Lesson 2 of 8 · 15 min
Uniform distributions and estimates from data
Model equally likely outcomes with a uniform discrete random variable, and use relative frequencies from data as point estimates of a probability function when no model is known.
What you will be able to do
- Recognise a uniform discrete random variable and write its probability function, .
- Tell a uniform random variable from a non-uniform one in context.
- Use relative frequencies from data as point estimates of .
- Judge whether data are consistent with a proposed model.
Some random variables come from a model you can reason about, like a fair die. Others come from the real world, like the number of people in each car crossing the Story Bridge, where nobody can tell you the probabilities in advance. This lesson covers both: the simplest model there is, and how to estimate a probability function from data.
Definition
Uniform discrete random variable
A discrete random variable is uniform when each of its values is equally likely:
The score on a fair die is uniform on , with .
A uniform distribution has bars of equal height. Turn on the cumulative view: what is ?
A fair spinner has eight equal sectors numbered 1 to 8. Let be the number it lands on. Find .
A model is uniform only when the values of are equally likely, not just the outcomes underneath. A fair die with faces 1, 1, 2, 3, 3, 3 has six equally likely faces, but takes the values 1, 2 and 3 with probabilities , and . That is a non-uniform random variable.
Which random variable is uniform?
For real data there is often no model. Instead, observe many repetitions and use relative frequencies as point estimates of the probabilities:
The estimated probabilities automatically add to 1, so they form a probability function you can use like any other.
Worked example
Estimate a probability function from data
A traffic survey records the number of people in each of 200 cars crossing the Story Bridge in peak hour.
| People in car | 1 | 2 | 3 | 4 | 5 |
|---|---|---|---|---|---|
| Number of cars | 92 | 58 | 26 | 16 | 8 |
Let be the number of people in a randomly chosen car.
(a) Use the data to estimate the probability function of .
(b) Estimate the probability that a car carries at least three people.
(c) Estimate how many of the next 1000 cars will carry at least three people.
Step 1: Divide each frequency by the total
The total is .
1 2 3 4 5 0.46 0.29 0.13 0.08 0.04 Relative frequencies are point estimates of the probabilities. They add to 1 because the frequencies add to the total.
Step 2: Add the estimates for 3, 4 and 5 people
Step 3: Scale up to 1000 cars
cars.
The expected number of times an event happens in trials is .
Try it yourself first, then check the working one step at a time.
Answer
The estimated probabilities are 0.46, 0.29, 0.13, 0.08 and 0.04; ; about 250 of the next 1000 cars.
Method
Find the total number of observations.
Divide each frequency by the total to estimate .
Use the estimates like any probability function, and call the answers estimates.
Watch out for
Dividing by the number of categories (5) instead of the number of cars (200).
Stating an estimate as if it were an exact probability.
Estimated probability function of , the number of people in a car, from 200 cars.
Using the estimates in the graph, estimate the probability that a car crossing the bridge carries at most two people.
Note
Estimates are not exact
A different 200 cars would give slightly different relative frequencies. Larger samples give better estimates, which is why you should say "estimate" and not "the probability is". In Unit 4 you measure how far such estimates can be trusted.
Data can also test a model. If a die is fair, 120 rolls should give each score about times. Counts of 18, 23, 19, 21, 17 and 22 are consistent with a fair die: they are all close to 20, and chance alone produces differences that size. Counts of 8, 12, 15, 20, 25 and 40 are not: a six turning up 40 times in 120 rolls would make you doubt the model.
Exam tip
Say what your estimate is based on
When you estimate a probability from data, show the relative frequency as a fraction of the total () and call it an estimate. When you compare data with a model, compare the observed counts with the expected counts and conclude in words.
Summary
A uniform discrete random variable has equally likely values, each with .
Equally likely outcomes don't always make equally likely values: check the values of .
Relative frequencies from data are point estimates of , and they add to 1.
Estimates vary from sample to sample; larger samples give better estimates.
Compare observed counts with the counts a model predicts to judge whether the model is reasonable.
Finished the lesson?
Try the 3 quick checks above and the lesson completes itself, or mark it complete now.