quick question on moderation

[Stephen J. Toglia]

My initial hypothesis was to test a mediational model but despite what the literature suggested I didn't find a relation between my IV and DV (step 1). So my professor then suggested looking at a moderation effect instead. 

Can I still test for moderation in the absence of any relation between my IV and DV?  If so, could I simply dichotomize the proposed mediator (e.g. low group and high group) and then run a covariate test (r) to check for any change in r between the two mediator groups (low & high)?

For example,  let's test for the moderation of religiosity on alcohol consumption and days missed work.  Say I split religiosity into two groups, low and high, and then ran a covariate between alcohol consumption and days missed work.  If the low religiosity group had a significant correlation and the high group had no significance, then do I conclude that religiosity is moderating or effecting the relation?? 

Anyone?
Thanks,
Stephen

[Poling, Taylor Leigh]

Yes, you can still test for moderation in absence of a simple correlation between your IV and DV. I would suggest you center your IV and the moderator variable (i.e., run descriptive on these 2 variables and check the save standardized variable box to create 2 new z-score variables). Next, create an interaction variable, which would be zscore(IV)*zsore(moderator). Then run a regression, enter your two main effects (i.e., the IV, and the moderator) then enter the new interaction variable you created. If the interaction variable is significant, your results support an interaction effect. Then you can start looking at the means at different levels of the moderator and create plots at different levels of the moderator (e.g., plot the slope between the IV and DV at +1 SD of the moderator and -1SD of the moderator) to help interpret the interaction effect. Anyway, that would be my initial thought of how to test for moderation. Good luck.

---

From: SPSSX(r) Discussion on behalf of Stephen J. Toglia
Sent: Mon 2/23/2009 3:41 PM
To: [hidden email]
Subject: quick question on moderation

My initial hypothesis was to test a mediational model but despite what the literature suggested I didn't find a relation between my IV and DV (step 1). So my professor then suggested looking at a moderation effect instead. 

Can I still test for moderation in the absence of any relation between my IV and DV?  If so, could I simply dichotomize the proposed mediator (e.g. low group and high group) and then run a covariate test (r) to check for any change in r between the two mediator groups (low & high)?

For example,  let's test for the moderation of religiosity on alcohol consumption and days missed work.  Say I split religiosity into two groups, low and high, and then ran a covariate between alcohol consumption and days missed work.  If the low religiosity group had a significant correlation and the high group had no significance, then do I conclude that religiosity is moderating or effecting the relation?? 

Anyone?
Thanks,
Stephen

[Swank, Paul R]

Some people argue that you can test for mediation even in the absence of a significant correlation between the IV and the DV. See more on Dave Kinney’s website. It is best to test the mediation effects (a*b or c-c1) to determine whether or not mediation is presence. Kris Preacher has a SPSS macro for using bootstrapping to test the mediation hypothesis.

 

Dr. Paul R. Swank,

Professor and Director of Research

Children's Learning Institute

University of Texas Health Science Center-Houston

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Poling, Taylor Leigh
Sent: Monday, February 23, 2009 3:05 PM
To: [hidden email]
Subject: Re: quick question on moderation

 

Yes, you can still test for moderation in absence of a simple correlation between your IV and DV. I would suggest you center your IV and the moderator variable (i.e., run descriptive on these 2 variables and check the save standardized variable box to create 2 new z-score variables). Next, create an interaction variable, which would be zscore(IV)*zsore(moderator). Then run a regression, enter your two main effects (i.e., the IV, and the moderator) then enter the new interaction variable you created. If the interaction variable is significant, your results support an interaction effect. Then you can start looking at the means at different levels of the moderator and create plots at different levels of the moderator (e.g., plot the slope between the IV and DV at +1 SD of the moderator and -1SD of the moderator) to help interpret the interaction effect. Anyway, that would be my initial thought of how to test for moderation. Good luck.

 

---

From: SPSSX(r) Discussion on behalf of Stephen J. Toglia
Sent: Mon 2/23/2009 3:41 PM
To: [hidden email]
Subject: quick question on moderation

My initial hypothesis was to test a mediational model but despite what the literature suggested I didn't find a relation between my IV and DV (step 1). So my professor then suggested looking at a moderation effect instead. 

Can I still test for moderation in the absence of any relation between my IV and DV?  If so, could I simply dichotomize the proposed mediator (e.g. low group and high group) and then run a covariate test (r) to check for any change in r between the two mediator groups (low & high)?

For example,  let's test for the moderation of religiosity on alcohol consumption and days missed work.  Say I split religiosity into two groups, low and high, and then ran a covariate between alcohol consumption and days missed work.  If the low religiosity group had a significant correlation and the high group had no significance, then do I conclude that religiosity is moderating or effecting the relation?? 

Anyone?
Thanks,
Stephen