iSixSigma

Graphing Attrition on a control chart

Six Sigma – iSixSigma Forums Old Forums General Graphing Attrition on a control chart

This topic contains 3 replies, has 3 voices, and was last updated by  MikeP 10 years, 10 months ago.

Viewing 4 posts - 1 through 4 (of 4 total)
  • Author
    Posts
  • #51439

    MJ Spearman
    Participant

    I need to graph Attrition percentage or rate in a control chart, first what type of chart to I use, and is it considered Attribute or Continuous Data?
    Thank you
    MJ Spearman

    0
    #178324

    MikeP
    Participant

    Attrition is attribute data, is analogous to defectives, and the most appropriate control chart would likely be the p-chart.
    A p-chart is assuming a binomial distribution for the underlying data.  You could run a chi-square goodness-of-fit test against a binomial distribution for your data to check that the binomial distribution is an adequate representation.

    0
    #178327

    lin
    Participant

    I disagree on using the p control chart.  The probability of attrition is not the same for all employees.  For example, younger employees tend to switch jobs more often.  Thus, one basic condition for using the binomial distribution for the control limits in the p chart is not met.
    I would use an individuals control chart.

    0
    #178330

    MikeP
    Participant

    I agree that the reality of attrition does not precisely match the conditions for the binomial distribution.  However, it is unlikely to match the conditions for a normal distribution either, making application of the individuals chart questionable.  An individuals chart is not robust to deviations from normality.  For a low percentage of attrition, I would expect the binomial distribution to be a closer fit, recognizing it is unlikely to be perfect.
    It would still be wise to test the data for conformance to a distribution, either binomial or normal, to validate the selection of a control charting approach.  I’m not sure how many samples of data are present, but if there aren’t many, the dataset may pass both tests.  If it doesn’t pass either, I would explore a transformation such as Box-Cox before applying an individuals chart to the data.

    0
Viewing 4 posts - 1 through 4 (of 4 total)

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