Your first data analysis with AI

I livestreamed while figuring out how to teach a lesson called "Your first data analysis with AI," using arrests data from the Deportation Data Project. I had Claude describe the data, then checked its claims myself, and ended up settling on having it write a Quarto document so I could run and review the code.

You'll learn:

• How to ask AI to describe a dataset first (what a row represents, the time period, oddities) without writing analysis code yet
• A way to check AI's claims about your data, like the date range and the number of distinct IDs, by running quick checks yourself
• How to have AI write a Quarto exploratory data analysis document so you can run and review the code behind each output
• How to ask AI to add plain-English comments to each line of code so it is easier to review
• Why giving AI the codebook or documentation can change its recommendation, such as switching to a different data file

Tools mentioned: Claude, Claude Code, Positron, Quarto, ggplot2, Air, arrow

Key moments:

• 01:53 Claude-generated livestream run sheet
• 02:48 Introducing the arrests data and parquet import
• 08:26 Asking Claude to describe the data before analysis
• 15:51 Checking Claude's claims in R
• 17:41 Asking about duplicate drop rows and the codebook
• 21:11 Trust but verify approach explained
• 22:08 Dictating a request for a Quarto EDA document
• 29:38 Reviewing the monthly arrests code and chart
• 33:37 Asking for code to answer a question, not just a yes or no
• 38:30 Testing a question the data can't answer
• 44:34 Adding plain-English comments to the code
• 46:46 Deciding to frame the lesson around the Quarto workflow
• 47:47 Next stream preview on data security with AI