inclusive-experiment-analysis

Inclusive Experiment Analysis

Use this skill to make sure experiment design and readouts consider the range

of users affected by a product change. It focuses on subgroup impact,

accessibility, representation, data dimensions, and unintended harm.

Source Traceability

Primary source: Practical A/B Testing by Leemay Nassery. Guidance is

transformed and paraphrased from chapter 1 lines 719-912 and related subgroup

analysis context from lines 639-718. Metric and eligibility context comes from

chapter 2 lines 1564-1735.

Related Advanced Skills

• trustworthy-experiment-insights: use when subgroup findings may be

underpowered, false positives, or false negatives.

• experiment-verification-monitoring: use when inclusion risks depend on

assignment, exposure, device, geography, accessibility, or segment monitoring.

• adaptive-experimentation-strategy: use cautiously when contextual bandits or

personalization could create uneven user impact across groups.

Reference Routing

| Need | Read |

|------|------|

| Inclusive experiment concepts | references/core/knowledge.md |

| Design and analysis rules | references/core/rules.md |

| Segment examples | references/core/examples.md |

| Review workflow | workflows/review-inclusive-impact.md |

Workflow

1. Identify which user groups could experience the change differently.
2. Choose dimensions that are relevant, ethical, and available.
3. Check test/control balance for important dimensions when possible.
4. Include accessibility, bandwidth, device, privacy, geography, and usage-level

concerns where relevant.

1. Analyze subgroup outcomes without cherry-picking.
2. Recommend launch, mitigation, follow-up testing, or deeper research.

Output Format

# Inclusive Experiment Review

## Change Under Review
[What is changing and who may be affected.]

## User Dimensions
| Dimension | Why It Matters | Data Available? | Use In Analysis? |
|-----------|----------------|-----------------|------------------|

## Balance And Impact
| Segment | Control | Test | Result | Concern |
|---------|---------|------|--------|---------|

## Risks
- Accessibility:
- Device or bandwidth:
- Privacy or consent:
- Representation:
- Data limitations:

## Recommendation
[Ship | Ship with mitigation | Do not ship | Investigate] because [reason].

Quality Bar

• Do not use sensitive attributes casually; explain why a dimension is needed.
• Do not claim inclusive impact when the data lacks relevant representation.
• Do not average away harm to a meaningful subgroup.
• Pair quantitative subgroup analysis with qualitative or accessibility review

when metrics cannot capture the risk.