When a Pay Floor Raises the Top First
This series studies a specific question about Italian collective bargaining agreements: what happens to the wage structure when an agreement raises its contractual pay floors? The focus is not whether unions are generally good or bad for wages. It is how an update inside a representative industry-level agreement travels through the wage ladder, and what that means for daily wage rates, monthly earnings, compliance, and employment. The practical judgment is equally specific: before assuming that a pay-floor increase must mainly help the lowest-paid workers, we need to understand which kind of floor is being raised and how firms and workers respond.
The central result is counterintuitive. In the Italian study examined here, a roughly 2.2% increase in the median contractual minimum raises average full-time-equivalent daily wages by about 2.2%. But the estimated response is close to zero at the 10th percentile, about 2.4% at the median, and about 4.5% at the 90th percentile. A floor rises, yet the strongest measured wage-rate response appears higher up the distribution. That does not mean that low-paid workers receive no benefit, or that the top is deliberately targeted. It means that a collectively bargained floor is not the same institution as a single statutory minimum that primarily constrains the bottom.
The surprising question is about a wage distribution
The phrase “pay floor” invites a simple picture: place a line below which no wage may fall, lift the line, and the workers nearest to it should experience the largest change. That picture is often useful for thinking about a legal minimum wage. It is incomplete for a collective agreement.
A collective bargaining agreement usually contains a schedule of minima. Different occupations, skill levels, seniority grades, and job classifications can each have their own contractual floor. The schedule is a wage ladder, not one horizontal line. An update negotiated for the agreement can therefore move several rungs at once. It may raise the fixed base-pay component for a worker whose actual wage is already above the applicable minimum, not merely for someone pressed directly against the bottom rung.
That design changes the question. Instead of asking only, “How many workers are lifted over a legal threshold?”, we must ask, “How does an agreement-wide update pass through an internal wage structure?” The answer can differ across grades and across agreements. It can also differ between a wage rate—the amount associated with a day of full-time-equivalent work—and monthly earnings, which depend on both the rate and how many days are worked.
This distinction matters because an apparently weak lower-tail wage response can have more than one meaning. It may reflect less effective implementation of the contractual floor among low-paid workers. It may reflect a stronger employment response in low-wage jobs, changing who remains in the observed wage distribution. Or it may coexist with higher monthly earnings for workers who remain employed and work more days. A serious explanation has to keep these possibilities separate rather than compressing them into the statement that “the minimum wage went up.”
The study's subject is the large, representative part of Italy's private non-agricultural labor market where national industry-level agreements are the relevant wage-setting institution. In this setting, agreements negotiated by unions and employers' associations are automatically extended to non-signatory parties, so contractual coverage can be close to universal within the relevant segment even though union membership itself is not. That is why a membership statistic cannot answer the study's question. The object being changed is the contractual pay schedule, and the object being measured is the distribution of pay among workers exposed to that schedule.
A contractual floor is a ladder, not a line
The contrast with a statutory minimum is not a claim that one institution always compresses inequality and the other always widens it. It is a claim about where the institution places the moving parts. A statutory minimum normally establishes a common legal lower bound. A CBA can establish a coordinated set of lower bounds for different jobs, and those bounds may be revised together while retaining differences between grades.
The Italian evidence makes the ladder visible. The average actual wage in the study is about 18% higher than the median contractual minimum. Within agreements, the distance between the average highest and lowest contractual floors is close to 60%. These figures do not mean that every worker is paid exactly according to a single schedule; they show that the schedule covers a substantial internal wage structure. The floors are not isolated legal markers at the very bottom of the labor market.
At the same time, the ladder is coordinated. When a contract's median floor rises, the floors for different job titles tend to move in the same direction. Correlations of positive floor growth across the minimum, median, and maximum positions are roughly 0.84 to 0.90. That makes the median floor a useful summary of the timing and direction of an agreement-level update. It does not make the update identical for every grade. Over the study period, annualized nominal growth was about 1.8% at the bottom of the contractual-floor distribution, 2.0% at the median, and 2.2% at the top.
This combination—coordination across grades and variation in the size and timing of updates—is the key institutional fact. Coordination gives researchers a recognizable agreement-level event. Variation makes it possible to compare agreements whose floors rose with agreements whose floors were unchanged over the same adjacent period. Variation across grades also means that the contractual ladder can transmit an update beyond the workers closest to its lowest point.
The evidence comes from a large administrative panel. The researchers link Italian Social Security Institute records on earnings, days worked, and job characteristics to a reconstructed archive of contractual minima covering about 160 national agreements. They focus on May and November, months chosen because they tend to have fewer vacation interruptions and are closer to ordinary base pay than months with common one-off payments. The resulting analysis covers 2006–2017 and contains more than 52 million worker-semester observations and more than 5 million individuals.
Two outcomes must be read differently. Full-time-equivalent daily wages are intended to capture the wage rate more directly. Monthly earnings combine the rate with the amount of labor supplied during the month. The paper's descriptive trends show why that matters: daily-wage inequality was broadly stable, while monthly-earnings inequality widened. The bottom decile of monthly earnings fell about 15% relative to 2006, and average days worked in the lowest daily-wage quartile declined from roughly 21.5 per month to about 19.5. In the highest quartile, days worked stayed close to 24. The amount earned over a month can therefore diverge from the price attached to a day of work.
The map from contract to outcome
The study does not simply compare workers who belong to unions with workers who do not. It exploits the timing of contractual-floor updates within agreements. In rolling panels of two consecutive periods, workers covered by an agreement whose median nominal floor rose by at least 1% are treated as exposed to an update. Workers covered by agreements with no nominal growth form the comparison group; modest increases below 1% are left out. Among treated agreements, the average median-floor increase is about 2.19%.
The preferred comparison also checks whether the two groups were already moving differently before the update. The study uses the preceding period as a placebo window, when the relevant floors were unchanged. Small, statistically insignificant placebo effects support the interpretation that the later difference is not simply a continuation of a pre-existing divergence. This is evidence for a short-run causal response under the design's assumptions, not proof that every feature of bargaining timing is random.
The result is a distributional pattern rather than a single average. The mean daily wage rises, but the response is muted at the 10th percentile and stronger at the median and 90th percentile. The implied 90–10 gap is positive, although its estimate is imprecise; the 50–10 gap is also positive and more precisely estimated, while the 90–50 gap is positive but not statistically significant. The careful conclusion is therefore not that every inequality measure is definitively widened. It is that the strongest and clearest estimated wage-rate movements are away from the lower tail, producing an inequality-enhancing pattern in this setting.
The diagram captures why the headline cannot be interpreted from the contract schedule alone. A contractual update creates exposure across several grades, but the observed outcome also depends on whether the scheduled floor is implemented and which jobs remain available. Those margins will be examined in detail later. For now, the important point is that the study is tracing a path from an agreement-level change to several different distributions, not reporting one universal “union wage effect.”
There is also a boundary around the result. The authors estimate what happens when contractual minima change within a centralized, representative Italian bargaining system. They do not estimate what would happen if Italy replaced collective bargaining with firm-level or individual bargaining, nor do they compare the update directly with a statutory-minimum reform. Nor is the estimate a direct raise received by every worker who began at the 10th percentile. The reported percentiles are unconditional distributional statistics: workers can move relative to one another, and the composition of observed employment can change.
That boundary is not a technical footnote. It tells us what the paradox means. The finding is not that collective bargaining has an inherent tendency to reward the top more than the bottom. It is that, in this institutional setting, a coordinated update to a multi-rung contractual ladder passes through the wage distribution unevenly. To understand why, we have to look inside the ladder: how its floors differ, how workers within the same agreement respond, and how the researchers distinguish those responses from differences between agreements.