Interaction term

[Vera van Oijen]

Hi Everyone!

I try to run a regression with an interaction variable in Stata. The interaction term consists of a dummy (value either 0 or 1) and a normal variable. However, the coefficient gives the value 0 or 1.
What am I doing wrong?

Thanks in advance!
Vera

[Rich Goldstein]

without showing EXACTLY the command you used, and possibly providing a data example (using -dataex- and CODE blocks), it is not possible to answer your question; please read the FAQ and try again

[Vera van Oijen]

Okay I will try to be more clear.

First of all, I tried to generate an interaction term with the following command:
generate Emorec1=Happy*Recency1
generate Emorec2=Happy*Recency2

Happy is the dummy variable and if it is happy it has the value 1 and if it is sad it gets the value 0.
Recency is a variable that reflects the return.

Afterwards I try to run the regression and I get the following result:
I can't upload the image, but I typed in regress Recency1 Happy Emorec1
And I obtained a coefficient of 0.193435 for the variable Happy
And a coefficient of 1 for the variable Emorec1

However, the coefficient Emorec1 shouldn't be 1.

[Vera van Oijen]

. regress Recency2 Happy Emorec2

Source | SS df MS Number of obs = 213
-------------+---------------------------------- F(2, 210) = 252.85
Model | 9.63195739 2 4.81597869 Prob > F = 0.0000
Residual | 3.99990196 210 .019047152 R-squared = 0.7066
-------------+---------------------------------- Adj R-squared = 0.7038
Total | 13.6318594 212 .064301223 Root MSE = .13801

------------------------------------------------------------------------------
Recency2 | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
Happy | -.0017099 .018916 -0.09 0.928 -.0389994 .0355796
Emorec2 | 1 .0444796 22.48 0.000 .9123163 1.087684
_cons | .0017099 .0133421 0.13 0.898 -.0245916 .0280115
------------------------------------------------------------------------------

[Ken Chui]

Interaction model does not operate this way. Usually it involves at least three variables: the dependent (y), and two independent variables (x1, x2). An interaction model would look like:

y = b0 + b1*x1 + b2*x2 + b3*(x1*x2)

In your case, you're predicting a variable with an interaction term derived from it, kind of like:

x1 = b0 + b1*x2 + b2*(x1*x2)

Because the values in (x1*x2) is the same as x1 when x2 == 1, that means it's partially identical to x1 when it's not 0, so the best-fit coefficient is 1.