[Anshul Anand]

Thank you very much, Sir! You helped me a lot, otherwise I would have gone the wrong direction.

[Anshul Anand]

I have seen that I should have use the categorical variable -arrival- as dummies. So, each immigrant cohort arrival as dummy. Therefore, if all immigrant cohort arrival have value zero, then the intercept will be the natives. So, my model looks like (ignore variable tento4, since this is only variable equal 10^(-4)). is051 is variable indicating if immigrant(1) or native(0):

svy: regress lnhourlyw_w c.age c.age#c.age c.age#c.age#c.age#c.tento4 c.age#i.is051
>  c.age#c.age#i.is051 c.age#c.age#c.age#i.is051#c.tento4 i.arvpre1980 i.arv1980 i.ar
> v1985 i.arv1990 i.arv1995 i.arv2000 if year==2004
(running regress on estimation sample)

Survey: Linear regression

Number of strata   =         1                  Number of obs     =     10,726
Number of PSUs     =    10,726                  Population size   =  1,317,293
Design df         =     10,725
F(  12,  10714)   =     244.61
Prob > F          =     0.0000
R-squared         =     0.2189


Linearized
lnhourlyw_w       Coef.   Std. Err.      t    P>t     [95% Conf. Interval]

age    .1339664   .0152074     8.81   0.000     .1041572    .1637757

c.age#c.age   -.0022769      .0004    -5.69   0.000    -.0030609    -.001493

c.age#c.age#c.age#
c.tento4    .1261596   .0333069     3.79   0.000     .0608718    .1914473

is051#c.age
foreign     .1088956   .0338962     3.21   0.001     .0424528    .1753384

is051#c.age#c.age
foreign    -.0024298   .0008359    -2.91   0.004    -.0040683   -.0007913

is051#c.age#c.age#
c.age#c.tento4
foreign     .1797599   .0657532     2.73   0.006     .0508714    .3086484

1.arvpre1980    .2623956   .0966542     2.71   0.007     .0729355    .4518557
1.arv1980    .2941179   .0927958     3.17   0.002      .112221    .4760148
1.arv1985    .2821505    .087123     3.24   0.001     .1113733    .4529277
1.arv1990    .2810173   .0846227     3.32   0.001     .1151412    .4468933
1.arv1995    .4809988   .0861537     5.58   0.000     .3121216     .649876
1.arv2000    .6242478   .0946557     6.59   0.000      .438705    .8097905
_cons   -.9649148   .4808887    -2.01   0.045    -1.907546    -.022284

Now, I would like to predict the regression at age=40 for each immigrant cohort arrival and native and then look at the wage differential between each immigrant cohort arrival with native.

margins arvpre1980 arv1980 arv1985 arv1990 arv1995 arv2000, at(age =    (40)) pwcompare

Pairwise comparisons of predictive margins

Number of strata   =         1                  Subpop. no. obs   =    10,725
Subpop. size      =    .
Design df         =    10,725
Model VCE    : Linearized

Expression   : Linear prediction, predict()
at           : age             =          40


Delta-method         Unadjusted
Contrast   Std. Err.     [95% Conf. Interval]

arvpre1980
1 vs 0     .2623956   .0966542      .0729355    .4518557

arv1980
1 vs 0     .2941179   .0927958       .112221    .4760148

arv1985
1 vs 0     .2821505    .087123      .1113733    .4529277

arv1990
1 vs 0     .2810173   .0846227      .1151412    .4468933

arv1995
1 vs 0     .4809988   .0861537      .3121216     .649876

arv2000
1 vs 0     .6242478   .0946557       .438705    .8097905

However, this looks strange, since the wage differential at age=40 are same as the coefficient of each immigrant cohort arrival in the above regression.

[Clyde Schechter]

Well, the -margins- results you are getting are completely consistent with your model. The model is linear, and the various arv* variables are not interacted with anything else in the model. Each of the arv* variables is a 0/1 variable designating an arrival cohort. So, in fact, the marginal effect of each such variable does equal its regression coefficient in this model.

Why were you expecting anything different? Perhaps your model is not consistent with your understanding of the process.