# Covariance question.

1. Jun 9, 2013

### peripatein

Hi,
1. The problem statement, all variables and given/known data
A fair die is rolled n times. X denotes the number of times '1' is obtained. Y denotes the number of times '6' is obtained.
I am first asked to state how X and Y are distributed (marginally) and to find their variance.

2. Relevant equations

3. The attempt at a solution
Aren't X and Y distributed Binomially, with p = 1/6 and n? I.e. isn't the Variance of each then equal to:
np(1-p) = 5n/36?
Moreover, how may I determine their marginal distribution?
I am not sure and would appreciate any feedback. Thanks!

2. Jun 9, 2013

### Staff: Mentor

Right
Do you know what the marginal distribution is? If yes, where do you get a problem?

3. Jun 9, 2013

### peripatein

I know what a marginal distribution is but I am still not sure how to approach this.
Here is an attempt. Could it be:
(n over x)(n-x over y)/(n over (x+y))?

4. Jun 9, 2013

### Ray Vickson

You have already told us what are their marginal distributions! Do you mean that you are asked to find their joint distribution?

Last edited: Jun 9, 2013
5. Jun 9, 2013

### peripatein

No, the question is phrased thus:
A) How are X and Y distributed (marginally)? What are their variances?
B) Suppose X_i = 1 if the i-th roll is '1' and 0 otherwise. Y_i = 1 if the j_th roll is '1' and 0 otherwise. Why is Cov(X_i,Y_j)=0 for i different than j? What is Cov(X_i,Y_j)=0 for i=j?
C) What are Cov(X,Y) and RHO_XY?

A) Is the marginal distribution of X and Y then Binom(n,1/6) and Var=5/36?
B) Cov(X_i,Y_i)=E(X_iY_j) - E(X_i)E(Y_i). If i is different than j then E(X_i)E(Y_i)=E(X_i)E(Y_j) as these are two independent events and Cov(X_i,Y_j) would be zero. Is that really why? As i=j, E(X_iY_i)=0 and Cov(X_i,Y_j) = -(n^2)/36. Is that correct?
C) Cov(X,Y) should therefore be -(n^3)/36. Correct?
Isn't then RHO_XY = Cov(X,Y)/SQRT(Var(X)Var(Y))=-(n^2)/5?

6. Jun 9, 2013

### Ray Vickson

There is something wrong with the problem setup: first you refer to X and Y as counting 1s and 6s, but later you refer to X_i and Y_j as both referring to 1s. Are you sure you do not mean that X_i = 1 if toss i is '1' and Y_i = 1 if toss i is '6'? At some point you WILL need both types of indicators because you need to count both 1s and 6s.

7. Jun 9, 2013

### peripatein

X_i is 1 if we get '1' on the ith roll and it is 0 othetwise.
X_j (or Y_j) is 1 if we obtain a '6' on the jth roll and it is 0 otherwise.
I apologise for any previous typos. I am sending this via my mobile.
Is it more comprehensible now?

8. Jun 9, 2013

### peripatein

More importantly, is it clear enough for you to be able to comment on my attempt?

9. Jun 9, 2013

### Ray Vickson

No: your definition of Y_j is still problematic, especially since you have already defined X_i in a certain way. Look over what you wrote, and think about whether it makes sense.

10. Jun 9, 2013

### peripatein

Xi=1 if we get '1' on the ith roll and Xi=0 otherwise.
Yj=1 if we obtain a '6' on the jth roll and Yj=0 otherwise.
This is the formulation in my textbook. Is it still sloppy? As in that case I truly would have no idea how to render this clearer for you, Ray.

11. Jun 9, 2013

### Ray Vickson

No, NOW it is not sloppy, but this is not what you wrote before!

So, you need to be able to compute $E(X_i Y_j )$ for both cases $i = j$ and $i \neq j$.

What is $E(X_i Y_i)$ (the i = j case)?

How would you compute $E(X_i Y_j)$ for $i \neq j?$ Hint: what event must occur in order to have $X_i Y_j > 0$?

12. Jun 9, 2013

### peripatein

For i=j, E(XiYj) = 0. For i different than j, is E(XiYj)=(n over 2)*(1/6)^2*(5/6)^(n-2)?

13. Jun 10, 2013

### Ray Vickson

You tell me. How did you get that? Show your work in detail.

14. Jun 10, 2013

### peripatein

Well, if i=j there could only be one possible outcome, either '1' or '6'. Therefore, the probability of both of these happening at once has to be NULL. As these are indicators that is also E(XiYj).
When i≠j, wouldn't E(XiYj) be equal the probability of obtaining '1' on the i-th roll and '6' on the j-th? And isn't that nC2 * (1/6)2(5/6)n-1?

15. Jun 10, 2013

### Ray Vickson

Why should the probability of obtaining '1' on the second toss and '6' on the fifth toss depend on whether we make 100, 200 or 300 tosses? (That is what your answer indicates.)

16. Jun 10, 2013

### peripatein

So is it simply 1/36?

17. Jun 10, 2013

### Ray Vickson

You tell me.

18. Jun 10, 2013

### peripatein

Hmm, I guess I was mistaken as these are not entirely two independent events. If I get x '1's, that means that I could at most get n-x '6's.
But if that is the case, why couldn't I use the Binomial formula above?

19. Jun 10, 2013

### peripatein

I infer from your silence that either you are busy, or I am way off, or both ;-).

20. Jun 10, 2013

### Ray Vickson

It is summer, and the weather is lovely outside. I want to walk my dog and sit outside at the coffee shop. I do not always care to give advice FREE OF CHARGE, especially when it seems that no matter how much advice I give it is never enough!

I will say things one last time: assuming the successive tosses of the die are independent, the different X_i (i=1,...,n) are mutually independent, as are the different Y_i. Furthermore, for i≠j the random variables X_i and Y_j are independent; however, X_i and Y_i are dependent (because if one of them = 1 the other = 0; however, both could = 0, of course.) Take it from there.