The Quantitative Psychology colloquium series meets weekly in the Autumn and Spring semesters. The primary activity is presentation of ongoing research by students and faculty of the quantitative psychology program. Often, guest speakers from allied disciplines (e.g., Education, Statistics, and Linguistics) within and external to The Ohio State University present on contemporary quantitative methodologies. Additionally, discussions of quantitative issues in research and recently published articles are often conducted during colloquium meetings.
This year, we are deviating from our usual weekly meetings for this series and instead focusing on participating in the joint Quantitative Brown Bag Series, held every month, as we have graduated all of our graduate students. The joint Quantitative Brown Bag Series is organized in collaboration with several other programs across the nation, including:
- University of Notre Dame
- University of Maryland, College Park
- University of North Carolina at Chapel Hill
- University of Virginia
- Vanderbilt University
- University of South Carolina
For more information on how to attend these events via Zoom, please contact Dr. Jolynn Pek.
Faculty coordinator: Dr. Jolynn Pek
Venue: Psychology 35 and online
Time: 12:30-1:30pm
September 9, 2024
Speaker: Dr. Victoria Stodden, Department of Industrial and Systems Engineering, University of Southern California
Title: AI-enabled Discovery: The Digital Scholarly Record
Abstract:
Abstract:
Colossal cloud infrastructure investments supporting near-ubiquitous global mobile technologies have trickled down to scientific research through cloud compute and storage, I/O tools, data analysis tools, and frameworks, which in turn have generated broad and expanding communities of users and supporters. These new technological innovations are deeply disruptive to the research community, since they open new paths to knowledge creation that were previously inaccessible and largely culturally unknown. In this talk, I examine these changes and their implications including: AI-enabled discovery pipelines; disruption in our scientific norms of transparency, accountability, and reproducibility; and the emergent digital scholarly record.
October 7, 2024
Speaker: Dr. Susu Zhang, Joint Assistant Professor of Psychology and Statistics, University of Illinois Urbana-Champaign
Title: Informing Educational Measurement with Test-Taking Process Data
Abstract: Computerized testing affords the collection of additional behavioral data beyond scores on test questions. One such instance is process data, the computer-recorded log files generated from test-taker interactions with a computerized assessment, e.g., keystrokes and clickstreams, in pursuit of solving a problem. This type of data offers rich insights into the cognitive processes underlying problem-solving, opening new avenues to address existing measurement and educational questions and explore novel ones. However, like constructed responses on open-ended questions with infinitely possible answers of different lengths, process data are highly unstructured and often noisy. This precludes the direct application of many well-established tools and psychometric models for structured test response data. In this talk, I discuss how sequential features extracted from test-taking process data can be incorporated into latent variable modeling frameworks commonly used in educational measurement, to address common questions in measurement and educational research, such as improving score reliability and understanding the math performance gap between demographic groups.
Title: Informing Educational Measurement with Test-Taking Process Data
Abstract: Computerized testing affords the collection of additional behavioral data beyond scores on test questions. One such instance is process data, the computer-recorded log files generated from test-taker interactions with a computerized assessment, e.g., keystrokes and clickstreams, in pursuit of solving a problem. This type of data offers rich insights into the cognitive processes underlying problem-solving, opening new avenues to address existing measurement and educational questions and explore novel ones. However, like constructed responses on open-ended questions with infinitely possible answers of different lengths, process data are highly unstructured and often noisy. This precludes the direct application of many well-established tools and psychometric models for structured test response data. In this talk, I discuss how sequential features extracted from test-taking process data can be incorporated into latent variable modeling frameworks commonly used in educational measurement, to address common questions in measurement and educational research, such as improving score reliability and understanding the math performance gap between demographic groups.
October 14, 2024
Speaker: Dr. Sacha Epskamp, Department of Psychology, National University of Singapore
Workshop title: Psychometric network modeling with the psychonetrics R package
Workshop title: Psychometric network modeling with the psychonetrics R package
Workshop abstract: The freely available psychonetrics package for R provides an encompassing framework for psychometric network modeling, combining typical practices in Structural Equation Modeling (SEM) with undirected network modeling now commonly used in network psychometrics. The psychonetrics package can be used for various types of data (cross-sectional, time-series and panel data), and not only allows researchers to explore relations between observed and latent variables through the use of network models, but also allows researchers to perform confirmatory tests on given network structures and to test for homogeneity in (latent) network structures across groups. This workshop will introduce participants to the psychonetrics package and will teach participants to estimate psychometric network models from cross-sectional data. Familiarity with R and having R and the psychonetrics package installed are recommended for attending the workshop.
October 15, 2024
Speaker: Dr. Sacha Epskamp, Department of Psychology, National University of Singapore
Research talk title: An Introduction to Network Psychometrics
Research talk abstract: Recent years have seen a rise in the network approach to psychology: modeling human behavior as a complex system of interacting psychological, biological, and sociological components. This approach has since been utilized in various topics of psychological research, such as mental health, attitude formation and intelligence. While the approach originated from a complex systems point of view, it has since inspired new psychometric modelling frameworks that are made available in open-source and accessible software packages – Network Psychometrics. Network psychometrics is both an alternative to and complementary with traditional psychometric modeling techniques such as Structural Equation Modeling (SEM). In this seminar, I will introduce network psychometrics and showcase some recent applications of psychometric network modeling, such as meta-analytic modeling of multiple datasets and separating within-person dynamics from between-person relations in longitudinal panel data.
November 4, 2024
Speaker: Dr. Yi Feng, Assistant Professor of Quantitative Psychology, UCLA
Title: Ask What You Mean and Mean What You Ask: Strategic Reparameterization of Latent Growth Curve Models
Abstract: Latent growth modeling (LGM) has been widely used in longitudinal studies. Falling within the structural equation modeling (SEM) framework, LGM allows the researchers to examine individuals’ longitudinal growth in measured or latent outcome variables. Although linear growth models are most commonly seen in practice, LGM is actually a far more flexible analytical framework than that. LGM is not only able to accommodate nonlinear growth trajectories, but also the change of different functional forms over time. More importantly, it is also possible to strategically reparameterize the model, such that the growth aspects of focal research interest can be directly estimated as fixed model parameters or random coefficients. During this presentation, the SEM-based structured latent curve modeling (SLCM) approach for modeling nonlinear trajectories (Blozis, 2004, 2007) will first be introduced, followed by the general approach outlined by Preacher and Hancock (2012, 2015) that allows us to convert almost any aspect of change into a fixed or random coefficient in LGM.
Examples of model reparameterization using the above approaches will be provided, giving particular attention to strategically reparameterizing a piecewise latent growth model, which can be used to model growth trajectories that are bounded by a floor and a ceiling at the two ends of the observation period (Feng et al., 2019). With the proposed reparameterized model, researchers will be able to directly examine the transition points, floor levels, and/or ceiling levels as fixed or random coefficients. Real data examples will be presented to demonstrate the implementation of the reparameterized models.
November 25, 2024
Speaker: Dr. Feng Ji, Department of Applied Psychology and Human Development, University of Toronto
Title: Valid standard errors for Bayesian quantile regression with clustered and independent data
Abstract: In Bayesian quantile regression, the most commonly used likelihood is the asymmetric Laplace (AL) likelihood. The reason for this choice is not that it is a plausible data-generating model but that the corresponding maximum likelihood estimator is identical to the classical estimator by Koenker and Bassett (1978), and in that sense, the AL likelihood can be thought of as a working likelihood. AL-based quantile regression has been shown to produce good finite-sample Bayesian point estimates and to be consistent. However, if the AL distribution does not correspond to the data-generating distribution, credible intervals based on posterior standard deviations can have poor coverage. Yang, Wang, and He (2016) proposed an adjustment to the posterior covariance matrix that produces asymptotically valid intervals. However, we show that this adjustment is sensitive to the choice of scale parameter for the AL likelihood and can lead to poor coverage when the sample size is small to moderate. We therefore propose using Infinitesimal Jackknife (IJ) standard errors (Giordano & Broderick, 2023). These standard errors do not require resampling but can be obtained from a single MCMC run. We also propose a version of IJ standard errors for clustered data. Simulations and applications to real data show that the IJ standard errors have good frequentist properties, both for independent and clustered data. We provide an R-package, IJSE, that computes IJ standard errors for clustered or independent data after estimation with the brms wrapper in R for Stan.
Abstract: In Bayesian quantile regression, the most commonly used likelihood is the asymmetric Laplace (AL) likelihood. The reason for this choice is not that it is a plausible data-generating model but that the corresponding maximum likelihood estimator is identical to the classical estimator by Koenker and Bassett (1978), and in that sense, the AL likelihood can be thought of as a working likelihood. AL-based quantile regression has been shown to produce good finite-sample Bayesian point estimates and to be consistent. However, if the AL distribution does not correspond to the data-generating distribution, credible intervals based on posterior standard deviations can have poor coverage. Yang, Wang, and He (2016) proposed an adjustment to the posterior covariance matrix that produces asymptotically valid intervals. However, we show that this adjustment is sensitive to the choice of scale parameter for the AL likelihood and can lead to poor coverage when the sample size is small to moderate. We therefore propose using Infinitesimal Jackknife (IJ) standard errors (Giordano & Broderick, 2023). These standard errors do not require resampling but can be obtained from a single MCMC run. We also propose a version of IJ standard errors for clustered data. Simulations and applications to real data show that the IJ standard errors have good frequentist properties, both for independent and clustered data. We provide an R-package, IJSE, that computes IJ standard errors for clustered or independent data after estimation with the brms wrapper in R for Stan.