Perspectives on Data Sciences

Data science approaches vary widely depending on context, discipline, background, goals, and available tools.

My Journey

• Pure mathematics → Statistics → Data Science.
• Informal coding background driven by real-world research problems.
• Evolution from statistical consulting to interdisciplinary collaboration.
• Systems trajectory: Ecology → Agriculture → Health → Environment.

Key Concepts

• Data sciences bridge computational thinking, quantitative methods, and domain expertise.
• Focus on how communities use data to understand health, environment, and local decision-making.
• Part of the broader effort to Document Digital Tools.

Blog Perspectives: byandell.github.io

Insights from personal blog publications on the evolution of data science:

What is Data Science?

Defining the Discipline & Roles

• Core Triangle: Blends statistics, computer science, and applied math with deep domain inquiry.
• Data Literacy vs. Science: Literacy reads and communicates with data; data science develops methods to reveal new insights.
• Roles in Practice: Beyond titles (engineer vs. analyst vs. scientist), researchers are domain investigators tackling data mountains.

Historical Lineage & Thought Forms

• Biometry (Fisher, 1948): Active pursuit of biological knowledge by quantitative methods to understand variable phenomena.
• Data Analysis (Tukey, 1962): Moving beyond pure mathematical theory to exploratory analysis and computer-aided display.
• Iterative Investigation (Box, 1980): Statisticians as proactive partners in iterative deductive-inductive scientific inquiry.

Data Evolve

Data as a Process & Dynamic Verb

• Data as a Verb: Inspired by Lakota language (Jhon Goes-in-Center), viewing data not as static nouns but as living, evolving processes.
• Contextual Meaning: A single datum doesn’t change, but its context, metadata, and provenance evolve its value over time.
• Harmonization: Small data augmented by Big Data continually evolves the questions we are able to ask.

The Evolution of Software & Tools

• “Software Rots”: Code cannot remain static; it must continuously adapt as data, systems, and questions shift.
• Daryl Pregibon’s Progression:
  • Equation → Algorithm → Code/Function → Package → Standalone Interactive Widgets.
• Living Workflows: Tools must facilitate direct collaboration and active exploration for domain scientists.

Data Sciences Collaboratory

From Consulting to Deep Collaboration

• Transformational Partnerships: Evolving beyond one-off statistical advice to embedded, co-designed research programs.
• Team-Based Science: Modern complex challenges (health, environment, energy) require interdisciplinary expertise across boundaries.
• Early Engagement: Framing and design planning before encountering IT and computation bottlenecks.

Education, Training & Campus Infrastructure

• Workforce Retraining: Developing curriculum targeted at domain researchers swamped by high-throughput data.
• Co-Advising Model: Biometry MS model pairing students with mentors in both quantitative and biological fields.
• Campus Collaboratories: Open creative spaces linking campus research with high-throughput cyberinfrastructure (CHTC, ACI).

Data Science Across Liberal Arts

Data Literacy as a 21st Century Civic Need

• Civic Literacy: In an information age, liberal arts education requires data science to enable active, informed civic life.
• Storytelling with Data: Training a new generation to extract generalizable knowledge and tell meaningful, ethical stories.
• Democratizing Access: Expanding quantitative skills across humanities, journalism, sociology, and the arts.

Deep Grounding to Address Wicked Problems

• Context Dependence: Technical methods are incomplete without deep grounding in history, culture, and domain knowledge.
• Beyond Monolithic Degrees: Fostering versatile domain experts equipped with agile computational and communication skills.
• Tackling Complexities: Addressing “wicked problems” through collaborative, cross-disciplinary synthesis.

Strategic Planning: Data, Models & Statistics

Statistics at the Center of the Data Surge

• Sense-Making in the Information Age: IT advances generate massive data, but statistical thinking translates noise into meaning.
• Quantitative Core: Statistical theory, experimental design, and modeling remain the bedrock of rigorous inquiry.
• Methodological Gaps: Massive modern datasets require developing new statistical methods rather than relying on legacy tools.

George Box’s Departmental Vision

• Core & Joint Architecture: A strong central departmental core combined with specialists holding joint appointments across campus.
• Direct Domain Impact: Embedding statistical faculty in forefront research (virology, genetics, agriculture, medicine).
• Institutional Strategy: Continuous curriculum review and interdisciplinary alignment to meet emerging scientific needs.

Useful Data Science Quotes

Modeling, Skepticism & Design

• George E.P. Box: “All models are wrong, but some models are useful.” — Avoid falling blindly in love with models.
• Gertrude M. Cox: The greatest statistical contribution is often persuading the investigator to explain why they do an experiment.
• Douglas M. Bates: “Statistics is the science of variation.”

Judgment, Reproducibility & Humility

• John M. Barry: Scientific progress relies on judgment to distinguish reproducible, expandable findings from tortured data.
• Karl W. Broman: High-dimensional biology requires tools enabling domain scientists to carry out their own investigations.
• William Cowper: “Knowledge is proud that he has learned so much, Wisdom is humble that he knows no more.”

Significance vs. Importance

Scatter Plots & Relational Visualization

Scatter (X-Y) plots investigate relationships among measurements, augmented with color, symbol shape, trend lines, and facets.

scattyr repo plots in R (ggplot2) and Python (plotnine) with color, symbol shape, and facets to reveal data patterns.

Analytical Pitfalls to Avoid

• Spurious Association: Confounding variables causing apparent relationships.
• Multiple Testing: Uncorrected testing across vast arrays of features guarantees false discoveries.
• \(p\)-Hacking: Data dredging or selective reporting to reach \(p < 0.05\).

Spatial & Temporal Autocorrelation

Complex environmental and biological systems present unique analytical challenges:

Cumulative Trajectories

Balancing Open Data & Data Sovereignty

There is an ongoing tension between the open data movement and data sovereignty—the inherent right of Indigenous communities and local groups to govern data collected from their people, lands, and resources.

Indigenous Data Governance Frameworks

Key initiatives and frameworks shaping Indigenous data science:

Understanding Big Data: The 3 Vs, 6 Vs, and 3 Us

Big data refers to datasets exceeding the processing capacity of traditional database systems and analysis tools.

Open Data Repositories & Resources

Platforms and repositories supporting modern data science workflows:

Summary & Learning Path

byandell.github.io/Documentation