0
$\begingroup$

I'm quite new to this topic, so this is probably very basic for most of you:

I want to analyse the relationship of multiple variables via partial correlation as an alternative for a regression analysis. I don't only want to find out how the independent variables relate to the dependent variable but also I want to analyse the relationship between the ivs.

I have one dichotomous dv and eight ivs that are both qualitative and quantitative. Is it possible to run a partial correlation analysis though the variables have different scales? Is there something I need to consider? For example: do I have to test two-tailed or one-tailed? Do I have to run bivariate correlation/chi square tests too?

$\endgroup$
1
$\begingroup$

Jennifer, first of all, to me it seems that you should be more clear about your substantive question. Are you interested in the causal effect of one IV on the DV, or in predicting the DV, or just in the correlations between all your variables (and why would that be?).

Then, partial correlation is not an "alternative" to multiple regression. The coefficients of a multiple regression are directly linked to partial correlations! See here: Multiple regression or partial correlation coefficient? And relations between the two

$\endgroup$
  • $\begingroup$ Well.... my plan was to do a logistic regression analysis but since my sample size is too small it doesn't work out to do that properly because SPSS just won't let me analyse interactions and so on. So I thought another way to do that is using partical correlations. $\endgroup$ – Jennifer Feb 14 '14 at 13:40
  • $\begingroup$ Coefficients of logistic regressions are also only some sort of partial correlations. If you have so few data, you cannot squeeze them endlessly anyway; why not look at some descriptive graphics and estimate models withouth the interactions and clearly state in the final presentation how limited the analysis is? $\endgroup$ – Julian Schuessler Feb 14 '14 at 14:15
  • $\begingroup$ Of course I will have to discuss the weakness of my small sample size. But even though - wouldn't correlations give me a bit more information than graphics? I'm really struggeling with statistics in general, so I don't even know what kind of graphics are best to use in that case. $\endgroup$ – Jennifer Feb 15 '14 at 12:59
0
$\begingroup$

Why partial correlations?

In any case, there are different correlations for different combinations of types of variables:

With one dichotomous and one continuous, you can use point-biserial correlation.

With both continuous you can use Pearson or Spearman correlation, depending on whether you want to use ranks or actual numbers (which depends on the shape of the distribution of the variables)

With both qualitative, correlation is meaningless, but you could look at chi-square.

Next: Correlation will cope with different scales

One-tailed vs. two-tailed is something only you can answer.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.