iSixSigma

Hypothesis Test for Attribute Data? Help

Six Sigma – iSixSigma Forums Old Forums General Hypothesis Test for Attribute Data? Help

Viewing 17 posts - 1 through 17 (of 17 total)
  • Author
    Posts
  • #34826

    Marz
    Participant

    I have ppm defect data for twelve months.  I want to compare the performance before and after some team activities.  I am using a two sample t test.  Is that correct?

    0
    #96466

    Bill W
    Participant

    Hello, it looks as if you are comparing proportions here. I would suggest a Chi-Square for this. T-test are best used on variable data.

    0
    #96478

    GDW
    Participant

    Are you trying to compare the defect % in month 1, compared to the defect % in month 2 after you’ve made improvements?  If so, you should use a 2 proportion test.

    0
    #96480

    Statman
    Member

    Lets see, sequential months of data and I want to know if there has been a shift in the process……I don’t know maybe I could use a
    CONTROL CHART
    Try either a C-chart or a U-chart.
    And fire the trainer that led you to believe that this is an application of hypithesis testing.
    Statman

    0
    #96493

    Bob J
    Participant

    From my experience GDW is right….  I would use a two proportions test (aggregate before and after)…
    Hope this helps…
    Bob J

    0
    #96498

    Gabriel
    Participant

    Very well said.
    Just thinking loudly, my only concern is that the standard control limits are set to avoid overreacting on process “common” variation. That is, the focus is put in a good type I error, not in agood type II. Specially in this case, with only 12 points and a known change in the middle, it could be difficult to set accurate control limits. Maybe using 2 sigmas control limits instead of 3 sigma ones can help. Ok, there will be more risk to detect a change in behavior where there was no one, but there will also be a smaller risk to fail to detect a change where there was one. You don’t need to prove thae change with a 99.7% of confidence before implementing it.

    0
    #96500

    Statman
    Member

    Gabriel,
     
    Yes, the control limits are conservative and biased towards preventing a type I error.  This of course makes sense we have a situation where the cost of adjustment to a process is high relative to the cost of small shifts in the process.  We tend to treat these alpha risks as if they were handed down by God and surely, I think we could agree that God likes 0.05 as much .0027. 
     
    Your suggestion of increasing the alpha risk to a 2 sigma decision line is one option.  One way to accomplish this is using EWMA or CUSUM type of charts.  Another option is to increase the number of data points by decreasing the area of opportunity.  In other words, look at defects per week rather than per month.  This will give us 52 data points and higher confidence in detecting a shift when using the run rules.  If the defect rate is very small than using an exponential control chart and assessing units between occurrences is another option.
     
    My objection is in using traditional hypothesis testing (an enumerative method) in an analytic application.  In hypothesis testing, we decide whether the unusual values are simply different because of random sampling error or they because they are truly different from others. Reference distributions have been developed that tell us what the probability of this sampling error is in a random sample obtained from a population.  In order to apply this principle to a sequential sample from an ongoing process, there must be the assurance that the process is in statistical control.  In order to assure the process is in statistical control, one must control chart the process.  If you are control charting the process, then hypothesis testing is, at best redundant and can be misleading. 
     
    Also, how do we determine the before and after sampling frame?  We could continue to increase the before by going back in history until we have reached a sample size that will make any change, real or random, significant.  This makes the practice irrelevant.
     
    I feel like I have posted this before. 
     
    Cheers,
     
    Statman

    0
    #96508

    Marz
    Participant

    Statman
    I’ve been using a p-chart you jerk because my sample size is different each time.
    Not having much experience with attribute data I was wondering how I can tell if I’m really seeing an improvement or if it just random variation. 
    Do I do some form of hypothesis test (hypithesis – you know how to spell?) or perhaps change ppm to a percent and do an ANOM.
    I don’t know the answer, I thought I used this forum to ask for advice, not a lecture.
    Kevin

    0
    #96509

    Mikel
    Member

    Try test of proportions. You will find it in Minitab under stat>basic stats.
    If you are unclear what to do the help menus are great or post your twelve months of data here (including the variable sample sizes) and I’ll show you what to do.
    And be gentler with Statman, calling him a jerk hurt my feelings – that is a title usually reserved for Stan.

    0
    #96511

    Statman
    Member

    I gave you an answer.  If you don’t like it follow the other advise that you got.  It appears that you do not know enough about what you are doing to follow a lecture.
    What is your P-chart telling you?  How are you using a P-chart if you have varing sample size? Shouldn’t you use an np-chart for varing sample size?
    Statman

    0
    #96512

    Bob M.
    Participant

    L’ego your ego.

    0
    #96513

    Marz
    Participant

    Statman
    I’m sensing a little hostility here, you gotta let it go man.
    I wouldn’t say I don’t know what I’m doing seeing I’m getting paid to work on this stuff.  I’m using a p-chart because I’m dealing with defective attribute data and variable sample size.
    I don’t have 9 consecutive points descending, and I don’t have 6 consecutive points below the mean.
    Kevin
     

    0
    #96514

    Marz
    Participant

    Stan
    Thanks I’ll try it.
    Kevin

    0
    #96900

    Chris Seider
    Participant

    Kevin, in the rush to lessen the hostility, I believe you got the two numbers reversed–according to the Western Electric rules.  You would be highly interested if 9 consecutive points were above or below that historical mean or 6 points consecutively increasing or decreasing from a single point–many think this means 6 points but one must have a relative line of 7 points since one must have a reference to start increasing or decreasing from.
     
    I want to also say Statman shouldn’t have been called a jerk in an earlier post by someone else.  First, a public, fairly confidential, forum leaves room for this mischief and second, Statman posts many points of good advice–or at least advice that should be considered.  No I don’t always agree with Statman myself but appreciate the time he puts into the discussion board.  If people are worried about being called names on a board, they probably aren’t going to post their experience and the usefulness of this forum might drop off from lack of insightful thoughts.

    0
    #96904

    Ken Feldman
    Participant

    Statman, guess u r a little flustered by the name calling.  Based on previous posts I know u know that varying defective sample sizes requires a p chart not a np chart.

    0
    #96938

    S K Nath
    Member

    2- sample t test is applicable if the data (i.e. defects in ppm) follows normal distribution. If the data are non-normal, then moods median test  need be conducted to check the performance before and after the project.

    0
    #97120

    JackG.
    Participant

    If you historical data on good vs. bad, then the proportion test is the way to go.  Often we overlook this test in favor of CHI square test. 

    0
Viewing 17 posts - 1 through 17 (of 17 total)

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