0
$\begingroup$

I try to identify a nolinear relationship between a dependent variable and independent varaibles. In the literature, to detect this relation, we introduce the term. When I make a simple Regression ( OLS) with the introduction of variables (without their squared), the model gives not significant results, for the indépendents variables. On the other hand, it becomes significant, (similar for basic variables) when I add the square term. The problem that I do not know how interpret the sign of the variable coefficient in a non linear relationship.

$\endgroup$
1
1
$\begingroup$

First, adding a squared term is only one of many ways to look for a nonlinear relationship.

Second, don't over-rely on significance.

Now, to your question: There are various ways that a quadratic relationship can mask a linear one. I'll give an example with only one IV, for simplicity. Suppose $Y = x^2 + e$ and x ranges from -10 to 10. Then trying to fit a linear relationship will result in a coefficient near 0 (and a p value near 1).

If $y = x^2 + x + e$ then it's possible to get a nonsignificant linear relationship that becomes significant when the quadratic term is added.

Graphics can help interpret these models.

$\endgroup$

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.