iSixSigma

Sample size determination

Six Sigma – iSixSigma Forums Old Forums General Sample size determination

Viewing 10 posts - 1 through 10 (of 10 total)
  • Author
    Posts
  • #33344

    Murray
    Participant

    Some plain speaking help is needed!My company issues contracts to all our customers every time they make a transaction with us. We have a customer complaints department which records the percentage of complaints to contracts issued, these complaints occurring when customers contact us to let us know there is an error.We would like to check these contracts before they are sent out to try to eliminate errors before the contracts are sent to the customers, but since we produce a large number of contracts, it is not possible for us to check all of them. How do I determine how many contracts we need to check in order to ‘catch’ most of the errors before it is too late?I’m not statistically minded, so would appreciate any responses in plain English.

    0
    #90032

    Ron
    Member

    For those not statistically adept the best document avaialbe was Mil-std-105e. You can obtain a copy by searching the internet.
    There is also an ISO version of this document and the sampling plans are identical.
    What you will find is a listing of quantity of items produced, the quality level you wish to detect, then look at a matrix to select the sample size.
     
    Try it I think you’ll like it. The statistical portion is already done for you. Instructions for utilization are in the document.
     

    0
    #90718

    HUGH
    Participant

    You can also use ANSI/ASQC Z1.4-1993.  The tables are the same as MIL-STD-105E, but the text is a little easier to understand.  I also recommend searching the iSixSigma website for the words “sampling plan” and “AQL.”  You should find a couple of good articles and some definitions that will be helpful in understanding either the MIL or ANSI/ASQC standards.

    0
    #90720

    Mikel
    Member

    The suggestions of MIL-STD-105 are bad.
    Sampling procedures are only useful when the system is not predictable. In your case, it appears that the system is predictable, just not good.
    Read Deming – look at 100% or none. Spend your energy improving the system instead of a sampling placebo.

    0
    #90756

    Gabriel
    Participant

    If you want to catch most of the errors, check most of the contracts. If you want to catch all the errors, check all the contracts. Sampling will not work in this case. (note: with “Error” I mean “a contract with 1 or more errors”)
    I am assuming that errors are made at random (errors are just made from time to time withou a specific trigger and without a specific pattern), then the number of errors found in a sample is ONLY a matter of chance.
    If this is not true and the errors are not made at random, then sampling could make sense. For example, if under specific circumstances many errors are made in a short period of time and in any other circumstance very few errors are made, then sampling may help you find that specific circumstances and fix them. In this case get a copy of the SPC handbook from AIAG (can be ordered from the AIAG’s website and is inexpensive) and try a p chart. But beware cause to use a p chart you need at least 5 errors per sample, If the rate of errors is low, then this could lead to huge sample sizes.

    0
    #90770

    Yashwant M Joshi
    Member

    Dear Friend
    First of all you need to collect and analyse recent past data. Find out which type of errors are most frequent. Then try to discover the root causes for the same with the help of collective wisdom of the team.
    Solutions to avoid errors could be thro Mistake Proofing , training, cretaing awareness and motivation,proper software etc.
    Till such time the solutions are implemented, prepare a checklist with focus on most frequently committed errors discovered thro the above analysis. Initially you may have to inspect 100 % contracts and then go on reducing the number. Finally when the error rate reduces substantially, install an audit system.
     

    0
    #90776

    Titu John
    Member

    Errors in a  contract or a document or for the matter in a book follows Poisson distribution hence the user should take into consideration the properties of Poisson distribution before taking any samples , and there after use relevent statiatical tools.

    0
    #90812

    Cheryl
    Participant

    1.  I believe the first step is determining how many different TYPES of contracts there are.  Are all the contracts somewhat unique or are most of them a “standard” format?
    1(a)  Review several different “unique” ones and record how much time it takes – of course, this will not deduce that all the contracts will take the same time, but will be a good starting point and at least provide you with some idea as far as review time.
    2.  Second step: 
    2(a) if this effort is more of a project with a starting and ending date, determine your timeline in relation to reviewing these contracts.  For example, you should have a target date as to when this initial review will be completed.  This timeline will be a determining factor in how many can actually be reviewed.
    2(b)if this effort will be more of an ongoing effort, determining the average time it takes to review a contract, the number of individuals that will be involved and THEIR available time to review these, will be determining factors as to how many can be reviewed.
    It is very important that you take into consideration the time availability of the resources targeted to review these contracts.  And of course, that they provide input as to the approach used for this effort.
     
     
     
     

    0
    #90820

    Bank
    Participant

    Nick, all of the replies are pretty on target in terms of looking at what the defects are and improving the process, determine which fields of data are important, the fact that you can’t “catch most of the defects unless you 100% inspect(then you will still miss 20% due to human error), etc.  But in the end, after all the improvements, you still have a human process – typing – that you would like to control.  Here’s what we do with our data entry processes: Because there are multiple opportunities for defects per page, we chart the defects on a C-chart.  Based on historical data of defect rates (5%), we sample 50 contracts/month/typist(you have to sample large enough to “see” defects – in this case, we see on ave 50x.05=2.5 defects per sample), and chart both the overall count data, and the individual data entry person.  The results are used as part of an incentive program – those under 6% for the month receive an incentive. This keeps the data entry person focused on the quality of their output, as well as requesting help to improve their results if they miss the incentives.  As a result, we are driving down the error rate

    0
    #90824

    Mikel
    Member

    The 20% due to human error is rhetoric and simply not true. If it were true there would be no need for Attribute R&R.
    Human systems, with error proofing, can be near perfect – certainly better than five sigma.

    0
Viewing 10 posts - 1 through 10 (of 10 total)

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