Saving coefficients from a regression and later used them as independent variable in another regression

[Silvana Builes]

Hi,

I am running this model: regress ln_inglab_hora experiencia experiencia_cua escolaridad period2 period3 i.id_region_birt_3, where i.id_region_birt_3 represent dummies for cohorts I have created. I only want to use the betas associated to such dummies in another regression but I am struggling on how to do it..

Can anyone guide me on this?

Thank you

[Ken Chui]

One way could be regsave. Check it out and download it using search regsave. A tutorial can be found here: https://www.stata.com/meeting/boston...ton10_reif.pdf

[George Ford]

matrix B = e(b)

or,

estimates store e1

and then

estimates restore e1

to get the results back.

[Carlo Lazzaro]

Silvana:
as an aside to other valuable advice, you may want to consider -statsby- (caveat: save a copy of the master dataset beforehand):

. sysuse auto.dta
(1978 automobile data)

. statsby, by(foreign): regress mpg gear turn
(running regress on estimation sample)

      Command: regress mpg gear turn
           By: foreign

Statsby groups:
.

. list

     +--------------------------------------------+
     |  foreign   _b_gea~o     _b_turn    _b_cons |
     |--------------------------------------------|
  1. | Domestic   3.814768   -.7819742   41.52745 |
  2. |  Foreign   6.566766   -2.327916   84.17068 |
     +--------------------------------------------+

[Silvana Builes]

Dear Ken, George ad Carlo
Thank you very much


What about obtaining fixed effects from the following regression and store them in a b vector as if they were the estimated parameters of a regrssion?

xtset id_region_birt_3 period
xtreg ln_inglab_hora experiencia experiencia_cua escolaridad ethnici1 edumo2 edumo3 edufa2 edufa3 female region_naci_Car region_naci_Ori region_naci_Cen region_naci_Pac region_naci_Otro period2 period3, fe r

[Carlo Lazzaro]

Silvana:
it is the same as your first regression code with -id_region- as a predictor.

[George Ford]

Carlo's method will do it and get you a matrix, though you may have to remove parts of it as it will include the coefficients on the Xs.

Another way is

predict fe , u

after xtreg. It creates a variable equal to the fixed effect coefficients.

or use

reghdfe .... , absorb(id , savefe)

which also creates a variable.

This would work to create a matrix of the FE:

use https://www.stata-press.com/data/r18/nlswork, clear
keep in 1/1000

xtreg ln_w age  tenure , fe
predict fe , u

preserve
     keep idcode fe
     collapse (first) fe, by(idcode)
     mkmat fe , matrix(FE)
restore

matrix list FE

[Silvana Builes]

Dear George Ford,
thank you.


I am actually running this regression

reghdfe ln_inglab_hora experiencia experiencia_cua escolaridad, abs(id_region_birt_3 period, savefe) vce(cluster id_region_birt_3)

but I'm not sure how to obtain those fe that I've saved with savefe. I understand how to do it with predict when you have an output from xtreg, but I don't get it for reghdfe

[Carlo Lazzaro]

Silvana:
the following toy-example may help:

. use https://www.stata-press.com/data/r18/nlswork.dta
(National Longitudinal Survey of Young Women, 14-24 years old in 1968)

. reghdfe ln_w grade age ttl_exp tenure not_smsa south , absorb(FE1=idcode FE2=year)
(dropped 550 singleton observations)
note: grade is probably collinear with the fixed effects (all partialled-out values are close to zero; tol = 1.0e-09)
(MWFE estimator converged in 8 iterations)
note: grade omitted because of collinearity

HDFE Linear regression                            Number of obs   =     27,541
Absorbing 2 HDFE groups                           F(   5,  23375) =     261.99
                                                  Prob > F        =     0.0000
                                                  R-squared       =     0.6762
                                                  Adj R-squared   =     0.6185
                                                  Within R-sq.    =     0.0531
                                                  Root MSE        =     0.2939

------------------------------------------------------------------------------
     ln_wage | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
       grade |          0  (omitted)
         age |   .0114497   .0099824     1.15   0.251    -.0081165    .0310159
     ttl_exp |   .0323758   .0015046    21.52   0.000     .0294266     .035325
      tenure |   .0104689   .0009264    11.30   0.000     .0086531    .0122847
    not_smsa |  -.0914148   .0096386    -9.48   0.000    -.1103071   -.0725225
       south |  -.0640471   .0110539    -5.79   0.000    -.0857134   -.0423808
       _cons |   1.161841    .290702     4.00   0.000     .5920463    1.731636
------------------------------------------------------------------------------

Absorbed degrees of freedom:
-----------------------------------------------------+
 Absorbed FE | Categories  - Redundant  = Num. Coefs |
-------------+---------------------------------------|
      idcode |      4147           0        4147     |
        year |        15           1          14     |
-----------------------------------------------------+

.

. list FE1 FE2 if idcode==1

       +------------------------+
       |       FE1          FE2 |
       |------------------------|
    1. | .41666994    .10056242 |
    2. | .41666994    .11992289 |
    3. | .41666994    .10192435 |
    4. | .41666994    .07946882 |
    5. | .41666994    .04549987 |
       |------------------------|
    6. | .41666994    .05055576 |
    7. | .41666994    .04673756 |
    8. | .41666994   -.01141121 |
    9. | .41666994   -.08817449 |
   10. | .41666994   -.11250592 |
       |------------------------|
   11. | .41666994   -.17160416 |
   12. | .41666994   -.17099379 |
       +------------------------+

.