+ Reply to Thread
Results 1 to 2 of 2

Thread: Normally/non-normally distributed data - initial significance testing

  1. #1
    Points: 4, Level: 1
    Level completed: 7%, Points required for next Level: 46

    Posts
    1
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Thumbs up Normally/non-normally distributed data - initial significance testing




    Hi all,

    I would be grateful if anyone can direct me please.

    I am looking to 1. look at predictors of my DV (when controlling for confounders) and 2. to investigate a potential mediating effect of one IV between my other IV and DV. N = 54.

    - I assessed each variable's distribution using z-scores. For those not normally distributed, I log transformed them and re-examined z-scores. This worked in some instances but not for others. Square root transformations did not work for the others either. This means I was left with two measures with non-normally distributed data (one an IV and the other a confounder).

    - I want to run initial correlations between all variables to see which are significant and would therefore be added into the multiple regression. This is where Iím a bit stuck. I know I can run Pearsonís correlations on all normally distributed variables and those log transformed but what should I do with the others? It seems strange, somehow, to run separate Spearmanís Rho correlations for these two, with all other variables. Iíve noticed SPSS gives an option to bootstrap data for correlations and wondered if this would be a better approach?

    Thanks http://www.talkstats.com/images/icons/icon14.png

  2. #2
    TS Contributor
    Points: 18,889, Level: 87
    Level completed: 8%, Points required for next Level: 461
    CowboyBear's Avatar
    Location
    New Zealand
    Posts
    2,062
    Thanks
    121
    Thanked 427 Times in 328 Posts

    Re: Normally/non-normally distributed data - initial significance testing


    - I want to run initial correlations between all variables to see which are significant and would therefore be added into the multiple regression.
    This is probably not the best approach to select variables. A variable may have a significant bivariate correlation with the DV, but not be a significant predictor in a multiple regression, and vice versa. It's probably better to specify your main regression model using existing knowledge and theory in the area. Regression of course does not assume that your variables are normally distributed. A paper some of us wrote about this: http://www.pareonline.net/getvn.asp?v=18&n=11

    If you are interested in looking at correlations for some other reason, using untransformed variables and bootstrapping sounds fine.

+ 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