Partial correlation controlling variable?

Moderators: statman, Analyst Techy, andris, Fierce, GerineL, Smash

xbar
Posts: 7
Joined: Sun Jan 04, 2015 10:34 pm

Partial correlation controlling variable?

Postby xbar » Sun Apr 26, 2015 1:07 pm

After a bivariate correlation at time 2 all three of my variables turned out to be positively related, so the next step is a partial correlation. I did not hypothesize that this would happen, at time 1 it did not.

Time 1
Variable1 +Variable 2 = related
Variable1 + Variable 3 = unrelated
Variable 2 + Variable 3= related

Time 2 all variables related.

So which variable do I control for Variable 2 because it was related at Time1 as well, or should I choose Variable 3 because it was not related as a control variable for partial correlation at time 2?
GerineL
Moderator
Posts: 1477
Joined: Tue Jun 10, 2008 4:50 pm

Re: Partial correlation controlling variable?

Postby GerineL » Tue Apr 28, 2015 10:25 am

more information needed, not clear what your goal is, what analysis you want to run, what you mean by "control" etc.

In short, if you have indication that all variables might be related, and you have not only statistical but also theoretical reasons to "control" for some of these variables, do it.
xbar
Posts: 7
Joined: Sun Jan 04, 2015 10:34 pm

Re: Partial correlation controlling variable?

Postby xbar » Tue Apr 28, 2015 11:43 am

I am running a partical correlation.

This is, as I was told, the next step after bivariate correlation if all variables are related. I have data from two time points which I conducted bivariate correlation with. However the first time I did the bivariate the data was not all related so my question is if that should influence my choice in controlling variable when running the partial correlation for data at time 2?

(I investigated correlation of three variables with my analysis. The theory and my hypothesis did not support all variables being related)
GerineL
Moderator
Posts: 1477
Joined: Tue Jun 10, 2008 4:50 pm

Re: Partial correlation controlling variable?

Postby GerineL » Tue Apr 28, 2015 12:18 pm

I think I understand now:

you want to now if 2 (or 3?) variables are associated, controlling for another variable.
At t1, you see that there is no association at all between 2 of the variables, at T2, all variables are associated.
Variable1 +Variable 2 = related
Variable1 + Variable 3 = unrelated
Variable 2 + Variable 3= related
you want to know if you should still control for variable 3 in the relation between variable 1 and variable 2, given that variable 1 and varialbe 3 are unrelated at T1, is that correct?

In addition, you state:
The theory and my hypothesis did not support all variables being related
My first question is: why would you test it then?

Second, you know that the variables are all related at t2, thus, there is still a possiblity that the relation between V1 and V2 are partly due to overlap with V3 (even if this relation does not reach significance at T1). So I think it would be better to still control vor variable 3, even if it does not reach significance (because you know it could in fact have an impact, based on T3).

Also, this decision should not be based solely on statistics, but also on theory. Apparently you have a reason to control for V3, otherwise you would not have included it, right?

Who is online

Users browsing this forum: No registered users and 2 guests

cron