Battle Cards 5–8
Card 5 — Bias vs Poor Data Quality vs Invalid Comparison
| Problem | What makes result untrustworthy | Typical clue |
|---|---|---|
| Bias | Systematic inclination/distortion in collection, selection, sampling, analysis, context or presentation. | preferred conclusion, cherry-picking, skewed sample, group disparity |
| Poor Data Quality | Data is inaccurate, incomplete, stale, unreliable or unfit for intended use. | wrong/missing values, fitness failure |
| Invalid comparison / unclear definitions | Values/populations/terms are treated as comparable despite mismatched meaning/context. | same label hides different population or denominator |
Deciding distinction: systematic preference vs unreliable values vs mismatched meaning.
Examples: - collect only supporting evidence → Bias; - wrong eligibility date causes adverse decision → DQ with ethical consequences; - two accurate but differently defined populations compared → Invalid comparison.
Trap: calling every misleading result “bad data quality.”
Memory hook: Bias = slant. Quality = reliability/fitness. Comparison = meaning mismatch.
Card 6 — Timing vs Visualization vs Definitions
| Pattern | Distortion mechanism | First repair |
|---|---|---|
| Timing | Selected observation window. | Use appropriate window + disclose timing. |
| Misleading visualization | Scale, graphic choice, omitted points. | Honest scale/context/conventions. |
| Unclear definitions / invalid comparison | Different population, denominator, label or category meaning. | Define terms and compare like with like. |
Deciding distinction: all can mislead without fabricating numbers; identify how meaning was distorted.
- y-axis starts at 98 to magnify tiny change → Visualization.
- only last ten minutes shown because it looks favorable → Timing.
- full-time workers treated as identical to benefit recipients → Invalid comparison.
Trap: choosing “bias” merely because the output is misleading when a more specific §3.4 category fits.
Memory hook: Timing = when. Visualization = how shown. Definitions = what values mean.
Card 7 — Obfuscation / Redaction vs Actual Safety
Obfuscation/redaction: removes, masks or aggregates identifiers/sensitive detail to reduce exposure.
Actual ethical safety: requires context-aware protection across access, downstream combination, use, sensitivity and Governance.
Deciding distinction: obfuscation can reduce risk; it does not guarantee anonymity or permission to stop governing the data.
Scenario: names/emails removed, but location traces + public information re-identify people. Governance review is still necessary.
Trap: identifiers removed = permanently anonymous = zero ethical risk.
Memory hook: masking is a layer, not a magic shield.
Typical users: Security/privacy/Governance teams, Data Owners/Stewards and sharing/research teams.
Card 8 — Governance vs Legal Counsel vs Practitioner Responsibility
| Role | Chapter 2 responsibility |
|---|---|
| Data Governance | Set/oversee data-handling standards and policies; review handling practices and BI/analytics/Data Science decisions. |
| Legal counsel | Interpret applicable law, legal obligations and legal risk; coordinate with Governance as law changes. |
| Practitioner | Recognize day-to-day ethical risk, use judgment, follow standards and raise/escalate concerns. |
Deciding distinction: responsibility is distributed. Governance structures/oversees; Legal interprets law; practitioners still must act.
Scenario: a consequential customer-scoring proposal has possible fairness harm. Governance should review with Legal, while the project team must surface and explain the risks.
Trap: outsourcing ethics to a committee or lawyer and treating technical practitioners as responsibility-free implementers.
Memory hook: Governance sets the system; Legal interprets the law; practitioners still have a duty to act.
Master discrimination prompts
Complete from memory:
- Ethical differs from merely legal because…
- Privacy differs from Data Ethics because…
- Justice differs from Beneficence because…
- A control differs from a principle because…
- Bias differs from poor Data Quality because…
- Timing differs from visualization because…
- Obfuscation differs from actual safety because…
- Governance differs from legal counsel because…
Source anchors: Chapter 2 pp. 51–67.