Latent transition analysis (LTA) is increasingly used to understand how individuals move among latent classes measured at multiple time points. LTA models often include auxiliary variables (e.g., covariates or distal outcomes) to explain class membership, predict transition probabilities, or examine later outcomes. Although measurement invariance is commonly assumed in LTA, it is not required for model estimation but is helpful for interpreting transitions as stability or change in the same construct. When auxiliary variables are included, testing and properly specifying measurement invariance becomes more complex if invariance is assumed. Implementing measurement invariance with recommended multi-step procedures, the ML three-step (ML 3-step) and the Bolck-Croon-Hagenaars (BCH) method, requires a sequence of model tests and modeling decisions. This tutorial provides a fully worked demonstration of how to evaluate measurement invariance in LTA and how to incorporate auxiliary variables using the ML 3-step and BCH approaches within Mplus. We introduce a practical workflow, address frequent modeling challenges, and provide annotated syntax using R and MplusAutomation for reproducibility. The goal of this tutorial is to offer applied researchers a transparent and accessible framework for drawing valid conclusions about subgroup stability and change.
Confirmatory factor analysis (CFA) is a statistical method that allows researchers to quantify how well the hypothesized factorial structure of a psychological measure fits the associated data. Fit indices are statistical measures that evaluate the factorial structure by quantifying the degree of model/data fit or misspecification. Researchers generally use traditional or fixed cutoff values of fit indices to judge the appropriateness of their CFA models. However, many authors have discussed the limitations of using these cutoffs, given that they cannot be generalized to all models. Recently, McNeish and Wolf (2023) developed the Dynamic Fit Index (DFI) approach, which generates cutoff values of fit indices that are tailored to the characteristics of a specific model. In this tutorial, a CFA of the Attainment of School Achievement Goal Scale (A-SAGS) was conducted using the lavaan package in R. Next, cutoff values for the fit indices were generated using the dynamic package in R. The model fit of the A-SAGS was mostly satisfactory when interpretations were based on fixed cutoff values. However, the hypothesized model did not reach the dynamic cutoff values, prompting a more thorough investigation into the possible sources of model misspecification. The DFI approach is easily implemented, highly promising, and offers researchers valuable insights as they attempt to improve the construct validity of a given psychological measure.
The 14th annual "Quantitative Methods in Human and Health Sciences" (MQSHS 2025) conference took place on June 5-6, 2025, at the MIL campus of the Universite de Montreal. This edition introduced two key innovations: the integration of a health sciences component through a collaboration between the Faculty of Nursing and the Faculty of Education, and an extended two-day format combining methodological workshops with original academic presentations. The first day featured four workshops : Patrick Desrosiers (Universite Laval) led a session on network analysis in social sciences, neuroscience, and health; Massimiliano Orri (McGill University) presented a workshop on network meta-analysis; Pier-Olivier Caron (Universite TELUQ) guided attendees through latent class and latent profile analysis using R and Mplus; and, Christophe Chenier (Universite de Montreal) explored validity as a cornerstone of measurement in social and health sciences, addressing the conceptual relationship between validity and validation in instrument development. The second day was dedicated to scholarly presentations and began with a keynote address by Alexandre J. S. Morin (Concordia University), who discussed the synergy between theoretical and methodological frameworks in social sciences. He emphasized the importance of bridging the gap between theory and quantitative techniques to yield richer, more precise research outcomes. Subsequent presentations examined various methodological challenges. Elise Desilets (Universite du Quebec a Trois-Rivieres) discussed the limitations of parametric tests, advocating for rank-based and permutation-based alternatives when assumptions of normality are not met. Vincent Arel-Bundock (Universite de Montreal) introduced marginaleffects, a package designed to enhance interpretability of statistical models by computing predictions, risk ratios, and contrasts across multiple analytical frameworks. Julien Bureau (Universite Laval) discussed person-centered analyses and their implications for theoretical development, particularly within self-determination theory. Andre Achim (Universite du Quebec a Montreal) compared three approaches to estimating factor scores (regression, Bartlett, and ten Berge) highlighting their respective trade-offs in terms of bias and accuracy. Eric Dion (Universite du Quebec a Montreal) discussed the measure of change and proposed three alternative psychometric criteria: the ordered nature of change, the internal consistency of items to reflect change, and the equivalence of measurement versions. The use of artificial intelligence in qualitative research was presented by Joel Gagnon who promoted its quick and systematic treatment of massive amount of text but pointed out challenges of occasional bullshit by emphasizing strict rigor, reliability, and human supervision. Jean Theroux (Universite de Montreal) validated a French-Canadian adaptation of the Pain Coping Questionnaire designed for adolescents with musculoskeletal conditions. Eric Frenette (Universite Laval) highlighted several adverse effects of the race for publications but focused mainly on the drawback of encouraging innovative statistics rather than relevant ones. Aoudou Njingouo Mounchingam (Universite du Quebec a Montreal) employed machine learning to construct infant mortality risk profiles based on demographic surveys from Cameroon. Dorothee Morand-Grondin (Universite de Montreal) proposed a new empathy measurement tool leveraging dynamic video stimuli and physiological markers. Talagbe Akpo (Institut national de la recherche scientifique) presented a historical analysis of quantitative methods, exploring their advancements and contemporary challenges. Bilkis Vissandjee (Universite de Montreal) examined the collection, interpretation, and dissemination of sensitive data on female genital mutilation (FGM/E), and concluded by emphasizing the need for inclusive terminology and a trauma-sensitive methodological framework to reduce the risk of re-traumatization and promote a more nuanced approach to the issue. The 2025 MQSHS conference demonstrated the relevance of quantitative methods in human and health sciences while fostering critical discourse on their ethical and practical implications.
The digit span task has long been a vital component in assessing working memory in both clinical and research contexts. Martin's (1978) two-experiment research established that digit span performance reflects order-information retention rather than mere memory capacity. The following study aimed to replicate Martin's second study methodology using a larger and more diverse sample to enhance generalizability, employing modernized testing procedures. English and French participants (n = 79) between 17 and 74 years of age, with varying levels of education and backgrounds, were asked to complete a 45-to 60-minute study, which included order recall and digit span tasks. While our results generally support the hypothesis that digit span is a predictor of memory performance, they only partially align with Martin's original findings. We observed significant but weaker correlations between digit span scores and strict recall scores (r = 0.32, p = 0.004) compared to Martin's original report (r = 0.63, p = 0.01). Conversely, we found a significant increase in correlation with lenient recall scores (r = 0.34, p = 0.002), indicating a shift in performance patterns. Notably, our participants exhibited nearly a 50% decline in average performance compared to those in the original study, suggesting potential external influences on outcomes. Future research should consider integrating a verbal span test to simultaneously assess item and order memory and explore how factors such as age and demographic variables influence performance. This approach would provide insights into generational effects and contextual influences on memory processing.
Le 14e colloque annuel « Méthodes Quantitatives en Sciences Humaines et Santé » (MQSHS 2025) s’est tenu les 5 et 6 juin 2025 au campus MIL de l’Université de Montréal. Cette édition a été marquée par deux nouveautés majeures : l’intégration d’un volet santé à travers une collaboration entre les Facultés des sciences infirmières et des sciences de l’éducation, ainsi qu’une organisation sur deux jours, combinant ateliers méthodologiques et communications scientifiques.. Le premier jour a accueilli quatre ateliers couvrant l’analyse de réseaux, la méta-analyse en réseau, l’analyse de classes latentes et de profils, ainsi que la validité des instruments de mesure en sciences sociales et de la santé. Le second jour a été consacré aux communications scientifiques originales, débutant avec la conférence d’ouverture d’Alexandre J. S. Morin sur les synergies théorie-méthode en sciences sociales. Les présentations qui ont suivi ont abordé des enjeux méthodologiques et appliqués, notamment les tests de permutations, l’interprétation statistique, les analyses centrées sur la personne, les propriétés des scores factoriels, la psychométrie des mesures du changement, les dérives en méthodologies quantitatives et l’intelligence artificielle pour l’analyse thématique. Les présentations plus appliquées ont porté sur l’adaptation francophone du Pain Coping Questionnaire, la détection de la mortalité infantile par l’apprentissage automatique, et l’évaluation de l’empathie par un nouvel outil écologique et ont suscité des discussions approfondies. Le colloque a permis des échanges critiques et a promu rigueur et éthique dans les analyses quantitatives. (english abstract at the end)
