An important aspect of psychometric model application is the choice of a suitable algorithm for parameter estimation. Particularly in the case of large datasets and high dimensional models, conventional algorithms may be inapplicable or relatively slow. As a remedy, variational inference was suggested in machine learning. While amortized variational inference algorithms are already established for several psychometric models, there has been little research on unamortized variational inference algorithms. In this article, we investigate the accuracy of this estimation method for a multidimensional variation of the joint model for responses and response times proposed by Van der Linden (Psychometrika 72: 287-308, 2007). Our results indicate that the proposed algorithm can be used for accurate item parameter estimation of these models even in relatively small datasets. In addition it is computationally fast in large datasets. We applied the method to a multidimensional empirical dataset in clinical psychology.
In a single-case experimental design (SCED), it is recommended that interrupted time-series analysis (ITSA) be applied in the AB design. This paper demonstrates that sample size determination in ITSA can be conducted using the Fisher information matrix (FIM) to predict estimation precision and statistical power. The simulation results corroborate that the FIM accurately predicts the empirical statistical power. Because of its low computational load, researchers can check the statistical power of various parameters (effect sizes) and sample sizes using FIM. Consequently, FIM serves as a valuable tool for optimizing the experimental design of SCED. Furthermore, we developed a graphical user interface (GUI) application capable of calculating the predicted statistical power.
This paper aims to examine the effect of rating scale orientation and labels on the measurement of subjective social position. Literature on the effects of layout and labels on responses frequently reported the left-side selection bias. Among the theoretical hypotheses that may explain this bias, we focused on the pseudoneglect and the “Left and top means first” heuristic. To examine the prediction from each hypothesis, we conducted a web survey experiment (Study 1). The results, however, did not seem to support either of the two hypotheses. Therefore, we proposed another hypothesis that assumes the “Moving rightward means going up” heuristic and conducted a web survey experiment (Study 2). The results partially supported our hypothesis. Based on the results of our two experiments, we recommended that the measurement of subjective social position should use a vertically oriented rating scale with the verbal anchors “Upper—Lower” and the descending numerical labels.
Count data frequently arise in applied research but are often challenging to model due to overdispersion, underdispersion, excess zeros, and unobserved heterogeneity. This study analyzes the Fisman_Miguel dataset using seven models: Poisson, negative binomial (NB), zero-inflated Poisson (ZIP), zero-inflated negative binomial (ZINB), Poisson hurdle (PH), negative binomial hurdle (NBH), and Conway–Maxwell Poisson (COM-Poisson) to model diplomatic misconduct. Parameters are estimated via maximum likelihood estimation, while model fit is evaluated using information criteria (AIC and BIC) and log-likelihood values. The results indicate that models accounting for overdispersion and excess zeros, particularly the ZINB and NBH, provide a superior fit compared to standard models. Moreover, the inclusion of random intercepts reveals substantial cross-country heterogeneity, highlighting the hierarchical structure of the data. Overall, the findings emphasize the importance of adopting flexible count data models in policy-oriented analyses, where reliance on standard Poisson assumptions may lead to misleading conclusions.
This study investigated the effects of topic variation on the accuracy of authorship analysis using Japanese stylometric features including n-grams of characters, parts-of-speech (POS), postpositional particles, and function words. We classified 4,574 blog texts from 2,287 authors into six topic groups and further divided them into similar and dissimilar topic groups using Latent Dirichlet Allocation (LDA) and hierarchical cluster analysis (HCA). Bayesian modeling and effect size analysis revealed that n-grams of characters were highly susceptible to topic variations, with increasing n exacerbating topic effects. In contrast, postpositional particles remained robust across all n-gram levels, whereas n-grams of POS and function words exhibited smaller topic effects than n-grams of characters despite significant topic effects at higher n-gram levels. These findings provide useful insights into authorship analyses, underscoring the importance of non-content-based stylometric features.
