Training opportunities in monopsonistic labour markets

ABSTRACT

This paper studies the empirical relationship between the extent of monopsonistic power observed in occupational labour markets and the training opportunities available to workers in those markets; using data from the American National Longitudinal Survey of Youth of 1979. The results reveal a positive and significant association between monopsony power and training availability. The estimated association is found to be stronger for individuals with a college degree, with longer tenure in their jobs, and higher wages. These results are consistent across several econometric specifications that control for time, occupation, and individual fixed effects.

KEYWORDS:

• Monopsony
• training

JEL CLASSIFICATION:

• J42
• J24

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

¹ In the limit, as $\frac{\lambda }{\delta }$ approaches infinity, the distribution of wages F(w) collapses to the perfectly competitive case and all workers get paid their marginal product.

² In the model, one can find the closed form solution for this key fraction to equal $\left[\frac{\frac{\lambda }{\delta }}{1+\frac{\lambda }{\delta }}\right]\left[\frac{1}{ln\left(1+\frac{\lambda }{\delta }\right)}\right]$. The figure plots values for the inverse of that number, for the different parameter values.

³ These answers convey the individual’s recollection of the employers’ policy, which is not necessarily an accurate account of the actual policy. Measurement error, however, would tend to bias our estimates towards zero.

⁴ See U.S. Census Bureau (2003) for a history of the relevant Census occupational classification lists up to that year.

⁵ I preferred to use the NLSY79 over the CPS because the CPS does not contain continuous records of employment which are key for generating our approximation.

⁶ Using the replication file that accompanies this paper, interested readers may verify similar results are obtained when the variables are entered in non-logarithmic form. The log-transformation, however, facilitates the convergence of the maximum likelihood expectation maximization algorithms when conducting logit regressions.