Rarely Updated Blog
Variational Inference for MRP with Reliable Posterior Distributions
Part 4- Importance Sampling
MRP
Variational Inference
This is section 4 in my series on using Variational Inference to speed up relatively complex Bayesian models like Multilevel Regression and Poststratification without the approximation being of disastrously poor quality.
Variational Inference for MRP with Reliable Posterior Distributions
Part 3- Alternatives to KL Divergence
MRP
Variational Inference
Note: I’ve gotten a lot more pessimistic about how generally useful the alternatives to simple KL-Divergence are on their own since writing this post. I still think these are really useful ideas to think to build intuition about VI, and techniques like CHIVI are useful for some lower dimensional problems or as part of an ensemble of techniques for high dimensional ones. However, this paper from Dhaka et al. is very convincing that CHIVI and currently available similar algorithms are in practice very hard to optimize for high dimensional, and that some of the intuitive benefits shown about CHIVI below in low dimensions don’t really generalize the way we’d expect to higher dimensions.
Variational Inference for MRP with Reliable Posterior Distributions
Part 2- The errors of our ways
MRP
Variational Inference
This is the second post in my series on using Variational Inference to speed up relatively complex Bayesian models like Multilevel Regression and Poststratification without the approximation being of disastrously poor quality.
Variational Inference for MRP with Reliable Posterior Distributions
Introductions- things to do, places to be
MRP
BART
Variational Inference
This post introduces a series I intend to write, exploring using Variational Inference to massively speed up running complex survey estimation models like variants of Multilevel Regression and Poststratification while aiming to keep approximation error from completely ruining the model.
BART with varying intercepts in the MRP framework
From Old Website
MRP
BART
This is the first of a few posts about some of the substantive and modeling findings from my master’s thesis, where I use Bayesian Additive Regression Trees (BART) and Poststratification to model support for a border wall from 2015-2019. In this post I explore some of the properties of using BART with varying intercepts (BART-vi) within the MRP framework.
Convention Prediction with a Bayesian Hierarchical Multinomial Model
From Old Website
Stan
Here, I use a Bayesian hierarchical multinomial model to predict the first ballot results at the 2018 DFL (Democratic) State Convention, with data aggregated to the Party Unit level (ex: State Senate district) to guarantee anonymity. While using aggregated data obviously isn’t ideal, this sort of strategy shows a lot of promise, especially if individual level predictors could be harnessed as another level of the hierarchical model. As it stands, this is mostly a proof of concept for Bayesian hierarchical models in this context. To use something like this in practice, one could use prior predictive simulation to game out the convention under various assumptions, or condition on the first ballot data and use it to analyze trends in support and predict subsequent ballots as your floor team collects further data.
Is Voting Habit Forming? Replication, and additional robustness checks
From Old Website
causal inference
This post walks through my replication of the fuzzy regression discontinuity portion of Coppock and Green’s 2016 paper Is Voting Habit Forming? New Evidence from Experiments and Regression Discontinuities, and details some additional robustness checks I conducted. While I was able to reproduce all of their estimates fairly easily due to great replication materials, my additional robustness checks suggest that their results are more sensitive to bandwith choices than their testing suggests. Additionally, Coppock and Green argue the effects they find are likely due to habit alone, whereas I’m unconvinced that’s the sole mechanism involved. This is my work from Jennifer Hill and Joe Robinson-Cimpian’s Causal Inference class at NYU, and I’m grateful for both their feedback on the project.
Predicting race part 1- Bayes’ rule method and extensions
From Old Website
Race is a defining part of political identity in the United States, and so it should be no surprise that accurately modeling race can be beneficial for many political campaign activities. For instance, many organizations work to improve turnout in specific communities of color, or want to target persuasion on a given issue to certain racial group. Alternatively, race and ethnicity might be desired as an input to a larger voting or support likelihood model, given that race is generally predictive of both voting likelihood and candidate support.
No matching items