Hi folks

A regression question - perhaps better placed on a different forum, but I thought I'd try here first :)

Partial correlations allow one to assess the unique contribution of a given set of predictors to the variance in Y outcome.

Let's say predictors x1, x2, x4, and x5 all stuck in a regression model accounted for 30% variance in Y outcome. Then you conduct partial correlations to assess their unique effects. Would it be possible for some of those predictors (let's say x1 and x2) to cause no unique variance? (i.e., weak and non significant correlations to Y outcome) - even though they were significant predictors in the overall model?

I think so but I am driving myself a bit mad!

Hopefully someone will be able to answer! : )

Not an expert but...yes because variables can be weak by themselves but become significant in combination with others. Obv you'd been to be sure there's no collinearity which could invalidate your results 1/2

1/2 but I guess you would report the result as being x variable becomes significant in the presence of the other predictor variables. You might then need to consider (qualitatively) why that would be the case?

I use logistic regression in my thesis and I found something similar btw. I put it down to the variables generally being quite weak predictors, and the one that is not sig by itself is helped along by others.

Finally, I should also add, (sorry, I can only write 2 line replies for some reason), you need to be super careful when reaching conclusions based on the p-value.

Thanks pd1598! Very helpful responses (in bite size chunks as well!). I haven't found this with my data, but I came across some results in a paper that made me question it... Quite an interesting notion really if it is the case. I mean, it would mean that if you had not included some other relevant predictors in your model, it may appear that a given set of predictors are accounting for the variance in the outcome. When actually they only appear to because they are correlated with some more important variables that you have missed. Do you see what I mean?

Yes, and you'd be right. The more stats I've used in my thesis the more a disbelieve everything about statistics.