Week 2 Notes - Course Introduction

Published

September 15, 2025

Key Concepts Learned

  • Algorithms are structured sets of rules used in both everyday life and government decision-making

  • Algorithmic decision-making in public policy can embed biases through data cleaning, coding, collection, and interpretation choices

  • Real-world failures (healthcare costs proxy, COMPAS, Dutch welfare fraud detection) highlight how “objective” systems still reflect human subjectivity

  • Census vs. ACS: differences in scope, frequency, and level of detail

  • Margins of error (MOE) are essential for evaluating the reliability of ACS estimates

  • Importance of reproducibility and transparency in policy data analysis

Coding Techniques

  • Learned to use get_acs() in the tidycensus package to directly access ACS data

  • Explored output structure: GEOID, NAME, variable, estimate, MOE

  • Data cleaning functions (str_remove(), str_extract(), str_replace()) to tidy up geographic names

  • Using case_when() to categorize data reliability based on MOE

  • Formatting professional tables with kable() for clarity and presentation

Questions & Challenges

  • Still clarifying how margins of error should be interpreted when comparing two different geographies—what’s the best way to handle overlapping error ranges?

  • Quarto integration with R code chunks for displaying ACS outputs smoothly

Connections to Policy

  • Algorithms used in government (e.g., predictive policing, healthcare prioritization) rely heavily on census and ACS data. If the underlying data is unreliable, it can perpetuate inequity

  • Reporting MOEs alongside estimates is a matter of professional credibility—decision-makers must see not just numbers but their uncertainty

  • Policy tools built on flawed proxies or biased data can unintentionally harm marginalized communities

Reflection

  • The most interesting takeaway was seeing how algorithmic bias emerges not from code alone but from human choices at each stage of data preparation

  • I’ll apply this by being cautious and transparent in my own analyses: always noting assumptions, reporting MOEs, and considering who might be excluded from the data

  • This week emphasized that ethical, reproducible methods aren’t optional—they’re essential to creating trustworthy analyses for real policy decisions