British Journal of Mathematical & Statistical Psychology最新文献

Item response theory (IRT) model is a widely appreciated statistical method in exploring the relationship between individual latent traits and item responses. In this paper, a sparse IRT model is established to address the sparsity of factor loadings. A global and local shrinkage prior is imposed to penalize the factor loadings: the global parameter controls the amount of shrinkage at the column levels, while the local parameter adjusts the penalty of factor loadings within each column. We develop a variational Bayesian procedure to conduct posterior inference. By exploiting a stochastic representation for logistic function, we frame sparse IRT model as a mixture model mixing with Pólya-Gamma distribution. Such a strategy admits a conjugate posterior for the latent quantity, thus leading to a straightforward posterior computation. We assess the performance of the proposed method via a simulation study. A real example related to personality assessment is analysed to illustrate the usefulness of methodology.

Recent advances in computerized assessments have enabled the use of innovative item formats (e.g., drag-and-drop, scenario-based), necessitating a flexible model that can capture systematic influence of item types on action counts. In this study, we present a refinement scheme that can explicitly model common features of items and allows inference on the item-type effects. We apply multifaceted parameterization to characterize the common and unique features of items and implement the formulation in two existing models, the Rasch and Conway-Maxwell-Poisson count models. The inference procedures for the proposed models are presented using Stan and validated for estimation accuracy. Numerical experimentation with simulated data suggest that the proposed inferential scheme adequately recovers the underlying model parameters. Empirical application demonstrated that the proposed refinement holds practical relevance when data exhibit distinct item-type effects. Based on the findings from the empirical investigation, we discuss practical considerations in applying the Poisson models for analysing count data.

Constructed-response (CR) items are widely used to assess higher order skills but require human scoring, which introduces variability and is costly at scale. Machine learning (ML)-based scoring offers a scalable alternative, yet its psychometric consequences in rater-mediated models remain underexplored. This study examines how scoring design, rater bias, ML inconsistency and model specification affect the reliability of ability estimation in polytomous CR assessments. Using Monte Carlo simulation, we manipulated human and ML rater bias, ML inconsistency and scoring density (complete, overlapping, isolated). Five estimation models were compared, including the Partial Credit Model (PCM) with fixed thresholds and the Many-Facet Partial Credit Model (MFPCM) with and without free calibration. Results showed that systematic bias, not random inconsistency, was the main source of error. Hybrid human-ML scoring improved estimation when raters were unbiased or exhibited opposing biases, but error compounded when biases aligned. Across designs, PCM with fixed thresholds consistently outperformed more complex alternatives, while anchoring CR items to selected-response metrics stabilized MFPCM estimation. The real data application replicated these patterns. Findings show that scoring design and bias structure, rather than model complexity, drive the benefits of hybrid scoring and that anchoring offers a practical strategy for stabilizing estimation.

Hidden Markov diagnostic classification models capture how students' cognitive attributes evolve over time. This paper introduces a Bayesian Markov chain Monte Carlo algorithm for diagnostic classification models that jointly estimates time-varying Q matrices, latent attributes, item parameters, attribute class proportions and transition matrices across multiple occasions. Using the R package hmdcm developed for this study, Monte Carlo simulations demonstrate accurate parameter recovery, and an empirical probability-concept assessment confirmed the algorithm's ability to trace attribute trajectories, supporting its value for longitudinal diagnostic classification in both research and instructional practice.

Generalizability theory (G-theory) defines a statistical framework for assessing measurement reliability by decomposing observed variance into meaningful components attributable to persons, facets, and error. Classic G-theory assumes homoscedastic residual variances across measurement conditions, an assumption that is often violated in psychological and behavioural data. The main focus of this work is to extend G-theory using a mixed-effects location-scale model (MELSM) that allows residual error variance to vary systematically across conditions and persons. By modeling heteroscedasticity, we can extend the computation of condition-specific generalizability ( $G_t$ ) and dependability ( $D_t$ ) coefficients to reflect local reliability under varying degrees of measurement precision. As an illustration, we apply the model to empirical data from an EEG experiment and show that failing to account for variance heterogeneity can mask meaningful differences in measurement quality. A simulation-based decision study further demonstrates how targeted increases in measurement density can improve reliability for low-precision conditions or participants. The proposed framework retains the interpretative character of classical G-theory while enhancing its flexibility. We argue that it supports finer-grained insights on conditions that influence reliability and better-informed design decisions in psychological measurements. We discuss implications for individualized reliability assessment, adaptive measurement strategies, and future extensions to multi-facet designs.

The widespread adoption of smartphones creates the possibility to passively monitor everyday behaviour via sensors. Sensor data have been linked to moment-to-moment psychological symptoms and mood of individuals and thus could alleviate the burden associated with repeated measurement of symptoms. Additionally, psychological care could be improved by predicting moments of high psychopathology and providing immediate interventions. Current research assumes that the relationship between sensor data and psychological symptoms is constant over time - or changes with a fixed rate: Models are trained on all past data or on a fixed window, without comparing different window sizes with each other. This is problematic as choosing the wrong training window can negatively impact prediction accuracy, especially if the underlying rate of change is varying. As a potential solution we compare different methodologies for choosing the correct window size ranging from frequent practice based on heuristics to super learning approaches. In a simulation study, we vary the rate of change in the underlying relationship form over time. We show that even computing a simple average across different windows can help reduce the prediction error rather than selecting a single best window for both simulated and real world data.

Although individuals may exhibit both gradual and abrupt changes in their dynamic properties as shaped by both slowly accumulating influences and acute events, existing statistical frameworks offer limited capacity for the simultaneous detection and representation of these distinct change patterns. We propose a Bayesian regime-switching (RS) modelling framework and an entropy measure adapted from the frequentist framework to facilitate simultaneous representation and testing of postulates of gradual and abrupt changes. Results from Monte Carlo simulation studies indicated that using a combination of entropy and information criterion measures such as the Bayesian information criterion was consistently most effective at facilitating the selection of the best-fitting model across varying magnitudes of abrupt changes. We found that slight lower entropy thresholds may be helpful in facilitating the selection of longitudinal models with RS properties as this class of models tended to yield lower entropy values than conventional thresholds for reliable classification in cross-sectional mixture models-even under satisfactory parameter recovery and classification results. We fitted the proposed models and other candidate models to the data collected from an intervention study on the psychological well-being (PWB) of college-attending early adults. Results suggested abrupt, regime-related transitions in the intra-individual variability levels of PWB dynamics among some participants following the intervention period. Practical usage of the entropy measure in conjunction with other model selection measures, and guidelines to enhance simultaneous detection of true abrupt and gradual changes are discussed.

The growing popularity of the ecological momentary assessment method in psychological research requires adequate statistical models for intensive longitudinal data (ILD), with multilevel latent state-trait (ML-LST) models based on the latent state-trait theory revised (LST-R theory) as one possible alternative. Besides the traditional LST-R coefficients reliability, consistency and occasion-specificity, ML-LST models are also suitable for estimating reliability at Level 1 ("within-subject reliability") and Level 2 ("between-subject reliability"). However, these level-specific coefficients have not yet been defined in LST-R theory and, therefore, their interpretation has been unclear from the perspective of LST-R theory. In the current study, we discuss the interpretation and identification of these coefficients based on the (multilevel) versions of the Multistate-Singletrait (MSST), the Multistate-Indicator-specific trait (MSIT)  and the Multistate-Singletrait model with M-1 correlated method factors (MSST-M-1). We show that, in the MSST-M-1 model, the between-subject coefficient is a measure of the indicator-unspecificity of an item (i.e. the portion of between-level variance that a specific item shares with a common trait) or the unidimensionality of a scale. Moreover, we highlight differences between occasion-specificity and within-subject reliability. The performance of the ML-MSST-M-1 model and the corresponding theoretical findings are illustrated using data from an experience sampling study on the within-person fluctuations of narcissistic admiration (Heyde et al., 2023).

Latent variable models typically require large sample sizes for acceptable efficiency and reliable convergence. Appropriate informative priors are often required for gainfully employing Bayesian analysis with small samples. Power priors are informative priors built on historical data, weighted to account for non-exchangeability with the current sample. Many extant power prior approaches are designed for manifest variable models, and are not easily adapted for latent variable models, for example, they may require integration over all model parameters. We examined two recent power prior approaches straightforward to adapt to these models, Mahalanobis weight (MW) priors based on Golchi (Use of historical individual patient data in analysis of clinical trials, 2020), and univariate priors, based on Finch (The Psychiatrist, 6, 2024, 45)'s application of Haddad et al. (Journal of Biopharmaceutical Statistics, 27, 2017, 1089) and Balcome et al. (bayesdp: Implementation of the Bayesian discount prior approach for clinical trials, 2022). We applied these approaches along with diffuse and weakly informative priors to a latent variable mediation model, under various sample sizes and non-exchangeability conditions. We compared their performances in terms of convergence, bias, efficiency, and credible interval coverage when estimating an indirect effect. Diffuse priors and the univariate approach lead to poor convergence. The weakly informative and MW approach both improved convergence and yielded reasonable estimates, but MW performed poorly under some non-exchangeable conditions. We discussed the issues with these approaches and future research directions.

Interrater reliability plays a crucial role in various areas of psychology. In this article, we propose a multilevel latent time series model for intensive longitudinal data with structurally different raters (e.g., self-reports and partner reports). The new MR-MLTS model enables researchers to estimate idiographic (person-specific) rater consistency coefficients for contemporaneous or dynamic rater agreement. Additionally, the model allows rater consistency coefficients to be linked to external explanatory or outcome variables. It can be implemented in Mplus as well as in the newly developed R package mlts. We illustrate the model using data from an intensive longitudinal multirater study involving 100 heterosexual couples (200 individuals) assessed across 86 time points. Our findings show that relationship duration and partner cognitive resources positively predict rater consistency for the innovations. Results from a simulation study indicate that the number of time points is critical for accurately estimating idiographic rater consistency coefficients, whereas the number of participants is important for accurately recovering the random effect variances. We discuss advantages, limitations, and future extensions of the MR-MLTS model.