1 October 2022 in Psychological Methods
Structural equation models (SEMs) are widely used to handle multiequation systems that involve latent variables, multiple indicators, and measurement error. Maximum likelihood (ML) and diagonally weighted least squares (DWLS) dominate the estimation of SEMs with continuous or categorical endogenous variables, respectively. When a model is correctly specified, ML and DWLS function well. But, in the face of incorrect structures or nonconvergence, their performance can seriously deteriorate. Model implied instrumental variable, two stage least squares (MIIV-2SLS) estimates and tests individual equations, is more robust to misspecifications, and is noniterative, thus avoiding nonconvergence. This article is an overview and tutorial on MIIV-2SLS. It reviews the six major steps in using MIIV-2SLS: (a) model specification; (b) model identification; (c) latent to observed (L2O) variable transformation; (d) finding MIIVs; (e) using 2SLS; and (f) tests of overidentified equations. Each step is illustrated using a running empirical example from Reisenzein's (1986) randomized experiment on helping behavior. We also explain and illustrate the analytic conditions under which an equation estimated with MIIV-2SLS is robust to structural misspecifications. We include additional sections on MIIV approaches using a covariance matrix and mean vector as data input, conducting multilevel SEM, analyzing categorical endogenous variables, causal inference, and extensions and applications. Online supplemental material illustrates input code for all examples and simulations using the R package MIIVsem. (PsycInfo Database Record (c) 2022 APA, all rights reserved).
1 January 2020 in Journal of Educational Psychology
15 February 2023 in International Journal of Methods in Psychiatric Research
Objective: Symptom counts as the basis for Post‐Traumatic Stress Disorder (PTSD) diagnoses in the DSM presume each symptom is equally reflective of underlying disorder severity. However, the “equal weight” assumption fails to fit PTSD symptom data when tested. The present study developed an enhanced PTSD diagnosis based on (a) a conventional PTSD diagnosis from a clinical interview and (b) an empirical classification of full PTSD that reflected the relative clinical weights of each symptom.Method: Baseline structured interview data from Project Harmony (N = 2658) was used. An enhanced diagnosis for full PTSD was estimated using an empirical threshold from moderated nonlinear factor analysis (MNLFA) latent PTSD scale scores, in combination with a full conventional PTSD diagnosis based on interview data.Results: One in 4 patients in the sample had a PTSD diagnosis that was inconsistent with their empirical PTSD grouping, such that the enhanced diagnostic standard reduced the diagnostic discrepancy rate by 20%. Veterans, and in particular female Veterans, were at greatest odds for discrepancy between their underlying PTSD severity and DSM diagnosis.Conclusion: Psychometric methodologies that differentially weight symptoms can complement DSM criteria and may serve as a platform for symptom prioritization for diagnoses in future editions of DSM.
1 June 2021 in Psychometrika
There recently has been growing interest in the study of psychological and neurological processes at an individual level. One goal in such endeavors is to construct person-specific dynamic assessments using time series techniques such as Vector Autoregressive (VAR) models. However, two problems exist with current VAR specifications: (1) VAR models are restricted in that contemporaneous relations are typically modeled either as undirected relations among residuals or directed relations among observed variables, but not both; (2) current estimation frameworks are limited by the reliance on stepwise model building procedures. This study adopts a new modeling approach. We first extended the current unified SEM (uSEM) framework, a widely used structural VAR model, to a hybrid representation (i.e., “huSEM”) to include both undirected and directed contemporaneous effects, and then replaced the stepwise modeling with a LASSO-type regularization for a global search of the optimal sparse model. Our simulation study showed that regularized huSEM performed uniformly the best over alternative VAR representations and/or modeling approaches, with respect to accurately recovering the presence and directionality of hybrid relations and reliably removing false relations when the data are generated to have two types of contemporaneous relations. The present study to our knowledge is the first application of the recently developed regularized SEM technique to the estimation of huSEM, which points to a promising future for statistical learning in psychometric models.
1 January 2024 in advances.in/psychology
Network psychometric approaches in clinical psychology seek to model the relations between symptoms of a psychological disorder as a network. This approach allows researchers to identify central symptoms, examine the dynamic structure of a disorder, and propose targeted intervention strategies. However, little work has been done assessing the consequences of two core limitations of most network psychometric models: a) that these models do not protect against omitted variable bias and b) these models rarely explicitly model measurement error. Measurement error or the omission of important variables could lead to decreased power and/or increased false positive rates with respect to hypotheses tested using a network psychometric framework, as they have in more traditional psychometric modeling frameworks. In the present study, we evaluate two methods for estimating cross-sectional psychometric networks, EBICglasso and LoGo-TMFG with respect to their robustness to measurement error and omitted variable bias. We found that overall, EBICglasso tends to be more robust to the negative impacts of omitted variables/miss-measured variables and that the relative performance of the two methods tend to equalize with larger network sizes. Notably, sample size did not change the magnitude of the impact or omitted variables/measurement error, with large sample sizes showing similar performance to smaller sample sizes.
23 August 2021 in Structural Equation Modeling: A Multidisciplinary Journal
Substantive theory rarely provides specific enough information to guide our selection of the optimal model for longitudinal data. Instead, researchers are more likely to rely on models common to their field, even if they are not appropriate. The purpose of our study is to assess whether researchers can use overall goodness-of-fit measures from structural equation models to correctly find the data generating model (DGM) from among a broad set of different longitudinal models. We use four different DGM adapted from published empirical studies. We compare goodness-of-fit statistics (e.g., p-value, CFI, RMSEA, etc.) of the DGM with those of six alternative models. Overall, the Bayesian Information Criterion (BIC) performed best in selecting the DGM, though no fit statistic was flawless. In the absence of substantive theory, we recommend that researchers begin with the most general longitudinal model and test whether it can be simplified by eliminating parameters.