1
$\begingroup$

I have a problem regarding interpretation of a multiple regression. For $Y=a+bX_1+cX_2$, the influence of independent variable $X_1$ on $Y$, the dependent variable, is positive (coefficient $b$ is positive). This means that an increase of one unit of $X_1$ leads to an increase of one unit of $Y$, which is consistent with the economic interpretation.

But in my dataset, $X_1$ has also negative values. For those values, the interpretation is not correct anymore. An increase with 1 unit of $X_1$ will lead to a decrease of Y, meaning an inverse relation. What am I doing wrong? Is it better to use the data as absolute values ABS(X)?

$\endgroup$
  • $\begingroup$ Do you mean coefficient b or coefficient a? $\endgroup$ – Gala Jul 2 '13 at 11:20
  • $\begingroup$ coefficient b (the coefficient of the independent variable X1); sorry for the mistake $\endgroup$ – Maria Jul 2 '13 at 11:32
2
$\begingroup$

The interpretation is the same on both sides of 0. Ignoring the rest of the equation, with $b > 0$, $-1*b > -2*b$ so an increase from -2 to -1 in $X_1$ also corresponds to an increase in $Y$. The thing is, while it might feel a little strange when writing it, going from -2 to -1 is an increase and going from -1 to -2 is a decrease.

Now, if you have theoretical reasons to expect the relationship not to hold for negative values, i.e. if you think that a decrease, say from 0 to -1, should result in a higher $Y$, then a simple linear regression might not be appropriate (transforming the variable could be solution but a quadratic term, splines or an interaction with an ad hoc binary variable could also capture such relationships).

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Thank you very much! Your answer was very helpful. I understand now what I was doing wrong. Best regards, Maria $\endgroup$ – Maria Jul 2 '13 at 12:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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