Sampling statistics?

[ianben]

Toms post on the accuracy of samples in relation to triffid has me wondering on all samples.

I have mentioned the Coors malting group we have here. We had a meeting last Tuesday at the maltings and found they had rejected 17% of barley delivered as being off spec.

All barley is tested on farm and only barley which passes is delivered. Coors believe we are unable to get representative samples from bulk stores to explain this.

Coors test each load, 30tonnes, with a robot spear which takes six samples from load. This is then the sample from which all further tests are done and the only sample Coors will accept and they will never re-sample the load. Germination test is done on just 50 grains

Anybody know or know where I could find the stats on the accuracy of Coors test?

[wmoebis]

A representative sample is the most important thing in a sample of grain. If it is not truely representative and accurate, the sample of grain becomes just that a sample.

If we take a sample that is not representative we must be willing to accept what the value of that sample is. If we deliver something that is of a different value we must accept that also.

After the representave sample has been drawn, how that sample is handled is also very important and it is up to us producers to see to it that we are happy with the way it is handled to be able to get a proper analyses of said sample.

I am sorry this doesn't answer your question but I felt it is important that everyone has to realise the importance of a representative sample.

[TOM4CWB]

Ianben,

Since Coors is calling the shots... they can make the rules.

If there is a significant difference from samples taken by other methods... that would be something to take up with standards governance and regulatory bodies you deal with.

If everyone is treated the same... it is hard to argue with them!

What happens if a sample is contaminated with an obvious admix that is not in the lot? Reducers have been know to hang up and cause problem!

[oneoff]

You've got the same problem as there is with Triffid tests. As long as representative samples are taken; and as long as there are no cheaters and liars and accidents at any stage of the test procedure; and the test is accurate and properly conducted it would seem logical that you could always count on the outcome.
BUT besides and because of the above factors where you're not completely sure of all the details at various stages; you seriously need to consider the science of statistics. Its the same science that applies to winning with lottery tickets.
Its an odds game; but luck seems to be more important. You can lose 10 times in a row with heads and tails. On your eleventh try you've still only got a one in two chance of being correct. And so on...
So it is with sampling barley; or sampling Triffid flax. When you have a sensitive test that can detect 1 in 10000 to yield a positive test; there is a slim chance that your subsample of the sample sent in (which was a sample of a much larger bin or field or farm); you could have 10000 negative seeds all ground up= negative tests. If there was any triffid present in the sample and you repeated the test on a subsample drawn from what was left; there is a chance the test could be positive because there was one or more positve seeds in the 10000 just tested. Thats the bad luck part.
Oh; and do you think someone counted out 10000 seeds. I think not. It would be based on weight per 10000 seeds on average. And that varies as farmers know from dealing with plants populations per acre. Another little source of error in the perfect test.
And errors are cumulative.
And somethings are grey.
And of course there is a little bit of everything in anything.

[ianben]

This Coors contract is a bit of a supply chain thing where we try to see the others point of view and try to improve thing where possible.

This summer I was part of a small trial where the on farm samples where compared with loads delivered to Coors.

I did my best to get a reprentative sample taking samples of each load as the bin was filled and mixing them in a bucket.

I had one sample for three loads delivered. No load matched my sample and had quite a variation but the average of the three did. This was true over three bins,nine loads.

I have had one sample tested three times these tests show the same variation as the loads tested by Coors.

I did not get a rejection but two loads were very close although my sample and the average was well within spec.

It will be very interesting to see if we can make any headway on this issue.
Until you see the logistics and feel the passion they have for brand and quality.I think for them rejecting good loads is better than risking ofspec material through.

I was just hoping to get some figures on how many samples are needed and how many tests per sample to give a meaningful result.

[oneoff]

Good point. We get paid by the load at the elevator(usually); but unless you put each load in a seperate bin at harvest; there is no way to have any confidence of sample results matching. (other than the combined elevator samples of delivered loads should match the overall bin sample).
Never really thought of it in that way; and have never seen elevators too keen to do much averaging. That concession takes a bit of pressure and burns up your good will points.