iSixSigma

Atribute data analysis

Six Sigma – iSixSigma Forums Old Forums General Atribute data analysis

Viewing 7 posts - 1 through 7 (of 7 total)
  • Author
    Posts
  • #46607

    Rob Duivis
    Member

    I am working a project to evaluate and improve test rejects. Over the last 5 years we had 78 rejects over 1000 produced units. About 200 units produced per year. Gives

    0
    #154319

    Rob Duivis
    Member

    Let me finish my earlier message; i want to analyze my atribute defectives data of 78 rejected units in 5 year of 1000 tested units. Approx. 16 rejects per 200 units produced a year. Calculating FTY etc. is not the issue, but can we make a capability plot of this data and calculate Z lt, Zst, Z-Bench, and a Z-shift? Assume data is binomial distributed?

    0
    #154323

    jberillajr
    Participant

    Rob,
    At first blush I would tell you this:Z score is a  measure of the difference in standard deviations of a sample from the mean:   (X – X bar) / sigma.
    Hence, it requires continuous data that conforms to the assumption of normality (As do Cp, Cpk, Pk, Ppk, Cpm, and all things ´Z´).  Since you are using porportional data (ie attribute data), this is not an option for you, nor are any of the other ¨Z¨calcualations.
    Here is the part you want to double-check, as it has been awhile:  You can take your yield measures and normalize them out and hence, back your way into a process capability metric….although many will not like this approach (including myself). 
    I like to always use real process output to identify a yield position as a baseline (FTY, RTY, DPU or DPMO), but concentrate on the statistical difference in a continuous, independent variable.  I find it is almost always possible and desireable to work with project Y that is continuous, limiting attribute data for stratification elements within the data collection plan. Others experience may differ so remember:  
    Trust but verify.

    0
    #154359

    The Force
    Member

    Whether your data is binomial (defectives) or poisson (defects), you can compute for Z. The only difference is that when you use mtb for the binomial, it automatically computes for z but for poisson, you need to estimate %defectives from the mean dpu, covert it to ppm then estimate Z.

    0
    #154363

    Mikel
    Member

    Who cares about Z lt, Zst, Z-Bench, and a Z-shift? What you have is
    long term data (the only kind of attribute data that makes any sense.
    You can convert that to a z score if you want, but why? The rest of the
    scores are based on a bunch of silly assumptions. You have an 8%
    defect rate – go improve it and forget about the rest of the nonsense.

    0
    #154387

    accrington
    Participant

    Stan, I think you’re flogging a dead horse. Most of the posters still ask about DPMO, capability indices, 3.4 sigma shifts, sigma levels, arbitrary targets and similar mumbo – jumbo. I don’t think they will ever get what process improvement is about (i.e hard work, understanding, knowledge, learning, etc.)

    0
    #154398

    Allthingsidiot O
    Participant

    Consider the  relation :Ppk = Zmin./3

    0
Viewing 7 posts - 1 through 7 (of 7 total)

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