Regression maintaining variable constant

[Lorena Alzamora]

Hello, how can i make a regression maintaining one variable constante? specifically i need to show the expected change in hours worked per week of the household head by an additional household member holding the age of the household head constant. Thank you so much

[Carlo Lazzaro]

Lorena:
if one of your predictors is=K it will be omitted from the output table, as reported in the following toy-example:

use "https://www.stata-press.com/data/r16/auto.dta"
g mpg_k=mpg[1]
regress price i.foreign mpg_k
. regress price i.foreign mpg_k
note: mpg_k omitted because of collinearity

      Source |       SS           df       MS      Number of obs   =        74
-------------+----------------------------------   F(1, 72)        =      0.17
       Model |  1507382.66         1  1507382.66   Prob > F        =    0.6802
    Residual |   633558013        72  8799416.85   R-squared       =    0.0024
-------------+----------------------------------   Adj R-squared   =   -0.0115
       Total |   635065396        73  8699525.97   Root MSE        =    2966.4

------------------------------------------------------------------------------
       price |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
     foreign |
    Foreign  |   312.2587   754.4488     0.41   0.680    -1191.708    1816.225
       mpg_k |          0  (omitted)
       _cons |   6072.423    411.363    14.76   0.000     5252.386     6892.46
------------------------------------------------------------------------------

.

Therefore, the results of your OLS will be the same as the ones obtained without the predictor=K:

. regress price i.foreign

      Source |       SS           df       MS      Number of obs   =        74
-------------+----------------------------------   F(1, 72)        =      0.17
       Model |  1507382.66         1  1507382.66   Prob > F        =    0.6802
    Residual |   633558013        72  8799416.85   R-squared       =    0.0024
-------------+----------------------------------   Adj R-squared   =   -0.0115
       Total |   635065396        73  8699525.97   Root MSE        =    2966.4

------------------------------------------------------------------------------
       price |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
     foreign |
    Foreign  |   312.2587   754.4488     0.41   0.680    -1191.708    1816.225
       _cons |   6072.423    411.363    14.76   0.000     5252.386     6892.46
------------------------------------------------------------------------------

.

[Nick Cox]

Cross-posted at https://www.reddit.com/r/stata/comme...able_constant/

Please note our cross-posting policy, which is that you should tell us about it. https://www.statalist.org/forums/help#crossposting

I read the question differently from Carlo Lazzaro. It's not about how to do the regression; it's a matter of reaching for margins and marginsplot afterwards.

[Carlo Lazzaro]

Nick is correct.
I was probably off-target (or partially so ).

[Lorena Alzamora]

Thank you Nick! I didnt know about that rule, thanks!

[Lorena Alzamora]

Thank you Carlo, you helped me!

[Jeff Wooldridge]

Based on my teaching, I took this to be more of a basic question about multiple regression. It's important to know that multiple regression, in effect, allows you to do the thought experiment of holding one variable fixed while changing another. Of course, that can't be true in the data set. But the estimated equation allows us to do this. So, if you use

[CODE]reg hhheadhours hhmembers hhage, vce(robust)/CODE]

then the coefficient on hhmembers has the ceteris paribus interpretation you're looking for. So if its coefficient is 4.3 then, holding hhage fixed, another household member increases the hours worked by the head by 4.3 hours, on average. The regression does that for you. If you run the simple regression

reg hhheadhours hhmembers, vce(robust)

then you are not "controlling for" or "holding fixed" hhage. If you want a fancier model, you can interact hhmembers and hhage and then use the margins command and marginsplot, as suggested by Nick.