When to Push Back on Requests
You'll learn how to identify ethically problematic analysis requests and develop strategies to decline or reframe work that could cause harm.
Not every request you receive as a data scientist deserves a "yes." Sometimes stakeholders ask for analyses that—intentionally or not—could violate privacy, amplify bias, enable discrimination, or mislead decision-makers. Knowing when and how to push back is a crucial professional skill.
Recognizing Red Flags
Watch for requests that:
- Target protected classes in ways that could enable discrimination (e.g., "find which demographic pays late most often" for pricing decisions)
- Encourage cherry-picking results you've already learned to avoid
- Violate consent boundaries or expand data use beyond its original purpose (secondary use and scope creep)
- Lack transparency about how results will be used
- Could have dual use implications you haven't discussed
- Ask you to omit uncertainty or present preliminary findings as definitive
Strategies for Pushing Back
Reframe, don't just refuse. Instead of "I can't do that," try:
- "I can answer a related question that avoids [specific ethical issue]..."
- "Before proceeding, we need to conduct a Data Protection Impact Assessment"
- "This analysis could have unintended consequences—let's apply red team thinking first"
- "The fairness-accuracy tradeoffs here need stakeholder input"
Invoke organizational values and legal obligations. Reference GDPR principles, your organization's conflicts of interest policy, or industry standards rather than personal objections.
Know your exit options. Sometimes you must escalate to leadership or, in extreme cases, consider responsible disclosure or changing roles. Maintaining your honest broker role means preserving your integrity even when it's uncomfortable.
Key Takeaway: Ethical data science requires the courage to say "no" or "not like that" when requests could cause harm—reframe problematic questions, invoke established principles, and prioritize long-term trust over short-term convenience.