Hi,

So I'm attempting to solve a problem at work where I need to estimate the useful life of a product given incomplete data. We don't have a large amount of terminal data, and most of the assets we have on file have not yet reached their full useful life, although a material percentage have.

So hypothetically, let's say that 30% of the products have been installed and replaced. Useful life for this subcategory is easy.

But what about the 70% that haven't been replaced yet. The data in this subset is of widely varying ages, some 30days or less, some five years or more. I'm doing my best to recall my stats classes, and I think my best attempt at a reasonable estimate is to use confidence intervals (any input on this assumption welcome). I guess what I'm asking is can anyone give me guidance on how to use statistical measures to incorporate the useful lives that I know, and use the distribution of lengths in service of the assets that have not yet been replaced? Any help is greatly appreciated.

Thanks,

1/28/2015 2:26:08 PM

asked a different way...

If 20% of the product has a known useful life, with an average useful life of 2.4yrs, 20% is older than 3 years, and 60% is between 0 and 3 years, but still functioning, is there a way that we can estimate the expected useful life of the entire population assuming a normal distribution?

1/29/2015 5:21:49 PM

so

20% - 2.4
20% - 3..infinity
60% - 0..3

?

I would start with P(x) = 0.2(2.4) + 0.6(1.5) + 20((infinity-3)/2)

You can replace infinity with the area under that distribution. I also don't know stats at all.

I'm Krallum and I approved message.

1/29/2015 6:52:43 PM

That's a good approach, thanks.

1/29/2015 7:05:18 PM

From what I can tell, this isn't just basic stats. This goes into "survival/reliability analysis". If you have access to JMP or SAS Enterprise Guide, it can do it for you easily.

You won't get a single number as the expected life, you can't, because 70% or so haven't failed yet. That's called as censored data, right censored in this case. What you will get is a survival curve which will estimate the cumulative probability of failure as a function of time. You can also get the "hazard rate", which is the instantaneous probability of failure given that an item has survived up till some time t.

I learned this last semester, and in fact today had a full day workshop on survival analysis using SAS. Workshop continues tomorrow.

1/30/2015 12:00:58 AM

Quote :

"From what I can tell, this isn't just basic stats."

No shit. Basic stats are essentially useless

I'm Krallum and I approved this message.

1/30/2015 5:10:40 PM

Quote :

"You won't get a single number as the expected life, you can't, because 70% or so haven't failed yet. That's called as censored data, right censored in this case. What you will get is a survival curve which will estimate the cumulative probability of failure as a function of time. You can also get the "hazard rate", which is the instantaneous probability of failure given that an item has survived up till some time t."

This sounds like what you want to do.

1/31/2015 9:42:20 AM

You could also simulate it and test how assumptions you make about the failure distribution affect your business decisions.

2/11/2015 1:52:37 AM

Quote :

"You could also simulate it and test how assumptions you make about the failure distribution affect your business decisions."

This is basically how any non publicly traded company works

I'm Krallum and I approved this message.

2/11/2015 4:53:12 PM

Why does it have to be a non-publically traded company?

2/12/2015 2:22:37 AM

If it's public, your priority is to chase the short-term profits.

2/18/2015 10:55:17 AM

The name of this thread is blatant false advertising

2/28/2015 10:18:27 PM