iSixSigma

Attirbute Data Analysis – Question

Six Sigma – iSixSigma Forums Old Forums General Attirbute Data Analysis – Question

Viewing 5 posts - 1 through 5 (of 5 total)
  • Author
    Posts
  • #41127

    Yadav
    Member

    Hi
    I have attribute data (y/n) as a response and have several predictor variables (also y/n) but the data set is not complete. That is if I include all the predictor variables I won’t have a value for every data point.  Can I break the groups down to a few subset sets of complete data and use binary linear regression on each?
    Would the results still be valid?
    The goal is to understand if my predictors are significant….
    Tina 
     

    0
    #128538

    Robert Butler
    Participant

      I think you need to provide some clarification before anyone can offer any meaningful thoughts.  As written you give the impression that you either have a block of data with a group of X variables that are independent(?) of each other and that some of the “experiments” failed or that you have a block of data with a group of X’s (again independence is assumed) and that some of the responses are simply non-responses (as in a survey).
      If it is the first case then the question you need to address is this – given the data for which you have both X’s and Y’s does the resultant matrix of X’s still provide enough separation so that the X’s are still reasonably independent of one another?  If this is the case then you can get on with the analysis.  If it isn’t then you need to identify those X’s which cannot be examined because of the resulting colinearity.
      If it is the second case then instead of a y/n response you now have three responses yes, no, and no response offered.  Non-response is not the same as missing data and if it is treated as such the final conclusions will be in error.

    0
    #128543

    Yadav
    Member

    Thanks for your feedback.It’s not an experiment.  So it would fall under a survey type response. However, it’s not a no response, but not rather that piece of information was not available for some of the data collected.  Does that make sense?

    0
    #128566

    Robert Butler
    Participant

    If I understand what you are saying then do you have a spread sheet that looks like this
    y/n      X1     X2     X3
    y         1        -1       -1
    .        -1        -1         1
    n         1         1         1
    n       -1         1        -1
    y       -1        -1         1
    where you are missing the occasional data point or is it like this:
    y/n      X1     X2     X3
    .         1        -1       -1
    n       -1        -1         1
    .         1         1         1
    n       -1         1        -1
    y       -1        -1         1
     where your missing responses correspond to having completely missed one of the two levels of a given X?
      If it is the first case then, as mentioned previously, you need to see how that impacts the co-linearity of your X’s.  If it is the latter then you have lost the ability to run an analysis with X1 present since responses for X1 are either present at a single level of X1 or they are missing.

    0
    #128572

    Yadav
    Member

    I acutally had data like
    y/n   x1 x2
    1     1
    0     1   0
    1      0   1
    But I have since been able to resolve the missing information in x2, so it’s working…..thanks for your help.

    0
Viewing 5 posts - 1 through 5 (of 5 total)

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