Multinomial process tree models for discrete and continuous variables: benefits and applicability
- Autores: Anahi Gutkin de Freitas
- Directores de la Tesis: Manuel Suero Suñe (dir. tes.), Juan Botella Ausina (dir. tes.)
- Lectura: En la Universidad Autónoma de Madrid ( España ) en 2024
- Idioma: español
- Número de páginas: 250
- Títulos paralelos:
- Modelos de árbol de procesamiento multinomial para variables discretas y continuas: beneficios y aplicabilidad
- Tribunal Calificador de la Tesis: Ricardo Olmos Albacete (presid.), María José Contreras Alcalde (secret.), Edgar Erdfelder (voc.)
- Programa de doctorado: Programa de Doctorado en Psicología Clínica y de la Salud por la Universidad Autónoma de Madrid
- Materias:
- Enlaces
- Tesis en acceso abierto en: Biblos-e Archivo
- Resumen
This doctoral thesis focuses on Multinomial Processing Tree Models (MPT) that estimate continuous and discrete variables simultaneously (MPT-DC). The general goals are: a) to investigate both the methodological and substantive advantages of MTP-DCs; b) to explore their potential for capturing individual differences of model selection in recognition memory; c) to explore their potential for including continuous variables in the modelling of non-discrete cognitive processes; and d) to compare parametric and non-parametric MPT-DC approaches.
The first chapter contextualizes this thesis within the research framework in which it is embedded, the mathematical modeling of cognitive processes, in the form of MPTs, and, more specifically, through MPT-DCs. This introduction provides an overview of some unsolved limitations, which will be synthesized in the second chapter as general objectives of the thesis. Chapters 3, 4 and 5 present the manuscripts published or under review that address these objectives. Finally, chapter 6 discusses the most relevant results derived from these studies and their main conclusions, which also point towards future research directions.
The thesis comprises three studies, the first one has already been published as an article, and the other two are under review in specialized scientific journals.
In the first study (chapter 3), we analyzed the benefits of MPT-DC. For this purpose, signal detection theory (SDT) and two-high-threshold (2HT) models were fitted to data in a memory recognition paradigm, in which we modeled categorical response frequencies, response times (RT), and confidence levels (CL) based on both empirical (Juola et al., 2019) and simulated data. The results showed that the joint inclusion of informative categorical and quantitative dependent variables (DVs): a) reduces the standard error (SEs) of estimated parameters, b) increases the rate of correct selection of the data-generating models and c) allows for the detection of interaction effects that classical versions of MPT models are unable to capture. Thus, it emphasizes the methodological and substantive advantages of MPT-DC in disentangling the latent cognitive processes that underlie observed responses. Chapter 3 includes this publication in a scientific journal specialized in the area of memory (Gutkin et al., 2024; Journal Citation Reports) and is available in open access.
The second study (chapter 4) includes an expanded theoretical analysis of the results from the first one. Based on the simulations of the first study, we found that the inclusion of CLs and RTs in a recognition paradigm significantly improved the comparison between competing models. However, based on empirical data, none of the fitted models was able to account for the data of the entire sample of participants. Therefore, the second study presents an alternative model that integrates discrete process elements like the 2HT model together with the continuous process aspects of the SDT in a hybrid model based on the Atkinson-Juola (A-J) model (Atkinson & Juola, 1973, 1974; Juola et al., 1971). The extended A-J model resulted in a superior fit to most participants' data, including model-consistent predictions for the distributions of RTs and CLs. Despite these achievements, it is noted that not all participants fit the same model, suggesting that none of the three models considered can account for all datasets. Thus, individual differences in participants' characteristics or in their response strategies may be a decisive factor in model selection. Therefore, we developed a more general model that could accommodate various cognitive strategies and models influencing human memory search, decision-making, and response selection in recognition memory experiments. This article has been submitted to the Journal of Experimental Psychology: Learning, Memory, and Cognition and is currently under review.
The third study (chapter 5) is motivated by the fact that there are two MPT-DC approaches, also called extended-MPTs: parametric (Heck et al., 2018) and non-parametric (Heck & Erdfelder, 2016). As far as we know, they have never been systematically compared. In this study, both approaches were compared in terms of power and robustness, through three simulations based on the weapon identification task (WIT). In this context, several statistically equivalent models can be applied if the order of their processes (their response times, RTs) is not considered, such as the preemptive-conflict-resolution model (PCRM) and the default-interventionist model (DIM). The first simulation evaluates the calibration of the goodness-of-fit test and the robustness of different models in accommodating the simulated data. The second simulation compares nested models to study the power of the approaches in testing the validity of each model's RT assumptions. The third one focuses on model-recovery performance for the two non-nested models. We simulated data (categorical and RTs) following the DIM or PCRM and manipulated the size of the effects to be detected, the nature of these discrepancies (e.g., shape of the distribution, location, etc.), the simulated and fitted models, the sample size, and the parametric assumptions. The results indicate that the parametric approach is powerful but very sensitive to incorrect distributional assumptions. In contrast, the non-parametric approach is more robust but less powerful, especially with small samples. The study provides recommendations on when to use each procedure and highlights the importance of selecting between extended-MPT methods for validating underlying cognitive processes and model selection. This study was submitted to the Journal of Behavior Research Methods and is now under review.
The contributions of this doctoral thesis are of both methodological and theoretical interest, supported by the use of empirical and simulated data. Our aim is to integrate both fields, allowing the methods to address real problems in the empirical world based on firm theoretical foundations. We also seek to ensure that the results obtained are not limited to the specialized field of mathematical modeling but are disseminated to a wider audience that can benefit from them, thus fostering research based on enriching bidirectional feedback.
