Latent Covariates Research Articles

Interactions between latent variables in count regression models

In psychology and the social sciences, researchers often model count outcome variables accounting for latent predictors and their interactions. Even though neglecting measurement error in such count regression models (e.g., Poisson or negative binomial regression) can have unfavorable consequences like attenuation bias, such analyses are often carried out in the generalized linear model (GLM) framework using fallible covariates such as sum scores. An alternative is count regression models based on structural equation modeling, which allow to specify latent covariates and thereby account for measurement error. However, the issue of how and when to include interactions between latent covariates or between latent and manifest covariates is rarely discussed for count regression models. In this paper, we present a latent variable count regression model (LV-CRM) allowing for latent covariates as well as interactions among both latent and manifest covariates. We conducted three simulation studies, investigating the estimation accuracy of the LV-CRM and comparing it to GLM-based count regression models. Interestingly, we found that even in scenarios with high reliabilities, the regression coefficients from a GLM-based model can be severely biased. In contrast, even for moderate sample sizes, the LV-CRM provided virtually unbiased regression coefficients. Additionally, statistical inferences yielded mixed results for the GLM-based models (i.e., low coverage rates, but acceptable empirical detection rates), but were generally acceptable using the LV-CRM. We provide an applied example from clinical psychology illustrating how the LV-CRM framework can be used to model count regressions with latent interactions.

Measurement and Uncertainty Preserving Parametric Modeling for Continuous Latent Variables With Discrete Indicators and External Variables

Research in education and behavioral sciences often involves the use of latent variable models that are related to indicators, as well as related to covariates or outcomes. Such models are subject to interpretational confounding, which occurs when fitting the model with covariates or outcomes alters the results for the measurement model. This has received attention in models for continuous observable variables but to date has not been examined in the context of discrete variables. This work demonstrates that interpretational confounding can occur in models for discrete variables, and develops a multistage Bayesian estimation approach to deal with this problem. The key features of this approach are that it is (a) measurement preserving, in that it precludes the possibility of interpretational confounding, and (b) uncertainty preserving, in that the uncertainty from the earlier stage of estimating the measurement model is propagated to the second stage of estimating the relations between the latent variable(s) and any covariates or outcomes. Previous work on these methods had only considered models for continuous observed variables, and software was limited to models with a single latent variable and either covariates or outcomes. This work extends the approach and software to a more general class of solutions, including discrete variables, illustrating the procedures with analyses of real data. Functions for conducting the analyses in widely available software are provided.

Prosthesis usability experience is associated with extent of upper limb prosthesis adoption: A Structural Equation Modeling (SEM) analysis.

Factors associated with upper limb prosthesis adoption are not well understood. In this study, we explored how prosthesis usability experience relates to the extent of prosthesis adoption through the development of a structural equation model (SEM). First, items related to prosthesis usability were developed and refined using cognitive testing and pilot testing and employed in a survey of 402 prosthesis users (mean age 61.7 (sd 14.4), 77.1% Veterans). The SEM examined two unidimensional latent constructs: Prosthesis Usability Experience and Prosthesis Adoption-and each had multiple measured indicators. SEMs tested direct as well as moderating and mediating effects between the latent constructs and covariates related to demographics and prosthesis type. SEM found a significant positive association between Prosthesis Usability Experience and Extent of Prosthesis Adoption. Several covariates had direct effects on prosthesis adoption: 1) Extent of Prosthesis Adoption was lower for those with transhumeral and shoulder amputation, and higher for those with bilateral amputation, compared to the reference group with unilateral transradial amputation and 2) Myoelectric multiple degree of freedom (multi-DOF) prosthesis use was associated with lower Extent of Prosthesis Adoption, compared to body-powered prosthesis use. Myoelectric multi-DOF use also modified the effect of Prosthesis Usability Experience on Extent of Prosthesis Adoption. For those with bilateral ULA, the strength of the relationship between Prosthesis Usability Experience and Extent of Prosthesis Adoption was reduced. Findings suggest that in order to increase prosthesis adoption, prosthetics developers and rehabilitation providers should focus on implementing strategies to improve prosthesis usability experience. New Prosthesis Usability Experience measures could be used to identify persons at greater risk for poor prosthesis adoption and target interventions to increase prosthesis use.

An encoding generative modeling approach to dimension reduction and covariate adjustment in causal inference with observational studies

In this article, we develop CausalEGM, a deep learning framework for nonlinear dimension reduction and generative modeling of the dependency among covariate features affecting treatment and response. CausalEGM can be used for estimating causal effects in both binary and continuous treatment settings. By learning a bidirectional transformation between the high-dimensional covariate space and a low-dimensional latent space and then modeling the dependencies of different subsets of the latent variables on the treatment and response, CausalEGM can extract the latent covariate features that affect both treatment and response. By conditioning on these features, one can mitigate the confounding effect of the high dimensional covariate on the estimation of the causal relation between treatment and response. In a series of experiments, the proposed method is shown to achieve superior performance over existing methods in both binary and continuous treatment settings. The improvement is substantial when the sample size is large and the covariate is of high dimension. Finally, we established excess risk bounds and consistency results for our method, and discuss how our approach is related to and improves upon other dimension reduction approaches in causal inference.

Hypoperfusion Intensity Ratio and Hemorrhagic Transformation in Patients with Successful Recanalization after Thrombectomy.

Hemorrhagic transformation remains a potentially devastating complication of acute ischemic stroke. We aimed to evaluate whether the hypoperfusion intensity ratio, a parameter derived from CT perfusion imaging, is associated with the development of hemorrhagic transformation in patients with anterior large-artery occlusion who had undergone thrombectomy. We retrospectively reviewed data from patients with consecutive acute ischemic strokes who had achieved successful recanalization (Thrombolysis in Cerebral Infarction score ≥2b) between January 2020 and December 2023. HIR was defined as the ratio of the volume of lesions with a time-to-maximum (Tmax) >6 seconds to those with a Tmax >10 second delay. The primary outcome, based on the European Cooperative Acute Stroke Study, was hemorrhagic transformation, diagnosed by follow-up imaging assessment in 24-hour windows, and radiologically classified as hemorrhagic infarction and parenchymal hematoma. The secondary outcome was a 3-month mRS score of ≥3. Among 168 patients, 35 of 168 developed hemorrhagic transformation; 14 of 168 developed hemorrhagic infarction, and 21 of 168 developed parenchymal hematoma PH. After adjusting the latent covariates, increased hypoperfusion intensity ratio (per 0.1, adjusted OR [aOR] 1.68, 95% CI 1.26-2.25), ASPECTS (aOR 0.44, 95% CI 0.27-0.72), onset-to-puncture (aOR 1.01, 95% CI 1.00-1.02), and cardioembolism (aOR 5.6, 95% CI 1.59-19.7) were associated with hemorrhagic transformation in multivariable regression. The receiver operating characteristic curve indicated that hypoperfusion intensity ratio can predict hemorrhagic transformation accurately (area under the curve = 0.81; 95% CI, 0.738-0.882; P < .001) and predict parenchymal hematoma (area under the curve = 0.801; 95% CI, 0.727-0.875; P < .001). Upon admission, hypoperfusion intensity ratio, an imaging parameter, predicted hemorrhagic transformation after reperfusion therapy in this patient population.

Estimating the size of a closed population by modeling latent and observed heterogeneity.

The paper extends the empirical likelihood (EL) approach of Liu etal. to a new and very flexible family of latent class models for capture-recapture data also allowing for serial dependence on previous capture history, conditionally on latent type and covariates. The EL approach allows to estimate the overall population size directly rather than by adding estimates conditional to covariate configurations. A Fisher-scoring algorithm for maximum likelihood estimation is proposed and a more efficient alternative to the traditional EL approach for estimating the non-parametric component is introduced; this allows us to show that the mapping between the non-parametric distribution of the covariates and the probabilities of being never captured is one-to-one and strictly increasing. Asymptotic results are outlined, and a procedure for constructing profile likelihood confidence intervals for the population size is presented. Two examples based on real data are used to illustrate the proposed approach and a simulation study indicates that, when estimating the overall undercount, the method proposed here is substantially more efficient than the one based on conditional maximum likelihood estimation, especially when the sample size is not sufficiently large.

Improvements in lung function following vitamin C supplementation to pregnant smokers are associated with buccal DNA methylation at 5 years of age

BackgroundWe previously reported in the “Vitamin C to Decrease the Effects of Smoking in Pregnancy on Infant Lung Function” randomized clinical trial (RCT) that vitamin C (500 mg/day) supplementation to pregnant smokers is associated with improved respiratory outcomes that persist through 5 years of age. The objective of this study was to assess whether buccal cell DNA methylation (DNAm), as a surrogate for airway epithelium, is associated with vitamin C supplementation, improved lung function, and decreased occurrence of wheeze.MethodsWe conducted epigenome-wide association studies (EWAS) using Infinium MethylationEPIC arrays and buccal DNAm from 158 subjects (80 placebo; 78 vitamin C) with pulmonary function testing (PFT) performed at the 5-year visit. EWAS were performed on (1) vitamin C treatment, (2) forced expiratory flow between 25 and 75% of expired volume (FEF25–75), and (3) offspring wheeze. Models were adjusted for sex, race, study site, gestational age at randomization (≤ OR > 18 weeks), proportion of epithelial cells, and latent covariates in addition to child length at PFT in EWAS for FEF25–75. We considered FDR p < 0.05 as genome-wide significant and nominal p < 0.001 as candidates for downstream analyses. Buccal DNAm measured in a subset of subjects at birth and near 1 year of age was used to determine whether DNAm signatures originated in utero, or emerged with age.ResultsVitamin C treatment was associated with 457 FDR significant (q < 0.05) differentially methylated CpGs (DMCs; 236 hypermethylated; 221 hypomethylated) and 53 differentially methylated regions (DMRs; 26 hyper; 27 hypo) at 5 years of age. FEF25–75 was associated with one FDR significant DMC (cg05814800), 1,468 candidate DMCs (p < 0.001), and 44 DMRs. Current wheeze was associated with 0 FDR-DMCs, 782 candidate DMCs, and 19 DMRs (p < 0.001). In 365/457 vitamin C FDR significant DMCs at 5 years of age, there was no significant interaction between time and treatment.ConclusionsVitamin C supplementation to pregnant smokers is associated with buccal DNA methylation in offspring at 5 years of age, and most methylation signatures appear to be persistent from the prenatal period. Buccal methylation at 5 years was also associated with current lung function and occurrence of wheeze, and these functionally associated loci are enriched for vitamin C associated loci.Clinical trial registration ClinicalTrials.gov, NCT01723696 and NCT03203603.

A copula formulation for multivariate latent Markov models

We specify a general formulation for multivariate latent Markov models for panel data, where outcomes are possibly of mixed-type (categorical, discrete, continuous). Conditionally on a time-varying discrete latent variable and covariates, the joint distribution of outcomes simultaneously observed is expressed through a parametric copula. We therefore do not make any conditional independence assumption. The observed likelihood is maximized by means of an expectation–maximization algorithm. In a simulation study, we argue how modeling the residual contemporary dependence might be crucial in order to avoid bias in the parameter estimates. We illustrate through an original application to assessment of poverty through direct and indirect indicators in a cohort of Italian households.

Changes in Teachers’ Perceptions of School Quality During COVID-19

Abstract: The COVID-19 pandemic brought forth unique challenges for schools, requiring a sudden shift to remote learning. However, there have been no empirical studies examining the effects of the pandemic on school quality. This study aims to address this research gap by examining teacher data from a longitudinal study ( n = 2,616 teachers in 120 schools in Germany) on various aspects of school quality. Using multiple-group multiple-indicator univariate latent change score modeling based on latent covariate propensity score balancing, we investigate whether COVID-19 impacted school quality – that is, school climate, collective teacher efficacy, teacher cooperation, and distributed leadership. With a few exceptions, we could not detect significant differences in the development of teachers’ perceptions of school quality during the pandemic compared to developments that occurred prior to the pandemic. For example, adaptive teaching increased more than before the pandemic, while social support between teachers decreased more than before the pandemic.

Estimation quality and required sample sizes in three-level contextual analysis models

In multilevel analysis, Level-1 predictors that also explain variance at a higher level are called contextual predictors. In the multilevel manifest covariate model, the Level-2 component is modeled as the average of the Level-1 predictor scores within a cluster. In the multilevel latent covariate model, the predictor is decomposed into two latent variables at Level-1 and Level-2. Performance conditions of these modeling approaches for three-level models are largely unexplored. We investigate the two approaches’ performance with respect to bias, coverage, and power in a three-level random intercept model. Results reveal differences in estimation quality and required sample sizes. We provide sampling recommendations for both approaches.

Coping Behaviors in Response to the COVID-19 Pandemic Among Essential Workers of Color: Latent Classes and Covariates

Coping Behaviors in Response to the COVID-19 Pandemic Among Essential Workers of Color: Latent Classes and Covariates

3-D Gait Identification Utilizing Latent Canonical Covariates Consisting of Gait Features

Biometric gait recognition is a lesser-known but emerging and effective biometric recognition method which enables subjects' walking patterns to be recognized. Existing research in this area has primarily focused on feature analysis through the extraction of individual features, which captures most of the information but fails to capture subtle variations in gait dynamics. Therefore, a novel feature taxonomy and an approach for deriving a relationship between a function of one set of gait features with another set are introduced. The gait features extracted from body halves divided by anatomical planes on vertical, horizontal, and diagonal axes are grouped to form canonical gait covariates. Canonical Correlation Analysis is utilized to measure the strength of association between the canonical covariates of gait. Thus, gait assessment and identification are enhanced when more semantic information is available through CCA-based multi-feature fusion. Hence, Carnegie Mellon University's 3D gait database, which contains 32 gait samples taken at different paces, is utilized in analyzing gait characteristics. The performance of Linear Discriminant Analysis, K-Nearest Neighbors, Naive Bayes, Artificial Neural Networks, and Support Vector Machines was improved by a 4% average when the CCA-utilized gait identification approach was used. A significant maximum accuracy rate of 97.8% was achieved through CCA-based gait identification. Beyond that, the rate of false identifications and unrecognized gaits went down to half, demonstrating state-of-the-art for gait identification.

An Offenders-Offenses Shared Component Spatial Model for Identifying Shared and Specific Hotspots of Offenders and Offenses: A Case Study of Juvenile Delinquents and Violent Crimes in the Greater Toronto Area

ObjectivesWe attempted to apply the Bayesian shared component spatial modeling (SCSM) for the identification of hotspots from two (offenders and offenses) instead of one (offenders or offenses) variables and developed three risk surfaces for (1) common or shared by both offenders and offenses; (2) specific to offenders, and (3) specific to offenses.MethodsWe applied SCSM to examine the joint spatial distributions of juvenile delinquents (offenders) and violent crime (offenses) in the York Region of the Greater Toronto Area at the dissemination area level. The spatial autocorrelation, overdispersion, and latent covariates were adjusted by spatially structured and unstructured random effect terms in the model. We mapped the posterior means of the estimated shared and specific risks for identifying the three risk surfaces and types of hotspots.ResultsResults suggest that about 50% and 25% of the relative risks of juvenile delinquents and violent crimes, respectively, could be explained by the shared component of offenders and offenses. The spatially structured terms attributed to 48% and 24% of total variations of the delinquents and violent crimes, respectively. Contrastingly, the unstructured random covariates influenced 3% of total variations of the juvenile delinquents and 51% for violent crimes.ConclusionsThe Bayesian SCSM presented in this study identifies shared and specific hotspots of juvenile delinquents and violent crime. The method can be applied to other kinds of offenders and offenses and provide new insights into the clusters of high risks that are due to both offenders and offenses or due to offenders or offenses only.

Robust analysis of cancer heterogeneity for high-dimensional data.

Cancer heterogeneity plays an important role in the understanding of tumor etiology, progression, and response to treatment. To accommodate heterogeneity, cancer subgroup analysis has been extensively conducted. However, most of the existing studies share the limitation that they cannot accommodate heavy-tailed or contaminated outcomes and also high dimensional covariates, both of which are not uncommon in biomedical research. In this study, we propose a robust subgroup identification approach based on M-estimators together with concave and pairwise fusion penalties, which advances from existing studies by effectively accommodating high-dimensional data containing some outliers. The penalties are applied on both latent heterogeneity factors and covariates, where the estimation is expected to achieve subgroup identification and variable selection simultaneously, with the number of subgroups being apriori unknown. We innovatively develop an algorithm based on parallel computing strategy, with a significant advantage of capable of processing large-scale data. The convergence property of the proposed algorithm, oracle property of the penalized M-estimators, and selection consistency of the proposed BIC criterion are carefully established. Simulation and analysis of TCGA breast cancer data demonstrate that the proposed approach is promising to efficiently identify underlying subgroups in high-dimensional data.

Using monotonicity restrictions to identify models with partially latent covariates

Using monotonicity restrictions to identify models with partially latent covariates

Genomic selection using random regressions on known and latent environmental covariates

Key messageThe integration of known and latent environmental covariates within a single-stage genomic selection approach provides breeders with an informative and practical framework to utilise genotype by environment interaction for prediction into current and future environments.This paper develops a single-stage genomic selection approach which integrates known and latent environmental covariates within a special factor analytic framework. The factor analytic linear mixed model of Smith et al. (2001) is an effective method for analysing multi-environment trial (MET) datasets, but has limited practicality since the underlying factors are latent so the modelled genotype by environment interaction (GEI) is observable, rather than predictable. The advantage of using random regressions on known environmental covariates, such as soil moisture and daily temperature, is that the modelled GEI becomes predictable. The integrated factor analytic linear mixed model (IFA-LMM) developed in this paper includes a model for predictable and observable GEI in terms of a joint set of known and latent environmental covariates. The IFA-LMM is demonstrated on a late-stage cotton breeding MET dataset from Bayer CropScience. The results show that the known covariates predominately capture crossover GEI and explain 34.4% of the overall genetic variance. The most notable covariates are maximum downward solar radiation (10.1%), average cloud cover (4.5%) and maximum temperature (4.0%). The latent covariates predominately capture non-crossover GEI and explain 40.5% of the overall genetic variance. The results also show that the average prediction accuracy of the IFA-LMM is 0.02-0.10 higher than conventional random regression models for current environments and 0.06-0.24 higher for future environments. The IFA-LMM is therefore an effective method for analysing MET datasets which also utilises crossover and non-crossover GEI for genomic prediction into current and future environments. This is becoming increasingly important with the emergence of rapidly changing environments and climate change.

Focus on Self-Presentation on Social Media across Sociodemographic Variables, Lifestyles, and Personalities: A Cross-Sectional Study.

Upward social comparison and aspects of self-presentation on social media such as feedback-seeking and strategic self-presentation may represent risk factors for experiencing negative mental health effects of social media use. The aim of this exploratory study was to assess how adolescents differ in upward social comparison and aspects of self-presentation on social media and whether these differences are linked to sociodemographic variables, lifestyle, or personality. The study was based on cross-sectional data from the “LifeOnSoMe” study performed in Bergen, Norway, including 2023 senior high school pupils (response rate 54%, mean age 17.4, 44% boys). Nine potentially relevant items were assessed using factor analysis, and latent class analysis was used to identify latent classes with distinct patterns of responses across seven retained items. The retained items converged into one factor, called “focus on self-presentation”. We identified three groups of adolescents with a low, intermediate, and high focus on self-presentation. Associations between identified latent classes and covariates were assessed using regression analyses. Being a girl, higher extraversion, lower emotional stability, more frequent alcohol consumption, and having tried tobacco were associated with membership in the high-focus group. These results suggest some characteristics that are associated with a higher focus on self-presentation and that could inform targeted interventions.

Distributionally Robust Losses for Latent Covariate Mixtures

Reliable Machine Learning via Structured Distributionally Robust OptimizationData sets used to train machine learning (ML) models often suffer from sampling biases and underrepresent marginalized groups. Standard machine learning models are trained to optimize average performance and perform poorly on tail subpopulations. In “Distributionally Robust Losses for Latent Covariate Mixtures,” John Duchi, Tatsunori Hashimoto, and Hongseok Namkoong formulate a DRO approach for training ML models to perform uniformly well over subpopulations. They design a worst case optimization procedure over structured distribution shifts salient in predictive applications: shifts in (a subset of) covariates. The authors propose a convex procedure that controls worst case subpopulation performance and provide finite-sample (nonparametric) convergence guarantees. Empirically, they demonstrate their worst case procedure on lexical similarity, wine quality, and recidivism prediction tasks and observe significantly improved performance across unseen subpopulations.

Indirect effects of bullying on school mathematics achievement in Chile

Indirect effects of bullying on school mathematics achievement in Chile

Causal effect analysis in nonrandomized data with latent variables and categorical indicators: The implementation and benefits of EffectLiteR.

Instead of using manifest proxies for a latent outcome or latent covariates in a causal effect analysis, the R package EffectLiteR facilitates a direct integration of latent variables based on structural equation models (SEM). The corresponding framework considers latent interactions and provides various effect estimates for evaluating the differential effectiveness of treatments. In addition, a user-friendly graphical interface customizes the implementation of the complex models. We aim to enable applications of EffectLiteR in more contexts, and therefore generalize the framework for incorporating latent variables measured with categorical indicators. This refers, for instance, to achievement tests in educational large-scale assessments (LSAs), which are typically constructed in the tradition of item response theory (IRT). We review different modeling strategies for incorporating latent variables from IRT models in an effect analysis (i.e., individual score estimates, plausible values, SEM for categorical indicators). The strategies differ in the handling of measurement error and, thus, have different implications for the accuracy and efficiency of causal effect estimates. We describe our extensions of EffectLiteR based on SEM for categorical indicators and illustrate the model specification step-by-step. In addition, we present a hands-on example, where we apply EffectLiteR in LSA data. The practical benefit of using latent variables in comparison to proficiency scores is of special interest in the application and discussion. (PsycInfo Database Record (c) 2024 APA, all rights reserved).