When the Same Extra Time Produces a Different Job

The Census interaction estimates test the central question directly: does an additional quarter of potential UI work differently when the local labor market is less or more concentrated? The answer is asymmetric. Concentration changes the estimated earnings return more than it changes the length of nonemployment, although the earnings estimates themselves require careful qualification. This pattern supports a bounded bargaining-power interpretation: extra time may create room to search, but it does not create enough employer choice to make that room equally valuable everywhere.

Duration stays almost flat while the earnings return narrows

The Census paper measures local concentration with a quarterly payroll HHI on a zero-to-one scale. In its preferred interaction model, it estimates the effect of an additional quarter of UI at different points in that distribution. At HHI 0.15—the upper boundary of an unconcentrated market under the paper’s 2010 reference—the extension increases nonemployment duration by about 0.82 weeks. At HHI 0.25—the lower boundary of a highly concentrated market—it increases duration by about 0.83 weeks.

The difference is negligible. The interaction between the UI extension and HHI is neither economically nor statistically significant for nonemployment duration. This result does not say that concentration has no effect on job finding in general. It says that, within this policy variation and study design, the incremental effect of the extension on time without work is essentially flat across the observed concentration range.

The earnings pattern is different. At HHI 0.15, the estimated reemployment-earnings effect of an additional quarter is about 1.0 percent. At HHI 0.25, it is about 0.5 percent—approximately half as large. The interaction term is statistically significant in the preferred level-HHI specification, even though neither of those marginal earnings effects is individually statistically different from zero across the full range of concentration. The paper reports that the difference is driven mainly by the extremes of the HHI distribution.

Those qualifications belong in the headline, not in a footnote. The study does not establish that every worker in a highly concentrated market loses 0.5 percent relative to a worker in an unconcentrated market. It estimates a moderation pattern: the positive point estimate associated with additional UI time becomes less positive as concentration rises. The interaction is informative about how the policy effect varies, while the individual marginal estimates remain uncertain.

The result still changes the question. If concentration merely made workers wait longer, we would expect the duration response to be larger in more concentrated markets. Instead, the measured duration response is almost unchanged while the earnings return becomes smaller. That pattern is consistent with extra time having less power to improve the destination job when the worker has fewer credible employers to compare.

An interaction estimate should be read as a comparison of policy effects, not as a stand-alone wage penalty. The question is how much the outcome changes when the policy changes at one level of HHI relative to how much it changes when the same policy changes at another level.