Hierarchical Multiple Regression VS Crosstabs

Home Forums Methodspace discussion Hierarchical Multiple Regression VS Crosstabs

Viewing 4 posts - 1 through 4 (of 4 total)
  • Author
  • #1381

    Hi Everyone,

    My thesis is due in a week and I’m still unsure whether I should report on the parametric test (hierarchical multiple regression) or non-parametric test (cross tabs – is this even right????). I’ve ran both.

    I’m considering to go with the non-para as my data does not fulfil some assumptions for the hierarchical multiple regression. I’m sorry if my understanding of stats sounds too basic (well, it is!) but here is what the textbooks say I have wrong:

    > the residuals aren’t normally distributed against the predicted values (illustrates non-linearity)

    > the residuals on the P-P plot of regression standardised residuals do not tightly cling to the diagonal line

    However, there were no outliers present and the Mahalanobis Distance is fine. So two of the four assumptions is violated while the other two are ok.

    Do I stick to the hierarchical regression? Or do you advise me to report on the Cross tabs instead?

    Any help will be MUCH appreciated!


    Dave Collingridge

    Crosstabs or Chi-square test for independence as it is sometimes called, is for categorical variables. Multiple regression is different in that the outcome/dependent variable is continuous, not categorical. So they are really two separate analyses and I have a difficult time seeing how they can accomplish the same thing, unless you’ve reduced your outcome variable into categories when running the crosstabs.

    If your regression does not show normality of residuals, you should consider transforming the outcome variable with something like a log transformation and then re-run the regression and check assumptions. If this works and you need someone to help you interpret the coefficients, let me know and I can provide you with a formula that will allow you to properly interpret the coefficients. If you don’t want to do a transformation then use your failed normality of residuals as a rationale for skipping the regression in favor of other analyses.


    Thanks so much David!!

    I tried a few different transformations but my variable remained skewed. How inconvenient!

    My thesis is due in two days so I don’t have the time to do another analysis, so I shall have to stick to the regression. Do you know of a reference I can cite as rationale for proceeding with the regression with the failed normality of residuals?  


    I’m new to method space, but just wanted to say that I found this discussion very useful

Viewing 4 posts - 1 through 4 (of 4 total)
  • You must be logged in to reply to this topic.