Say a sample $(X_1,X_2,X_3)$ is taken from a $Ber(p)$ distribution, we need to check the conditional distribution of $(X_1,X_2,X_3)$ given a value of $X_1X_2+X_3$, i.e. we need to check $P(X_1=x_1,X_2=x_2,X_3=x_3|X_1X_2+X_3=s)$.

Now, $s$ can take values $0,1,2$.

I am kinda confused here, I am used to tackle conditional probabilities as follows :

I see the 'given' part ($X_1X_2+X_3=s$ here.) as a 'thing' that restricts our sample space. Here our sample space consists of 8 points $(0,0,0),(1,0,0),(0,1,0),(0,0,1),(1,1,0),(0,1,1),(1,0,1),(1,1,1)$.

So when we say $P(X_1=0,X_2=0,X_3=0|X_1X_2+X_3=0)$, we say - "First get into the points that satisfy $X_1X_2+X_3=0$, that will be $(0,0,0),(1,0,0),(0,1,0)$, and then check the probability of $(0,0,0)$, which comes out to be $\frac{1}{3}$. "

Now, if we proceed in a 'conventional' manner, i.e. like $P(A|B)=\dfrac{P(A\cap B)}{P(B)}$, the term above comes out to be in the form of $p$ (As given in the solution).

So I wanted to know where does the first method fails ?

Or is it completely wrong ?

  • $\begingroup$ Try substituting $p=\frac 12$ in the “official” answer to see if you get your “simple” answer which assumes that all eight points are equally likely, which is equivalent to assuming that the random variables are $Ber(1/2)$. $\endgroup$ May 3, 2018 at 11:29
  • $\begingroup$ Make a table of eight rows, one per outcome, in which the first three columns give the values of $X_1,X_2,X_3.$ In a fourth column write the probabilities of each outcome. With simple calculations, taking very little work, you can compute a fifth column giving the values of $X_1X_2+X_3,$ which you might as well label $s$. Focusing on those last two columns, can you now compute the chances $\Pr(s=0),$ $\Pr(s=1),$ and $\Pr(s=2)$? $\endgroup$
    – whuber
    May 3, 2018 at 12:53

1 Answer 1


I am not sure I understand your question, however I think both methods are ok and give the same result:

P(B) = P((0,0,0),(1,0,0),(0,1,0)) = 3/8

P(A∩B) = P((0,0,0)) = 1/8

so that

P(A∩B)/P(B) = 1/3

In general:

$$P(X_i=0) = 1-p,$$


$$P(X_1=0,X_2=0,X_3=0) = (1-p)^3$$


$$P(X_1X_2+X_3=0) = (1-p)* [ (1-p)^2+2p*(1-p)]$$


$$P(A∩B)/P(B) = \frac{1-p}{1+p}$$

  • $\begingroup$ The solution says something like this: $P(X_1=0,X_2=0,X_3=0 | X_1X_2+X_3=0)=\dfrac{1-p}{1+p}$ $\endgroup$
    – User9523
    May 3, 2018 at 9:44
  • $\begingroup$ so, if p=1/2, as you have assumed, the solution will be 1/3. $\endgroup$
    – fabiob
    May 3, 2018 at 12:04
  • $\begingroup$ So the 'flaw' I am looking in the first method is that, I am assuming each point is equally likely, right ? $\endgroup$
    – User9523
    May 3, 2018 at 14:25
  • 1
    $\begingroup$ Yes, however let me insist that both methods are valid, if applied correctly. If p=1/2 then you are applying the first method correctly. If p $\neq 1/2$ then, as you are saying, it would be wrong/a flaw to consider that (0,0,0) and (1,0,0) have the same probability to occur. Instead their probabilities to occur would be respectively: $(1-p)^3$ and $p(1-p)^2$. $\endgroup$
    – fabiob
    May 3, 2018 at 15:16

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.