QSG this week features two talks. The title of the first talk is “An Introduction to Penalized Spline“ by Chris Qiu. The title of the second talk is “Text for Prediction: A Bayesian Supervised Topic Model with Covariates“ by Tyler Wilcox. The abstracts for both talks are given below.
An Introduction to Penalized Spline by Chris Qiu.
Penalized spline (P-spline) has been used as a flexible tool for modeling nonlinear patterns nonparametrically and semiparametrically. It strikes a balance between fitting complex patterns using lower-order polynomials and maintaining a relatively smooth form. Moreover, by expressing P-spline in a linear mixed-effect (LME) models framework, P-spline models can not only be fitted using commonly used software (e.g., nlme in R and PROX MIXED in SAS) but be extended to a wide range of usages. In this talk, we provide an overview of P-spline, describe the connection between P-spline and LME models, and extend P-spline to other modeling purposes.
Text for Prediction: A Bayesian Supervised Topic Model with Covariates by Tyler Wilcox.
Although ubiquitous, text remains an underused source of data in psychological research. While most approaches to utilizing text within psychology rely on top-down dictionary-based methods or model-free algorithms such as latent semantic indexing, recent development of probabilistic models, commonly known as topic models, have opened up new possibilities for the use of textual data. I extend the supervised topic model of Blei and McAuliffe (2010) to simultaneously model textual data with a latent variable topic model and estimate a regression model of an outcome variable on both the latent topics and a set of manifest variables. This model (sLDAX) improves upon existing two-stage approaches in psychology and other areas. I propose a Bayesian estimation algorithm for fitting the sLDAX model for either a normal and dichotomous outcome and discuss an in-development of an R package (psychtm) to facilitate the use of this and other related models for researchers. I illustrate the sLDAX model in an empirical study of the relationship between nonsuicidal self-injury and (1) narratives regarding neutral and negative interpersonal interactions and (2) a measure of emotional regulation. Finally, I discuss open questions regarding the sLDAX model and future directions for research.
QSG meets on Thursdays from 3:30-4:45pm in Corbett 378. All are welcome, and we hope to see you there.
For a complete list of speakers for Spring 2019, please visit this link.
The primary objective is to provide students the opportunity to develop their critical thinking skills, presentation abilities, and knowledge of the most recent developments in quantitative and statistical methods and techniques. The seminar format of this course is designed to stimulate and foster the intellectual environment of the program and department as well as to engage students at all levels. This is one of the required courses for non-quantitative students to get their minor in quantitative psychology. Quant minor requirements: https://psychology.nd.edu/graduate-programs/areas-of-study/quantitative/minor/