Source
Adapted from my own handwritten and hand-typed notes from Year 1 to Year 3 of my EEE degree
These two results handle problems where the sample space is split into several cases, such as several machines or several suppliers.
Partition of S
is a partition of when:
- the are disjoint (mutually exclusive)
- together they cover everything: , so
Total probability theorem
For any event ,
Idea: must happen together with exactly one of the , so add up each path, built with the multiplication rule from Conditional Probability.
Bayes theorem
To reverse the condition (find the cause given the effect):
The denominator is just from the total probability theorem.
Method
Steps:
- Name the partition (the possible "sources") and the observed event
- Write down the priors and the likelihoods
- Use total probability to get
- Use Bayes to get
A tree diagram helps: each branch product is .
Worked example
Machines make 50%, 30% and 20% of the output, with defect rates 2%, 3% and 5%. An item is defective (). Which machine probably made it?
Solution:
- similarly and
Machine 3 makes only 20% of items but accounts for about a third of the defects.