[Carlo Lazzaro]

Enrico:
for the sake of precision, the same source (page 29) states that -vce(cluster clusterid)- standard errors "...allows arbitary correlations of disturbances within clusters as well as arbitrary heteroskedasticity across clusters."
As expected, there's no difference between standard errors calculated via -newey- or -vce(cluster clusterid)- as we have -lag(0)- in (one wave) cross-sectional dataset:

ysuse auto.dta
. g id=_n
. g time=1

. tsset id time

Panel variable: id (strongly balanced)
 Time variable: time, 1 to 1
         Delta: 1 unit

. newey price mpg, lag(0)

Regression with Newey–West standard errors      Number of obs     =         74
Maximum lag = 0                                 F(  1,        72) =      17.28
                                                Prob > F          =     0.0001

------------------------------------------------------------------------------
             |             Newey–West
       price | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
         mpg |  -238.8943   57.47701    -4.16   0.000    -353.4727    -124.316
       _cons |   11253.06   1376.393     8.18   0.000     8509.272    13996.85
------------------------------------------------------------------------------

. regress price mpg, vce(cluster id)

Linear regression                               Number of obs     =         74
                                                F(1, 73)          =      17.28
                                                Prob > F          =     0.0001
                                                R-squared         =     0.2196
                                                Root MSE          =     2623.7

                                    (Std. err. adjusted for 74 clusters in id)
------------------------------------------------------------------------------
             |               Robust
       price | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
         mpg |  -238.8943   57.47701    -4.16   0.000    -353.4459   -124.3428
       _cons |   11253.06   1376.393     8.18   0.000     8509.914    13996.21
------------------------------------------------------------------------------

.

[Enrico Azzini]

I was also doubtful about the advantages of using xtreg but maybe there were some features that I did't know that could be useful. Thankyou for you suggestion for newey I will deepen the topic.

[Carlo Lazzaro]

Enrico:
cross-sectional datasets have one wave of data only.
This means that you do not have within panel variation, for instance (that is, what you usually investigate via -fe- specification) and that you have -insufficient observations- if you go -re-.

[Enrico Azzini]

Hi Carlo excuse me if I take advantage of your availability. I was reading about the methods used to investigate functional form misspecification. I read on introductory econometrics by Wooldridge (p.306) about this test.
I was surprised by the fact that I seem to have understood that in order to test for this possibility I should adding in my model the square and the cube of the linear prediction of the model. I was wondering why you in the #5 post included in the model the fitted values and its square and not the square and the cube of fitted values?

[Carlo Lazzaro]

Enrico:
see -linktest- and, more substantively, do the two tests aimed at investigating possible misspecification of the functional form of the regressand produce opposite results?

[Enrico Azzini]

Thank you for you answer. I haven't checked for both tests yet. However I read on Microeconometrics Using Stata that in this case I should use as regressors only the fitted values and its square and omit the regressors used in the original model

[Carlo Lazzaro]

Enrico:
correct.
The -linktest- entry in Stata .pdf manual proposes something similar too.
The substantive issue is that sq_fitted values reaching statistical significance mean that something is wrong with the right-hand side of your regressoon equation (even though the test formally focuses on the regressand; this is covered in the -linktest' entry mentioned above).