 0.07 0.944 -.0403564 .043345
---------------+----------------------------------------------------------------
sigma_u | 0
sigma_e | .08792706
rho | 0 (fraction of variance due to u_i)
--------------------------------------------------------------------------------

. estimate store re

. hausman fe re

---- Coefficients ----
| (b) (B) (b-B) sqrt(diag(V_b-V_B))
| fe re Difference S.E.
-------------+----------------------------------------------------------------
CP | .0675011 .0901471 -.022646 .0200261
mfglaborgr~h | .9766219 1.041938 -.0653165 .054721
------------------------------------------------------------------------------
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(2) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= 1.28
Prob>chi2 = 0.5277
(V_b-V_B is not positive definite)

. xttest0

Breusch and Pagan Lagrangian multiplier test for random effects

MfgGrowth[Object,t] = Xb + u[Object] + e[Object,t]

Estimated results:
| Var sd = sqrt(Var)
---------+-----------------------------
MfgGrowth | .0116923 .108131
e | .0077312 .0879271
u | 0 0

Test: Var(u) = 0
chibar2(01) = 0.00
Prob > chibar2 = 1.0000

[Jared Greathouse]

You've got panel data, so OLS via xtreg is what empirical.


You wanting fixed or random effects is a question of both theory and empirics.

[Carlo Lazzaro]

Rama:
welcome to this forum.
Both F-test appearing as a footnote below the -xtreg,fe- outcome table and -xttest0- after -xtreg,re- do not support the evidence of a panrl- wise effect.
Hence you shoyod go pooled OLS.

[Rama Dhani]

Rama:
welcome to this forum.
Both F-test appearing as a footnote below the -xtreg,fe- outcome table and -xttest0- after -xtreg,re- do not support the evidence of a panrl- wise effect.
Hence you shoyod go pooled OLS.

Dear Mr Greathouse and Mr Lazzaro
Thank you for the reply, very much appreciate it
If I may, since I'm really a noobs in stata,
based on the hausman test, how to interpret this data?

chi2(2) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= 1.28
Prob>chi2 = 0.5277
(V_b-V_B is not positive definite)

CMIIW if the prob>chi2 = 0.0000 ; the fixed model can be accepted, but how about in this case where prob>chi2 = 0.5277 . Does it mean, while the test doesn't satisfy fixed approach, the logic is to accept the counter part on the test which is the random model. Assuming I haven't done the Breusch and Pagan Lagrangian test

Sincerely

[Carlo Lazzaro]

Rama:
the issue here is the lack of panel-wise effect.

[Rama Dhani]

Rama:
the issue here is the lack of panel-wise effect.

I'm sorry, perhaps if you don't mind you could elaborate more
appreciate it

[Carlo Lazzaro]

Rama:
-xtreg,fe-:

F test that all u_i=0: F(1, 56) = 1.27 Prob > F = 0.2641

If the null (i.e., all u_i=0) is not rejected, there's no evidence of a panel-wise effect.

-xtreg,re-:

Prob > chibar2 = 1.0000

The outcome of -xttes0- does not reject the null (i.e., Var(u) = 0): again, there'sno evidence of a panel-wise effect.

In addition, -hausman- outcome is not reliable as

(V_b-V_B is not positive definite)

If what above sounds still obscure, youmay want to read -xtreg- entry in Stata .pdf manual along with a decent textbook on panel data econometrics.

[Rama Dhani]

Dear Carlo I've tweaking with the data and I got the result as follow

-xtreg,fe-

 Prob > F          =     0.0000

F test that all u_i=0: F(4, 138) = 0.97                      Prob > F = 0.4284

-xtreg,re-

Prob > chi2       =     0.0000

but the hausman test shown

 Test:  Ho:  difference in coefficients not systematic

                  chi2(2) = (b-B)'[(V_b-V_B)^(-1)](b-B)
                          =        0.05
                Prob>chi2 =      0.9767

My question is, which model should I use, since both test now shown panel-wise effect? is it fixed or random? or not both of it?

[Carlo Lazzaro]

Rama:
1)

F test that all u_i=0: F(4, 138) = 0.97                      Prob > F = 0.4284

shows no evidence of panel-wise effect;
2) as far as -xtreg,re- is concerned, have you run -xttest0- after -xtreg,re-?

[Rama Dhani]

Dear Carlo
yes the xtest0 is as follow

 Test:   Var(u) = 0
                             chibar2(01) =     2.23
                          Prob > chibar2 =   0.0676

does it mean i have to go OLS?


Suppose I have to go with OLS instead of fixed or random effect, since there is no prove of panel wise effect. Does it mean I have to breakdown my panel database onto separate database of time series data? then run the regression? (right now my data is consist of 6 different countries panel data)



if I have to go with OLS, in terms of equation , do I have to divide the independent variable onto two separate equation?
in this case I have one dependent variable and two independent variable...
then the equation will be

Y=a+ Xb+e and Y=a=+Xd+e

or

Y=a+Xb+Xd+e

[Rama Dhani]

q

[Carlo Lazzaro]

Rama:
not quite.
You should go pooled OLS coding up as follows:

reg MfgGrowth CP mfglaborgrowth, vce(cluster panelid)