5
$\begingroup$

I find a lot of resources online which explains the dummy variable trap and that you should remove 1 category of your dummy variable before fitting it into a multilinear model to avoid multicollinearity. While I understand what you should do I don't understand why you should do it in term of mathematical explanation. I mean, let's take a concrete example: I have a variable Gender with values Male or Female. If I take the multilinear model equation I get:

$$y = B_0 + x_1B_1 + x_2B_2$$

with $x_1 = 1$ and $x_2 = 0$. So I get: $y=0+1\times1 + 0\times1$ so how is it different from $y=0+1\times1$ (which the second dummy variable removed ? Could someone give me a concrete mathematical example of how this "trap" works? Thanks

$\endgroup$
2
  • 8
    $\begingroup$ You seem to describe a model for just a single observation! With $n$ observations, $B_0$ is a vector of ones, $B_1$ is a binary vector, $B_2$ is a binary vector, and (by construction) $B_0 = B_1+B_2$: that's collinearity. Regardless, your question itself contains the concrete example you ask for. $\endgroup$
    – whuber
    Commented Apr 13, 2018 at 14:39
  • 2
    $\begingroup$ Oh ok I see, that makes sense. Thanks for your answer :) $\endgroup$
    – E-Kami
    Commented Apr 14, 2018 at 11:32

1 Answer 1

1
$\begingroup$

We can see from Wikipedia that:

Multicollinearity refers to a situation in which two or more explanatory variables in a multiple regression model are highly linearly related.

In your case, that means that
$$x_1 = 1-x_2$$ and hence, your equation becomes \begin{align} y &= B_0 + x_1B_1 + x_2B_2 \\ &= B_0 + B_1(1-x_2) + B_2 x_2 \\ &= (B_0+B_1) + (B_2-B_1)x_2 \end{align}

It is obvious that $(B_0+B_1) + (B_2-B_1)x_2$ is equivalently $\alpha + \beta x$ which entails only one variable.

Reference: Dummy Variable Trap

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.