Moderation Analyses

[PsychResearch]

---

Members do not see this ad.

Moderation analysis can be quite challenging. One minute I have the 'aha' moment that I finally got it, and the next minute I'm confussed again about how to interpret my findings.

Suppose I have a regression with three predictors:

parental acceptance, age, parental acceptance X age (product term). all predictors are mean centered. the outcome variable is depression.

path from parental acceptance to depression = -.300 (sign of path is sig; path is significant)

path from age to depression (non-significant) = not significant

path from product term parental acceptanceXage = .50 (sign of path is positive, path is significant)

How do I interpret the product term path? As age increases the strenght of the relationship between parental acceptance and depression increases? Does depression become worse with every year increase in age as a result of parental acceptance?

I'm confussed as to the interpretation. If anyone can help me with this, I would appreciate it.

 

[airmont]

To understand the interaction, you need to graph it. There is a nice excel macro that does it for you called modgraph (just google it). All you need to do is plug in the regression weights (beta's) and some descriptive stats and it does all the work for you.

 

[ADDICTED2STATS]

This is going to depend some on your intercept and the sign/magnitude of the slope of the moderator (even though it isn't significant). At any rate, based on what you've given here I would say that you have a negative association between parental acceptance and parental depression that varies as a function of age (this is assuming PA is your predictor and age is your moderator). If age is your predictor and PA is your moderator then you'll have something like the association between age and depression is positive at high levels of PA and negative at low levels. Without knowing your other coefficients I can't tell you for sure, but it will be something like that.

The interaction coefficient is only important in so much as it tells you that the relationships vary as a function of one another. When I have interactions in regression, I usually interpret the effect of the predictor at various values of the moderator (e.g., +/-1 SD).

Check out Aiken & West 1991 or the following website:
http://www.people.ku.edu/~preacher/interact/interactions.htm

 

[PsychResearch]

This is going to depend some on your intercept and the sign/magnitude of the slope of the moderator (even though it isn't significant). At any rate, based on what you've given here I would say that you have a negative association between parental acceptance and parental depression that varies as a function of age (this is assuming PA is your predictor and age is your moderator). If age is your predictor and PA is your moderator then you'll have something like the association between age and depression is positive at high levels of PA and negative at low levels. Without knowing your other coefficients I can't tell you for sure, but it will be something like that.

The interaction coefficient is only important in so much as it tells you that the relationships vary as a function of one another. When I have interactions in regression, I usually interpret the effect of the predictor at various values of the moderator (e.g., +/-1 SD).

Check out Aiken & West 1991 or the following website:
http://www.people.ku.edu/~preacher/interact/interactions.htm

"This is going to depend some on your intercept and the sign/magnitude of the slope of the moderator (even though it isn't significant). At any rate, based on what you've given here I would say that you have a negative association between parental acceptance and parental depression that varies as a function of age (this is assuming PA is your predictor and age is your moderator). "

That is correct. Path from parental acceptance to depression is negative in sign and significant. This is part of a mediation model. So, what I have is moderated mediation.

Low parental acceptance predicts high depression. (B = -.300)

AgeXparental acceptance predicting depression is positive and significant. (B = .50)

Slope of the moderator is negative and non-significant (-.004)

What I know is that one adds the -.300+.50= -.250. I'm really confussed. Does this actually mean that the strenght of the path decreases with age? Or increases?

Thanks to both for the resources. I'll check it out. I already did the plot with the program.

 

[PsychResearch]

The interaction coefficient is only important in so much as it tells you that the relationships vary as a function of one another. When I have interactions in regression, I usually interpret the effect of the predictor at various values of the moderator (e.g., +/-1 SD).

Check out Aiken & West 1991 or the following website:
http://www.people.ku.edu/~preacher/interact/interactions.htm[/QUOTE]


Does that mean I +/- 1 SD from the mean centered moderator?
When I + 1 SD from the mean centered age (moderator) = that means I get the older people ??
When I - 1 SD from the mean centered age (moderator) = that means I get the younger people ??

Will I do the same for the predictor variable? +/- 1 SD

and then end up with 4 product terms?


low/low
low/high

high/low
high/high

??