Confirmation Bias Detection in Performance Monitoring
Context
Performance monitoring often leans on human annotators reviewing and labelling data. When the same people review the same entities over time, confirmation bias can creep in: earlier assessments nudge the current one, and the evaluation slowly drifts.
What I built
- Designed and ran an experiment to detect confirmation bias in the annotation process
- Set up controlled comparisons to isolate the effect of prior exposure on current assessments
- Analyzed results to quantify the bias and recommend process changes