19
$\begingroup$

I'm seeking a better understanding of the following problem from Blitzstein and Huang (2015) (Chapter 1, Exercise 31, p. 35):

A jar contains $r$ red balls and $g$ green balls, where $r$ and $g$ are fixed positive integers. A ball is drawn from the jar randomly (with all possibilities equally likely), and then a second ball is drawn randomly.

(a) Explain intuitively why the probability of the second ball being green is the same as the probability of the first ball being green.

Now, I get that in theory, before an experiment is conducted, there is no evidence to condition on, and the two probabilities are equivalent. No particular ball has any particular likelihood to be selected over any other. But in practice, I just don't buy it.

The odds of the first ball being green is $g/(r + g)$. There are strictly two possible outcomes when a ball is drawn once: it is either green or red. If it's green, the odds of the 2nd ball being green is $(g-1) / (r + g - 1)$. If the first ball is red, the odds of the second being green is $g/ (r + g -1)$. Neither of these is equivalent to $g/(r + g)$. While it is true that this is a claim about an experiment that has already happened, versus one that has not yet taken place, it still makes me feel like the author's claim actually contains no information, something like: "We don't know how to adjust the odds before the experiment because the experiment hasn't happened yet."

Can anyone help me understand, not why I'm wrong, but why the author's claim carries useful information?

$\endgroup$
3
  • 2
    $\begingroup$ Here is an argument between people who do understand the author's claim and people who don't: reddit.com/r/magicTCG/comments/24kfjb/…. Understanding the claim makes you better at Magic: The Gathering. (Note that this link may not be helpful to people who don't know the rules of Magic: The Gathering.) $\endgroup$
    – fblundun
    Commented Jul 8 at 12:36
  • 4
    $\begingroup$ Consider: Suppose you draw all but one of the balls, discarding them without looking at them. What is the probability of the last ball being green? $\endgroup$
    – Sneftel
    Commented Jul 8 at 12:39
  • 3
    $\begingroup$ You aren't answering the question. You (correctly) compute conditional probabilities. By following the laws of probability, you will perform the proper multiplication to discover (perhaps non-intuitively) that at least those laws imply the unconditional chances are the same. $\endgroup$
    – whuber
    Commented Jul 9 at 20:09

6 Answers 6

23
$\begingroup$

Let $\mathcal{G}_1, \mathcal{G}_2, \mathcal{G}_3, ...$ be the events that each consecutive drawn ball is green. You are correct to point out that:

$$\begin{align} \mathbb{P}(\mathcal{G}_1) &= \frac{g}{g+r}, \\[6pt] \mathbb{P}(\mathcal{G}_2| \mathcal{G}_1) &= \frac{g-1}{g+r-1}, \\[8pt] \mathbb{P}(\mathcal{G}_2| \bar{\mathcal{G}_1}) &= \frac{g}{g+r-1}. \\[6pt] \end{align}$$

Where you are going wrong is that you are looking at the conditional probabilities for the second ball being green (conditional on the two possible outcomes for the first ball) but you are not looking at the marginal probability that the second ball is green. Conditioning on the outcome of the first drawn ball and applying the law of total probability using the probabilities you have already specified gives:

$$\begin{align} \mathbb{P}(\mathcal{G}_2) &= \mathbb{P}(\mathcal{G}_2| \mathcal{G}_1) \cdot \mathbb{P}(\mathcal{G}_1) + \mathbb{P}(\mathcal{G}_2 | \bar{\mathcal{G}_1}) \cdot \mathbb{P}(\bar{\mathcal{G}_1}) \\[6pt] &= \frac{g-1}{g+r-1} \cdot \frac{g}{g+r} + \frac{g}{g+r-1} \cdot \frac{r}{g+r} \\[6pt] &= \frac{g(g-1)}{(g+r)(g+r-1)} + \frac{gr}{(g+r)(g+r-1)} \\[6pt] &= \frac{g(g-1) + gr}{(g+r)(g+r-1)} \\[6pt] &= \frac{g(g+r-1)}{(g+r)(g+r-1)} \\[6pt] &= \frac{g}{g+r} \\[12pt] &= \mathbb{P}(\mathcal{G}_1). \\[6pt] \end{align}$$

As you can see, although the conditional probabilities of the second ball being green are not the same as the probability that the first ball is green, the marginal probability that the second ball is green is the same as the probability that the first ball is green.

In regard to your specific question about the intuitive reasoning, the law of total probability essentially captures the idea of "adjusting the odds before the experiment because the experiment hasn't happened yet". Using this law, we consider the odds of an event of interest under each possible outcome of an experiment and we weight these odds by the probability of each possible outcome of the experiment to get an overall sense of the odds of the event of interest. It is this weighting of the conditional probabilities (to come up with an overall marginal probability) that you were missing in your own analysis.

$\endgroup$
1
  • 2
    $\begingroup$ To intuitively grasp the marginal probability it can be helpful to think about this in a frequentist way. Given some large number of jars that each have r red balls and g green balls and drawing two balls from each jar, what proportion of jars do we expect to have a green ball removed on second draw? This approach can give an intuitive understanding why the weighing makes sense. $\endgroup$
    – blues
    Commented Jul 10 at 9:37
14
$\begingroup$

The solutions for exercises marked with the circled S symbol are available from this page: https://projects.iq.harvard.edu/stat110/strategic-practice-problems

The intended solution is:

This is true by symmetry. The first ball is equally likely to be any of the g + r balls, so the probability of it being green is g/(g + r). But the second ball is also equally likely to be any of the g + r balls (there aren’t certain balls that enjoy being chosen second and others that have an aversion to being chosen second); once we know whether the first ball is green we have information that affects our uncertainty about the second ball, but before we have this information, the second ball is equally likely to be any of the balls. Alternatively, intuitively it shouldn’t matter if we pick one ball at a time, or take one ball with the left hand and one with the right hand at the same time. By symmetry, the probabilities for the ball drawn with the left hand should be the same as those for the ball drawn with the right hand.

$\endgroup$
0
6
$\begingroup$

What's "intuitive" to one may be less so to another. I find @Flounderer's answer to be quite intuitive, but here is an alternative if that didn't do it for you:

An equivalent process is to draw two balls simultaneously, then randomly choose one of the two (with equal probability) to designate as the first, and the other as the second.

The two balls chosen are either both red, both green, or mixed, each with some probability, call it $p_1, p_2, p_3$. In the "both green" case, the probability that the ball labeled "first" is green is 1, and in the "mixed" case, the probability that the ball labeled "first" is green is 50%, for an average of $1 \cdot p_2 + 0.5 \cdot p_3$. By symmetry, the probability that the ball labeled "second" is green is the same.

$\endgroup$
4
$\begingroup$

The abstract mathematical machinery is intuitive for many people provided it is clearly set up in a way that parallels the original probability problem.

Before reading any more of this post, though, please skip to the end: if the argument made there is anything but intuitive to you, then you can return here and read the details.


Analysis

The idea is to create an obviously correct mathematical model of the draws by mechanically and directly representing those draws with mathematical objects. At this stage of the analysis we will minimize the amount of thought and insight needed, so we can be sure of capturing any subtleties that might elude our intuition.

The first step is to represent the experimental outcomes accurately and completely. The first draw will be one of the $n = r+g$ balls; the second draw will be a different one of the balls. Thus, if we were to number the balls $1, 2, \ldots, n$ to distinguish them, then the possible draws are represented by all ordered pairs of indices $(i,j)$ where $1\le i\le n,$ $1\le j \le n,$ and $i\ne j.$ The first entry in an ordered pair designates the first ball drawn and the second entry designates the second ball drawn.

This is the sample space $X.$

The second step is to determine the probabilities. You know two things about them:

  1. All balls are equally likely to be the first draw. Thus, every subset of the form $$\mathcal E_i = \{(i,1), (i,2),\ldots,(i,i-1),\quad (i,i+1),\ldots, (i, n)\}$$ has the same probability. Since there are $n$ such events, $$\mathbb P(\mathcal E_i) = \frac{1}{n}$$ for every $i.$

  2. All remaining balls are equally likely to be the second draw. This says that within each $\mathcal E_i,$ each of the $n-1$ ordered pairs is equally likely. Allocating the total probability $\mathbb P(\mathcal E_i)$ to each of its members thereby gives each of those members a probability $$\mathbb P(\{(i,j)\})=\frac{1}{n-1}\times\mathbb P(\mathcal E_i) = \frac{1}{n(n-1)}.$$

We have thereby constructed a probability space out of this sample space and endowed it with a probability measure $\mathbb P$ that models the problem. This space has the salient characteristics that

  1. All elements of $X$ have probabilities and

  2. Those probabilities are all equal (to $1/(n(n-1))$): all singletons are equiprobable.

(Having found this result, we could have argued at the outset that all pairs are equally probable--and thereby avoid doing any algebra at all, as you will see.)


As is often the case with finite sample spaces, the solution to any question of probabilities often devolves to counting. These events are:

  • $\mathcal G_1:$ the first ball is green.

  • $\mathcal G_2:$ the second ball is green.

Let $G$ be the subset of indices in $\{1,2,\ldots, n\}$ corresponding to the green balls. $G$ therefore has $g$ elements. Use it to express the first event explicitly:

$$\mathcal G_1 = \{(i,j)\mid i\in G,\ 1\le i\le n,\ 1 \le j\le n,\ i\ne j\}.$$

The second event is expressed in the same way, merely with the roles of $i$ and $j$ switched:

$$\mathcal G_2 = \{(i,j)\mid j\in G,\ 1\le i\le n,\ 1 \le j\le n,\ i\ne j\}.$$

Since the question asks us to see why two events have the same chances, we would be well served by some insightful observation rather than mechanical counting or applying some combinatorial formula. This motivates studying the permutation $\pi:X\to X$ given by swapping the indices:

$$\pi((i,j)) = (j,i)$$

for all $(i,j)\in X.$

Here, then, is the full rigorous argument:

$\pi$ does not change the probability of any event, since every $(i,j)$ is equiprobable. Because $\pi(\mathcal G_1)=\mathcal G_2$ and $\pi(\mathcal G_2) = \mathcal G_1,$ $\mathcal G_1$ and $\mathcal G_2$ have the same probabilities.

If you care to write this formally (which is one method mathematics uses to develop insight) it might go like this:

$$\begin{aligned} \mathbb P(\mathcal G_1) &= \sum_{i\in G}\sum_{j\ne i} \mathbb P(i,j)&\text{Axiom}_1\\ &= \sum_{i\in G}\sum_{j\ne i} \mathbb P(\pi(i,j))&\pi\text{ preserves }\mathbb P \\ &= \sum_{i\in G}\sum_{j\ne i} \mathbb P(j,i)&\text{Definition of }\pi \\ &= \sum_{j\in G}\sum_{i\ne j} \mathbb P{(i,j)}&\text{Relabel the subscripts}\\ &= \mathbb P(\mathcal G_2).&\text{Axiom}_1 \end{aligned}$$

"$\text{Axiom}_1$" is the probability axiom that the probability of a finite union of disjoint events is the sum of the probabilities of those events.

Solution

By reflecting on this demonstration and the insight afforded by $\pi,$ you might want to capture the intuition in words. Here's one effort:

Because there are just as many pairs of draws in which the first ball is green as there are pairs of draws in which the second ball is green, and every pair of draws has the same chance of occurring, the probability the first ball is green equals the probability the second ball is green.

As with many mathematical discoveries, in retrospect the result looks obvious. That's what is meant by "intuitive."

$\endgroup$
3
$\begingroup$

Imagine there are only two balls, one of each color.

If the first ball is red, the probability of the second one being green is $100\%$.

If the first ball is green, the probability of the second one being green is $0\%$.

But you wouldn't argue that the overall probability of the second ball being green is either $100\%$ or $0\%$, would you? It must lie somewhere in between. Where exactly?

In one out of two scenarios the probability is 0%. In the other it's $100\%$. So the overall probability is $\frac{1}{2} 0\% + \frac{1}{2} 100\%$ which is, of course $50\%$

With this logic in mind, we build the general case for $r$ and $g$ balls of each color as in Ben's answer

$\endgroup$
0
$\begingroup$

Let $b_1, \dots, b_r$ be the $r$ red balls and let $b_{r+1}, \dots, b_{r+g}$ be the $g$ green balls. Let $\mathcal{B_1}, \mathcal{B_2}$ denote the first and the second drawn ball, respectively. If we ignore the color, everything about the $r+g$ balls is symmetric, each ball has the same properties. I.e., the probability distribution is symmetric for these $r+g$ balls. By symmetricity, we have that each of the balls must have equal probability of being pulled on the second try - $$p(\mathcal{B}_2 = b_1) = p(\mathcal{B}_2 = b_2) = \dots = p(\mathcal{B}_2 = b_{r+g}).$$ Since total probability must sum to one, we have that for each $i \in \{1, \dots, r+g\}$, $p(\mathcal{B}_2=b_i) = \frac{1}{r+g}$. Then we get that $$p(\mathcal{B}_2 \text{ is green}) = p(\mathcal{B}_2=b_{r+1}) + \dots + p(\mathcal{B}_2=b_{r+g}) = \frac{g}{r+g}.$$

$\endgroup$

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.