Library · Causal inference & statistics
What Causal Mediation Analysis got wrong, from 21 theses
Summary written by a model from the records below: The records describe methodological and structural challenges encountered when applying causal mediation analysis across various observational and experimental settings. Researchers frequently rejected or revised mediation models due to poor structural equation fit, unfulfilled prerequisite associations, cross-sectional design limitations, and unaddressed confounding. These records come from graduate theses, 2021 to 2026, each labeled with its degree where the repository states it, and each links to its thesis. They were extracted by language models reading the full text, so treat each as a lead to read, not a verdict.
Plus 9 problems the authors left open, listed at the end.
Failure modes, by theses reporting them
Select a mode to jump to its records.
Structural and moderated mediation models fail to achieve acceptable goodness of fit or numerical convergence
Complex structural equation specifications and moderated mediation configurations frequently yield unacceptable fit metrics such as elevated root mean square errors of approximation. These models also suffer from convergence failures during bootstrapping or underperform when compared to simpler direct or single-mediator models.
Tried and failed
path analysis mediation model applied to cross-sectional behavioral survey data. Outcome: no signal. Reason: The specified structural mediation model exhibited very poor global fit and non-significant indirect mediation paths.
Family communication patterns and technology and face-to-face integration behaviors in parent-college student relationships · UT Austin · PhD
Tried and failed
continuous interaction modeling in structural equation models applied to longitudinal behavioral mediation analysis. Reason: produced unacceptable model fit indices including very high RMSEA
The intervening effects of screen time and physical activity on the relation between perceived environmental stress and the internalizing problems of adolescents · UT Austin · PhD
Tried and failed
moderated multigroup structural equation modeling applied to moderated mediation psychometric cascade pathways. Reason: models produced extremely poor fit indices across multigroup moderation configurations compared to direct mediation models
Mindfulness and its relationship to race-related stress, racial identity, age, gender, and ses across multiple racial minorities · Georgia Tech · PhD
Considered and rejected
Considered and rejected: Rejected moderated mediation models including interaction terms (dosage x working alliance and dosage x program satisfaction) due to poor model fit (CFI = .86, RMSEA = .10) and convergence failure in bootstrapping
Examining the Effects Between Teacher Qualifications and Program Implementation on Improving Social Skills among Students in K-3rd Grade · University of Oregon · PhD
Considered and rejected
Considered and rejected: Moderated mediation SEM models were rejected due to poor fit (RMSEA > .45) and non-significant likelihood ratio tests compared to mediation
Subjective social status, symbolic capital, and psychopathology in adolescence. · University of Washington · PhD
Considered and rejected
Considered and rejected: Mediation model separating intergenerational support into two observed variables (providing support and receiving support) was rejected due to poor fit (chi2(20) = 512.58, p < .001, CFI = .70, RMSEA = .14).
The impact of adult children's economic hardship on older parents' psychological well-being · Iowa State · PhD
Mediation analyses are abandoned when prerequisite bivariate correlations or main effects are absent
Researchers discard planned mediation procedures when direct relationships between predictors, treatments, and outcomes fail to reach statistical significance. Without demonstrable baseline associations or strong posterior confidence in main effects, the requisite assumptions for testing indirect pathways are violated.
Considered and rejected
Considered and rejected: Rejected mediation analysis via Hayes Process Macro due to failure of requisite bivariate correlations between CTQ-SF total, BIS-11 total, and CWI4 to reach statistical significance
Exploring childhood adversity, impulsivity, and cognitive functioning within forensic populations · Nottingham · Doctorate
Considered and rejected
Considered and rejected: Rejected planned mediation analyses because no treatment variables significantly correlated with competency restoration.
Juvenile Competency Restoration in Texas: Factors Associated with Remediation in Inpatient Settings · Sam Houston State University · PhD
Considered and rejected
Considered and rejected: Omitted formal statistical tests of indirect effects (mediation) for research questions 3a, 4a, 5a, 6a, 7a, and 8a because main effects lacked strong posterior confidence.
Preschoolers’ Non-risky and Risky Play Types, Social Competence, and Emotion Regulation: Considering Proximal and Distal Influences · Texas Tech · PhD
Considered and rejected
Considered and rejected: Discarded mediation models for DA mediating conformity motives on alcohol quantity and Level III binge drinking due to violation of the assumption requiring significant X-Y relations.
Examining the mediating effects of experiential avoidance on the relations between drinking motives and problematic alcohol consumption · Texas Tech · PhD
Full mediation models are rejected in favor of partial mediation or simpler direct models
Chi-square difference tests and fit indices demonstrate that constraining direct paths to zero provides an inadequate fit to observed data. Consequently, researchers favor partial mediation or direct effect baseline models that achieve equal or superior fit with lower structural complexity.
Lost to a baseline
Direct-effect SEM models were statistically preferable over indirect mediation models via chi-square difference tests (p = .059 to .859) due to equal fit with lower complexity.
Lost to a baseline
Measurement model without regression paths fit parent-report data significantly better than the proposed structural mediation model (chi-square difference = 33.70, p < .001).
Pathways to Parenting Stress and Challenging Behaviors for Preschoolers with Developmental Delays: The role of emotion regulation · University of Oregon · PhD
Considered and rejected
Considered and rejected: In Study 3, a full mediation model fixing the direct path from MORE treatment to Grade 4 math to zero was rejected in favor of a partial mediation model based on a chi-square difference test (χ²(1) = 8.79, p < .01).
Considered and rejected
Considered and rejected: A full mediation model of OBSE was rejected because it did not provide an adequate fit to the data (χ² = 10.21, df = 4, RMSEA = .09, CFI = .90, TLI = .77) compared to the partial mediation model.
Exploring the Impact of Servant Leadership on Thriving at Work and Adaptive Performance · Carleton University · PhD
Mediation analysis is rejected on cross-sectional data due to the lack of temporal ordering
Investigators decide against formal mediation modeling when cross-sectional study designs cannot establish temporal precedence among variables. These analyses are abandoned in favor of experimental manipulations, simple indirect effect estimation, or sequential covariate adjustment to avoid methodological bias.
Considered and rejected
Considered and rejected: Rejected performing full mediation analysis on Study 1 cross-sectional data due to methodological invalidity.
Emotion Dynamics of Solitude: How Spending Time Alone Affects the Way We Experience and Manage Affect · Carleton University · PhD
Considered and rejected
Considered and rejected: Rejected using classic mediation analysis for mental effort due to methodological failure concerns in experimental psychology (Lemmer & Gollwitzer, 2017), opting instead for experimental manipulation.
Leistungserwartungen und mentale Anstrengung bei Menschen mit depressiven Störungen · University of Marburg · PhD
Considered and rejected
Considered and rejected: Rejected formal cross-sectional mediation modeling in favor of estimating indirect effects, citing severe methodological bias and lack of temporal ordering in cross-sectional designs
The influence of perceived administrator support on the relationship between aggression against teachers and teachers’ reports of well-being in the workplace · Texas Tech · PhD
Considered and rejected
Considered and rejected: Decided against conducting a full mediation analysis due to the limiting nature of cross-sectional survey data, instead implementing sequential thematic covariate adjustment models.
Birth Matters: Examining Contextual Risk and Protective Factors for Maternal Postpartum Depression · Harvard · PhD
Causal mediation assumptions are violated by intermediate confounding and overlapping constructs
Mediation models are discarded when intermediate confounding between interrelated mediators or bidirectional reinforcement undermines causal validity. In addition, measurement tools that fail to separate mediator actions from outcomes or surrogate exposure violations prevent proper identification of causal pathways.
Considered and rejected
Considered and rejected: Rejected using dietary biomarkers directly as surrogate exposures in standard regression calibration without accounting for mediation pathways, as biomarkers often act as mediators violating surrogacy.
Causal Selection of Covariates in Regression Calibration for Mismeasured Continuous Exposure · Harvard · PhD
Considered and rejected
Considered and rejected: Mediation models were considered based on prior literature but rejected in favor of moderation models due to insufficient causal justification from bidirectional reinforcement.
What did you expect? The role of alcohol outcome expectancies in the relationship between psychopathic traits and alcohol-related aggression · Carleton University · Master's
Considered and rejected
Considered and rejected: Rejected using natural effects mediation analysis due to bias from intermediate confounding among interrelated school/family mediators, choosing an interventional effects approach instead.
Identifying Pathways from Gender and Sexual Orientation Stigma to Sexual Assault and Dating Violence in Adolescence and Young Adulthood · Harvard · PhD
Left open by the authors
Problems the authors named and did not get to.
Left open
Adapt mediational E-value sensitivity analysis methodology to evaluate unmeasured confounding in network causal models using simulations and public frameworks. Blocker: None
Networked resources and health in South Africa: Towards a Materialist Social Network Epidemiology · Harvard · PhD
Left open
Perform mediation analysis assessing whether gut microbiota-derived enterolignans mediate dietary lignan associations with gout risk. Blocker: Requires cohort biospecimen/metabolomics data (plasma or urinary enterolactone and enterodiol measurements)
Plant-Based Diets, Lignan Intake, Weight Change, and the Risk of Gout · Harvard · PhD
Left open
Develop causal mediation analysis methodology that jointly accounts for left-censoring and measurement error in viral persistence mediators. Blocker: None
Methods for causal mediation analysis with applications in HIV and cardiorespiratory fitness · Boston University · PhD
Left open
Investigate mediation mechanisms linking ethnic enclaves and neighborhood poverty to adverse maternal and birth outcomes. Blocker: Requires access to restricted Texas vital statistics records and clear mediator variables not specified.
Neighborhood Asian American composition and poverty concentration : associations with maternal and birth outcomes among Asian Americans in Texas · UT Austin · PhD
Left open
Perform Mendelian randomisation and mediation analysis to assess causal roles of identified metabolite levels in disease aetiology using GWAS summary statistics. Blocker: None
Common 'Inborn Errors' of Metabolism in the General Population · Cambridge · PhD
Left open
Perform mediation analysis to determine if inflammatory biomarkers mediate associations between prenatal PFAS exposure and pregnancy or child complications. Blocker: Requires access to restricted longitudinal pregnancy cohort data (MIREC/MIREC-ENDO)
Examining the impact of poly- and perfluoroalkyl substances (PFAS) exposure on metabolic health · Carleton University · PhD
Left open
Extend pragmatic mediation to settings with unobserved confounders, soft or high-dimensional interventions, or partially known abstract feature sets. Blocker: The task is an open-ended research direction without concrete mathematical targets, formal formulations, or specific evaluation methods provided.
Casual and trustworthy machine learning: methods and applications · Oxford · PhD
Left open
Replicate neural mediation analyses of peer victimization and drinking motives on racially and culturally diverse fMRI and behavioral datasets. Blocker: Requires access to controlled-access neuroimaging datasets (e.g., ABCD study) with diverse cohorts and relevant behavioral phenotypes
Early life peer victimization : associated neural structure and reactivity to social-emotional stimuli, drinking motives, and alcohol use in young adulthood · UT Austin · PhD
Left open
Perform a mediation analysis evaluating how average institutional SAT scores mediate effects on the percentage of Pell Grant recipients. Blocker: None
The relationships between unfunded tuition discounts at public universities in Texas and access, revenue, and prestige · Texas Tech · Doctorate
Who has worked on this
Authors of the public theses cited on this page, most-cited first.
- Joshua Baer Gilbert · Harvard · PhD · 2026
Synthesizing Psychometrics and Causal Inference: Applications of Latent Variable Models to Treatment Heterogeneity, Psychological Networks, and Learning Transfer - Beth Joy Dunn · Texas Tech · PhD · 2026
Preschoolers’ Non-risky and Risky Play Types, Social Competence, and Emotion Regulation: Considering Proximal and Distal Influences - Luca-Lars Hauser · Tübingen · PhD · 2026
Exploring Functional and Dysfunctional Features of Talent Development Environments in German Olympic and Paralympic Sports. Implications for the Holistic Development of Elite Youth Athletes - Oliver Johnson · Nottingham · Doctorate · 2025
Exploring childhood adversity, impulsivity, and cognitive functioning within forensic populations - Helen Gädeke · University of Marburg · PhD · 2025
Leistungserwartungen und mentale Anstrengung bei Menschen mit depressiven Störungen - Mackenzie Moore · Carleton University · Master's · 2025
What did you expect? The role of alcohol outcome expectancies in the relationship between psychopathic traits and alcohol-related aggression - Jana Palaniyandi · Carleton University · PhD · 2025
Examining the impact of poly- and perfluoroalkyl substances (PFAS) exposure on metabolic health - Elizabeth Glenn · University of Oregon · PhD · 2024
Pathways to Parenting Stress and Challenging Behaviors for Preschoolers with Developmental Delays: The role of emotion regulation - Grazia Mirabito · Georgia Tech · PhD · 2023
Mindfulness and its relationship to race-related stress, racial identity, age, gender, and ses across multiple racial minorities - Lameya Ahmed · Harvard · PhD · 2023
Birth Matters: Examining Contextual Risk and Protective Factors for Maternal Postpartum Depression - Wenze Tang · Harvard · PhD · 2023
Causal Selection of Covariates in Regression Calibration for Mismeasured Continuous Exposure - Muge Bakkaloglu · Carleton University · PhD · 2023
Exploring the Impact of Servant Leadership on Thriving at Work and Adaptive Performance
Checking a claim in this area?
We can run the same search on any method or claim. If nothing turns up, we will say so, and that proves nothing on its own.