Learn what your exams actually test.
Upload past exams for a course. We sort every question by topic, explain the concepts it tests and show you a solution.
Free to try. No account needed.
An animation of a statistics course: three past exams break into questions, the questions are sorted by topic into a course map, a question on one topic opens with its solution, then a concept it tests opens with a diagram, a worked example and a slider to try.
Upload your past exams
See what they actually test
Open any question, see the solution
Learn the concept behind it
Work through an example
Try it yourself
Question 6
7 pointsBayesian InferenceA screening test is used where 1 in 200 people have a disease. It detects 98% of cases and wrongly flags 3% of healthy people. Someone tests positive. What are the posterior odds that they have the disease, and what probability is that?
The prior odds are and the likelihood ratio is , so the posterior odds are . That is a probability of .
Bayes' theorem
How to work out how likely something is once you have seen new evidence.
So a positive result puts you among 198 people, of whom only 99 are ill: a 50% chance.
The same test (positive for 99% of ill people, for 1% of healthy people) is used on 10,000 people, but now 2% of them have the illness. What is the probability that a person who tests positive is ill?
Ill: of . Healthy: .
Ill and positive: of . Healthy and positive: of .
Move the slider until a positive test is a coin flip: a 50% chance of being ill.
- how common it is (%)
- 0.1
- chance you're ill (%)
- 9.0
At 1%, the 99 ill people who test positive exactly match the 99 healthy people flagged by mistake. Any rarer, and most positive tests are false alarms.
Learn each concept your way
Start with the idea and a picture, work through an example step by step, or go straight to the definition.
STAT110Probability and Statistics
ExampleBayes' theorem
How to work out how likely something is once you have seen new evidence.
Picture 10,000 people screened for an illness that 1 in 100 has, with a test that is right 99% of the time. Of the 100 ill people 99 test positive, and so do 99 of the 9,900 healthy people, by mistake.
So a positive result puts you among 198 people, of whom only 99 are ill: a 50% chance.
The 10,000 people split by illness, then by test result; the highlighted groups are everyone who tests positive, 99 ill and 99 healthy.
How to spot it
The question gives the chance of a result for each possible cause (a positive test if ill, a positive test if healthy) and asks for the chance of the cause once you have seen the result.
Worked example
The same test (positive for 99% of ill people, for 1% of healthy people) is used on 10,000 people, but now 2% of them have the illness. What is the probability that a person who tests positive is ill?
Split the population into the possible causes using how common each is.
Ill: of . Healthy: .
For hypotheses that partition the sample space and data with :
The denominator is the law of total probability. is the prior, the likelihood, and the posterior.
In general .
On the 10,000 people: .
Same test: it flags 99% of ill people and wrongly flags 1% of healthy people. Now you choose how common the illness is among the 10,000.
Move the slider until a positive test is a coin flip: a 50% chance of being ill.
- how common it is (%)
- 0.1
- chance you're ill (%)
- 9.0
At 1%, the 99 ill people who test positive exactly match the 99 healthy people flagged by mistake. Any rarer, and most positive tests are false alarms.