[Paris Rira]

Prof Ford, They are control variables.

. reghdfe ln_labor_productivity_w c.immi_sh#i.sector logsize lavg_firm_age lage, a( year id) cl(id)
(dropped 50256 singleton observations)
(MWFE estimator converged in 7 iterations)

HDFE Linear regression                            Number of obs   =  1,464,334
Absorbing 2 HDFE groups                           F(  11, 237899) =     189.63
Statistics robust to heteroskedasticity           Prob > F        =     0.0000
                                                  R-squared       =     0.7017
                                                  Adj R-squared   =     0.6438
                                                  Within R-sq.    =     0.0039
Number of clusters (id)      =    237,900         Root MSE        =     0.5282

                                   (Std. Err. adjusted for 237,900 clusters in id)
----------------------------------------------------------------------------------
                 |               Robust
ln_labor_produ~w |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-----------------+----------------------------------------------------------------
sector#c.immi_sh |
              3  |   .0024782   .0353288     0.07   0.944    -.0667652    .0717217
              6  |  -.0420726   .0253802    -1.66   0.097    -.0918171    .0076719
              7  |  -.0431337   .0218272    -1.98   0.048    -.0859144    -.000353
              9  |   .0524663   .0204169     2.57   0.010     .0124497    .0924829
             10  |  -.0788029   .0826174    -0.95   0.340    -.2407308     .083125
             11  |   .0017283   .0531121     0.03   0.974      -.10237    .1058267
             12  |  -.0441454   .0353662    -1.25   0.212    -.1134622    .0251714
             13  |  -.0171878   .0400251    -0.43   0.668    -.0956359    .0612603
                 |
         logsize |  -.0760921   .0023878   -31.87   0.000    -.0807722    -.071412
   lavg_firm_age |     .09495   .0026204    36.23   0.000      .089814     .100086
            lage |  -.0155982   .0089625    -1.74   0.082    -.0331645    .0019681
           _cons |   9.569129   .0340675   280.89   0.000     9.502357      9.6359
----------------------------------------------------------------------------------

Absorbed degrees of freedom:
-----------------------------------------------------+
 Absorbed FE | Categories  - Redundant  = Num. Coefs |
-------------+---------------------------------------|
        year |        10           1           9     |
          id |    237900      237900           0    *|
-----------------------------------------------------+
* = FE nested within cluster; treated as redundant for DoF computatio

[George Ford]

For 6, 7, and 9 you've got something. It appears that the statistically insignificant coefficient is, in part, due to sectors offsetting each other.

Do the affected sectors make sense? What sector is 9?

[Paris Rira]

3 =Manufacturing
6 =Construction
7= Trade
9 =Accomadation
10 =Real estate
and services

[George Ford]

So productivity is higher in accommodation, but lower in construction and trade, and not much evidence of an effect in other sectors.

Add region to absorb, then add sector (it may fall out due to firm FE; not sure).

[Paris Rira]

. reghdfe ln_labor_productivity_w c.immi_sh#i.sector logsize lavg_firm_age lage, a(  region year id) cl(id)
(dropped 50256 singleton observations)
(MWFE estimator converged in 19 iterations)

HDFE Linear regression                            Number of obs   =  1,464,334
Absorbing 3 HDFE groups                           F(  11, 237899) =     189.52
Statistics robust to heteroskedasticity           Prob > F        =     0.0000
                                                  R-squared       =     0.7017
                                                  Adj R-squared   =     0.6438
                                                  Within R-sq.    =     0.0039
Number of clusters (id)      =    237,900         Root MSE        =     0.5282

                                   (Std. Err. adjusted for 237,900 clusters in id)
----------------------------------------------------------------------------------
                 |               Robust
ln_labor_produ~w |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-----------------+----------------------------------------------------------------
sector#c.immi_sh |
              3  |   .0024444   .0353263     0.07   0.945    -.0667942    .0716829
              6  |  -.0423324   .0253764    -1.67   0.095    -.0920695    .0074047
              7  |  -.0431497   .0218201    -1.98   0.048    -.0859167   -.0003828
              9  |   .0524784   .0204167     2.57   0.010     .0124621    .0924947
             10  |  -.0790082   .0825961    -0.96   0.339    -.2408944     .082878
             11  |   .0014724   .0531149     0.03   0.978    -.1026314    .1055763
             12  |  -.0438945   .0353711    -1.24   0.215    -.1132209    .0254319
             13  |  -.0171264   .0400312    -0.43   0.669    -.0955865    .0613338
                 |
         logsize |  -.0760605   .0023872   -31.86   0.000    -.0807394   -.0713816
   lavg_firm_age |   .0948893   .0026202    36.21   0.000     .0897537    .1000249
            lage |  -.0156925   .0089619    -1.75   0.080    -.0332576    .0018727
           _cons |   9.569573   .0340647   280.92   0.000     9.502807    9.636339
----------------------------------------------------------------------------------

Absorbed degrees of freedom:
-----------------------------------------------------+
 Absorbed FE | Categories  - Redundant  = Num. Coefs |
-------------+---------------------------------------|
      region |         8           1           7     |
        year |        10           1           9     |
          id |    237900      237900           0    *|
-----------------------------------------------------+
* = FE nested within cluster; treated as redundant for DoF computation

. reghdfe ln_labor_productivity_w c.immi_sh#i.sector logsize lavg_firm_age lage, a(sector region year id) cl(id)
(dropped 50256 singleton observations)
(MWFE estimator converged in 28 iterations)

HDFE Linear regression                            Number of obs   =  1,464,334
Absorbing 4 HDFE groups                           F(  11, 237899) =     190.40
Statistics robust to heteroskedasticity           Prob > F        =     0.0000
                                                  R-squared       =     0.7019
                                                  Adj R-squared   =     0.6441
                                                  Within R-sq.    =     0.0039
Number of clusters (id)      =    237,900         Root MSE        =     0.5279

                                   (Std. Err. adjusted for 237,900 clusters in id)
----------------------------------------------------------------------------------
                 |               Robust
ln_labor_produ~w |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-----------------+----------------------------------------------------------------
sector#c.immi_sh |
              3  |   .0225389   .0351905     0.64   0.522    -.0464335    .0915113
              6  |  -.0715308   .0251798    -2.84   0.005    -.1208825   -.0221791
              7  |  -.0455842   .0217838    -2.09   0.036    -.0882799   -.0028885
              9  |    .059597   .0204167     2.92   0.004     .0195809    .0996132
             10  |  -.0606581   .0816049    -0.74   0.457    -.2206017    .0992854
             11  |   .0141336   .0533489     0.26   0.791    -.0904288    .1186959
             12  |  -.0416892   .0350635    -1.19   0.234    -.1104127    .0270343
             13  |  -.0088793    .040106    -0.22   0.825    -.0874861    .0697275
                 |
         logsize |    -.07599   .0023827   -31.89   0.000      -.08066     -.07132
   lavg_firm_age |    .094846   .0026189    36.22   0.000      .089713    .0999791
            lage |  -.0163256   .0089591    -1.82   0.068    -.0338853    .0012341
           _cons |   9.571929   .0340518   281.10   0.000     9.505189     9.63867
----------------------------------------------------------------------------------

Absorbed degrees of freedom:
-----------------------------------------------------+
 Absorbed FE | Categories  - Redundant  = Num. Coefs |
-------------+---------------------------------------|
      sector |         8           1           7     |
      region |         8           1           7     |
        year |        10           1           9    ?|
          id |    237900      237900           0    *|
-----------------------------------------------------+
? = number of redundant parameters may be higher
* = FE nested within cluster; treated as redundant for DoF computation

[George Ford]

This looks fairly stable over FE specifications.

Is that all the X vars you want to use?

[Paris Rira]

Is that all the X vars you want to use?

Yes Prof, they are part of literature. I should hold their effect in order to evaluate the main effect of the main explanatory var.

[George Ford]

Ok. The last model presented looks good. 3 immi_sh coefficients are stat sig.

There's heterogeneity in the effect across sectors. The 0 effect is a consequence of treating it as a homogeneous effect.

Why immi_sh reduces productivity in some sectors and increases it in others is something you'll have to explain.

question: immi_sh is firm specific?

[Paris Rira]

question: immi_sh is firm specific?

It denotes share of immigrants in Firm i time t.

[George Ford]

Looking up for ya, Paris, except for the need to explain the negative/positive results. Do you have a capital/labor ratio by firm, or something that might proxy for it?

[Paris Rira]

In the Firm level only there are: firm age, firm size ( its based on the employment)

[Paris Rira]

Looking up for ya, Paris, except for the need to explain the negative/positive results.

highly educated immigrants are mostly present in these sectors, also according to other regressions, highly skilled have a postive impact on productivity, while low skilled negative

[George Ford]

how is labor productivity measured?

[Paris Rira]

LP at firm i year t: is defined as gross value added (GVA) per worker, calculated as the difference between gross output and material inputs.

[Paris Rira]

Dear Professor @Jeff Wooldridge
It is a great honor for me to participate in the summer school at Braga/Minho University in Portugal from July 15th to 18th. I hope there will be an opportunity to discuss regression with high levels of mixed effects during this event.