Robert Chew

Robert Chew, John Bollenbacher, Michael Wenger et al.

Deductive coding is a widely used qualitative research method for determining the prevalence of themes across documents. While useful, deductive coding is often burdensome and time consuming since it requires researchers to read, interpret, and reliably categorize a large body of unstructured text documents. Large language models (LLMs), like ChatGPT, are a class of quickly evolving AI tools that can perform a range of natural language processing and reasoning tasks. In this study, we explore the use of LLMs to reduce the time it takes for deductive coding while retaining the flexibility of a traditional content analysis. We outline the proposed approach, called LLM-assisted content analysis (LACA), along with an in-depth case study using GPT-3.5 for LACA on a publicly available deductive coding data set. Additionally, we conduct an empirical benchmark using LACA on 4 publicly available data sets to assess the broader question of how well GPT-3.5 performs across a range of deductive coding tasks. Overall, we find that GPT-3.5 can often perform deductive coding at levels of agreement comparable to human coders. Additionally, we demonstrate that LACA can help refine prompts for deductive coding, identify codes for which an LLM is randomly guessing, and help assess when to use LLMs vs. human coders for deductive coding. We conclude with several implications for future practice of deductive coding and related research methods.

Christoph Kern, Stephanie Eckman, Jacob Beck et al.

When training data are collected from human annotators, the design of the annotation instrument, the instructions given to annotators, the characteristics of the annotators, and their interactions can impact training data. This study demonstrates that design choices made when creating an annotation instrument also impact the models trained on the resulting annotations. We introduce the term annotation sensitivity to refer to the impact of annotation data collection methods on the annotations themselves and on downstream model performance and predictions. We collect annotations of hate speech and offensive language in five experimental conditions of an annotation instrument, randomly assigning annotators to conditions. We then fine-tune BERT models on each of the five resulting datasets and evaluate model performance on a holdout portion of each condition. We find considerable differences between the conditions for 1) the share of hate speech/offensive language annotations, 2) model performance, 3) model predictions, and 4) model learning curves. Our results emphasize the crucial role played by the annotation instrument which has received little attention in the machine learning literature. We call for additional research into how and why the instrument impacts the annotations to inform the development of best practices in instrument design.

Robert Chew, Stephanie Eckman, Christoph Kern et al.

Supervised machine learning assumes that labeled data provide accurate measurements of the concepts models are meant to learn. Yet in practice, human labeling introduces systematic variation arising from ambiguous items, divergent interpretations, and simple mistakes. Machine learning research commonly treats all disagreement as noise, which obscures these distinctions and limits our understanding of what models actually learn. This paper reframes annotation as a measurement process and introduces a statistical framework for decomposing labeling outcomes into interpretable sources of variation: instance difficulty, annotator bias, situational noise, and relational alignment. The framework extends classical measurement-error models to accommodate both shared and individualized notions of truth, reflecting traditional and human label variation interpretations of error, and provides a diagnostic for assessing which regime better characterizes a given task. Applying the proposed model to a multi-annotator natural language inference dataset, we find empirical evidence for all four theorized components and demonstrate the effectiveness of our approach. We conclude with implications for data-centric machine learning and outline how this approach can guide the development of a more systematic science of labeling.

Rob Chew, Michael Wenger, Caroline Kery et al.

SMART is an open source web application designed to help data scientists and research teams efficiently build labeled training data sets for supervised machine learning tasks. SMART provides users with an intuitive interface for creating labeled data sets, supports active learning to help reduce the required amount of labeled data, and incorporates inter-rater reliability statistics to provide insight into label quality. SMART is designed to be platform agnostic and easily deployable to meet the needs of as many different research teams as possible. The project website contains links to the code repository and extensive user documentation.

Jorge Cisneros, Timothy Wojan, Matthew Williams et al.

Public-use microdata samples (PUMS) from the United States (US) Census Bureau on individuals have been available for decades. However, large increases in computing power and the greater availability of Big Data have dramatically increased the probability of re-identifying anonymized data, potentially violating the pledge of confidentiality given to survey respondents. Data science tools can be used to produce synthetic data that preserve critical moments of the empirical data but do not contain the records of any existing individual respondent or business. Developing public-use firm data from surveys presents unique challenges different from demographic data, because there is a lack of anonymity and certain industries can be easily identified in each geographic area. This paper briefly describes a machine learning model used to construct a synthetic PUMS based on the Annual Business Survey (ABS) and discusses various quality metrics. Although the ABS PUMS is currently being refined and results are confidential, we present two synthetic PUMS developed for the 2007 Survey of Business Owners, similar to the ABS business data. Econometric replication of a high impact analysis published in Small Business Economics demonstrates the verisimilitude of the synthetic data to the true data and motivates discussion of possible ABS use cases.

Robert Chew, Matthew R. Williams, Elan A. Segarra et al.

Differential privacy (DP) is becoming increasingly important for deployed machine learning applications because it provides strong guarantees for protecting the privacy of individuals whose data is used to train models. However, DP mechanisms commonly used in machine learning tend to struggle on many real world distributions, including highly imbalanced or small labeled training sets. In this work, we propose a new scalable DP mechanism for deep learning models, SWAG-PPM, by using a pseudo posterior distribution that downweights by-record likelihood contributions proportionally to their disclosure risks as the randomized mechanism. As a motivating example from official statistics, we demonstrate SWAG-PPM on a workplace injury text classification task using a highly imbalanced public dataset published by the U.S. Occupational Safety and Health Administration (OSHA). We find that SWAG-PPM exhibits only modest utility degradation against a non-private comparator while greatly outperforming the industry standard DP-SGD for a similar privacy budget.