Fairly biased

Health equity through a journalism and data science lens


Welcome to West Athens

Fun fact: USC isn’t the southern border of Los Angeles

Grocery/pharmacy run

Where’s the pharmacy?

No car?

All routes point to our past

L.A. air quality vs. L.A. redlining

L.A. air quality vs. L.A. COVID vulnerability

How do you find your inequity topic?

  • Compare your ‘normal’ with the ‘normal’ of others

  • Switch up your route to work or the store

  • Follow up on something surprising you heard

  • Go with your natural curiosity

    • Passionate about a topic? Get exploring

    • Know a personal blindspot? Challenge yourself

Let’s talk data science

There’s bias around every corner

  • Antiquated attitudes

  • Non-representative sampling

  • Exclusionary methodologies

  • Uninterpretable AI models

“Even if AI systems are designed by unbiased coders striving for neutrality, those systems derive data from and exist within a medical system that has its own antiminority culture; those views are embedded in the patters that AI learns.”

Race is more than a data column

Race is more than a data column

  • Even when race is removed from a dataset, statistical models can reliably predict a patient’s race from the way the physician wrote about them. (Keeling et al)

  • If bias is already creeping through, a statistical model will only exacerbate it.

Questions to explore