PhD Thesis Defence - Stephen Kasica

Name: Stephen Kasica

Date: Monday, October 5, 2026

Time: 09:00 am 

Location: Zoom https://ubc.zoom.us/j/67791358769?pwd=RJRg7ayz13VD3TAgKNV275LavO3opC.1

Supervisor’s name: Tamara Munzner 

Title of the thesis: Data Wrangling in Journalism: Artifacts, Interviews, and an Interactive Visualization Tool for Handling Complexity Arising from Multiple Table Workflows

Abstract

To produce original reporting, data journalists routinely wrangle data, the iterative process of transforming and exploring data to enable analysis or visualization; however, existing data wrangling frameworks and interactive systems do not address the principal source of complexity, wrangling involving multiple tables. This dissertation asks what challenges journalists face when wrangling data, especially in a multi-table context, and how interactive visualization tools can empower journalists without programming expertise.

Three studies address this question. Table Scraps, a technical observation study of 50 public code repositories authored by data journalists, yields two descriptive taxonomies of wrangling actions and processes, grounded in 165 open and axial codes, and a Multi-Table Framework of 21 operations organized by data object (tables, columns, rows) and operation class (create, delete, transform, separate, combine). This framework shows multi-table wrangling is pervasive in journalistic practice but absent from prior conceptual models. Dirty Data in the Newsroom extends this foundation through a hybrid thematic analysis combining deductive codes from 16 data science workflow papers with inductive codes from 36 semi-structured interviews across 31 newsrooms. It contributes an augmented model of 23 data preparation activities, a model-discrepancy taxonomy of 60 dirty data issues characterizing data quality problems as mismatches between a worker's mental model and the underlying data model, and a typology of four multi-table integration challenges: regional, diachronic, fragmented, and disparate datasets.

Building on these findings, OpenRoundup translates the insight that multi-table wrangling is both pervasive and under-supported into a designed, implemented, and evaluated system: an open-source, browser-based tool that lets journalists consolidate multiple tables without programming. This system instantiates two design principles, eager table consolidation and a minimal declarative vocabulary called Stack and Pack, similar to SQL's \texttt{UNION} and \texttt{JOIN}. A replication study demonstrates the system's expressiveness across 17 published journalistic workflows spanning 2 to 51 source tables, while a deployment study with four professional journalists confirms its utility and usability in authentic reporting contexts.

Together, these contributions close a persistent gap in the tool landscape and research literature on data wrangling, showing that multi-table complexity in data journalism can be studied empirically, understood theoretically, and addressed through design.