Dyadic Data Analysis in R

A practical guide to the Actor–Partner Interdependence Model with multilevel models and structural equation models

Dyadic Data Analysis in R

A complete, hands-on guide to estimating, interpreting, and reporting Actor–Partner Interdependence Models (APIM) using R.

Learn how to model two-person data correctly — when the people in your study are not independent.

What is the APIM?

The Actor–Partner Interdependence Model is the standard framework for analysing data in which two people’s responses are inherently linked — romantic partners, parent–child dyads, coworkers, doctor–patient encounters, even dyadic experiments.

It separates two kinds of effects:

Actor effects — the effect of my characteristic on my outcome.
Partner effects — the effect of my characteristic on my partner’s outcome.

Distinguishing the two requires accounting for the non-independence within each dyad. That is the central methodological move of the APIM, and it is what the rest of this site teaches you to do in R.

Who is this for?

This site is for applied researchers and graduate students who want to analyse dyadic data with R. You should be comfortable with linear regression and the basics of R. Basic or fundamental knowledge of multilevel models and structural equation modelling is assumed, as is working familiarity with the lme4 and lavaan packages.

How the site is organised

Foundations

Non-independence, the intraclass correlation, distinguishability, the four k-patterns, and the data structures APIM can handle.

Start here →

Tutorials — Indistinguishable dyads

Multilevel and SEM approaches when the two members of each dyad are interchangeable. Pooled slopes, no gender effect.

Tutorials →

Tutorials — Distinguishable dyads

Add gender (or any distinguishing variable) as a moderator. Compare MLM and SEM specifications. Test equality constraints.

Tutorials →

Tutorials — Moderated APIM

The full model: a dyad-level moderator (e.g. presence of children) shaping the partner crossover effect. Replication of Hahn, Binnewies, & Dormann (2014).

Tutorial →

Exercises

A nine-section problem set on a separate simulated dataset. Each section asks you to fit a model, interpret it, and reflect on the result. Hints and worked solutions are provided.

Practise →

Data & references

Variable lists, data-generating parameters, and the canonical reading list for APIM work in the social and organisational sciences.

Browse → · References →

A two-minute example

The model you will learn to fit looks like this in long format (one row per person):

\[\begin{align} \text{satisfaction}_{ij} = &\;\beta_0 + \beta_1 \text{wnc}_{ij} + \beta_2 \text{partner\_wnc}_{ij} \\ &+ \beta_3 \text{recovery}_{ij} + \beta_4 \text{partner\_recovery}_{ij} \\ &+ \beta_5 \text{has\_children}_j + \beta_6 \text{dual\_earner}_j \\ &+ u_{0j} + \varepsilon_{ij} \end{align}\]

where person $i$ is nested in dyad $j$, $u_{0j}$ is the dyad-level random intercept, and $\varepsilon_{ij}$ is the residual.

$\beta_1$ is the actor effect of work–nonwork conflict.
$\beta_2$ is the partner effect of conflict crossover.

The tutorials work through how to estimate, interpret, and report this model, and how to extend it to distinguishable dyads and moderation.

Run it yourself

Every tutorial is a Quarto document. The R code is executed when the site is rendered, so output (model summaries, fit indices, plots) is always current with the data.

To run locally, install the R packages and preview the site:

# From the repository root
make deps        # install lme4, lmerTest, lavaan, dplyr, interactions, semPlot, knitr, rmarkdown
make preview     # live-reload local preview at http://localhost:4200
make site        # build static site into docs/_site/

Or with Quarto directly:

quarto preview docs
quarto render docs

Acknowledgements

These materials were originally developed for the EAOHP 2026 conference workshop Dyadic Data Analysis Using R. The simulation data structures emulate the work of Hahn, Binnewies, & Dormann (2014). The APIM framework rests on the work of Kenny, Kashy, Cook, and Ledermann, among many others; see the references for the full list.

Materials are by George Michaelides, released under CC BY 4.0.

--- title: "Dyadic Data Analysis in R" subtitle: "A practical guide to the Actor–Partner Interdependence Model with multilevel models and structural equation models" format: html: toc: false page-navigation: false --- ::: {.hero} # Dyadic Data Analysis in R A complete, hands-on guide to estimating, interpreting, and reporting Actor–Partner Interdependence Models (APIM) using R. **Learn how to model two-person data correctly — when the people in your study are not independent.** ::: ## What is the APIM? The Actor–Partner Interdependence Model is the standard framework for analysing data in which two people's responses are inherently linked — romantic partners, parent–child dyads, coworkers, doctor–patient encounters, even dyadic experiments. It separates two kinds of effects: - **Actor effects** — the effect of *my* characteristic on *my* outcome. - **Partner effects** — the effect of *my* characteristic on *my partner's* outcome. Distinguishing the two requires accounting for the non-independence within each dyad. That is the central methodological move of the APIM, and it is what the rest of this site teaches you to do in R. ## Who is this for? This site is for **applied researchers and graduate students** who want to analyse dyadic data with R. You should be comfortable with linear regression and the basics of R. Basic or fundamental knowledge of multilevel models and structural equation modelling is assumed, as is working familiarity with the `lme4` and `lavaan` packages. ## How the site is organised ::: {.card-grid} ::: {.card} ### [Foundations]{.card-title-link} Non-independence, the intraclass correlation, distinguishability, the four k-patterns, and the data structures APIM can handle. [Start here →](foundations/index.html) ::: ::: {.card} ### [Tutorials — Indistinguishable dyads]{.card-title-link} Multilevel and SEM approaches when the two members of each dyad are interchangeable. Pooled slopes, no gender effect. [Tutorials →](tutorials/indistinguishable/index.html) ::: ::: {.card} ### [Tutorials — Distinguishable dyads]{.card-title-link} Add gender (or any distinguishing variable) as a moderator. Compare MLM and SEM specifications. Test equality constraints. [Tutorials →](tutorials/distinguishable/index.html) ::: ::: {.card} ### [Tutorials — Moderated APIM]{.card-title-link} The full model: a dyad-level moderator (e.g. presence of children) shaping the partner crossover effect. Replication of Hahn, Binnewies, & Dormann (2014). [Tutorial →](tutorials/moderated/index.html) ::: ::: {.card} ### [Exercises]{.card-title-link} A nine-section problem set on a separate simulated dataset. Each section asks you to fit a model, interpret it, and reflect on the result. Hints and worked solutions are provided. [Practise →](exercises/index.html) ::: ::: {.card} ### [Data & references]{.card-title-link} Variable lists, data-generating parameters, and the canonical reading list for APIM work in the social and organisational sciences. [Browse →](data/index.html){.card-link} · [References →](references/index.html) ::: ::: ## A two-minute example The model you will learn to fit looks like this in long format (one row per person): ```{=tex} \begin{align} \text{satisfaction}_{ij} = &\;\beta_0 + \beta_1 \text{wnc}_{ij} + \beta_2 \text{partner\_wnc}_{ij} \\ &+ \beta_3 \text{recovery}_{ij} + \beta_4 \text{partner\_recovery}_{ij} \\ &+ \beta_5 \text{has\_children}_j + \beta_6 \text{dual\_earner}_j \\ &+ u_{0j} + \varepsilon_{ij} \end{align} ``` where person $i$ is nested in dyad $j$, $u_{0j}$ is the dyad-level random intercept, and $\varepsilon_{ij}$ is the residual. - $\beta_1$ is the **actor effect** of work–nonwork conflict. - $\beta_2$ is the **partner effect** of conflict crossover. The tutorials work through how to estimate, interpret, and report this model, and how to extend it to distinguishable dyads and moderation. ## Run it yourself Every tutorial is a Quarto document. The R code is executed when the site is rendered, so output (model summaries, fit indices, plots) is always current with the data. To run locally, install the R packages and preview the site: ```r # From the repository root make deps # install lme4, lmerTest, lavaan, dplyr, interactions, semPlot, knitr, rmarkdown make preview # live-reload local preview at http://localhost:4200 make site # build static site into docs/_site/ ``` Or with Quarto directly: ```bash quarto preview docs quarto render docs ``` ## Acknowledgements These materials were originally developed for the EAOHP 2026 conference workshop *Dyadic Data Analysis Using R*. The simulation data structures emulate the work of Hahn, Binnewies, & Dormann (2014). The APIM framework rests on the work of Kenny, Kashy, Cook, and Ledermann, among many others; see the [references](references/index.html) for the full list. Materials are by [George Michaelides](about/index.html), released under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/).