Gls problem

[Gabriella Miniero]

Hi Forum!
I have a problem with GLS method. I'm not sure that the Stata command "xtgls" is correct for me, just because I haven't panels data. My response variable is the "consumption", my obs are the Italian provinces, and my predictors are unemployment, the average wage, etc. Can you help me?

. regress consumilog disoccup inattivitàlog retrib_medialog componenti_famsqr

      Source |       SS       df       MS              Number of obs =     107
-------------+------------------------------           F(  4,   102) =  200.05
       Model |  5.43680568     4  1.35920142           Prob > F      =  0.0000
    Residual |  .693034164   102  .006794453           R-squared     =  0.8869
-------------+------------------------------           Adj R-squared =  0.8825
       Total |  6.12983984   106  .057828678           Root MSE      =  .08243

-----------------------------------------------------------------------------------
       consumilog |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
------------------+----------------------------------------------------------------
         disoccup |  -.0132551   .0027748    -4.78   0.000     -.018759   -.0077512
    inattivitàlog |  -.6670955   .0805296    -8.28   0.000    -.8268255   -.5073654
  retrib_medialog |   .4781122   .1531212     3.12   0.002     .1743971    .7818273
componenti_famsqr |   .0401237   .0133465     3.01   0.003      .013651    .0665963
            _cons |    5.11111   1.678655     3.04   0.003     1.781507    8.440714
-----------------------------------------------------------------------------------

[Carlo Lazzaro]

Gabriella:
if you do not have a panel dataset, you should go -regress- (provided tha your regressand is continuous, as I can surmise from your post. By the way, for those on this forum who do not speak Italian, Gabriella's regressand is logged consumption).
That said, what's the matter with your data/results?

[Gabriella Miniero]

First, I have to test the normality assumption and then run this "generalized least square". Is the "xtgls" command correct for my data?

[Carlo Lazzaro]

Gabriella:
- in OLS setting, normality assumption refers to residuals distribution only;
- GLS is a particular case of OLS when the residual distribution shows heteroskedastcity.In Stata, it simply implies to replace default with -robust- standard errors;
- being conceived for long panel datasets, -xtgls- is not the way to go when you have one wave of data only.

[Gabriella Miniero]

Thank you so much!

[Gabriella Miniero]

Carlo:
I’m actually doing a project and I’m "obligated" to choose between ridge regression and gls. But I don’t think the latter can fit a model with continuous variables. Using robust standard error could be a solution ? I thought the GLM might be an alternative. How do you propose ?


Kind regards,
Gabriella

[Carlo Lazzaro]

Gabriella:
questions like this one are impossible to reply positively.
I find difficult to envisage how ridge regerssion can compete with gls for the same research topic, especially in an OLS setting.
I would discuss these topics with your supervisor, if any.

[Abid Jahangir]

Hello Everyone!
I have a short panel T =12 and N = 262. The properties of the dataset require the use of the FE approach for estimation. do I need to test for the normality of the distribution of error terms? what if they are not normally distributed? how will it affect the generalization of my findings?

[Carlo Lazzaro]

Abid:
normality is a weak requirement of the residual distribution. You can skip it.
It is more relevant that the epsilon is not heteroskedastic and/or autocorrelated or correlated across panels.
For the future, pleass start a new thread, as the title of this one has nothing to do with your query. Thanks.