Section 2 — Indistinguishable dyads: MLM

Section 2 — Indistinguishable dyads: multilevel modelling

What you are practising: the basic Actor–Partner Interdependence Model in long format, treating dyad members as exchangeable.

Reference: Indistinguishable dyads: MLM tutorial.

Goal

Fit the full APIM for each of the three outcomes, compare with a partner-free baseline, and write a two-sentence interpretation for each.

Tasks

Fit the full APIM for engagement. Include all three actor predictors (affect, sdt, job_crafting), all three partner predictors (partner_affect, partner_sdt, partner_job_crafting), the three moderators (live_together, years_together, time_spent_this_morning_together), and a random intercept for dyad_id. Save as apim_eng.
Fit a reduced model that omits the three partner predictors. Save as actor_only_eng. Compare the two via anova().
Repeat for performance and creativity. Save as apim_perf, apim_creat, and their corresponding actor-only versions.
Profile-likelihood confidence intervals for each of the three full models. Use confint(..., method = "profile").
Interpretation. For each outcome, write a two-sentence interpretation of the actor and partner effects.

Reflection prompt

For which outcome is the partner block (partner_affect, partner_sdt, partner_job_crafting) jointly most important? Does this match the data-generating process?

Hints & solution outline

Tutorial reference: Indistinguishable MLM. The full APIM pattern is in Step 2; the partner-block LRT is in Step 3.

Substitutions from the tutorial: - wnc → affect, also add sdt and job_crafting - recovery → sdt (the conceptual analogue), also add job_crafting - satisfaction → engagement (or performance / creativity) - has_children → live_together, also add years_together and time_spent_this_morning_together - ddl → ddl2

What to expect. All three DGP intercept gaps are non-zero, and the DGP has partner effects for sdt and job_crafting on all three outcomes. So the LRT for the partner block should reject for all three outcomes, and the strongest rejection is typically on engagement (which has the most focal moderated-crossover signal in the DGP).

What to record. For each outcome: “The partner block is jointly significant (LRT χ² = XX.X, df = 3, p = .XXXX), and [the focal partner predictor] is the strongest partner effect.”

--- title: "Section 2 — Indistinguishable dyads: MLM" --- # Section 2 — Indistinguishable dyads: multilevel modelling **What you are practising:** the basic Actor–Partner Interdependence Model in long format, treating dyad members as exchangeable. **Reference:** [Indistinguishable dyads: MLM tutorial](../../tutorials/indistinguishable/mlm.html). ::: {.callout-note} ## Goal Fit the full APIM for each of the three outcomes, compare with a partner-free baseline, and write a two-sentence interpretation for each. ::: ## Tasks 1. **Fit the full APIM for `engagement`.** Include all three actor predictors (`affect`, `sdt`, `job_crafting`), all three partner predictors (`partner_affect`, `partner_sdt`, `partner_job_crafting`), the three moderators (`live_together`, `years_together`, `time_spent_this_morning_together`), and a random intercept for `dyad_id`. Save as `apim_eng`. 2. **Fit a reduced model** that omits the three partner predictors. Save as `actor_only_eng`. Compare the two via `anova()`. 3. **Repeat for `performance` and `creativity`.** Save as `apim_perf`, `apim_creat`, and their corresponding actor-only versions. 4. **Profile-likelihood confidence intervals** for each of the three full models. Use `confint(..., method = "profile")`. 5. **Interpretation.** For each outcome, write a two-sentence interpretation of the actor and partner effects. ## Reflection prompt For which outcome is the partner block (`partner_affect`, `partner_sdt`, `partner_job_crafting`) jointly most important? Does this match the data-generating process? ::: {.callout-tip collapse="true"} ## Hints & solution outline **Tutorial reference:** [Indistinguishable MLM](../../tutorials/indistinguishable/mlm.html). The full APIM pattern is in Step 2; the partner-block LRT is in Step 3. **Substitutions from the tutorial:** - `wnc` → `affect`, also add `sdt` and `job_crafting` - `recovery` → `sdt` (the conceptual analogue), also add `job_crafting` - `satisfaction` → `engagement` (or `performance` / `creativity`) - `has_children` → `live_together`, also add `years_together` and `time_spent_this_morning_together` - `ddl` → `ddl2` **What to expect.** All three DGP intercept gaps are non-zero, and the DGP has *partner* effects for `sdt` and `job_crafting` on all three outcomes. So the LRT for the partner block should reject for all three outcomes, and the *strongest* rejection is typically on `engagement` (which has the most focal moderated-crossover signal in the DGP). **What to record.** For each outcome: *"The partner block is jointly significant (LRT χ² = XX.X, df = 3, p = .XXXX), and [the focal partner predictor] is the strongest partner effect."* :::