Why the Upper Tail Moves More

The Italian evidence shows an upward gradient in wage-rate responses, but that pattern could arise in two different ways. Higher-paid collective agreements might simply receive larger effective increases than lower-paid agreements. Or workers on higher rungs inside the same agreement might respond more strongly than workers on lower rungs. This chapter separates those possibilities by looking first within agreements and then across them. The result is a two-part explanation, not a single mechanical rule.

The gradient survives inside the agreement

The simplest challenge to the headline is that high-wage agreements are different from low-wage agreements in every respect, including the size of their contractual updates. If the aggregate 90th-percentile response is larger only because workers in generally high-paying agreements received bigger increases, the upper-tail pattern might disappear once those agreement-level wage differences are removed.

The researchers address this by recentering each worker's wage around the average wage for that contract and event panel. The question then becomes: within an agreement that is experiencing an update, how does the distribution of workers' wages relative to that agreement-level average respond? This does not compare a low-paid industry with a high-paid industry simply because their wage levels differ. It asks whether the internal ladder itself shows an uneven response.

It does. In the recentered analysis, the estimated change at the 10th percentile of the within-contract distribution is 0.012, with a standard error of 0.004. The estimate at the median is 0.023, with a standard error of 0.004, and the estimate at the 90th percentile is 0.033, with a standard error of 0.013. The upward ordering is less dramatic than in the unconditional distribution, but it remains: lower-paid positions inside the same agreement respond less, while the upper part responds more.

The associated 90–10 gap effect falls from about 0.041 log points in the unconditional analysis to about 0.023 log points after recentering. Relative to the corresponding within-contract gap, that is about a 2.7% increase. The 50–10 gap effect falls to about 0.013 log points, or roughly 3.6% of that gap. The 90–50 estimate remains positive, about 0.010 log points, but is not statistically distinguishable from zero. Re-centering therefore removes a meaningful portion of the aggregate inequality-enhancing pattern without removing all of it.

That attenuation is informative. It says that differences between agreements matter. Workers in agreements with different median wage levels contribute to the overall distributional result, and removing those level differences reduces the apparent tilt. But the remaining within-agreement gradient rules out the idea that the entire result is just a comparison between a high-paying contract and a low-paying contract. Something about responsiveness along the internal wage ladder also differs.

The interpretation is not that a lower-grade worker is necessarily denied a raise while a higher-grade worker is guaranteed one. The estimates are distributional and averaged across many contracts and workers. They show a difference in the movement of percentile thresholds relative to the agreement-level average.