State of the Art Speakers
Kathleen Gates, University of North Carolina at Chapel Hill, USA
Katie Gates’s overarching aim is the development of quantitative methods that quantify individuals' dynamic processes across time. Arriving at personalized dynamic models often requires the use and adaptation of exploratory methods such as unsupervised classification, feature selection, and model-building. Much of her work involves psychophysiological, ecological momentary assessments, and human brain data. She’s an assistant professor in the Thurstone Psychometric Lab at the University of North Carolina (Chapel Hill) where she guides the ongoing maintenance and development of the algorithm GIMME (http://gimme.web.unc.edu/) as well as other freeware. Her lab’s primary funding sources is the National Institute for Biomedical Imaging and Bioengineering. For more information see http://gateslab.web.unc.edu/.
Assessing individual differences in non-traditional data structures
Time series data bring new opportunities – and problems – for psychological researchers. Examples of this data are plentiful and include: self-assessments obtained multiple times a day for numerous days; psychophysiological measures (e.g., functional MRI); and passive data continually captured from devices. Ideally, researchers can use this data to investigate underlying processes of interest in a way that previously was unattainable. New questions can be asked, and new types of information learned.
In reality researchers face numerous challenges that can diminish the full potential of this data. At its most fundamental, it can be difficult to formulate research questions. Often little prior work has been done that can inform hypotheses on dynamic processes of interest. For this reason, many methods used on these data have a data-driven component. The question then becomes, which method to use? Even if one does have hypotheses, translating them into quantifiable and testable research questions can be daunting given that the available methods and measurement approaches may be unknown. The present talk provides an overview of the types of questions that are currently being answered with this data and examples of a few state-of-the-art methods.
Anna-Lena Schubert, Heidelberg University, Heidelberg, Germany
Anna-Lena Schubert works as an assistant professor at the Section of Personality Research at Heidelberg University in Germany. She completed her PhD on the relationship between mental speed and mental abilities at Heidelberg University in 2016 and has since been working on the neurocognitive processes underlying general cognitive abilities at Heidelberg University and the University of California, Irvine. Her research interests include the measurement and manipulation of neurocognitive processes related to general intelligence. For this purpose, she has been working on the psychometrics of electrophysiological data and mathematical models of cognition to overcome measurement problems with the goal to link theoretically guided measures of cognitive processes with individual differences in cognitive abilities. Her work integrates psychometrics with cognitive psychology, electrophysiology, and individual differences research.
Psychometric and mathematical modeling approaches to the measurement of individual differences in cognitive processes
The aim of cognitive psychometrics is to measure individual differences in parameters of cognitive processes. Often, these parameters are measured as performances indicators (e.g., response times or accuracies) in tasks supposedly engaging one specific cognitive process. This approach presumes that a specific task provides a process-pure measure of a single cognitive process—an assumption that is often violated as most cognitive tasks do not measure one specific cognitive process, but rather a combination of several cognitive processes. In this talk, I will discuss latent change and bifactor models as modeling approaches that may help to overcome this conceptual problem by modeling interindividual differences in intraindividual differences between experimental conditions. Special consideration will be given to recent studies suggesting negligible variance and substantial interindividual heterogeneity in these intraindividual experimental effects (Rey-Mermet, Gade, & Oberauer, 2018; Rouder & Haaf, 2018). Moreover, I will demonstrate how mathematical models of cognition can be used to directly quantify individual differences in specific cognitive processes without relying on the assumption of pure insertion. I will highlight several cognitive models that may be of particular interest for cognitive psychometrics and discuss particularities of the psychometric modeling of model parameters, such as their hierarchical nature and their typically low-to-moderate consistencies.
Minjeong Jeon, University of California, Los Angeles, USA
Dr. Jeon is an Assistant Professor of Advanced Quantitative Methods at the department of Education of UCLA. Prior to coming to UCLA, she was an Assistant Professor of Quantitative Psychology at the Ohio State University. She obtained her Ph.D in Quantitative Methods and Evaluation and MA in Statistics from UC Berkeley. Her research interests include developing, applying, and estimating a variety of statistical/latent variable models, such as multilevel models, structural equation models, item response theory models, and growth models. She is also interested in developing computational algorithms and software. Her recent interests include network analysis, item response tree/process models, and joint modeling of multivariate data (such as behavior, psychological, and neural data).
A Latent Space Modeling Approach to Unveiling Respondents’ and Items’ Dependence Structures in Item Response Analysis
I will present a novel statistical framework for analyzing item response data. The proposed framework leverages ideas and tools from state-of-art latent space modeling approaches and aims to capture unknown, complex item and respondent dependence structures that may be undetectable with existing methods. Specifically, the proposed method understands item response data as a function of the distances between items and respondents, between items, or between respondents. The positions of individual items and respondents are displayed in a in a low-dimensional Euclidean space, showing sub-groups of items and respondents that may be too nearby or distant from each other. Since similarities and differences are explicitly modeled, the traditional (local) independence assumptions for items and for respondents are no longer needed in the proposed framework. I will first present a simple latent space Rasch model that explicitly incorporates the effects of item and respondent latent positions on the probability of a correct response and then explain a more flexible approach that directly estimates similarities and differences between items as well as between respondents. Lastly, I will describe a hierarchical extension of the latent space item response model that accommodates hierarchical data structures and captures dependence structures for higher-level units, such as classrooms and schools. Empirical examples are provided to illustrate the use of the proposed models in practice.
Dani Gamerman, Instituto de Matemática, Universidade Federal do Rio de Janeiro, Brazil
Dr. Gamerman is Professor of Statistics at UFRJ and Director of their Graduate Program in Statistics. He is also visiting faculty at University College London, Duke University (USA), University of Connecticut (USA), Universidad Rey Juan Carlos (Spain), and ITAM (México). In 1987, he received his Ph.D. in Statistics from University of Warwick. Author of Monte Carlo Markov Chain: Stochastic Simulation for Bayesian Inference (with H. F. Lopes) and Statistical Inference: an Integrated Approach (with H. S. Migon and F. Louzada), both in their 2nd edition. Dr. Gamerman has authored more than 60 papers, published in Biometrika, JRSS B, Statistics & Computing, Multivariate Analysis, JBES, Applied Statistics and other (mostly statistical) journals and book chapters, including for the Handbook of IRT (edited by W. v. d. Linden). He is Associate Editor for the International Statistical Review, Environmetrics, Statistical Modelling, Statistical Methods & Applications, the Brazilian Journal of Probability and Statistics, and formerly for JBES. Lastly, Dr. Gamerman has given plenary talks at a number of scientific meetings, including Valencia International Meeting on Bayesian Statistics and ISBA World Meeting and seminars at universities in the USA, Europe, and Australia.
Dynamic generalized structural equation modeling, with application to the effect of pollution on health
Structural equation modeling (SEM) is a very useful tool for psychometrists as relations between latent constructs are frequently built, eg the effect of stress level in the student proficiency. This tool is also valuable in many other areas of Science. Important requirements for the SEM framework is to be able to incorporate effects associated with the passage of time in the so-called dynamic SEM and the generalization to handle non-Gaussian measurements. Dynamic generalized SEM is geared towards accommodating these extensions. We show how to: a) set up a model in state-space format; b) to perform inference; c) to make model selection and predictions and; d) to summarize results obtained from the analysis. Inference is performed with a Bayesian perspective, facilitating the use of MCMC methods and other useful modelling tools, including parsimony and identification. An application on the effect of pollution on health is used to illustrate many of these issues in the context of a real data set from northern Italy. Joint work with Luigi Ippoliti and Pasquale Valentini (Pescara).
Ernesto San Martin, Pontificia Universidad Católica de Chile
Ernesto San Martín is Full Professor at the Faculty of Mathematics, Pontificia Universidad Católica de Chile and Invited Professor at the Economics School of Louvain, Université Catholique de Louvain, Belgium. His research interests focus on the statistical modeling of social phenomena, particularly in education. Examples of his research are the following: identification problems in psychometric model construction; school effectiveness and value-added models; credible evaluations of educational public policies; learning spaces and classroom diversity. This research, mainly developed at the Interdisciplinary Laboratory of Social Statistics (LIES), is pursuit through scientific collaborations with psychologists, educational scientists and mathematicians.
How to broker the evaluation of public policies? A proposal based on partial identification
In this talk, we propose a way to broker the evaluation of public policies. Educational researchers, psychometricians or statisticians become honest brokers when they clarify all policy options and their associated uncertainties. In the context of public policy evaluation, we argue that a partial identification analysis is a way to make explicit different non-testable assumptions. Each of these assumptions combined with observational evidence will provide different policy recommendations. These recommendations imply different volitional attitudes of the policy maker, which can be considered as the final position of scientific research in the sense of Neyman, that is, the inductive behavior. In this context covariates are no longer viewed as controls, but as contexts in which the policy eventually works. In passing, it is shown that the ignorability conditions typically used in the evaluation of public policies are logically strong but incredible. (This is joint work with Trinidad González-Larrondo.)
Gunter Maris, ACTNext, Iowa City, IA, USA
Gunter Maris is ACTNext’s director of advanced psychometrics. Prior to joining the ACTNext team, Gunter was a full professor of psychological methods at the University of Amsterdam (for 10 years), and a principal research scientist with CITO (for 16 years). Gunter’s research focuses on when and why learning does, or does not, happen, and when education does, or does not, work. As a by-products of this research, Gunter contributed to the founding of network psychometrics as an independent field of research, and to the advancement of online real-time rating systems for educational measurement. Gunter earned his MSc in theoretical psychology from the University of Leuven, and a PhD from the University of Nijmegen. He lives in Arnhem (the Netherlands) with his wife and two teenage daughters.
The Wiring of Intelligence
The positive manifold of intelligence has fascinated generations of scholars in human ability. In the past century, various formal explanations have been proposed, including the dominant g-factor, the revived sampling theory, and the recent multiplier effect model and mutualism model. In this article we propose a novel idiographic explanation. We formally conceptualize intelligence as evolving networks, in which new facts and procedures are wired together during development. The static model, an extension of the Fortuin-Kasteleyn model, provides a parsimonious explanation of the positive manifold and intelligence’s hierarchical factor structure. We show how it can explain the Matthew effect across developmental stages. Finally, we introduce a method for studying growth dynamics. Our truly idiographic approach offers a new view on a century-old construct, and ultimately allows the fields of human ability and human learning to coalesce.