2
$\begingroup$

In Linear Regression, we have the Conditional mean independence assumption:

E(u|x) = 0, where u is the error in the linear relationship.

May I clarify: my understanding of this is that implications of this assumption are that:

  1. u is normally distributed around 0, therefore expectation = 0 (I understand that it may not be a normal distribution now after seeing an answer here, but at the very least the average value is at 0)

  2. x doesn't influence anything about u (i.e. gives no information about u, and by extension its expectation) since x and u are independent (so this shows the independence)

Is my understanding correct?

$\endgroup$

1 Answer 1

2
$\begingroup$

Thanks for your questions.

In words, the assumption $E(u|x_1, ..., x_k)=E(u)=0$ states that the error term $u$ has an expected value of zero given any value of the independent variables.

  1. Therefore, the zero conditional mean assumption itself does not make a statement about which distribution $u$ has, only a statement about its expected value/mean.

For example, if you check the textbook "Introductory Econometrics" by Wooldridge you can compare assumptions MLR.4 and MLR.6. Only in assumption MLR.6, it is assumed that the error term follows a normal distribution. However, the more important assumption is MLR.4 which is needed for the OLS estimator to be unbiased.

  1. If the assumption $E(u|x_1, ..., x_k)=0$ holds $u$ and $x$ are said to be mean independent (technically, they must not be fully independent). An implication of this is that $u$ and $x$ are not correlated.

I'm not sure what you mean by the statement

x doesnt influence anything

But assume that the true model is $y=b_0 + b_1x + u$. Here, by definition, $x$ has an effect on $y$ of $b_1$, irrespective of whether the zero conditional mean assumption holds. If the assumption holds the OLS-estimator $\hat{b1}$ is an unbiased estimator of $b_1$. If the zero conditional mean assumption does not hold, this is not the case.

$\endgroup$
3
  • $\begingroup$ I think I meant to say in the second point that another implication of the zero mean conditional assumption is that "x gives no information of u" or like that "having a given value of x will not tell us anything of u and therefore not it's expectation as well" $\endgroup$
    – jojorabbit
    Oct 13, 2020 at 14:23
  • $\begingroup$ thank you so much for your answer nevertheless! I dont have the textbook but i'll see if i can get a hold of it $\endgroup$
    – jojorabbit
    Oct 13, 2020 at 14:24
  • 1
    $\begingroup$ Yes. Interpretations such as "x gives no information about u" are useful in thinking about the zero conditional mean assumption. You can see an overview of the assumptions in Wooldridge here . Compare MLR.4 and MLR.6 to see the difference. If I answered your question, please consider checking my answer as right. Thanks! $\endgroup$
    – curious
    Oct 13, 2020 at 14:31

Your Answer

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

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