A reliable data-analysis workflow

Good analysis starts before the chart or model. First make sure the question, data and definitions agree.

1. Define the question

Specify the population, time period and outcome. “What were sales?” is incomplete without those definitions.

2. Inspect the dataset

Check row count, columns, types, missing values, duplicates and representative records. Look for impossible values.

3. Define the grain

Ask what one row represents. Errors often occur when tables with different grains are combined without accounting for duplication.

4. Clean with traceable rules

Document type conversions, missing-value treatment, filters and deduplication. Avoid silently deleting unusual observations.

5. Analyze and compare

Separate counts from rates and totals from averages. Choose summaries that match the question.

6. Communicate uncertainty

Explain assumptions, limitations and data gaps. A precise-looking number is not automatically a precise conclusion.

Validation checklist

• Reconcile totals where possible.
• Sample records after transformations.
• Check filters for unintended exclusions.
• Repeat key calculations independently.
• Keep raw data separate from cleaned data.