Omitted (time) dummy variables

[Elisabeth Maes]

Hello all,

While running a panel regression I am encountering a collinearity problem, but do not understand where the collinearity is coming from. My dataset is sufficiently large (>6000 ids).

The problem pops up when I want to include time dummies; without time dummies there is no issue. For each year in my dataset I created a dummy. Due to the dummy trap one dummy is always omitted, but in my case in total 3 time dummies are omitted. I already checked whether there are observations which have a unit value for several year dummies or zeros for all years, which is clearly not the case. The observations are also quite evenly distributed over time.

I understand that perhaps one or more of the time dummies can be written as a linear combination of the other regressors. But how can that be in case of time dummies?

Many thanks in advance.

BR
Elisabeth

[Attaullah Shah]

First of all, Elisabeth please register yourself with full name. We can help if we you share the correlation matrix and definitions of various variables. Some macro level variables closely covary with year dummies, these might be hidden culprits.

Regards
Attaullah Shah

[Elisabeth Maes]

Hello Attaullah Shah,

Thanks for your quick reply.

I want to explain the long-term debt ratio. The regressors are firm characteristics (firm age (i.e. ln(age), size (ln(total assets), risk (st.dev of net earnings over the last 3 years), profitability (ebit/total assets), collateral (fixed assets/total assets) and growth (sales growth, measured relatively to previous year)); while the dependent 'LTschuld' represents LT debt scaled by total assets.
The Pearson correlation matrix looks as follows:
| LTschuld Leeftijd Ebit Groei Grootte Onderp~d Risico
-------------+---------------------------------------------------------------
LTschuld | 1.0000
|
|
Leeftijd | 0.0011 1.0000
| 0.8363
|
Ebit | -0.1020* -0.0277* 1.0000
| 0.0000 0.0000
|
Groei | 0.0339* -0.0796* 0.1924* 1.0000
| 0.0000 0.0000 0.0000
|
Grootte | 0.2175* 0.2017* -0.0046 0.0975* 1.0000
| 0.0000 0.0000 0.3870 0.0000
|
Onderpand | 0.5107* 0.0416* -0.0881* 0.0166* 0.1226* 1.0000
| 0.0000 0.0000 0.0000 0.0017 0.0000
|
Risico | -0.1126* -0.1236* -0.1397* -0.1177* -0.2949* -0.0939* 1.0000
| 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
|
So, collinearity between the continuous regressors is of no concern here.
Performing pooled OLS gives the following (where y* stands for the time dummies and panel covers the period 2004-2010):

regress LTschuld L.Leeftijd L.Ebit L.Groei L.Grootte L.Onderpand L.Risico y2005 ... y2010

LTschuld | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
Leeftijd |
L1. | -.0100332 .0012856 -7.80 0.000 -.012553 -.0075133
|
Ebit |
L1. | -.0940291 .0071702 -13.11 0.000 -.1080832 -.079975
|
Groei |
L1. | .0177338 .0047171 3.76 0.000 .0084879 .0269797
|
Grootte |
L1. | .02193 .0008137 26.95 0.000 .020335 .023525
|
Onderpand |
L1. | .3614957 .0046434 77.85 0.000 .3523943 .370597
|
Risico |
L1. | -.0809053 .0156659 -5.16 0.000 -.1116115 -.0501991
|
y2005 | 0 (omitted)
y2006 | 0 (omitted)
y2007 | .0006116 .0027443 0.22 0.824 -.0047675 .0059907
y2008 | .0025904 .0027309 0.95 0.343 -.0027623 .007943
y2009 | -.0006751 .0027178 -0.25 0.804 -.0060022 .004652
y2010 | -.0043153 .0027585 -1.56 0.118 -.0097221 .0010916
_cons | -.3017464 .0154462 -19.54 0.000 -.332022 -.2714707

[Attaullah Shah]

As a general advice on posting results, please following the instruction in FAQs. For example, when typing the message, click on the underlined A on the upper right, then click #. Copy and paste your code and output between the two code tags. Since the year dummies are dropped, you should include them in the correlation analysis and see the results. Second, you should be less concerned about the year dummies as none of these seems to be significant. Even if they are significant, journals usually do not publish coefficients of these dummies. So do not worry if they are dropped by Stata.

[Jeff Wooldridge]

Elisabeth: Here is what I think is happening. First, you are lagging variables by using the "L" operator, so the data for your equation really starts in 2005, not 2004. That is why you are losing the 2005 dummy. I'm unsure about why y2006 is being dropped, but it may have to do with the log(age) variable. If you were including firm fixed effects then this would be the problem: Once you account for different firm ages in the first year, age -- which always increases by one -- is perfectly collinear with the year dummies. But usually a variable like age does not drop out when you don't have firm fixed effects.

Try dropping y2005 to see what happens, and then maybe you can figure out why y2006 is dropped. JW

[Elisabeth Maes]

Thanks for your helpful advice!
I figured out why y2006 was dropped. One of my continuous regressors was missing for each panel for one particular year. Once I solved this, the model ran fine.

Elisabeth

[Roy Steinvoort]

Hi guys, I was wondering the same as Elisabeth above. If I run my panel data (2007-2013) regression with year and firm fixed effects 2 extra time dummies are dropped because of collinearity (see output below). I already dropped one time dummie myself to avoid dummy variable trap. Could one please have a look at why those years are also dropped from the regression output.

I have also included a correlation matrix for you to check whether something is odd.

If one needs explanation on the variables I am happy to provide upon request.

Thanks in advance,

Best,

Roy Steinvoort

 areg LnCEO_compensation  LnMarketcap BtM ROA ChangeInROA LagAnnualizedMonthlyReturn AgeCEO TenureCEO  NewCEO  
> Boardsize CEOisChair  PercOutside PercBusy PercOld PercApptdAfterCEO YEAR*, a(gvkey) cluster(gvkey)
note: YEAR7 omitted because of collinearity
note: YEAR2 omitted because of collinearity

Linear regression, absorbing indicators           Number of obs   =       6752
                                                  F(  18,   1638) =      19.98
                                                  Prob > F        =     0.0000
                                                  R-squared       =     0.8421
                                                  Adj R-squared   =     0.7908
                                                  Root MSE        =     0.4464

                                             (Std. Err. adjusted for 1639 clusters in gvkey)
--------------------------------------------------------------------------------------------
                           |               Robust
        LnCEO_compensation |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
---------------------------+----------------------------------------------------------------
               LnMarketcap |   .0306365   .0084857     3.61   0.000     .0139924    .0472805
                       BtM |  -.0028283   .0017949    -1.58   0.115    -.0063488    .0006921
                       ROA |   .7570996   .2868026     2.64   0.008     .1945611    1.319638
               ChangeInROA |   .8247792   .2488013     3.32   0.001      .336777    1.312781
LagAnnualizedMonthlyReturn |   .0826548   .0167448     4.94   0.000     .0498114    .1154982
                    AgeCEO |   .0030136   .0034373     0.88   0.381    -.0037284    .0097556
                 TenureCEO |   .0439201   .0189114     2.32   0.020     .0068271    .0810132
                    NewCEO |  -.0019386   .0282041    -0.07   0.945    -.0572584    .0533812
                 Boardsize |  -.0073345   .0103377    -0.71   0.478    -.0276111    .0129421
                CEOisChair |   .0534148   .0323341     1.65   0.099    -.0100058    .1168353
               PercOutside |   .1392485   .1696608     0.82   0.412    -.1935264    .4720235
                  PercBusy |   .2141589   .0981868     2.18   0.029     .0215739    .4067438
                   PercOld |    .126415   .0903798     1.40   0.162    -.0508572    .3036872
         PercApptdAfterCEO |   .2963825   .1399405     2.12   0.034     .0219013    .5708637
                     YEAR2 |          0  (omitted)
                     YEAR3 |  -.0676431   .0278097    -2.43   0.015    -.1221894   -.0130969
                     YEAR4 |   .0007112   .0392487     0.02   0.986    -.0762718    .0776941
                     YEAR5 |   .0247186   .0552832     0.45   0.655    -.0837146    .1331518
                     YEAR6 |     .00256   .0724646     0.04   0.972     -.139573     .144693
                     YEAR7 |          0  (omitted)
                     _cons |   7.289132    .275787    26.43   0.000       6.7482    7.830065
---------------------------+----------------------------------------------------------------
                     gvkey |   absorbed                                    (1639 categories)

  pwcorr LnCEO_compensation  LnMarketcap BtM ROA ChangeInROA LagAnnualizedMonthlyReturn AgeCEO TenureCEO  NewCEO
>   Boardsize CEOisChair  PercOutside PercBusy PercOld PercApptdAfterCEO YEAR*

             | LnCEO_~n LnMark~p      BtM      ROA Change~A LagAnn~n   AgeCEO
-------------+---------------------------------------------------------------
LnCEO_comp~n |   1.0000
 LnMarketcap |   0.5894   1.0000
         BtM |  -0.0519  -0.1029   1.0000
         ROA |   0.1028   0.1762  -0.1353   1.0000
 ChangeInROA |   0.0702   0.0604  -0.0279  -0.2806   1.0000
LagAnnuali~n |   0.0991   0.0972  -0.0860   0.0886   0.2078   1.0000
      AgeCEO |   0.0032   0.0036   0.0048  -0.0349   0.0020  -0.0155   1.0000
   TenureCEO |  -0.0577  -0.0496  -0.0274  -0.0137   0.0053   0.0085   0.2010
      NewCEO |   0.0009  -0.0047   0.0435  -0.0022  -0.0140  -0.0192   0.1019
   Boardsize |   0.3145   0.3855   0.1006  -0.0633   0.0206  -0.0356   0.0153
  CEOisChair |   0.1349   0.1106   0.0024   0.0240   0.0022  -0.0302   0.2038
 PercOutside |   0.2542   0.1712   0.0116  -0.0207   0.0175  -0.0158  -0.0712
    PercBusy |   0.3677   0.3415  -0.0280   0.0300   0.0067   0.0023  -0.0711
     PercOld |  -0.0121   0.0104  -0.0022  -0.0246   0.0105  -0.0085   0.3094
PercApptdA~O |  -0.0021  -0.0329  -0.0331  -0.0038   0.0078   0.0295   0.1191
       YEAR2 |  -0.0466  -0.1203   0.0685   0.0275  -0.0717  -0.0771  -0.0600
       YEAR3 |  -0.0738  -0.0509  -0.0055  -0.0088  -0.0971  -0.3762  -0.0509
       YEAR4 |   0.0077   0.0063  -0.0216  -0.0706   0.1599   0.2665  -0.0189
       YEAR5 |   0.0331  -0.0081  -0.0079   0.0201   0.0400   0.1177   0.0566
       YEAR6 |   0.0390   0.0719   0.0171   0.0272  -0.0283  -0.1280   0.0761
       YEAR7 |   0.0772   0.1073  -0.0364   0.0046  -0.0097   0.0429   0.0829

             | Tenure~O   NewCEO Boards~e CEOisC~r PercOu~e PercBusy  PercOld
-------------+---------------------------------------------------------------
   TenureCEO |   1.0000
      NewCEO |  -0.2020   1.0000
   Boardsize |  -0.1142   0.0230   1.0000
  CEOisChair |   0.1861   0.0323   0.0714   1.0000
 PercOutside |  -0.1307   0.0219   0.1518   0.1298   1.0000
    PercBusy |  -0.1667   0.0149   0.1855   0.0816   0.2875   1.0000
     PercOld |   0.1502  -0.0114   0.0608  -0.0139  -0.1275  -0.0626   1.0000
PercApptdA~O |   0.8260  -0.2310  -0.1025   0.1942  -0.0427  -0.1172   0.0173
       YEAR2 |  -0.1061  -0.0101   0.0050   0.0722  -0.0282   0.0371  -0.0757
       YEAR3 |  -0.0532  -0.0016  -0.0023   0.0656  -0.0356  -0.0002  -0.0500
       YEAR4 |   0.0012   0.0243  -0.0007   0.0566   0.0039  -0.0152  -0.0069
       YEAR5 |   0.0536   0.0411  -0.0047  -0.0457   0.0252  -0.0158   0.0413
       YEAR6 |   0.1050   0.0605   0.0020  -0.0537   0.0344  -0.0324   0.0838
       YEAR7 |   0.1568  -0.1132   0.0089  -0.0734   0.0491  -0.0196   0.1002

             | PercAp~O    YEAR2    YEAR3    YEAR4    YEAR5    YEAR6    YEAR7
-------------+---------------------------------------------------------------
PercApptdA~O |   1.0000
       YEAR2 |  -0.0879   1.0000
       YEAR3 |  -0.0516  -0.1663   1.0000
       YEAR4 |  -0.0190  -0.1670  -0.1674   1.0000
       YEAR5 |   0.0456  -0.1669  -0.1673  -0.1680   1.0000
       YEAR6 |   0.0872  -0.1673  -0.1677  -0.1684  -0.1683   1.0000
       YEAR7 |   0.1377  -0.1681  -0.1685  -0.1691  -0.1691  -0.1695   1.0000

[Carlo Lazzaro]

Roy (the usual, seemingly pedantic, remark about re-registering according to the strong preference for real full names still holds. Thanks).
- Stata automatically avoids the -dummy trap- if you go i.year in -long- format with a pooled OLS (just out of curiosity: why don't you simply rely upon the capabilities of the wide suite of -xt- built-in Stata commands explicitly conceived for dealing with panel datasets?);
- why did you invoke -pwcorr- instead of -estat vce, corr- after -areg- to investigate multicollinearity?

[Clyde Schechter]

I can't give a specific answer because I don't know what the variables in your model are. But most likely it is a variant of this:

. webuse grunfeld, clear

. xtset
       panel variable:  company (strongly balanced)
        time variable:  year, 1935 to 1954
                delta:  1 year

. xtreg mvalue invest i.year, fe

Fixed-effects (within) regression               Number of obs     =        200
Group variable: company                         Number of groups  =         10

R-sq:                                           Obs per group:
     within  = 0.5383                                         min =         20
     between = 0.8572                                         avg =       20.0
     overall = 0.6801                                         max =         20

                                                F(20,170)         =       9.91
corr(u_i, Xb)  = 0.6904                         Prob > F          =     0.0000

------------------------------------------------------------------------------
      mvalue |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      invest |   1.717841   .1969516     8.72   0.000     1.329056    2.106627
             |
        year |
       1936  |   322.5244   112.1003     2.88   0.005     101.2364    543.8124
       1937  |   559.3822   112.3838     4.98   0.000     337.5346    781.2298
       1938  |   131.4099   111.9601     1.17   0.242    -89.60131    352.4211
       1939  |   359.4182   111.9666     3.21   0.002     138.3942    580.4422
       1940  |   356.9408   112.2402     3.18   0.002     135.3767    578.5048
       1941  |    265.428   112.7305     2.35   0.020     42.89614    487.9599
       1942  |   87.62209    112.387     0.78   0.437    -134.2317    309.4759
       1943  |   210.3145   112.3071     1.87   0.063    -11.38151    432.0106
       1944  |   237.5371   112.3575     2.11   0.036     15.74143    459.3328
       1945  |   344.3146   112.4131     3.06   0.003     122.4092      566.22
       1946  |   344.2699   113.3083     3.04   0.003     120.5975    567.9423
       1947  |   92.03748   112.9107     0.82   0.416    -130.8502    314.9251
       1948  |   50.01684   113.0927     0.44   0.659      -173.23    273.2636
       1949  |   95.84998   112.7196     0.85   0.396    -126.6604    318.3603
       1950  |   135.0513   113.0137     1.19   0.234    -88.03958    358.1421
       1951  |   282.7271   114.7093     2.46   0.015     56.28914    509.1651
       1952  |   287.0239   115.8533     2.48   0.014      58.3275    515.7203
       1953  |   421.8682   118.8701     3.55   0.001     187.2166    656.5198
       1954  |   385.1248   118.7513     3.24   0.001     150.7077    619.5418
             |
       _cons |   582.5049   80.45101     7.24   0.000     423.6933    741.3165
-------------+----------------------------------------------------------------
     sigma_u |  1026.8384
     sigma_e |  250.34154
         rho |  .94389698   (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(9, 170) = 128.45                    Prob > F = 0.0000

. gen byte early = (year < 1944)

. xtreg mvalue invest i.early i.year, fe
note: 1954.year omitted because of collinearity

Fixed-effects (within) regression               Number of obs     =        200
Group variable: company                         Number of groups  =         10

R-sq:                                           Obs per group:
     within  = 0.5383                                         min =         20
     between = 0.8572                                         avg =       20.0
     overall = 0.6801                                         max =         20

                                                F(20,170)         =       9.91
corr(u_i, Xb)  = 0.6904                         Prob > F          =     0.0000

------------------------------------------------------------------------------
      mvalue |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      invest |   1.717841   .1969516     8.72   0.000     1.329056    2.106627
     1.early |  -385.1248   118.7513    -3.24   0.001    -619.5418   -150.7077
             |
        year |
       1936  |   322.5244   112.1003     2.88   0.005     101.2364    543.8124
       1937  |   559.3822   112.3838     4.98   0.000     337.5346    781.2298
       1938  |   131.4099   111.9601     1.17   0.242    -89.60131    352.4211
       1939  |   359.4182   111.9666     3.21   0.002     138.3942    580.4422
       1940  |   356.9408   112.2402     3.18   0.002     135.3767    578.5048
       1941  |    265.428   112.7305     2.35   0.020     42.89614    487.9599
       1942  |   87.62209    112.387     0.78   0.437    -134.2317    309.4759
       1943  |   210.3145   112.3071     1.87   0.063    -11.38151    432.0106
       1944  |  -147.5877   115.9332    -1.27   0.205    -376.4417    81.26637
       1945  |  -40.81016   115.7694    -0.35   0.725    -269.3409    187.7206
       1946  |  -40.85485   114.1246    -0.36   0.721    -266.1388    184.4291
       1947  |  -293.0873   114.7011    -2.56   0.011    -519.5091   -66.66544
       1948  |  -335.1079   114.4168    -2.93   0.004    -560.9685   -109.2473
       1949  |  -289.2748    115.049    -2.51   0.013    -516.3835   -62.16604
       1950  |  -250.0735   114.5354    -2.18   0.030    -476.1683   -23.97868
       1951  |  -102.3976   112.9058    -0.91   0.366    -325.2756    120.4804
       1952  |  -98.10087   112.3841    -0.87   0.384    -319.9489    123.7471
       1953  |   36.74347   111.9567     0.33   0.743    -184.2609    257.7479
       1954  |          0  (omitted)
             |
       _cons |   967.6297   95.78427    10.10   0.000     778.5499    1156.709
-------------+----------------------------------------------------------------
     sigma_u |  1026.8384
     sigma_e |  250.34154
         rho |  .94389698   (fraction of variance due to u_i)
------------------------------------------------------------------------------
F test that all u_i=0: F(9, 170) = 128.45                    Prob > F = 0.0000

Notice that when we add a new variable that is defined in terms of the year variable, we lose one year to multicollinearity. My guess is that two of your other variables work in a way similar to this, and that is the cause.

By the way, you will make your life easier going forward if you use factor variable notation (-help fvvarlist-), as I have done, instead of creating your own indicator variables. Not only will you need to do less typing to set up your models, you will then be able to use the versatile -margins- commands to get predicted values and marginal effects after.

Finally, please click on "CONTACT US" in the lower right corner of your screen and request the system administrators to change your user id on this site to Roy Steinvoort. In this community it is the norm to use real names as user ids. Thank you.

[Roy Steinvoort]

Carlo and Clyde, thank you for your responses. Unfortunately I am not much wiser now.

To maybe give you a clearer idea of what is going on I have tried the tips you provided. I think it might have something to do with the specification of the variable "CEO tenure".

. tab fyear

      fyear |      Freq.     Percent        Cum.
------------+-----------------------------------
       2007 |      1,226       12.69       12.69
       2008 |      1,356       14.04       26.73
       2009 |      1,415       14.65       41.39
       2010 |      1,437       14.88       56.26
       2011 |      1,438       14.89       71.15
       2012 |      1,416       14.66       85.81
       2013 |      1,370       14.19      100.00
------------+-----------------------------------
      Total |      9,658      100.00

. xtset gvkey fyear, year
       panel variable:  gvkey (unbalanced)
        time variable:  fyear, 2007 to 2013, but with gaps
                delta:  1 year

. xtreg LnCEO_compensation  LnMarketcap BtM ROA ChangeInROA LagAnnualizedMonthlyReturn AgeCEO TenureCEO  NewCEO
>  Boardsize CEOisChair  PercOutside PercBusy PercOld PercApptdAfterCEO i.fyear, fe cluster(gvkey)
note: 2013.fyear omitted because of collinearity

Fixed-effects (within) regression               Number of obs      =      6669
Group variable: gvkey                           Number of groups   =      1558

R-sq:  within  = 0.1039                         Obs per group: min =         1
       between = 0.0031                                        avg =       4.3
       overall = 0.0078                                        max =         6

                                                F(18,1557)         =     26.44
corr(u_i, Xb)  = -0.3723                        Prob > F           =    0.0000

                                             (Std. Err. adjusted for 1558 clusters in gvkey)
--------------------------------------------------------------------------------------------
                           |               Robust
        LnCEO_compensation |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
---------------------------+----------------------------------------------------------------
               LnMarketcap |    .030474   .0073589     4.14   0.000     .0160397    .0449084
                       BtM |  -.0028329   .0015627    -1.81   0.070    -.0058981    .0002323
                       ROA |   .7570686   .2496137     3.03   0.002     .2674542    1.246683
               ChangeInROA |    .824451   .2160089     3.82   0.000      .400752     1.24815
LagAnnualizedMonthlyReturn |    .083084   .0146168     5.68   0.000     .0544133    .1117547
                    AgeCEO |   .0030077   .0029908     1.01   0.315    -.0028587     .008874
                 TenureCEO |   .0437457   .0164498     2.66   0.008     .0114796    .0760117
                    NewCEO |  -.0012613   .0245529    -0.05   0.959    -.0494216    .0468989
                 Boardsize |  -.0074259    .008995    -0.83   0.409    -.0250696    .0102177
                CEOisChair |   .0533571   .0281486     1.90   0.058    -.0018561    .1085703
               PercOutside |   .1401098   .1475574     0.95   0.342    -.1493223    .4295419
                  PercBusy |   .2158994   .0854419     2.53   0.012     .0483061    .3834927
                   PercOld |   .1261479   .0786555     1.60   0.109     -.028134    .2804298
         PercApptdAfterCEO |   .2980932   .1218312     2.45   0.015     .0591226    .5370638
                           |
                     fyear |
                     2009  |  -.0680739   .0241446    -2.82   0.005    -.1154332   -.0207146
                     2010  |   .0001542   .0341648     0.00   0.996    -.0668596     .067168
                     2011  |   .0254965   .0480934     0.53   0.596    -.0688381    .1198312
                     2012  |   .0021286   .0630334     0.03   0.973    -.1215106    .1257678
                     2013  |          0  (omitted)
                           |
                     _cons |   7.294346   .2400827    30.38   0.000     6.823426    7.765265
---------------------------+----------------------------------------------------------------
                   sigma_u |   .9705203
                   sigma_e |  .44649751
                       rho |  .82531733   (fraction of variance due to u_i)
--------------------------------------------------------------------------------------------

. estat vce, corr

Correlation matrix of coefficients of xtreg model

             |                                                                                          
        e(V) | LnMark~p       BtM       ROA  Change~A  LagAnn~n    AgeCEO  Tenure~O    NewCEO  Boards~e
-------------+------------------------------------------------------------------------------------------
 LnMarketcap |   1.0000                                                                                 
         BtM |   0.1675    1.0000                                                                       
         ROA |  -0.0393   -0.0090    1.0000                                                             
 ChangeInROA |  -0.0663    0.0502    0.8017    1.0000                                                   
LagAnnuali~n |  -0.0754   -0.0627   -0.2128   -0.2493    1.0000                                         
      AgeCEO |   0.0656    0.0415    0.0059   -0.0208   -0.0482    1.0000                               
   TenureCEO |  -0.0003    0.0491    0.0278    0.0535   -0.0130   -0.0315    1.0000                     
      NewCEO |   0.0607   -0.0787   -0.0264   -0.0055    0.0584   -0.0616    0.0087    1.0000           
   Boardsize |   0.0385   -0.1871    0.0421    0.0367    0.0221    0.0169    0.1233    0.1452    1.0000
  CEOisChair |   0.0739    0.1353    0.0017    0.0454   -0.0920    0.1109    0.0991   -0.0373    0.0928
 PercOutside |   0.0527   -0.0317   -0.0140    0.0086   -0.0222    0.0703   -0.0089    0.0406    0.0131
    PercBusy |   0.1004    0.0928   -0.1051   -0.0959    0.1031    0.0478    0.1058    0.0301    0.0962
     PercOld |   0.0062    0.0021    0.0667    0.0073   -0.0196    0.0674   -0.1271    0.0224   -0.1894
PercApptdA~O |  -0.1228   -0.2050   -0.0567   -0.0623    0.1251   -0.3415   -0.2260    0.0353   -0.2014
  2009.fyear |  -0.0247   -0.0140    0.0937    0.0901    0.2834   -0.0202   -0.5175    0.0602    0.0839
  2010.fyear |  -0.0699   -0.0399    0.0963    0.0344   -0.0672   -0.0467   -0.8085   -0.0512   -0.0205
  2011.fyear |  -0.0374   -0.0264    0.0124   -0.0149   -0.0395   -0.0772   -0.8836   -0.0494   -0.0233
  2012.fyear |  -0.0664   -0.0321    0.0045   -0.0095    0.0332   -0.0874   -0.9024   -0.0390   -0.0389
 2013o.fyear |        .         .         .         .         .         .         .         .         .
       _cons |  -0.3148   -0.0076   -0.1021   -0.0719    0.0550   -0.7077   -0.2738   -0.0525   -0.4038

             |                                                       2009.     2010.     2011.     2012.
        e(V) | CEOisC~r  PercOu~e  PercBusy   PercOld  PercAp~O     fyear     fyear     fyear     fyear
-------------+------------------------------------------------------------------------------------------
  CEOisChair |   1.0000                                                                                 
 PercOutside |   0.0355    1.0000                                                                       
    PercBusy |  -0.0212   -0.1899    1.0000                                                             
     PercOld |   0.0301   -0.0617   -0.0227    1.0000                                                   
PercApptdA~O |  -0.2309   -0.2215   -0.1311    0.2353    1.0000                                         
  2009.fyear |  -0.0207    0.0096    0.0131    0.0053    0.0358    1.0000                               
  2010.fyear |  -0.0895   -0.0739   -0.0261   -0.0239    0.0764    0.6893    1.0000                     
  2011.fyear |  -0.0489   -0.0414   -0.0419   -0.0547    0.0476    0.6232    0.9107    1.0000           
  2012.fyear |  -0.0439   -0.0175   -0.0386   -0.0530    0.0450    0.6387    0.9015    0.9591    1.0000
 2013o.fyear |        .         .         .         .         .         .         .         .         .
       _cons |  -0.2182   -0.4941   -0.0878   -0.0085    0.4177    0.0757    0.2763    0.3138    0.3288

             |    2013o.          
        e(V) |    fyear     _cons
-------------+--------------------
 2013o.fyear |        .           
       _cons |        .    1.0000

Here I have omitted the CEO tenure variable and now the regression keeps 2013 in the model. However, in the model above and below the years 2007 and 2008 are also omitted and I have no clue why (one obviously to prevend omitted variable bias)...

xtreg LnCEO_compensation  LnMarketcap BtM ROA ChangeInROA LagAnnualizedMonthlyReturn AgeCEO NewCEO  Boardsize
> CEOisChair  PercOutside PercBusy PercOld PercApptdAfterCEO i.fyear, fe cluster(gvkey)

Fixed-effects (within) regression               Number of obs      =      6704
Group variable: gvkey                           Number of groups   =      1567

R-sq:  within  = 0.1044                         Obs per group: min =         1
       between = 0.1261                                        avg =       4.3
       overall = 0.1192                                        max =         6

                                                F(18,1566)         =     26.66
corr(u_i, Xb)  = 0.1599                         Prob > F           =    0.0000

                                             (Std. Err. adjusted for 1567 clusters in gvkey)
--------------------------------------------------------------------------------------------
                           |               Robust
        LnCEO_compensation |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
---------------------------+----------------------------------------------------------------
               LnMarketcap |   .0306472   .0073632     4.16   0.000     .0162045    .0450899
                       BtM |  -.0028289   .0015614    -1.81   0.070    -.0058916    .0002337
                       ROA |   .7498007   .2494652     3.01   0.003     .2604796    1.239122
               ChangeInROA |   .8311876   .2169715     3.83   0.000     .4056023    1.256773
LagAnnualizedMonthlyReturn |   .0827663   .0145514     5.69   0.000      .054224    .1113086
                    AgeCEO |   .0030093   .0029904     1.01   0.314    -.0028563    .0088749
                    NewCEO |  -.0011906   .0245491    -0.05   0.961    -.0493432    .0469619
                 Boardsize |  -.0076066   .0089576    -0.85   0.396    -.0251767    .0099635
                CEOisChair |   .0532522   .0281416     1.89   0.059    -.0019471    .1084514
               PercOutside |   .1349533   .1474511     0.92   0.360    -.1542691    .4241757
                  PercBusy |   .2181248   .0849761     2.57   0.010     .0514459    .3848036
                   PercOld |   .1249191   .0781931     1.60   0.110    -.0284551    .2782934
         PercApptdAfterCEO |   .2993863   .1213933     2.47   0.014     .0612758    .5374968
                           |
                     fyear |
                     2009  |  -.0249349   .0209489    -1.19   0.234    -.0660258     .016156
                     2010  |   .0884912   .0206666     4.28   0.000      .047954    .1290283
                     2011  |   .1569294   .0233754     6.71   0.000     .1110791    .2027797
                     2012  |   .1765662   .0283534     6.23   0.000     .1209517    .2321808
                     2013  |   .2186899   .0821929     2.66   0.008     .0574703    .3799096
                           |
                     _cons |   7.458374   .2308479    32.31   0.000     7.005571    7.911178
---------------------------+----------------------------------------------------------------
                   sigma_u |  .85316989
                   sigma_e |  .44585107
                       rho |  .78548934   (fraction of variance due to u_i)
--------------------------------------------------------------------------------------------

Finally, I have just contacted the administrators to change my username.

Best,

Roy

[Clyde Schechter]

Well, 2007 is omitted because it is the base year for your indicator variables. The mystery is why 2008 is also omitted. It is not due to collinearity because Stata would have given you a message about that right below your command line. (Notice how it did that in your original model.) The most likely explanation (actually, the only one I can think of) is that after taking into account that observations containing missing values for any variable in the model are eliminated from estimation, there just aren't any observations left with year = 2008 (even though you started out with plenty of them.) So I would run:

tab year if e(sample)

to see if 2008 appears among the years in the estimation sample. I imagine it will not. Then you have to figure out why that is. It may be just that the haphazard scattering of missing data among your variables coincidentally left you with nothing usable in 2008. More likely, there may be some variable in your model which is all missing values in the year 2008, perhaps some construct that was not measured in that year, or was omitted from the 2008 data for some reason.

[Carlo Lazzaro]

Roy:
Thanks for re-registering.
The substantive issue is the very high correlation between the variables -CEO tenure- and -years-. Hence, you should decide which one to keep in the model.

[Roy Steinvoort]

Clyde, if I use the commands as you specified above I get the following:

. tab fyear

      fyear |      Freq.     Percent        Cum.
------------+-----------------------------------
       2007 |      1,226       12.69       12.69
       2008 |      1,356       14.04       26.73
       2009 |      1,415       14.65       41.39
       2010 |      1,437       14.88       56.26
       2011 |      1,438       14.89       71.15
       2012 |      1,416       14.66       85.81
       2013 |      1,370       14.19      100.00
------------+-----------------------------------
      Total |      9,658      100.00

. tab fyear if e(sample)

      fyear |      Freq.     Percent        Cum.
------------+-----------------------------------
       2008 |      1,307       19.60       19.60
       2009 |      1,337       20.05       39.65
       2010 |      1,333       19.99       59.63
       2011 |      1,343       20.14       79.77
       2012 |      1,296       19.43       99.21
       2013 |         53        0.79      100.00
------------+-----------------------------------
      Total |      6,669      100.00

I think it is strange that so little observations from 2013 are left in the sample and that 2008 is not even presented to us in the regression output while there are 1,307 observations in the sample from a possible 1,356...

I hope you have any more ideas regarding this.

Tomorrow, I will have another look on the specification of my "CEO tenure" variable since it is indeed very highly correlated with the time variable in the data. Furthermore, I will have another look at the variable "Change in ROA" since there might be a problem. I will let you know tomorrow when I had the chance to look at it.

Best,

Roy

[Clyde Schechter]

This seems very bizarre. I can't comment on the small number of observations with 2013 in the estimation sample: I can only assume that it's some peculiarity related to missing values in your data and leave it to you to investigate that. But the failure of 2008 to appear in the output is a mystery. The only other thing I can think of is if that variable is somehow set in such a way as to cause that. What does Stata say about fyear if you run -fvset report-?

Added: one more thing that's just a general standby--is your Stata up to date? Try -update query- to see if there may be an update since your last.

[Roy Steinvoort]

Thanks for your time and effort.

I have managed to fix the data and now it appears like the following:

xtreg  LnCEO_compensation  LnMarketcap BtM ROA ChangeInROA LagAnnualizedMonthlyReturn AgeCEO TenureCEO NewCEO
>  CEOisChair Boardsize PercOutside PercBusy PercOld PercApptdAfterCEO i.fyear, fe cluster(gvkey)

Fixed-effects (within) regression               Number of obs      =      9176
Group variable: gvkey                           Number of groups   =      1564

R-sq:  within  = 0.1186                         Obs per group: min =         1
       between = 0.0150                                        avg =       5.9
       overall = 0.0274                                        max =         7

                                                F(20,1563)         =     39.16
corr(u_i, Xb)  = -0.2153                        Prob > F           =    0.0000

                                             (Std. Err. adjusted for 1564 clusters in gvkey)
--------------------------------------------------------------------------------------------
                           |               Robust
        LnCEO_compensation |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
---------------------------+----------------------------------------------------------------
               LnMarketcap |   .0394152    .006716     5.87   0.000      .026242    .0525884
                       BtM |  -.0104713   .0033253    -3.15   0.002    -.0169939   -.0039487
                       ROA |   .8594852   .1666709     5.16   0.000     .5325631    1.186407
               ChangeInROA |   .6459091   .2180657     2.96   0.003     .2181769    1.073641
LagAnnualizedMonthlyReturn |   .0994488   .0145736     6.82   0.000      .070863    .1280347
                    AgeCEO |   .0026646   .0023262     1.15   0.252    -.0018982    .0072275
                 TenureCEO |   .0393694   .0123448     3.19   0.001     .0151552    .0635836
                    NewCEO |   .0401013   .0253753     1.58   0.114     -.009672    .0898746
                CEOisChair |   .0316066   .0228494     1.38   0.167    -.0132121    .0764254
                 Boardsize |   .0015619   .0075025     0.21   0.835    -.0131541    .0162779
               PercOutside |   .2239039   .1183024     1.89   0.059    -.0081441     .455952
                  PercBusy |   .1380394   .0660623     2.09   0.037     .0084594    .2676195
                   PercOld |     .04945   .0641041     0.77   0.441     -.076289    .1751891
         PercApptdAfterCEO |    .069633   .0966974     0.72   0.472    -.1200373    .2593032
                           |
                     fyear |
                     2008  |  -.0028282   .0189923    -0.15   0.882    -.0400813    .0344248
                     2009  |  -.0497692   .0258332    -1.93   0.054    -.1004405    .0009021
                     2010  |   .0298012   .0296231     1.01   0.315     -.028304    .0879065
                     2011  |   .0769887   .0374103     2.06   0.040     .0036091    .1503683
                     2012  |   .0694241   .0485345     1.43   0.153    -.0257755    .1646237
                     2013  |   .0950013   .0579988     1.64   0.102    -.0187625     .208765
                           |
                     _cons |   7.179339   .1870619    38.38   0.000     6.812421    7.546258
---------------------------+----------------------------------------------------------------
                   sigma_u |  .92318895
                   sigma_e |  .45285629
                       rho |  .80604575   (fraction of variance due to u_i)
--------------------------------------------------------------------------------------------


 tab fyear if e(sample)

      fyear |      Freq.     Percent        Cum.
------------+-----------------------------------
       2007 |      1,195       13.02       13.02
       2008 |      1,322       14.41       27.43
       2009 |      1,346       14.67       42.10
       2010 |      1,344       14.65       56.75
       2011 |      1,363       14.85       71.60
       2012 |      1,322       14.41       86.01
       2013 |      1,284       13.99      100.00
------------+-----------------------------------
      Total |      9,176      100.00

The only thing that is left for me is to figure out why "CEO tenure" and "New CEO" are highly correlated with the variable "fyear". I have to code them in a different way I guess.

At least, the problem of omitted year dummies is resolved.

Best,

Roy