Between Effects regression on panel data - Do I need dummies for years?

[Zach Rodgers]

Hi All,

Thanks in advance for your help!

I cannot find definitive information on "year dummies" for between-effects regression.

I am analyzing organizations in an industry where the organizational-level characteristic of interest does not change over time. Hence, I am using between-effects regression. Consequently, I understand that between-effects regression is also known as "time fixed-effects" regression (in contrast to what we normally call "fixed effects" regression, which is really "entity fixed-effects" regression), which controls for omitted variables that vary across time but not within entities. (Stock & Watson, 2003; Alison, 2009)

So, I think that means that in a between effects regression (xtreg ...., be), I do *not* need to add year dummies, that by setting the xtreg to cluster by year and firm it is already adjusting for anything associated with individual years. Is that true? Do I need to add year dummies for a BE regression? Is there a text that I can refer to in order to justify including or excluding year dummies?

Thanks!
Very Best,
Zach

[Carlo Lazzaro]

Zach:
there's no mandatory recommendation about including -i.time- among -xtreg- predictors.
However, you may be interested in including -tijmevar- as a predictor if you think that time can contribute to explain variation (across panels, in your case) of the dependen variable as time goes by, as in the following toy-example, where the joint statistical significance of -i.year- is tested via -testparm- after -xtreg, be-:

. use "http://www.stata-press.com/data/r15/nlswork.dta"
(National Longitudinal Survey.  Young Women 14-26 years of age in 1968)

. xtreg ln_wage age i.race i.year, be

Between regression (regression on group means)  Number of obs     =     28,510
Group variable: idcode                          Number of groups  =      4,710

R-sq:                                           Obs per group:
     within  = 0.0640                                         min =          1
     between = 0.1203                                         avg =        6.1
     overall = 0.0749                                         max =         15

                                                F(17,4692)        =      37.73
sd(u_i + avg(e_i.))=  .3989205                  Prob > F          =     0.0000

------------------------------------------------------------------------------
     ln_wage |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         age |   .0143467   .0020342     7.05   0.000     .0103587    .0183347
             |
        race |
      black  |  -.1187728   .0130181    -9.12   0.000    -.1442945   -.0932512
      other  |   .1007168   .0533526     1.89   0.059    -.0038792    .2053129
             |
        year |
         69  |   .0826589   .0661738     1.25   0.212    -.0470728    .2123906
         70  |  -.0149261   .0614792    -0.24   0.808    -.1354542    .1056021
         71  |  -.0205426   .0550792    -0.37   0.709    -.1285237    .0874386
         72  |   .0718167   .0619946     1.16   0.247    -.0497219    .1933552
         73  |   .0546877   .0605577     0.90   0.367    -.0640338    .1734092
         75  |  -.1084746   .0591837    -1.83   0.067    -.2245025    .0075533
         77  |   .1230523   .0612431     2.01   0.045     .0029869    .2431176
         78  |   .3572142    .068961     5.18   0.000     .2220183    .4924102
         80  |   .2508506   .0731485     3.43   0.001     .1074451    .3942561
         82  |  -.0740984   .0708411    -1.05   0.296    -.2129802    .0647833
         83  |   .1933064    .067161     2.88   0.004     .0616393    .3249734
         85  |   .2859533   .0625017     4.58   0.000     .1634207     .408486
         87  |   .2261934    .064248     3.52   0.000      .100237    .3521497
         88  |    .116136   .0559644     2.08   0.038     .0064194    .2258526
             |
       _cons |   1.166478   .0545962    21.37   0.000     1.059444    1.273513
------------------------------------------------------------------------------

. testparm i.year

 ( 1)  69.year = 0
 ( 2)  70.year = 0
 ( 3)  71.year = 0
 ( 4)  72.year = 0
 ( 5)  73.year = 0
 ( 6)  75.year = 0
 ( 7)  77.year = 0
 ( 8)  78.year = 0
 ( 9)  80.year = 0
 (10)  82.year = 0
 (11)  83.year = 0
 (12)  85.year = 0
 (13)  87.year = 0
 (14)  88.year = 0

       F( 14,  4692) =    6.19
            Prob > F =    0.0000

As an aside, please note that -be- specification is rarely used (in favour or -re-) when the researcher's intererest is in comparing panels across time.

[Zach Rodgers]

Thanks for your help!!