Chapter Four · failure evidence
What Structural Equation Modeling got wrong, from 33 dissertations
Researchers frequently encountered methodological obstacles when implementing structural equation modeling across complex survey and longitudinal datasets. Common points of failure included poor overall goodness of fit, estimation non-convergence, inadequate sample sizes, and violations of strict model assumptions. These records come from PhD theses at 16 institutions, 2021 to 2026. 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.
Structural equation models frequently fail to meet standard goodness of fit thresholds
Hypothesized multi-construct and multilevel structural equation models repeatedly exhibited poor global fit when applied to survey and longitudinal data. Researchers reported inadequate fit statistics across both structural pathways and measurement models, often forcing the abandonment of unified multi-domain representations.
Tried and failed
multilevel structural equation modeling applied to longitudinal survey diary data. Reason: Poor model fit indices for latent variables
Examining the Benefits of Cisgender Allyship Behavior for Allies of Transgender and Gender Non-conforming Individuals · Texas Tech
Tried and failed
Structural equation modeling of self-determination theory applied to occupational burnout and well-being data. Reason: The hypothesized multi-construct path model exhibited poor global goodness-of-fit to the cross-sectional survey data
Using self-determination theory to evaluate well-being and burnout of community pharmacists · UT Austin
Tried and failed
latent congruence structural equation modeling applied to multi-informant psychological longitudinal outcomes. Outcome: no signal. Reason: models connecting latent agreement factors to symptom trajectories failed to achieve adequate fit or significant predictive paths
Tried and failed
latent congruence structural equation modeling applied to multi-informant longitudinal dyadic ratings. Reason: No adequately fitting structural equation model could be established across measurement sessions
Considered and rejected
Considered and rejected: Structural equation models (SEMs) rejected because literature shows almost none meet standard goodness-of-fit thresholds and suffer poor underlying assumption handling.
Understanding transport behaviour and policies to increase walking and cycling · Oxford
Considered and rejected
Considered and rejected: Rejected evaluating a single unified structural equation model encompassing all five social adjustment domains simultaneously due to poor fit (RMSEA = .064, CFI = .772, TLI = .698).
Considered and rejected
Considered and rejected: Multilevel structural equation modeling (MSEM) latent variable approach rejected due to poor measurement model fit across mental health and additional benefit constructs.
Examining the Benefits of Cisgender Allyship Behavior for Allies of Transgender and Gender Non-conforming Individuals · Texas Tech
Considered and rejected
Considered and rejected: Initial plan of conducting a pathway structural equation model of self-regulation mediating social-emotional skills and academic achievement, rejected due to poor fit statistics.
Complex latent specifications cause numerical non-convergence during estimation
Iterative maximum likelihood estimation frequently failed to reach convergence when models incorporated complex latent interactions, auxiliary variables, or distal predictors. These persistent convergence failures routinely forced investigators to abandon latent structural models in favor of standard regression techniques.
Considered and rejected
Considered and rejected: Rejected a latent variable SEM approach for SRM-dependent Response Surface Analysis because the structural equation models failed to converge.
Stereotypes and Social Decisions: The Interpersonal Consequences of Socioeconomic Status · Scholars' Bank
Considered and rejected
Considered and rejected: Rejected Structural Equation Modeling (SEM) in favor of multiple linear regression analyses due to persistent model non-convergence.
The Role of Parent-Adolescent Relationships and Romantic Relationships in Type 1 Diabetes Management Outcomes in Adulthood · Virginia Tech
Tried and failed
incorporating distal exposure constructs into structural equation modeling applied to behavioral intention prediction. Outcome: did not converge. Reason: direct effects lacked signal and inclusion caused model non-convergence and severe fit deterioration
Media and memorable messages: beyond traditional barriers to organ donation in contemporary China · UT Austin
Tried and failed
fully latent product-indicator structural equation modeling applied to moderated mediation analysis. Outcome: did not converge. Reason: numerical convergence issues with complex latent interaction estimation
Career Calling in Older Adults: A Socioemotional Selectivity Perspective · Georgia Tech
Considered and rejected
Considered and rejected: Exclusion of auxiliary variables from the structural equation model to avoid model non-convergence with maximum likelihood estimators.
Diversity and Inequality in Context: Schools, Neighborhoods, and Adolescent Development · DukeSpace
Poor indicator quality and construct overlap undermine measurement model validity
Latent constructs suffered from low or negative factor loadings, near-zero explained variance, and severe multicollinearity that compromised discriminant validity. Pruning or co-varying problematic low-loading indicators failed to resolve fit deficiencies or eliminate spurious relationships.
Tried and failed
structural equation modeling measurement model applied to survey scale latent factor estimation. Reason: poor fit due to items having extremely low or negative factor loadings
A mixed methods evaluation of the relationship between purity culture and sexual shame · UT Austin
Tried and failed
Structural equation modeling with correlated latent constructs applied to survey-based consumer perception measurement. Reason: Severe multicollinearity and poor discriminant validity between highly correlated constructs forced variable removal
Social media influencers’ social cause communication: Message attributes and consumer responses · Iowa State
Tried and failed
pruning low loading indicators in structural equation modeling applied to survey-based latent variable models. Reason: removing or co-varying low-loading items failed to resolve model fit issues or spurious relationships
Multi-Generational Hotel Branding: Investigating Brand Equity in Millennial-Friendly Hotels · Texas Tech
Tried and failed
Bifactor structural equation modeling applied to Hierarchical psychopathology dimensions. Outcome: did not converge. Reason: Failed to achieve acceptable fit and externalizing indicators had near-zero explained variance
Adverse Childhood Experiences and Adult Psychopathology: An Examination of Transdiagnostic Mechanisms · Texas Tech
Insufficient sample sizes prevent reliable model identification and convergence
Studies with small cohorts failed to meet standard sample size requirements, leading to severe parameter overfitting and convergence constraints. Researchers were forced to abandon structural equation modeling or switch to alternatives like ordinary least squares regression on factor scores.
Tried and failed
structural equation modeling with composite score indicators applied to small sample census tract data. Outcome: data insufficient. Reason: High parameter count relative to sample size (N=35) caused severe overfitting, multicollinearity, and poor fit indices.
Heritage and Housing: The Impact of Historical Designation in Miami’s Urban Development. · Harvard
Considered and rejected
Considered and rejected: Rejected structural equation modeling for grade-level comparisons with sociodemographic covariates due to sample size convergence constraints, switching to OLS regressions on factor scores
Considered and rejected
Considered and rejected: Rejected traditional structural equation modeling (SEM) in favor of Exploratory Structural Equation Modeling (ESEM) to better handle real-world social science data cross-loadings and small sample size.
Expansion of Motivation Models of Engineering Doctoral Student Populations · unevada
Considered and rejected
Considered and rejected: Structural Equation Modelling (SEM) rejected due to insufficient sample sizes failing minimum requirements of 100-200 participants
Examining the Temporal Stability of Evaluative Attitudes Toward Violence · Carleton University Institutional Repository
Rigid parametric and structural assumptions conflict with empirical data characteristics
Structural equation modeling was discarded when data violated strict assumptions such as continuous variable scaling or stable a priori factor structures. Methodological requirements such as homogeneous treatment effects and unrealistic exact-fit null hypotheses further restricted model applicability.
Considered and rejected
Considered and rejected: Rejected the standalone use of the chi-square (χ2) goodness-of-fit test for evaluating model fit in structural equation modeling due to its unrealistic exact-fit null hypothesis and severe sample size sensitivity.
Evaluating the Performance of Existing and Novel Equivalence Tests for Structural Equation Modeling · YorkSpace
Considered and rejected
Considered and rejected: Rejected confirmatory factor analysis (CFA) and structural equation modeling (SEM) because provisional factor structures in prior literature were too unstable for a priori constraints
An Exploration of Fear of Sleep and Experiential Avoidance in the Context of PTSD and Insomnia Symptoms · Scholars' Bank
Considered and rejected
Considered and rejected: Rejected standard two-stage least squares / structural equation modeling due to requiring strong parametric assumptions, homogeneous treatment effects across units, and correct models for both treatment and outcome processes.
Doubly Robust Causal Inference With Complex Parameters · Penn
Considered and rejected
Considered and rejected: Covariance-based and Partial Least Squares Structural Equation Modeling (SEM) were rejected because the policy intervention variable was binary rather than continuous
Left open by the authors
Problems the authors named and did not get to.
Left open
Model the simultaneous interaction effects of privacy concerns and three sustainability types on consumer adoption using structural equation modeling. Blocker: Requires collecting new survey participant data or accessing the original thesis dataset.
Privacy Suspension with Sustainability and Trust in Consumer Adoption of Smart Technology · Virginia Tech
Left open
Test predicted health behaviors as mediators or moderators between negative healthcare experiences and health outcomes using structural equation modeling. Blocker: Requires the private survey dataset of sexual and gender minority adults used in the thesis
Investigating the healthcare setting as uniquely stressful for sexual and gender minority adults · UT Austin
Left open
Fit structural equation models to VR experimental data to evaluate simultaneous predictive pathways between individual/task traits and cybersickness. Blocker: Requires private participant experimental data collected during the thesis study
Cybersickness in virtual reality as a function of individual differences and scalable task workload · Iowa State
Left open
Fit structural equation models on the ECLS-K:2011 dataset to test and evaluate the modified teacher input theory. Blocker: None
Left open
Validate ISM and MICMAC structural models of aerospace manufacturing ecosystems using structural equation modelling (SEM). Blocker: Requires the specific structural models, expert elicitation data, or survey data used in the thesis for SEM testing
Analysis of the evolution of aerospace manufacturing ecosystems · Cranfield
Left open
Quantify out-of-distribution generalization within the structural equation modeling (SEM) and invariance framework for reweighting fairness methods. Blocker: Lacks specific target metrics, datasets, or concrete evaluation protocols
Enhancing Fairness in Machine Learning through Reweighting · Publikationssystem UB Tuebingen
Left open
Perform structural equation modeling with latent measurement structures on the thesis survey data using lavaan instead of observed mean path analysis. Blocker: Requires the private student survey dataset collected for this specific PhD thesis.
Left open
Fit a structural equation model to identify a common latent variable across depressive symptoms, self-compassion, and self-efficacy predicting academic outcomes. Blocker: Requires private survey data collected from the Texas Tech college student sample.
Left open
Examine the role of teachers' beliefs on student mathematics performance using structural equation modeling of survey and performance data. Blocker: None
Left open
Incorporate undergraduate GPA, GRE scores, and graduate GPA into the structural equation model for physics graduate student retention. Blocker: Requires access to private student academic records and longitudinal tracking across graduate programs.
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.