Need help with getting results in fixed effects model

[Ismoil Ismoilov]

Dear Statalist,

I am running regression (N>T) on the determinants of economic growth of countries (DV) and one of the variables (IV) is the country`s initial level of GDP per capita.

xtset id year
xtreg growth inflation(-1) trade government_consumption(-1) population_growth initial_GDPPC, fe rob cluster(id)

As we know, the data for the initial level of GDP per capita is the same for a particular country during the whole period. And when I run the regression, STATA omits this variable because of the collinearity issue.

Most of the research on this topic used a fixed effects model and showed the results for all independent variables (without omitting). Is there any method or command toget the result for initial level of GDP per capita?

Thank you.

[Joro Kolev]

What you are showing

reg Y X(-1)

is not valid Stata syntax and should generate an error.

As you explain in fixed effects regressions everything that is time constant gets dropped.

Maybe you have misunderstood, and the literature regresses on lagged values of some independent variables, say the lagged value of inflation?

[Ismoil Ismoilov]

Maybe you have misunderstood, and the literature regresses on lagged values of some independent variables, say the lagged value of inflation?

Dear Joro,

Yes, authors used lag of IVs, almost all of them used lag of Government consumption and some added lag of inflation

[Joro Kolev]

Well, those lagged values are not constant across time. So you do something like

 
 xtreg growth l.inflation trade l.government_consumption population_growth initial_GDPPC, fe rob cluster(id)

[Ismoil Ismoilov]

Well, those lagged values are not constant across time. So you do something like

xtreg growth l.inflation trade l.government_consumption population_growth initial_GDPPC, fe rob cluster(id)

Joro, thank you for your response and attention!

I`m sorry I couldn`t clarify what I need in my question.
I need to get the results for IV with constant value within some period (Initial GDPPC). Most research has the results for this variable and I`m struggling on it

[Carlo Lazzaro]

Ismoil:
I share Joro's concern about what you're actually after.
That said, if you want to stick with -fe- the only way that might get what you want (but not always, because of collinearity) is using -regress. instead of -xtreg,fe- (an approach that rarely worths consdering):

. use "https://www.stata-press.com/data/r17/nlswork.dta"
(National Longitudinal Survey of Young Women, 14-24 years old in 1968)

. regress ln_wage i.year i.race i.idcode if idcode<=5
note: 5.idcode omitted because of collinearity.

      Source |       SS           df       MS      Number of obs   =        61
-------------+----------------------------------   F(18, 42)       =      1.48
       Model |  2.91651475        18  .162028597   Prob > F        =    0.1461
    Residual |  4.59521911        42  .109409979   R-squared       =    0.3883
-------------+----------------------------------   Adj R-squared   =    0.1261
       Total |  7.51173386        60  .125195564   Root MSE        =    .33077

------------------------------------------------------------------------------
     ln_wage | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
        year |
         69  |   .1847799   .3307718     0.56   0.579    -.4827446    .8523044
         70  |   .0639881   .2909513     0.22   0.827    -.5231754    .6511516
         71  |  -.0128146    .282324    -0.05   0.964    -.5825676    .5569384
         72  |   .1124264    .282324     0.40   0.692    -.4573265    .6821794
         73  |   .1928424    .282324     0.68   0.498    -.3769106    .7625954
         75  |   .1510399    .282324     0.53   0.595     -.418713    .7207929
         77  |   .1594544   .2908195     0.55   0.586    -.4274432     .746352
         78  |   .3482667   .2908195     1.20   0.238    -.2386309    .9351643
         80  |   .1820693    .282324     0.64   0.523    -.3876837    .7518222
         82  |    .033479   .2909513     0.12   0.909    -.5536844    .6206425
         83  |   .2247867   .2959791     0.76   0.452    -.3725232    .8220966
         85  |   .3874729   .2959791     1.31   0.198    -.2098371    .9847828
         87  |    .328272   .2959791     1.11   0.274    -.2690379    .9255819
         88  |   .4046278   .2959791     1.37   0.179    -.1926821    1.001938
             |
        race |
      Black  |   .1086187   .1458739     0.74   0.461    -.1857668    .4030042
             |
      idcode |
          2  |  -.3498364    .136466    -2.56   0.014     -.625236   -.0744368
          3  |  -.4278554   .1310835    -3.26   0.002    -.6923926   -.1633183
          4  |   .0866698   .1500993     0.58   0.567    -.2162428    .3895824
          5  |          0  (omitted)
             |
       _cons |   1.719945   .2433232     7.07   0.000     1.228899    2.210992
------------------------------------------------------------------------------

. xtreg ln_wage i.year i.race if idcode<=5, fe
note: 2.race omitted because of collinearity.

Fixed-effects (within) regression               Number of obs     =         61
Group variable: idcode                          Number of groups  =          5

R-squared:                                      Obs per group:
     Within  = 0.1714                                         min =         11
     Between = 0.0052                                         avg =       12.2
     Overall = 0.1263                                         max =         15

                                                F(14,42)          =       0.62
corr(u_i, Xb) = -0.0190                         Prob > F          =     0.8325

------------------------------------------------------------------------------
     ln_wage | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
        year |
         69  |   .1847799   .3307718     0.56   0.579    -.4827446    .8523044
         70  |   .0639881   .2909513     0.22   0.827    -.5231754    .6511516
         71  |  -.0128146    .282324    -0.05   0.964    -.5825676    .5569384
         72  |   .1124264    .282324     0.40   0.692    -.4573265    .6821794
         73  |   .1928424    .282324     0.68   0.498    -.3769106    .7625954
         75  |   .1510399    .282324     0.53   0.595     -.418713    .7207929
         77  |   .1594544   .2908195     0.55   0.586    -.4274432     .746352
         78  |   .3482667   .2908195     1.20   0.238    -.2386309    .9351643
         80  |   .1820693    .282324     0.64   0.523    -.3876837    .7518222
         82  |    .033479   .2909513     0.12   0.909    -.5536844    .6206425
         83  |   .2247867   .2959791     0.76   0.452    -.3725232    .8220966
         85  |   .3874729   .2959791     1.31   0.198    -.2098371    .9847828
         87  |    .328272   .2959791     1.11   0.274    -.2690379    .9255819
         88  |   .4046278   .2959791     1.37   0.179    -.1926821    1.001938
             |
        race |
      Black  |          0  (omitted)
       _cons |   1.630989   .2401092     6.79   0.000     1.146429    2.115548
-------------+----------------------------------------------------------------
     sigma_u |  .19540886
     sigma_e |  .33077179
         rho |  .25871288   (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(4, 42) = 4.39                       Prob > F = 0.0047

.

[Ismoil Ismoilov]

Ismoil:
I share Joro's concern about what you're actually after.
That said, if you want to stick with -fe- the only way that might get what you want (but not always, because of collinearity) is using -regress. instead of -xtreg,fe- (an approach that rarely worths consdering):

Thank you, Carlo!