Easy to Ask Hard to Answer
Abstract
Chris and Fred discuss questions that are easy to ask … but hard to answer. And these questions are often hard to answer because the person asking the question hasn’t thought about the REAL question they are trying to answer. So how many samples do I need for everything to be OK???
Key Points
Join Chris and Fred as they discuss questions that are hard to ask but difficult to answer. This happens a lot in reliability engineering. Perhaps the most common question is … how many samples do I need to test?
Good question. Actually … it isn’t.
Topics include:
- What is the decision you are trying to inform? For example, if you are trying to test two candidate materials for an AMAZING new product … and your DECISION is only about selecting the best material, then you might not need many samples at all. Especially if one material is clearly superior to the other. We aren’t trying to estimate reliability to three decimal places. We are trying to work out which material we are going to use.
- How are you going to USE these samples? If we are trying to find out the probability of a tire lasting 40 000 km, one approach is to test sample tires for 40 000 km and work out how many have failed in the 40 000 km (or not). A much better approach is to test all samples until they fail. Let’s say you only test three tires, but they fail at 100 467, 98 629 and 98 756 km respectively. Do you need to test many more tires to know that there is a really good probability of tires lasting 40 000 km? … but what happens if you stop testing at 40 000 km and simply find out that the tree tires didn’t fail the test?
- Confidence is a measure of you … so what is it? Data analysis that investigates a random process (like failure) is inherently uncertain. That is why we see confidence bounds on many outputs. So what confidence do you need to make your decision? Hint … you can never get 100 % statistical confidence.
- … and how reliable is your product? If your product is REALLY, REALLY reliable, then you don’t need nearly as many samples to (for example) demonstrate that you meet your system requirements. But if your product is only JUST more reliable than the requirement, then you need LOTS of samples.
- What question do you WANT to hear? This is a big one. It is called bias. And when we have bias, we sometimes do whatever we can to pretend the answer we are hearing is not possible. Perhaps we say to those of us giving us these uncomfortable answers that ‘you need to PROVE it.’ Perhaps we don’t believe our subordinates and get a bunch of highly paid consultants. What is your (organization’s) bias?
So … have you answered all these questions before you asked your ‘easy’ question?
Enjoy an episode of Speaking of Reliability. Where you can join friends as they discuss reliability topics. Join us as we discuss topics ranging from design for reliability techniques to field data analysis approaches.
- Social:
- Link:
- Embed:
Show Notes
Leave a Reply