Being analysis-driven means looking beyond what a dataset says. Examine where the data came from, how it was gathered, what may be missing, and which incentives or changing conditions shaped it. section 10.19
Use this quick interrogation checklist whenever you receive data:
- What does the data measure? section 10.19
- Who made it, and who or what is represented? section 10.19
- Where and when was it gathered? section 10.19
- Why was it gathered, and how was it gathered? section 10.19
- Which biases, missing values, environmental changes, or incentives could steer the results? section 10.19 section 2.7
Bias is inevitable rather than automatically bad. Disclose suspected biases, sample as randomly as possible, and recognize that deliberately correcting representation also reflects modeling choices. section 2.7
