Session 3

Data Collection

Author
Affiliations

Moritz Mähr

University of Bern

University of Basel

Published

February 27, 2025

Modified

May 31, 2025

TLDR

  • Recap Session 2
  • Leseauftrag gemeinsam anschauen
  • Datenbegriff klären

Recap Session 2

mündlich

Leseauftrag “All Data Are Local”

“All Data Are Local” – A Critical Review

Introduction: Why Data Are Never Neutral

In All Data Are Local, Yanni Alexander Loukissas challenges the common belief that data are neutral, universal, and objective. He argues that data are deeply embedded in local, historical, and institutional conditions, making it impossible to separate them from their context. Instead of treating data sets as isolated and self-contained, he urges readers to examine data settings—the environments in which data are produced, organized, and used.

📌 Key Themes & Arguments

1. The Locality of Data: Four Case Studies

Loukissas introduces four examples to illustrate how data are shaped by their origins:

  • Harvard’s Arnold Arboretum: A data record for a cherry tree mistakenly attributes its collection to a botanist who had died years earlier, highlighting inconsistencies in institutional data.
  • Digital Public Library of America (DPLA): Different institutions contribute metadata in varying formats, causing classification inconsistencies.
  • NewsScape (TV News Archive): Data are inseparable from the algorithms that process them, shaping what information is surfaced or obscured.
  • Zillow (Real Estate Data): Zillow provides transparency in the housing market, yet masks structural inequalities.
2. The Problem with Data as “Sets”
  • The term data set implies that data are complete, standardized, and universally applicable, which is misleading.
  • Instead, data settings acknowledge the social, institutional, and technological environments that shape data collection and interpretation.
  • Understanding the context of data creation is essential to prevent misinterpretation.
3. The Rise of Data Skepticism
  • In the early 2010s, skepticism toward data neutrality increased, with concerns about:
    • Algorithmic bias (e.g., Google’s search algorithms reinforcing stereotypes).
    • Misinformation & manipulation (e.g., the role of fake news in the 2016 U.S. election).
    • P-Hacking in academic research, where scientists manipulate statistical analyses for misleadingly significant results.
4. From Data Collection to Critical Data Practices
  • Identifying bias isn’t enough—we must change how we engage with data.
  • Recognizing locality in data allows for mitigation of biases, context-aware findings, and ethical use.
  • Loukissas advocates for a reflexive, comparative, and critical approach to data work.
5. Case Studies as a Framework for Critical Thinking

Each chapter explores six core principles through real-world examples: - Data are attached to places (Arnold Arboretum). - Data come from heterogeneous sources (DPLA). - Data and algorithms are intertwined (NewsScape). - Interfaces shape data perception (Zillow). - Later chapters offer practical guidelines for ethical data use.

6. Resisting Digital Universalism
  • The myth of digital universalism assumes that technology transcends place and context.
  • This belief, rooted in Silicon Valley ideology, ignores the cultural, economic, and political power structures embedded in data systems.
  • Loukissas calls for resisting universalism by acknowledging the locality of data, ensuring it remains accountable to its origins and impacts.

🗂 Summary of Chapter 3: “Collecting Infrastructures”

How Data Infrastructures Shape Knowledge

This chapter explores data infrastructures, particularly the DPLA, and questions whether data can truly be separated from their local origins.

📌 Key Insights:
  • Data Standardization vs. Context Loss:
    • The DPLA standardizes data across institutions, but this often erases unique local contexts.
    • Its MAP (Metadata Application Profile) forces diverse data sources into a rigid structure.
  • The Role of Locality in Data:
    • Institutions classify data differently (e.g., the term Upstate means different things in different regions).
    • Historical bias: Some institutions categorize race inconsistently or exclude certain demographics.
  • Data Visualization as a Critical Tool:
    • The Library Observatory uses a tree-map visualization to show how different institutions contribute to the DPLA.
    • The Temporalities Project highlights inconsistencies in date formatting across data sources.
  • The Influence of Vannevar Bush’s Memex:
    • Bush’s 1945 vision of a universal digital archive influences modern data infrastructures.
    • Loukissas critiques this ambition, arguing that knowledge cannot be divorced from its social and institutional context.
  • The Political & Ethical Stakes of Data Infrastructures:
    • Large, well-funded institutions like Smithsonian or Getty dominate data collection, reinforcing power imbalances.
    • Loukissas calls for counterdata infrastructures that challenge dominant narratives and promote inclusive histories.

🗂 Summary of Chapter 7: “Beyond Data Sets”

Why Open Data Isn’t Enough

Loukissas argues that data accessibility does not guarantee understanding. Instead of simply providing open data, we need contextualized guides that explain their origins and limitations.

📌 Rethinking the Goals of Data Work

Loukissas contrasts traditional data objectives with alternative, locally grounded goals:

Traditional Goals Local Alternatives Explanation
Orientation Place-Making Data should not just help users navigate but also reveal the institutions behind them (e.g., Arnold Arboretum).
Access Restraint Open data can be misleading if context is missing (e.g., misinterpretations of the 2016 U.S. election polls).
Analysis Reflexivity Algorithms are not neutral—we must critically engage with them (e.g., Google’s biased autocomplete).
Optimization Contestation Data-driven decisions often ignore competing interests (e.g., Zillow optimizing the housing market while hiding its inequalities).
📌 A Five-Step Approach to Critical Data Practices

Loukissas proposes a methodology for engaging with data critically:

  1. Read: Examine the dataset for inconsistencies or unusual features.
  2. Inquire: Consult experts, data collectors, or subjects to understand the dataset’s background.
  3. Represent: Use visualizations to highlight patterns and biases.
  4. Unfold: Investigate how data are collected, processed, and normalized.
  5. Contextualize: Analyze who uses the data and what ethical concerns arise.

This approach treats data as an ethnographic inquiry, emphasizing critical engagement rather than passive consumption.

🔍 Conclusion: Data as a Social & Ethical Responsibility

Loukissas closes with a call to action: we must change how we engage with data. Instead of treating data as abstract, portable facts, we should see them as points of contact between people, institutions, and power structures.

He warns against digital universalism, arguing that data should be understood within their specific historical, institutional, and social contexts. Instead of prioritizing optimization and efficiency, we should focus on social justice, accountability, and transparency.

💡 Final Takeaway

“Do not mistake the availability of data as permission to remain at a distance.”

Loukissas urges us to engage with data deeply, ethically, and contextually—not just as raw information but as a socially embedded artifact requiring care and responsibility.

💬 Discussion Questions

  • How do data infrastructures reinforce social inequalities?
  • What would ethical open-data policies look like in practice?
  • Is it possible to create truly neutral datasets, or is all data inherently biased?

Was sind Daten?

TBD

Back to top