RMME/STAT Joint Colloquium

thinkCausal: One Stop Shopping for Answering your Causal Inference Questions

Dr. Jennifer Hill
New York University

Friday, April 30th, at 12:00PM ET

https://uconn-cmr.webex.com/uconn-cmr/j.php?MTID=m8c032f2f335a1c377fcd8a293df02bbc

Causal inference is a necessary tool in education research for answering pressing and ever-evolving questions around policy and practice. Increasingly, researchers are using more complicated machine learning algorithms to estimate causal effects. These methods take some of the guesswork out of analyses, decrease the opportunity for “p-hacking,” and are often better suited for more fine-tuned causal inference tasks such as identifying varying treatment effects and generalizing results from one population to another. However, these more sophisticated methods are more difficult to understand and are often only accessible in more technical, less user-friendly software packages. The thinkCausal project is working to address these challenges (and more) by developing a highly scaffolded multi-purpose causal inference software package with the BART predictive algorithm as a foundation. The software will scaffold the researcher through the data analytic process and provide options to access technology-based teaching tools to understand foundational concepts in causal inference and machine learning. This talk will briefly review BART for causal inference and then discuss the challenges and opportunities in building this type of tool. This is work in progress and the goal is to create a conversation about the tool and role of education in data analysis software more broadly.

Loading...

Taking too long?

Reload document

| Open in new tab

RMME/STAT Joint Colloquium

Causal Inference Under Interference in Dynamic Therapy Group Studies

Dr. Susan Paddock
NORC University of Chicago

Friday, April 16th, at 12:00PM ET

https://uconn-cmr.webex.com/uconn-cmr/j.php?MTID=mdbdf7c1935ee0cdcc88e0a90573ea2fc

Group therapy is a common treatment modality for behavioral health conditions. Patients often enter and exit groups on an ongoing basis, leading to dynamic therapy groups. Examining the effect of high versus low session attendance on patient outcomes is of interest. However, there are several challenges to identifying causal effects in this setting, including the lack of randomization, interference among patients, and the interrelatedness of patient participation. Dynamic therapy groups motivate a unique causal inference scenario, as the treatment statuses are completely defined by the patient attendance record for the therapy session, which is also the structure inducing interference. We adopt the Rubin Causal Model framework to define the causal effect of high versus low session attendance of group therapy at both the individual patient and peer levels. We propose a strategy to identify individual, peer, and total effects of high attendance versus low attendance on patient outcomes by the prognostic score stratification. We examine performance of our approach via simulation, apply it to data from a group cognitive behavioral therapy trial for reducing depressive symptoms among patients in a substance use disorders treatment setting, and discuss the strengths and limitations of this approach.

Loading...

Taking too long?

Reload document

| Open in new tab

RMME/STAT Joint Colloquium

Bayesian Pyramids: Identifying Interpretable Deep Structure Underlying High-dimensional Data

David Dunson
Duke University

Friday, March 26th, at 12:00PM ET

https://uconn-cmr.webex.com/uconn-cmr/j.php?MTID=m09a58d2d0b8f3973e89583e46454fbfa

High-dimensional categorical data are routinely collected in biomedical and social sciences. It is of great importance to build interpretable models that perform dimension reduction and uncover meaningful latent structures from such discrete data. Identifiability is a fundamental requirement for valid modeling and inference in such scenarios yet is challenging to address when there are complex latent structures. We propose a class of interpretable discrete latent structure models for discrete data and develop a general identifiability theory. Our theory is applicable to various types of latent structures, ranging from a single latent variable to deep layers of latent variables organized in a sparse graph (termed a Bayesian pyramid). The proposed identifiability conditions can ensure Bayesian posterior consistency under suitable priors. As an illustration, we consider the two-latent-layer model and propose a Bayesian shrinkage estimation approach. Simulation results for this model corroborate identifiability and estimability of the model parameters. Applications of the methodology to DNA nucleotide sequence data uncover discrete latent features that are both interpretable and highly predictive of sequence types. The proposed framework provides a recipe for interpretable unsupervised learning of discrete data and can be a useful alternative to popular machine learning methods.

Loading...

Taking too long?

Reload document

| Open in new tab

RMME/STAT Joint Colloquium:

Partially Ordered Responses and Applications

Edward Ip
Wake Forest University

Friday, February 26^th, at 12:00PM EST

https://uconn-cmr.webex.com/uconn-cmr/j.php?MTID=mc1ed9d99b7e6b63ab2796d729867e365

Partially ordered set (poset) responses are prevalent in fields such as psychology, education, and health. For example, the psychopathologic classification of no anxiety (NA), mild anxiety (MA), anxiety with depression (AwD), and severe anxiety (SA) form a poset. Due in part to the lack of analytic tools, poset responses are often collapsed into other data forms such as ordinal data. During such a process, subtle information within a poset is inevitably lost. In this presentation, a longitudinal latent-variable model for poset responses and its application to health data will be described. It is argued that latent variable modeling enables the integration of information from both ordinal and nominal components in a poset. Using the abovementioned example, NA>{MA,AwD}>SA form the ordinal component, and MA and AwD form the nominal component. Specifically, it will be demonstrated that the latent variable model “discovers” implicit ordering within the nominal categories. This is possible because both intra-person and inter-person information are borrowed to reinforce inference. Some potential applications of the poset model will also be highlighted.

Loading...

Taking too long?

Reload document

| Open in new tab

RMME faculty member, Dr. Chris Rhoads, will deliver a (virtual) keynote address at the LEAD retreat on April 16, 2021. The title of his talk is: Research Design for Educational Effectiveness Studies:  Statistical and Practical Considerations. The LEAD retreat is sponsored by the LEAD Graduate School and Research Network, which is based in Tubingen, Germany.

RMME/STAT Joint Colloquium:

The Bayesian Causal Forest Model: Regularization, Confounding, and Heterogeneous Effects

Richard Hahn
Arizona State University

January 29, 2021, at 12:00 EST

https://uconn-cmr.webex.com/uconn-cmr/j.php?MTID=mc19e545b14cc3a980ffc36760a5ce5f4

This talk will describe recent work on Bayesian supervised learning for conditional average treatment effects. Dr. Hahn will motivate the proposed Bayesian causal forest model in terms of fixing two specific flaws with previous approaches. One, our model allows for direct regularization of the treatment effect function, providing lower variance estimates of heterogeneous treatment effects. Two, by including an estimate of the propensity score as a control variable in our model we mitigate a phenomenon called “regularization induced confounding” that leads to substantial bias in previous approaches. Dr. Hahn will conclude with a detailed discussion of designing simulation studies to systematically investigate and validate machine learning models for causal inference.

Note: Dr. Hahn may also talk about this tutorial: https://math.la.asu.edu/~prhahn/xbcf_demo.html

Register Now! RMME faculty member, Dr. Bianca Montrosse-Moorhead will give a 3-module workshop in January 2021, sponsored by Encompass Learning Center.

This workshop will focus on best practices for combining different evaluation methods or frameworks. It is open to evaluators working all over the globe.

As a continuation of the presentation series this fall, the University of Connecticut’s Research Methods, Measurement, & Evaluation (RMME) program and Statistics department will jointly sponsor several additional RMME/STAT colloquia, starting in January of 2021. So, mark your calendar now! And as always, be sure to check the RMME website for more information as these talks approach!