Section 3

Probability and Random Variables

Sample Spaces
We now will investigate probability through looking at various occurrences/processes that are
subject to chance. We call such a process that leads to one of several possible outcomes an -
____________________. All such possible outcomes form a set called the ______________________, usually
denoted S.
Example: Write out the sample spaces for the following experiments:

Flip a coin. S=

Roll a six-sided die. S=

Flip two coins in sequence. S=

Flip three coins in sequence. S=

When examining the outcomes of an experiment, we often focus on specific outcomes of interest.
We call such a subset of a sample space an ____________. We can then turn to probability to find how
likely something interesting to us is!
Example: Find the probability of getting exactly two heads when flipping three coins in
sequence.

Now, find the probability of getting three heads.

Probability Notation
The complement of an event A, denoted Ac or A’, is

Complement Rule: P(A’) =

The intersection of events A and B, denoted as A ∩ B, is the event where an outcome is from both
events A and B.

The union of events A and B, denoted as A ∪ B, is the event where an outcome is from either event A
or event B.

Properties of Probabilities
There are three basic axioms of probability that define how probability works more generally. Let E
be an event of a sample space S.
1. 0 ≤ P(E) ≤ 1
2. P(S) = 1
3. For any disjoint events E1, E2
𝑃(𝐸1 ∩ 𝐸2 ) = 𝑃(𝐸1 ) + 𝑃(𝐸2 )
Note: Two events are disjoint or mutually exclusive if A ∩ B = Ø.
Conditional Probability
Example: Consider the following table of counts below:
Income Level
High (H) Mid (D) Low (L)
Marital Status Not Married (N) 30 40 30 100
Married (M) 60 50 40 150
90 90 70 250

Find the following probabilities:

P(N) =

P(H) =

P(N ∩ H) =

Above, we found the probability of selecting an unmarried person, a person of high income,
and an unmarried person of high income. How would we go about finding the probability of
selecting someone of high income, given that we know the person is unmarried?
P(H | N) =

The conditional probability of A given B is

P(A | B) =

We can think of these probabilities as working with known information. The event provided after
the “|” is something that we know is true, and are now working to find the probability knowing this.
Notice that the definition of conditional probability gives us information about the intersection too,
if we simply rearrange the terms from the conditional probability above.
P(A ∩ B) =

But sometimes, knowing some given event happens doesn’t actually affect the probability of
another event. Perhaps there’s a certain probability that you attend this class on a particular day.
There are probably many things that will change this probability – a rainy day may make you less
likely to attend, or maybe other personal events in your life may occur that would lower the
probability of your attendance. But there are some events that may not affect this too much: it’s
likely that the event of you receiving a spam call or the event that someone else in the USA winning
the Powerball lottery will not have an impact on how likely you attend class. If that is the case that
knowing one event happens doesn’t affect the probability of another event, we can say two events
are independent. Mathematically, we can write the condition for independence as:
P(A | B) = (for independent events A, B)

We can also write out an alternative rule for independence, based on the multiplication rule for
intersections above.
P(A ∩ B) = (for independent events A, B)

Important: While they sound similar, independent events are not the same as mutually exclusive
or disjoint events! In fact, mutually exclusive events are very dependent. If A and B are mutually
exclusive and A occurs, then you know for a fact that B did not happen!
As your instructor, I would sincerely hope that the event you attend this class and the event that it
is rainy are independent events, but experience has unfortunately told me that this isn’t true.

Random Variables
We now define new tools to convey a numerical description of items in the sample space and their
associated probabilities.
Let S be a sample space. A _____________________ is a function X : S ⟶ ℝ, equating outcomes from a
sample space to a numerical value.
Example: Consider the experiment of flipping 3 coins, and let X be a random variable for the
number of heads.

Random variables give us a way to relate the outcomes of an experiment to a numeric value, but it
doesn’t tell us anything about probability. We need to assign another function to determine the
probability of certain numeric values for a random variable.
The type of function we assign to a variable differs depending upon what kind of variable we are
measuring in the first place. There are two types of random variables: __________________ and
_________________. In this section, we will only talk about discrete random variables.
Discrete Random Variables
If the random variable we are working with has a _________________ range, that is, the possible
numeric values the random variable can be is _________________, we say that it is a discrete random
variable.
To talk about probabilities for a discrete random variable, we define a probability mass function
(pmf). The probability mass function for a discrete random variable is defined pX : ℝ ⟶ [0, 1],
where pX(x) = P(X = x), and must satisfy the property:

∑ 𝑝𝑋 (𝑥) = 1
𝑎𝑙𝑙 𝑥

All of this mathematical jargon is really just masking the main axioms of probability we discussed
previously in terms of a new probability function. Specifically, these just make sure that for our
probability function that all probabilities should always be between 0 and 1, and that the
probabilities of all outcomes must add up to 1.
Example: For the previous experiment of flipping 3 coins, write out the pmf of X.

Describing and Summarizing Discrete Random Variables

We now explore several summary measures of a random variable. The first one we will examine is
the expected value, which is synonymous with the mean value of the random variable. We define the
expected value as
μ = E(X) =

Example: For the previously discussed experiment of flipping three coins, find the expected
value of X.

We can also look at measures of variability on random variables. Similar to the sample variance
measure we defined previously on a set of data, we can define the variance of a random variable as
σ2 = Var(X) = E[(X – μ)2] =
Computationally, this is not too fun to do. But we can make it slightly easier on ourselves by doing
some rearranging of the definition above:

Thus, we get the alternative formula for the variance of X below.

Var(X) =

When we defined the sample variance, we defined the sample standard deviation as the square root
of the sample variance. That relationship holds up in the random variable case as well!

σ = SD(𝑋) = √Var(𝑋)

Example: For the previously discussed experiment of flipping three coins, find the variance
and standard deviation of X.

Example: For the pmf given in the table below, find E(X) and Var(X).

x 0 1 2 3 4
p(x) 0.4 0.2 0.15 0.15 0.1
When we’re working with expected values and variances, there are nice properties we can leverage
with linear functions of a random variable. Let a, b be real numbers, and X a random variable.
E(aX + b) =

Var(aX + b) =

These properties also hold for linear combinations of random variables:

E(X ± Y) = E(X) ± E(Y)
Var(X ± Y) = Var(X) + Var(Y)

Notice the ± in the equation for variance above doesn’t stay on the right side of the equation. Why is
that the case – specifically, why would taking the difference of two random variables result in a
larger variance?

The Binomial Distribution

When modeling real world experiments, we can often map these to common probability models or
distributions. There are many distributions we will not cover in this class (e.g. geometric, Poisson),
but we will focus on the binomial distribution.
The Bernoulli distribution is not common to use, as it only looks at the outcome of a single event.
However, repeating these events many times is a more applicable probability experiment, as this is
done in random sampling with selecting multiple individuals from a population. These kinds of
experiments use the binomial distribution.
The binomial distribution: X ~ Bin(n, p)
𝑛!
𝑝𝑋 (𝑥) = 𝑥!(𝑛−𝑥)!
𝑝 𝑥 (1 − 𝑝)𝑛−𝑥 , x = 0, 1, … , n
Used to count the number of successes in n independent trials that have a probability p each of
succeeding.
Characteristics of a binomial experiment:

• There are always a fixed number of trials, n.

• Each trial has two outcomes: success and failure.
• The probability of success is fixed from trial to trial.
• Each trial is independent – that is, results from previous trials don’t influence the
outcome of other trials.
Example: A bike shop is ordering new bikes from a supplier. Before selling bikes, this shop
always does an inspection of the bikes to ensure all components are aligned and working
properly. Based on their history with the manufacturer, 90% of all bikes shipped to them
are ready to be sold without any further work done.
If X is a random variable for the number of bikes ready to be sold immediately, does X
represent a binomial experiment?

The bike shop has two customers that have placed pre-orders for this bike. What is the
probability that this shipment will have at least two bikes ready to sell with no additional
work to be done?
To find the mean and variance of a binomial random variable, we can start with the case of when n
= 1. In this case, there would only be two possible outcomes for X, 0 and 1. Thus, we can easily find
the following:
E(X) =

Var(X) =

Now, we can see that to get to a binomial distribution with any n, we could consider adding up
many binomial random variables with n = 1. So long as these outcomes are independent, their sum
would equal the total number of successes (or 1’s) that occurred in each of the individual trials.
Thus, using our rules for expected values of sums, we can find that for any binomial distribution:
E(X) =

Var(X) =

Example: Find the expected value and variance for the number of bikes ready to be sold
from the previous example.

We can also compute binomial probabilities in R. The various functions you can use with the
binomial distribution are given below.
dbinom(x, n, p) #gives the probability P(X = x)
pbinom(x, n, p) #gives the probability P(X ≤ x)
qbinom(x, n, p) #finds the value k value where P(X ≤ k) = x
rbinom(x, n, p) #randomly generates x data points that come
from a Bin(n, p) distribution
Example: Use R to find the probability from the previous bike example.
Continuous Random Variables
Last chapter, we talked about discrete random variables and common models like the binomial
distribution. This chapter, we will talk about continuous random variables.
If the random variable we are working with has a _________________ range, that is, the possible
numeric values the random variable can be is _________________, we say that it is a continuous random
variable.
For discrete random variables, we assigned to each possible outcome a probability using the
probability mass function. However, we now have an uncountable set of possible values for the
range of our random variable, like all of ℝ or an interval like [0, 1]. If we were to try to assign
probabilities to an uncountably infinite number of values, we wouldn’t be able to make the
probabilities add up to 1 – they would greatly exceed that!
Thus, we need a new tool to allow us to take probabilities over ranges of values. We now use
functions known as probability density functions to do this. These functions allow you to find
probabilities for ranges of values for the distribution by calculating the area underneath the
function within that range. To preserve the idea that all of the probability must “add up” to 1, a
probability density function must have all area under the function equal to 1.
Let’s start with a basic example of a probability density function!
Example: A student takes MTD’s Green (5) bus into campus. They haven’t memorized the
schedule, but they know that during the middle of the day that the bus leaves every 15
minutes. Assuming that the bus arrives on time, what is the probability that they will wait
no more than 8 minutes for the bus to arrive?

So, areas under functions are easy to find when they’re familiar geometric shapes like rectangles.
But limiting our probability modeling to geometric area formulas is not very versatile.
Unfortunately, we need calculus to compute the area under functions generally, and this is not a
prerequisite for this class. (I’m guessing most people are thinking this is more fortunate than
unfortunate, and I can respect that. Enjoy your calculus in Stat 400, stat majors!)
However, R is quite good at finding areas under functions, especially functions for well-known
distributions. And the most well-known distribution is…
The Normal Distribution
The normal distribution is probably the most loved, known, used and misused distribution of them
all. You’ve probably heard of bell-curves before in relation to modeling test scores, human heights,
and many other natural phenomena. You might have also heard it in terms “curving” grades in a
class – in practice, this has little to do with making grades look like a normal distribution, but
instead, ends up in the instructor being nice and giving out more points than originally earned.
Just for fun, let’s take a look at the function that defines the normal distribution:
1 𝑥−𝜇 2
1 − ( )
𝑓𝑋 (𝑥) = 𝑒 2 𝜎 , –∞ < x < ∞
𝜎√2𝜋

In the normal distribution above, the parameters μ and σ are the mean and standard deviation of
that normal distribution. In fact, there is no way to use calculus to determine probabilities with this
function in a closed form, so we have to use other tools to compute probabilities.
Before the widespread use of computers, the primary method was to convert everything to a
standard normal distribution. Usually denoted with the random variable Z, the standard normal
distribution is a normal distribution with mean 0 and standard deviation 1. When drawn, this looks
something like the picture below:

The usefulness of a standard normal distribution is realized by the following fact: If a random
variable X is distributed N(μ, σ), then the random variable Z, defined as

𝑍=

has a standard normal distribution. This quantity is typically referred to as a ___________________,

which tells you how many standard deviations one data value is from the mean.

Thus, for any normal distribution, we could use this to convert something from any normal
distribution to a standard normal distribution, and then use a probability table to find the solution.
Such methods are irrelevant with the use of computers! Like with the binomial distribution, R gives
ways to calculate the probabilities of a normal distribution.
dnorm(x, μ, σ) #gives the height of the density function
pnorm(x, μ, σ) #gives the probability P(X ≤ x)
qnorm(x, μ, σ) #finds the value k where P(X ≤ k) = x
rnorm(x, μ, σ) #randomly generates x data points that
come from a N(μ, σ) distribution
If you leave the fields for the mean and standard deviation blank, R will assume a standard normal
distribution.
Let’s try some examples of normal probabilities that use these R functions.
Example: Assume that for a specific population, heights are normally distributed with μ = 68
inches and σ = 2 inches. What percentage of the population is taller than 72 inches?

Example: Assume that speeds on Interstate 57 are normally distributed with μ = 68 mph
and σ = 4 mph. Find the 85th percentile of these speeds.

Example: For the Interstate 57 speed example, find the 2 speeds that contain the middle
90% of all speeds on Interstate 57.
Additional Practice
Example: An archer can hit the bullseye with an arrow 40% of the time. If the archer takes 6
shots in a given round, what is the probability they hit at most 1 bullseye?

What is the probability they hit 4 or more bullseyes?

What is the expected number of bullseyes the archer will hit? What is the variance?

Example: Measuring blood pressure provides many challenges due to the variation in
measurements depending on the time within a cardiac cycle it is taken. Adults are often
diagnosed for treatment of high blood pressure when they report a blood pressure of 140
mm Hg. For a patient whose average blood pressure is 130 mm Hg and standard deviation
is 13 mm Hg, what is the probability that a doctor will diagnose this patient with high blood
pressure? Assume that this patient’s distribution of blood pressure measurements is
normally distributed.
What blood pressure reading would represent the largest measurement among the lowest
25% of measurements? What blood pressure reading would represent the smallest
measurement among the highest 25% of measurements?