Sampling Problem

Six Sigma – iSixSigma Forums General Forums General Sampling Problem

Viewing 7 posts - 1 through 7 (of 7 total)
  • Author
  • #54052



    I really hope that someone can help me with a problem I am facing. (I know that I should know how to do this, but…)

    I have a process with a measurement on the process outputs (products)and the criteria for an ok measurement.

    Now I am starting a new process to make the same product and I can’t seem to be able to define how big a sample I need. The criteria for accepting the new process is that yield shall be over 93%. I know that one way I can do it is to treat it as attributes, but there must be some way to use the continuous data that I have.

    I’m stuck in MiniTab’s power and sample size, but I realise that the answer isn’t in there since I can’t define neither “sample size” nor “difference”.

    Any information would be very helpful!




    Can you better define what kind of data you do have and what defines your yield? Usually yield is a percentage.


    Chris Seider


    Think of the difference as to the precision of how you want to describe your process mean yield. Would you want to know the process yield within +/- X% of the true yield when you only sample. You will need to know the standard deviation of the yields from a process capability study.


    Thomas Whitney

    How about looking up a simple outgoing quality sampling plan. These are not taught so much anymore but they certainly work. Minitab has a selection for developing good sampling plans. Try it out. If you get stuck, I will help you further, but I am a really hard teacher and will make you work!



    The outgoing quality sampling plan was a really good idea – thanks!

    The product being produced is electronical and fairly complex so we are dealing with small series. Now I am trying to determine, not if a specific batch is ok, but instead if this new process is capable of making this product at a reasonable quality level going forward.

    Historically we have been just above 93% in yield (approved/not approved). I have dug into the data and found the numbers that make it approved/not approved so it is continuous data. However when I try to use the outgoing sampling plan feature in MiniTab I get really “penalised” for having an AQL that is that close to the RQL. The number of devices gets ridiculously high, much higher than a traditional batch size.

    Any ideas?



    For small volumes you could check 100%.
    If these are custom electronics they probably get tested 100% and the price will probably also be fairly high.


    Mike Paulonis

    Based on the information here, it seems that you are dealing with an attribute capability analysis. The way it sounds to me is that your continuous measurements result in a pass/fail assessment for each individual item produced.

    Not knowing what your “traditional batch size” is, let me use 100 as an example.

    If you have data like this:

    Batch Count Defective
    A 100 5
    B 102 6
    C 97 5
    D 99 4

    you can use something like a binomial capability analysis in Minitab (the version of this in the Assistant in Minitab 16 is very nice) to come up with a 95% CI on the % defective. If the upper limit of this CI is greater than 7, your process is not capable of 93% yield.

    Note that Minitab recommends at least 25 subgroups (rows in the data table above) to be sure that long-term process variability is well represented.

    With this kind of problem, there is just no way that you will be able to determine the process capability without a significant amount of production data to work with.

Viewing 7 posts - 1 through 7 (of 7 total)

You must be logged in to reply to this topic.