Week 5: Analyzing AI Exposure Through Linear Regression
Hey everyone, welcome back to Week 5 of my senior project! This week, I’ve been running regressions in RStudio to examine which demographics are most vulnerable to AI-induced job displacement. As I mentioned in previous posts, we can analyze this by looking at the relationship between demographic factors (such as education level) and AI exposure using different types of regressions—such as linear and polynomial.
In this post, I’ll walk you through my findings using linear regression, and next week, I’ll compare these results to those from polynomial regression to see if any conclusions change.
Overview of Data and Methodology
To quantify AI exposure, I used the AI Occupational Exposure Scores provided by the U.S. Department of Treasury (Table B.3: Complete List of Fields by AI Exposure). These scores range from 0 to 1, with higher values indicating greater exposure to AI.
For demographic and occupational data, I pulled information from the IPUMS USA database, which compiles American Census Survey data. Using R, I created the variable AIExposure, assigning each occupation an AI Exposure Score. Then, I ran linear regressions for different demographic variables.
Regression Results
Full regression results available here
Race and AI Exposure
The first question I explored: Which racial groups are most exposed to AI?
In the IPUMS dataset, the RACHSING variable categorizes race in the U.S. as follows:
- White
- African American
- American Indian/Alaskan Native
- Asian/Pacific Islander
- Hispanic/Latino
Key Takeaways:
- African Americans, American Indians, and Hispanics tend to hold jobs that are significantly less exposed to AI than White workers (as indicated by the negative coefficients).
- Asian workers, however, are more exposed to AI than White workers, as reflected by the positive coefficient.
- The statistically significant t-values (all greater than 2) confirm the reliability of these results.
Conclusion: Asian workers are the most exposed to AI, followed by White workers, then African Americans, American Indians, and Hispanics.
Age and AI Exposure
Next, I analyzed whether older workers are more or less exposed to AI-driven job displacement.
Key Takeaways:
- The positive coefficient for age suggests that older workers tend to hold jobs with higher AI exposure.
- The high t-value confirms that the result is statistically significant.
Sex and AI Exposure
For this analysis, the SEX variable assigns:
- 1 for male workers
- 2 for female workers
Key takeaways:
- The positive coefficient for sex means that female workers are more exposed to AI than male workers.
- This aligns with previous studies predicting that women’s overrepresentation in administrative and customer service roles makes them more vulnerable to AI-driven job displacement.
Education and AI Exposure
I then explored the relationship between education level and AI exposure.
Key Takeaways:
- The positive coefficient indicates that higher education levels correlate with increased AI exposure.
- This aligns with studies showing that AI is more likely to replace white-collar jobs (which often require a college degree) than manual labor jobs.
Residency in the U.S. (for immigrants) and AI Exposure
For immigrants, I examined whether years spent in the U.S. (YRSUSA1) affected AI exposure.
Key Takeaways:
- The negative coefficient suggests that the longer an immigrant has lived in the U.S., the less exposed they are to AI.
Income and AI Exposure
Finally, I looked at whether income level correlates with AI exposure.
Key Takeaways:
- The positive coefficient for income indicates that higher-income workers are more exposed to AI.
- This supports the theory that high-paying white-collar jobs (e.g., finance, legal, tech) are more susceptible to AI than lower-wage manual labor jobs.
Final Thoughts
These findings highlight that AI exposure is not evenly distributed. Groups with higher exposure—such as women, Asians, older workers, and those with higher education—may face greater disruption in the workforce. Understanding these patterns is crucial for policymakers and businesses as they plan for an AI-driven job market.
What’s Next?
Next week, I’ll compare these linear regression results with polynomial regression to see if the patterns hold up or change. Thanks for following along—I’m excited to share more insights with you soon!
Let me know if you have any questions or thoughts in the comments!
Comments:
All viewpoints are welcome but profane, threatening, disrespectful, or harassing comments will not be tolerated and are subject to moderation up to, and including, full deletion.