Independent categorical variable in mutliple linear regression

[Resa Pfu]

Dear all,

For my master thesis I am building a linear regression model with a continous dependent variable (satisfaction_score) and three independent variables (age/continous, country/categorical, device/categorical). You can find my code below. Country has 4 categories Cambodia (1), Ethiopia (2), South Sudan (3), Uganda (4) and device has five categories Contec (1), Devon (2), Lifebox (3), Utech (4), Masimo (5). I have the problem, that i cannot get rid of a baseline level for the categorical variables so they are always compared to Cambodia and Contec. I have tried the bn command, but with this the last categories Uganda (4) and Masimo (5) are automatically omitted. Is there another way to do it and why are the last categories omitted with ibn.?

This is my code

regress Mean_Score ibn. country ibn. device hwage

Thanks in advance,

Theresa

[Maarten Buis]

Regression coefficients are by their very nature comparisons*, so when you look at a categorical variable, then you have to decide with whom you want to compare. So a reference category is inevitable. There are ways around that, but you should not do that unless you really know what you are doing, as it changes the meaning of the coefficients. You could look at help contrast to get other comparisons.

* as always exceptions exist

[Carlo Lazzaro]

Resa,
an as aside to Maarten's helpful advice, you can use -lincom- to get what you're presumably after:

. use "https://www.stata-press.com/data/r17/auto.dta"
(1978 automobile data)

. regress price i.foreign i.rep78 if rep78>=4

      Source |       SS           df       MS      Number of obs   =        29
-------------+----------------------------------   F(2, 26)        =      0.51
       Model |  4454354.32         2  2227177.16   Prob > F        =    0.6071
    Residual |   113826247        26  4377932.56   R-squared       =    0.0377
-------------+----------------------------------   Adj R-squared   =   -0.0364
       Total |   118280601        28  4224307.17   Root MSE        =    2092.4

------------------------------------------------------------------------------
       price | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
     foreign |
    Foreign  |   835.4296   844.6543     0.99   0.332    -900.7821    2571.641
     5.rep78 |  -424.3185   844.6543    -0.50   0.620     -2160.53    1311.893
       _cons |   5653.785   649.2909     8.71   0.000     4319.149    6988.422
------------------------------------------------------------------------------

. mat list e(b)

e(b)[1,5]
            0b.          1.         4b.          5.           
       foreign     foreign       rep78       rep78       _cons
y1           0   835.42963           0  -424.31852   5653.7852

* via -lincom- you can compare -domestic- with -4rep.78- (=_cons) vs. -foreign- with -5rep.78-*
. lincom _b[_cons] +  _b[4.rep78] + _b[0.foreign] -  _b[_cons] + _b[1.foreign] + _b[5.rep78]

 ( 1)  0b.foreign + 1.foreign + 4b.rep78 + 5.rep78 = 0

------------------------------------------------------------------------------
       price | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
         (1) |   411.1111   986.3437     0.42   0.680    -1616.347     2438.57
------------------------------------------------------------------------------

.

[Resa Pfu]

Thanks for the quick answers that helps! I was then now wondering, if I compare to one country/device in the regression, is there a command to set the base level to the country/device with the highest mean of dependant variable?

[Carlo Lazzaro]

Resa:
not that I know.
If you find that approach fruitful and methodologically sound, you can first calculate the mean yourself and then set the level of categorical variable the highest mean refers to as the reference category vis -fvvarlist- notation.

[Resa Pfu]

Alright! Thanks a lot!