Research Decisions Log — Template
Cleaning and variable-construction choices are research decisions. This log makes them visible — including the ones an AI agent made on your behalf: when an agent writes your cleaning code, audit it for imputations, dropped observations, and recodes, and log each one below as if you had made it yourself. (You did.)
Dataset Information
- Dataset: _______________________________________________
- Date: _______________________________________________
- Analyst: _______________________________________________
- Original sample size: _______________________________________________
Instructions
For each data quality issue you identify, document your decision below. There are no universally “correct” answers to cleaning decisions — but every decision must be justified and its consequences understood.
Before filling this out, explore the data using summarize, tabulate, codebook, or equivalent commands. Ask AI to help you flag potential issues, but make all final decisions yourself.
Cleaning Decisions
| Issue # | What I Found | Decision | Justification | Alternative Considered | Sample Size Impact |
|---|---|---|---|---|---|
| 1 | |||||
| 2 | |||||
| 3 | |||||
| 4 | |||||
| 5 | |||||
| 6 | |||||
| 7 | |||||
| 8 |
Column Guide
- What I Found: Describe the issue precisely (e.g., “3 respondents with earnings > 500,000; median earnings is 17,000”)
- Decision: What you did (e.g., “Winsorized at 99th percentile”, “Dropped observation”, “Recoded to missing”, “Kept as-is”)
- Justification: Why this is the right call for your analysis (e.g., “Likely data entry error based on context”, “Value is plausible for a business owner”)
- Alternative Considered: What else you could have done and why you didn’t
- Sample Size Impact: How many observations are affected (e.g., “3 observations recoded”, “Dropped 12 obs, N goes from 500 to 488”)
Final Sample Summary
- Final sample size: _______________________________________________
- Total observations dropped: _______________________________________________
- Total observations with recoded values: _______________________________________________
- Percentage of sample affected by any cleaning decision: _______________________________________________
Every cleaning decision is a research decision. If you can’t justify it, don’t do it.