Union wage effects

The union wage gap is the percentage by which union workers’ wages exceed those of comparable non-union workers, $D = (w_U - w_N)/w_N$. Even for truly comparable workers, it equals the union’s effect on its members only if the non-union wage $w_N$ is what they would earn in a world without unions, $w^*$. Spillover effects push $w_N$ below $w^*$ and inflate the measured gap; threat effects push $w_N$ above $w^*$ and shrink it.

The measured gap and the true gain

The gap a researcher can compute compares two observed wages. The union’s true wage gain compares the union wage with a wage nobody observes:

$$D = \frac{w_U - w_N}{w_N} \qquad \text{versus} \qquad \Delta = \frac{w_U - w^*}{w^*}.$$

Three problems separate them.

1. Unions change the non-union wage. Workers who lose union jobs crowd into non-union jobs (spillover), and non-union employers may raise pay to keep unions out (threat).
2. Union workers are not a random sample. Who wants union jobs, and whom union employers hire, depends on skill.
3. Union status is misreported. In survey panels a few misclassified workers create spurious switches into and out of unions, which biases estimates that rely on switchers toward zero.¹

Example: without unions every worker earns $20; a union raises its members’ wage to $23, a true gain of 15 percent. If displaced workers drive the non-union wage to $18.50, the measured gap is 24.3 percent. If non-union firms instead match part of the raise and pay $21.20, the measured gap is 8.5 percent.

Spillover and threat effects

Assume a fixed workforce that moves freely between a union sector and a competitive non-union sector, with both paying $w^*$ before the union arrives. The union then raises the wage in its sector, as in the monopoly union model, and union employers cut jobs along their demand curve.

Union wage (w_U)

$23.00

Threat effect: share of the wage gap non-union firms close

0%

True union gain (w_U vs w*)

15.0%

Measured gap (w_U vs w_N)

24.3%

Non-union wage w_N

$18.50

Union jobs lost

7.5k

Workers without a job

0.0k

Open in Graph Lab

Figure 1. A union wage of $23 cuts union employment from 30 to 22.5 thousand. With pure spillover the 7.5 thousand displaced workers lower the non-union wage to $18.50, so the measured gap (24.3 percent) exceeds the true gain (15 percent); the purple triangle is output lost to misallocation. When non-union firms close part of the gap to deter organizing, the non-union wage rises, some workers go without jobs and the measured gap falls below the true gain.

• Spillover. Displaced workers raise labour supply in the non-union sector and lower its wage. The measured gap overstates the true gain.
• Threat. A non-union employer that fears organizing pays more than the market-clearing wage. The non-union wage rises, so the measured gap understates the true gain. Workers who cannot find jobs at the higher wages may wait for union openings.

The spillover outcome also wastes output. In equilibrium the last union worker produces $w_U$ while the last non-union worker produces $w_N$, so moving a worker back to the union sector would raise output by $w_U - w_N$. The triangle between the two demand curves is the deadweight loss of misallocated labour. Freeman and Medoff estimated the US output loss from union monopoly power at 0.2 to 0.4 percent of GNP.²

Key idea

Spillover lowers $w_N$ and makes the measured gap too large; threat raises $w_N$ and makes it too small. A researcher never observes $w^*$, so the sign of the bias depends on which effect dominates.

Selection and wage compression

Union contracts set standard rates: one wage per job, raised by seniority rather than individual merit. The union pay scale is therefore flatter in skill than the non-union wage. This union wage compression gives the largest gains to the least skilled and can leave the most skilled worse off than in a non-union job.

Slope of the union pay scale (1 = no compression)

0.70

Union premium for a $15 worker

36.7%

Union premium for a $40 worker

−5.0%

Skill above which a union job pays less

$33.33

Open in Graph Lab

Figure 2. A union scale that pays 10 percent more at average skill but has a slope of 0.7 gives a worker who could earn $15 elsewhere a 37 percent premium and a $40 worker a 5 percent cut. Workers below the crossing point ($33.33) queue for union jobs; workers above it avoid them.

Compression drives selection in both directions:

• Less-skilled workers want union jobs, so jobs are rationed. Employers facing a queue hire the best applicants, so union workers are positively selected within low-skill groups and the raw gap overstates the union effect for them.
• Highly skilled workers avoid union jobs unless their non-union options are poor, so union workers are negatively selected at the top and the raw gap understates the effect there.

Card’s longitudinal study of US men found exactly this pattern: positive selection among the less skilled, negative selection among the more skilled, and a true union wage effect that is larger for less-skilled workers. For all workers pooled, the biases roughly offset and a simple cross-section estimate was close to the corrected one.³

A newer approach compares workplaces where a union narrowly won or lost a certification vote. DiNardo and Lee found estimates of the wage effect of new unionization close to zero, with small effects on employment, output and survival.⁴ Frandsen found that close elections are not as good as random and, after correcting for it, that unionization reduced payroll, employment and establishment survival, mainly through changes in who works there.⁵

Worked example

Union-sector demand is $w = 32 - 0.4E_U$ and non-union demand is $w = 34 - 0.2E_N$, with $E$ in thousands and $E_U + E_N = 100$. These are the curves in Figure 1.

1. Wage without the union. Set the two demand prices equal: $32 - 0.4E_U = 34 - 0.2(100 - E_U)$, so $18 = 0.6E_U$, $E_U = 30$, $E_N = 70$ and $w^* = 20$.

2. Union employment. At $w_U = 23$, $E_U = (32 - 23)/0.4 = 22.5$, so 7.5 thousand workers are displaced.

3. Spillover. Non-union employment rises to 77.5 thousand, so $w_N = 34 - 0.2(77.5) = 18.50$.

4. Measured gap and true gain. $D = (23 - 18.50)/18.50 = 24.3$ percent; $\Delta = (23 - 20)/20 = 15$ percent.

5. Deadweight loss. The triangle between the demand curves over the displaced workers: $\tfrac{1}{2}(23 - 18.50)(7.5) = 16.9$ thousand dollars an hour.

6. Threat instead. If non-union firms pay $21.20, they employ $(34 - 21.20)/0.2 = 64$ thousand. The measured gap is $(23 - 21.20)/21.20 = 8.5$ percent, below the true 15 percent, and $100 - 22.5 - 64 = 13.5$ thousand workers are without jobs.

Evidence on the size of the gap

United States. H. Gregg Lewis’s survey of 165 studies concluded that the US union wage gap averaged about 15 percent from 1967 to 1979. Blanchflower and Bryson found that it averaged 17 to 18 percent from 1973 to 2002, with the private-sector gap falling from 21.5 percent in 1983 to 1988 to 17.0 percent in 1996 to 2001.²

Canada. In the 1999 Workplace and Employee Survey, union workers earned $20.36 an hour and non-union workers $17.82, a raw premium of 14.3 percent; controlling for worker and workplace characteristics cut it to 7.7 percent.⁶ In 2025 the Labour Force Survey shows an average hourly wage of $38.74 for employees with union coverage and $35.37 without, a raw gap of 9.5 percent. The gap in medians is larger, 26 percent ($36.00 against $28.50), because union pay is compressed: the non-union distribution has more workers near the bottom and a longer top tail that pulls up its mean.⁷

Average hourly wage by union coverage, Canada

Employees aged 15 and over, current dollars

• Union coverage
• No union coverage

Average hourly wages of Canadian employees with and without union coverage. The raw gap fell from 31 percent in 1997 to 9.5 percent in 2025; it compares different workers in different jobs, so it is not the union wage effect.Source: Statistics Canada, Table 14-10-0066-01, Employee wages by job permanency and union coverage, annual. Retrieved 2026-10-23.

Show the data and notes

Average hourly wage rate of employees covered by a collective agreement (members or not) and of employees with no union coverage. Labour Force Survey, both genders, permanent and temporary jobs. The gap compares different workers in different jobs, so it is not the causal effect of unions on wages.

Period	Union coverage	No union coverage
1997	$18.52	$14.10
1998	$18.68	$14.36
1999	$19.05	$14.80
2000	$19.46	$15.33
2001	$19.89	$15.94
2002	$20.51	$16.31
2003	$20.98	$16.65
2004	$21.57	$17.07
2005	$22.15	$17.64
2006	$23.17	$18.77
2007	$24.13	$19.56
2008	$24.93	$20.47
2009	$25.79	$21.18
2010	$26.61	$21.49
2011	$27.07	$22.05
2012	$27.69	$22.65
2013	$28.42	$23.11
2014	$28.88	$23.58
2015	$29.45	$24.30
2016	$30.05	$24.83
2017	$30.36	$25.29
2018	$31.07	$26.08
2019	$31.68	$26.89
2020	$32.87	$28.77
2021	$33.45	$29.46
2022	$34.47	$30.88
2023	$35.72	$32.62
2024	$37.15	$34.35
2025	$38.74	$35.37

Unions and wage inequality

Unions affect inequality through three channels. Standard rates compress wages within the union sector. The union premium widens the gap between union and non-union workers. And union coverage is concentrated in the middle of the skill distribution, so the premium mostly raises middle wages.⁸

For men, compression dominates. Lemieux’s estimates imply that unionization reduced the variance of wages among Canadian men by about 17 percent.⁸ Card, Lemieux and Riddell found that if the union effect on US men had stayed at its 1973 to 1974 level, the variance of male wages would have grown 31 percent less than it did through 2001 (14 percent less with skill controls); the decline in union compression explained 9 to 29 percent of the rise in the UK. Unions do not reduce wage inequality among women: unionized women are more concentrated in the upper part of the wage distribution, and the union premium is not larger for less-skilled women.⁸ Including threat effects on non-union employers doubles the estimated contribution of de-unionization to rising US male wage inequality.⁹

Union coverage and unionization rates, Canada

Employees aged 15 and over, annual averages

Canadian union coverage fell only about three points from 1997 to 2025. Card, Lemieux and Riddell found that the union compression effect for Canadian men declined from 1984 to 2001 while overall Canadian wage inequality stayed stable.Source: Statistics Canada, Table 14-10-0129-01, Union status by geography. Retrieved 2026-10-23.

Show the data and notes

Coverage rate: employees who are union members and/or covered by a collective agreement, as a share of all employees. Unionization rate: union members only. Labour Force Survey, both genders, 10 provinces. Statistics Canada cautions that 2006 changes to LFS processing may have shifted levels; compare across 2006 with care.

Period	Union coverage rate
1997	33.7%
1998	33.0%
1999	32.3%
2000	32.3%
2001	32.3%
2002	32.2%
2003	32.2%
2004	31.7%
2005	32.0%
2006	31.6%
2007	31.3%
2008	31.0%
2009	31.5%
2010	31.4%
2011	30.8%
2012	30.9%
2013	30.9%
2014	30.2%
2015	30.4%
2016	30.0%
2017	30.0%
2018	29.6%
2019	29.8%
2020	31.0%
2021	30.7%
2022	30.4%
2023	30.5%
2024	30.2%
2025	30.6%

Non-wage effects

Freeman and Medoff described two faces of unions. The monopoly face raises wages and misallocates labour. The collective voice face gives workers a way to fix problems without quitting: a grievance procedure replaces exit.¹⁰

• Quits. Union members quit less and stay longer with their employer.¹¹
• Benefits. Unionism raises the share of compensation paid as fringe benefits, especially pensions, vacation pay and insurance; in percentage terms the union effect on benefits exceeds the effect on wages.¹²
• Productivity. Brown and Medoff found higher output per worker in unionized manufacturing, under assumptions they could not test.¹³ Later evidence is mixed: a meta-analysis found a positive association in the United States, including US manufacturing, and a negative one in the United Kingdom, and Hirsch concludes that the average US effect is near zero.¹⁴¹⁵
• Profits and investment. Union firms typically earn profits 10 to 20 percent lower and invest less in physical capital and R&D, with little evidence of more business failures.¹⁵

Common mistakes

Traps on exams

• Treating the measured gap as the union effect. $D$ compares union and non-union workers; the true gain compares the union wage with $w^*$, which nobody observes.
• Getting the sign of the bias wrong. Spillover lowers $w_N$ and overstates the effect; threat raises $w_N$ and understates it.
• Saying unions always increase inequality. They widen the union-non-union gap but compress pay within the union sector; for men the net effect is lower inequality.
• Equating higher productivity with higher profit. Even where union productivity is higher, profits are typically lower.

On the exam

For spillover, draw the union and non-union markets with $w^*$ in both. Raise the union wage, mark the jobs lost, shift non-union supply right by the same number of workers, and label the new $w_N$. State the measured gap and the true gain with $w^*$ in the denominator of the second.

Check yourself

1.

A union raises wages in its sector, union employers cut jobs, and the displaced workers take jobs in the non-union sector. How does this spillover affect the measured union wage gap?

1It lowers the non-union wage, so the measured gap overstates the union's true wage gain2It raises the non-union wage, so the measured gap understates the union's true wage gain3It has no effect on the measured gap, because the union wage is unchanged4It lowers both wages equally, so the measured gap is unbiased

Check answer

2.

Without unions, a group of workers would earn $25 an hour. A union raises its members' wage to $30, and because displaced workers crowd into non-union jobs, comparable non-union workers now earn $24. What union wage gap, in percent, would a researcher measure?

Your answer (percent)

Check answer

3.

Card (1996) found that union workers are positively selected among less-skilled men and negatively selected among highly skilled men. Which mechanism explains this pattern?

1Less-skilled workers queue for compressed union pay and employers hire the best of them, while highly skilled workers take union jobs only if their non-union options are poor2Unions recruit the most able workers at every skill level3Employers prefer to unionize their least productive workers4Measurement error in union status makes union workers look more skilled than they are

Check answer

Practise more questions on this topic

Sources

1. Freeman, R. B. (1984). Longitudinal analyses of the effects of trade unions. Journal of Labor Economics, 2(1), 1 to 26. doi:10.1086/298021. ↩

2. Blanchflower, D. G. and A. Bryson (2004). What effect do unions have on wages now and would Freeman and Medoff be surprised? Journal of Labor Research, 25(3), 383 to 414. doi:10.1007/s12122-004-1022-9. Summarizes Lewis, H. G. (1986), Union Relative Wage Effects: A Survey, University of Chicago Press, and the output-loss estimate in Freeman and Medoff (1984). ↩ ↩²

3. Card, D. (1996). The effect of unions on the structure of wages: A longitudinal analysis. Econometrica, 64(4), 957 to 979. PDF. ↩

4. DiNardo, J. and D. S. Lee (2004). Economic impacts of new unionization on private sector employers: 1984 to 2001. Quarterly Journal of Economics, 119(4), 1383 to 1441. doi:10.1162/0033553042476189. ↩

5. Frandsen, B. R. (2021). The surprising impacts of unionization: Evidence from matched employer-employee data. Journal of Labor Economics, 39(4), 861 to 894. doi:10.1086/711852. ↩

6. Fang, T. and A. Verma (2002). Union wage premium. Perspectives on Labour and Income, 3(9), Statistics Canada catalogue no. 75-001-XIE. ↩

7. Statistics Canada, Table 14-10-0066-01, Employee wages by job permanency and union coverage, annual. Employees aged 15 and over, 2025. ↩

8. Card, D., T. Lemieux and W. C. Riddell (2004). Unions and wage inequality. Journal of Labor Research, 25(4), 519 to 559. PDF. Reports the estimate of Lemieux, T. (1998), Journal of Labor Economics, 16(2), 261 to 291. ↩ ↩² ↩³

9. Fortin, N. M., T. Lemieux and N. Lloyd (2021). Labor market institutions and the distribution of wages: The role of spillover effects. Journal of Labor Economics, 39(S2), S369 to S412. doi:10.1086/712923. ↩

10. Freeman, R. B. and J. L. Medoff (1984). What Do Unions Do? New York: Basic Books. ↩

11. Freeman, R. B. (1980). The exit-voice tradeoff in the labor market: Unionism, job tenure, quits, and separations. Quarterly Journal of Economics, 94(4), 643 to 673. NBER working paper 242. ↩

12. Freeman, R. B. (1981). The effect of unionism on fringe benefits. Industrial and Labor Relations Review, 34(4), 489 to 509. doi:10.1177/001979398103400401. ↩

13. Brown, C. and J. Medoff (1978). Trade unions in the production process. Journal of Political Economy, 86(3), 355 to 378. doi:10.1086/260677. ↩

14. Doucouliagos, C. and P. Laroche (2003). What do unions do to productivity? A meta-analysis. Industrial Relations, 42(4), 650 to 691. doi:10.1111/1468-232X.00310. ↩

15. Hirsch, B. T. (2004). What do unions do for economic performance? Journal of Labor Research, 25(3), 415 to 455. IZA discussion paper 892. ↩ ↩²