Panel data within standard error

[Micky Lu]

Dear all,

I have a set of panel data called u, and it looks like this:

year country industry     u
-------------------------------
2010  A         01T        1
2010  B         01T        1.5
2010  C         01B        1.2
2010  A         02T         3
2010  B         02T        2.5
2010  C         02B       1.1
....
2018  A         01T         2
2018  B         01T        3.5
2018  C         01B        2.1
2018  A         02T        2.1
2018  B         02T        1.5
2018  C         02B        2.1

I want to see if u's trend over time, more specifically, to see if variable u has a great within variation over time.

I have tried the following:

 xtset upstreamness year

I got error: repeated time values within panel
r(451);

I read this FAQ but I'm still not sure how to fix this problem. https://www.stata.com/support/faqs/d...d-time-values/

Or can I just do:

collapse(mean) u, by(year)
reg u year

and see the standard error?

Your advice would be highly appreciated.

[Carlo Lazzaro]

Micky:
you can easily fix this issue by -xtset-ting your dataset with the -panelid- only.
This workaround comes at the cost of making time-series operators unavailable (you can still add -i.year- as a predictor in the right-hand side of your regression equation, though).
In addition, if you're intrested in a non-linear relationship between -year- and the regerssand, you may want to explore the following interaction, that replaces -i.year- (the code assumes that you go -re- with default standard errors):

 
 xtset upstreamness xtreg uniform c.year##c.year

[Micky Lu]

Micky:
you can easily fix this issue by -xtset-ting your dataset with the -panelid- only.
This workaround comes at the cost of making time-series operators unavailable (you can still add -i.year- as a predictor in the right-hand side of your regression equation, though).
In addition, if you're intrested in a non-linear relationship between -year- and the regerssand, you may want to explore the following interaction, that replaces -i.year- (the code assumes that you go -re- with default standard errors):

xtset upstreamness xtreg uniform c.year##c.year

Dear Professor Lazzaro,

Thank you for your response!

Yes, it works with:

xtset upstreamness

But when I run:

xtreg upstreamness year

or

xtreg upstreamness c.year##c.year

Stata shows: the panel variable upstreamness may not be included as an independent variable
r(198);

[Carlo Lazzaro]

Micky:
Stata is right and I was wrong.
I mistyped my previous code.
Two amended options follow:

xtset industry
 xtreg upstreamness c.year##c.year

or:

xtset country
 xtreg upstreamness c.year##c.year

As an aside, please call me Carlo just like all on (and many more off) this forum do. Thanks.

[Micky Lu]

Dear Carlo,
I tried

xtset industry

and

xtset country

, but got error: string variables not allowed in varlist; industry/country is a string variable.
Sorry I should have specified industry is a string variable too.

[Carlo Lazzaro]

Micky:
see (carefully, as it may show nasty behaviors) -help encode-.

[Micky Lu]

Micky:
see (carefully, as it may show nasty behaviors) -help encode-.

Dear Carlo,
Thank you! I tried encoding industry and It works now.
May I ask a follow-up question for #2? In order to see the within variation of upstreamness, should I use random effect or fixed effect? How should I read the result table?

For example:

Random-effects GLS regression                   Number of obs     =    917,815
Group variable: industry_new                    Number of groups  =         25

R-sq:                                           Obs per group:
     within  = 0.0000                                         min =     27,680
     between = 0.0001                                         avg =   36,712.6
     overall = 0.0000                                         max =     38,025

                                                Wald chi2(1)      =       1.65
corr(u_i, X)   = 0 (assumed)                    Prob > chi2       =     0.1994

-------------------------------------------------------------------------------
 upstreamness |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
--------------+----------------------------------------------------------------
         year |  -.0001876   .0001462    -1.28   0.199    -.0004741    .0000989
c.year#c.year |          0  (omitted)
        _cons |   2.618334   .3174491     8.25   0.000     1.996145    3.240522
--------------+----------------------------------------------------------------
      sigma_u |  .59397366
      sigma_e |  .36140742
          rho |  .72980987   (fraction of variance due to u_i)
-------------------------------------------------------------------------------

Does it mean that the within std. error of upstreamness over time is 0.59397366 (sigma_u)?

Best,
Micky

[Carlo Lazzaro]

MIcky:
you have a more relevant issue to deal with.
Yo actually have a T>N (aka long) panel dataset (25 groups (N), that is panels, were observed for a minimum of 27,680 times (T)).
Hence, you should consider -xtregar- or -xtgls-.

[Micky Lu]

MIcky:
you have a more relevant issue to deal with.
Yo actually have a T>N (aka long) panel dataset (25 groups (N), that is panels, were observed for a minimum of 27,680 times (T)).
Hence, you should consider -xtregar- or -xtgls-.

Dear Carlo,
Thank you very much for your help!

Best,
Micky