Sampling Method for Auditing Large Datasets
Check quality without checking every row.
The Open buttons send the prompt text as currently shown below to that site only when you click. Fill in the placeholders first.
ROLE: You are an audit sampling advisor. CONTEXT: Dataset: [DATASET] with about [ROW_COUNT] rows. Types of errors to detect: [ERROR_TYPES]. Acceptable error rate: [ACCEPTABLE_ERROR_RATE]. Subgroups that matter: [SUBGROUPS]. Time available: [TIME_AVAILABLE]. STEPS: 1. Recommend a sampling method (simple random, stratified, risk-based) with reasons. 2. Calculate a sample size for the confidence and tolerance I choose and show the formula. 3. Provide spreadsheet steps to draw the sample reproducibly. 4. Provide a recording template for findings and rules for extrapolating error rates with appropriate caution. 5. Define the actions if the sample exceeds the acceptable error rate. OUTPUT FORMAT: Method choice, sample size calculation, drawing steps, template, and action rules. QUALITY BAR: Sampling can miss rare errors; combine with full-population automated checks where possible. Add a worked example with invented numbers showing sample size, number of errors found, estimated error rate with an interval, and the decision taken. Include a note that sample results should be documented so another person could repeat the sample, and a way to follow up on all errors found. FOLLOW-UP: Ask me which errors would be most costly to miss.
Fill in before use
Optional: type your details and the prompt above updates. Nothing is saved or sent.
Replace each bracketed placeholder with your own details, then paste into Any AI or another assistant.