p-value and wald test issue

[Klodi Gjini]

Hello, I have two questions regarding the p-value and the wald test. What would you have to say, if all the coefficients in my regression are non statistically significant (p-value>0.05) and if I run the test var1 var2 I get a probability higher that 0.05?? Is this model inappropriate? This model works fine for other countries, this issue arises only for one country in particular... what do you suggest? thanks

[Carlo Lazzaro]

Klodi:
you don't provide enough data to support any helpful reply (as usual, posting what you typed and what Stata gave you back would have helped enormously).
That said, the first suspect is that, for that country, your sample size is too small due, for instance, to missing data.

[Klodi Gjini]

Carlo thank you for your reply.. indeed that for this particular country I have few observations.. below you can find my code..

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(lntfp age size export sector) int employees float foreign
        . 2.0794415 1 0 0   7 0
        . 1.7917595 2 0 0  11 0
 2.467846 1.7917595 2 0 0  44 0
        . 1.3862944 2 0 0  32 0
        . 2.0794415 1 0 0   2 0
        . 2.6390574 2 0 0  24 0
        .  1.609438 4 . 0   . 0
        . 2.3025851 1 0 0   7 0
1.8328762  2.890372 3 0 0 108 0
        .  2.890372 1 0 0   5 0
 4.704588 1.0986123 2 0 0  12 0
        .  1.609438 2 . 0  10 0
        . 1.0986123 2 0 0  16 0
        . 1.3862944 2 0 0  14 0
        .  1.609438 4 0 0   . 0
 3.439583 2.6390574 2 1 0  33 1
3.5590935 2.6390574 1 0 0   6 0
        . 1.3862944 3 . 0 123 1
        .  .6931472 1 . 0   5 0
        .  2.564949 1 0 0   9 0
end
label values size cat_size
label def cat_size 1 "Micro", modify
label def cat_size 2 "Small", modify
label def cat_size 3 "Medium", modify
label def cat_size 4 "Large", modify
label values sector cat_sector
label def cat_sector 0 "Service", modify
label values foreign cat_fo
label def cat_fo 0 "Domestic", modify
label def cat_fo 1 "Foreign", modify

[Carlo Lazzaro]

Klodi:
what kind of regression model did you run?
Please, provide what you typed. Thanks.

[Klodi Gjini]

my dependent var is lntfp, and independent vars are age size export employees foreign sector.. I run regress lntfp age size sector foreign employees export

[Carlo Lazzaro]

Klodi:
your dataset excerpt is plagued with missing values:

. regress lntfp age size sector foreign employees export
note: sector omitted because of collinearity
note: export omitted because of collinearity

      Source |       SS           df       MS      Number of obs   =         5
-------------+----------------------------------   F(4, 0)         =         .
       Model |  4.85520692         4  1.21380173   Prob > F        =         .
    Residual |           0         0           .   R-squared       =    1.0000
-------------+----------------------------------   Adj R-squared   =         .
       Total |  4.85520692         4  1.21380173   Root MSE        =         0

------------------------------------------------------------------------------
       lntfp |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         age |   1.644308          .        .       .            .           .
        size |   4.311552          .        .       .            .           .
      sector |          0  (omitted)
     foreign |  -1.582149          .        .       .            .           .
   employees |  -.1055153          .        .       .            .           .
      export |          0  (omitted)
       _cons |  -4.458789          .        .       .            .           .
------------------------------------------------------------------------------

Deal with them before considering any regression model.

[Klodi Gjini]

Thank you Carlo for your time and suggestions..maybe the collinearity issue is because of the 20 first observations that I included in the dataex.. when I run regress for the whole sample, I don't get any collinearity problems..

[Carlo Lazzaro]

Klodi:
yes, it may be.
Please consider that, as a rule of thumb, you should have 10-20 observations per coefficients to have reliable coefficients.
In your case, missing values make things even harder.