Consider a Cobb-Douglas model :

$$ y = \beta_0x_1^{\beta_1}x_2^{\beta_2} $$

I would like to know if we can talk about interaction in this model concerning the variables $x_1$ and $x_2$. When thinking of this with $x_1$ the labour, $x_2$ the capital and $y$ the the total production we like to think of $\beta_1$ and $\beta_2$ as output elasticities and at first I thought this gives us nothing in terms of interaction. One way to see interactions is to talk about the fact that when we have two variables $x_1$ and $x_2$ that influences another variable $y$, we could see the interaction between $x_1$ and $x_2$ as follows : a change in $x_1$ depends of $x_2$ and so the impact of a change in $x_1$ on $y$ will not be the same given different levels of $x_2$. We can formalize this by saying that the partial derivatives of $y$ with respect to $x_1$ depends on $x_2$. From this, we see that in a Cobb-Douglas model, there are interactions between explanatory variables.

Here is my thoughts, I would like to know if it is correct ?

Thank you a lot !


1 Answer 1


This is more a matter of terminology than of substance. Yes, in the form you show, there is a product between $x_1$ and $x_2$. One might consider such a product to represent an "interaction," as the association between one predictor and outcome depends on the value of the other predictor.

But if you do a log transformation, you have:

$$\log y = \log\beta_0 + \beta_1 \log x_1+ \beta_2 \log x_2, $$

which is an equation for a linear (in the coefficients) regression without an interaction term. The association of neither log-transformed predictor with outcome depends on the value of the other. So I'm not sure that it's helpful to think of an "interaction" here.

  • $\begingroup$ Thank you for your answer ! It is true that a log change annihilates this interaction... I would like to be able to explain why but I am not aha. Beyond that, I'm interested in interactions because I'm trying to represent them (and ideally quantify them, which is going to be complicated). $\endgroup$
    – G2MWF
    Commented Jun 8, 2023 at 6:39
  • 1
    $\begingroup$ @coboy the simplest way to represent the apparent "interaction" in the Cobb-Douglas formula is to work in the log-transformed scale where they disappear. In this case, the fundamental information is the same in either situation. The question is how well the Cobb-Douglas formula will actually represent the data. $\endgroup$
    – EdM
    Commented Jun 8, 2023 at 13:31
  • $\begingroup$ Yes I had the same thoughts on the invariance of "fundamental information" by a log transformation, the model is pretty relevant for my purpose. $\endgroup$
    – G2MWF
    Commented Jun 8, 2023 at 14:43

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.