I agree with both of you when you say that the initial qsmso's question is an important one, that's why I really wanted to understand the answer.
It has not been easy maybe for language problems as I'm not a native speaker of English (by the way , JSW05 when in message #5 you write "either the mode or the cause" do you mean both of them or just one out of the two?).
Now I think that I have understood how things go, thanks to RAHarpster's answer too.
As to the extimate of (O) from available data, I want to express my very personal thinking.
I write "very personal" because I've never heard people saying this, even if I thing it's an important topic to discuss about.
In statistics there is a golden rule: all the times you need to do an extimate, you need to plan (i.e. to design it) a poll and you cannot browse informations gotten for other reasons and make a late supposition!
A real example I know:
A company had sampling plans to test incoming products.
They followed a sampling scheme very similar to MLT STD 105E, so the sampling number was not a constant.
Every month they happend to make an extimate of their supplier rate of defects, and guess how did they extimate that rate?
From the the only information they had of course!
So if lot#1 had 10000 items and they checked 100 items out of them and just one was defective they assumed a rate of 1%.
Suppose lot#2 was in free pass, so it was not taken into account (what if the rate of defectives was 10%?); lot#3 had a sampling number of 10 and no one defective (rate 0%), and so on...
Just imagine what was the value of those statistics!!!
And a lot of high managers looked at those diagrams pretending to understand everything and making future plans based on them.
So be sincere... do you really think that most of the data, people referr to when they try to extimate (O) is better that data in my example?
Making a reliable poll is a very difficult science and you have to prevent all the possible causes of biasness.
I think that virtually all the infomations companies have are biased.
Maybe the reasons of biasness are very different from the ones I've shown but are always in place.
BTW, I don't think they could not be useful, you just have to know their limits, how to use them and expecially what the aim of
FMEA is: that is not to make a scientific extimation of (O), but just to have an idea in order to prioritize actions.
Subjectiveness is fundamental, and data you have are just useful to help you having a more precise idea about the numbers.
The more you try to get objective your extimate from biased data the more you are wasting your time (and risking to abtain a less real extimate).
Just my 0,02.