Sorry about the the stupid question but I needed to be sure.
I am investigating the relationship between self-compassion and distress, while controlling for bad childhood, stress, and attachment. I got positive correlations for self-compassion and my control variables. So what implications does that have have for self-compassion? Does it mean the results are unclear, because distress could have been caused by any of these factors?
Martyna, my understanding is that if the independent variables are correlated then they both measure some common shared variance, and if you put them both into the regression model, in effect you count the shared variance twice - once for each correlated variable. This is called multicollinearity. As a rule I manage this by limiting the number of independent variables used if correlated. For example, where I work, age and years of education are always correlated because of the stepped introduction of more mandatory years of education in the 1970s. To avoid the problem of multicollinearity, I use either age or education. There are other solutions.
Dear Martyna.
Im the author of the german version of the Culture-Fair Intelligence Test (CFT 20-R) and an expert for test development and a lot of publicatons and many aticles of media psychology.
In the article in appendix I would also see, like my culture fair test (CFT 20-R) has effects for school-achievement in mathematics und german language in comparision whith social or migration states. Both achievement variables are correlated (r= ca. 0.40). I used a hierarchical path analysis. Important is to give not both aim variables in one model, important is to separate aim variables and put in the two separated models not more than 8-10 moderator-variables. These must be found before hypothesis-leaded-not accidently. The best way is to give moderator-variables into a factor-analysis and find the best with good loadings.
Best greetings from Germany-Baden-Württemberg
Yours Dr. Rudolf Weiss
No question is stupid, a person I know once told me. No need to apologize for our questions, that is :-)
Back to your question now. A regression analysis is conducted exactly to resolve the problem that you are dealing with. If A correlates with C and B correlates with C, would A and B still correlate with C if we take them both into account?
Given your description, I would say that we cannot be sure if your correlations are giving you trouble. What I would suggest doing is, first, calculate a bivariate correlation matrix and check for the sizes of the correlations between your independent variables. If they are high (say, higher than .80 - rather unlikely, in psychological constructs), then you may have multicollinearity issues.
Then run your regression analysis, but also ask for multicollinearity statistics. I don't know what statistics software you are using, but this should not be difficult in any software. If the multicollinearity test shows there could be multicollinearity issues, then you would need to make a decision as to which of the correlating variables is more improtant to you and stick to these only, following the ideas of Richard Fielding.
Let me know if I can be of more help.
I think the procedure of Georgios Vleioras is a good way.
Rudolf Weiß
- Badges
- Science topic
- Similar topics
- Psychology
- Developmental Psychology