Please rotate your device to landscape mode to view the charts.

Productivity and Selection of Human Capital with Machine Learning

Journal: American Economic Review

Date: 20160501

Author: Chalfin, Aaron; Danieli, Oren; Hillis, Andrew; Jelveh, Zubin; Luca, Michael; Ludwig, Jens

Abstract:
Economists have become increasingly interested in studying the nature of production functions in social policy applications, with the goal of improving productivity. Traditionally models have assumed workers are homogenous inputs. However, in practice, substantial variability in productivity means the marginal productivity of labor depends substantially on which new workers are hired--which requires not an estimate of a causal effect, but rather a prediction. We demonstrate that there can be large social welfare gains from using machine learning tools to predict worker productivity, using data from two important applications - police hiring and teacher tenure decisions.

Link: Google Scholar

Key Findings

Police Misconduct Reduction

Using machine learning to screen police officers could reduce shootings by 4.81% compared to traditional hiring methods

Teacher Performance Impact

ML-based teacher selection improved student test scores by 0.0167σ in Math and 0.0111σ in ELA compared to current systems

Cost Effectiveness

ML-based teacher selection is 2-3 times more cost-effective than reducing class sizes by one-third

Impact on Police Misconduct Rates

  • ML-based selection reduced police shootings by 4.81%
  • Current police system selection showed increased shootings by 1.92%
  • ML showed improvements across all misconduct categories

Impact on Student Test Scores

  • ML selection showed greater improvements in both Math and ELA scores
  • Largest gain was in Math with 0.0167σ improvement
  • Benefits apply across entire student population

Cost-Effectiveness Comparison

  • ML teacher selection is 3x more cost-effective than class size reduction
  • Implementation requires only 6% increase in teacher wages
  • Provides broader systemic benefits compared to class size intervention

Contribution and Implications

  • Demonstrates significant potential for ML in improving public sector hiring decisions
  • Provides empirical evidence for cost-effective alternatives to traditional reform approaches
  • Highlights importance of using ML tools to inform specific, concrete decisions rather than general predictions

Data Sources

  • Police misconduct chart based on data from Figure 1 showing changes in police misconduct rates
  • Student test score improvements based on data reported in text showing gains in Math (0.0167σ) and ELA (0.0111σ)
  • Cost-effectiveness comparison derived from authors' calculations comparing ML implementation to Tennessee STAR experiment results