Prepare Data for Exploration
Core concepts
Preparation is about finding data that is reliable and appropriate. Classify sources by party: first-party data is collected directly by you, second-party data is collected by another organization and shared with you (often a partner), and third-party data comes from an outside aggregator or provider.
Judge credibility with ROCCC: Reliable, Original, Comprehensive, Current, and Cited. A dataset that fails any letter should be flagged before analysis.
Know common file formats - CSV (plain text, comma-separated), JSON (key-value pairs, common in web APIs), XML (nested tags) - and where data lives: spreadsheets, relational databases (tables linked by primary and foreign keys), and data warehouses. Metadata describes data (who created it, when, how), and data governance sets the policies controlling access, privacy, and security. Open data is freely available, but free does not mean credible.
Understand sampling: random (every member has an equal chance), systematic (every nth member), stratified (random samples taken within each group), and convenience (whoever is easiest to reach - the weakest option). Match each to sampling bias (a sample that is not representative), observer bias, interpretation bias, and confirmation bias.
SQL fundamentals appear here: SELECT, WHERE, ORDER BY, AS for aliasing, and LIKE with % (any number of characters) and _ (exactly one character).
Traps
Questions test party labels, ROCCC letters, and sampling names. Convenience sampling is never random, even when results look reasonable. Also remember % and _ are not interchangeable in LIKE patterns.
Rule of thumb
Before trusting any table, run ROCCC plus one bias check: who is missing from this dataset, and would their absence change the answer?