Re: Find a distribution!
From: N10 (limbic_lesion_at_hotmail.com)
Date: 10/27/04
- Next message: Derek: "Map range to range"
- Previous message: Richard Ulrich: "Re: Statistical Analysis in Politics"
- In reply to: Lurker: "Re: Find a distribution!"
- Messages sorted by: [ date ] [ thread ]
Date: Wed, 27 Oct 2004 22:12:19 +0000 (UTC)
"Lurker" <spamkill@spamkill.co.uk> wrote in message
news:2t9ehjF1ss850U1@uni-berlin.de...
> "LWn" <lars.wahlgren.pleasenospam@stat.lu.se> wrote in message
> news:7bzbd.106157$dP1.396840@newsc.telia.net...
>>
>> "viktor eriksson" <viktor_eriksson@hotmail.com> skrev i meddelandet
>> news:j1oesizgn9c5@legacy...
>> > I am working on a project at a company in Sweden where the amount of
>> > bacterias in food are simulated in Excel. The problem is to find a
> god
>> > distribution to my strange indata:
>> >
>> > minimum: 0.04.
>> > mean: 0.04
>> > maximum:1 000 000
>> >
>> > What kind of distribution is best adapted to this strange (and
>> > impossible) indata? I think you have to choose a mean value that is
> a
>> > bit bigger than 0.04 but what kind of distributionh is that, and how
>> > can I use it in Excel to find samples?
>>
>> The exponential and the gamma distribution I think can give you quit
>> skewed distributions. I don't remember about the parameters but
>> "trial and error" might give you some hints to get it positively
> skewed.
>>
>> hth / LWn
>>
>>
> Tried to post this yesterday, but I didn't see it arrive.
> This time I'm cross-posting to the food-science and
> microbiology groups.
> ===============
> Sorry about the anonymous post, it has to be!
>
> 1) Microbiological concentrations are often modelled by a
> logNormal distribution.
>
> 2) Microbiological concentration are often measured by
> counting organisms in small amounts of material. Beware
> of the "censoring" this can introduce. e.g. 0.04 cfu/g
> may correspond to 1 cfu in 25 g. If this is a pathogenic
> organism I would suspect that "mean 0:04" means the
> vast majority of samples resulted in a count of zero. It
> is possible to make inferences from censored data, but
> the techniques are not elementary, I wouldn't try in Excel.
>
> 3) If you're looking at pathogens in food it may be that you
> need different distributions for "contaminated" and
> "non-contaminated" foods. It is likely that contamination
> occurs by "discontinuous" processes.
>
> 4) You should not base risk management on any kind of
> statistical model; you cannot get the required confidence.
> Risk management should be based on understanding the
> ways in which the hazard arises and blocking them, by
> a structured approach such as HACCP.
>
> 5) If you are assessing risk, rather than managing it, you
> should look at the huge volume of prior art. I suggest
> starting at http://www.foodriskclearinghouse.umd.edu/
>
> 6) All except the most recent versions of Excel gave
> wrong results when calculating distributions in the tails.
> (Recent version may have improved, I don't know).
>
> HTH
>
> A lurker
>
>
If you really look at this data I think you might agree it is fundimentally
flawed and therefore useless. Dont mean to be harsh but factually something
is wrong here.
Best N10
- Next message: Derek: "Map range to range"
- Previous message: Richard Ulrich: "Re: Statistical Analysis in Politics"
- In reply to: Lurker: "Re: Find a distribution!"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|
