Hi everyone,
This Friday, June 26 at noon Eastern time, the International Methods
Colloquium will host two methodology presentations originally slated for
the 2020 Annual Meeting of the Midwest Political Science Association that
was cancelled due to the outbreak of COVID-19. These presentations are:
****************************************************************************
Presentation #1: How to Cautiously Uncover the 'Black Box' of Machine
Learning Models
Authors: Soren Jordan (Auburn), Hannah L. Paul (CU Boulder) and Andrew
Philips (CU Boulder)
Abstract: Machine learning models, especially ensemble and tree-based
approaches, have been criticized as `black-box' techniques due to their
difficulty in visualizing the effect of predictors on the outcome of
interest. While political scientists have started to utilize visual aids to
interpret these effects, they have exclusively used plotting strategies
which rely on strong assumptions that are not likely to be met in applied
data. In this paper we survey several recent graphical tools that involve
less assumptions and can better show effect heterogeneity. We also identify
a number of issues with these plots that can arise when working with social
science data, and propose solutions. We illustrate this with several
applied data examples. We recommend that authors test the robustness of
their findings using a variety of graphical approaches when interpreting
machine learning models.
****************************************************************************
Presentation #2: Forecasting rare events: Two-stage classification of
terrorist, civil wars, and coups
Author: Jared Edgerton (Ohio State) and Grant Buckles (Gallup)
Abstract: Predicting civil wars, terrorist attacks, and coup attempts
remains a primary goal of policy makers. In turn, a more precise
understanding of where these political events are likely to happen could
greatly benefit the United States and its allies, as well as help protect
civilians. In the present analysis, we demonstrate a new framework for
forecasting rare events. Specifically, we employ a new two-stage framework
and permutation upsampling to identify countries which are more likely to
experience terrorism. The presented two-stage framework uses novel
micro-level data gathered by the Gallup World Poll initiative. The Gallup
World Poll survey asks respondents over 200 questions relating to health
outcomes, views on governance, economic security, among others. Through
this process, we improve on existing machine learning classification
processes for predicting rare events.
****************************************************************************
To tune in to the presentations and participate in the discussion after the
talks, visit http://www.methods-colloquium.com/ and click "Watch Now!" on
the day of the talk. To register for the talk in advance, click here:
https://zoom.us/webinar/register/WN_AOQw2rbESp2PY1nzqg17GQ
The IMC uses Zoom, which is free to use for listeners and works on PCs,
Macs, and iOS and Android tablets and phones. You can be a part of the talk
from anywhere around the world with access to the Internet. The
presentations and Q&A will last for a total of one hour.
I hope to see you there!
-JE
--
Dr. Justin Esarey
Associate Professor of Politics and International Affairs
Wake Forest University
Voice: (678) 383-9629
Fax: (336) 758-6104
E-mail: <[log in to unmask]>[log in to unmask]
Web: www.justinesarey.com
**************************************************************
Political Methodology E-Mail List
Editor: Erin Rossiter <[log in to unmask]>
**************************************************************
Send messages to [log in to unmask]
To join the list, cancel your subscription, or modify
your subscription settings visit:
https://www.cambridge.org/core/membership/spm/mailing-list
**************************************************************
|
