iSixSigma

Process Capability

Six Sigma – iSixSigma Forums Old Forums General Process Capability

Viewing 24 posts - 1 through 24 (of 24 total)
  • Author
    Posts
  • #39847

    abans
    Participant

    Hi there,
                I would appreciate if anybody would let me know how to do capability analysis on attribute data. Is there some formula like continuous data????
    Thanks.
     
     
     

    0
    #122254

    Dr. Scott
    Participant

    Abans,
    What sort of attribute data are you using? Poisson or binomial?
    Dr. Scott

    0
    #122255

    abans
    Participant

    Dr. Scott,
    Thanks for replying…..I am sure its not Poisson, its binomial data, but not normally distributed.
    Thanks
     
     

    0
    #122256

    Dr. Scott
    Participant

    Abans,
    And are you using Minitab?
    Dr. Scott

    0
    #122257

    abans
    Participant

    Dr. Scott,
                No, I am not using MINITAB…That’s the problem…guide me
    Thanks

    0
    #122259

    Dr. Scott
    Participant

    Abans,
    Do you have constant or changing sample sizes? From what you have said so far, I assume you are looking at percent defectives given either a constant or changing sample size.
    Dr. Scott

    0
    #122260

    abans
    Participant

    Dr. Scott,
    Sample size is same all the time. Actually, I have to find out variablity in the cleaning process of one of the Radisson hotels in my area. I have data on daily basis-Total Number of rooms cleaned and uncleaned. I guess here my sample size would be total number of days for the data I have. I wanted to use control chart, but cannot as I don’t have per housekeeper data. So, I decided to go for attribute chart with percentage of clean number or rooms as my data. Just guide me if I am on the right track….I would appreciate it.
    Thanks.

    0
    #122261

    Dr. Scott
    Participant

    Abans,
    Sample size is same all the time. Actually, I have to find out variablity in the cleaning process of one of the Radisson hotels in my area. I have data on daily basis-Total Number of rooms cleaned and uncleaned.
    Your sample size then is the total number of rooms daily TOO be cleaned (cleaned plus uncleaned). Your defective is the number of rooms that day that were not cleaned.
    I guess here my sample size would be total number of days for the data I have.
    No, the sample size is as I described above. The total number of rooms that were supposed to be cleaned daily. This is not likely to be constant. The mean is the total cleaned divided by the total not cleaned across all of the days.
    I wanted to use control chart, but cannot as I don’t have per housekeeper data.
    I am not understanding why this would make a difference (at this point) if you are comparing facilities and not people.
    So, I decided to go for attribute chart with percentage of clean number or rooms as my data. Just guide me if I am on the right track….I would appreciate it.
    You can do a P-chart for each facility on daily data, clean rooms divided by total rooms to clean. This will give you a way to compare.
    As to your original question, you can simply take the percent defective (number of uncleaned rooms divided by total rooms to be cleaned) and translate it into a Z-score or capability index (e.g., Cp, Cpl, Cpu, or Cpk which I assume will be the same as Cpl).
    You really don’t seem to need a Capability analysis so much as a way to compare. Let me know if I am wrong. I really need more to know about what you want to do.
    Hope I can help,
    Dr. Scott

    0
    #122263

    abans
    Participant

    Dr. Scott,
     I guess I was doing the wrong way…I was calculating proportionate number of rooms cleaned by dividing total number of rooms cleaned by total number of rooms and then I was plotting those points on the graphs (run chart and p chart). It means I can find the mean of each day by cleaned/uncleaned and then use those points on the graph. As I have got mean for all the days, can I use Control Charts as I am more comfortable in control chart instead of attribute p chart. I was not using it before because I was not able to figure out how to find mean on daily basis as in control charts we use means as our plotted points…If I am right. Anyways, I will try the way you suggested and then let you know the results I came up with.
    I really appreciate your reply. As I need to finish this project in a day or so.
    Thanks again
     

    0
    #122272

    A.S.
    Participant

    If you are considering the total no of rooms to be cleaned (both occupied and vacant) as your subgroup,you can use NP chart instead P chart since your Sample size is constant.Hope that no of rooms to be cleaned is not varying on daily basis.No of rooms not cleaned can eb considered as Defective.
    Through control chart you can ensure the stability of your process and you can identify the special causes.In my view it is adequate for this process.
    Talking about Process capability is more useful when you have variable data and your process is just meeting the specifications.If you have more non conformities,doing process capability study won’t give you the fruitful results.Monitoring the control charts will help you
    anbu
       

    0
    #122276

    abans
    Participant
    #122277

    Anurag
    Participant

    Dear,
    1) If u have drawn the C.C for the attribute then, process capability is the mean value i.e. Control Line (C.L)
    2) its very interesting to do analysis on attribute data
    a) Binomial distributaries is an discreet distribution, it has two outcome “success ” or “failure” i.e.  pass or fail,
    p and np chart follow binomial distribution
    p-chart : proportion of defective items per sample
     
    np-chart : number of defective items per sample
     
    But there are many instances where an item will contain nonconformities but the item itself is not classified as nonconforming, In that case  number of nonconformities in a given area can be modeled by the Poisson distribution
    But there are many instances where an item will contain nonconformities but the item itself is not classified as nonconforming, In that case number of  nonconformities in a given area can be modeled by the Poisson distribution
     
     If u deal with this kind of situation ( use c/u chart), or where u can find u can categorize the abnormality in major ,minor or critical u go for capital U chart ( U-chart)
     

    0
    #122278

    Aries Suhariyono
    Participant

    Dear,
    You can calculate the attribute data with this formula :
    Sigma Level =0.8406 + square root (29.37-2.221 x ln (DPMO)
    DPMO = defect per million opportunity
    or you can calculate sigma level with the Microsoft EXCEL :
    Sigma Level = NORMSINV ((1000000-DPMO)/1000000)+1.5
     
    If you have any question please do not hesitate to email me.
     
    Regards,
    Aries Suhariyono

    0
    #122287

    Whitehurst
    Participant

    Hi Aries
    The formula u mentioned I think is wrong the correct one is
    =NORMSINV(1- DPMO /1000000) + 1.5
    Where  Zhift = 1.5 is added
    Joe

    0
    #122290

    Whitehurst
    Participant

    https://www.isixsigma.com/forum/showmessage.asp?messageID=47790 
    Please use this link to know more about attribute data process capability

    0
    #122310

    abans
    Participant

    Thanks Suhariyono for your reply.
    But how to find DPMO in an excel sheet. Have you seen my data on which I am working. What this Sigmalevel will tell me about my data once I calculate it. Guide me.
    Thanks again
     

    0
    #122312

    Dr. Scott
    Participant

    Abans,
    If you are considering the total number of dirty rooms as your defective and the total number of rooms as your area of opportunity, then simply find your percent dirty by dividing the total number found dirty by the total number of rooms inspected (which is 2434/3510=.693447) and multiply this by 1,000,000. So your DPMO or more accurately PPM Defective is 693,447.
    To get the Z-value or sigma level for this process from Excel, merely use the formula “-1*(normsinv(.693447)”. Your Z-value or sigma level will be a poor -.50565 (not good at all).
    One side note: A P-chart or NP-chart will not do you much good when your denominator (N or total number of rooms) is so large. It makes everything look like a special cause. An I-MR chart will likely be more meaningful for you. It shows that you are consistently very bad.
    Also, I noted (not surprisingly) that as the number of occupied rooms rises so does the number of dirty rooms:
    The relationship between Occupied Rooms and Total Dirty Rooms is R-square=48.7, p-value=.004.
    The relationship between Occupied Rooms and Dirty Occupied Rooms is R-square=86.1, p-value=.000.
    There is no relationship between Occupied Rooms and Dirty Vaccant Rooms (R-square=3.1, p-value=.533).
    So it would simply appear that the busier you get, the more dirty rooms you leave behind. Not surprising. But not at all good either.
    Hope this helps,
    Dr. Scott

    0
    #122356

    abans
    Participant

    Dr. Scott,
    I am confused about the mean: how come total rooms cleaned divided by  total  uncleaned will give me mean of each day.
    Thanks.

    0
    #122358

    abans
    Participant

    Dr. Scott,
      Now I am trying to make XbarR Control chart of the same data and I calculated Xbar that is average as you said: Total number of cleaned rooms divided by uncleaned rooms. Now my range would be for each day maximum – minimum, but here I have only twovariables which are cleaned and uncleaned rooms, so I guess I will take the difference of those two as a range.
    Please suggest me.
    Thanks again
    Anubha

    0
    #122359

    Dr. Scott
    Participant

    Abans,
    It won’t. I believe I said the following in my previous post:
    If you are considering the total number of dirty rooms as your defective and the total number of rooms as your area of opportunity, then simply find your percent dirty by dividing the total number found dirty by the total number of rooms inspected (which is 2434/3510=.693447) and multiply this by 1,000,000. So your DPMO or more accurately PPM Defective is 693,447.
    Hope this helps,
    Dr. Scott

    0
    #122360

    Dr. Scott
    Participant

    Abans,
    Again, read my previous instructions again. It is total dirty divided by total number of rooms.
    I would not use and XbarR chart. Use an I-MR. You can find instructions on doing this most anywhere, including on isixsigma.
    Dr. Scott

    0
    #122361

    abans
    Participant

    Thanks. I will try to do I-MR now. I hope it will do any better to me.
    Thanks for your patience with me.

    0
    #122362

    Dr. Scott
    Participant

    Abans,
    An I-MR chart for the total of rooms dirty each day will yield the following:
    UCL = 319.6
    Mean = 162.3
    LCL = 4.972
    An I-MR chart for the proportion of rooms dirty each day will look the same but have the following parameters:
    UCL = 1.366
    Mean = .6934
    LCL = .02125
    Note that the upper control limit in both cases is not possible. Since you only have 234 total rooms you can’t have 319 of them not clean. This miscalculation is the result of the data not being normally distributed. However, ways to correct for this in this instance aren’t feasible since you have only 15 data points (days of data) to consider.
    Dr. Scott
     

    0
    #125942

    Manians
    Participant

    If you do not use MINITAB, then for calculating process capability of an attribute data, you can first findout the defects per unit(proportion). Later apply this in the formula to get the Z-score(Short term), which is the process capability of your data.
     Z Score = Normsinv(1-dpu)+1.5
    The more the Z-score (over 2+), the capability of process is better. Correct me if I’m wrong.
    Manians
     

    0
Viewing 24 posts - 1 through 24 (of 24 total)

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