Data Bias: Modalities, Detection, and Mitigation
Prof. Dr. David B. Blumenthal · 6 concepts · 10 questions
Concept 1 / 6
What Is Data Bias? Definition, Consequences, and the Data-vs-Algorithmic Distinction
Data bias: a systematic distortion in the data. It arises from how data is collected, measured, labeled, or represented, and often reflects and perpetuates existing social, historical, or institutional inequities. It is typically unintentional and hard to detect without critical analysis.
Core principle: Systematic distortion in the data (data bias) can lead to systematic distortion in the results.
Bias vs. variance (archery-target analogy):
- Bias = how far the cluster of shots sits from the bullseye (a systematic offset).
- Variance = how scattered the shots are around their own center (random spread).
Real-world consequences:
| Domain | Example consequence |
|---|---|
| Healthcare | Diagnostic algorithms misdiagnose certain populations due to non-representative training data |
| Criminal justice | Predictive policing reinforces racial profiling and over-policing in marginalized communities |
| Finance | Credit scoring denies loans via proxies correlated with socio-economic background (e.g., zip code) |
| Hiring | Resume screening perpetuates socio-economic, ethnic, or gender biases in historical hiring data |
Bottom line: biased data is not just a technical issue — it is a societal and ethical challenge with tangible impacts on people's lives.
Data bias vs. algorithmic bias:
| Data bias | Algorithmic bias | |
|---|---|---|
| Origin | The data collection process | How algorithms process data |
| Nature | Reflects existing prejudices / systemic inequalities | Can amplify or introduce new biases, even with unbiased data |
| Examples | Biased survey responses, underrepresented groups | Biased recommenders, unfair credit-scoring models |
In ML, where algorithms learn from data, data bias often leads to algorithmic bias.
Three modalities of data bias (the structure of this lecture): structural bias, measurement bias, and representational bias.