To: Seafood HACCP Mailing List
Fr: Pamela Tom, UC Sea Grant Extension Program
Happy New Year to all 850 of our subscribers representing over 50 nations.
I'm forwarding John Wekell's reply to Mike Strang.
A reminder...the correct e-mail address to post your comments is
seafood@ucdavis.edu
==========================================================================
Reply-To: <john.c.wekell@noaa.gov>
From: "John C. Wekell" <john.c.wekell@noaa.gov>
To: "'Strang, Mike'" <Mike.Strang@us.fjord.com>
Cc: <owner-seafood@ucdavis.edu>
Subject: RE: Sample size
Date: Fri, 10 Jan 2003 13:34:53 -0800
Organization: NOAA-Fisheries
Mike,
Of course the typical statistical reply is make n=1000 or 10,000. Of
course this is nonsense on a daily practical basis. Most statisticians
dearly love very large numbers of sample.
Determining a statistically valid sample size is very difficult to answer.
Much depends on what particular parameter you are measuring and how much
variation exists between the members of the sample, i.e., from fish to
fish, or clam to clam, or whatever item comprises your lot. In some cases
the measurement technique may have quite a bit of built in variation. For
example, in the mouse bioassay for Paralytic Shellfish Poison, the
variation from one mouse to the next is at least 20% or worse. In
additon, the variation from clam to clam in these assays may well be 100
to 200%. Analytical techniques for simply measured chemical compounds can
have very low variability, for example perhaps as low as 1-2% (in some
cases it be 0.01%).
But even if you have a very good chemical measurement, you need to deal
with the fish to fish variation. This can also be influenced by how your
"lot" is defined. Are all of the fish in the lot you are defining all from
exactly the same school or area. Was the catch, and therefore the lot, all
dumped into one holding tank, and your fishing itinerary covered a hundred
miles? Are all the fish then from one universe.
There are ways to back calculate N from the t-test formula, but these tend
to be iterative processes, but they do give you a ball park for N. This
pre-supposes that you know ahead of time what the estimated variance (or
variability) is from fish to fish (assuming that your analytical technique
has minimal variation).
While N (or sample size) is important, how you sample your lot may be far
more important. You must insure that you are getting a truly random
sample when you pick your samples. Stratification of samples is extremely
common. For example, you wouldn't pick fish from the top layer of fish
being held in a RSW unit for sodium analysis. The tendency is to grab
samples that are the easiest to reach--another bad idea. Another tendency
is to grab samples that are obviously poor quality because they do not
represent high cost for your company ("We were just going to throw them
away anyway"). In any event, a random number table or generator is an
absolute must!
Without knowing exactly what you are attempting to sample, it is
impossible to give you a specific number for samples from a lot. In some
cases, FDA might be able to give some guidance on sampling size, since
they too have check on what you are doing.
Remember, it is sometimes better to "over sample" because you can always
throw data away but can't create data from samples that were never taken.
John C. Wekell, Ph.D.
National Oceanic & Atmospheric Agency
NOAA-Fisheries
Northwest Fisheries Science Center
Environmental Conservation Division
2725 Montlake Blvd. East
Seattle, WA 98112
Phone: (206)-860-3388
FAX: (206)-860-3335
http:\\www.nwfsc.noaa.gov\hab
http:\\www.nwfsc.noaa.gov\ORHAB
-----Original Message-----
From: owner-seafood@ucdavis.edu [mailto:owner-seafood@ucdavis.edu] On
Behalf Of Strang, Mike
Sent: Friday, January 10, 2003 12:35 PM
To: HACCP mailing list (E-mail)
Subject: Sample size
Happy New Year!!
Looking for some advice in developing a quantitative fish inspection
program.
What would be a statistically valid sample size for a given lot?
Any advice would be appreciated.
Mike Strang
Quality Control Manager
Ducktrap River Fish Farm, LLC
57 Little River Drive
Belfast, Maine 04915
Mike.Strang@us.fjord.com
Phone: 207 338-6280
Fax: 207 338-6288
This archive was generated by hypermail 2b29 : Fri Jan 10 2003 - 19:36:16 PST