When to control for a variable?

#1
Hi. I'm trying to analyze the relationship between child abuse and personality disorders.
I have two scales which have their own subscales, all on ratio scales/basis.
I am doing many different tests and analyses. My question is:
should I confound for gender?
In my data gender was not associated with any of the child abuse scales (emotional physical or sexual abuse or neglect) but is associated with a bunch of PD's..

Thanks in advance!
 

noetsi

Fortran must die
#2
You don't confound for a variable, it either is a confound or is not :p

You should base whether it is in the model or not based on theory (either your own or someone else's). I don't know what you mean by PD's but is sounds like you already have it in your model, if so it already is a control. Anything in a regression or ANOVA model is effectively a control.

Are you talking about an interaction effect (which is different than a control variable).
 
#3
:) yeah, I meant to say "control for a variable" :)
I'll try to clarify:
PD is personality disorders.
Research shows child abuse(CA) is related to personality disorders (PD). Research also shows that gender is related to both of these variables.
So when testing for the relationship between CA and PD I want to control for gender, right?
But in my data gender does not have a significant relationship with CA. It does with PD.
Do I still control for gender when analyzing the relationship between CA and PD?
Thanks a bunch.
 

noetsi

Fortran must die
#4
You would want to control for a variable if theory says it is correlated with the dependent variable and your predictors.
I assume PD is your dependent variable not a predictor. If gender is not related to another predictor then it is not a control in the way that is normally used. It is actually "causing" the change in the DV (at least relative to the variables you have in your model).

I have no expertise in your field but to me this suggest some other factor (something that differs by gender) not in your model is influencing the dependent variable and you have not found it. Unless you think gender itself causes personality disorder. One way or the other you need to address this issue in your analysis.
 
#5
:) yes gender does not cause personality disorders but it is found to be a related. For example more men tend to have antisocial personality disorder than women do.

The cause for any psychological phenomena is not agreed upon. Child abuse and genetics lead the race as far as I know.
I replied to your helpful comments on the other thread :) Thanks..
 

hlsmith

Not a robit
#6
For full disclosure, I read none of the above content. Ideally predictors are independent and you don't have to control for them to find an effect. However, if the predictor and the outcome are both effects of a variable you need to control for it to unconfound the effect. You need to know the relationship between variables before controlling for them. You would not want to control for a mutual effect of a variable and the outcome or exogenous variable. All depends on the purpose.