arXiv subjects

Graham Tierney

Publications and source records attributed to Graham Tierney.

8 recordsLinked to original sources

arXiv2026

The Confounder Trap: Treatment-Encoding Representations in Causal Inference with Text

Estimating causal effects of linguistic properties from observational text is difficult because the same document can contain both the treatment of interest and the non-treatment textual attributes needed for adjustment. Existing approaches often learn representations from the full text to capture latent confounding, but when treatment status is itself encoded by words in the text, these representations can directly encode treatment. This creates a confounder trap: richer representations can make treated and control documents separable, inducing overlap violations even when the underlying causal problem satisfies overlap. We study latent text treatments that are encoded through lexicons or other treatment-defining lexical information, and propose masking-based adjustment representations that remove this lexical treatment signal before representation learning. We formalize representation-induced overlap failure, prove that deletion masking preserves overlap for bag-of-words/topic-model representations, and characterize replacement masking as a natural relaxation for large language models that hides treatment-defining tokens while preserving word order and context. Across simulations, masking improves overlap diagnostics, stabilizes treatment effect estimates, and reduces bias relative to adjustment methods that learn from the unmasked text.

stat.ME↗

arXiv2025

A Design-based Solution for Causal Inference with Text: Can a Language Model Be Too Large?

Many social science questions ask how linguistic properties causally affect an audience's attitudes and behaviors. Because text properties are often interlinked (e.g., angry reviews use profane language), we must control for possible latent confounding to isolate causal effects. Recent literature proposes adapting large language models (LLMs) to learn latent representations of text that successfully predict both treatment and the outcome. However, because the treatment is a component of the text, these deep learning methods risk learning representations that actually encode the treatment itself, inducing overlap bias. Rather than depending on post-hoc adjustments, we introduce a new experimental design that handles latent confounding, avoids the overlap issue, and unbiasedly estimates treatment effects. We apply this design in an experiment evaluating the persuasiveness of expressing humility in political communication. Methodologically, we demonstrate that LLM-based methods perform worse than even simple bag-of-words models using our real text and outcomes from our experiment. Substantively, we isolate the causal effect of expressing humility on the perceived persuasiveness of political statements, offering new insights on communication effects for social media platforms, policy makers, and social scientists.

stat.ME↗

arXiv2024

Compositional dynamic modelling for causal prediction in multivariate time series

Theoretical developments in sequential Bayesian analysis of multivariate dynamic models underlie new methodology for causal prediction. This extends the utility of existing models with computationally efficient methodology, enabling routine exploration of Bayesian counterfactual analyses with multiple selected time series as synthetic controls. Methodological contributions also define the concept of outcome adaptive modelling to monitor and inferentially respond to changes in experimental time series following interventions designed to explore causal effects. The benefits of sequential analyses with time-varying parameter models for causal investigations are inherited in this broader setting. A case study in commercial causal analysis-- involving retail revenue outcomes related to marketing interventions-- highlights the methodological advances.

stat.ME↗

arXiv2023

Reply to Discussions of "Multivariate Dynamic Modeling for Bayesian Forecasting of Business Revenue"

We are most grateful to all discussants for their positive comments and many thought-provoking questions. In addition, the discussants provide a number of useful leads into various areas of the literatures on time series, forecasting and commercial application within which the work in our paper is, of course, just one contribution linked to multiple threads. Our view is that, collectively, the discussion contributions nicely expand on the core of the paper and together -- with multiple additional references -- provide an excellent point-of-entr\'ee to the broader field of retail forecasting and its research challenges. Interested readers are encouraged to dig deeply into the discussions and our responses here, and explore referenced sources. There are several themes that recur across discussants, as well as a range of specific points/questions raised. Following some "big-picture" comments on our perspectives on Bayesian forecasting systems, we comment in turn on some specifics in each contribution.

stat.ME↗

arXiv2023

Multivariate Bayesian dynamic modeling for causal prediction

Bayesian forecasting is developed in multivariate time series analysis for causal inference. Causal evaluation of sequentially observed time series data from control and treated units focuses on the impacts of interventions using contemporaneous outcomes in control units. Methodological developments here concern multivariate dynamic models for time-varying effects across multiple treated units with explicit foci on sequential learning and aggregation of intervention effects. Analysis explores dimension reduction across multiple synthetic counterfactual predictors. Computational advances leverage fully conjugate models for efficient sequential learning and inference, including cross-unit correlations and their time variation. This allows full uncertainty quantification on model hyper-parameters via Bayesian model averaging. A detailed case study evaluates interventions in a supermarket promotions experiment, with coupled predictive analyses in selected regions of a large-scale commercial system. Comparisons with existing methods highlight the issues of appropriate uncertainty quantification in casual inference in aggregation across treated units, among other practical concerns.

stat.ME↗

arXiv2022

Bias and Excess Variance in Election Polling: A Not-So-Hidden Markov Model

With historic misses in the 2016 and 2020 US Presidential elections, interest in measuring polling errors has increased. The most common method for measuring directional errors and non-sampling excess variability during a postmortem for an election is by assessing the difference between the poll result and election result for polls conducted within a few days of the day of the election. Analyzing such polling error data is notoriously difficult with typical models being extremely sensitive to the time between the poll and the election. We leverage hidden Markov models traditionally used for election forecasting to flexibly capture time-varying preferences and treat the election result as a peak at the typically hidden Markovian process. Our results are much less sensitive to the choice of time window, avoid conflating shifting preferences with polling error, and are more interpretable despite a highly flexible model. We demonstrate these results with data on polls from the 2004 through 2020 US Presidential elections and 1992 through 2020 US Senate elections, concluding that previously reported estimates of bias in Presidential elections were too extreme by 10\%, estimated bias in Senatorial elections was too extreme by 25\%, and excess variability estimates were also too large.

stat.AP↗

arXiv2021

Multivariate Dynamic Modeling for Bayesian Forecasting of Business Revenue

Forecasting enterprise-wide revenue is critical to many companies and presents several challenges and opportunities for significant business impact. This case study is based on model developments to address these challenges for forecasting in a large-scale retail company. Focused on multivariate revenue forecasting across collections of supermarkets and product Categories, hierarchical dynamic models are natural: these are able to couple revenue streams in an integrated forecasting model, while allowing conditional decoupling to enable relevant and sensitive analysis together with scalable computation. Structured models exploit multi-scale modeling to cascade information on price and promotion activities as predictors relevant across Categories and groups of stores. With a context-relevant focus on forecasting revenue 12 weeks ahead, the study highlights product Categories that benefit from multi-scale information, defines insights into when, how and why multivariate models improve forecast accuracy, and shows how cross-Category dependencies can relate to promotion decisions in one Category impacting others. Bayesian modeling developments underlying the case study are accessible in custom code for interested readers.

stat.ME↗

arXiv2021

Author Clustering and Topic Estimation for Short Texts

Analysis of short text, such as social media posts, is extremely difficult because of their inherent brevity. In addition to classifying topics of such posts, a common downstream task is grouping the authors of these documents for subsequent analyses. We propose a novel model that expands on the Latent Dirichlet Allocation by modeling strong dependence among the words in the same document, with user-level topic distributions. We also simultaneously cluster users, removing the need for post-hoc cluster estimation and improving topic estimation by shrinking noisy user-level topic distributions towards typical values. Our method performs as well as -- or better -- than traditional approaches, and we demonstrate its usefulness on a dataset of tweets from United States Senators, recovering both meaningful topics and clusters that reflect partisan ideology. We also develop a novel measure of echo chambers among these politicians by characterizing insularity of topics discussed by groups of Senators and provide uncertainty quantification.

cs.IR↗