+ Reply to Thread
Results 1 to 4 of 4

Thread: Categorical IV and DV: What test?

  1. #1
    Points: 595, Level: 11
    Level completed: 90%, Points required for next Level: 5

    Posts
    2
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Lightbulb Categorical IV and DV: What test?




    Hi everyone,

    I am running an analysis on some data to test the accuracy of subjects at predicting their own attachment style. Based on test scores and some qualitative data, the researchers have categorised each individual (n=86 for our client sample, and n=70 for a sample of therapists) as being either securely or anxiously attached. This is our DV.
    Each subject has classified themselves as being either securely or anxiously attached (IV) and I want to see if subjects are accurate at predicting their own attachment style at a significant level. Here, I have conducted both a chi-squared and binary logistic analysis but am wondering which, if either, is more appropriate.
    Alternatively, I have run a linear regression on this data but instead of using the research determined overall attachment style (secure or anxious), I have used their Secure or Anxious attachment score (these are two separate scores, not two ends of the same scale).
    Is it possible to include the results of both the linear and logistic analysis?
    What statistics are essential to report for logistic analysis?

    Cheers for your input guys.

    P.S anyone who is working on attachment out there please don't be alarmed by this post, we are actually looking at different scales but I chose these variables for familiarity and ease of use.

  2. #2
    R must die
    Points: 23,888, Level: 94
    Level completed: 54%, Points required for next Level: 462
    Awards:
    Activity Award
    noetsi's Avatar
    Posts
    4,381
    Thanks
    256
    Thanked 688 Times in 663 Posts

    Re: Categorical IV and DV: What test?

    I am not sure you have enough data to run logistic regression (normally you want a hundred cases plus more for each IV). But, if power is not a major issues, then logistic regression would certainly appear appropriate. It will give you something chi square won't, the specific impact the IV has on the DV.

    I don't see how you can run linear regression with this data. While categorical data sometimes is used as a dependent variable (for example Likert scale data) the absolute minimum number of categories is probably 5 (many suggest ten seperate categories or more). It does not seem like you have this.

    You could include the results of linear and logistic regression. But people would probably wonder why you did. Normally if you can run linear regression you do (it is much easier to interpret and has well understood diagnostics which are lacking in many cases for logistic regression or at least are harder to find in commerical software. For example a meaningful Rsquared). If your data does not support it, and I doubt yours does, then you should not do so.

    People vary on what they think are important. Probably the most useful thing to report in logistic regression is the Odds ratio and the Wald test of paramater signficance. Also your overall model signficance test (I think that is a -negative two log liklihood test, but my memory fails me at times). Whatever commerical software you use should tell you where to find this (and what they call them). Some like the pseudo Rsquare tests as well (although not all by any means). Slopes are basically useless in logistic regression other than the sign and whether they are signficant (which is why Odds ratios and relative risk are used instead).

    I suggest looking at "Using Multivariate Statistics" 5th ed by Barbara Tabachnick and Lidna Fidel. They have a chapter on logistic regression including data clean up. They are easy to understand and a wealth of information including on software for methods. Chapter 10 is on logistic regression.

  3. #3
    Points: 595, Level: 11
    Level completed: 90%, Points required for next Level: 5

    Posts
    2
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Re: Categorical IV and DV: What test?

    Thanks noetsi.

    I should have made it clearer that I have two sets of data for my DV.
    One is the categorical (secure or anxious attachment style), and one is the score for each individual on the security scale, and their score on the anxiety scale.
    So for the linear regression I was using say, the security score (continuous) as my DV, and then the self nominated attachment style as a categorical IV.
    Does that make sense?
    The reason I asked about reporting both is linear and logistic regression results is that the logistic regression gave me some nice results overall which don't come out when I regress the actual scores. So maybe it's better just to use the logistic?
    Cheers again for your time.

  4. #4
    TS Contributor
    Points: 9,764, Level: 66
    Level completed: 29%, Points required for next Level: 286
    Karabiner's Avatar
    Location
    Schalke 04, Germany
    Posts
    1,486
    Thanks
    27
    Thanked 370 Times in 352 Posts

    Re: Categorical IV and DV: What test?


    Each subject has classified themselves as being either securely or anxiously attached (IV) and I want to see if subjects are accurate at predicting their own attachment style at a significant level. Here, I have conducted both a chi-squared
    This means a dependent measures design with a dichotomous outcome variable, so
    you need to perform McNemars test.
    and binary logistic analysis but am wondering which, if either, is more appropriate.
    That depends on your research question(s) and theoretical considerations. Interval
    scale measurements are of course much more informative than dichotomies, but
    dichotomies are easy to present to the audience and (hopefully) easy to interpret.

    Regards

    K.
    Last edited by Karabiner; 10-18-2011 at 04:34 PM. Reason: typo

+ Reply to Thread

           




Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts






Advertise on Talk Stats