11 replies to this topic

[okido]

Last October Regulation 1935/2000 entered into force.
This regulation is about traceability of food contact materials.
The advice is to keep samples of raw materials end final products.
If you take samples from raw materials and final products, how much do you take?
What sample size is normal 1% or 0,001%, do you evaluate the risk first and determine sample sizes?

Who can share his experience with this issue?

Every disadvantage has its advantage, a Dutch soccer player

Okido

[Simon]

Last October Regulation 1935/2000 entered into force.

I think you mean 1935/2004

This regulation is about traceability of food contact materials.
The advice is to keep samples of raw materials end final products.
If you take samples from raw materials and final products, how much do you take?
What sample size is normal 1% or 0,001%, do you evaluate the risk first and determine sample sizes?

I'm no statistician, but I suppose just enough to provide sufficient backwards traceability and to allow adequate testing in the event of a customer complaint. I can only give a practical example...a printer of reels of food contact packaging such as lidding material would probably keep something like a 1 metre strip of unprinted raw material and a 1 metre strip of printed material from every reel of material produced. So if you said a reel was 2,500 metres and 2x 1 metre samples were retained that would be a sample size of 0.08%. Who can tell me the confidence level of this?

Simon

[Lalith Gunatillake]

Dear Simon, Below eg is not related to food but it gives an idea about confidence level (and also confidence interval).
By the way I think another important aspect related to sampling is "representative sampling". If not we may not get the proper picture of the population (in our case end product/raw meterial as a whole).

The confidence interval is the plus-or-minus figure usually reported in newspaper or television opinion poll results. For example, if you use a confidence interval of 4 and 47% percent of your sample picks an answer you can be "sure" that if you had asked the question of the entire relevant population between 43% (47-4) and 51% (47+4) would have picked that answer.

The confidence level tells you how sure you can be. It is expressed as a percentage and represents how often the true percentage of the population who would pick an answer lies within the confidence interval. The 95% confidence level means you can be 95% certain; the 99% confidence level means you can be 99% certain. Most researchers use the 95% confidence level. When you put the confidence level and the confidence interval together, you can say that you are 95% sure that the true percentage of the population is between 43% and 51%.

Regards,
Lalith

[Charles.C]

Dear All,

I also don’t quite understand what “confidence level” means in the context of traceability ( an item is either traceable or not ?) although it obviously exists , eg –

“The Eurofins-TAG traceability system begins with the establishment of a sampling plan that is statistically appropriate to provide the desired level of confidence of traceability. Reference samples are collected from each animal and a reference database is created.” (Whatever that means?, each animal !)
http://www.dairyrepo...lity-technology


Saw one semi-statistical suggestion for amount of retention sample ( for organic products though) which seemed reasonable (I have used something similar for mineral products) but not sure if this logic can apply to packaging –

http://www.ceres-cer...aceability.html

This article on Guidelines for traceability of food contact materials - 1935/2004 seems good and relevant but couldn’t see any sample sizes –

http://crl-fcm.jrc.i...m...view&gid=39.
(7MB slow file, maybe on IFSQN site already ?)

It's an interesting thread IMO, hopefully some more input possible.

Rgds / Charles.C

[Simon]

Thanks Lalith and Charles. The description of confidence levels takes me back to my school days. Way back. I agree it's an interesting discussion. Okido (as the thread starter) how did you get on with this problem?

Reagrds,
Simon

[okido]

Hi Simon,

The problem of keeping samples of finished products and raw materials originated from new traceability regulations.
We are redesigning the traceability system into a risk based model were food safety, product quality and processing quality come together.
I do not choose the statistical approach for the time being.
We started sampling raw materials on a regular basis mainly to gather information and experience.
Sampling size is determined by the tiny containers we still had on stock.
Finished product was already kept as a rule.
Finally I plan to use the risk based traceability model for determing sampling needs.
High risk = sampling plan based on statistics
Medium risk = straight forward sampling
Low risk = sampling can be done but not necessary

Regards, Okido

[Charles.C]

Dear Okido,

I hope you have a good supply of intelligble statisticians (unless you are one yourself) !.

As per Simon’s comment, seems to me the answer will strongly depend on what you are measuring, for example if it’s a demonstration that a warning regarding “allergen content” is present on all labels in a particular lot, you will need to hold a 2x, presumably unseen ( ) , randomly selected quantity of at least the number of labels corresponding to a specified (allowed?) risk/probability of finding at least one defective or enabling a declaration that the percentage of defects is below X%. Sounds like a heck of a lot of labels. (The 2x is in case the first sample is not helpful.)

Medium risk = straight forward sampling.

I thought even simple sampling was also based on statistics, or do you mean like Simon's ?

Rgds / Charles.C

[Sankara narayanan]

Dear Okido,

II thought even simple sampling was also based on statistics

I agree with you Charles... One of the simple ways we used to sample when we received say 100 carbouys of Hydrogen Peroxide was to take samples from √ +1 ie in this case from 11 carbouys....

Best Regards,

A Sankara Narayanan

[okido]

Hi Charles,

I have no supply of statisticians; I have to do it myself Simon.
Charles, you are right that simple sampling has also a statistical component, all sampling has.
I meant that a would just determine the sample size by experience rather than making a sampling plan based on statistics.
Basically I use the 10% rule, is the population large I decrease a little, this is close to the √ +1.
When you have some expectation/understanding of how the problem is distributed in the population this works fine.
I will do the statistical calculations when the risk model is completed.

Best regards, Okido

[Charles.C]

Dear Okido,

Actually, I should probably slightly apologise for querying yr “straightforward” since I mentally translated straightforward as simple and then confused “simple” with “single” (as compared to 2-stage sampling etc) in my original comment.
This is a topic where there are an almost infinite number of fascinating cookbook procedures, some of which actually have a vaguely statistical basis like the square root thing even if unknown (and uncaring) to the user. Similar to yourself, I most frequently use a 5, 2, 1, or even 0.5 percent basis for sampling cartons of product depending on experienced variability, quantity, value and sometimes my intuitive (lack of) trust in the producer. Absolutely no statistical basis I’m sure. In the theory for the AQL tables, I seem to remember there is a delightful comment from the originators like “the relationship between the various sample sizes does not quite follow the theoretical line since we ‘empirically’ adjusted up the values for larger quantities to minimise the chance of an expensive error !” Hmmm.

I am just looking at a text on sampling petrochemicals which has this nice one – in order to ascertain with at least 95% certainty that the consignment does not contain more than x% defectives, take samples from 300/x packages selected at random, or from the whole consignment, whichever is less (the practical requirement being zero defectives found in the sample). I guess this might work for the labels also. What is an acceptable max. defect rate for not printing allergens I wonder, 0.001% ?? the result is of course that sampling cannot routinely do it, a situation similar to sampling for zero tolerance for Salmonella at equivalent levels (In fact, the typical high end sample size of 60 units / negative result gives a 95% certainty of max. 5% defective units from above formula = "free of Salmonella")

Rgds / Charles.C

[Simon]

I think a sample frequency / size based on 'knowing the supplier' is probably as accurate as that provided by any mathematical equation. I'm sure a statistician could quite easily prove me wrong.

[Charles.C]

Dear Simon,

I think the bookies would agree with yr method. Otherwise I believe called (politely) as "knowing a wrong 'un".

Nonetheless one has to have some respect for a profession which could show that the probability of the sun rising tomorrow is 1,826,214 :1.

Rgds / Charles.C

PS sorry, this is going OT