Sampling/Measuring fresh produce against specs
Hi everyone,
I'm still new to QS etc and am now comfortable with the sanitation/GMP side, and moving on to the technical gobbledy gook. (I've just started Six Sigma etc and am trying to wrap my head around it all).
I work for a fresh produce distributor and we are now having to introduce product specs and monitor supplier conformance and verify that incoming produce from the farmers meet our specifications. Note that there are no existing formal procedures and I'm starting everything from scratch. Everything I know so far is self taught & I'm missing quite a bit of basic foundation in silly little things.
My specs are still very basic ie: tomato must be round, red, free from excess debris & pest infestation, x to y cm diameter, x to y kg weight etc...
Most of the sampling is based on visual observation, but I'm having a problem with the following:
1. Sampling size : I'm unsure how to determine an accurate sampling size per lot. I have found numerous charts online, but the qty they recommend is more than our little operation can take (I can't cut up that many tomatoes!! & it would take my QC guy forever, we work on deadines and fast turnaround) ie, for a batch of 3000 units, sample 279.....I'd be grateful if anyone has any suggestions on how I could calculate this?
2. Measuring: at the moment,we are taking 3 samples per batch, weighing and measuring the diameter with a scale & a ruler. This works ok for 3 units, but if I have to do more, then time becomes a problem .... wondering if anyone has a faster, more accurate suggestion for doing this? (also has to be relatively cheap, my Directors are giving me practically zip to work with) .... I had thought of getting plastic "moulds/templates"with cut outs of the ideal size, but this doesn't quite work with a range of sizes, and we have over 70 inventory items so not the most effective way to my thinking....
I've completely hit a blank here :uhm: , grateful for any clever ideas from the gurus out there ! Thank you!! :w00t:
Dear Tan,
Sampling unfortunately tends to a book sized topic. And notably so for highly variable commodities, eg like fresh produce ?
The basic requirements, as you probably already know, depends on things like (a) the characteristic you wish to measure, (b) its typical variability (including measurement error), and © the desired accuracy of yr estimation. (a,b) obviously inter-relate to the specification.
IMEX, the buyer/seller agreed specification often defines the sample size / limits so statistical theory can become somewhat irrrelevant officially. But one still needs internal control to know what the released data actually means. :smile:
Internally, the first priority is usually to set a specification. Then find a way to ("adequately") measure the critical characteristics. Then maybe iterate.
For control purposes, the usual approach is to first evaluate the variability of the average value of the item(s) you are interested in. IF POSSIBLE. Often (very) approximately done by a random sample of say 20-30 units from a few lots, do the calculation for sigma, and compare the results. Basic book equations will now give the required sample size for a given acceptable error/accuracy. This is sampling 101 of course and detailed in thousands of text books. Sophisticated sampling methods can improve the accuracy / reduce the sample size but may be restricted by manpower/accessability/time/cost etc.
The basic one-stage formulae assuming random sampling / small size samples are pretty simple and maybe you know them already. As you say the usual problem is too many samples, if so, you have to decide where to compromise, eg acceptable error = +/-10%, 20%? As you get more lot-by-lot data, the calculation can be refined and the error should reduce (somewhat).
Another pre-developed procedure is via standard AQL (acceptable quality level) methodology. Experts have done all the calculations and offer accept/reject tables of sample size vs desired accuracy. It's good to quickly guess the sampling options for your problem, eg illustrate the compromises which may be possible between sample size and accuracy. Note that this method is intended to allow a direct accept/reject decision, not be the prelude to a negotiation. Tables exist for attribute and variable data.
Sometimes, unfortunately, a practical compromise cannot avoid a large error. If so, the risk, eg acceptance/rejection should be evaluated within the context of, for example, the specification / financial/reputation consequences.
Rgds / Charles.C
PS - added - from memory, a very, very approx. generic rule of thumb to assess yr sampling capability is that the CV (coefficient of variation) should ideally be < 5% but what specific caveats exist, if any, i don't remember offhand.
Dear Charles, thanks very much for the great input and detail! I've never heard of IMEX or AQL so that will be great to research also!
haha I guess my hopes for a shortcut were unrealistic, especially as consistency of supply is not something I have the luxury of dealing with, so trying to maintain any form of spec and is incredibly difficult, but hopefully with a more accurate sampling plan I will be able to achieve something ....
All other comments welcome! :)
IMEX = In my experience
AQL = Acceptable quality limits.
Dear Tan,
I have attached a little reading material related to my previous post. I should have clarified that most sampling is done for one of two possibilities – (a) to enable the estimation of some quality related factor, or (b) to make a direct decision as to lot acceptance/rejection. I guess you are currently more interested in (a).
sa1 - determination sample size.pdf 950.46KB 173 downloads
sa2 - codex sampling plans, guidelines.pdf 509.63KB 175 downloads
sa3 - inspection fresh cut produce USDA 1993.pdf 2.55MB 153 downloads
sa4 - FSA, sampling of feeding stuffs, 2004.pdf 96.89KB 88 downloads
sa5 - SOP sampling,analysis agricultural products plant origin, 2005.doc 227.5KB 126 downloads
Referring to the attachments above, sa1 illustrates option (a) and the comments i made previously on this aspect are quantitatively presented particularly from pg 434 onward. In contrast sa2 is a detailed look at plans for option (b)( including micro.issues).
As you mentioned, a typical problem is that required sample sizes in published plans are often impractically large, many organisations seem to adjust for this by using a “hotch-potch” of ad hoc plans or semi- statistical schemes, examples sa3 – sa5 look like such approaches, mainly for the estimation of random quality factors.
IMEX for estimation in the field, inspectors often use an intuitive percentage scale based on total lot size and then distribute the number generated proportionately across any sub-varieties within the lot. Other variations based on square roots of lot size have also been long popular. Nonetheless, it can be auditorially useful IMO to use a method “based” on a quotable reference source.
Rgds / Charles.C
PS - and a couple of fairly readable links -
IMEX = In my experience
AQL = Acceptable quality limits.
....thanks Snookie .haha, feel a little silly now :roflmao:
Charles, If I could give you a million thanks/ votes right now I would!! :thumbup:
I've had a quick look at the files and they are exactly what I need, all my other reference info is more generic, and yes it will be great to have an actual reference ..I am reading up on sampling & variance and standard deviations etc, and it's great to have more material...will make for some interesting weekend reading!
Dear Tan,
I'm glad it helped.
The difficulty IMEX is that most textbooks on sampling and statistics are only readily understood by Statisticians and Probabilists. :smile:
The theoretical problem unfortunately rapidly gets more complex when you also have heterogeneous products like mineral ores and fresh produce. Hence the approximations which are often slid in between the lines without much fanfare or explanation but are often critical.
A lot of the original theory was aimed at controlled engineering production lines where more homogeneous systems existed so that large samples were less unrealistic but they tend to collapse when you have a truckload of unpacked raw food in front of you. :smile: For the AQL aspect, one particular area of interest is the "S" Grades where the sample sizes sudddenly become more realistic (eg <10) for many lots. Another complexity is where 2-stage sampling occurs, eg sampling X inner units from each of Y outer cartons in a total lot of Z outers. There is a lot of theory (2-stage sampling) for "general X" but I still recall being amazed when i saw a proof that if X=1, the equation is the same as direct 1-stage sampling. Intuitively obvious maybe (?) but mathematically less so IMO.
One fortunate thing is that IMEX, even if you only have a basic understanding of sampling theory, you will probably know more than most auditors. In some cases a lot more. :smile:
Rgds / Charles.C