Using Language Models to Understand Wage Premia

Explore a Stanford HAI seminar where postdoctoral fellow Sarah Bana presents her research on using natural language processing to predict salaries from job posting text. Delve into the methodology of creating a model that accurately estimates wage premia for various job characteristics. Learn about the application of NLP techniques to turn job postings into context-dependent vectors, facilitating a modern approach to hedonic regression. Discover insights on the relationship between job posting text and salary predictions, including the significance of specific certifications and skills. Gain understanding of the model's performance, structure, and the implications of text injection experiments. Compare the findings with data from the Current Population Survey and examine the broader implications for understanding labor market dynamics through computational methods.

Intro
Job Postings Give Us Insight Into Workers' Skills and Activit
Do These Roles Pay Different Salaries? How Different?
Growing Set of Tools Available to Researchers
NLP Turns Words into Context-Dependent Vectors
Elements Create a 21st Century Version of Hedonic Regress
This Paper: Use NLP to Predict Salaries
Two Sources of Job Posting Data
Some Job Board User Interfaces Ask Recruiters To Input SE
How Selected are the Posted Salaries? Discussion and Comparison to the Current Population Survey (CPS)
Simple Regression for Salary Prediction
This Approach Facilitates Turning a Posting into a Matrix Each Token (Word) is Represented by a Vector
Model Performance
Model Structure
Text Injection Experiments
Example: International Institute of Business Analysis - Agile Certification (IIBA-AAC)
The Text of the Posting Matters!