A Pay Floor Is Also a Bargaining System
The Italian study answers the opening question in a deliberately conditional way: a bargained pay floor can raise the top more than the bottom when the floor is part of a multi-rung wage ladder, when pass-through differs within and between agreements, and when enforcement and employment responses are weakest or most selective in lower-paid segments. This final synthesis reconnects those relationships so the reader can distinguish a causal short-run estimate from a descriptive trend, and a useful policy implication from an unsupported generalization.
The causal map: a ladder with filters
The shortest accurate answer is not that collective bargaining inherently favors high earners. It is that the contractual object being updated is broader than a bottom-only legal minimum. In the Italian agreements studied, different occupations, skill levels, seniority positions, and job classifications have their own contractual minima. The minima are coordinated enough that an agreement-level update can be identified, but they are not identical. Because they are fixed components of base pay, the update can reach workers already earning above a particular floor.
That design creates exposure across the wage ladder. In treated agreements, the median contractual minimum rises by about 2.19% on average, compared with about 1.92% at the lowest floor and 2.33% at the highest. The modestly larger movement at the top may contribute to the result, but it is not large enough to explain the full contrast in observed wage responses. The important fact is not simply where the schedule begins. It is that one negotiated event can move several rungs and then be transmitted unevenly into actual pay.
The study's preferred stacked two-period comparison makes the event more precise. Agreements with a median nominal-floor increase of at least 1% are compared with agreements whose median floor is unchanged over the same adjacent periods; cases with a smaller increase are left out. The estimate is a short-run within-system response to that contrast. Individual and agreement differences are absorbed through the fixed-effect structure, and the preceding unchanged period supplies a placebo check. Under the identifying assumptions, this supports a causal reduced-form interpretation of the update, not a general estimate of “the union effect.”
The observed wage-rate result is clear in its broad shape. Mean log full-time-equivalent daily wages rise by about 2.2%. The estimated response at the 10th percentile is 0.004 and not statistically distinguishable from zero; at the median it is 0.024, approximately 2.4%; and at the 90th percentile it is 0.045, approximately 4.5%, with more uncertainty. The 50–10 gap increases by about 0.020 log points and is more precisely estimated. The 90–10 gap is positive at about 0.041 log points but less precise, while the 90–50 gap is positive without being statistically distinguishable from zero.
Those numbers support an upper-tail tilt, but not every sentence that might be built from it. A near-zero 10th-percentile estimate is not proof that no low-paid worker received a raise. A 4.5% 90th-percentile estimate is not a 4.5% raise for every high-paid worker. The estimates describe movements in unconditional distributional thresholds relative to flat-floor agreements. Workers can change rank, and the people observed after an update need not be identical to those occupying the same percentile before it.
The decomposition then explains why the aggregate result cannot be reduced to one comparison between low-wage and high-wage agreements. When wages are recentered around each contract and event-panel average, the within-contract estimates remain ordered: about 0.012 at the 10th percentile, 0.023 at the median, and 0.033 at the 90th. The within-contract 90–10 effect falls to about 0.023 log points, but the asymmetry persists. Removing agreement-level wage differences removes part of the upper-tail pattern; it does not remove the evidence that positions within the same contractual ladder respond differently.
A second component appears across agreements. Contract-specific average treatment effects are positively related to the agreement's median wage, with a reported correlation of approximately 0.32.