This paper develops a three-step method to estimate “government-job skills”: (1) extract noisy individual skill signals from wages in private-sector jobs comparable to government roles, adjusting for experience and a time-varying skill price; (2) use machine learning (GBM preferred) to learn the mapping from rich observable skill-related covariates to the noisy skill signal; (3) predict out-of-sample skills for government workers and all other workers. Applying the method to IFLS (1988–2014) shows (i) government workers are positively selected but their relative skills decline vs. the private sector over time (driven by the private sector’s growing right tail), (ii) government wages are somewhat less informative about skills but—after accounting for life-cycle profiles—still predictive, and (iii) a conditional government wage premium of ≈43% (partly explained by a large private-sector gender wage gap).
| Method | Out-of-fold R2 |
|---|---|
| GBM | 0.11 |
| LASSO | 0.10 |
| RF | 0.10 |
| Edu OLS | 0.088 |
| Basis | -0.06 (overfit) |
元数据与概览(引言与报告概览)
逐节深度解读
图表与数据深度解读(逐图逐表)
估值/预测方法学的细节解读(方法学、模型与机器学习)
风险因素评估(作者识别出的风险与影响)
批判性视角与方法局限(审慎评价)
结论性综合(要点与政策含义)
附:关键图像(本文中分析所参照的主要图表)
總結性的評價(最後一句)
2604.15819.pdf