math.wiki
Probability › Random variables › Expectation, linearity and indicators
Sign in

Expectation, linearity and indicators

A count such as how many of your bingo numbers have been called has a tangled distribution, and a waiting time takes each single value with probability zero. What does the average of such a quantity mean, and can it be found without the whole distribution?

Where you are 5 in Random variables

Why this exists

The school lesson Expected value weighted each value by its probability, and Result 6 of Discrete distributions: binomial, geometric, Poisson made that exact for a discrete variable. Two questions remain. A variable with a density gives every single value probability 0 (Densities and cumulative distribution functions), so a sum has nothing to weight. And many counts have tangled distributions: after 30 calls in a bingo game, how many rows of Zoe’s card are complete?

The first question is answered by an integral against the density. The second is answered by linearity: the expectation of a sum is the sum of the expectations, whether or not the terms are independent. Written as a sum of indicators, a count has a mean equal to a sum of probabilities, and its distribution never has to be found.

Definition

Let X be a random variable. If X is discrete with pmf p_X, its expected value, or mean, is E(X) = \sum_x x \, p_X(x), summed over the support, provided \sum_x |x| \, p_X(x) converges. If X has a density f_X, its expected value is

E(X) = \int_{-\infty}^{\infty} x \, f_X(x) \, dx,

provided \int_{-\infty}^{\infty} |x| \, f_X(x) \, dx converges. A mixed variable, with masses p_X(x) at countably many points and a density part g carrying the rest of its probability (Example 5 of Densities and cumulative distribution functions), has E(X) = \sum_x x \, p_X(x) + \int x \, g(x) \, dx, provided both converge absolutely.

When the absolute sum or integral diverges, X has no finite mean; for X \ge 0 we then write E(X) = \infty. The mean is also written \mu_X. Absolute convergence makes the value independent of the order in which values are listed (Absolute and conditional convergence).

For a function g, g(X) is the random variable whose value at the outcome \omega is g(X(\omega)). The indicator of an event A is the random variable \mathbf{1}_A with \mathbf{1}_A(\omega) = 1 for \omega \in A and \mathbf{1}_A(\omega) = 0 for \omega \notin A.

Intuition

Spread one unit of mass along a line with density f_X. E(X) is where the line balances: the moment \int (x - \mu_X) f_X(x) \, dx about it is zero, as the bars of a pmf balance at the weighted mean (Figure 1).

0 2 4 6 8 10 0.2 median
E(T) = \tfrac{20}{3}
f_T(t)
t
Figure 1The density \displaystyle f_T(t) = \frac{t}{50} of Example 1. The shaded region balances on the triangle at \displaystyle E(T) = \frac{20}{3}, left of the median \displaystyle 5\sqrt{2}.

An indicator is a switch that reads 1 when its event happens. A count is a sum of switches, and the mean of a switch is the chance that it is on, so the mean of a count is the sum of the chances, however the switches affect one another.

Results

Result 1 (the mean of a function of X). Let g be a real function. If X is discrete, then g(X) has a finite mean exactly when \sum_x |g(x)| \, p_X(x) converges, and then

E(g(X)) = \sum_x g(x) \, p_X(x).

If X has a density and g is continuous except at finitely many points of each bounded interval, then whenever \int |g(x)| \, f_X(x) \, dx converges,

E(g(X)) = \int_{-\infty}^{\infty} g(x) \, f_X(x) \, dx.

The density case is proved below for monotone g and stated without proof in general.

Result 2 (linearity). If X and Y have finite means and a, b, c are constants, then aX + bY + c has a finite mean and

E(aX + bY + c) = a \, E(X) + b \, E(Y) + c.

By induction, E(X_1 + \dots + X_n) = E(X_1) + \dots + E(X_n) when every term has a finite mean. No independence is assumed. Proved below for discrete variables and for aX + c; Joint distributions proves it for densities. The case b = 0 is the school rule of Linear transformations of a random variable.

Result 3 (indicators). E(\mathbf{1}_A) = P(A). If N is the number of the events A_1, \dots, A_n that happen, then N = \mathbf{1}_{A_1} + \dots + \mathbf{1}_{A_n} and

E(N) = P(A_1) + \dots + P(A_n).

Result 4 (the tail formula). If X has a density, then \int_0^\infty x \, f_X(x) \, dx and \int_0^\infty P(X > x) \, dx are equal, both finite or both infinite. So if X \ge 0, E(X) = \int_0^\infty P(X > x) \, dx, the area under the curve x \mapsto P(X > x). Exercise 14 of Discrete distributions: binomial, geometric, Poisson is the same fact for whole-number variables.

Result 5 (uniform and exponential means). If X \sim \operatorname{Uniform}(a, b), then \displaystyle E(X) = \frac{a + b}{2}. If T \sim \operatorname{Exponential}(\lambda), then \displaystyle E(T) = \frac{1}{\lambda}: events at a rate of \lambda per unit of time mean an average wait of \displaystyle \frac{1}{\lambda}.

Why it is true

Result 1, discrete. g(X) takes countably many values y, and \{g(X) = y\} is the disjoint union of the events \{X = x\} with g(x) = y, so countable additivity gives P(g(X) = y) = \sum_{x : g(x) = y} p_X(x). Hence

\begin{aligned}\sum_y |y| \, P(g(X) = y) &= \sum_y \sum_{x : g(x) = y} |g(x)| \, p_X(x) \\ &= \sum_x |g(x)| \, p_X(x),\end{aligned}

because a series of nonnegative terms may be grouped in any way without changing its sum. So one side converges exactly when the other does, and then the same regrouping is allowed without absolute values, since the series converges absolutely.

Result 1, density, for monotone g. Suppose f_X is 0 outside an open interval I, and on I the function g has a continuous derivative g' > 0. Then g maps I one-to-one onto an open interval J, its inverse h has h'(y) = 1/g'(h(y)), and for y \in J, F_{g(X)}(y) = P(X \le h(y)) = F_X(h(y)). This is continuous, and by the chain rule its derivative is f_X(h(y)) \, h'(y) wherever f_X is continuous at h(y), so by Result 4 of Densities and cumulative distribution functions Y = g(X) has that density on J and 0 off it. The substitution y = g(x) gives

\int_J |y| \, f_X(h(y)) \, h'(y) \, dy = \int_I |g(x)| \, f_X(x) \, dx,

and the same without absolute values. For g' < 0 the argument runs with |h'| in place of h'. A g that is monotone on pieces is handled in Functions of a random variable; the general case needs the measure-theoretic integral and is stated here without proof.

Result 2. Let X and Y be discrete. The pair (X, Y) takes countably many values (x, y), with probabilities p(x, y) = P(X = x, Y = y), and countable additivity gives \sum_y p(x, y) = p_X(x) and \sum_x p(x, y) = p_Y(y). The discrete proof of Result 1 never used that the values of X were numbers, so it applies to the pair and g(x, y) = ax + by + c. The absolute sum is at most

\begin{aligned}&\sum_{x, y} \big(|a| |x| + |b| |y| + |c|\big) \, p(x, y) \\ &\quad = |a| \sum_x |x| p_X(x) + |b| \sum_y |y| p_Y(y) + |c|,\end{aligned}

which is finite, and the same computation without absolute values gives a E(X) + b E(Y) + c. If X has a density and a \ne 0, Result 1 with the monotone g(x) = ax + c gives E(aX + c) = a \int x f_X + c \int f_X = a E(X) + c. For a pair (X, Y) with a joint density, E(X + Y) = E(X) + E(Y) is proved in Joint distributions; for every pair it is a theorem of measure-theoretic probability, stated here without proof.

Result 3. \mathbf{1}_A is discrete with P(\mathbf{1}_A = 1) = P(A) and P(\mathbf{1}_A = 0) = P(A^c), so E(\mathbf{1}_A) = 0 \cdot P(A^c) + 1 \cdot P(A). At each outcome, \mathbf{1}_{A_1} + \dots + \mathbf{1}_{A_n} adds 1 for each event that contains it, which is the value of N. Result 2 finishes.

Result 4. Fix b > 0 and integrate by parts on [0, b] with u = x and v = F_X(x) - 1 = -P(X > x). Here v is continuous and v' = f_X except at finitely many points, so the parts formula (Integration by parts) holds on each piece and the boundary terms at the break points cancel:

\begin{aligned}\int_0^b x \, f_X(x) \, dx &= \Big[ -x \, P(X > x) \Big]_0^b + \int_0^b P(X > x) \, dx \\ &= \int_0^b P(X > x) \, dx - b \, P(X > b).\end{aligned}

If \int_0^\infty x \, f_X(x) \, dx converges, then

0 \le b \, P(X > b) = \int_b^\infty b \, f_X(x) \, dx \le \int_b^\infty x \, f_X(x) \, dx,

which tends to 0 as b \to \infty, being the tail of a convergent integral; so both sides have the same limit. If it diverges, the first display gives \int_0^b P(X > x) \, dx \ge \int_0^b x \, f_X(x) \, dx, so the tail integral diverges too. When X \ge 0, f_X is 0 on (-\infty, 0) except at isolated points, so \int_0^\infty x f_X is all of E(X).

Result 5. For the uniform, the integral is over a bounded interval:

\int_a^b \frac{x \, dx}{b - a} = \frac{b^2 - a^2}{2(b - a)} = \frac{a + b}{2}.

For t \ge 0, P(T > t) = e^{-\lambda t} (Result 6 of Densities and cumulative distribution functions), so Result 4 gives E(T) = \int_0^\infty e^{-\lambda t} \, dt, which is \displaystyle \frac{1}{\lambda}.

Worked examples

Example 1

How long a game lasts

Arjun runs a bingo hall. He models the time T, in minutes, from the first call of a game to the first bingo by the density \displaystyle f_T(t) = \frac{t}{50} for 0 \le t \le 10, and 0 otherwise. Find E(T) and compare it with the median. Then find E(T^2) and compare it with E(T)^2.

Solution

The support is bounded, so the integral converges absolutely, and

E(T) = \int_0^{10} t \cdot \frac{t}{50} \, dt = \frac{1000}{150} = \frac{20}{3} \approx 6.67.

On [0, 10], \displaystyle F_T(t) = \frac{t^2}{100}, and \displaystyle \frac{m^2}{100} = \frac12 gives the median \displaystyle m = 5\sqrt{2} \approx 7.07. The early minutes carry little probability, but it sits far from the bulk and pulls the balance point below the median (Figure 1).

T^2 is a new random variable, and g(t) = t^2 has g' > 0 on (0, 10), so Result 1 gives its mean without its distribution:

E(T^2) = \int_0^{10} t^2 \cdot \frac{t}{50} \, dt = \frac{10{,}000}{200} = 50,

while \displaystyle E(T)^2 = \frac{400}{9} \approx 44.44. Expectation passes through linear functions (Result 2) but not through squaring; the gap is the variance of T, the subject of Variance and covariance.

Example 2

A wait that is cut off

Before the session Zoe sells cards at the door. The gap T, in minutes, from one player’s arrival to the next is \operatorname{Exponential}(0.4). After each sale Zoe waits for the next player, but leaves the table after 4 minutes if nobody comes, so her wait is W = \min(T, 4). Find E(T) and E(W).

Solution

By Result 5, \displaystyle E(T) = \frac{1}{0.4} = 2.5 minutes. W is mixed: P(W = 4) = P(T \ge 4) = e^{-1.6} \approx 0.2019, and the rest of its probability lies on (0, 4) with density part 0.4 e^{-0.4t}. By the definition,

E(W) = 4 e^{-1.6} + \int_0^4 t \cdot 0.4 e^{-0.4 t} \, dt.

Integrating by parts with u = t and v = -e^{-0.4t},

\begin{aligned}\int_0^4 0.4 \, t e^{-0.4 t} \, dt &= \Big[ -t e^{-0.4t} \Big]_0^4 + \int_0^4 e^{-0.4t} \, dt \\ &= -4 e^{-1.6} + 2.5 \left(1 - e^{-1.6}\right).\end{aligned}

The terms \pm 4e^{-1.6} cancel, so

E(W) = 2.5 \left(1 - e^{-1.6}\right) \approx 1.9953.

This is Result 1 with g(t) = \min(t, 4). The cutoff shortens the mean wait by 2.5 e^{-1.6} \approx 0.5047 minutes.

Example 3

The first and last of four friends

Four of Zoe’s friends will join her at the hall. Arjun models their arrival times, in minutes after the doors open, as independent \operatorname{Uniform}(0, 15) variables U_1, U_2, U_3, U_4. Find the mean of the first arrival time S = \min(U_1, \dots, U_4) and of the last, L = \max(U_1, \dots, U_4).

Solution

The first arrival comes after t exactly when every arrival does. As for the minimum in Example 5 of Independence of events and variables, independence gives, for 0 \le t \le 15,

P(S > t) = \prod_{i=1}^{4} P(U_i > t) = \left(1 - \frac{t}{15}\right)^4,

and P(S > t) = 0 for t > 15. So F_S is continuous, with a continuous derivative except at 0 and 15, and S has a density by Result 4 of Densities and cumulative distribution functions. Since S \ge 0, Result 4 gives

E(S) = \int_0^{15} \left(1 - \frac{t}{15}\right)^4 dt = \frac{15}{5} = 3,

the shaded area in Figure 2. In the same way \displaystyle P(L \le t) = \left(\frac{t}{15}\right)^4 on [0, 15], so

E(L) = \int_0^{15} \left(1 - \frac{t^4}{15^4}\right) dt = 15 - 3 = 12.
0 3 6 9 12 15 0.5 1
E(S) = 3
P(S > t)
t
Minutes after the doors open
Figure 2The tail \displaystyle P(S > t) = \left(1 - \frac{t}{15}\right)^4 of Example 3. By Result 4 the shaded area is the mean, E(S) = 3.
Example 4

Numbers called and rows completed

Zoe’s card holds 24 different numbers from 1 to 75 and a free center square. Its five rows hold 5 numbers each, except the middle row, which holds 4 numbers and the free square. Arjun calls the 75 balls one at a time without replacement, every order equally likely. After 30 calls, find the expected number of Zoe’s numbers called and the expected number of complete rows.

Solution

For each number k on the card, let A_k be the event that ball k is among the first 30 calls. Ball k is equally likely to be in any of the 75 positions, so \displaystyle P(A_k) = \frac{30}{75}. The number of Zoe’s numbers called is N = \sum_k \mathbf{1}_{A_k}, and Result 3 gives

E(N) = 24 \cdot \frac{30}{75} = 9.6.

The events A_k are not independent: once one happens, fewer of the first 30 places are left for the other balls. N is hypergeometric, as in Counting models. Neither fact matters to Result 3.

A row with j numbers is complete when all j of its balls are among the first 30. Their positions form an ordered choice of j different positions, all (75)_j equally likely, and (30)_j of them lie among the first 30. With one indicator per row,

\begin{aligned}E(\text{complete rows}) &= 4 \cdot \frac{(30)_5}{(75)_5} + \frac{(30)_4}{(75)_4} \\ &\approx 4 (0.008257) + 0.022547 \\ &\approx 0.0556.\end{aligned}
Example 5

Every dauber back to its owner

At the end of the night, n players each hand Arjun their dauber, and he hands the daubers back in a uniformly random order, one to each player. Prove that the expected number of players who get their own dauber back is 1, whatever n is.

Solution

Let \Omega be the n! equally likely ways to hand the daubers back, and let A_i be the event that player i gets their own. In A_i dauber i is fixed and the other n - 1 go in any order, so \displaystyle P(A_i) = \frac{(n - 1)!}{n!} = \frac{1}{n}. The number of players who get their own back is N = \mathbf{1}_{A_1} + \dots + \mathbf{1}_{A_n}, so by Result 3

E(N) = n \cdot \frac{1}{n} = 1.

This is the expected number of matches promised in Counting models, where the chance of no match was shown to approach e^{-1}. The distribution of N changes with n (for n = 2, N is 0 or 2), but the mean does not, and the dependence of the A_i never entered.

Example 6

Collecting dauber colors

Each card Zoe buys comes with a free dauber whose color is equally likely to be any of 6 colors, independently from card to card. Find the expected number of different colors among her first 10 cards, and the expected number of cards she buys until she holds all 6 colors.

Solution

For each color c, let B_c be the event that color c appears among the first 10 cards. By independence \displaystyle P(B_c^c) = \left(\frac56\right)^{10}, and Result 3 gives

\begin{aligned}E(\text{colors}) &= 6 \left(1 - \left(\tfrac56\right)^{10}\right) \\ &\approx 6 (1 - 0.1615) \approx 5.031.\end{aligned}

For the second question, let G_k be the number of cards bought after she first holds k - 1 colors, up to and including the card that brings the k-th, for k = 1, \dots, 6. While she holds k - 1 colors, each new card brings a new color with probability \displaystyle p_k = \frac{7 - k}{6}, independently of earlier cards, so G_k \sim \operatorname{Geometric}(p_k) and \displaystyle E(G_k) = \frac{6}{7 - k} by Result 6 of Discrete distributions: binomial, geometric, Poisson. By Result 2,

\begin{aligned}E(G_1 + \dots + G_6) &= 6 \left(1 + \tfrac12 + \tfrac13 + \tfrac14 + \tfrac15 + \tfrac16\right) \\ &= 6 \cdot \tfrac{49}{20} = 14.7.\end{aligned}

Boundaries

A mean need not exist. The density \displaystyle \frac{1}{\pi (1 + x^2)} of Exercise 4 in Densities and cumulative distribution functions is symmetric about 0, yet \displaystyle \int_0^b \frac{x \, dx}{1 + x^2} = \frac12 \ln(1 + b^2) grows without bound, so it has no mean. A nonnegative variable may have mean \infty (Exercise 2).

Linearity is about sums. In general E(g(X)) \ne g(E(X)) (Example 1, Exercise 1), and E(XY) need not equal E(X) E(Y) (Exercise 18); for independent variables it does, as Variance and covariance shows.

Every mean in the sum must be finite. If X has the density above and Y = -X, then X + Y = 0 has mean 0, but E(X) + E(Y) has no value.

A mean is not a typical value. 9.6 numbers (Example 4) is not a possible count, and the mean does not fix the distribution: every n in Example 5 gives mean 1.

Scope. The definition here covers discrete, density and mixed variables; measure theory gives one definition for all, and these results hold there.

Where this goes next

Variance and covariance measures spread by E\big((X - \mu_X)^2\big), computed with Result 1, and finds the variance of a count from its indicators. Functions of a random variable finds the whole distribution of g(X). Joint distributions computes E(g(X, Y)) as a double integral, and Conditional expectation averages one variable given another. The The law of large numbers and the gambler's fallacy explains why the mean is the long-run average, Markov and Chebyshev inequalities turns means into bounds on probabilities, and Moment generating functions gather the means E(X^k) into M_X(t) = E(e^{tX}).

Exercises

Means and functions of a variable

  1. X has density f_X(x) = 3x^2 for 0 \le x \le 1, and 0 otherwise. Find E(X), E(X^2) and E(1/X), and compare the last with 1/E(X).
  2. X has density \displaystyle f_X(x) = \frac{3}{x^4} for x \ge 1, and 0 otherwise. Find E(X) and E(X^2), and show that E(X^3) = \infty.
  3. The time T, in minutes, until the hall’s snack counter gets its next customer is \operatorname{Exponential}(0.25). Find E(T), E(T^2) and E(e^{-T}).
  4. The number X of cards a player buys has P(X = 1) = 0.2, P(X = 2) = 0.5 and P(X = 3) = 0.3. Each card costs \$4, plus a \$3 entry fee. Find E(X), the expected amount a player spends, E(X^2) and E(1/X), and compare the last with 1/E(X).
  5. U \sim \operatorname{Uniform}(0, 20). Find E(\min(U, 12)).
  6. X \ge 0 has distribution function F_X(x) = 1 - (1 + x)^{-3} for x \ge 0. Find E(X) with Result 4, then check it with the density.
  7. X \sim \operatorname{Uniform}(a, b) has E(X) = 7 and P(X > 9) = \tfrac14. Find a and b.

Indicators and linearity

  1. A game runs to 45 calls from 75 balls. Find the expected number of Zoe’s 24 numbers called, the expected number of the 4 corner numbers of her card called, and the expected number of complete columns. Her card’s columns hold 5 numbers each, except the middle column, which holds 4 numbers and the free square.
  2. All 75 balls are called in a random order. For i = 1, \dots, 74, the numbers i and i + 1 form a neighbor pair if they are called one right after the other, in either order. Find the expected number of neighbor pairs.
  3. Each of 30 players picks a lucky number from 1 to 75, uniformly and independently. Find the expected number of pairs of players with the same lucky number.
  4. Call j is a record if its number is larger than every number called before it, so call 1 is always a record. Find the expected number of records among the first 8 calls.
  5. The first column of Zoe’s card holds 5 numbers, and all 75 balls are called in a random order. Find the expected position of the first call that hits one of those 5 numbers.

Proofs

  1. Let X \sim \operatorname{Binomial}(n, p). Prove that E(X) = np using indicators.
  2. A bag holds M balls, K of them marked, and n \le M are drawn without replacement. Prove that the expected number of marked balls drawn is \displaystyle \frac{nK}{M}.
  3. Let X have a density with f_X(c + u) = f_X(c - u) for all u, and a finite mean. Prove that E(X) = c, and explain why the finite mean is needed.
  4. Let X have a density and a finite mean. Prove that E(X) = I_+ - I_-, where I_+ = \int_0^\infty P(X > x) \, dx and I_- = \int_{-\infty}^0 P(X \le x) \, dx.
  5. Let X \ge 0 have a density, and let c > 0. Prove that E(\min(X, c)) = \int_0^c P(X > x) \, dx.
  6. Refute the claim that whenever X and Y have finite means, E(XY) = E(X) E(Y).
  7. In Example 5 with n \ge 2, prove that E(N(N - 1)) = 1, and deduce E(N^2).
Answers
  1. E(X) = \tfrac34, E(X^2) = \tfrac35, E(1/X) = \tfrac32, while 1/E(X) = \tfrac43. By Result 1 these are \int_0^1 3x^3 \, dx, \int_0^1 3x^4 \, dx and \int_0^1 3x \, dx.
  2. E(X) = \tfrac32 and E(X^2) = 3, from \int_1^\infty 3x^{-3} \, dx and \int_1^\infty 3x^{-2} \, dx. For X^3 \ge 0, \int_1^b x^3 \cdot 3x^{-4} \, dx = 3 \ln b \to \infty, and the substitution in the proof of Result 1 equates this integral with the one for the mean of X^3.
  3. E(T) = 4, E(T^2) = 32, E(e^{-T}) = \tfrac15. Result 5 gives \displaystyle \frac{1}{0.25}. By parts, \int_0^\infty t^2 \lambda e^{-\lambda t} \, dt equals \displaystyle \int_0^\infty 2t e^{-\lambda t} \, dt = \frac{2}{\lambda^2}, here 32. Last, \displaystyle \int_0^\infty 0.25 e^{-1.25 t} \, dt = \frac{0.25}{1.25}.
  4. E(X) = 2.1; spending 4X + 3 has mean 4(2.1) + 3 = \$11.40 by Result 2; E(X^2) = 0.2 + 2 + 2.7 = 4.9; E(1/X) = 0.2 + 0.25 + 0.1 = 0.55, while 1/E(X) \approx 0.4762.
  5. 8.4. The variable is mixed, with mass P(U > 12) = 0.4 at 12 and density part \displaystyle \frac{1}{20} on (0, 12): \displaystyle \int_0^{12} \frac{u}{20} \, du + 12(0.4) = 3.6 + 4.8. Exercise 17 agrees: \displaystyle \int_0^{12} \left(1 - \frac{u}{20}\right) du = 12 - 3.6.
  6. E(X) = \tfrac12. Result 4 gives \int_0^\infty (1 + x)^{-3} \, dx = \tfrac12. The density is 3(1 + x)^{-4}, and with u = 1 + x, \int_1^\infty (u - 1) \, 3u^{-4} \, du = 3 \left(\tfrac12 - \tfrac13\right).
  7. a = 3, b = 11. Result 5 gives a + b = 14, and \displaystyle \frac{b - 9}{b - a} = \frac14 gives 3b + a = 36; subtracting, 2b = 22.
  8. 14.4; 2.4; about 0.4057. Each number is called with probability \displaystyle \frac{45}{75} = 0.6, giving 24(0.6) and 4(0.6); columns, as in Example 4, give \displaystyle 4 \cdot \frac{(45)_5}{(75)_5} + \frac{(45)_4}{(75)_4} \approx 4(0.07079) + 0.12258.
  9. \displaystyle \frac{148}{75} \approx 1.973. Balls i and i + 1 occupy a uniformly random pair of the \binom{75}{2} = 2775 pairs of positions, 74 of which are adjacent, so the chance is \displaystyle \frac{74}{2775} = \frac{2}{75}; Result 3 gives \displaystyle 74 \cdot \frac{2}{75}.
  10. \displaystyle \frac{29}{5} = 5.8. Two given players share with probability \displaystyle \sum_{k=1}^{75} \left(\frac{1}{75}\right)^2 = \frac{1}{75} by independence, and there are \binom{30}{2} = 435 pairs: \displaystyle \frac{435}{75}.
  11. \displaystyle \frac{761}{280} \approx 2.718. The largest of the first j balls is equally likely to be any of them, so call j is a record with probability \displaystyle \frac{1}{j}, and Result 3 gives \displaystyle 1 + \frac12 + \dots + \frac{1}{8}.
  12. \displaystyle \frac{38}{3} \approx 12.67. The position is 1 plus the number of the other 70 balls called before all 5. A given other ball comes first among itself and the 5 with probability \displaystyle \frac16, so the mean is \displaystyle 1 + \frac{70}{6}.
  13. The mean depends only on the distribution, so let X count successes in n independent trials of success probability p. Let A_i be success on trial i, so P(A_i) = p and X = \mathbf{1}_{A_1} + \dots + \mathbf{1}_{A_n}. Result 3 gives E(X) = np; independence is not used for the mean.
  14. Let A_i be the event that draw i is marked. The ordered draws are equally likely to be any of the (M)_n sequences of different balls, and K \cdot (M - 1)_{n - 1} of them have a marked ball in place i: choose that ball, then fill the other places in order. So \displaystyle P(A_i) = \frac{K}{M}, and Result 3 gives \displaystyle \frac{nK}{M}, the mean of the hypergeometric distribution of Counting models.
  15. Since \int f_X = 1, E(X) - c = \int (x - c) f_X(x) \, dx, and x = c + u splits this into \int_0^\infty u f_X(c + u) \, du and \int_{-\infty}^0 u f_X(c + u) \, du. Putting u = -v in the second and using the symmetry, it is minus the first. Both are finite, being at most \int |x| f_X + |c|, so they cancel and E(X) = c. Without a finite mean both pieces can be infinite, and \infty - \infty has no value; the density \displaystyle \frac{1}{\pi(1 + x^2)} is such a case.
  16. Split E(X) into \int_0^\infty x f_X(x) \, dx and \int_{-\infty}^0 x f_X(x) \, dx, both finite. Result 4 turns the first into I_+. Y = -X has F_Y(y) = 1 - F_X(-y), so it has the density f_X(-y) by Result 4 of Densities and cumulative distribution functions. Result 4 for Y, then x = -y, gives -\int_{-\infty}^0 x f_X(x) \, dx = \int_0^\infty P(X < -y) \, dy = I_-, since P(X < x) = P(X \le x) by Result 1 of Densities and cumulative distribution functions.
  17. If P(X > c) = 0, then \min(X, c) = X with probability 1, and this is Result 4, as P(X > x) = 0 for x \ge c. Otherwise \min(X, c) is mixed, with mass P(X > c) at c and density part f_X on (0, c), so its mean is \int_0^c x f_X(x) \, dx + c \, P(X > c). The integration by parts in the proof of Result 4, with b = c, turns the integral into \int_0^c P(X > x) \, dx - c \, P(X > c), and the terms \pm c \, P(X > c) cancel.
  18. Let A be an event with \displaystyle P(A) = \frac12, say Zoe’s first card wins, and X = Y = \mathbf{1}_A. Then XY = \mathbf{1}_A, so \displaystyle E(XY) = \frac12, while \displaystyle E(X) E(Y) = \frac14.
  19. If N = m, then m(m - 1) counts the ordered pairs (i, j), i \ne j, of players who both get their own, so N(N - 1) = \sum_{i \ne j} \mathbf{1}_{A_i \cap A_j}. Two daubers fixed gives \displaystyle P(A_i \cap A_j) = \frac{(n - 2)!}{n!} = \frac{1}{n(n - 1)}, and there are n(n - 1) ordered pairs, so Result 3 gives E(N(N - 1)) = 1. Since N^2 = N(N - 1) + N, Result 2 gives E(N^2) = 1 + 1 = 2, which Variance and covariance turns into a variance of 1.
math.wiki · CC-BY-SA-4.0