Survival analysis
And the legend of the exploding barbeque…..
The problem• Barbeques, while tasty, use gas bottles. • From time to time, and with no warning, these
bottles randomly explode.
• This tends to put a dampner on parties.
• How to design an experiment to test the exploda-bility of 2 BBQs?
• Could get 20 of each BBQ, use them for 40 days, look at number of days till explosion.
The problem
Red Neck Roaster T-Bone Terminator
Day Exploded? Day Exploded?
32 yes 40 no
18 yes 40 no
40 no 39 yes
10 yes 40 no
•How do we analyse these data?•What about ones that didn’t explode?
• We could just remove the data that didn’t explode
• But they’re an important part of the analysis• Need to instead work out a hazard function,
the chance of each BBQ type exploding on each day.
• E.g. we go back to our data, and find the Red Neck Roaster has a 1% chance of exploding per day
• For The T-Bone terminator, it’s 0.1%
The problem
Results..
Redneck roaster
T-Bone Terminator
Survival analysis• Hazard function forms the basis of survival
analysis• Used for any “time to event” data
Everything eventually dies/fails
Looking for differences in rate