A curated list of readings about data science as a practice — how it actually gets done, where it goes wrong, and what it does to the people in the data. It is a work in progress.
For practical how-to material (Python, pandas, git, visualization tutorials, cheat
sheets), see the companion Resources repo.
New here? Start with What is the question?, Tidy Data, and Ten simple rules for responsible big data research.
A note on access. Most entries link to a freely readable copy. Entries tagged (UCSD access) are paywalled — sign in through the UCSD Library or the campus VPN to read them. Entries tagged (book) are worth requesting through the library if you don't want to buy them.
Papers and essays
- Foundational
- How Data Science Actually Gets Done
- Data and Data Organization
- Thinking Carefully With Data
- Ethics and Accountability
- Privacy, Anonymity, and Consent
- Computing Practices
- Visualization
- Communicating Results
- Machine Learning and Generative AI
Longer reads and ongoing sources
- 50 Years of Data Science — Donoho (JCGS, 2017). Traces where "data science" came from and argues for what the field would look like if it took its own science seriously; the standard starting point for the "is this just statistics?" debate. (UCSD access)
- Exploratory Data Analysis — Tukey (1977). The book that made looking hard at your data — before modeling it — a respectable activity; skim the early chapters for the mindset rather than the hand-computation methods. (book · free to borrow)
- Statistical Modeling: The Two Cultures — Breiman (Statistical Science, 2001). The essay that named the split between modeling data to explain it and modeling it to predict; published with commentary from statisticians who disagreed. (open access)
- How Data Science Workers Work with Data — Muller et al. (CHI, 2019). Interview study showing that practitioners don't "find" data so much as actively construct it through discovery, capture, curation, and design. (UCSD access)
- "Everyone wants to do the model work, not the data work" — Sambasivan et al. (CHI, 2021). How undervaluing data collection produces "data cascades" — small upstream problems that compound into downstream failures in deployed systems. (free author PDF)
- Tidy Data — Wickham (JSS, 2014). One variable per column, one observation per row, one table per unit of observation — and why every downstream step gets easier when you do that first. (open access)
- Data Organization in Spreadsheets — Broman & Woo (The American Statistician, 2018). Concrete rules for the spreadsheet stage that most data still passes through; the companion tutorial is free and equally useful. (UCSD access)
- Datasheets for Datasets — Gebru et al. (CACM, 2021). Proposes documenting every dataset's motivation, collection process, and intended uses; a useful checklist for describing your own project data. (open access)
- "Raw Data" Is an Oxymoron — Gitelman, ed. (MIT Press, 2013). Essay collection on the central point that data is always collected by someone, for some purpose, and never arrives raw; the introduction is freely readable. (book)
- What is the question? — Leek & Peng (Science, 2015). Two pages arguing that the most common error in data analysis is mistaking which of six question types you're actually asking. (UCSD access)
- Many analysts, one data set — Silberzahn et al. (AMPPS, 2018). Twenty-nine teams analyzed the same data on referee bias and got a wide spread of effect sizes; the clearest demonstration that analytic choices are results. (open access)
- The Garden of Forking Paths — Gelman & Loken (working paper, 2013). Why a researcher can inflate false-positive rates without ever running multiple tests — the choices you would have made on other data are enough. (open access)
- Why Most Published Research Findings Are False — Ioannidis (PLOS Medicine, 2005). The paper that kicked off the replication crisis: how power, bias, and the number of hypotheses tested determine whether a published claim is likely true. (open access)
- The ASA Statement on p-Values — Wasserstein & Lazar (The American Statistician, 2016). The American Statistical Association's six principles on what a p-value does and does not tell you. (open access)
- Measurement and Fairness — Jacobs & Wallach (FAccT, 2021). Recasts fairness disputes as measurement problems: the gap between the abstract thing you care about and the variable you actually recorded. (open access)
- Measurement Schmeasurement — Flake & Fried (AMPPS, 2020). Names the questionable measurement practices that quietly invalidate a study, and the questions to ask to avoid them. (open access)
- Calling Bullshit — Bergstrom & West (2020). Book and full course site — syllabus, case studies, and videos — on spotting misleading data claims without needing to re-do the analysis. (book · free course site)
- Ten simple rules for responsible big data research — Zook et al. (PLOS Comp Bio, 2017). Ten concrete commitments — data are people, privacy is contextual, guard against reidentification — that translate directly into project decisions. (open access)
- Dissecting racial bias in an algorithm used to manage the health of populations — Obermeyer et al. (Science, 2019). A widely deployed algorithm used health costs as a proxy for health need, and thereby understated the illness of Black patients; the canonical example of a biased target variable. (UCSD access)
- Gender Shades — Buolamwini & Gebru (FAT*, 2018). Audit of commercial gender classifiers showing error rates near zero for lighter-skinned men and above 30% for darker-skinned women — aggregate accuracy hides everything. (open access)
- Model Cards for Model Reporting — Mitchell et al. (FAT*, 2019). Companion to Datasheets: report model performance disaggregated by group, alongside intended use and known limitations. (open access)
- Data Feminism — D'Ignazio & Klein (MIT Press, 2020). How power decides what gets counted, by whom, and for whose benefit; the full text is free online. (free book)
- Weapons of Math Destruction — O'Neil (2016). Case studies of opaque scoring systems — teacher evaluation, credit, policing, college rankings — that scale harm while appearing objective. (book)
- Automating Inequality — Eubanks (2018). Ethnographic account of what automated eligibility, prediction, and ranking systems do to poor and working-class people who cannot opt out. (book)
- Robust De-anonymization of Large Sparse Datasets — Narayanan & Shmatikov (IEEE S&P, 2008). Re-identified Netflix Prize users from a little outside knowledge; the reason "we removed the names" is not a privacy plan. (open access)
- Differential Privacy: A Primer for a Non-Technical Audience — Wood et al. (2018). Plain-language explanation of what differential privacy guarantees, what it costs, and what it doesn't do. (open access)
- Differential Perspectives: Epistemic Disconnects Surrounding the US Census Bureau's Use of Differential Privacy — boyd & Sarathy (HDSR, 2022). Why the 2020 Census privacy fight was not really a technical disagreement — data users and privacy engineers hold incompatible ideas of what census data is. (open access)
- Good Enough Practices in Scientific Computing — Wilson et al. (PLOS Comp Bio, 2017). The realistic baseline: data management, project organization, version control, and collaboration habits any researcher can adopt this week. (open access)
- Best Practices for Scientific Computing — Wilson et al. (PLOS Biology, 2014). The more ambitious companion piece: write for people first, automate repetition, test, and don't optimize before you measure. (open access)
- Software development skills for data scientists — Causey (2015). The five habits — modular code, documentation, version control, testing, logging — that separate analysis scripts from code other people can use. (blog)
- PEP 8 — Style Guide for Python Code — van Rossum, Warsaw & Coghlan. The conventions that make Python readable to other Python programmers; skim it once, then let a linter enforce it. (docs)
- The Grammar of Graphics — Wilkinson (2nd ed., 2005). The formal system underneath
ggplot2,plotnine, and most modern plotting APIs: charts as compositions of data, scales, and geometry rather than a menu of chart types. (book · UCSD access) - Graphical Perception: Theory, Experimentation, and Application — Cleveland & McGill (JASA, 1984). The experiments that ranked how accurately people read position, length, angle, and area — the empirical reason to prefer bars over pies. (open access)
- The Science of Visual Data Communication: What Works — Franconeri, Padilla, Shah, Zacks & Hullman (PSPI, 2021). Book-length review of what the perception and cognition literature actually supports about chart design. (UCSD access)
- Same Stats, Different Graphs — Matejka & Fitzmaurice (CHI, 2017). The Datasaurus: datasets with identical summary statistics and wildly different shapes. (open access)
- Fundamentals of Data Visualization — Wilke (2019). Free, figure-heavy book on why some charts work and others mislead. (free book)
- How Charts Lie — Cairo (2019). How to read charts skeptically — misleading scales, dubious aggregation, and the ways a technically accurate figure still deceives. (book)
- Points of View: Storytelling — Krzywinski & Cairo (Nature Methods, 2013). One page on relating your data to the world your reader already knows. (UCSD access)
- Ten Simple Rules for Better Figures — Rougier, Droettboom & Bourne (PLOS Comp Bio, 2014). Practical rules for figures that carry an argument: know your audience, pick colormaps deliberately, don't mislead. (open access)
- Leakage and the reproducibility crisis in ML-based science — Kapoor & Narayanan (Patterns, 2023). Catalogs eight kinds of leakage found across 329 papers in 17 fields, with model info sheets for catching them before publication. (open access)
- Depths of learning — Buchanan (Nature Physics, 2018). Short column on adversarial examples: imperceptible changes to an input can flip a network's prediction, and detection schemes keep getting evaded. (UCSD access)
- On the Dangers of Stochastic Parrots — Bender, Gebru, McMillan-Major & Mitchell (FAccT, 2021). The costs of ever-larger language models — environmental, documentation, and the harms of training on unvetted web text. (open access)
- What Should Data Science Education Do with Large Language Models? — Tu, Zou, Su & Zhang (HDSR, 2024). Argues the data scientist's role shifts toward problem formulation and validation when the code comes cheap, and what that means for how the field is taught. (open access)
Books listed in the topic sections above aren't repeated here.
- Data Science from Scratch — Grus (2nd ed., 2019). Builds the tools from first principles instead of calling a library you don't understand yet; the harder book, and you'll grow a lot working through it. Free through your UCSD login via O'Reilly for Higher Education. (book · UCSD access)
- Python Data Science Handbook — VanderPlas (2nd ed., 2023). Short and to the point on NumPy,
pandas, matplotlib, and scikit-learn; text and code are both free. (free book) - Probability and Statistics for Data Science — Fernandez-Granda (NYU, 2017). Full lecture notes from NYU's Center for Data Science: probability through hypothesis testing and regression, with the math written out. (free notes)
- Better Explained — Intuition-first explainers for the math that underlies data science. (blog)
- Simply Statistics — Irizarry, Leek & Peng on statistics, data science, and how academic research works. (blog)
- Datawrapper Blog — Data-visualization writing with great worked examples; the "what to consider when" posts are excellent (and used in lecture). (blog)
- Statistical Modeling, Causal Inference, and Social Science — Gelman's long-running blog; where statistical claims in the news get taken apart. (blog)
- Simon Willison's Weblog — Close, practical tracking of what LLM tooling can and can't do, from someone who builds with it daily. (blog)
- Harvard Data Science Review — Open-access journal aimed at "everything data science and data science for everyone"; the most readable venue on this list. (open access)
- Journal of Open Source Software — Publishes short papers about research software, making tool-building citable work. (open access)
- Scientific Data — Nature journal for descriptions of reusable datasets; good models for documenting data. (open access)
- GigaScience — Publishes large-scale data, the code that produced it, and the analyses together. (open access)
- Distill — Archive of exceptionally clear interactive machine-learning explainers; the journal ran 2016–2021 and is on indefinite hiatus, but the back catalog is still some of the best explanatory writing in the field. (archive)
Suggestions welcome — especially from students. If a link is dead, a citation is wrong, or something you read for this course belongs here, open an issue or a pull request.
When adding an entry, please match the existing format:
- **[Title](https://example.com)** — Author(s) (*Venue*, Year). One or two lines on what it
argues and why someone in COGS 108 should read it. *(tag)*Tags in use: open access, UCSD access, book, free book, free notes, blog, docs, archive.