Ex-Post Harmonisation and “Statistical” Data Provenance

SoDa Lab Meeting, LMU Munich, 16 Jun 2024

Cynthia A. Huang

Department of Econometrics and Business Statistics, Monash Business School

Introduction

💱 Previously:
- Economics at the University of Melbourne
- Tutoring undergraduate economics
- Assisting with data collection & curation for empirical economists

👩🏻 Outside of Research:
- 🧗🏻‍♀️ Climbing, 🧘🏻‍♀️ Yoga, 👩🏻‍🍳 Foodie
- 🎙️ Regular host on The Random Sample podcast

📊 Research Interests
- 🌰 Statistically sound, well-documented and low-friction adaptation of “alternative” data for research purposes.
- 🖇️ Data provenance models that capture both statistical decisions, and computational implementation details.
👩‍🎓 Thesis: Unified Statistical Principles and Computational Tools for Data Harmonisation and Provenance
- Conceptual framework for redistributing numeric mass between categories in related statisical classifications
- Software implementation in R

📋 Collaborative work:
- Review of Data Provenance approaches across CS and Statistics
- Adapting web-scraped retail product & price data for public health research
- Human in the Loop verification for data extraction from spreadsheets using Generative AI
💡 Reproducible and reusable research and teaching tools:
- learnr.academy built on Quarto & webR
- Reproducibility for Economists also with Quarto

Opportunities to combine existing data for analysis abound,
Existing literature exists on a spectrum from conceptual to applied,
with keywords such as data preprocessing, cleaning, fusion, integration, harmonisation etc.

Defining or selecting mappings between classifications or taxonomies,

Implementing and validating mappings on given data,

Documenting and analysing the implemented mapping.

Approaches and insights are scattered across “data science” fields:
- Multi-resolution, multi-phase and multi-source inference (Meng 2014)
- Data wrangling workflows and domain specific languages (Kandel et al. 2011; Wickham 2014)
- Guidelines for data quality and provenance for trustworthy research (Gebru et al. 2021; Peng 2011; Wilkinson et al. 2016)
- Frameworks for considering alternative preprocessing decisions(Steegen et al. 2016; Blocker and Meng 2013)

Many attempts at domain-specific and general purpose tools exist:
- Survey Data Harmonisation (Kołczyńska 2022)
- Data validation tools (van der Loo and de Jonge 2021)
- Data provenance tracking and documentation (Lucchesi et al. 2022)
- Interactive tools for Multiverse Analysis (Sarma et al. 2018; Liu et al. 2021)