Interaction versus moderation effect

[noetsi]

I know what these are I think. But I really don't know how to tell them apart in practice. The way I think of it is if there are two predictors X and X1 a moderation effect would be X ->X2 -> Y. For interaction the impact of X on Y varies at levels of X1 [maybe that is saying the same thing].

How would you run moderation in statistical software? I know how to run interaction effects.

 

[noetsi]

How would you model this. You have variable with five levels (a variable with 5 types of disability which you can model as 4 dummies showing if you have that disability or not). You think that they effect the impact on the dependent variable. So age has an impact on a DV say at one level (one disability type) but not at another level. Or the impact is in one direction at one disability but another direction for another disability.

How do you model that. An interaction between age and each of the 4 dummy variables?

 

[hlsmith]

Yup just insert disability into model as cat var, also insert age and the product of age and disability. In SAS you can visualize it if you turn on ODS GRaphics, I did this yesterday looking at age groups * time since covid predicting antibody level.

There is a methods paper in a epidemiology journal in last 5 years on moderation vs interaction. The distinction is confusing. I think one of the following was an author it: Stephen Cole, Daniel Westreich or Tyler Vander weele.

 

[hlsmith]

Referenced paper: https://journals.lww.com/epidem/Ful...nction_Between_Interaction_and_Effect.16.aspx

 

[hlsmith]

From reference:
'Interaction is defined in terms of the effects of 2 interventions whereas effect modification is defined in terms of the effect of one intervention varying across strata of a second variable.'

Dr. VanderWeele literally wrote the book on interactions : VanderWeele. T.J. (2015). Explanation in Causal Inference: Methods for Mediation and Interaction. Oxford University Press

 

[noetsi]

Thanks hlsmith although I am not sure I understand that quote. I will try to get that book and read the article you posted.

 

[noetsi]

hlsmith have you used the lsmean code? It seems a lot easier than the contrast statement.

 

[hlsmith]

Yes I have and it (lsmeans) is. The book is really good. It goes into additive verses multiplication interaction. I wasn't recommending you buy it, but I guess you can.

 

[noetsi]

I will get it through the state library. I am sure all of at least 100 people have read it

I am still not sure what LS mean really is telling you.

 

[noetsi]

https://stats.idre.ucla.edu/sas/seminars/analyzing-and-visualizing-interactions/

I never heard of proc plm before

 

[hlsmith]

I have only used it for scoring data. Thou it stores data as some weird SAS file - I remember that irking me.

 

[noetsi]

I still have not been able to figure out if lsmeans generates simple effects or not.