Building the Analysis Story

Introduction

The code is not the product. The understanding it gives someone is. A good analysis notebook reads like a short report: it says what you wanted to know, how you found out, what you found and what to do. This lesson shows how to structure one, using the library project.

The structure

1. Title and summary: the answer in two or three sentences, for readers who stop there.
2. Question and audience: what decision does this support, and for whom?
3. Data: source, period, size, known problems, and the data dictionary.
4. Method: the main steps, in plain language, including cleaning decisions.
5. Results: the key numbers and charts, each with a sentence of interpretation.
6. Limits: what the analysis cannot tell us.
7. Recommendations and next steps.
8. Appendix: extra tables and the code.

Put the answer first. Readers are busy.

Example summary

Summary: Between January and March 2024, the library recorded 30 loans. Crime books were the most borrowed genre (11 loans) and were kept longest, at 19.7 days on average. Monthly loans rose from 7 to 13. These findings come from a small sample, so they should be tested with a longer period.

Every number can be traced to a query you ran: 11 Crime loans, the 19.7 average, and the monthly counts 7, 10 and 13.

Each chart needs a sentence

A chart without interpretation makes the reader do the work. Give every chart:

• A title that states the point, for example "Crime books were borrowed most"
• A one-line takeaway beneath it
• A note of the data source and period

Avoid charts that do not support a point. Fewer, clearer charts beat many.

Notebooks and scripts

A notebook mixes text, code and output, which is excellent for exploring and presenting. Habits that keep it trustworthy:

• Restart and run all cells from top to bottom before sharing, to be sure it works.
• Keep imports and settings at the top.
• Put reusable code into functions, and later into files.
• Do not leave secrets, personal data or big raw outputs in the notebook.
• Save the key outputs, such as charts and tables, as files.

Check before sharing

• Do the numbers match across summary, charts and tables?
• Can someone re-run it from raw data?
• Are all assumptions and limits stated?
• Is personal data removed or protected?
• Did someone else read it? A fresh reader finds unclear parts.
• Is the tone honest, not overselling small findings?

Recommendations

Good recommendations are specific, linked to evidence and realistic: "Stock more Crime titles, since they account for 37 percent of loans, and track loans per member by city next quarter". Avoid recommendations the data cannot support.

Try it yourself

Write a one-page report on the library data using the structure above, with a three-sentence summary, two charts with takeaways, a limits section and two recommendations. Ask a friend to read it and say what was unclear.

Summary

• Structure an analysis as a story: summary, question, data, method, results, limits and recommendations.
• Lead with the answer, and give every chart a title that states the point and a one-line takeaway.
• Re-run notebooks from scratch, check the numbers match, protect personal data and have someone read it.