Independence of events and variables
When does learning that one thing happened tell us nothing about another, and how do we say that for a whole list of events, or for two random quantities at every one of their values at once?
Where you are 4 in Foundations
Why this exists
The school lesson Independent events and the multiplication rule multiplies the probabilities of stages that cannot affect each other, such as a coin and a number cube. Conditioning and Bayes' rule, rigorously gives the idea a precise form: learning B tells us nothing about A when P(A\mid B)=P(A). Four questions remain. What does independence mean when P(B)=0 and P(A\mid B) is undefined? When are three or more events independent, and is it enough to check them in pairs? What does it mean for two tests to be independent “given that the package is counterfeit”, as Example 5 of Conditioning and Bayes' rule, rigorously assumed? And what does it mean for two random variables, where learning Y means learning one of many possible values?
Independence is rarely deduced inside a model; it is how most models are built. The binomial count of Discrete distributions: binomial, geometric, Poisson rests on independent trials, and every repeated measurement and sample in this course rests on independent random variables. This lesson defines all four notions with one device, a product of probabilities, and proves the facts that make them usable.
Definition
Throughout, (\Omega,P) is a probability space.
Two events. Events A and B are independent when
Otherwise they are dependent. The product form needs no condition on P(B) and treats A and B alike; Result 1 links it to P(A\mid B)=P(A).
Many events. Events A_1,\dots,A_n are (mutually) independent when the product rule holds for every subcollection: for every choice of k\ge 2 indices i_1<\dots<i_k,
That is 2^n-n-1 equations. An infinite family of events is independent when every finite subfamily is. Events are pairwise independent when every two of them are independent, the case k=2 only.
Conditional independence. Let P(C)>0. Events are conditionally independent given C when they are independent for the probability P(\cdot\mid C), which is a probability by Result 1 of Conditioning and Bayes' rule, rigorously. For two events this reads P(A\cap B\mid C)=P(A\mid C)\,P(B\mid C).
Random variables. Random variables X and Y on \Omega are independent when, for all real x and y,
A comma inside P(\ ) means “and”: P(X\le x,\ Y\le y)=P(\{X\le x\}\cap\{Y\le y\}), the joint distribution function of Joint distributions. Random variables X_1,\dots,X_n are independent when P(X_1\le x_1,\dots,X_n\le x_n)=F_{X_1}(x_1)\cdots F_{X_n}(x_n) for all real x_1,\dots,x_n, and a sequence X_1,X_2,\dots is independent when every finite collection of its terms is. Variables that are independent and share one distribution function are independent and identically distributed, iid for short. Random variables are conditionally independent given C when they are independent for P(\cdot\mid C).
No list of subcollections is needed for variables: letting x_i grow without bound turns F_{X_i}(x_i) into 1 and the event \{X_i\le x_i\} into \Omega, which drops X_i from both sides (continuity, Result 8 of Probability axioms and sample spaces).
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 otherwise.
Intuition
When P(B)>0, the equation P(A\cap B)=P(A)P(B) says P(A\cap B)/P(B)=P(A): A takes the same share of B as it takes of \Omega. Learning B rescales the probabilities on B, and independence says that this leaves the chance of A where it was.
For two discrete variables, draw a unit square and give X a column for each value i, as wide as P(X=i). Split each column across in the proportions P(Y=j\mid X=i), so that each piece has area P(X=i,\ Y=j). Independence says that every column is split in the same proportions, those of Y alone, so the cuts run straight across the square (Figure 1, left). Cuts that step up and down between columns mean that knowing X changes the chances for Y (right). One column cut at the common height, like X=0 on the right, makes one pair of events independent but not the variables.
The product also explains why variables need one equation where events need many. The definition for variables holds for every x, including x so large that \{X\le x\} is nearly all of \Omega, so it already contains every subcollection; for events, the subcollections must be listed. Result 6 makes the link exact.
Results
Result 1 (two events). If P(B)>0, then A and B are independent exactly when P(A\mid B)=P(A). An event of probability 0 or 1 is independent of every event, itself included.
Result 2 (complements). If A_1,\dots,A_n are independent, they stay independent when any of them are replaced by their complements. In particular
Result 3 (rectangles). Random variables X and Y are independent exactly when, for all a<b and c<d,
Then P(X\in I,\ Y\in J)=P(X\in I)\,P(Y\in J) for all intervals I and J. The same holds for every set of reals met in this course (every Borel set), but that needs measure theory and is not proved here.
Result 4 (discrete variables). Let X and Y be discrete with supports S_X and S_Y. They are independent exactly when, for all i and j,
and then P(X\in A,\ Y\in B)=P(X\in A)\,P(Y\in B) for all sets A and B of reals. The same holds for X_1,\dots,X_n, with n values and n sets.
Result 5 (functions of independent variables). If discrete X and Y are independent, then g(X) and h(Y) are independent for any functions g and h. More generally, if discrete X_1,\dots,X_n are independent, then functions of disjoint groups of them, such as g(X_1,X_2) and h(X_3,\dots,X_n), are independent. The same holds for any random variables and every function met in this course (Borel functions), again by measure theory.
Result 6 (indicators). Events A_1,\dots,A_n are independent exactly when the random variables \mathbf{1}_{A_1},\dots,\mathbf{1}_{A_n} are independent.
Why it is true
Result 1. If P(B)>0, divide P(A\cap B)=P(A)P(B) by P(B). If P(B)=0, monotonicity gives 0\le P(A\cap B)\le P(B)=0, so both sides are 0. If P(B)=1, then P(A\cap B^c)\le P(B^c)=0, so P(A\cap B)=P(A)-P(A\cap B^c)=P(A)=P(A)P(B).
Result 2. It is enough to replace one event, say A_1, by A_1^c; repeating the step replaces as many as we like. A subcollection without A_1 is unchanged. For one containing it, let D be the intersection of its other events, so P(D) is the product q of their probabilities. D splits into A_1\cap D and A_1^c\cap D, so
For the union, De Morgan’s law gives (A_1\cup\dots\cup A_n)^c=A_1^c\cap\dots\cap A_n^c, whose probability is \prod_i\bigl(1-P(A_i)\bigr) by what was just proved.
Result 3. Write G(x,y)=P(X\le x,\ Y\le y). The event E=\{X\le b,\ Y\le d\} contains U=\{X\le a,\ Y\le d\} and V=\{X\le b,\ Y\le c\}, with U\cap V=\{X\le a,\ Y\le c\}, and the outcomes of E outside U\cup V are exactly those with a<X\le b and c<Y\le d. Monotonicity and the addition rule (Results 4 and 5 of Probability axioms and sample spaces) give
If G(x,y)=F_X(x)F_Y(y), the right side factors as stated. Conversely, take a=c=-m and let m\to\infty: the events \{-m<X\le b,\ -m<Y\le d\} grow to \{X\le b,\ Y\le d\} while F_X(-m) and F_Y(-m) tend to 0, so continuity gives G(b,d)=F_X(b)F_Y(d). Every other interval is a growing union or a shrinking intersection of intervals (a,b]; for instance [b,b]=\{b\} is the intersection of the intervals (b-\tfrac{1}{m},b]. Applying continuity to both sides, one interval at a time, carries the product to X\in I and Y\in J.
Result 4. If X and Y are independent, Result 3 with I=[i,i] and J=[j,j] gives the product for the pmf. Conversely, suppose the pmf factors. The events \{X=i,\ Y=j\} for i\in A\cap S_X and j\in B\cap S_Y are countably many and disjoint, and their union differs from \{X\in A,\ Y\in B\} only by outcomes with X\notin S_X or Y\notin S_Y, an event of probability 0. Countable additivity gives
The double sum factors because a sum of nonnegative terms may be taken in any order and grouping (Result 7 of Probability axioms and sample spaces), and each factor is Result 1 of Discrete distributions: binomial, geometric, Poisson. Taking A=(-\infty,x] and B=(-\infty,y] gives the definition. With n variables the sums run over n indices.
Result 5. g(X) and h(Y) are discrete. For values u and v, let A=\{x: g(x)=u\} and B=\{y: h(y)=v\}. Then \{g(X)=u\}=\{X\in A\} and \{h(Y)=v\}=\{Y\in B\}, so the set form of Result 4 gives P(g(X)=u,\ h(Y)=v)=P(g(X)=u)\,P(h(Y)=v), and Result 4, read the other way, makes g(X) and h(Y) independent. For groups, the same argument runs with A and B sets of tuples and sums over tuples of values.
Result 6. The indicator \mathbf{1}_{A_i} takes the values 0 and 1, with \{\mathbf{1}_{A_i}=1\}=A_i and \{\mathbf{1}_{A_i}=0\}=A_i^c. If the events are independent, Result 2 says that for every choice of B_i\in\{A_i,A_i^c\}, P(B_1\cap\dots\cap B_n)=P(B_1)\cdots P(B_n); that is the joint pmf of the indicators factoring, so they are independent by Result 4. Conversely, if the indicators are independent, the set form of Result 4, with the set \{1\} for each index in a chosen subcollection and the whole real line for the others, gives P(A_{i_1}\cap\dots\cap A_{i_k})=P(A_{i_1})\cdots P(A_{i_k}).
Worked examples
A fairing and a tire
From Elena’s hangar records, an aircraft arriving for an overnight check has a cracked fairing (event A) with probability 0.15 and needs a tire change (event B) with probability 0.4, and P(A\cup B)=0.49. Are A and B independent? Find P(A\mid B) and the probability that neither happens. What changes if instead P(A\cup B)=0.5?
SolutionThe addition rule gives P(A\cap B)=0.15+0.4-0.49=0.06, and P(A)P(B)=0.15\cdot 0.4=0.06. The two agree, so A and B are independent, and by Result 1 P(A\mid B)=P(A)=0.15: a tire change says nothing about the fairing. By Result 2, A^c and B^c are independent too, so P(A^c\cap B^c)=0.85\cdot 0.6=0.51; directly, it is 1-P(A\cup B)=1-0.49=0.51, as it must be.
With P(A\cup B)=0.5 instead, P(A\cap B)=0.05, which is not 0.06, so the events are dependent, and P(A\mid B)=\tfrac{0.05}{0.4}=0.125, below 0.15. Independence is one equation between numbers, and a change of 0.01 in one probability breaks it.
Reading a joint table
At each check Mateo records the number X of tires changed and whether the brakes were also replaced (Y=1) or not (Y=0). His records give the joint pmf below. Find the distributions of X and Y, decide whether X and Y are independent, and decide whether the events \{X=0\} and \{Y=1\} are independent.
| X=0 | X=1 | X=2 | |
|---|---|---|---|
| Y=1 | 0.20 | 0.06 | 0.14 |
| Y=0 | 0.30 | 0.24 | 0.06 |
Adding columns, P(X=0)=0.5, P(X=1)=0.3 and P(X=2)=0.2; adding rows, P(Y=1)=0.4 and P(Y=0)=0.6, the margins of Joint and marginal probability. By Result 4, independence needs all six cells to equal the products of their margins. The cell X=1, Y=1 fails, since 0.06\ne 0.3\cdot 0.4=0.12, so X and Y are dependent. The conditional probabilities show how: P(Y=1\mid X=i) is \tfrac{0.20}{0.5}=0.4, \tfrac{0.06}{0.3}=0.2 and \tfrac{0.14}{0.2}=0.7 for i=0,1,2, the cut heights in Figure 1, right.
Yet P(X=0,\ Y=1)=0.20=0.5\cdot 0.4, so \{X=0\} and \{Y=1\} are independent, and by Result 2 so are \{X=0\} and \{Y=0\}, \{X\ge 1\} and \{Y=1\}, and \{X\ge 1\} and \{Y=0\}. One independent cell is not enough; Result 4 needs every cell. Independent variables with the same margins would have 0.20, 0.12, 0.08 in the top row and 0.30, 0.18, 0.12 below (Figure 1, left).
Two aircraft, three events
Mateo sends each of two arriving aircraft to the east bay or the west bay by a fair coin toss, the tosses independent. Let A be “the first goes east”, B “the second goes east” and C “both go to the same bay”. Show that A, B, C are pairwise independent but not independent. Then, if each aircraft goes east with probability p instead, find every p for which A and C are independent.
SolutionBy the independence of the tosses, the four outcomes EE, EW, WE, WW each have probability \tfrac{1}{4}. So P(A)=P(B)=\tfrac{1}{2}, and C=\{\text{EE},\text{WW}\} has P(C)=\tfrac{1}{2}. Each of A\cap B, A\cap C and B\cap C is the single outcome EE, of probability \tfrac{1}{4}=\tfrac{1}{2}\cdot\tfrac{1}{2}, so every pair is independent. But A\cap B\cap C is also EE, and P(A\cap B\cap C)=\tfrac{1}{4}\ne\tfrac{1}{8}=P(A)P(B)P(C). Learning A and B together settles C, though each alone says nothing about it.
With probability p, P(A)=p, P(C)=p^2+(1-p)^2 and P(A\cap C)=P(\text{EE})=p^2. Independence of A and C means
which rearranges to p\,(2p-1)(p-1)=0. So p=0, p=\tfrac{1}{2} or p=1. The first and last are the trivial case of Result 1; among 0<p<1, only the fair coin makes A and C independent. Independence is a fact about the numbers, not about how the events are described.
A joint pmf that factors
(a) Let X and Y be discrete with P(X=i,\ Y=j)=g(i)\,h(j) for all i in a countable set S and j in a countable set T, where g,h\ge 0, and P(X=i,\ Y=j)=0 for every other pair. Prove that X and Y are independent. (b) The numbers X and Y of rivets rejected in the trays Elena and Mateo inspect have P(X=i,\ Y=j)=c\left(\tfrac{1}{3}\right)^i\left(\tfrac{1}{4}\right)^j for i,j=0,1,2,\dots. Find c, the pmf of each, and P(X=Y).
Solution(a) Write G=\sum_{i\in S}g(i) and H=\sum_{j\in T}h(j). The joint probabilities total 1, and a double series of nonnegative terms may be summed in any order, so GH=1. Summing over j gives P(X=i)=g(i)H for i\in S, and summing over i gives P(Y=j)=G\,h(j) for j\in T. Hence
for i\in S and j\in T, and both sides are 0 for other pairs. By Result 4, X and Y are independent. No constant needs to be found first: a split of the joint pmf into a factor in i times a factor in j, on a product set S\times T, settles it.
(b) Here g(i)=c\left(\tfrac{1}{3}\right)^i and h(j)=\left(\tfrac{1}{4}\right)^j on the product set of pairs of whole numbers, so X and Y are independent. The geometric series give \sum_i\left(\tfrac{1}{3}\right)^i=\tfrac{3}{2} and \sum_j\left(\tfrac{1}{4}\right)^j=\tfrac{4}{3}, so 2c=1 and c=\tfrac{1}{2}. Then P(X=i)=\tfrac{1}{2}\cdot\tfrac{4}{3}\left(\tfrac{1}{3}\right)^i=\tfrac{2}{3}\left(\tfrac{1}{3}\right)^i and P(Y=j)=\tfrac{3}{4}\left(\tfrac{1}{4}\right)^j. The events \{X=k,\ Y=k\} are disjoint, so
Waiting for both, waiting for one
An aircraft leaves the hangar when Elena’s avionics check and Mateo’s engine check are both done. Their times in hours, X and Y, are independent, with F_X(x)=\tfrac{x}{4} for 0\le x\le 4 (0 before, 1 after) and F_Y(y)=1-e^{-y/2} for y\ge 0 (0 before). Find the distribution function of the release time T=\max(X,Y), then P(T\le 3), and the probability that at least one check is done within the first hour.
SolutionT\le t exactly when X\le t and Y\le t, so the definition of independence gives F_T(t)=F_X(t)\,F_Y(t) directly:
So P(T\le 3)=\tfrac{3}{4}\bigl(1-e^{-3/2}\bigr)\approx 0.583. No check is done within the first hour exactly when \min(X,Y)>1, that is, when X>1 and Y>1. The events \{X\le 1\} and \{Y\le 1\} are independent by the definition, so their complements are too, by Result 2:
and the answer is 1-\tfrac{3}{4}e^{-1/2}\approx 0.545. Only distribution functions were used; Densities and cumulative distribution functions gives such times densities, and Joint distributions treats independence through joint densities.
Pads from one box
A box of brake pads comes from supplier S_1 with probability 0.6 and from supplier S_2 with probability 0.4. A pad from S_1 is defective with probability 0.05, one from S_2 with probability 0.2, and the pads in one box are conditionally independent given the supplier. Mateo fits two pads from one box; let D_i be “the ith pad is defective”. Are D_1 and D_2 independent? Find P(D_2\mid D_1).
SolutionBy the The law of total probability, P(D_1)=0.6\cdot 0.05+0.4\cdot 0.2=0.11, and P(D_2)=0.11 in the same way. Conditional independence makes P(D_1\cap D_2\mid S_k) the product of the two conditional probabilities, so
Since 0.11^2=0.0121, the events are dependent, and
above P(D_2)=0.11. The supplier is the reason. By Bayes’ rule P(S_2\mid D_1)=\tfrac{0.08}{0.11}=\tfrac{8}{11}, against 0.4 before, and the second pad is judged under that posterior: \tfrac{3}{11}\cdot 0.05+\tfrac{8}{11}\cdot 0.2=\tfrac{7}{44} again. The repeated tests of Example 5 in Conditioning and Bayes' rule, rigorously behave the same way: independent given counterfeit and given genuine, yet dependent under P, since each positive raises the chance of the hypothesis that makes the next positive likely. Conditional independence and independence are different properties, and neither implies the other (Exercise 7 runs the other way).
Boundaries
- Independent is not disjoint. Disjoint events with positive probabilities are never independent, since P(A\cap B)=0<P(A)P(B). Learning that one happened rules the other out, which is as far from “tells us nothing” as it gets.
- Pairs are not enough, and neither is one product. Example 3 has three pairwise independent events that are not independent, and Exercise 18 has P(A\cap B\cap C)=P(A)P(B)P(C) with two of the events dependent. Independence of n events needs all 2^n-n-1 equations.
- Every value. Two variables are independent only if the product holds at every pair of values; one independent pair of events, as in Example 2, is not enough. A joint pmf that is constant on a set that is not a product set, such as the pairs with i<j, never gives independent variables (Exercise 15).
- A property of P. Independence is about the probability, not the events. Conditioning can create or destroy it (Example 6, Exercise 7), and in Example 3 the bias of the coin decides it.
- Separation is a modeling claim. Draws without replacement are dependent, as in the hypergeometric counts of Counting models, and that tests on one package or pads from one box are independent is an assumption, often true only given the hidden cause.
- Beyond intervals. Results 3 and 5 are proved here only for intervals and for discrete variables; the general versions need measure theory. Two variables with zero covariance can still be dependent, as Variance and covariance shows.
Where this goes next
Most models in the rest of the course are declared through independence. Joint distributions shows that independent variables with densities have a joint density equal to the product of the two, and conditions one variable on another. Indicators carry counting problems into Expectation, linearity and indicators. For independent variables the expectation of a product is the product of the expectations, so variances add (Variance and covariance, school version Combining random variables). The distribution of a sum of independent variables is a convolution (Sums of independent variables), as in Exercise 16 of Discrete distributions: binomial, geometric, Poisson, and its moment generating function is a product (Moment generating functions). Averages of iid variables settle down to their mean in The law of large numbers.
Exercises
Independent events
- Events A and B are independent with P(A)=0.3 and P(B)=0.7. Find P(A\cap B), P(A\cup B), P(A\cap B^c) and P(A^c\mid B).
- A number N is chosen uniformly from \{1,2,\dots,12\} to pick a tool drawer. Let A be “N is even”, B “N is a multiple of 3” and C “N is a multiple of 4”. Which of the three pairs are independent? Are A, B, C independent?
- Events A and B are independent, P(A)=0.6 and P(A\cup B)=0.8. Find P(B).
- A part passes four independent checks with probabilities 0.9, 0.95, 0.8 and 0.85. Find the probability that it passes all four, that it fails at least one, and that it fails exactly one.
- Each of several independent inspections misses a given crack with probability 0.3. How many inspections are needed for the probability that all of them miss it to be at most 0.001?
Conditional independence
- A fuel tank is contaminated with probability 0.1. Each of two tests on samples from the tank is positive with probability 0.9 if it is contaminated and 0.2 if not, the tests conditionally independent given each state. Let T_i be “test i is positive”. Find P(T_1), P(T_1\cap T_2) and P(T_2\mid T_1), and decide whether T_1 and T_2 are independent.
- Elena and Mateo each take a certification check and pass independently, with probabilities 0.8 and 0.7. Let E and M be their passing, and C “exactly one of them passes”. Find P(E\mid C), P(M\mid C) and P(E\cap M\mid C), and decide whether E and M are conditionally independent given C.
Independent random variables
- The number X of engines on an aircraft in for a check and the number Y of panels replaced have P(X=1,\ Y=j) equal to 0.12, 0.18, 0.30 and P(X=2,\ Y=j) equal to 0.08, 0.12, 0.20 for j=0,1,2. Are X and Y independent? Find P(X+Y=3).
- X takes values in \{0,1\} and Y in \{0,1,2\}, they are independent, and P(X=0,\ Y=0)=0.1, P(X=0,\ Y=1)=0.15 and P(X=1,\ Y=0)=0.15. Complete the joint pmf.
- X\sim\operatorname{Geometric}\left(\tfrac{1}{3}\right) and Y\sim\operatorname{Geometric}\left(\tfrac{1}{4}\right) are independent. Find P(\min(X,Y)>k) for k=0,1,2,\dots, name the distribution of \min(X,Y), and find P(X=Y).
- X and Y are independent, each equal to -1, 0 or 1 with probability \tfrac{1}{3} each. Let Z=XY. Find the pmf of Z. Are X and Z independent? Are |X| and Y?
- X and Y are independent, with F_X(x)=x^2 for 0\le x\le 1 and F_Y(y)=y for 0\le y\le 1 (each 0 below and 1 above). Find P(\max(X,Y)\le\tfrac{1}{2}), P(\min(X,Y)>\tfrac{1}{2}) and P(\tfrac{1}{4}<X\le\tfrac{3}{4},\ Y>\tfrac{1}{3}).
- X_1,\dots,X_5 are iid with F(x)=1-e^{-x} for x\ge 0 and 0 before. Find the distribution function of \max(X_1,\dots,X_5), then P(\max(X_1,\dots,X_5)\le 2) and P(\min(X_1,\dots,X_5)>0.2).
- (a) P(X=i,\ Y=j)=c\,ij for i\in\{1,2,3\} and j\in\{1,2\}. Find c and the pmfs of X and Y, decide whether they are independent, and find P(X>Y). (b) Find c and decide independence when instead P(X=i,\ Y=j)=c\,(i+j) on the same pairs.
- Mateo picks 2 of the bays numbered 1 to 4 at random; X is the smaller number and Y the larger, so P(X=i,\ Y=j)=\tfrac{1}{6} whenever 1\le i<j\le 4. Find the pmfs of X and Y and show that they are dependent, though the joint pmf is constant where it is positive.
Proofs
- Prove that if A and B are disjoint and independent, then P(A)=0 or P(B)=0.
- Prove that A is independent of itself exactly when P(A) is 0 or 1.
- Let N be uniform on \{1,\dots,8\}, let A=B=\{N\le 4\} and let C=\{N\in\{1,5,6,7\}\}. Show that P(A\cap B\cap C)=P(A)P(B)P(C) although A and B are dependent.
- Prove that if A, B, C are independent, then A\cup B and C are independent.
- Prove that if A is independent of B and of C, and B and C are disjoint, then A and B\cup C are independent.
- Let X and Y be independent and discrete. Prove that P(X=Y)=\sum_k P(X=k)P(Y=k), the sum over the support of X. If X and Y are iid, prove that P(X<Y)=\tfrac{1}{2}\bigl(1-P(X=Y)\bigr).
- Prove that a random variable X independent of itself is constant: there is a c with P(X=c)=1.
Answers
- P(A\cap B)=0.21, P(A\cup B)=0.79, P(A\cap B^c)=0.09, P(A^c\mid B)=0.7. The first is 0.3\cdot 0.7; the addition rule gives 0.3+0.7-0.21; A and B^c are independent by Result 2, so 0.3\cdot 0.3; and P(A^c\mid B)=P(A^c) by Results 1 and 2.
- A and B are independent: P(A\cap B)=P(N\in\{6,12\})=\tfrac{1}{6}=\tfrac{1}{2}\cdot\tfrac{1}{3}. B and C are independent: P(N=12)=\tfrac{1}{12}=\tfrac{1}{3}\cdot\tfrac{1}{4}. A and C are dependent: C\subseteq A, so P(A\cap C)=\tfrac{1}{4}\ne\tfrac{1}{8}. So A, B, C are not independent.
- P(B)=0.5. The addition rule and independence give 0.8=0.6+P(B)-0.6P(B), so 0.4P(B)=0.2.
- 0.5814, 0.4186 and 0.34315. By Result 2 every mix of passes and failures multiplies: all pass is 0.9\cdot 0.95\cdot 0.8\cdot 0.85=0.5814, at least one failure is 1-0.5814, and exactly one failure is 0.0646+0.0306+0.14535+0.1026, one term for each check that can be the failure.
- 6 inspections. We need 0.3^n\le 0.001, and 0.3^5=0.00243 while 0.3^6=0.000729.
- P(T_1)=0.1\cdot 0.9+0.9\cdot 0.2=0.27; P(T_1\cap T_2)=0.1\cdot 0.81+0.9\cdot 0.04=0.117; P(T_2\mid T_1)=\tfrac{0.117}{0.27}=\tfrac{13}{30}\approx 0.433. Dependent, since 0.27^2=0.0729.
- P(C)=0.8\cdot 0.3+0.2\cdot 0.7=0.38, so P(E\mid C)=\tfrac{0.24}{0.38}=\tfrac{12}{19} and P(M\mid C)=\tfrac{0.14}{0.38}=\tfrac{7}{19}, while P(E\cap M\mid C)=0 because E\cap M\cap C is empty. Not conditionally independent, since \tfrac{12}{19}\cdot\tfrac{7}{19}=\tfrac{84}{361}. Independent events can be dependent given C.
- Independent. The margins are P(X=1)=0.6, P(X=2)=0.4, P(Y=0)=0.2, P(Y=1)=0.3, P(Y=2)=0.5, and each of the six cells is the product of its margins, for example 0.6\cdot 0.5=0.30. Then P(X+Y=3)=0.30+0.12=0.42, from the cells (1,2) and (2,1).
- P(Y=0)=0.1+0.15=0.25, so P(X=0)=\tfrac{0.1}{0.25}=0.4 and P(X=1)=0.6; P(Y=1)=\tfrac{0.15}{0.4}=0.375, so P(Y=2)=0.375. The missing cells are P(X=0,\ Y=2)=0.15, P(X=1,\ Y=1)=0.225 and P(X=1,\ Y=2)=0.225.
- P(\min(X,Y)>k)=\left(\tfrac{2}{3}\right)^k\left(\tfrac{3}{4}\right)^k=\left(\tfrac{1}{2}\right)^k, so \min(X,Y)\sim\operatorname{Geometric}\left(\tfrac{1}{2}\right), and P(X=Y)=\tfrac{1}{6}. The minimum exceeds k when both do; Result 4 multiplies the geometric tails of Discrete distributions: binomial, geometric, Poisson. Then P(X=Y)=\sum_{k\ge 1}\tfrac{1}{3}\cdot\tfrac{1}{4}\left(\tfrac{1}{2}\right)^{k-1}=\tfrac{1}{12}\cdot 2.
- P(Z=1)=P(Z=-1)=\tfrac{2}{9} and P(Z=0)=\tfrac{5}{9}. X and Z are dependent: P(X=0,\ Z=1)=0 but P(X=0)P(Z=1)=\tfrac{2}{27}. |X| and Y are independent by Result 5.
- \tfrac{1}{8}, \tfrac{3}{8} and \tfrac{1}{3}. The first is F_X(\tfrac{1}{2})F_Y(\tfrac{1}{2})=\tfrac{1}{4}\cdot\tfrac{1}{2}; by Result 2 the second is \left(1-\tfrac{1}{4}\right)\left(1-\tfrac{1}{2}\right); by Result 3 the third is \left(\tfrac{9}{16}-\tfrac{1}{16}\right)\left(1-\tfrac{1}{3}\right).
- \left(1-e^{-t}\right)^5 for t\ge 0 and 0 before; \left(1-e^{-2}\right)^5\approx 0.4833; \left(e^{-0.2}\right)^5=e^{-1}\approx 0.3679. The maximum is at most t when all five are, and the minimum exceeds 0.2 when all five do, by Result 2 for the events \{X_i\le 0.2\}.
- (a) c=\tfrac{1}{18}, since the values ij total 6\cdot 3=18; P(X=i)=\tfrac{i}{6} and P(Y=j)=\tfrac{j}{3}; independent by Example 4; P(X>Y)=\tfrac{2+3+6}{18}=\tfrac{11}{18}, from the pairs (2,1), (3,1), (3,2). (b) c=\tfrac{1}{21}, and dependent: P(X=1,\ Y=1)=\tfrac{2}{21} but P(X=1)P(Y=1)=\tfrac{5}{21}\cdot\tfrac{9}{21}=\tfrac{5}{49}.
- P(X=1)=\tfrac{1}{2}, P(X=2)=\tfrac{1}{3}, P(X=3)=\tfrac{1}{6}, and P(Y=2)=\tfrac{1}{6}, P(Y=3)=\tfrac{1}{3}, P(Y=4)=\tfrac{1}{2}. Dependent: P(X=3,\ Y=2)=0 but P(X=3)P(Y=2)=\tfrac{1}{36}. The positive cells form a triangle, not a product set, so Example 4 does not apply.
- P(A)P(B)=P(A\cap B)=P(\varnothing)=0, and a product of two real numbers is 0 only when one factor is.
- Independence of A with itself says P(A)=P(A\cap A)=P(A)^2, that is, P(A)\bigl(1-P(A)\bigr)=0, so P(A) is 0 or 1. Conversely, Result 1 makes an event of probability 0 or 1 independent of every event, itself included.
- P(A)=P(B)=P(C)=\tfrac{1}{2}, and A\cap B\cap C=\{N=1\} has probability \tfrac{1}{8}=\left(\tfrac{1}{2}\right)^3. But A\cap B=A, so P(A\cap B)=\tfrac{1}{2}\ne\tfrac{1}{4}=P(A)P(B).
- By the distributive law and the addition rule, P((A\cup B)\cap C) equals P(A\cap C)+P(B\cap C) minus P(A\cap B\cap C). By independence this is P(C)\bigl(P(A)+P(B)-P(A)P(B)\bigr), and the bracket is P(A\cup B) by the addition rule and the independence of A and B.
- A\cap B and A\cap C are disjoint, because B and C are, and their union is A\cap(B\cup C). So P(A\cap(B\cup C)) is P(A)P(B)+P(A)P(C), that is, P(A)\bigl(P(B)+P(C)\bigr), which is P(A)P(B\cup C) by additivity.
- Up to the event \{X\notin S_X\} of probability 0, \{X=Y\} is the disjoint union of the countably many events \{X=k,\ Y=k\}, k\in S_X; countable additivity and Result 4 give the sum. If X and Y are iid with pmf p on S, Result 4 gives P(X<Y)=\sum_{i<j}p(i)p(j) and P(Y<X)=\sum_{i<j}p(j)p(i), sums over i,j\in S, which are the same sum. The events \{X<Y\}, \{X=Y\} and \{X>Y\} partition \Omega, so 2P(X<Y)+P(X=Y)=1.
- Write F=F_X. Independence of X and X gives F(x)=P(X\le x,\ X\le x)=F(x)^2, so F(x) is 0 or 1 for every x. Since F is nondecreasing and tends to 0 at -\infty and to 1 at \infty, the set where F=1 is nonempty, bounded below, and runs up to \infty; let c be its infimum. Right-continuity (Result 2 of Discrete distributions: binomial, geometric, Poisson) gives F(c)=1, and F(x)=0 for x<c. Continuity then gives P(X=c)=F(c)-\lim_{m\to\infty}F\left(c-\tfrac{1}{m}\right), which is 1-0=1.