Data science turns observations into a defensible answer about a process, population, or decision. The analysis must preserve how records were collected and what each measure actually represents.
Choose a starting point
Start by defining the question, unit of analysis, and data contract. Continue through cleaning, exploration, statistical comparison, experiments, and communication of uncertainty. Projects join the steps in complete workflows.
- Data Science for Beginners
- Data science begins with a decision, population and clock
- Dataset grain and join cardinality: protect the unit of analysis
- Sampling frames and coverage error: who could enter the analysis?
- Column profiles and domain constraints before analysis
- Cohort entry and survivorship bias: count the cases that could fail
- SQL null semantics: count records without inventing outcomes
- Product event contracts: identity, deduplication and eligibility
- Quantity units and conversion contracts for mixed datasets
Common Mistakes
A chart or p-value cannot repair a biased sample or a denominator that changed halfway through collection. Use the linked audits to check assumptions before making a decision.