Abstract:Sibling comparison designs are increasingly used to address unmeasured familial confounding in observational studies. We propose that three key interpretational fallacies - sibling characteristics, exposure correlation and non-shared confounding, and unmet life course model assumptions - can mislead causal conclusions. We demonstrate these fallacies by investigating childhood family income and mental health. A nationwide Danish cohort of individuals born between 1986 and 1996 (n = 643,814; 404,179 siblings) was followed-up from age 15 until onset of severe mental disorders. Population-wide and within-sibling adjusted hazard ratios (aHR) between childhood family income and offspring mental disorders were estimated, supplemented by descriptive statistics and pseudo-sibling analyses. A $15,000 increase in family income at age 14 was associated with a reduced rate of severe mental disorders (aHR = 0.78; 95% CI: 0.76-0.81), with comparable estimates across measurement ages 1-14 (range: 0.67-0.82). Null results were observed in both a pseudo-sibling cohort of unrelated individuals with the same income differences as the true sibling cohort (aHR = 0.93; 95% CI: 0.85-1.01) and the true sibling cohort (aHR = 1.02; 95% CI: 0.94-1.11). Siblings were typically born three years apart, with an average monthly income difference of $496 at age 14 (IQR;$150-$641). This study advocates for cautious causal interpretation of null results in sibling comparison studies because (1) it may not capture meaningful differences in family income across siblings due to minor income fluctuations ; (2) the pseudo-sibling cohort showed evidence of unmeasured non-shared confounding; (3) sibling comparison designs test a critical periods life course model, but the results favour an accumulating/vulnerable model. We present guidelines and R syntax to assess these interpretational fallacies.

Assessing Outcome-to-Outcome Interference in Sibling Fixed Effects Models

Familial confounding or measurement error? How to interpret findings from sibling and co-twin control studies

Estimating sibling spillover effects with unobserved confounding using gain-scores

Extended Sensitivity Analysis for Heterogeneous Unmeasured Confounding with An Application to Sibling Studies of Returns to Education

Modeling Interference Via Symmetric Treatment Decomposition

The ties that bind II: Reunification for siblings in out-of-home care using a statistical technique for examining non-independent observations

Semiparametric Estimation of Treatment Effects in Observational Studies with Heterogeneous Partial Interference

Estimating Causal Effects Under Interference Using Bayesian Generalized Propensity Scores

Causal Effects in Twin Studies: the Role of Interference

Interpreting polygenic score effects in sibling analysis

Regression adjustments for estimating the global treatment effect in experiments with interference

A Note on Modelling Bidirectional Feedback Loops in Mendelian Randomization Studies

Bipartite Causal Inference with Interference

Causal inference on networks under continuous treatment interference

Treatment Effect Analysis for Pairs with Endogenous Treatment Takeup

A Nonparametric Test of Heterogeneous Treatment Effects under Interference

Doubly robust matching estimators for high dimensional confounding adjustment

Interpretational fallacies in sibling comparison designs with social exposures: A case study of childhood income and mental disorders

Spatial Interference Detection in Treatment Effect Model

Heterogeneous Treatment and Spillover Effects under Clustered Network Interference

Efficient Nonparametric Estimation of Stochastic Policy Effects with Clustered Interference