Data Archaeology

Today you start working with found data — a table you didn't make, full of words real people wrote for their own reasons. Before you load a single row in pandas, one idea has to land: that data is not raw. It's an artifact. Someone produced it, by making choices, and reading it well is closer to archaeology than to chemistry.

Reading traces of a life

An archaeologist never has the thing itself — the meal, the ritual, the household. They have what survived: a pottery shard, a posthole, a midden of bones. From those traces they reconstruct how people lived.

Pottery shards → ancient dining habits. Dataset rows → cultural patterns.

You're making the same move. Our 93 comments are a dig site — traces a community left while arguing about a law. You'll excavate patterns from them, but like any archaeologist you have to remember you're holding shards, not the meal: a partial record of something larger you'll never fully recover. (That's the noumena pipeline again — what reaches you is already several steps downstream of the lived thing.)

You found it; you didn't make it

There are two kinds of data, and this course runs almost entirely on the second.

• Created data — surveys you design, experiments you run. You control the questions.
• Found data — traces that already exist because culture left them: comments, reviews, lyrics, posts. Culture chose the questions, not you.

Found data is thrilling because it's real — nobody was performing for a researcher. It's treacherous for the same reason: it wasn't built for your question, its gaps aren't random, and you weren't there when it was made. "Found" never means "neutral."

"Raw data" is an oxymoron

We say "raw data" as if it arrived straight from nature, untouched. It never did. By the time a comment reaches your notebook, someone chose which platform to pull from, when, which field counted, and what got dropped — remember the fragments in our own corpus: the scraper broke at least one comment in two, so "Of Texas" arrived on its own, and our cleaning cut long comments apart at their line breaks. Every dataset is the survivors of a collection process, and the ones who didn't survive — the people who never posted, the comment that was deleted, the language the scraper couldn't parse — leave no row at all. The empty seats aren't in the file. Treat "raw" as a warning label, not a fact.

Hear, don't extract

Listen to the verbs data culture uses: mine, scrape, harvest, crawl, extract. Every one treats people like ore — raw material to be stripped out and processed. This course asks you to change the verb to hear. "Hearing" assumes the other end is a someone, with something to say and a claim on being received with care.

That one swap reframes the ethics of collection — which you'll practice in code today (robots.txt, fair use, attribution). The legal question is "am I allowed to take this?" The humanist question underneath is "am I listening, or am I extracting?" Both matter; only the second is about the people in the data.

Keep a data biography

So before the analysis, become the artifact's biographer. An archaeologist logs an object's provenance; you log your data's. Four things, written plainly in your notebook:

• Origin — who made this, when, and why?
• Journey — how was it collected, processed, shared?
• Transformations — what did you clean, filter, or drop?
• Limitations — what's missing, and who's missing?

This isn't busywork; it's credibility. A reader can only trust your reading if they can see where the data came from and what you did to it — and remember that a single row is layered (surface: who/what/when; cultural: the trend it hints at; societal: the values underneath). It's the habit the whole course keeps asking for: make your choices visible. You're interpreting human cultural behavior — handle it with care.

Going deeper (optional): Lisa Gitelman, ed., "Raw Data" Is an Oxymoron; Catherine D'Ignazio & Lauren Klein, Data Feminism; Johanna Drucker, "Humanities Approaches to Graphical Display" (data as capta, not "given").