Assessment Validation
Validate AI assessment accuracy against human evaluators
Q1 2026 Concurrent Validation — Power Systems Engineers
concurrent · Power Systems
AI-Human Correlation
87%
Accuracy
84%
False Positives
8%
False Negatives
6%
Findings: AI scores correlated strongly with human consensus (r=0.87). Minor over-scoring on communication competency for candidates with strong technical backgrounds.
Recommendations: Adjust communication weight by -0.5 for technical roles. Re-validate after Q3 2026.
Predictive Validity — Hired Cohort 2025
predictive · Mixed Engineering
AI-Human Correlation
79%
Accuracy
78%
False Positives
12%
False Negatives
9%
Findings: AI assessment scores predicted 6-month performance ratings with r=0.79. Candidates scoring above 75% had 85% retention rate vs 62% for those below.
Recommendations: Maintain current scoring model. Consider raising pass mark to 70% for safety-critical roles.
Content Validity — SCADA Engineer Question Bank
content · SCADA & ICT
AI-Human Correlation
82%
Accuracy
81%
False Positives
7%
False Negatives
11%
Findings: Question bank covers 92% of required competencies. Gap identified in cybersecurity assessment items for SCADA systems.
Recommendations: Add 15 cybersecurity-focused questions. Review and update SCADA vendor-specific items.
Construct Validity — Leadership Assessment
construct · Leadership
AI-Human Correlation
74%
Accuracy
72%
False Positives
14%
False Negatives
10%
Findings: Leadership construct showed moderate correlation. AI may over-reward assertiveness over collaborative leadership styles.
Recommendations: Rebalance leadership scoring weights. Add situational judgment scenarios for Zambian cultural context.