BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Drupal: Date iCal//NONSGML kigkonsult.se iCalcreator 2.16.12//
METHOD:PUBLISH
X-WR-CALNAME;VALUE=TEXT:Ical
BEGIN:VTIMEZONE
TZID:America/New_York
BEGIN:STANDARD
DTSTART:20201101T020000
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20210314T020000
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:calendar.33.field_date.0@polmeth2020.org
DTSTAMP:20200924T172123Z
DESCRIPTION:\n\n \n Fast and Accurate Estimation of Non-Nested Binomial Hiera
rchical Models Using Variational Inference\n \n\n \n Max Goplerud\n \n\n
\;\n\n\n\n Chair: Ludovic Rheault (University of Toronto)\n\n \;\n\n\n C
o-Host: Mikaela Karstens (Penn State University)\n\n\n\n Fast and Accurate
Estimation of Non-Nested Binomial Hierarchical Models Using Variational In
ference\n\n\n\n Download Paper\n\n\n\n Presentation Slides\n\n\n\n Author(s
): Max Goplerud\n\n\n\n Discussant: Justin Grimmer (Stanford University)\n
\n \;\n\n\n Estimating non-linear hierarchical models can be computatio
nally burdensome in the presence of large datasets and many non-nested ran
dom effects. Popular inferential techniques may take hours to fit even rel
atively straightforward models. This paper provides two contributions to s
calable and accurate inference. First\, I propose a new mean-field algorit
hm for estimating logistic hierarchical models with an arbitrary number of
non-nested random effects. Second\, I propose “marginally augmented varia
tional Bayes” (MAVB) that further improves the initial approximation throu
gh a post-processing step. I show that MAVB provides a guaranteed improvem
ent in the approximation quality at low computational cost and induces dep
endencies that were assumed away by the initial factorization assumptions.
I apply these techniques to a study of voter behavior. Existing estimatio
n took hours whereas the algorithms proposed run in minutes. The posterior
means are well-recovered even under strong factorization assumptions. App
lying MAVB further improves the approximation by partially correcting the
under-estimated variance. The proposed methodology is implemented in an op
en source software package.\n\nAdd to Calendar \;
DTSTART;TZID=America/New_York:20200716T143000
DTEND;TZID=America/New_York:20200716T160000
LAST-MODIFIED:20200804T075554Z
SUMMARY:Virtual Room 1: Hierarchical Models
URL;TYPE=URI:https://polmeth2020.org/event/hierarchical-models
END:VEVENT
END:VCALENDAR