Effects of Gender-Specific Workweek Restrictions

The 20th century saw the rise and fall of a legal regime ostensibly aimed at protecting women that had the effect of limiting women’s labor market opportunities. One type of such legislation explicitly limited the number of hours a woman could work per day or per week. By 1950, nearly every state had such a gender-specific hours restriction (GHR). Within two and a half decades, these laws were eliminated through legislative, judicial, and executive actions. We study the consequences of ending these laws for both women and men by exploiting their staggered repeal to estimate the effects on weekly hours, employment, and earnings.

We first develop a model extending that in Landes (1980). The model distinguishes between total employment and hours per worker as well as allowing for male and female labor to enter as complements or substitutes to production. We conceptualize the repeal of a GHR as a positive shock to the labor supply of women. What effect this has on men depends on whether male and female labor are complements. If so, then male labor goes up along with wages. If male and female labor are substitutes, then the opposite happens with male labor and wages falling. What happens to women’s labor market outcomes is more complicated depending on, for example, the substitutability of employment and hours per worker.

To empirically estimate the impact of GHRs, we exploit their staggered state-level repeals in a difference-in-differences framework. First, as for the quantity of labor, we find that repeal increased the probability that women worked beyond the prior GHR limit by about 1 percentage point, roughly doubling the pre-repeal mean. Average weekly hours for women also rose by around 0.4 to 0.9 hours per week. Both results are consistent with our interpretation of repeal operating as a positive labor supply shock for women. Perhaps more surprisingly, men’s average workweek also rose by around 0.5 hours per week. On the extensive margin of employment, we find that repeal reduced flows out of industries covered by the GHR by roughly 1.2 to 1.4 percentage points for women and 1.1 to 1.5 percentage points for men. Our findings echo Goldin (1988)’s analysis of the introduction of these laws, which found that introducing GHRs reduced hours worked by both women and men, suggesting men and women are complements in production.¹ Marchingiglio and Poyker (2026), in contrast, find that men’s employment increased after the introduction of gender-specific minimum wages.

Turning to labor earnings, we find that women’s hourly and annual earnings both fell following the repeal of these laws by around 5 percent. This is consistent with the increase in female labor supply swamping any feedback effects on female labor demand from the rise in male employment. As for men, from the viewpoint of the model when male and female labor are complements, the reason male labor inputs rise following repeal is that a positive shock to female labor supply acts as a positive demand shock for male labor. However, we find no evidence that male annual or hourly earnings rose following repeal. If anything, there is some evidence that hourly earnings actually fell leaving annual earnings basically flat. These results suggest, if anything, that repeal led to an increase in male labor demand and supply. One potential explanation for an increase in male labor supply is a complementarity in leisure between spouses. If spouses value shared leisure, then a change in wives’ work opportunities could shift the supply of husbands’ labor. Consistent with this idea, Goux, Maurin and Petrongolo (2014) find that husbands responded to their wives cutting work hours by cutting their own work hours. We report suggestive evidence for this mechanism by showing that hours rose more for married men who are presumably more affected by the leisure complementarity than for single men following GHR repeal.

The fall in women’s earnings is consistent with expanded female labor supply, but why would women work more hours and earn less overall? We hypothesize that an important but (unmeasurable) effect of repealing the GHR was to provide more flexibility to women in scheduling their workweek. This was particularly the case when daily limits were repealed. In this case, even if hours per week for a woman remained fixed, she might be willing to accept wages if it meant those hours could be scheduled at more convenient times that might not have been legal when the GHR was in place. If the scheduling of hours matters, then whether a woman is better off cannot simply be inferred by what happened to her total hours worked and earnings.

The mixed effects for women, with on average more work above the old limit but with lower hourly wages, help rationalize the fierce debate around repealing these laws. Women were not unanimous in wanting these restrictions repealed. Doepke et al. (2024) study the political economy of the introduction of these laws restricting hours and argue that the laws were about men trying to limit economic competition from women. Our evidence suggests that repeal was not simply a battle of the sexes: some women gained access to longer hours, while women competing in affected labor markets faced lower earnings, and men’s hours moved as well without clear earnings gains. Another driver of the debate over repeal may therefore have been disagreement among women over the value of limiting competition from other women who stood to take advantage of greater labor market opportunities. Our conclusion echoes that of Siegel (1987-1988), who called the debate over the origin of these laws a “struggle of women against women, in which women serve[d] as the primary agents of their own victimization.”²

Although our paper addresses the impact of the repeal of hours restrictions, it fits into a literature that empirically explores the effects of introducing and repealing gender-specific labor market regulations more broadly. Goldin (1988) and Landes (1980) examined the effects of the introduction of GHRs in the early 1900s. Kato and Kodama (2018) investigated the effects of relaxing overtime restrictions on female employment in Japan. Like these earlier papers, we also find similar effects on work hours for males and females from changes in regulations that legally applied to only one gender. We go beyond earlier work by also examining the effects on wages for both men and women, as these effects are particularly important for understanding who benefited from these laws. Zveglich and van der Meulen Rodgers (2003) studied the introduction of hours restrictions in Taiwan and found that employment and hours for women fell when an hours restriction was imposed, but there was no effect on wages. They argued that the null effect for wages reflects that women made up a small share of the labor force in their setting. Haddad and Kattan (2024) examined other types of protective laws for women in the labor force from 19th century America including seating, health and safety, and night-work regulations. They found that these laws regulating health and safety conditions of female work actually increased the likelihood of female employment by about 4 to 8 percent by making the jobs more attractive. These regulations tended to come in a package with workweek restrictions, so a key question for us will be to distinguish the effects of those regulations from the effects of the direct regulation on the workweek of women.

During the decade before World War I, several states enacted protective labor laws applying only to women, regulating their hours of work and conditions of employment. The laws specified maximum daily or weekly hours, minimum days of rest, the provision of meal and rest periods, and limitations on night work. At their peak in 1967, these gender-specific laws existed in 46 states. Some labor standards were established by statute; others by the orders of industrial commission boards. In the majority of cases, the GHRs applied only to women working in manufacturing and mercantile industries. Some states had much broader coverage of women. In Arizona, for example, nearly all female workers were subject to a limit of 8 hours per day and 48 hours per week, with a few exceptions for domestic workers and some others.³

After the passage of Title VII in the Civil Rights Act of 1964, which called for the elimination of discrimination on the basis of sex, a pressing question for states was how to reconcile these sex-specific regulations with the federal mandate not to discriminate on the basis of sex. A year before the passage of the Civil Rights Act, the President’s Commission on Status of Women had studied labor regulations applying to only women. Its recommendation was to extend the same protections to men and to incorporate statutes pertaining to lunch periods, weight-lifting limits, and occupational hazards into a comprehensive safety and health program applicable to men and women alike. In 1967, the Task Force on Labor Standards was established to study developments in these issues since passage of the act. In their final 1968 report, the task force, rather than suggesting the standards be applied to everyone, called for the repeal of these restrictions altogether.

The Civil Rights Act of 1964 created the Equal Employment Opportunity Commission (EEOC) to address workplace discrimination. At the same time, the act created the “Bona Fide Occupational Qualification” (BFOQ) exemption to the Act’s general ban on employment discrimination. Such a qualification is an attribute that employers are allowed to consider when making employment decisions. The question the EEOC struggled with during the latter half of the 1960s was exactly how the BFOQ applied when there were gender-specific labor regulations in place (Ross, 1970). In December 1965, the first EEOC guidance allowed restrictive laws to be used by employers as a BFOQ if employers acted in good faith and if the laws were protective and not discriminatory. The authors of the guidance believed that Congress had not intended to override the state protective laws with the Civil Rights Act. Subsequently, in August 1966, the EEOC announced that it would refrain from making a decision when Title VII conflicted with state laws. In February 1968, they rescinded this policy, returning to one in which they would determine whether the laws were discriminatory based on consultation with the states. In August 1969, it changed the guidance again, saying that no prohibitory law could be a BFOQ. In other words, the EEOC declared that state laws restricting the employment and hours of females were in conflict with the Civil Rights Act.

The EEOC’s opinions, “no matter how correct and persuasive,” were not binding because the EEOC was “only a policy and guideline-making, conciliatory and mediating agency” (Pitt, 1970). The EEOC’s action in August 1969, however, was the nudge needed to get the state-level legislative ball rolling. In 1970 Arizona repealed its maximum hours and weekly rest day law for females; New York released women from more than 21 restrictions on daily and weekly work hours and on night work; and Vermont released women ages 18 and over from its maximum hours law and repealed other provisions governing conditions of female employment. Other states, including California, Connecticut, Maryland, Massachusetts, New Mexico, North Carolina, Tennessee, and Virginia, adopted major amendments to these laws that reduced the scope or loosened the restrictive nature of restrictions. Several states made the restrictions inapplicable to employment that met or were covered by requirements of the Federal Fair Labor Standards Act (Weissbrodt, 1971). The map in Figure 1 shows when states legislatively repealed these laws.⁴

Courts at both the federal and state level eventually started ruling against these protective laws, but it took time. Even in 1968, several years after the passage of Title VII, “federal district courts in California and Louisiana refused to negate work hours limitations which restricted only women” (Bayh, 1972). A turning point was the case of Rosenfeld v. Southern Pacific Company heard initially at the end of 1968. The plaintiff, Leah Rosenfeld, applied for a position as agent-telegrapher in 1966. The defendant Southern Pacific denied her the job on two grounds: first, that women were not physically or “biologically” suited for such work; second, that giving the plaintiff the job would violate California’s maximum hours and weight restriction laws for women (Hill, 1979). The district court decided the case between Rosenfeld and Southern Pacific in favor of Rosenfeld with a summary judgment in which she was not awarded damages in large part because the company did not allow Rosenfeld to show that she could handle the long hours and extensive physical activity involved in the job. On appeal in 1971, the Ninth Circuit Court of Appeals decided that California’s labor restrictions for women should be disregarded. The judges stated that “[voiding the law] alone accords with the congressional purpose to eliminate subjective assumptions and traditional stereotyped conceptions regarding the physical ability of women to do particular work” and clearly enunciated Title VII’s mandate that women must be treated the same as men. As a consequence, the state of California announced it would no longer enforce its restrictions on hours and weight-lifting for women (Hill, 1979).⁵ The 1969 case of Weeks v. Southern Bell Telephone & Telegraph Co. involved a similar question and ended with a similar result. The Fifth Circuit Court of Appeals overturned the district court ruling and held that Georgia’s protective law violated Title VII.

One final case to mention is Caterpillar Tractor Co. v. Grabiec in 1970, which consolidated separate cases brought by the Caterpillar and the Illinois Bell Telephone Company against the state of Illinois challenging the Illinois Female Employment Act. The district court ruled in the companies’ favor, writing that “[m]ost, if not all, female employees of each plaintiff who were within the coverage of the Illinois Female Employment Act are fully capable of working in the operations of their respective employers for more than the maximum hours permitted by the Illinois Female Employment Act.” Subsequently, state female hours laws in Kentucky (1971), Louisiana (1971), Massachusetts (1971), Missouri (1971), Ohio (1972), and Pennsylvania (1971) were struck down by federal courts or state supreme courts (United States Women’s Bureau, 1974).⁶

Finally, in some states, the laws were invalidated simply on the basis of opinions by states’ attorneys general or administrative rulings. In general, these opinions stated that the hours laws were not applicable to employers covered under Title VII or the opinions modified the status of the laws directly. For example, in an opinion dated July 17, 1970, Wisconsin’s Attorney General Robert W. Warren advised the Department of Industry, Labor and Human Relations’ chairman that Sec. 103.02 of the statutes and administrative rules limiting the number of hours a woman could work “[was] superseded by provisions of the Civil Rights Act of 1964 as to employers covered by that Act, but other employers remain subject to the State law.” In support of this interpretation, he pointed to 1969 guidelines of the EEOC. We note that these different kinds of repeal were not necessarily mutually exclusive. For example, in a number of states, an administrative ruling came a few years before the law was repealed or modified by the legislature.

One question is whether these laws while in force had much of a bite. For example, Figure 2 shows that as of 1961 the state-specific workweek limits were in every case longer than 40 hours per week, and how many women really worked that long? The Report of the Task Force on Labor Standards to the Citizens’ Advisory Council on the Status of Women found that only 17 percent of married women workers worked more than 41 hours a week in 1966. However, the report did not conclude that hours limits were unimportant overall. Instead, it pointed to the daily hours limits as the bigger problem by “restrict[ing] women’s access to jobs, promotions and overtime pay.” Hedges (1971), in discussing a potential move to a 4-day workweek with 10-hour days that “intrigued management and is winning guarded support from labor organizations,” noted that GHR laws held back their adoption: “State restrictions on maximum daily hours for women mean that a 4-day week would necessitate a reduction in weekly hours.” Similarly, Poor (1970, p. 21) noted that “4-day schedules have raised legal problems for some firms, particularly where women employees’ hours of labor exceed state regulations.”

The records from several lawsuits challenging these laws also provide evidence for the discriminatory effects of the GHRs. In some cases, while defending themselves in court, employers explicitly cited these workweek restrictions as a basis for their discriminatory employment decisions. In the 1971 case Kober v. Westinghouse Electric Corporation (summarized in Gregory (1971)), Westinghouse defended the denial of the promotion of a woman on the grounds that promoting her would have conflicted with Pennsylvania law limiting work hours for women. Ratner (1980) quotes a statement of a United Auto Workers representative to the Equal Employment Opportunity Commission to the same effect:

The contracts we negotiate with employers provide equal pay, equal job opportunity, equal seniority, training, etc., but I couldn’t begin to estimate the number of grievances we have taken all the way to arbitration in an effort to enforce a contract only to be stymied by one or another of the so-called state “protective” laws [...] During war periods [laws] were honoured only in the breach. Yet when men were again available the employers resorted to the technique of combining two jobs into one so that it was beyond the state maximum weight law, or scheduling hours of work beyond the statutory limit for women in order to avoid hiring women employees.

In the 1972 case Jones Metal Products Co. v. Walker, the employer in admitting to discriminatory practices went so far as to contend that it was required to discriminate against women under Ohio law. On appeal, the firm argued that Title VII preempted state-level legislation and that a failure of the appellate court to invalidate the state law would put the firm in an impossible position:

According to appellants, continued compliance with these state statutes will subject them to a multiplicity of damage suits by their employees, in the federal courts, for violations of Title VII. Conversely, if they comply with the requirements of Title VII, they will subject themselves to suits by one or more of the appellees and will subject themselves to fines, imprisonment and the costs of defending such litigation.

The Supreme Court of Ohio agreed, and the state restrictions were invalidated. In the 1971 case Ridinger v. General Motors Corporation, the plaintiff contended that she had been “denied employment opportunities which are extended to males such as Saturday and Sunday overtime work and better paying jobs.” The defendants, General Motors and the International Union of Electrical, Radio and Machine Workers, did not dispute the charge, instead claiming that they were required to discriminate in this way.

The repeal of these protective laws was deeply intertwined with the debate over the Equal Rights Amendment (ERA). Initially, campaigners for the ERA faced pushback over fears that state protective laws would be disallowed following its passing (Kanowitz, 1979). To allay those concerns, early versions of the ERA specifically exempted state-level protective laws from consideration. The “Hayden rider”, introduced in the Senate in 1950 by Senator Carl Hayden of Arizona, stated that “The provisions of this article shall not be construed to impair any rights, benefits, or exemptions conferred by law upon persons of the female sex” (Neale, 2013). Versions of the ERA passed by the Senate over the 1950s included the rider. By the 1960s, the rider had fallen out of favor, and the 1964 Senate Judiciary Committee report stated that “[i]t is under the guise of so-called ‘rights’ or ‘benefits’ that women have been treated unequally and denied opportunities which are available to men.” Birch Bayh, the chairman of the Senate Subcommittee on Constitutional Amendments, in arguing for the merits of the ERA, noted that the state-level protective laws made it “more difficult for women to obtain work they desire and for which they are qualified, or to become supervisors.” Just reaching a vote on the ERA in 1971 required an unusual discharge petition since for 21 years, the chairman of the House Judiciary Committee, Emanuel Celler, blocked even a hearing on the law. According to the New York Times, Celler “argued mainly that the amendment should be defeated because it would probably render invalid laws aimed at protecting women”⁷ (Shanahan, 1971).

The heated debate about the consequences of these protective laws as part of the attempted enactment of the ERA shows that there was nowhere close to unanimity about the need to repeal these laws. As Time Magazine in August 1970 reported, “[m]ost of organized labor opposes [the ERA], arguing that it will destroy a broad carapace of laws that ‘protect’ women workers by regulating their hours and the kinds of work they can perform” (Time Magazine, 1970). Myra Wolfgang, vice president of the Hotel and Restaurant Employees and Bartenders International Union AFL-CIO, argued before the Senate that “frequently we [women] obtain real equality through a difference in treatment, rather than identity in treatment” and hence “the passage of an hours limitation law for women provided them with a shield against obligatory overtime to permit them to carry on their life at home as wives and mothers”. A spokesman for the Amalgamated Clothing Workers of America said that “local members of the union – of whom at least 80 per cent are women – would oppose a repeal of the law” and “[women] don’t want to work all the time.” These fears were not totally unwarranted. Following the repeal of the protective law in Michigan, Chrysler supposedly demanded such long work hours that, according to a female employee testifying before the Occupational Safety Standards Commission, “women dropped over from fatigue and exhaustion daily and had to be removed by stretcher” (Munts and Rice, 1970, p. 9).

According to a 1971 article in The New York Times, it was “working-class women” who were leading the charge against the GHR laws. These women argued that the laws “kept women from earning premium overtime pay that they wanted to earn, kept them out of better-paid job classifications and prevented their promotion to such jobs as foreman.” The United Auto Workers union, which had previously advocated for protective laws, became the first large union to hold that state laws were superseded by Title VII (Barnard, 2004, p. 398). In the end, unions came to a belief that “[s]tate protective laws have often been used by employers, and by some labor unions, to prevent women from holding better paying jobs, from working overtime, and from accumulating plantwide security and seniority” (Raphael, 1974, p. 30). It was ironic that unions eventually opposed these protective laws as they were the ones who had pushed for these laws in the first place. In a discussion of a book by Henry (1925) about women in trade unions, Landes (1980) noted the fears women in the early 20th century had about competition from female immigrants who were “asking for labor” and “being used to lower wages yet further for themselves and others.” In the end, unions at the time gave up trying to organize the new immigrants and instead decided that “[t]he one thing we can do to alleviate their hard lot is to secure legislation for shorter hours and for the minimum wage.” Our point is that the debate over these laws was just as much a debate among women as it was a debate between the sexes, as Doepke et al. (2024) suggest.

This period saw other state-level anti-gender discrimination actions. For example, by 1983, 42 states, the District of Columbia, and Puerto Rico had adopted Fair Employment Practices (FEP) laws covering private employment. Many of these laws prohibited discrimination on the basis of age while others went further, prohibiting discrimination on the basis of handicap, marital status, and sexual orientation. All these laws prohibited discrimination based on sex, with some providing stronger protection than was available under Title VII of the Civil Rights Act of 1964. If the adoption of these laws was correlated at the state level with the repeal of GHRs, then it is possible our estimates of the effects of the GHRs are actually due to the effects of the FEP instead.

Also complicating our analysis is that the state-level repeals of GHRs tended to happen as part of a package repealing other employment protections for women such as limits on how much they could be required to lift on the job. These other protections included required meal periods, limits on weight lifting, and bans on night work. While a state having a GHR did not necessarily mean it had, for example, a law protecting pregnant women, the reverse was almost always true with all of these provisions packaged into the same bill. This meant that when the non-workweek limit protections got repealed, the GHRs disappeared as well. Figure 3 plots the shift across states in their policies toward treating men and women equally in the workforce. Not surprisingly, there is a common trend in all these policies across states over time, but there are some differences in the exact timing. Therefore, in our most saturated specifications, we control for the most prominent of these gender-specific labor laws related to overtime and a gender-specific minimum wage.

To organize our empirical results, we build on the model of Landes (1980). We depart from the model’s original assumption that firms viewed men and women as perfect substitutes in production. Here, we consider both the perfect complements case and the perfect substitutes case. In fact, given the different types of work men and women did at the time, we would argue that the assumption of perfect complements is more reasonable. We also generalize the original model by considering two dimensions of preference heterogeneity: (1) along the extensive margin of whether a person works at all and (2) along the intensive margin of how many hours a person works, conditional on working at all. Though not explicitly stated, the original model had preference heterogeneity only on the extensive margin. Preference heterogeneity along the intensive margin allows us to capture the distribution of hours worked among those who work. More importantly, this second dimension of heterogeneity generates disagreement among women about whether the workweek limit should be repealed. We study the comparative statics of a marginal increase in the legally imposed workweek limit on women, denoted h̄.⁸ We also distinguish between employment and hours per worker in the production function and introduce a fixed payment for working that is separate from the hourly wage.

Labor Supply

An individual has an idiosyncratic value of not working δ and an idiosyncratic marginal disutility from working an additional hour β drawn from the joint distribution F(δ , β ).⁹ We assume that an individual’s preferences for work are private information, and firms do not try to screen workers based on these differences.¹⁰ The individual chooses to work at all if and only if

U +ω > δ, where U is the utility from working, which depends on the number of hours worked h and consumption c. Here ω is a fixed payment (“benefits”) from the employer that each individual receives for working any amount. Since we will assume that U is quasi-linear, we can assume that ω is measured in units of consumption.

The individual’s preferences over consumption and working hours are given by

α α+1 u(c, h) = c − β −1/α h α . α +1

Utility is maximized subject to the budget constraint c = wh, where w is the potentially hourly wage. Like the fixed payment for working ω, the hourly wage is only potentially conditioned on the individual’s gender. Note that with this utility function, δ can be interpreted as the level of consumption when h = 0. Conditional on working, the elasticity of an individual’s hours worked with respect to the wage is determined by the parameter α.

Conditional on working, the optimal number of hours worked given a workweek limit h̄ is

h = min{β wα , h̄}.

Hence, the utility from working is α α+1 U = wh − β −1/α h α . α +1 Individuals with a low marginal disutility of working an extra hour (a high value of β ) will either work up to the limit or if they have a high enough disutility from working at all, will not work at all. Individuals with a high disutility of the marginal hour (a low value of β ) will not be constrained by the workweek limit.

We now aggregate across individuals of a particular gender to calculate aggregate employment and average hours worked. The aggregate employment rate for gender g is

Lg = P [U + ωg > δ ] .

The average workweek among those employed for gender g is hg = E[h | U + ωg > δ ].

It is implicit that the expectation (and probability) above is over only individuals of gender g. For simplicity, we assume that the distribution of preferences F and the supply elasticity parameter α are the same for men and women. The only difference between men and women is that women are subject to the workweek limit h̄ while men are not.

Writing xb ≡ d ln x for log-changes, the linearized labor supply equations for hours and employment are

b bm − ηmh ω hm = ᾱmh w bm , b f − η hf ω h f = ᾱ hf w b b f + αh̄h , (1) bm = ᾱmL w L bm + ηmL ω bm , b f = ᾱ Lf w L b f + η Lf ω b f + αh̄L . (2)

Here αh̄h , αh̄L ≥ 0 are the elasticities of female hours and employment with respect to the workweek limit. In the appendix, we show that the elasticities ᾱgj ≥ 0 for j ∈ {h, L} and g ∈ { f , m}, meaning that hours and employment are increasing in the wage. The elasticities ηgh ≥ 0 capture the reduction in average hours caused by a higher ωg : the fixed payment attracts marginal workers with low β (low preference for long hours), who work fewer hours conditional on participating, pulling average hours per worker down. The employment elasticities ηgL are positive since a higher fixed payment makes participation more attractive. These terms were not present in Landes’ model, which did not have a fixed payment for working.

Labor Demand

Firms face perfectly competitive output and labor markets. The production function is

σ σ −1 Y= H σ , σ −1

where σ ≥ 1 controls the overall returns to scale of the production function. We consider a general CES aggregator for H of the form:   π π−1 π−1 π−1 H = (1 − s)Hm + sH f π π ,

Here π ≥ 0 controls the elasticity of substitution between male and female labor and s the relative importance of female to male labor inputs in production. We assume that the labor input for gender g is the Cobbγ 1−γ Douglas aggregator Hg = hg Lg of hours per worker and employment (not necessarily total hours hg Lg ).

Firms maximize profits max Y − L f (ω f + w f h f ) − Lm (ωm + wm hm ). h f ,hm ,L f ,Lm

The term ω f + w f h f represents the total compensation for a single worker comprising the fixed payment for working ω f and the variable payment for hours worked w f h f .

Note that what the firm chooses is average hours per worker and employment. Individual workers will work more or less than the average given the wage and their preferences. This is a shortcut to avoid modeling how individual worker’s hours enter the production function and specifying a distribution of wages across workers.

Letting Pg ≡ ∂∂Y Hg denote the marginal revenue product of gender-g labor, we have

−1 −σ −1 −1 −1 −σ −1 −1 Pf = s H π H −π f , Pm = (1 − s) H π Hm−π .

The first-order conditions for average hours and employment for gender g are

Hg Pg · γ = wg Lg (3) hg Hg Pg · (1 − γ) = ωg + wg hg . (4) Lg

The left-hand side of Eqn. (3) is the marginal revenue product of raising average hours by one unit; the right-hand side is the total wage bill across all Lg workers. The left-hand side of Eqn. (4) is the marginal revenue product of one additional worker; the right-hand side is total compensation per worker. Dividing the two conditions yields 1 − 2γ ωg = wg hg , γ which pins down the equilibrium ratio of the fixed payment to the (per worker) wage bill as a function of γ alone. Note that if γ > 1/2, then ωg < 0. This case will be ruled out for other reasons below. An important implication of this expression is that “variable” labor earnings wg hg is proportional to benefits ωg .

Linearizing the first-order conditions around the no GHR equilibrium and eliminating Pbg and H bg , we find:

b bg − w hg = ω bg , (5) 1 − π/σ b γ + π(1 − γ) γ(π − 1) bg = L H− bg − ω bg , w (6) 1 + γ(π − 1) 1 + γ(π − 1) 1 + γ(π − 1)

b = sf H bm is the log-change in the labor composite and (abusing notation )sg ≡ sCES b f +sm H (π−1)/π where H g (Hg /H) is the equilibrium quantity share of gender g. The presence of aggregate labor input H b couples male and female labor demand through the shared marginal revenue product. When σ > π, an increase in aggregate labor input H b raises Pg so employment rises.

Limiting cases for π

We consider three limiting cases for the value of π:

1. As π → 1, then male and female labor are neither complements nor substitutes. Instead, we have a Cobb-Douglas aggregation of the two.

2. As π → ∞, then male and female labor are perfect substitutes.

3. As π → 0, then male and female labor are perfect complements.

Equilibrium Responses

The system of supply and demand equations that define the equilibrium response can be written as Ax = b, where x stacks the eight within-gender variables (b hm , L bm , w bm , ω bm , b hf ,L bf ,w bf ,ω b f ) and b places the workweeklimit elasticities αh̄h , αh̄L in the two rows corresponding to the outcomes for women and zeros elsewhere.¹¹ For our comparative statics, we assume that H b > 0. A sufficient condition for this, which we derive in the appendix, is that ε Lf < γ/(1 − 2γ),

where ε Lf = α̃ Lf + η Lf (1 + η hf ) is the reduced-form elasticity of female employment.¹²

Table 1 summarizes the equilibrium outcomes assuming εgh ≥ 0 and H b > 0, which we derive in the appendix. In general, the labor market effects for women are ambiguous.¹³ The issue is that the direct effect of removing the workweek limit, which is to increase female labor supply, is offset by the indirect equilibrium effects. Which of these forces dominates depends on parameter values. On the other hand, the labor market effects for men are straightforward. They depend solely on the degree of substitutability between male and female labor because there is no direct effect on male labor supply from repealing the workweek limit. All that matters is the induced effect on male labor demand from the increase in female labor input.

We use the March Current Population Survey (CPS) from 1962 to 1980. We stop in 1980 because by then, all GHRs that were repealed had been repealed for over 5 years, a reasonable amount of time for any effects of the repeal to be present while limiting the number of potential confounders that come with including more years. One difficulty is that the CPS data prior to 1968 does not provide the detailed and consistent 1950 Census Bureau industrial classification system. This is a problem because many GHRs only applied to women working in particular industries. We therefore harmonize industry classifications across years into a coding system that is finer than the 2-digit NAICS code but coarser than the 3-digit NAICS code. For instance, we separate restaurants from the rest of the retail sector, but we cannot separate hotels and lodging places from laundries and other professional services due to data limitations in the 1963–1967 CPS. Therefore, there is some measurement error in our identification of who is subject to the GHRs. Throughout the primary analysis, we restrict attention to people employed in either manufacturing or mercantile industries based on our coding.¹⁴

Another issue in working with the CPS is the definitions of the employment and earnings variables. First, annual wage and salary incomes are reported for the previous calendar year while weekly hours are reported for the week before the survey was taken. In addition, weeks worked are reported in intervals.¹⁵ Taken together, this makes it difficult to accurately measure weekly or hourly earnings using the CPS. In our first approach to this problem, following Bailey, DiNardo and Stuart (2021), we simply divide the annual wage earnings for year t − 1 by the product of the midpoint of the interval for the number of weeks worked in year t − 1 and hours worked in the reference week of year t. As noted in Bailey, Helgerman and Stuart (2024), the hourly wage imputed in this way matches quantiles of reported hourly wages from the CPS Outgoing Rotation Group data above the minimum wage reasonably well. Because annual wage and salary income is reported for the previous calendar year while weekly hours refer to the survey reference week, our imputed hourly earnings measure combines information from different reference periods and uses intervalled weeks worked. We therefore report both imputed hourly earnings and annual wage-and-salary income. Annual earnings uses reported wage-and-salary income directly, while hourly earnings divides annual earnings by reference-week hours and the midpoint of the reported weeks-worked interval.

One final problem with the CPS data is that the publicly available data in many cases only identifies the state group rather than the state in 1962 and from 1968 to 1976, an important period for the repeals of GHRs. States were placed into groups that were relatively small and geographically close to each other (Derenoncourt and Montialoux, 2020). Nevertheless, this creates a challenge in our empirical analysis by making it difficult to identify whether a person living in a particular state group is actually treated, since several state groups include states with and without a GHR at any given point in time. For example,

Kentucky and Tennessee were grouped together from 1973 to 1976. The former repealed its GHR in 1974, while the latter never lifted its GHR during the sample period. As a result, between 1975 and 1976, we would not know whether a person living in a particular state group was actually subject to a GHR. Our preferred specifications therefore use the states that are consistently identified at the state level throughout the analysis windows: California, Connecticut, the District of Columbia, Florida, Illinois, Indiana, New Jersey, New York, Ohio, Pennsylvania, and Texas. In the appendix, we show that the results are robust to using the unbalanced panel of states and years. We also show that the results are robust to aggregating to the state-group level using the share of the population within a state group with a repealed GHR as the treatment variable.

Our primary empirical specification estimates the effect of state-level actions that invalidated a GHR using a difference-in-differences framework:

yit+1 = β Repealst + Controlsit + εit+1 , (7) where i indexes individuals, s states in the consistently state-identified CPS sample, and t years. We include individual-level demographic controls in the vector Controlsit including the linear and quadratic forms of age, educational attainment, marital status, and race as well as different sets of fixed effects. In our preferred specification, we include state-by-occupation-industry fixed effects and year-by-occupation-industry fixed effects. We restrict attention to people between 25 and 55 years old working in the private sector who were not self-employed, not working on a farm, and not working without pay for the family. Regressions are weighted using the CPS sample weights following Lemieux (2006). Because the preferred sample contains only eleven state clusters, we supplement conventional state-clustered standard errors with wild-cluster-bootstrap inference, which can provide more reliable finite-sample inference when the number of clusters is small (Cameron and Miller, 2015). Wild-cluster-bootstrap p-values for the coefficients of interest are reported in brackets below the conventional clustered standard errors.

We also control for changes in state-level labor policies. One such policy was requirements for overtime compensation. Even though these overtime rules applied uniformly to men and women, they might have had different effects for men and women given the differences between their average workweeks. By 1967, fifteen states and Washington, D.C., had laws or regulations (typically as a part of the minimum wage program) that provided for overtime compensation. These laws generally required the payment of premium rates for hours worked in excess of a daily or weekly standard. Such premiums acted as a deterrent to long hours and a nudge toward more equal workweeks across workers. As mentioned earlier, we also control for other gender-specific labor regulations like nighttime work restrictions.

The variable Repealst is an indicator for whether state s had repealed its GHR through any type of action by time t. For example, in Kentucky, there was a court decision striking down the GHR, an administrative ruling against it, and a legislative repeal. The earliest action was the court decision in 1971, so in this case, Repealst is coded as one starting in 1971.¹⁶ Because the date of repeal is the date of the action and not the actual date of implementing the repeal, the outcomes are one year in the future to account for a lag in implementation. Additionally, key CPS variables differ in their reference periods: workweek length corresponds to the total number of hours worked in the previous week, whereas wage income is measured as respondents’ total pre-tax wage and salary earnings received during the preceding calendar year. Consequently, the effective sample period varies slightly across outcome measures.

We further restrict our analysis to people working in the manufacturing and mercantile sectors, which were, in general, the only sectors affected by the workweek limit and its repeal. There were a few exceptions to this general rule. For example, in Colorado and South Carolina, the repeal only applied to certain industries. In these cases, because the number of people in these states who were not in those industries was so small, we treated the repeal as applying to all industries.¹⁷ This sample selection is why our treatment variable Repealst does not vary within a state-year across occupations, for example, or across industries in a state-year. As a consequence, our preferred specification only exploits two sources of variation associated with the repeal: (1) geographic variation across states and (2) the longitudinal variation over years. The primary reason we focus on workers in these two industries is that we are concerned about general equilibrium spillovers to workers in other industries not directly affected by the workweek limits. In other words, it is not clear if the group of people working outside these affected industries should be part of the treatment or control groups. We instead consider each state as a separate labor market, while considering industries within a state as part of that single integrated labor market.

We also estimate event-study specifications:

10 yit+1 = ∑ βτ 1{t − Ts = τ} + Controlsit + εit+1 , (8) τ=−5 where Ts is the year of repeal in state s and τ indexes years relative to repeal. The coefficients βτ identify the dynamic effects associated with the removal of the GHR. The recent concerns about staggered DiD designs apply in our case (Goodman-Bacon, 2021). For example, there are the potential issues of always-treated and never-treated units. To assess whether these issues drive our results, we re-estimate the main outcomes using a variety of staggered-DiD estimators that allow treatment effects to vary across repeal cohorts and calendar years.¹⁸

Obviously, states did not repeal their GHRs at random. This raises the concern that unobserved state-level changes could drive both repeal timing and subsequent labor-market outcomes. For example, growing demand for manufacturing goods could have created pressure from business owners to lift hours limits and could also have changed hours and employment directly. As a balancing test, Table 2 relates 1962 state-level characteristics to characteristics of the GHR policy: the year in which a state repealed its GHR through any type of action (conditional on having a GHR to start with), the old GHR weekly-hours limit, and whether the state never had a GHR. We include all states in this regression not just those that make up our consistent sample.

Overall, the demographics we consider are not particularly strong predictors of these characteristics of the GHRs. We find some evidence that states with a higher share of high school graduates repealed their GHRs earlier. There is also some evidence that states with higher union density were more likely to have had a GHR in the first place. These regressions are, of course, not a proof of exogeneity, but they show which observables are correlated with repeal timing after conditioning on Census region.

Workweek

We first estimate the effect of repealing the GHR on the probability of working longer than the previous legal workweek limit. In principle, women working above the limit before the law was repealed were doing something illegal, though there are potential measurement issues that could explain apparent violations. Nevertheless, we expect the number of such women to be close to 0. In fact, about 1.4 percent of women reported working longer than the limit. Note that an increase in this fraction does not reflect simply an increase from, say, 44 to 45 hours per week. Such a change, for example, in Washington would represent an increase in the share of women working longer than the previous workweek limit, but in other states such as Mississippi with a limit of 60 hours, this change in the average workweek would not.

Table 3 shows that the probability of working over the initial GHR limit statistically significantly increased among women. Our first specification includes demographic controls and state and year fixed effects.¹⁹ The second adds occupation-industry fixed effects. The third adds state-by-occupation-industry fixed effects, year-by-occupation-industry fixed effects, and year-specific demographic controls. The final and most demanding specification also includes controls for the presence of a state-level overtime law as well as a female-specific minimum wage. The magnitude and statistical significance are stable across all specifications. Given the rarity of working over the GHR before it was repealed, the increase of around 1 percentage point for women working past the GHR hours level is economically meaningful, as the rise represents a doubling of the mean rate before the GHR. The fact that the increase was only slightly more than a percentage point highlights that the workweek limits tended to be set rather high and were not binding for most working women. For men, the point estimate is, in general, positive but becomes smaller and statistically insignificant in the richest specification with state policy controls. Moreover, prior to repeal, over 14 percent of men worked longer than the GHR (which did not apply to them), so the percentage change in the share working beyond the limit is much smaller than for women.

Table 4 shows that the repeal of the GHR was associated with an increase of roughly 0.4 to 0.9 hours per week for female workers. These effects are economically meaningful and are statistically significant in the first three specifications.²⁰ Repeal also led to a meaningful and statistically significant increase of roughly 0.4 to 0.6 hours per week for men. This is one piece of evidence that male and female labor were complements in production.

Figure 4 shows that the effect of repealing the GHR on the average workweek was almost immediate for women and slightly delayed for men. For both genders, the effect had reached basically its maximum within

2 years and remained there for another 8 years, through the end of the time period we study (though some of the effects in later years are statistically insignificant for women). We take the shift in women’s workweek so closely matching the shift for men as additional suggestive evidence for complementarity between these two types of labor.

Employment

We next focus on the employment margin. Because our sample is composed of workers initially in the manufacturing or mercantile industries, the employment variable we examine is the probability of shifting out of the affected industries into other employed industries. There is no natural individual-level specification for flows into affected industries or for total employment in the affected industries. Therefore, we are not able to directly test whether employment went up or down in the affected industries after repeal. A lower outflow rate after repeal is only suggestive evidence on what happened to overall employment.

Table 5 summarizes our estimates of the effect on transitioning out of the treated industries into other employed industries using the same specifications as in Table 3. Here we rely on the CPS question that asked respondents about their industry of work at the time of the survey and in the previous year. Because of the timing of the transition variable, we shift the time frame of our sample from 1962–1980 to 1963–1981. A negative coefficient estimate indicates smaller flows out of manufacturing and mercantile employment following a repeal. Prior to repeal, an average of 16.4 percent of women and 18 percent of men transitioned from the affected industries into another employed industry. The estimates show that the outflow rate fell by about 1.2 to 1.4 percentage points for women and by about 1.1 to 1.5 percentage points for men. Like the results for the workweek, the magnitude of the effects on the transition rate are (perhaps surprisingly) quite similar for men and women. Unlike the workweek results, the baseline transition rates for men and women are also quite similar meaning the relative magnitude of the effects is also similar.

Figure 5 shows that starting in the second year after repeal, women experienced reductions in outflows of around 1 percentage point through the end of our time period. Men also experienced reductions in outflows in all but one post-repeal year, often between negative one and negative two percentage points. We do note that only one of the female coefficients is statistically significant, and none are for males. Again, that these dynamic effects are so similar for men and women we interpret as evidence for complementarity between male and female labor. Moreover, from the viewpoint of the model, there is nothing contradictory about increases in hours and lower outflows. When the overall demand for labor rises, firms want to use both margins of workweek and employment to meet that increase.

Earnings

Before discussing the results on earnings, it is important to remember that the hourly earnings variable combines annual earnings from the year prior to the survey and the hours worked variable in the reference week. Moreover, both measures, in effect, incorporate both straight-time and overtime earnings.²¹ This means that an increase in hourly earnings does not necessarily mean the average base rate went up. It could be because people are working longer hours and, potentially, earning more overtime. We view this as a plausible interpretation of the results. Nevertheless, it is worth keeping in mind.

The effects of repealing the GHR on hourly earnings and annual earnings are shown in Panels A and B of Table 6 for the same set of specifications as before. The point estimates for women’s earnings show that repeal was associated with reductions in hourly and annual earnings. Across specifications, repeal was associated with a 4.6 to 5.7 percent drop in hourly earnings and a 4.1 to 5.8 percent drop in annual earnings. The hourly estimates are statistically significant in each specification; the annual estimates are statistically significant in three of the four specifications. For men, there is no evidence that hourly or annual earnings went up on average following repeal. Male hourly earnings estimates are modestly negative, ranging from 0.7 to 2.3 percent in magnitude, and are statistically significant in three specifications but not in the most saturated specification. The male annual earnings estimates are basically 0 in statistical and economic magnitude.

Figure 6 shows that the effects on hourly earnings for men and women are slightly delayed following repeal relative to the effects on the workweek and employment transitions. Still, within 3 years after repeal, the effects for both men and women have nearly reached their largest magnitude and remain there through the end of our time period. Again, we find it striking that the effects for men and women so closely match each other in shape, which we interpret as evidence on the integration between the labor markets for men and women. These were clearly not disconnected markets, and it would not make sense to think of men as the control group while women are treated by the repeal of the GHR.

The decline in the hourly wage for women casts doubt on the importance of the repeal of other employment protections for women. The removal of these other protections such as limits on working in more physically demanding jobs presumably made work less attractive. This would tend to depress female labor supply. Therefore, under this interpretation, we would expect the average female wage to increase following the repeal of these other protections to compensate for the less desirable job conditions. This is not what we find at all. Instead, the most straightforward interpretation of the hourly earnings as well as workweek and transition results for women is that repealing the GHR increased the supply of female labor.

The puzzle that remains is why women worked more and earned less per hour, and why they earned less overall on an annual basis. This suggests that women, on average, were worse off following repeal even though they could have presumably continued to work the same as before. One possible explanation is that repealing the GHR changed the composition of the female workforce in such a way that earnings fell and hours worked rose, on average, but no woman who worked more after repeal was irrational. We cannot directly test this explanation since we cannot follow particular women over time. The best we can do is offer indirect evidence on whether the demographic composition of women changed over time.

In results we do not report, we find limited evidence that the composition of the female workforce changed following repeal. There were no effects on whether affected-industry workers were married or had own children in the household. However, the share of female workers with a high-school diploma or more rose by 1.7 percentage points, but this increase points away from a simple story in which repeal lowered average wages by drawing in less educated women. These estimates do not rule out selection on unobserved productivity or preferences, but they show that the wage decline is not accompanied by large observable shifts along these dimensions. One might wonder if non-wage compensation is increasing and offsets the decline in total wage compensation. Unfortunately, we do not observe these other forms of compensation. Furthermore, at least from the viewpoint of the model, wage and non-wage compensation are positively correlated so theoretically, we would predict that non-wage compensation is also falling. This makes the puzzle even more difficult to resolve.

Another possible explanation is that repealing the work limits not only provided women the opportunity to work more but also to do that work when they wanted to. On this point, it was potentially the repeal of the daily rather than weekly hours limit that mattered more. Every state with a female-hours limit in 1969 except Minnesota had both weekly and daily limits. In 1969, nineteen states had 8-hour daily limits, one had an 8.5-hour limit, twelve had 9-hour limits, and nine had 10-hour limits. These limits could make a four-day, 40-hour week illegal even when total weekly hours remained below the weekly cap. If flexibility in scheduling is sufficiently valuable, women might be willing to work more even if they get paid less. ²² For example, a woman might be happy to work two 10-hour days rather than three 5-hour days, even if she ends up making a bit less overall. In other words, there is an amenity value to schedule flexibility that people are willing to pay for in the form of lower earnings. Unfortunately, the CPS data only reports weekly hours, not the daily arrangement of those hours, making it impossible for us to observe whether the scheduling of hours changed after repeal. The other problem is that the repeal of the workweek and the daily limits were a package deal, making it hard to separate out the effects of each.²³

For men, the combination of more work and little evidence of earnings gains is hard to reconcile with a plain increase in male labor demand. One possibility is that production substitution and complementarity varied across job categories, producing offsetting effects on average earnings. A related possibility is that repeal changed the organization of work rather than only the relative demand for male and female labor. Evidence from changes in statutory hours in France shows that work-time rules can affect work intensity as well as measured hours (Askenazy, 2004). In our setting, such reorganization could help explain why male hours rise without detectable earnings gains. The explanation we now explore is whether the increase in female labor supply led to an increase in the labor supply of their husbands.

How do we square the rise in male labor inputs, which suggests a rise in demand for male labor, with, if anything, declines in male hourly earnings? As a matter of simple economics, it must be that the supply of male labor also rose, but why? We hypothesize that spouses value shared leisure or coordinate work schedules, meaning that repeal-induced changes in wives’ work opportunities could shift husbands’ hours as well. While the CPS allows us to identify couples, it is not a panel, so we are not able to follow couples over time and identify how husbands’ workweeks respond to changes in wives’ workweeks. Instead, we provide suggestive evidence for such a complementarity in leisure by examining differences in the effects of repeal by marital status and industry in which the wife worked.

Table 7 reports the differences in the effect of repeal on the workweek between married and unmarried men using the same sample and specifications used throughout the analysis. This is obviously a less-thanperfect comparison since married and unmarried men differ in many other ways besides having a wife. With this caveat in mind, we find that the average workweek for unmarried men also increased following repeal (though this estimate is less stable than for the differential effect for married men). However, married men’s average workweek rose by about 0.45 hours more than unmarried men’s following repeal. This is consistent with a complementarity in spouses’ leisure.

The comparison of married to unmarried men is less than ideal since not all married men have wives who were directly affected by repeal. The women might have been working in non-covered industries or not working at all. Married men with wives in these two categories are, in effect, closer to unmarried men in terms of how affected they should be by repeal. For this reason, our second analysis restricts attention to married men and separates those whose wives worked in the covered industries of manufacturing or mercantile from other married men. Again, this is a less-than-perfect comparison since men with wives in these industries presumably differ from men who don’t along many dimensions that we cannot control for.

With this caveat in mind, Table 8 shows that, in general, married men with directly affected wives were more sensitive to the effects of repeal. In fact, married men with affected-industry spouses worked about 0.5 to 0.7 more hours per week after repeal than other married men. We admit that this effect is somewhat imprecisely estimated and imperfect since we cannot exactly examine how a particular husband’s workweek responds to the change in his wife’s workweek. Nevertheless, the results are consistent with a complementarity in spouses’ leisure leading to a rise in male labor supply, offsetting the rise in male labor demand following repeal. ²⁴

Limits on the hours worked by women were a common form of Progressive Era labor regulation. After the Civil Rights Act of 1964, courts, agencies, and legislatures began to invalidate these laws state by state. We found that repealing these laws affected labor market outcomes for women but also for men. After repeal, women were about 1 percentage point more likely to work beyond the old statutory limit, and the point estimate for men was similarly positive, though smaller and not statistically significant in our richest specification. The average workweek for women and men rose, and there were fewer transitions out of industries initially covered by the GHR. The one difference in the effects of repeal was on earnings, for which women’s hourly earnings fell by roughly 4.6 to 5.7 percent, and annual earnings fell by roughly 4.1 to 5.8 percent, while men’s earnings were more or less unchanged. These results suggest that repealing a GHR expanded the effective female labor supply, which had the effect of increasing demand for male labor because of a complementarity in production between the genders.

There are two puzzles with this interpretation. First, if male labor demand rose, why did average hourly earnings for men fall? We argue that there was also an increase in male labor supply due to a complementarity between spouses’ leisure. In support of this, we show that the effect of repeal on hours worked was larger for married than for single men. The other puzzle is why it appears that women were, on average, working more and earning less following repeal. Here we suggest that repealing a GHR, particularly the daily limit, provided additional flexibility for women scheduling their work. The value of additional flexibility offset the costs of more hours and lower pay. Unfortunately, we do not have direct evidence on how schedules changed after repeal and leave this for future work.

Taken together, our results highlight the complicated nature of the political economy of repeal. This was not a simple battle of the sexes pitting all men against all women. Some women likely valued the old limits as protection from longer schedules and lower wages, while others saw the laws as barriers to overtime, promotion, and access to better jobs. Repeal opened one set of opportunities while exposing workers in affected labor markets to new competition. That mix helps explain why the historical debate divided women, employers, unions, and policymakers in ways that do not fit a single winner-loser story.