A framework for asking better questions of data

Making Data Sing ♫

Type → Pair → Move → Shape → Sharpen

The problem

Most analysis stops at group-by

What we usually do
group by name
sum the count
sort
→ “The most common name is James.”
True. But nobody leans forward.

Good questions don't come from knowing more functions. They come from seeing what kinds of columns you have and which ones appear together.

The model

Five notes, climbing

1Type

Name each column's job

Five jobs. Bottom row: the baby-names dataset, one column per job.

2Pair

Every pair of types asks a question

3Move

Nine moves, three families

3Moves in action · rescale + spotlight

Fad or classic?

Jennifer rose and fell within one generation. Elizabeth never left.

Who × When + peak, measured as a share (per 1,000 girls), so baby booms don't distort it.

3Moves in action · contrast + spotlight

Distinctive, not just common

Over-index finds what makes a place different. Reversal finds where the pattern breaks.

3Moves in action · flat vs. singing

“Parents are spreading out”

Flat: “Michael is the most common boys' name.” Singing: rank + share + concentration.

4Shape (advanced)

Knock on three doors

5Sharpen

Grow a seed into a full question

🧪 Test: could two classmates compute the same answer? If not, it's still a seed.

Your turn

For any new dataset

Typeevery column
Pairtwo outlined cells → two questions
Movefrom two different families
Shapeorder? pair? nest?
Sharpenthree seeds, expectation first

The gap between what you expected and what you found is where insight lives.

Site