History of Data

Before we touch a line of Python, it is worth asking a strange question: what is data, and why do humans collect it in the first place? The answer turns out to be older than writing itself, and understanding it will quietly make every later chapter make more sense.

The first datasets were bones

The oldest known dataset is the Ishango bone, found near a volcano in central Africa and dated to roughly 20,000 years ago. Carved along three rows of its surface are clusters of notches — groups of 11, 13, 17, 19 and others — that mathematicians still argue about today. Whether they recorded prime numbers, a lunar calendar, or simply a count of livestock, the impulse was the same: something happened repeatedly, and the carver wanted a durable record of it.

That impulse — to externalize counting so we can come back to it later — is the seed of every spreadsheet, every CSV file, and every SQL table you will ever see.

Clay tablets and the first ledgers

About 5,000 years ago, the Sumerians of Mesopotamia started pressing wedge-shaped marks into wet clay. Most of those tablets are not poetry or religious texts — they are receipts. Sheep delivered. Beer rationed. Land surveyed. Wages owed.

A typical Sumerian temple ledger looks shockingly modern. It has:

• A heading row (who, what, when).
• Repeated rows of transactions.
• Columns that align across rows so the scribe can scan down.
• A total at the bottom.

If you tilt your head, it is a spreadsheet. The technology has changed; the shape of the information has not.

Census, navigation, and the rise of statistics

For most of history, the largest datasets in the world were government records: censuses, tax rolls, military rosters, land registries. Rome counted citizens; Domesday Book counted English villages; Qing officials counted households across China.

Two big shifts happened in the 1600s and 1700s:

1. Probability theory was invented (Pascal, Fermat, then Bernoulli and Bayes), giving humans a way to reason about uncertain data.
2. Navigation demanded the first big precomputed tables — logarithms, trigonometric values, ephemerides — so that sailors could fix their position at sea.

By the 1800s, statistics was a discipline. Florence Nightingale used a polar area chart to convince Parliament that soldiers were dying from sanitation, not bullets. John Snow mapped cholera cases in London to a single contaminated water pump. These were not just numbers — they were data analyses with a conclusion, and they changed government policy.

Punch cards and the first machines

The 1890 US Census was a turning point. The country had grown so fast that hand-tabulating the previous census had taken eight years, and the next one was due before the previous one was finished. Herman Hollerith built an electromechanical tabulator that read punched cards — one per person — and the 1890 count was done in roughly one. The company he founded would later become IBM.

Punch cards were the dominant data medium for the next seventy years. Every payroll system, airline reservation system, and scientific simulation in the 1950s and 60s ate boxes of cards. The column on a card became the field in a record, which became the column of a database table, which became the column of a DataFrame. The vocabulary has barely changed.

The relational model

In 1970, an IBM researcher named Edgar F. Codd published a paper titled A Relational Model of Data for Large Shared Data Banks. It proposed a radical idea: instead of programmers hand-crafting the file layouts and traversal paths for each application, data should be stored as relations — sets of rows, each row a tuple of typed values — and queried with a mathematical algebra.

This is the model behind every SQL database in the world today, and it is also the model Pandas borrows from when it talks about joins, merges, groups, and aggregates. When we cover those operations later in the course, you will be using ideas that have been around for fifty-plus years and have survived because they work.

Why this story matters to you

A common mistake of new analysts is to think of "data" as something that lives inside a particular tool — an Excel file, a Pandas DataFrame, a Tableau workbook. It does not. Data is a record of something that happened in the world. The tools come and go (clay → cards → spreadsheets → DataFrames → cloud warehouses), but the work — turning recorded observations into useful decisions — is the same job humans have been doing for millennia.

Every time you load a CSV in this course, picture the Sumerian scribe with their reed stylus. The medium is different. The job is exactly the same.

A tiny historical experiment

Let us do our first piece of "data analysis" — on a dataset that looks a lot like a Sumerian ledger.

Take a moment to appreciate this. We just filtered by item, grouped by recipient, and aggregated by sum — three core operations that we will spend entire chapters on in Pandas. The scribe was doing the same calculation with a reed stylus on clay. You did it in twelve lines of Python. The underlying thought process — pick the rows you care about, slice by a category, total something up — is unchanged.

Check your understanding

QuestionSelect one

Which of the following is not an example of a tabular dataset in the historical sense used in this chapter?

A Sumerian clay tablet listing daily grain rations

A Roman census enumerating citizens by household

The 1890 US census tabulated on Hollerith punch cards

A handwritten love poem on parchment

QuestionSelect one

Why is the relational model (Codd, 1970) relevant to a course about Pandas?

Pandas is itself a relational database

Codd invented the CSV format

Pandas borrows the same conceptual vocabulary — tables, rows, columns, joins, group-bys — and the same mental model of operating on whole sets of rows at once

Pandas requires you to use SQL syntax internally