Choosing the Right Model for Multi-Perspective Emotional Analysis

Beyond Dyads: Modeling the Triadic System

Analyzing the husband-wife and wife-bull dyads with APIM provides critical insights, but it treats them as two separate phenomena. In reality, they are interconnected components of a single system. For instance, the husband’s general disposition (an individual effect) influences his reactions in both the H-W and H-B dynamics. A series of separate APIMs cannot capture this holistic structure. To do so, we need a model that conceptualizes the triad as the fundamental unit of analysis.

The Social Relations Model (SRM)

The Social Relations Model is a specialized statistical method derived from analysis of variance, designed specifically to analyze interpersonal perception and behavior in small groups. For your research, it offers a theoretically rich framework for deconstructing the emotional data from your triad.

The core principle of SRM is the decomposition of the total variance in an outcome (e.g., a jealousy rating) into three primary components:

1. Actor Effect: The portion of variance due to the perceiver's general tendencies. This represents a person's stable trait to feel or express an emotion.
   • Your Research Question: Does the husband, on average, report higher jealousy across all interactions compared to other participants?
2. Partner Effect: The portion of variance due to the target's general tendencies. This represents a person's stable tendency to elicit an emotion from others.
   • Your Research Question: Does the bull, on average, elicit lower feelings of guilt or jealousy from his partners (wife, husband) compared to others?
3. Relationship Effect: The portion of variance unique to the dyad, after accounting for actor and partner effects. This captures the specific "chemistry" between two people.
   • Your Research Question: Is the level of guilt the wife feels in relation to the bull uniquely different from what would be expected based on her general tendency to feel guilt (her actor effect) and the bull's general tendency to elicit guilt (his partner effect)?

As you can see, these components map directly onto the individual and dyadic perspectives you aim to study. Given your statistical background, you can think of SRM as a two-way random effects ANOVA, where actor and partner are the random effects and the relationship effect is their interaction.

The Asymmetric Block Design: A Powerful Analogy

While your triad is a small "round-robin" group where everyone can (in theory) perceive everyone else, a useful framework for conceptualizing the analysis comes from the "asymmetric block design." This design is often used when individuals from two distinct social categories interact with each other. The following resource provides an outstanding and practical guide to implementing SRM, using a speed-dating study (men interacting with women) as its primary example. This is highly analogous to your work, where "husband," "wife," and "bull" represent distinct social roles.

This paper demonstrates that SRM doesn't just partition variance. It also quantifies two types of reciprocity:

• Generalized Reciprocity (Actor-Partner Correlation): Assesses if individuals who exhibit a behavior also tend to elicit it. In your context: "Does a person who tends to feel high levels of jealousy (actor effect) also tend to make their partners feel more jealousy (partner effect)?"
• Dyadic Reciprocity (Relationship Correlation): Assesses if the unique feeling between two people is mutual. In your context: "If the husband feels a uniquely high level of jealousy specifically toward the bull, does the bull in turn feel a unique level of a corresponding emotion (e.g., competitiveness, respect) toward the husband?"

By implementing SRM, you can answer whether the emotional variance in your triad is primarily person-driven (large actor/partner variances) or relationship-driven (large relationship variance).

Cross-Classified Multilevel Models (CC-MLM)

An alternative, and more flexible, approach is the Cross-Classified Multilevel Model. While SRM is a specialized tool, CC-MLM is a general framework that can be configured to answer the same questions.

You're already familiar with hierarchical or nested models (e.g., repeated measures nested within individuals). A cross-classified structure arises when a lower-level unit belongs to two or more higher-level units that are not nested within each other.

This "pupil-in-school-and-neighborhood" example maps perfectly to your data. An emotional rating (e.g., jealousy) can be seen as being cross-classified by two entities: the person feeling the emotion (the perceiver) and the person the emotion is about (the target).

• Level 1: An individual emotional rating at a specific time.
• Level 2 (Crossed Factors):
  • Perceiver ID: The unique ID of the person providing the rating (e.g., Husband).
  • Target ID: The unique ID of the person being rated (e.g., Bull).

When you fit a CC-MLM with perceiver and target as crossed random effects, the model estimates the variance components for each. The interpretation of these components is functionally identical to the SRM:

• Perceiver Variance = Actor Variance
• Target Variance = Partner Variance
• Residual Variance = Relationship Variance + Measurement Error

The following video provides an excellent visual intuition for how mixed models, the family to which CC-MLM belongs, separate within-group and between-group effects.

Justifying a Model for Your Capstone Project

So, which model should you choose: SRM or CC-MLM?

• Social Relations Model (SRM):

  • Justification: Choose SRM if you want to align your work with the rich theoretical tradition of interpersonal perception research. Its terminology (actor, partner, reciprocity) provides a powerful and established language for describing your findings. It is the bespoke tool for round-robin data.
  • Limitation: Classic SRM analysis requires balanced data (e.g., every person rates every other person), which can be difficult to achieve in practice.

• Cross-Classified Multilevel Model (CC-MLM):

  • Justification: Choose CC-MLM for its flexibility. As a general MLM framework, it more easily handles messy, unbalanced data (e.g., missing ratings) and can be extended with additional predictors or levels of nesting (e.g., nesting ratings within specific events).
  • Limitation: It lacks the specific theoretical terminology of SRM, though the variance components are interpreted identically.

The Recommended Approach: A Synthesis

The best approach for your project is a synthesis. As the Kluger and Eastwick paper (LINK) demonstrates, you can implement the Social Relations Model using modern multilevel modeling software (like R's glmmTMB or brms packages).

This strategy gives you the best of both worlds:

1. You can frame your research questions and hypotheses using the precise, theoretically-grounded language of SRM (actor, partner, relationship, and reciprocity effects).
2. You can implement the analysis using the flexible and powerful CC-MLM framework, which can handle the complexities of your real-world data.

The Kluger and Eastwick paper is your blueprint. It provides the R syntax for specifying the crossed random effects structure needed to estimate the SRM parameters.

By adopting this approach, you can confidently justify your model selection: you are using a Cross-Classified Multilevel Model to estimate the parameters of the Social Relations Model, providing a comprehensive, flexible, and theoretically-grounded analysis of all six perspectives within the triad.

Conclusion

In this lesson, we have moved from analyzing isolated dyads to modeling the triad as a complete social system. You have learned how to select and justify a comprehensive model that can partition emotional variance across individuals and their unique relationships.

Key Takeaways:

• Analyzing dyads with separate APIMs is insufficient for understanding the full triadic system, as it ignores the interdependence among all participants.
• The Social Relations Model (SRM) is a specialized framework that decomposes variance into actor (perceiver), partner (target), and relationship components, and estimates both generalized and dyadic reciprocity.
• The Cross-Classified Multilevel Model (CC-MLM) is a flexible framework that can answer the same questions by modeling outcomes as being simultaneously nested within non-hierarchical groups (e.g., perceivers and targets).
• The most robust approach is to use MLM software to estimate the parameters of the SRM. This provides the theoretical richness of SRM with the flexibility and power of modern mixed-effects modeling.

You now have the final, and most comprehensive, piece of your analytical puzzle. In our next lesson, we will shift from statistical modeling to the practicalities of study execution. We will focus on developing a detailed data collection plan, covering everything from participant recruitment and measurement instruments to data security and procedural timelines.