Data Leveler

Six Sigma – iSixSigma Forums Old Forums General Data Leveler

Viewing 6 posts - 1 through 6 (of 6 total)
  • Author
  • #43001


    Here is my question: 
    I am analyzing attrition data.  I have two different groups of individuals.  Group #1 has a high school diploma and group#2 does not.  The group #1 is a larger population than group #2.  When I try to analyze the data to only look at why people quit, do I need to level the data or is a percentage of the population a stable call?  This is very important and I hope that it will spark some great conversations.



    Depending on your sample sizes, a 2-proportion test should do it for you.  There would be no need to “level” the data.



    Hi Jason,
    I am not a Six Sigma Expert, but have an exposure to the kind of project you have undertaken. In my experience, besides the traits of an individual, more importantly we need to look into all those factors, X’s, that lead an employee to make such a decision. Most of these Xs would the softer elements like job dissatisfaction, conduct of the superiors, general company culture, corporate values, etc. You can capture this data thru mechanisms such as ‘Exit Interviews’ and VOE surveys.
    Also, other key factors (depending on the industry) could be growth opportunities, dynamism in the job itself, competetion, industry compensation packages, etc.
    Hope this helps, best regards,
    – Neo.


    Ravi Bhatia

    I have done a similer project for my organization.. what we found out that apart from softer elements like behaviour of the manager etc.. few factors like number of previous jobs and average number of years / months will show a much better trend… However, one needs to continue to work on improving softer elements.



    Do you know what your Xs are?  If based on the pareto, FMEA & C-E diagram is degree an X driver of Y (associates leaving), or are they 2 distinct categories (Ys) with different Xs.   I think you have to answer these basic questions before you jump into what tool to use.



    Along that same line, you might think about trying to use something like binary logistic regression.  The goal is to find a mathematical relationship to help you predict the probability of a discrete event like Y={“Yes They Left” or “No They Stayed”} as a function of multiple input variables which can be either continuous or discrete:
    Discrete X’s:
    X1={Degree or No Degree},
    X3={HiCompetitionArea=Yes/No…i.e., New York or Silicon Valley vs. Skookumchuck, WA},
    X3={PerformingBoss=High,Medium,Low…you could use 360 degree evals if you have access to that},
    X4={TravelRequired=High,Medium,Low…maybe based on job description?}
    X5={IntellectualStim=Yes or No…cognitive work or redundant transactional work}
    Continuous X’s:
    X6={months on previous job},
    X7={age of employee},
    Of course, the first thing you would need to do to make sure you are getting a clear picture is to identify those individuals who are in your potential span of control to influence – i.e., review the exit interview docs and separate those folks who left due to family relocation, elder care requirements, winning the lottery, etc.  Good luck and have fun analyzing!  Shoot me an email or IM on Yahoo of you need help with Minitab binary logistic regression ([email protected]).

Viewing 6 posts - 1 through 6 (of 6 total)

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