Hi Jeffrey,

I think there is potential in the Social Relations Model.  I read up recently on work trying to revive it. 

It might be helpful to look on my profile of the original article demonstrating the use of a multilevel model for dyadic modeling in JPSP 1993 by Barnett et al.  Although in the 1995 Family Psychology article Steve Raudenbush and I lay out a longitudinal extension of it, many people find that article more useful because it is more methodological in nature.  The underlying "trick" is that in the level-one equation we omit the beta0 term such that the beta1 and beta2 terms become partner-specific intercepts in level-two equations.  The multilevel approach has pros and cons versus the SEM-based APIM.  I believe there is a Wendorf article comparing the approaches.  The point is that there is no difficulty expanding this model to a third member.  On my profile there is a 1998 article by Ozer et al., that expands to 2 outcomes per spouse, which would be the equivalent of a group of four persons with unique outcomes. Any multivariate outcome model using multilevel modeling would be the equivalent of a model for a social group with a fixed number of persons.  See, for example, the two articles on my profile in the Harvard Educational Review.  The principal advantages of the 1993 dyadic model were to accommodate the correlated outcomes in a dyad and to provide an unbiased test of the difference of regression coefficients when there are dependent samples. 

Bob

Hi Jeffrey,

I am pretty sure that Dave Kenny's Social Relations Model can assess triadic data. It's described in his Dyadic Data Analysis book (Kenny, Kashy & Cook, 2006).

Three persons mean a group.   Did you consider using random coefficient model?  Group can be used as a variable at the second level.   It depends on what you intend to study.  Just a thought.   

Hi Jeffrey,

I think there is potential in the Social Relations Model.  I read up recently on work trying to revive it. 

It might be helpful to look on my profile of the original article demonstrating the use of a multilevel model for dyadic modeling in JPSP 1993 by Barnett et al.  Although in the 1995 Family Psychology article Steve Raudenbush and I lay out a longitudinal extension of it, many people find that article more useful because it is more methodological in nature.  The underlying "trick" is that in the level-one equation we omit the beta0 term such that the beta1 and beta2 terms become partner-specific intercepts in level-two equations.  The multilevel approach has pros and cons versus the SEM-based APIM.  I believe there is a Wendorf article comparing the approaches.  The point is that there is no difficulty expanding this model to a third member.  On my profile there is a 1998 article by Ozer et al., that expands to 2 outcomes per spouse, which would be the equivalent of a group of four persons with unique outcomes. Any multivariate outcome model using multilevel modeling would be the equivalent of a model for a social group with a fixed number of persons.  See, for example, the two articles on my profile in the Harvard Educational Review.  The principal advantages of the 1993 dyadic model were to accommodate the correlated outcomes in a dyad and to provide an unbiased test of the difference of regression coefficients when there are dependent samples. 

Bob

Hi Jeffrey,

Here is a reference to a more recent paper that discusses the analysis of triads using Kenny's Actor-Partner Interdependence Model:

Ledermann, T., Rudaz, M., & Grob, A. (2017). Analysis of group composition in multimember multigroup data, Personal Relationships. Advance online publication. 

Tom