Did Unions End Long Work Hours?

Is it growth or organized labor that's more responsible for giving us shorter workdays?

The first general strike in American history took place in Philadelphia, Pennsylvania in 1835. The strike involved approximately 20,000 workers demanding two things:

1. Higher wages;

2. A ten-hour work day.

Banner of the Journeyman House Carpenters’ Association of Philadelphia promoting the ten-hour work day. United Brotherhood of Carpenters and Joiners of America Records, 1835.

The demand for higher wages is self-explanatory; the demand for a ten-hour work day, less so. The standard for labor hours at the time was dawn to dusk, sun up to sun down, for however long that might happen to be. Having to live with the hand nature deals isn’t a good situation, as nature deals different hands in different seasons and that causes trouble. In the summer months, the work days could be fifteen hours long, but as days dropped below ten hours in the winter months, wages became thin and the harsh winters became all the worse for want of money. The strikers’ demand was asking employers to set the work day by hours rather than by the sun, and to raise wages so families would could have more consistency throughout the year.

The strikers won and by the end of the year, the standard for city laborers across the U.S. was ten hours, except in Boston. European labor movements followed suit, demanding ten-hour work days in their own nations.

Boston, as it turns out, would be where some of the first major successes in earning an eight-hour work day would take place. In 1842, Boston’s ship carpenters—who were not unionized—earned the eight-hour work day. Chicago’s labor movement made the eight-hour day one of their central demands twenty years later, and Baltimore’s labor movement called for a national eight-hour work day resolution two years later in 1866. Two years after that, Congress made the eight-hour work day the law of the land for federal employees, but it wasn’t very effective right away, as they also slashed wages by 20%. When Ulysses S. Grant was elected the following year, he reversed the pay cuts.

Eight-hour Leagues cropped up all across the country and kept fighting for workers to work just eight hours a day, eventually achieving their goal by about 1919. Just take a look at this chart based on data from various sources:

The first thing that really stands out is the abrupt progress in the Huberman and Minns series. What, precisely, is going on between 1913 and 1938 that could lead to such a vast reduction in working hours? In economic historian Robert Whaples’ telling (see also), the major progress happened in a five-year period between 1914 and 1919, and there are two potential and non-exclusive explanations for that period’s rapid progress:

1. There was a major uptick in growth to 3.26pp per annum, far exceeding the 1.4pp annual growth that occurred between 1900 and 1914 and coinciding with a tight labor market that came with major real annual earnings improvements;

2. The federal government began intervening in the labor market at a previously unseen scale, enacting laws such as the Adamson Act (which limited railroad workers to a 48-hour work week) and mediating labor disputes in a manner favorable to workers, all while state governments took similar actions largely at the behest of successful labor movements that were managing to make their way into federal wartime agencies and other high political appointments.

These explanations suggest that the major contributors to the period’s work hour improvements came from growth or unions. But did both influences exist? Were they equal, or did one of these factors do the real heavy lifting? Perhaps we can develop a prior by doing a modern test.

The OECD provides data on union density (the percentage of workers in unions) and it also provides data on collective bargaining coverage (the percentage of workers who can engage in collective bargaining).¹ If we correlate union density or collective bargaining coverage with working hours in the years 2000 or 2017, we see large, significant negative correlations. For example, union density correlated at -0.502 and -0.488 with working hours in 2000 and 2017, respectively, while collective bargaining coverage correlated with working hours at -0.609 and -0.745 in the same years. From the outset, it seems that the presence of more organized labor is related to countries having fewer working hours.

But this alone cannot be decisive, as countries vary culturally in ways that might promote both less work and more union penetration and acceptance, as in the Nordics or Germany. To get to the meet of this, we need to look at the changes in both quantities to see if the rise or fall of unions augments the rate of decline in working hours seen across countries. That regression with trade density as the independent variable looks like so:

Most countries in the OECD saw simultaneous declines in working hours and trade union density. This chart isn’t so clear on the magnitude, but it does seem to be reduced compared to the cross-sectional correlation. A simple ordinary least squares regression produces a coefficient of -4.209, with a p-value of 0.059, or just barely nonsignificant. If we introduce population weighting by the square-root of the population size in 2000 in millions, the coefficient increases to -4.419 and the p-value also rises to 0.135. If we ignore population weighting, a robust regression with HC5 standard errors suggests a p-value of 0.081 and if we throw in weighting, the p-value on that becomes 0.201. Using the population in 2017, our OLS and robust p-values are 0.140 and 0.206 with a coefficient of -4.384.

So maybe this is evidence for an effect of unions such that more unions equates to fewer working hours. Maybe it isn’t, but it’s very hard to tell given the small number of countries and the potential for measurement error to have thrown off the results somewhat.² For even more clarity, let’s try this with the data on collective bargaining coverage.

Once again, both labor organizing and working hours declined and it looks like perhaps there’s a significant negative relationship such that the smaller the decline or the greater the increase in collective bargaining coverage, the more working hours fell. Appearances are deceiving, however, because the OLS coefficient of -0.786 had a p-value of 0.614, and the robust regression had an even worse p-value of 0.957, and population weighting doesn’t make a dent here at all. There was more missing data for this analysis relative to the previous one, but the result was a piddling correlation of -0.114 paired with a larger set of cross-sectional correlations, so who really knows.

To me, what this analysis suggests is that in the modern day, unions aren’t clearly responsible for huge impacts on working hours in the admittedly confounded setting of cross-national comparisons. But if they were in the past, I still think there should be clearer statistical signals. As an example of where one such signal might show up, consider old age labor force participation (which, like child labor force participation, had by then been declining for a long time). In contrast with the abrupt decline in working hours circa 1914-19 for the economy as a whole, the participation of old men in the economy was on the decline before 1910, but it actually increased somewhat between then and 1920, perhaps because of the promise of higher wages generated by the aforementioned tight labor market.

Gordon 2016, The Rise and Fall of American Growth, Figure 8–4. Labor Force Participation Rate by Sex, Ages 65–74, 1870-2010

One striking fact to consider is that unions could not have been ultimately responsible for a decline in annual working hours, because there had to be economic growth to give labor organizing something to demand. If people were toiling in fields and skipping work would be deadly, there could be no eight-hour work day; it simply wouldn’t be feasible without the ability to trade off wages and work.

Whaples assembled data from various sources to provide a more illuminating test of his two hypotheses than the ones I’ve outlined based on cross-sectional hints, international comparisons, and gesticulating about the possibility of impacts of unions on labor conditions. Using data on industry-level union growth, strikes, and wages, Whaples observed a backward-bending labor supply curve, that rising wages and manufacturing sector growth could explain almost half of the reduction in working hours, and that unionization explained about a third of the increase in wages which, combined with strikes, meant unions could explain about 14% of the fall in hours 1914-19.

Whaples also saw that a tight labor market was likely to have driven down hours more than unions based on the behavior of immigrants. Early-20ᵗʰ-century immigrants frequently made their way to America with the goal of returning to their home countries after working to amass a nest egg in America. Immigrants thus tended to work extremely long hours so they could go home. But the war strangled immigration and caused a draw-down on immigrant work in cities, both due to constraining the supply of immigrant workers and by reducing the work desires of existing immigrants. Consistent with this thesis, the disruption to immigration explains about 20% of the decline in working hours.

Women joining the labor force to help with the tight labor market and war effort, and their working fewer hours contributed a small amount to the decline in working hours. State laws regulating working hours also contributed to the decline, but not much because they tended to follow declines in working hours, effectively codifying what had already taken place for other reasons. Federal government regulations and labor decisions, such as those enacted on shipping and steel workers, also made modest contributions, putting the total effect of the government—at the state and federal levels—at around 10-15% of the decline. The revolution in managerial techniques and employers’ shifting preferences towards keeping workers for most of their lives instead of just temporarily might have accounted for another 4% of the decline, but the real kicker here is electrification, which accounted for almost 30% of the overall decline in Whaples’ industry-level estimation. To wit:

The most important force reducing the work week was the rise in wages, and there are a number of reasons why workers may have been especially willing to trade wages for leisure during this period. First, workers were accustomed to obtaining leisure through unemployment. They may have desired a shorter work day as a substitute for the unemployed leisure time lost during this period. Second, this period may have been especially prone to a burst in the purchase of leisure because consumption of discretionary goods was constrained…. Finally, it takes time and conscious decision making to change consumption patterns and form new tastes, but as experienced consumers of leisure workers can readily expand its use.

If we look internationally, it becomes clear that other Western nations experienced similar trends in working hours, albeit with some notable differences.

The suggestion that all of this could have been down to one labor movement is off the table, given the unrealistic required closeness of coordination and the known ideological differences in movements across countries. But, as noted earlier, the labor movement in America has certainly influenced the one in Europe, and there’s also other evidence for cross-pollination. So while the picture for 1914-19 clearly suggested that America’s progress in reducing working hours wasn’t a product of unions, who’s to say that they weren’t critical elsewhere or at other points in time?³

Well, perhaps, economists and historians. When Whaples surveyed them at the Economic History Association meeting, he asked about whether economic growth or labor unions were primary causes of the decline in American working hours before the Great Depression (1929), denoting the period in which the vast majority of the progress on reducing working hours was made.⁴ There was resounding agreement that the dominant force was economic growth rather than labor unions:

As described here:

Over eighty percent accepted the proposition that “the reduction in the length of the workweek in American manufacturing before the Great Depression was primarily due to economic growth and the increased wages it brought”. Other broad forces probably played only a secondary role. For example, roughly two-thirds of economic historians surveyed rejected the proposition that the efforts of labor unions were the primary cause of the drop in work hours before the Great Depression.

Other, more recent empirical work has yielded similar conclusions. For example, Vandenbroucke found, much like Whaples argued, that rising wages and declining costs of recreation could explain the overwhelming majority of the decline in working hours between 1900 and 1950, and though there was a spike in union activity in the 1920s, it arrived after the precipitous decline in working hours circa 1914-19 was completed and while continued decline was already underway. In fact, Vandenbroucke’s chart that conveyed his point about unions arriving after the fact shows quite clearly that unions really hit their stride long after almost all of the progress in shortening working hours had been made:

Vandenbroucke 2009, Fig. 5c.

At the end of the day, it’s hard to chalk up much progress in reducing working hours to the effects of organized labor and it’s very easy to attribute the vast majority of the progress to economic growth. So, while labor unions played a small role, every Labor Day from now on, you should be able to confidently tell people, “Unions didn’t give us shorter working hours, economic growth did.”

1. I used the earliest and latest available years from the linked sources for union density, working hour, and collective bargaining source, settling on 2000 and 2017 as the chosen years for maximum data completeness. Here’s how I handled missingness.

Trade Union Density: Australia’s 2017 value was the average of 2016 and 2018. France’s 2017 value was their 2016 value. Greece’s 2017 value was their 2016 value and their 2000 value was their 2001 value. Hungary’s 2017 value was the average of 2016 and 2018. I used Latvia’s 2003 value for 2000, Lithuania’s 2001 value for 2000, and Luxembourg’s 2003 value for 2000. Portugal’s 2017 value was their 2016 value and their 2000 value was their 2002 value. Slovenia’s 2017 value was their 2015 value.

Work Hours: No missingness.

Collective Bargaining Coverage: The UnadjCov variable was used unless a country only had values of UnadjCov_s, in which case that was used. Australia had no usable data, nor did France, Ireland, Israel, Italy, Latvia, Slovakia, or Slovenia. Chile, Colombia, Costa Rica, Denmark, Estonia, Lithuania, and New Zealand were missing values for 2000 and could not be imputed from an adjacent year. Luxembourg was missing a value for 2017. I did not pursue more advanced imputation strategies. Canada used UnadjCov_s as did Germany, New Zealand, Norway, Britain, and the United States. Chile used their 2016 value for 2017, and Denmark and Estonia used their respective 2018 and 2015 values for 2017. Norway’s 2000 value was taken from 1998 and Poland’s 2017 value was based on the average of their 2015 and 2019 values, while Switzerland’s 2000 value was based on the average of their 1999 and 2001 values and their 2017 value was based on the average of 2016 and 2018.

2. In other words, we get the ranks about right, but there’s more error in measurements of change.

3. There are also non-growth and non-union sources of reductions in working hours, such as elites responding to the potential for communist revolutions by appeasing workers.

4. That is, before the Great Depression.