I was refreshing my knowledge regarding the definition of joint probabilities and read a page from the book: 'Econometrics for dummies' which gave the following example.

enter image description here

What I intended to do was to calculate the joint probabilities $P(Y,X)$ from the unconditional probabilities $P(Y)$ and $P(X)$.

I came across the following two definitions of a joint probability:

Definition 1: $P(Y,X) = P(Y) \times P(X)$

Definition 2: $ P(Y,X) = P(Y|X) \times P(X) = P(X|Y) \times P(Y) $

Using Definition 1, I could not obtain the joint probabilities from the tables. For example multiplying $P(Y=1)=0.35$ and $P(X=2)=0.10$ did not give me $P(Y=1,X=2) = P(1,2) = 0$.

I do not understand why Definition 1 does not give me the right answer.

I believe the issue is that I am facing is:

I know that the conditional expectation of $P(Y|X)$ should not necesarily be equal to $P(Y)$. But does the second definition not imply that $P(Y|X)$ and $P(Y)$ are equal?


The first identity of the joint probability with the product of marginals

$$P(X,Y) = P(X) P(Y)$$

only holds when $X$ and $Y$ are independent. Based on the table and the fact that the product of the marginals do not equal the joint probabilities, should simply lead you to suspect a lack of independence.

The second identity is true by the definition of conditional probability where

$$P(X\lvert Y) := \frac{P(X,Y)}{P(Y)}$$ hence

$$P(X\lvert Y) P(Y) = P(X,Y).$$

And this second "definition" does not imply that $P(Y) = P(Y\lvert X)$, I do not know why you are lead to that conclusion.

  • $\begingroup$ I marked your answer as the correct one.Thanks for the clarification regarding the property of independence. $P(Y) = P(Y|X)$, given that X and Y are independent! It all makes sense now. I lead to that conclusion if you'd combine definition 1 and 2 (However, using definition 1 assumes independence): $P(Y,X) = P(Y|X) \times P(X) = P(Y) \times P(X)$ $\endgroup$ – Nadia Merquez Oct 18 '19 at 11:00
  • $\begingroup$ yes that is true $P(X\lvert Y) = P(X)$ and $P(Y\lvert X) = P(Y)$ given independence. So for two independent variables conditional and marginal probabilities are the same - knowing the value of one of the variables does not put you in a better position to predict the other. But of course the key again is that by using definition 1 you have already assumed independence (as you yourself state). $\endgroup$ – Jesper for President Oct 18 '19 at 11:05
  • $\begingroup$ In order to gain full clarification, given that there is a form of dependence between $X$ and $Y$ and one would like to obtain the joint probability $P(Y,X)$. Then, one would have to use the second identity $P(Y,X) = P(Y|X) \times P(X) = P(X|Y) \times P(Y)$, implying that at least one marginal probability and one conditional probability has to be known, right? $\endgroup$ – Nadia Merquez Oct 18 '19 at 11:29
  • $\begingroup$ To say that one "would have to" is probably to strong in general. But in this context where the choice is between rule 1 and rule 2, then I would say you are right. But note also that in the table you actually observe joint frequencies that can be used to estimate the joint probabilities directly, but I guess that is another story. $\endgroup$ – Jesper for President Oct 18 '19 at 11:40
  • $\begingroup$ Thanks for the insightful remarks! $\endgroup$ – Nadia Merquez Oct 18 '19 at 11:51

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.