Person-centered (or person-oriented) approach – in contrast to variable centered one – is more and more frequently used to analysis of individuals, e.g. personality traits. Did you meet any theoretically based papers and/or empirically sound studies where this analytical toolkit (e.g. various clustering methods, Latent Class Analysis and Latent Profile Analysis) was used for DYADIC data? Dyadic data is commonly analyzed by Actor Partner Interdependence Model (APIM) which is clearly a variable centered method. May be there examples of an alternative pattern oriented approach? How can these kind of results be interpreted? Any hints are highly appreciated!
Dear Colleagues, there was not much interest in this question, unfortunately. Has anyone an idea why?
Even more importantly: anyone interested in collaboration for working out the topic more in detail?
I think maybe would help to understand what kinds of dyadic data you're looking at. I know a little bit about SNA and a bit more about person centered stuff. The two communities don't meet all that often. What kinds of dyadic measures were you hoping to analyze?
Hi AnneMarie,
Thank you for the comment. I have married/cohabiting couples in my sample, and I did the assessment of their well-being, attachment style, etc.
I began to search a theoretical frame for a pattern oriented analyis of this kind of data (e.g., clustering the wife's and husband's attachment dimensions together). Patterns can be found in any kind of data but what do they mean, how can be they interpreted?
I found some papers that use this kind of pattern oriented analysis for dyadic data (partners, parents and siblings) but no real systematic reflection on the meaning of such patterns. I feel that INTRA-personal analysis and interpretation of a pattern-oriented approach is much more advanced.
I hope i could make my questions a bit clearer.
OK. I understand a bit more now. We do similar kinds of analyses with individuals and classrooms. One way is to cluster (or LPA--quite similar when specified the right way) at the level of the individual and then cross tab the partner 1 and partner 2 clusters. Where are the white spots and dark spots in the data? Are there particular pairings that are more common (statistics would be like a chi sq, testing whether the adjusted standardized residuals are greater than one would expect if the pairings were due to chance alone).
I think the Bergman et al. book will help you a lot (Studying individual development in an interindividual context: A person-oriented approach. L R Bergman, D Magnusson, B M Khouri). They use their freely-available SLEIPNER (which is the way I learned to do cluster analysis) with a set of modules for testing and comparing different cluster solutions, imputing missing data, etc. Note that MPlus is also my go-to package, but the questions you are asking about theory don't require any particular package to run. They also aren't new, though they're certainly advanced. I linked below to a paper I wrote using clustering with SLEIPNER. I can send it to you directly if you can't get the full text. Just message me for it.
There are a few publications by Steve Peck who is a genius at thinking about these multiple levels of analysis. My favorite is "TEMPEST in a Gallimaufry: Applying multilevel systems theory to person-in-context research." The reason you haven't found much done in this area is because most of us are still working it out and I, for one, am not comfortable publishing results from this person centered work until I have been able to replicate and test my findings across different samples and talk with others doing similar work about different ways to tweak the methods to get at the question. When it comes to getting support from the Muthens for work in Mplus, their responsiveness is quick and helpful on the stat model boards.
Article Patterns of Motivation Beliefs: Combining Achievement Goal a...
Thank you for sharing this - Peck's work is really interesting and it fits well to what I am looking for.
Hi Tamas! Did you figure out more about this since this post? I am interested in using LCA/LPA to look at patterns of spousal variables together - wondering how to account for the nonindependence?
Hi Lindsey, thank you for writing.
I have a few thoughts on this; however, it will be not exhaustive by no means.
1. As I see now, the basic pitfall of treating nonindependent data may be doing the analyses on the individual level. For example, in the case of a pattern-oriented approach, finding individual patterns in the partners separately. In this regard, looking for dyad level patterns/profiles is a correct choice; it treats nonindependent data connected.
2. However, indistinguishable dyads may mean a problem.
We are running now a systematic review of these types of analyses where pattern-oriented methods are applied to dyadic data. First lesson: people 'just do it' without much theoretical consideration on possible methodological inferences. Lesson two: all dyads in the analyses are distinguishable ones (heterosexual couples, child-parent dyads). It seems, there are no theoretically justified methods for treating indistinguishable dyads in clustering/LPA.
3. Meanwhile, we also published a paper in Frontiers Article Stress and Dyadic Coping in Personal Projects of Couples – A...
where we analyzed couple level patterns in heterosexual couples. (But not too much methodological considerations either.)
Again, thanks for your question and please share any information you may find elsewhere.
If you are close to publishing your paper, please share it with me that I could eventually include it in the review. (getting slowly forward with it, unfortunately...)
Best, Tamás
- Badges
- Science topic
- Similar topics
- Psychology