non parametric partial correlations

Home Forums Methodspace discussion non parametric partial correlations

Viewing 3 posts - 1 through 3 (of 3 total)
  • Author
    Posts
  • #923
    maria
    Member

    I have a stats problem and hope someone out there can help.

    I have 4 variables, most of which are non normal.  I wanted to do an ANCOVA but as the data was non-normal I ran a Quades analysis.

    Now I need to run some partial correlations.  I can run the bivariate correlations using Kendall’s tau on SPSS but am not sure what to do about the partial correlations?

    I have a small sample (n=55) with lots of tied scores.

    thanks

    maria

    #925
    Anonymous
    Inactive

    yes, there are extensions for the concept of ‘partial correlation’ for using Kendall’s tau/Spearman’s rho. but i doubt they can be calculated by default in SPSS. 

    now, when you say you have 4 variables which are non-normal… what do you exactly mean? please bare in mind that the distributional assumption of normality is on the residuals of the linear model and not the variables themselves

    #924
    Stephen Gorard
    Participant

    Or use regression/correlation based on mean absolute deviations (from either the mean or the median). These make no assumptions about normal distributions or even co-linearity. 

    ANCOVA is a glorified significance test. Avoid like the plague. My guess is you do not have a complete random sample anyway. Apologies if you do (it would be the first I have ever encountered). 

Viewing 3 posts - 1 through 3 (of 3 total)
  • You must be logged in to reply to this topic.