Attribute data and sample size

Six Sigma – iSixSigma Forums Old Forums General Attribute data and sample size

Viewing 5 posts - 1 through 5 (of 5 total)
  • Author
  • #28502

    Sam Sivalenka

    I have read about various ways to calculate the sample size for measuring attribute data. When a process ,for a given sample size, shows zero defects should I disregard that sample for calculating my average defective or value of Pbar?
    For example, if I take 20 samples of size 300, and sample # 19 shows sero defects does the sample need to be removed when I caluculate the grand average?i.e, divide by 19?
    All feedback is appreciated.


    Dave Strouse

    Sam –
    What reason to you have to cull the data?
    Would you not use a sample that was  for example 100% defective. How about one that is 1% defective? Where is your cutoff to decide what to retain?
    If you know of no reason that the sample was abnormally taken, you should use it. If it was abnormal, e.g. the metrology was acting up at the time and gave bad data, then you could justify not using it. Otherwise, “it is what it is”.
    I’m curious to find out what makes you think you would not use it. 



    I hope you are not suggesting that you are going to average the individual proportion of pass/fail data to come up with a Pbar for the sampling.  It’s not a good ideal  (habit to get into) to ever average proportional data.  Basically if the data collection method is sound, you have what you have and your Pbar (grand proportion defective) is simply the sum of the defect count / total samples inspected (in the example 6000).  If you are using the data in an attribute control chart, then you calculate the control limits based on that proportion and look for special cause if any.



    I agree with some views expressed above.
    To your pointed question, you would not ignore.
    Pbar for each sample is an indication of the defect proportion in that sample.Also evaluate what are the conditions in which these samples are getting picked up. Are u using Pbar to track by shift or by product variations, etcAs long as the purpose is clear, you dont need to eliminate any data.i guess the confusion comes from the fact that we dont use Capaility Sigma when one has zero defects.
    but in this case, go ahead and use the data.



    hi there
    based on my understanding, when it comes to attribute sampling, big4s always have their magical sample sizes ranging 24-27 at 90% confidence level, 5% sampling risk & 2.5% error rate
    likewise, i found a sampling worksheet (exel) which suggests exactly the same as above though i wish that some guy can help to ensure that formulas therein hold water!
    lotsa thx for reading, let alone responses

Viewing 5 posts - 1 through 5 (of 5 total)

The forum ‘General’ is closed to new topics and replies.