Design Experiments From Reported Results

 Quick Analysis Plan Copied

 Long Analysis Plan

Design Specification (Grounded in Supplied Methodological Evidence)

Honest evidence gap: None of the supplied records directly measures sibling caregiving, MI episodic disruption, or MI-sibling education/employment deficits. The plan below is therefore a methodological transplant: it borrows the longitudinal mediation architecture validated in the vaccination study Longitudinal moderated mediation template, the causal-inference framework for psychopathology Causal inference framework for mediation designs, and the sibling-comparison logic shown by Greene et al., who found social impairment predicted SUDs among 104 nonreferred ADHD siblings vs 82 controls after controlling for conduct disorder and class Sibling-design precedent and causality caution.

Three waves, three constructs: W1 (adolescence, ~14-17y), W2 (mid-20s, ~24-26y), W3 (midlife, ~40-45y). X = MI-specific episodic disruption (episode count/severity between waves) contrasted against DD-stability index (chronicity without episodic spikes). M = sibling caregiving hours/week (time-diary + caregiver-identity items). Y = education/employment continuity (credential attainment, unemployment months, occupational continuity). Family-level controls: parental diagnosis, household SES at W1, sibling birth-order and spacing.

Analytic Plan

Primary test: Counterfactual sibling fixed-effects regression within families (MI-sibling vs DD-sibling vs unaffected-sibling triads) estimating whether episodic disruption predicts Y beyond stable chronicity. This directly mirrors how Greene et al. isolated sibling-specific risk beyond diagnostic category Sibling fixed-effects covariate logic. Mediation: Sequential X(W1)→M(W2)→Y(W3) indirect effects with bootstrapped CIs (5,000 resamples), following the PROCESS Model 4/14 approach validated in the job-insecurity study, which also demonstrates the moderated-mediation index metric Bootstrapped mediation index template. Episodic-vs-stability discrimination: Latent growth mixture modeling of MI vs DD trajectories (accelerated longitudinal multilevel approach per the SHLSET disability-trajectory model Multilevel trajectory modeling precedent) to decompose episodic spikes from chronic level.

Bias, Attrition, and Falsification Controls

Attrition: The vaccination study lost 53.7% at T2 with systematic retention of more-stable participants 53.7% systematic attrition precedent. Plan: inverse-probability weighting, construct-propensity models including W1 caregiving and symptom severity. Cross-sectional fallacy guard: The job-insecurity and PCOS records both demonstrate why cross-sectional mediation is unsafe Cross-sectional mediation limitation precedent. All mediation here uses temporally lagged predictors. Pre-registration: hypothesize MI-episodic > DD-stability for indirect effect magnitude; report both CIs and moderated-mediation indices. Falsification: (a) If DD-stability path indirect effect ≥ MI-episodic path (CI overlap), episodic-disruption hypothesis is refuted; (b) if sibling fixed-effects coefficient collapses after adding W1 caregiving baseline, reverse-causation (pre-existing liability) likely explains the association; (c) if worry/neuroticism-type third-variable confounding mirrors the neuroticism→worry→avoidance pattern (r=.80 between neuroticism and worry in that study) Third-variable confounding caution, add trait-negative-affectivity covariate at W1.

Sample and Power Consideration

Target ≥600 families with ≥2 siblings enrolled (Greene precedent: 186 siblings detected a social-impairment effect with covariate adjustment; 5,121 in SHLSET powered trajectory modeling). Given 50%+ expected attrition across 25+ years, enroll ≥1,200 baseline families to retain ~600 at W3. Confidence in this design plan: moderate — the mediation architecture is well-validated but the specific MI/DD sibling outcome question has no supplied direct precedent.