Evaluating Workplace Mandates with Flows Versus …ecobth/Curtis_et_al-Oct2016-SEJ-final.pdf · Evaluating Workplace Mandates with Flows Versus Stocks: An Application to California

Evaluating Workplace Mandates with FlowsVersus Stocks: An Application to CaliforniaPaid Family LeaveE. Mark Curtis,* Barry T. Hirsch,† and Mary C. Schroeder‡

Employer mandates typically have small effects on wages and employment. Such effects shouldbe most evident using data on employment transitions and wages among new hires. QuarterlyWorkforce Indicators (QWI) provides county by quarter by demographic group data on thenumber and earnings of new hires, separations, and recalls (extended leaves). The QWI is usedto examine the effects of California�s 2004 paid family leave (CPFL) program, comparingoutcomes for young women in California to those for other workers within and outside ofCalifornia. CPFL had little effect on earnings for young women, but increased separations,hiring, and worker mobility.

JEL Classification: J32, J38

1. Introduction

There has been considerable attention given recently to the need for “family-friendly” work-place policies in the United States.1 Analysis of employer mandates, be they family leave, work-place safety, or health coverage requirements, depends crucially on reliable estimates of changes inworkplace wages, employment, and other outcomes. The costs of mandates are expected to beborne by employers and employees, with incidence determined by labor demand and supply elas-ticities and workers� valuation of benefits. A special case is one in which a workforce values thebenefits dollar-for-dollar and the full costs are shifted to workers. Under these circumstances,there need not be a distortion in employment or a deadweight welfare loss (Summers 1989; Gruber1994). Because mandates typically impact some groups of workers more than others, are imple-mented in some settings (e.g., states, countries) but not others, and are adopted at different times,evaluation studies often use difference-in-differences or triple-difference estimators to identify thetreatment effects of such policies (e.g., Ruhm 1998; Baum 2003).

This article examines wage and employment transitions following implementation ofCalifornia�s Paid Family Leave (CPFL) insurance program in July 2004, the first mandated paid

* Department of Economics, Wake Forest University, Winston-Salem, North Carolina 27109, USA; E-mail:[email protected].

† Department of Economics, Andrew Young School of Policy Studies, Georgia State University, Atlanta, Geor-gia 30302-3992, and IZA (Bonn), USA; E-mail: [email protected].

‡ Department of Pharmacy Practice and Science, College of Pharmacy, University of Iowa, Iowa City, Iowa52242, USA; E-mail: [email protected].

Received September 2015; accepted March 2016.1 In June 2014 there was a White House Summit on Working Families and the President wrote an op-ed on family

friendly policies (Obama 2014). The 2015 Economic Report of the President devoted a chapter to family-friendly work-place policies (Council of Economic Advisors 2015).

� 2016 by the Southern Economic Association 501

Southern Economic Journal 2016, 83(2), 501–526DOI: 10.1002/soej.12150

family leave program in the United States. The theoretical underpinnings and statistical methodsused in our analysis are similar to those used in prior studies examining workplace mandates, withone notable difference. Rather than focusing on changes in wages and employment among thestock of incumbent employees, we examine wage offers among new hires and employment flows,the latter including the number of new hires, permanent separations, and extended leaves withreturn to work. We examine changes in these outcomes following enactment of CPFL amongyoung women in California relative to young men and older women within the state, and relativeto young women and other workers elsewhere in the country. Data from the Quarterly WorkforceIndicators (QWI) (Abowd et al. 2009) are used to measure the pre-tax earnings and employmentof “stable” new hires, and provide information on separations and extended leaves, all by quarter,county, age, and sex.2

Why the focus on new hires and other labor market flows? A limitation of most studies isthat wage and employment effects resulting from workplace mandates occur gradually. Employ-ers do not instantly move to a new equilibrium employment level and/or rapidly change thedemographic composition of their workforce, nor do we expect to see substantive wage adjust-ments for an existing workforce. Although little short-run impact on incumbent employees isexpected, the effects of the policy should be quickly observed among new hires and otheremployment flows. As explained subsequently, we expect to see small wage decreases amongyoung women (the treated group) relative to other (non-treated) workers, while relative employ-ment for young women could decrease, remain constant, or increase, depending on the valua-tion of benefits and degree of cost shifting. To understand how universal paid leave affectsbehavior, we need to examine not just hiring and earnings, but also labor market outcomessuch as separations, recalls (extended leaves), and the demographic composition ofemployment.

Although our focus is on paid family leave, the implications are broader, applying to anyevent, behavior, or policy that shifts labor market demand or supply. Even were a workplace man-date to have a substantial impact, we suggest that it is difficult to measure the impact based onchanges in employment levels and average wages. A focus on new hire earnings and composition,along with employment flows, may allow researchers to detect the effects of workplace policiesshortly following their implementation.

2. Overview of California Paid Family Leave Policy

Overview/Coverage

CPFL policy was enacted August 30, 2002 and took effect July 1, 2004. Prior to the 2004implementation of CPFL, women had access to paid disability leave during pregnancy and shortlyafter birth. To understand the effect of CPFL program, one must recognize how it interacts withpre-existing programs and how multiple policies are used to receive leave that is both job protected

2 Previous analyses on mandates typically measure changes in wage and employment levels (stocks) by state and demo-graphic group using the Current Population Survey (CPS) data (e.g., see Card (1992) on minimum wages and Gruber(1994) on health insurance pregnancy coverage). Recent articles by Rossin-Slater, Ruhm, and Waldfogel (2013), Byker(2014), Das and Polachek (2015), and Baum and Ruhm (2016) use alternative data sets to examine various effects ofCalifornia�s paid family leave. The focus of these articles differs substantially from our work, as discussed below.Articles by Dube, Lester, and Reich (2016) and Gittings and Schmutte (forthcoming) use the QWI to examine the effectof minimum wages on employment flows (separations and hires).

502 Curtis, Hirsch, and Schroeder

and paid. As described below, CPFL has been typically used to extend paid leave among mothersby six weeks.

CPFL is administered by the California Employment Development Department (EDD),which also administers the State Disability Insurance (SDI) program (begun in 1977). SDI andCPFL are jointly financed by a mandatory payroll tax on employees, with no tax on employers.Both programs provide partial wage replacement. Coverage among private sector employees isnearly universal. Employees are required to participate if their employer has more than oneemployee and has paid an employee at least $100 in any quarter during a 12-month referenceperiod. Self-employed and state/local workers are not automatically enrolled, although some canelect coverage. No proof of citizenship is required.

Payroll Tax Financing

The SDI/CPFL employee tax rate and cap on total contributions have varied across years tomaintain funds to pay current benefits. As seen in Table 1, the payroll tax rate varied from 0.6 to1.2% between 2003 and 2011, while the cap on payments varied from a low of $500 in 2007 to$1120 in 2011. In 2003, prior to CPFL, the payment cap was $512. This was increased to $812 in2004. The cap rose sharply following the recession, to a high of $1120 in 2011. In 2011, the 1.2%employee SDI/CPFL contribution rate combined with a taxable wage ceiling of $93,316 to pro-duce a maximum annual contribution of $1120. The taxable wage base is adjusted, typically annu-ally, to reflect state wage growth.

SDI Wage Base and Benefit Calculation

SDI provides partial wage replacement, with benefits equal to 55% of workers� wages up to acap. The CPFL program uses the same benefit formula as SDI. Workers unable to work due to anon-work-related illness or injury, including pregnancy, may be eligible for SDI benefits. The SDIbenefit period is four weeks before the due date and six weeks postpartum for normal pregnancies,but up to eight weeks in the case of Caesarian births or other difficulties. The benefit amount iscalculated using a wage base equal to the highest paid quarter during the 12-month referenceperiod 5 to 17 months before the SDI disability claim (eligibility requires at least $300 in earningsduring the 12-month reference period). Average SDI pregnancy claim benefits in FY 2011 were$398 per week, lasting on average 10.7 weeks (time before and after birth). The 2011 benefitsranged from a floor of $50 to a ceiling of $987 per week. Because of SDI, women had access topaid maternity leave decades before CPFL was enacted.

CPFL Description

CPFL was created for mothers (or fathers) to bond with their newborns, although it also pro-vides benefits to workers to care for a seriously ill child, spouse, domestic partner, or for a newlyadopted child or recently placed foster child.3 California was the first state to provide paid family

3 In FY 2011, 87.3% of CPFL claims were for care of newborns. Effective July 1, 2014, the CPFL temporary disabilityprogram was expanded to include time off to care for a seriously ill grandparent, grandchild, sibling, or parent-in-law.Bartel et al. (2015) examines the increased use of CPFL among fathers.

Evaluating Paid Family Leave Flows 503

Table 1. Descriptive Statistics on California State Disability Insurance (SDI) and PaidFamily Leave (PFL)

SDI/PFL Claims andBenefits FY 2005 FY 2006 FY 2007 FY 2008 FY 2009 FY 2010 FY 2011

Total SDI pregnancy claimspaid

172,623 175,194 183,013 189,139 181,685 169,957 168,593

SDI claimstransitioning to PFLbonding claims

108,818 115,392 119,442 111,024 127,529

Estimated PFL/SDI share 0.655 0.631 0.636 0.614 0.655Average weekly

benefit, SDIpregnancy claims

$354 $368 $382 $397 $398

Average weeks per SDI preg-nancy claim

11.97 10.43 10.43 10.50 10.70

Average weeklybenefit, PFL claims

$409 $432 $441 $457 $472 $488 $488

Average weeks per PFLclaim

4.84 5.32 5.37 5.35 5.39 5.37 5.30

Total PFL claims filed 150,514 160,988 174,838 192,494 197,638 190,743 204,893Total PFL claims paid 139,593 153,446 165,967 182,834 187,889 180,675 194,777Total PFL benefits paid* $300.42 $349.33 $387.88 $439.49 $472.11 $468.79 $498.44% of PFL claims filed for

bonding87.7% 87.8% 87.6% 87.6% 88.8% 87.8% 87.3%

# of bonding claims filed bywomen

109,566 112,631 119,893 129,986 132,958 123,632 128,774

% of bonding claims filed bywomen

83.0% 79.7% 78.3% 77.1% 75.8% 73.8% 72.0%

SDI/PFL rules CY 2000 CY 2001 CY 2002 CY 2003 CY 2004 CY 2005 CY 2006Contribution rate 0.65% 0.70% 0.90% 0.90% 1.18% 1.08% 0.80%Taxable wage ceiling $46,327 $46,327 $46,327 $56,916 $68,829 $79,418 $79,418Maximum worker

contribution$324 $324 $417 $512 $812 $858 $635

Maximum weeklybenefits

$490 $490 $490 $603 $728 $840 $840

CY 2007 CY 2008 CY 2009 CY 2010 CY 2011Contribution rate 0.60% 0.80% 1.10% 1.10% 1.20%Taxable wage ceiling $83,389 $86,698 $90,669 $93,316 $93,316Maximum worker

contribution$500 $693 $997 $1,026 $1,120

Maximum weekly benefits $882 $917 $959 $987 $987

*Dollar amounts are in millionsSource: Data were compiled by authors from data provided on the website and by an analyst at the State ofCalifornia, Employment Development Department. Some but not all of these figures could be updated beyond 2011.


leave, but two others have followed with similarly structured programs.4 CPFL funds are adminis-tered jointly with SDI, employees covered by SDI also being eligible for CPFL benefits. Followingreceipt of six to eight postpartum weeks under SDI, a new mother is then eligible for up to sixadditional weeks of paid family leave using the same wage base and benefit formula describedabove for SDI. In FY 2011, the average CPFL payout was $488 a week for 5.3 weeks.

Job Protection vs. Paid Leave

Although providing partial pay replacement, neither SDI nor CPFL provides job protection.Job protection is in turn provided by state and federal laws guaranteeing unpaid leave. A combina-tion of SDI, CPFL, other state programs, and the federal FMLA provides workers with a“package” of protected leave with partial wage replacement. And of course some employers maychoose to provide paid maternity leave independent of any legal requirements.5 The most generousmandated package includes up to 28 weeks of job protection (up to 16 weeks of pregnancy disabil-ity covered by the state PDL concurrent with FMLA, plus 12 weeks protection from the CFRApostpartum) and 16-18 weeks of partial wage replacement (4 weeks pregnancy and 6–8 weekspostpartum under SDI, plus six weeks from CPFL). Figure 1 shows what is typically the maxi-mum use of wage replacement (top half of figure) and job protection (bottom half) before andafter child birth.

Both SDI and CPFL are available to covered employees, but some workers claim benefitsfrom only one of the programs. Receipt of CPFL but not SDI benefits may result from house-hold financial constraints or company-provided time off during pregnancy and/or after birth.Among those who receive SDI pregnancy benefits, many do not elect to receive further benefitsunder CPFL once the SDI benefits are exhausted. In 2011, 65.5% of beneficiaries receivingSDI transitioned to CPFL. The decision not to claim CPFL payments can occur if somewomen prefer to return to work or feel a financial need to receive full rather than partial pay.Workers in small companies (fewer than 50 employees) are not covered by the FMLA�s jobprotection provision and may risk loss of their job with a lengthy maternity leave. Even absentrisk of job loss, a new mother may choose to return to her job if her employer is highly depend-ent on her contribution.

In short, the main effect of CPFL was to extend the availability of paid maternity leave by sixweeks. Given the incremental nature of CPFL, identifying its impact using standard methods (i.e.,measuring changes in wage and employment levels) is likely to prove difficult. Focusing instead onnew hire wages and employment flows enhances chances of an informative analysis.

4 New Jersey passed PFL in May 2008, began collecting taxes in January 2009, and began disbursements in July 2009.Rhode Island�s Temporary Caregiver Insurance (TCI) Law, which began January 2014, provides four weeks of paidleave (with job protection) for bonding with a new child or for family or household member with a serious health con-dition. Washington passed a PFL bill in May 2007, planning to begin payouts in October 2009 and subsequently post-poned to October 2012 and then October 2015. In 2013, legislation was passed that delays implementation until thelegislature approves funding and program implantation (unlike the other three states, this program was to be fundedthrough the state budget rather than employee payroll taxes).

5 National Compensation Survey (NCS) data from the BLS show that the overall coverage of paid family leave in theU.S. private sector was 11% in 2012, compared to just 2% in 1992–1993 (the latter figure is exclusively for maternityleave), with higher coverage for full-time workers and those in large establishments (Van Giezen 2013). For earlier esti-mates of paid maternity leave compiled from CPS supplements, see Klerman and Leibowitz (1994). Byker (2014) pro-vides a nice discussion, coupled with nation-wide evidence on paid leave compiled from several years of data from theSurvey of Income and Program Participation (SIPP).


3. Studies on California Paid Family Leave and the Labor Market

We are aware of five studies (some unpublished) that use household data to analyze vari-ous effects of CPFL on labor market outcomes.6 Espinola-Arrendondo and Mondal (2010)examine CPFL employment effects using the March 2001–2007 Current Population Survey(CPS). They compare female employment changes in California following CPFL relative tochanges for women in other states with and without expanded FMLA provisions. Using numer-ous combinations of treatment and comparison groups, the authors conclude that all theirtreatment estimates are “both economically and statistically insignificant.” One possibility isthat the effects of CPFL are close to zero. Another is that CPFL effects cannot be readilyobserved in the CPS because of small sample sizes of treated employees and the long timeperiod required for changes in wage and employment levels to be detectable.

Rossin-Slater, Ruhm, and Waldfogel (2013) has as its focus the effect of CPFL on time offfrom work among young mothers with children. Their principal data source is the March CPS.Although the authors face difficulties in identifying those who are and are not treated by CPFL(time of a child�s birth cannot be precisely measured), they provide convincing evidence thatCPFL increased time off from work among mothers of young children. Although not the princi-pal focus of their article, the authors provide estimates of earnings and employment effects ofCPFL. They detected no changes in employment following CPFL, but there appeared to be anincrease in work hours (hours last week and in the prior year), conditional on employment. Theauthors note that future study is needed.

Byker (2014) examines the effects of paid family leave in California and New Jersey on wom-en�s labor force interruptions following birth of a child, using monthly longitudinal data from the

Figure 1. Timeline for maximum use of disability insurance and paid family leave in California.Notes: See text for discussion and greater detail. Acronyms shown are: SDI, California State Disability Insur-ance; CPFL, California Paid Family Leave; PDL, California Pregnancy Disability Leave; CFRA, CaliforniaFamily Rights Act; FMLA, Family Medical and Leave Act (federal).

6 There is a larger literature examining the effects of paid family leave outside the United States, some of it focused onwomen�s labor supply and some on health, education and other outcomes for mothers and children. Articles by Bakerand Milligan (2008, 2010) and Baker, Gruber, and Milligan (2008) use Canadian data and focus on mothers� employ-ment and early child development outcomes. Effects on child well-being are generally small. Using German data,Dustmann and Sch€onberg (2012) focus on child outcomes, while Sch€onberg and Ludsteck (2014) examine mothers�labor market outcomes. Using changes in laws governing parental leave in Austria, Lalive and Zweim€uller (2009)examine the effect of leave extensions on fertility and return to work, while Lalive et al. (2014) focus on mothers� workcareers and differentiate the effects from benefits versus those from job protection. Dahl et al. (2013) provide a criticalassessment of recent expansions in the length of paid maternity leave in Norway, concluding that the expansionincreased the length of leave but was costly and regressive, while having minimal effects on a range of labor market andchild outcomes. Carneiro, Løken, and Salvanes (2011) provide a more positive assessment of the Norwegian system�slong run impact on children�s subsequent education and earnings, although they do not compare benefits to costs.


Survey of Income and Program Participation (SIPP). Data from the two treatment states are usedjointly, with New York, Florida, and Texas forming the comparison group. She finds little effectfrom paid family leave on job attachment among college-educated mothers, those most likely tohave employer-provided paid leave absent a legal mandate. For mothers with less than a collegedegree, she concludes that paid leave reduces exits lasting less than six months, while having littleeffect on longer exits. Using public-use SIPP files, Byker observes flows into and out of employ-ment, but does not know if employment is with the same employer.

Baum and Ruhm (2016) use the National Longitudinal Survey of Youth (NLSY97) to exam-ine CPFL effects on use of leave surrounding child birth and subsequent labor market outcomes.They conclude that an average mother�s use of leave increased by about 2.4 weeks, typically atabout the time that disability benefits were exhausted. Fathers took a short amount of time offimmediately following birth. Baum and Ruhm find increased work probabilities for mothers nineto twelve months after birth and increased weeks and hours worked (and possibly wage increases)in the child�s second year of life.

Das and Polachek (2015) use CPS data aggregated to the state level and use difference-in-differences techniques to identify CPFL effects on labor force participation (LFP) and unemploy-ment among young women in California. They conclude that CPFL increases LFP among youngwomen, but also has the unintended effect of increasing their unemployment and unemploymentduration. Similar tests based on placebo laws were generally insignificant, strengthening theauthors� confidence that the results are robust and causal.

Although the focus and approaches by these authors differ, Das and Polachek (2015),Baum and Ruhm (2016), and (to a lesser extent) Byker (2014) each conclude that CPFLincreased labor force attachment among young women. Their results can be interpreted asbroadly consistent with the evidence we present on job flows. We find CPFL to be associatedwith higher separations among young women, but it also leads to increases in new hires andpossibly rates of recall (return to the same employer following time off a payroll for at leastthree months). Our interpretation of such evidence is that universal paid family leaveincreased the mobility of young women (i.e., reduced job lock) and led to efficiency-enhancing resorting in the labor market. Although the increased churn among young womenmay have increased unemployment rates (Das and Polachek 2015), we find no evidence of adecrease in their employment.

4. Expected Effects of CPFL on Wages, Employment, and Turnover

The costs of CPFL are nominally borne by employees through a payroll tax. The costsare attached to all employees, but with lower average costs per hour and zero marginal costamong those with earnings above the tax threshold (about $93,000 in 2011, an amountexceeded by few young workers). The payroll tax is independent of whether a worker is likelyto use and/or values paid family leave. Because the payroll tax on workers is levied at nearlyall California establishments, the costs to workers cannot readily be shifted to employers and/or consumers in those product markets where output prices are determined nationally orinternationally.

Apart from the payroll costs paid by workers, employers face “disruption costs” resultingfrom time off the job among employees. Leave taking reduces output and/or requires added hiring.Increased uncertainty as to whether and when a worker will return also adds cost. Such


uncertainty existed prior to CPFL, but six weeks of additional leave could increase the uncertainty.That said, the employer survey conducted by Applebaum and Milkman (2011) concluded thatafter more than five years� experience with paid family leave, the vast majority of employersreported it had a minimal effect on business operations.

The general equilibrium wage and employment effects resulting from CPFL can be eval-uated using a demand and supply “tax incidence” approach (Summers 1989). The costs of paidfamily leave places a “wedge” between labor demand (and the gross wage to which employersrespond) and labor supply (the net wage to which workers respond). To the extent that market-level labor supply is less elastic than demand, more costs are shifted to employees. The statutorypayroll cost facing employees from CPFL shifts labor supply upward for all workers. The valua-tion of benefits by young women (or others) then shifts labor supply outward. “Disruption” costsfacing employers cause a downward shift in demand for young women.

These demand and supply shifts are shown in Figure 2, separately for young women (thetreated group) and for other (non-treated) workers, with the assumption that the two groups ofworkers are imperfect substitutes (discussed subsequently).7 The non-treated “other workers”have a small increase in wages and decrease in employment due to the supply shift resulting fromthe payroll tax. For young women, wages unambiguously decrease as long as the valuation of leaveexceeds their payroll costs (i.e., if S2 is to the right of S1).8 A decrease in demand due to disruptioncosts (a shift from D1 to D2) further reduces wages for young women. Employment can rise or fallfrom the pre-mandate level, depending on the size of the supply increase and demand decrease.Relative employment of young women may increase or decrease.

It is worth noting that the effects of a PFL mandate are not independent of how financing isstructured. CPFL benefits are paid through a state agency funded by a mandatory payroll tax.Thus, CPFL financing costs to firms are independent of use given that payroll taxes are not experi-ence rated. Absent scheduling and productivity “disruption” costs that may accompany longerleaves, employers have no economic incentive to select employees who are less or more likely tocollect paid leave from the state fund. If a PFL program either mandated employers to directlyprovide and fund paid leave for their employees or the state payroll tax was fully experience rated,cost to businesses and their employees would differ according to the frequency of use. All else thesame, employers would prefer to hire workers least likely to use paid leave.

Given the funding mechanism and absent disruption costs, CPFL affects employers� choiceof employee mix only to the extent that it produces changes in relative market wages. As seen inFigure 2, labor supply shifts inward for all workers due to the payroll tax costs, while shifting out-ward for workers based on their valuation of CPFL benefits. The resulting shift in aggregate laborsupply is indeterminate. We subsequently show that even with full shifting of payroll tax costs toyoung women, the implied wage change is far too small to detect. Substantive negative wage

7 Throughout the article we use the term “young women” and “treated” synonymously. To the extent that workers otherthan young women value the availability of paid leave, our methods will slightly understate causal effects. The propor-tion of paid family leave taking for bonding with children by women has fallen from over 80% female during the yearsused in our analysis to about 70% currently (see Table 1). We do not have data on how duration of PFL differs amongmale and female recipients. Evidence in Baum and Ruhm (2016) indicates that father�s leave time is brief and immedi-ately follows a child�s birth. Bartel et al. (2015) examine fathers� use of CPFL.

8 For simplicity, we show a fixed dollar shift from S1 to S20. Because the payroll tax is proportional to earnings but

capped at a maximum, labor supply would rotate upward and resume its original slope at the maximum. Qualitativeoutcomes match those shown in Figure 1.


effects require sizable disruption costs (the decrease from D1 to D2) and a high valuation of paidleave benefits (the increase from S1

0 to S2).Although unlikely, it is useful to state the implications of a unified labor market wherein

young women and other workers are perfect substitutes. In contrast to Figure 2, a unified marketwould have a single aggregate labor supply curve that is the horizontal sum of supply among allworker groups. Thus, there should be no wage difference (i.e., no payroll cost shifting) betweenequally productive young women and other workers, absent a downward shift in labor demandfrom family leave disruption costs.

In addition to examining wage and employment effects, we examine evidence on separationsand extended leaves (referred to as “recalls” in our data set). Doing so provides a broader pictureof how paid leave affects labor market outcomes. Such evidence can strengthen (or weaken) confi-dence in empirical evidence since labor market flows are not independent of each other. For exam-ple, if paid family leave has an impact on separations, hiring should be similarly affected. Onepossibility is that longer leaves may prevent what would otherwise be quits (from the employer per-spective, it is not clear whether a long leave or a quit is more disruptive). A second possibility isthat universal PFL should reduce job lock and increase job mobility among young women. Evi-dence of reduced job lock and increased churn would provide support for a standard argumentmade for mandated benefits—that such policies may correct market failures due to asymmetricinformation and adverse selection (Summers 1989).

Starting from an equilibrium in which job matches and equilibrium wages are deter-mined based in part on company leave policies, the introduction of universal paid leaveunambiguously adds churn to the labor market following its introduction. Wages at firmsproviding paid leave prior to the mandate are initially too low; wages at firms that had notprovided leave are too high. Worker turnover should increase as worker sort on wages andattributes other than paid leave (sorting on paid leave would continue among firms withbenefits exceeding the mandate). In response to implementation of CPFL, we should see rel-atively higher levels of separations and new hires for workers who highly value paid leave.Long-run turnover rates for young women may remain higher than rates prior to the man-date. With paid family leave now universal within California, search frictions are lower (i.e.,it is easier for young women to find a good match) and equilibrium levels of job churn arelikely to be higher.

Figure 2. Wage-Employment Effects of CPFL in California with Separate Markets for Young Women andOther Workers.


5. How Large an Effect Might CPFL Have on Wages?

In the previous section, we discussed why the payroll costs of CPFL should be shifted toyoung women if they are imperfect substitutes with other workers. To assess the plausibility of ourestimates of relative wage changes, we ask the question: What effect might CPFL have had onwages if the full costs were borne by young women? A back-of-the-envelope calculation is inform-ative. Although there need not be full shifting, such a calculation provides an upper-bound on theexpected wage effect from the CPFL payroll tax.

Ideally we would like to account for the full costs of the CPFL program, including disruptioncosts. Our calculation incorporates only the direct costs of CPFL leave benefits, funded from pay-roll taxes. We observe payroll tax rates and revenues collected to fund the system. Hence we havereasonably good information on the direct cost of paid leave based on CPFL expenditures and rev-enues, payroll tax rates, and costs as a percentage of (taxable) earnings.

As seen in Table 1, the overall payroll tax rate for the state disability program has been about1.0%, but most of this is used to fund state disability programs other than paid family leave. Theinitial increase in the tax rate that accompanied the introduction of CPFL was about 0.3%, but alarge portion of this was used to start up the CPFL administrative structure. Longer run, paid fam-ily leave benefits account for a small share of total benefits of the combined SDI/PFL fund, 11.1%in 2012. In our calculation shown below, we assume a payroll tax cost of CPFL of 0.2%, anamount halfway between the 0.3% rise seen during the years of our analysis and the 0.11% of pay-roll needed to currently fund CPFL benefits.

To what extent would nominal wages need to decrease for young women and increase forother workers to fully shift the payroll tax burden? Letting C be the total payroll tax cost for aworkforce, Y the total taxable earnings for that workforce, t the administrative payroll tax rateused for CPFL, and Pf and (1 2 Pf) the shares of taxable payroll for young women and others (thetreated and non-treated), respectively, the total payroll cost across a workforce would be:

C5tY5Pf tYð Þ1ð12PfÞ tYð Þ:

We wish to solve for the percentage reduction in relative earnings required to load all costs C ontoyoung women. We designate this “tax” rate as tf, which collapses to the simple relationship:

tf5t=Pf ;

where t is the statutory tax rate and Pf the share of taxable earnings among young women.As stated above, we set t at 0.2% (0.002), an amount lower than the initial funding cost but

below the long-run cost. Using Current Population Survey (CPS) data for California in the twoyears prior to CPFL, we calculate the share of taxable payroll among young women. We obtain anestimate of 21.5%.9 Thus, the implied relative wage effect from full shifting would be t/Pf 5 0.20/0.215 5 0.93, or roughly one percent. Using instead the initial 0.3% tax cost, full shifting wouldhave resulting in a roughly 1.4% differential. Had we used the 2012 tax cost of 0.111%, full shiftingwould result in a differential of only a half of one percent. A wage gap between young women and

9 Using CPS data, we exclude each individual�s earnings above the taxable cap from the denominator in calculatingyoung women�s share of taxable payroll. Absent the exclusion, the share of young women�s earnings to total payroll is12.1%, as compared to 21.5% of taxable payroll. Using our QWI administrative earnings data, we obtain an estimated10.9% share of young women�s earnings to total payroll, similar to the CPS estimate.


others in the California labor market of, say, 1.0% could arise from a combination of young wom-en�s wages falling 0.8% and others� wages rising 0.2%. Were one comparing young women inCalifornia relative to young women (or others) outside of California, the differential would becloser to 0.8% rather than 1.0% since the comparison group is not levied the payroll tax.

Such back-of-the-envelope estimates are imprecise, but provide an approximation of thewage effects that might result from CPFL absent disruption costs. The suggestion is that relativewage effects resulting from the direct cost of CPFL (the payroll tax) should be small—one percentor less with full shifting. This amount provides an upper bound (ignoring disruption costs). Ifyoung women and other workers are close substitutes, relative wages change even less.

Causal wage effects on the order of one percent or less are nearly impossible to identify withstandard data sets, in particular if we are looking at wage levels (rather than new hire wages) andusing data sets based on relatively small samples. Even with our data set, which provides nearlyuniversal administrative earnings records for new hires by gender, age group, county, and quarter,obtaining reliable estimates of such small wage effects is difficult.

6. Data Description: The Quarterly Workforce Indicators

The earnings and employment flow variables central to our analysis are obtained from theQWI database. The QWI is publicly available data derived from the Local Employment Dynamics(LED) data program, which in turn is built on the confidential Longitudinal Employer-Household Dynamics (LEHD) program. The LEHD is based on state unemployment insurancedata and contains individual level quarterly earnings data that matches workers to firms. TheLEHD identifies when workers begin at a new firm and records their earnings. The data rely onstate participation and while all states now participate, five did not provide complete data over ourprimary period of analysis, which begins in 2002.10 The QWI provides employment and earningsmeasures at the state, metropolitan statistical area (MSA), and county levels. Based on individual-level LEHD data, these measures are aggregated into narrowly-defined demographic categoriesincluding age, sex, ethnicity, race and education within the geographic area. The data cover 98%of all private, non-agricultural wage and salary employment in the states for which data are avail-able (Abowd et al. 2009). Data cells with very low populations are suppressed.

In the analysis that follows, we utilize measures of the average monthly earnings for new hires ina quarter, and the number of new hires, separations, and recalls, within tightly defined sex-age group-ings, all observed at the county-by-quarter level.11 We examine these outcomes both in levels and inshares for young women. In primary results shown, we use data for 2002:3 through 2004:2 as thepre-CPFL period and 2004:3 through 2006:2 as the post-treatment period. Thus we have the samenumber and composition of quarters before and after implementation of the law in July 2004. Exam-ination of the data suggested a limited effect of the policy between passage and implementation. Asshown in Appelbaum and Milkman (2011), following passage of the law, Californians had a low rec-ognition of the law�s existence and content. Recognition grew over time, particularly among those

10 Data for California is available starting in 1991. In the analysis that follows, five states are excluded from analyses.Massachusetts provided no data during our period of analysis, while data for Arizona, Arkansas, Mississippi, andNew Hampshire were provided for some but not all quarters.

11 Age groupings identified in the QWI are 14–18, 19–21, 22–24, 25–34, 35–44, 45–54, 55–64, and 65–99. We do not usecells by education, race, or ethnicity since many cell would be suppressed. Demographic attributes change little overtime, while state and county fixed effects account for cross-sectional differences.


most likely to use it. To the extent that effects of the policy began prior to implementation, our esti-mated policy effects are understated, as seen in subsequent analysis wherein our control period ismoved back prior to adoption (quarters 2000:3–2002:2). A downside to reaching back to earlieryears is that the “dot-com bust” that followed the 1997 through early 2000 “dot-com bubble” mayhave affected relative earnings and employment among young women in California differently thanin other states (Dot-com bubble, n.d.). In addition, California state minimum wage increasesoccurred in January 2000 and 2001, but not in the period between passage and implementation.

The unit of analysis is at the demographic-location-quarter level where demographicgroups are defined by sex-age group categories and location is at the county level. These dataallow us to measure average monthly earnings of new employees for the first full quarter inwhich they are employed. We are able to distinguish between all new hires and all new “stable”hires, where stable hires are defined as employees who have worked at least a full quarter at thefirm where they were hired, as evidenced by their presence on that firm�s unemployment insur-ance records for three consecutive quarters. Our analysis includes employment and earningsdata only for stable hires. Among other things, the focus on stable hires largely avoids includinghires of temporary replacement workers at non-representative wages.

The narrowly defined demographic and geographic groupings over time in the QWI arewell suited to examine treatment effects from CPFL policy. If CPFL affects employment andearnings, we expect this to be most evident in relative new hire employment and new hire earn-ings among young women in California. The QWI panel allows us to examine changes thatoccurred following CPFL among young female treatment groups in California compared tochanges for other demographic groups within California, as well as compared to young womenand other demographic groups outside California.

In order to provide some feel for the QWI data, Table 2 shows average new hire monthly earn-ings (pre-tax), and the average monthly number of stable new hires, separations, and recalls (i.e.,extended leaves in which individuals observed on a firm�s payroll in quarter t were not there inquarter t 2 1, but had been on that firm�s payroll in quarters t 2 2, t 2 3, or t 2 4). Our four meas-ures are shown for young women (ages 19-34) and other demographic groups in California and fora group of all states other than California with complete QWI data during these years (Arizona,Arkansas, Massachusetts, Mississippi, and New Hampshire are excluded). For each of these fouroutcomes we also provide relative (or share) measures for young women. Specifically, we show theratio of young women�s new hire earnings to average earnings for all new hires, and the shares ofall new hires, separations, and recalls who are young women. These latter four measures are shownfor both the periods before and following implementation of CPFL.

Focusing first on the change in log earnings among new hires, we see that both in Californiaand other states, new hire earnings among young women grew somewhat more slowly than forother groups. For example, in California, the change in real earnings was 1.8%, similar to that foryoung men (2.4%) but less than the 4.5 among older women and 4.7% among older men.12 Newstable hires among young women in California increased by nearly seven percent between the twoperiods, as compared to 3% for young men and 4–5% for older women and men. Moreover, newhire earnings in California grew at a considerably faster rate than outside the state for all

12 In the article, we refer to the change in the log of mean earnings as the percentage change. It measures a percentagechange in earnings with an intermediate base in the denominator and has the advantage of being invariant to thebase. Of course, the difference in the log of mean earnings is not identical to the difference in the means of logearnings.


Tab

le2.

Des

crip

tive

Evi

denc

eon

QW

IN

ewH

ire

Ear

ning

s,E

mpl

oym

ent,

Sepa

rati

ons,

and

Rec

alls

,P

re-

and

Post

-CP

FL

New

Hir

eE

arni

ngs

(mon

thly

)N

ewH

ires

Sepa

rati

ons

Rec

alls

Pan

elA

Pre

-CP

FL

Post

-CP

FL

Log

Dif

fP

re-C

PF

LPo

st-C

PF

LL

ogD

iff

Pre

-CP

FL

Post

-CP

FL

Log

Dif

fP

re-C

PF

LPo

st-C

PF

LL

ogD

iff

Cal

iforn

iaY

oung

wom

en$2

,104

$2,1

590.

0175

1,84

5,79

52,

031,

023

0.06

621,

932,

557

2,09

7,88

70.

0519

226,

268

222,

909

20.

0434

You

ngm

en$2

,635

$2,7

190.

0238

2,04

9,89

92,

194,

502

0.03

102,

156,

837

2,23

0,92

82

0.00

4627

5,43

326

0,81

62

0.09

51O

lder

wom

en$2

,663

$2,7

980.

0452

1,47

1,21

81,

566,

733

0.04

861,

855,

683

1,91

2,75

60.

0150

340,

356

318,

617

20.

0738

Old

erm

en$4

,167

$4,3

830.

0472

1,74

4,36

71,

849,

186

0.04

132,

314,

796

2,34

2,63

42

0.00

5145

1,58

843

9,55

22

0.04

95A

llw

orke

rs$2

,879

$2,9

890.

0307

7,11

1,27

97,

641,

444

0.04

358,

259,

873

8,58

4,20

50.

0096

1,29

3,64

51,

241,

894

20.

0631

“All”

stat

esex

cept

CA

You

ngw

omen

$1,8

23$1

,851

0.01

0212

,615

,073

14,1

91,1

640.

0622

13,9

12,5

9315

,139

,958

0.02

472,

027,

650

1,94

2,78

32

0.07

51Y

oung

men

$2,4

53$2

,506

0.01

4613

,105

,511

14,7

66,7

980.

0558

14,4

76,1

1515

,547

,222

0.00

542,

419,

342

2,29

5,37

22

0.09

55O

lder

wom

en$2

,342

$2,4

120.

0263

10,0

97,3

6811

,458

,987

0.06

7213

,689

,056

14,5

21,1

090.

0007

3,10

5,35

22,

967,

092

20.

0674

Old

erm

en$3

,981

$4,0

860.

0247

11,0

19,3

7112

,566

,936

0.06

9515

,622

,413

16,3

03,9

782

0.01

363,

849,

258

3,68

6,24

52

0.07

03A

llw

orke

rs$2

,619

$2,6

850.

0218

46,8

37,3

2352

,983

,885

0.06

2357

,700

,177

61,5

12,2

672

0.00

0511

,401

,602

10,8

91,4

922

0.07

34

Rel

ativ

eN

ewH

ire

Ear

ning

sN

ewH

ire

Shar

eSe

para

tion

sSh

are

Rec

all

Shar

e

Pan

elB

Pre

-CP

FL

Post

-CP

FL

Dif

fP

re-C

PF

LPo

st-C

PF

LD

iff

Pre

-CP

FL

Post

-CP

FL

Dif

fP

re-C

PF

LPo

st-C

PF

LD

iff

Cal

iforn

iaY

oung

wom

en0.

7308

0.72

242

0.00

840.

2596

0.26

580.

0062

0.23

400.

2444

0.01

040.

1749

0.17

950.

0046

You

ngm

en0.

9154

0.90

962

0.00

580.

2883

0.28

722

0.00

110.

2611

0.25

992

0.00

120.

2129

0.21

002

0.00

29O

lder

wom

en0.

9249

0.93

610.

0112

0.20

690.

2050

20.

0019

0.22

470.

2228

20.

0018

0.26

310.

2566

20.

0065

Old

erm

en1.

4476

1.46

630.

0187

0.24

530.

2420

20.

0033

0.28

020.

2729

20.

0073

0.34

910.

3539

0.00

49“A

ll”st

ates

exce

ptC

AY

oung

wom

en0.

6961

0.68

932

0.00

680.

2693

0.26

782

0.00

150.

2411

0.24

610.

0050

0.17

780.

1784

0.00

05Y

oung

men

0.93

660.

9333

20.

0033

0.27

980.

2787

20.

0011

0.25

090.

2527

0.00

190.

2122

0.21

072

0.00

14O

lder

wom

en0.

8943

0.89

830.

0039

0.21

560.

2163

0.00

070.

2372

0.23

612

0.00

120.

2724

0.27

240.

0001

Old

erm

en1.

5202

1.52

200.

0018

0.23

530.

2372

0.00

190.

2708

0.26

512

0.00

570.

3376

0.33

850.

0008

Inad

diti

onto

excl

udin

gC

alifo

rnia

,th

e“A

ll”st

ates

grou

pdo

esno

tin

clud

eA

rizo

na,

Ark

ansa

s,M

assa

chus

etts

,M

issi

ssip

pi,

and

New

Ham

pshi

re.

You

ngw

omen

and

men

are

ages

19–3

4,an

dol

der

wom

enan

dm

enar

eag

es35

–65.

All

rati

osan

dsh

ares

incl

ude

valu

esfo

ral

lw

orke

rsin

the

deno

min

ator

and

valu

esfo

rth

eid

enti

fied

grou

p(e

.g.,

youn

gw

omen

)in

the

num

erat

or.

Log

diff

eren

ces

are

calc

ulat

edus

ing

the

mea

nof

the

logg

edco

unty

-by-

quar

ter

valu

esan

dno

tth

elo

gof

the

mea

ns,

cons

iste

ntw

ith

the

regr

essi

onan

alys

is.

Ear

ning

sar

ein

2010

dolla

rs.

New

hire

earn

ings

are

wei

ghte

dby

the

num

ber

ofne

whi

res

inea

chce

ll.Sh

ares

ofne

whi

res,

sepa

rati

ons,

and

reca

llsin

the

bott

ompo

rtio

nof

the

tabl

esar

ew

eigh

ted

byth

ece

llnu

mbe

rsof

new

hire

s,se

para

tion

s,an

dre

calls

.


demographic groups (overall rates being 3.1 vs. 2.2%). The results of our subsequent analysis,which indicate little relative change in earnings, but with increased hiring, separations, and recallsfor young women due to CPFL, can be gleaned to at least a limited degree from the information inTable 2.

7. Method of Analysis

As evident in the summary statistics shown in Table 2, there are three major sources of varia-tion that can be exploited to identify the impact of CPFL on young women in California—time,demographic group, and location. We begin by setting up a simple difference-in-differences (DD)model that uses demographic variation within California over time to identify the impact on newhires, new hire earnings, separations, and recalls. Then we progress to a model that includes datafrom other states, thus utilizing geographic variation in demographic differences over time to iden-tify estimated treatment effects on young women in California.13

Consider first the following simple econometric specification, which serve as the basis for ouranalysis of the labor market impacts of the CPFL within California.

ln Ydqc� �

5bT Postq 3 YoungFemd� �

1dd1cq1ac1�dqc (1)

In this specification only data from California is used. In equation 1 the unit of observation is atthe demographic-quarter-county level with ln Ydqc

� �representing one of the four log outcome

measures—average monthly new hire earnings, total new hires, separations, and recalls (extendedleaves), each measured for a given demographic group (d), in a given quarter (q), and in a givencounty (c). The coefficient of interest is bT, which measures the treatment on young-female out-comes following implementation of CPFL.14 The variable Post is an indicator variable equal toone for all observations in or after the third quarter of 2004, after CPFL went into effect. The vari-able YoungFem is an indicator variable equal to one for women in the 19–21, 22–24, and 25–34age categories.15 dd and cq represent full sets of demographic group and quarter indicator variablesto account for time invariant differences between demographic groups and common shocks thathit all demographic groups in a given quarter, plus county fixed effects. From this specification, we

13 In an earlier analysis we used as the control group just the four SDI states (Hawaii, New Jersey, New York, and RhodeIsland) whose disability programs provided partial wage replacement benefits for pregnancy, but not paid family leave,as was the case in California prior to its 2004 implementation of CPFL. Placebo tests convinced us that the SDI states(in particular, New York) provided a questionable control group for the immediate years around CPFL. This mayhave stemmed in part from introduction of state minimum wage increases beyond the federal minimum in all four ofthese states during our estimation period, increases most likely to affect young workers. There were no federal orCalifornia increases in minimum wages during these years. As a robustness check on the results shown in this article,we account for all state-specific minimum wage changes. We found no substantive differences between those resultsand the results reported in the article. We thank Ian Schmutte for providing information on state-by-quarter changesin minimum wages (see Gittings and Schmutte, forthcoming).

14 Throughout the article, we treat bT as the percentage effect of the treatment on outcome Y. It is an approximate per-centage, but differences are trivial when values of bT are small, as is the case for all our estimates.

15 These are the age groupings most likely to be impacted by the CPFL. Births per 1000 women in 2004 were 20.1 for15–17 year olds; 66.2, 96.3, 110.5, 97.7 for age groups 18–19, 20–24, 25–29, 30–34; and 46.5, 10.1 for age groups 35–39, 40–44 (Martin et al. 2011, Table 4). We include separate demographic fixed effects for the detailed age groups, but“treatment effect” estimates are for the combined 19–34 age group of young women.


can extract estimates for CPFL treatment effects on young women relative to both young men andolder women in California.

Equation 2 presents a DD model that expands the data to include other states, but restrictsthe comparison group and sample to observations for young women. Including other states (usingcounty observations) allows us to directly compare changes to hiring and wage offers for youngwomen in California with young women in states not impacted by CPFL.

ln Yqc� �

5 bT Postq 3 CA� �

1cq1ac1�qc (2)

In equation 2, bT provides an estimate of log differences in outcomes for young women in Califor-nia following CPFL, as compared to those for young women in other states, conditioned on fixedeffects for quarter q and county c. Equation 2 is estimated using county as the unit of observation,thus providing a comparison of California counties with counties in other states. Use of state-levelobservations provides highly similar results.

Finally, we extract estimated treatment effects from a more general triple-difference modelthat includes all counties across California, all states and all demographic groups, but now identi-fies bT off the comparison of time changes in new hires and earnings (among other outcomes) foryoung women relative to other demographic groups in California counties compared to thesesame relative changes over time for young women in counties in other states. It takes the form

ln Ydqc� �

5 bT Postq 3 CA 3 YoungFemd� �

1ddc1cqs1adq1�dqc (3)

where the variables ddc, cqs; and adq represent full sets of county-demographic, quarter-state (withs designating state), and demographic group-quarter indicator variables to control for time invari-ant differences between county-demographic groups as well as shocks to demographic groups andstates that occur in a given quarter.16

The inclusion of these large sets of indicator variables effectively controls for many of theworker differences that vary across demographic groups, counties, and years. Consider education,a crucial determinant of new hire earnings. If young women in California have different levels ofeducation than other demographic-county combinations these differences will be picked up by ddc

as long as they are time invariant over the estimation period. If state education levels or county-demographic group education levels are changing over time these changes are likely picked up bycqs and adq respectively.

County rather than state-level results naturally provide greater variation to the outcome vari-ables of interest and are likely to provide more precise estimates. There are two minor disadvan-tages. First, the county models become large given the substantial number of interaction variablesrequired in fixed effects models. Second, the QWI does not report data for very small data cells inorder to insure confidentiality (this involves a tiny proportion of total county-by-demographicobservations). All of our regression analyses are weighted. For the new hire earnings regressions,we weight by the number of new hires. For the new hire, separations, and recall regressions weweight by total employment. This has the effect of making our sample representative of the fullpopulation flows of new hires entrants, separations, and recalls.

16 Specifications using fixed effects based on more disaggregated geographic categories proved computationally unwork-able despite access to the considerable resources of Cornell�s Social Science Gateway. For those regressions where wewere able to use the full set of interactive county fixed effects, results were highly similar to those shown.


We attempt to be conservative in selecting the level at which to cluster standard errors, fol-lowing advice in recent literature (Cameron and Miller 2015). Much of our analysis is at thecounty-by-demographic group level, but the policy treatment is at the state level, with differenteffects across demographic groups. In our preferred triple-difference analysis (Table 5), we opt forclustering at the state-by-demographic group level. Given the concerns over inference indifference-in-differences models (Bertrand, Duflo, and Mullainathan 2004), we ran other models(not reported) clustering at the county-demographic, state, and demographic level. Clustering atthe state-by-demographic group level (Tables 5–7) proved to be a conservative choice.17 In thewithin-California analysis (Table 3) we cluster by demographic group. Because of the limited num-ber of clusters in the California-only analysis, we compute standard errors using the wild clusterbootstrap method suggested in Cameron, Gelbach, and Miller (2008). In the cross-state analysisincluding only young women (Table 4), we cluster by state.

8. CPFL Effects on New Hire Earnings and Employment Flows

Double Difference Estimates Using Within-California Analysis

Before discussing our preferred triple-difference specification, we consider simpler but lessinformative double difference estimates shown in equations 1 and 2. All analyses provide estimatesof “treatment” effects from CPFL on new hire earnings, new hire employment, separations, andrecalls among young (ages 19–34) women. Table 3 provides results from within-California analysesbased on changes in outcomes between the quarters prior to and following implementation ofCPFL, as shown in equation 1. Observations are at the quarter-by-county–by-demographic grouplevel. Panel A compares changes in outcomes (in log levels) for young women compared to olderwomen within California. Panel B does likewise using younger men as the comparison group,while Panel C compares young women to all demographic groups within the state (other thanyoung women). Included are fixed effects for quarter, county, and demographic group (i.e., sex-by-detailed-age dummies). Standard errors are clustered at the demographic group level using a wildcluster bootstrap method (Cameron, Gelbach, and Miller 2008).

Given the relatively large standard errors in Table 3, most coefficients are insignificant.Taking these insignificant coefficients at face value, the within-state analysis suggests decreases innew hire earnings of young women relative to other worker groups of between 1 and 2%, substan-tively larger than expected given our back-of-the-envelope guesstimates of wage decline with fullshifting. Skepticism regarding the within-state wage results is reinforced by our summary statisticsin Table 2 showing that wage growth for young women between the pre-CPFL and post-CPFLperiods was lower than for all workers not only in California (0.018 vs. 0.031) but also in the com-parison group states (0.010 vs. 0.022).

Estimates on within-state worker labor market flows (hires, separations, and recalls) amongyoung women appear to indicate substantive (but statistically insignificant) positive effects oneach of these flows resulting from paid family leave. Similar results are obtained in our preferredtriple-difference analysis, seen subsequently. The suggestion is that paid family leave led to moreseparations and extended leaves for young women bonding with their newborns. There was littleeffect on the composition of the workforce, however, as some women returned to the same

17 Given that the QWI data are compiled from something close to the population of private wage and salary employeesrather than a sample, there is some ambiguity as to interpretation of (or even need for) standard errors.


employer (i.e., recalls) and there were higher levels of new hires, some of whom may have perma-nently separated from other employers following family leave.

Double Difference Estimates Using Across State Analysis for Young Women

The prior within-California results are informative, but cannot rule out the possi-bility that young women as a whole, regardless of whether they were in California,might have been experiencing differences in labor outcomes relative to other demo-graphic groups for reasons unrelated to CPFL. The across-state double difference speci-fication shown in Table 4 compares young women in California to young women inother states to account for this possibility, as seen in equation 2. Included are fixedeffects for quarter and county, with standard errors clustered at the state level.While controlling for changes occurring to young women across the country, this specifi-cation has the disadvantage of not controlling for economic conditions in Californiathat differ from the rest of the country, differences readily evident in Table 2. Usingyoung women in other states as the control group yields similar qualitative results forhires, separations and recalls as does the within-state analysis, but indicates an increaseof about 1% in new hire earnings attributable to CPFL, a result opposite thatexpected from theory. These results are likely to reflect overall growth in California�seconomy that were not experienced in the rest of the country. These results reinforcethe need for a triple-difference specification accounting for changes both within andacross states.

Table 3. CPFL Double-Difference Effects on New Hire Earnings and Labor Market Flowsof California Young Women: Within-State Comparison to Other Workers

(1) (2) (3) (4)ln(NH Earn) ln(New Hires) ln(Seps) ln(Recalls)

Panel A: Older Women ComparisonPost 3 Young Fem 20.0204

(0.0157)0.0141

(0.0301)0.0380

(0.0317)0.0386*(0.0212)

Observations 5422 5455 5462 5365R2 0.983 0.998 0.997 0.979

Panel B: Younger Men ComparisonPost 3 Young Fem 20.0128

(0.0163)0.0240

(0.0320)0.0439

(0.0347)0.0414

(0.0314)Observations 5413 5444 5447 5314R2 0.990 0.998 0.998 0.984

Panel C: All Demographic ComparisonPost 3 Young Fem 20.0173

(0.0133)0.0186

(0.0311)0.0468*(0.0263)

0.0276(0.0212)

Observations 10,841 10,916 10,939 10,713R2 0.986 0.997 0.996 0.979

Note: *** p< 0.01, ** p< 0.05, * p< 0.1. Analysis is at the county-by-quarter-by demographic group levelwithin California. County, demographic group, and quarter fixed effects are included. Standard errors areclustered at the demographic level and computed using the wild cluster bootstrap method suggested byCameron, Gelbach, and Miller (2008). “NH” designates new hires, “Seps” separations, and “Recalls” extendedleaves of at least one quarter. Col. (1) regressions weighted by new hires. Cols. (2), (3), and (4) regressions weightedby employment.


Triple-difference Analysis Across States Using Multiple Demographic Groups

In this section, we turn to the preferred triple-difference analysis (equation 3) in which theexperience of young women in California is compared to those of other young women elsewherein the U.S., each relative to other demographic groups within their respective states. Panels Athrough C of Table 5 present results from three alternative triple-difference specifications withroughly a half million county-by-quarter-by-demographic group observations. The specificationsdiffer only in the density of the fixed effects. The results shown in Panel A include county-by-demographic group and quarter fixed effects. Panel B reports results with county-by-demographicand quarter-by-demographic group fixed effects. Panel C reports results with county-by-demographic, quarter-by-demographic, and state-by-quarter fixed effects. Standard errors areclustered at the state-by-demographic group level (standard errors are smaller when clustered atthe county-by-demographic level). Coefficient estimates are highly similar across these three speci-fications, as well as in unreported specifications including different combinations of geographic,demographic, and time fixed effects. As fixed effects are added moving from Panels A to C, stand-ard errors decrease as more variation is explained. Panel D provides an analysis identical to thatshown in C, except that the control period is moved backward two years to exclude the periodbetween CPFL passage and implementation.

Unlike the simpler double-difference results in Table 3, these large set of fixed effects are ableto account for changes during this time period that are unique to California and for changes occur-ring to earnings and employment flows of young women relative to other groups that are commonacross states. Inclusion of county-demographic fixed effects accounts for any time-invariant differ-ences in the unit of observation. Thus the impact of CPFL in these specifications is identifiedsolely off changes that occur within a county-demographic group over time. Although our pre-ferred specification is Panel C, we also focus attention on Panel B results given that specification Cis too computationally intensive to conduct placebo tests. The magnitude of coefficients is largelyinvariant to the combination of fixed effects.

As expected, column 1 of Table 5 shows new hire earnings effects that are essentially zero.The coefficient of about two-tenths of one percent, combined with a small standard error of abouta half of a percent, provides reasonably strong evidence that the policy had a minimal impact onthe earnings of young female new hires, the group for whom you would most likely see an impact.The tiny wage effect is likely to reflect some combination of a low monetary cost of CPFL (as apercent of payroll) coupled with weak disruption costs among employers from CPFL (i.e., smalldemand shifts).

Table 4. CPFL Double-Difference Effects on New Hire Earnings and Labor Market Flowsof California Young Women: Other-State Young Women Comparison


Post 3 CA 0.0130***(0.0038)

0.0053(0.0060)

0.0281***(0.0071)

0.0324(0.0279)

Observations 128,954 131,434 132,789 115,474R2 0.970 0.996 0.996 0.958

Note: *** p< 0.01, ** p< 0.05, * p< 0.1. Analysis is at the county-by-quarter level across all states in sample.County and quarter fixed effects are included. Robust standard errors in parentheses, clustered at the state level.“NH” designates new hires, “Seps” separations, and “Recalls” extended leaves of at least one quarter. Col. (1) regres-sions weighted by new hires. Cols. (2), (3), and (4) regressions weighted by employment.


While the impact on new hire earnings is minimal there are substantive impacts on workerflows. Estimates in Panel B indicate that CPFL increase new hires by 3%, separations by 2.3%,and recalls (extended leaves) by 2.7%. All three of these results are quantitatively significant andthose for new hires and separations are statistically significant. The recall coefficients have largestandard errors in Panels A and B, calling into question what weight ought to be attached to therecall evidence. We interpret findings in the next section.

Panel D is equivalent to C, except that we have moved backward by two years the time periodfor the pre-CPFL control period in order to exclude the period between CPFL passage and imple-mentation. If the effects of CPFL began following passage, such effects would be reflected in ourprimary control period (as in Panel C) and thus understate the size of CPFL effects. Evidence inPanel D using the earlier control period but same treatment period produces coefficient estimatessubstantively larger than those seen in Panel C. New hire earnings effect estimates are a positiveone percent rather than zero, but insignificant. The estimated effects of new hire and separationflows are a half percentage point higher than previously shown in Panel C, roughly 3.5% rather

Table 5. CPFL Triple-Difference Effects on New Hire Earnings and Labor Flows of YoungWomen


Panel A: County-Demographic Group, and Quarter FEPost 3 CA 3

Young Fem20.0015(0.0106)

0.0310(0.0217)

0.0241(0.0212)

0.0294(0.0297)

Observations 515,501 528,511 528,133 476,992R2 0.963 0.995 0.993 0.950

Panel B: County-Demographic Group, and Demographic Group-Quarter FEPost 3 CA 3

Young Fem20.0018(0.0047)

0.0299**(0.0132)

0.0229**(0.0092)

0.0274(0.0257)

Observations 515,501 528,511 528,133 476,992R2 0.965 0.995 0.993 0.951

Panel C: County-Demographic Group, Demographic-Quarter, and State-Quarter FEPost 3 CA 3

Young Fem20.00215(0.0041)

0.0304**(0.0125)

0.0236***(0.0070)

0.0295***(0.0102)

Observations 515,501 528,511 528,133 476,992R2 0.967 0.996 0.995 0.966

Panel D: Same as C but with Pre-Passage Control Period, 2000:Q3–2002:Q2 County-Demographic Group, Demographic-Quarter, and State-Quarter FE

Post 3 CA 3Young Fem

0.0095(0.0064)

0.0347**(0.0168)

0.0296**(0.0126)

0.0620***(0.0228)

Observations 485,273 497,624 517,281 448,615R2 0.965 0.995 0.994 0.968

Note: *** p< 0.01, ** p< 0.05, * p< 0.1. Analysis is at the county-by-quarter-by-demographic group level across allstates in sample. Panel A contains separate county-demographic group and quarter FE. Panel B contains county-demographic group and demographic-quarter FE. Panel C, the largest model we are able to run, contains county-demographic group, demographic-quarter and state-quarter two-way FE. By using these sets of fixed effects, all firstand second order interaction terms associated with a triple difference model are implicitly included in the above equa-tions. All models run include the lower order triple-difference interaction and indicator terms (Post x YoungFem, Postx CA, YoungFem x CA, Post, YoungFem, and CA) when they are not absorbed by the fixed effects. Robust standarderrors in parentheses, are clustered at the state-demographic group level. “NH” designates new hires, “Seps” separa-tions, and “Recalls” extended leaves of at least one quarter. Col. (1) regressions weighted by new hires. Cols. (2), (3),and (4) regressions weighted by employment.


than 3% for new hires and 3% rather than 2.5% for separations. Coefficient estimates for the twospecifications using less dense fixed effects (as in Panels A and B), but with the earlier controlgroup period, are highly similar to those shown in Panel D.

Although the coefficient differences seen between Panels C and D are consistent with therebeing small effects between passage and implementation of CPFL, we are reluctant to draw thatconclusion for two reasons. First, evidence indicates that workers had little knowledge of CPFLfollowing passage (Appelbaum and Milkman 2011) and, second, the earlier control period used inpanel D includes a state minimum wage increase and the “dot-com” boom and bust that may haveaffected young women (relative to other California workers) differently than in other states. Notealso that the sample sizes are somewhat lower in Panel D than C because state participation in theLEHD was less complete going back in time. Because estimates using the alternative control peri-ods are similar, choosing between Panels C and D is relatively unimportant. We prefer to empha-size the more conservative estimates based on the control period between passage andimplementation.

Comparing Earnings and Employment Stocks Versus Flows

We have argued in this article that examining employment flows rather than stocks providesa fruitful approach to examining labor market policy effects. To provide perspective on this argu-ment, in Table 6 we provide a triple-difference analysis identical to that shown in Table 5, exceptthat we instead use as our outcome measures (dependent variables) traditional measures of aver-age earnings and employment rather than the earnings for new hires and employment flows. Asdiscussed earlier, CPFL-induced labor market changes are unlikely to be picked up using levels ofearnings and employment, which are driven mainly by incumbent workers and adjust slowly overtime. Panels A-C of Table 6 present the results of the triple-difference specification in equation 3using average monthly earnings and employment levels as the dependent variables. Results arereported for both workers with stable employment (those who have been at the firm for at leastthree months) and for all workers (total number of workers with earnings in the quarter). Asexpected, the earnings level coefficients on the triple-difference variable produce point estimatesthat are effectively zero. For employment, point estimates are positive (about one percent), but thecoefficients are not significant. That said, estimates of employment level effects of about one per-cent reinforce our confidence in the conclusion that CPFL did not reduce employment amongyoung women.

Discussion of Results

The results from our preferred triple-difference specifications (Table 5) enhance our under-standing of the labor market effects of CPFL. As expected, we find a limited impact of the pro-gram on young women�s earnings. Point estimates show a near-zero earnings effect for newly hiredyoung women. The estimates are precisely estimated and preclude the possibility that the programhad large negative earnings effects. Near-zero effects on new hire earnings are consistent with ourback-of-the-envelope calculation showing that even with full wage shifting, substantive earningsimpacts were unlikely absent substantial disruption costs to employers and sizable increases inyoung women�s labor supply.


Although wage effects for young women were expected to be small, finding an increase inyoung female new hires was initially surprising absent evidence of much lower wages. A substantialincrease in young female new hires coupled with small increases in employment levels is betterunderstood once we examine separations and (to a lesser extent) recalls, both increasing substan-tially following implementation of CPFL (with the caveat that standard errors in the recall equa-tions are large). Establishment-level hiring and separations tend to move together, assuming thatmost separations are from quits rather than layoffs.

The magnitude of the coefficients for new hires, separations, and recalls in Table 5 are roughlysimilar, approximately 0.03, 0.025, and 0.03, respectively. These coefficients measure the approxi-mate percentage increases in these labor flows associated with CPFL. Because the magnitudes ofseparations and new hires (or new hires plus recalls) are similar, equivalent treatment effect coeffi-cients in the separations and new hire regressions imply similar size flows. As seen in Table 2, inthe post-CPFL period, mean quarterly flows among young women in California for separations,new hires, and recalls, were 2.10, 2.03, and 0.22 million. The small difference between new hiresplus recalls (2.25) minus separations (2.10) represents a 0.15 (or 150 thousand) increase in the levelof employment over the post-CPFL period.

In short, our results show that adoption of CPFL led to increased job churn. In work notshown, we find that these higher rates were sustained beyond our sample period. Churn can beefficiency enhancing if young women were previously staying in jobs that, apart from their paidleave policy, were an inferior match, with universal paid leave reducing job lock and allowingworkers to find better matches. Churn is not unambiguously good, however. An argument made

Table 6. CPFL Triple-Difference Effects on Earnings and Employment Levels Rather thanFlows

(1) (2) (3) (4)ln(Earn Stable) ln(Earn All) ln(Emp Stable) ln(Emp All)

Panel A: County-Demographic Group, and Quarter FEPost 3 CA 3 Young Fem 0.0032

(0.0080)0.0049

(0.0071)0.0133

(0.0237)0.0124

(0.0228)Observations 560,862 560,859 560,862 560,863R2 0.988 0.986 0.999 0.999

Panel B: County-Demographic Group, and Demographic-Quarter FEPost 3 CA 3 Young Fem 0.0024

(0.0038)0.00415(0.0047)

0.0112(0.0080)

0.0101(0.0073)

Observations 560,862 560,859 560,862 560,863R2 0.990 0.989 0.999 0.999

Panel C: County-Demographic Group, Demographic-Quarter, and State-Quarter FEPost 3 CA 3 Young Fem 0.0024

(0.0030)0.0042

(0.0039)0.0115

(0.0072)0.0103

(0.0065)Observations 560,862 560,859 560,862 560,863R2 0.991 0.990 0.999 0.999

Note: *** p< 0.01, ** p< 0.05, * p< 0.1. Analysis is at the county-by-quarter-by-demographic group level across allstates in sample. Panel A contains separate county-demographic group and quarter FE. Panel B contains county-demographic group and demographic-quarter FE. Panel C, the largest model we are able to run contains the county-demographic group, demographic-quarter and state-quarter two-way FE. Robust standard errors in parentheses, clus-tered at the state-demographic group level. We estimate the model on “stable” employment, defined as workers whohave worked at the firm for at least 3 months and on “all” (i.e., point-in-time) employment. All regressions weightedby employment.


for paid family leave is that it might increase job continuity for jobs that are a good match or thatentail firm-specific human capital. The increase in separations following CPFL raises concerns.

The finding that recalls may have increased following CPFL is consistent with results inRossin-Slater, Ruhm, and Waldfogel (2013) that indicate increased leave time for women as a resultof mandated paid leave. CPFL appears to have enabled some women with young children to sub-stantially increase time off from work beyond the period of mandated paid leave and, in somecases, return to their same employer following extended leaves. Although coefficient estimatesindicate similar percentage changes in recalls and separations, the number of recalls is about one-tenth the number of separations (see Table 2). Most separations are not followed by a recall.

9. Placebo Tests of Earnings, Employment, Separations, and Recalls

As a robustness check, we replicate the triple-difference model shown previously in Panel Bof Table 5 (with county-demographic and demographic-quarter fixed effects), but this time removethe California data and “replace” it with nine alternative placebo treatment groups of statesbundled together based on geography and group size.18 For each set of placebo tests, data fromplacebo states are excluded from the control group. These results are shown in Table 7 for new hireearnings, new hires, separations, and recalls. These placebo policy tests allow us to assess the reli-ability of our California estimates. Small and insignificant estimates for all or most of the placebopolicies would provide a valuable falsification test. A systematic set of results that either mimickedthose for California or produced numerous large and/or statistically significant estimates wouldsuggest our data and empirical approach might not be sufficiently informative to identify theimpact of CPFL.

The first row shows the treatment effect estimates for California previously reported in PanelB of Table 5. We first examine estimates for new hire earnings, where we previously obtained anear-zero estimate for California. Most of the placebo coefficient estimates for new hire earningsare larger in absolute value than the (near-zero) California estimate. One coefficient is statisticallysignificant, with a 20.016 estimate. We previously argued that theory suggests a small negativewage effect from CPFL, but that our estimates of new hire wage effects are statistically and sub-stantively indistinguishable from zero. Results from the placebo tests reinforce this conclusion.

The placebo tests on worker flow variables provide more definitive results. For the new hireemployment regressions, where we obtained a highly significant 3% treatment effect estimate, onlyone placebo estimate is significant (marginally so), with a negative coefficient small in magnitude.These results reinforce our conclusion that CPFL led to substantial increases in young female newhires. Such a result, however, is only plausible if there is increased churn in the labor market, assuggested by our results on separations and recalls. For separations, we find no significant or mar-ginally significant coefficients among the nine placebo tests, with no coefficient as large in magni-tude as the estimate for California. The recall results are less definitive. Although no placebogroup produces a significant estimate, four of the state groups produce coefficients the same orderof magnitude as that for California (but only one with a positive coefficient). Although the recall

18 Coefficient estimates in Panels A, B, and C of Table 5 are highly similar. Our choice of using the Panel B rather thanC specification for the placebo test is due to the heavy computational (time) demands required for estimation of thePanel C specification.


results are consistent with there being an increase in extended leaves following CPFL, estimatesare noisy.

Overall, the placebo tests reinforce our previous conclusions. The data are not powerfulenough to precisely identify earnings effects of, say, one percent or less. In contrast to the earningsresults, the placebo tests support the conclusion that there were substantive increases in labor mar-ket hiring, separations, and possibly recalls among young women following CPFL.

10. Conclusion

Mandated worker benefits not provided voluntarily, but politically viable, are likely to havemodest benefits and costs. Nonwage benefits highly valued by workers relative to their costs arethose most likely to be voluntarily provided by employers (with costs shifted to workers). Man-dates that have substantial costs relative to workers� valuation are those least likely to work theirway through the political process.

Few data sets are capable of reliably identifying small causal effects from employer mandates.Household data sets typically have small sample sizes of individuals by geographic location bytime period. Establishment data, on the other hand, rarely provide the demographic and geo-graphic breakdown needed to analyze mandates that differentially impact alternative groups ofworkers. More fundamentally, wages and employment across demographic groups or within busi-nesses change gradually. Incumbent workers are not likely to have their pay reduced substantially,nor will businesses quickly alter the demographic make-up of their trained workforces through dis-missals. The margin over which one is most likely to observe wage and employment adjustments iswith respect to new hires, both through changes in their demographic composition and in thewages offered, as well as changes in other worker flows.

The QWI data set provides a relatively new and underutilized resource that lends itself toevaluation of public policies that differentially affect employment and/or earnings with respect to

Table 7. Triple-Difference Outcome Estimates from State Group Placebo Policies


California (CA) 20.0018 0.0299** 0.0229** 0.0274Placebo State Groups:ME VT NY CT RI NJ 0.0078 0.0112 0.0074 0.0182PA DE WV OH MI 0.0044 20.0104 20.0083 0.0131MD DC VA NC SC 0.0085 20.0027 0.0050 20.0127GA FL AL 20.00001 0.0018 0.0161 20.0346IN KY TN IL MO 0.0027 20.0020 20.0162 20.0244WI MN IA KS MO ND SD NE 20.0048 20.0066 20.0110 0.0285LA TX OK 20.0162*** 20.0185* 20.0091 20.0266NM NV UT CO 0.0003 20.0145 20.0211 20.0124MT ID WY WA OR 0.0079 20.0098 20.0007 20.0125

Note: *** p< 0.01, ** p< 0.05, * p< 0.1. All standard errors clustered at the state-by-demographic group level (eachplacebo state group is treated as a single state). The first line provides CPFL estimates shown previously in Table 6,Panel B. We report above placebo results in which alternative groups of neighboring states, roughly similar in size toCalifornia, are designated as treated areas. Not included in the placebo state groups are Arizona, Arkansas, Massa-chusetts, Mississippi, and New Hampshire, which had incomplete LEHD records during the years of study, andAlaska and Hawaii, which are not geographically close to other states. “NH” designates new hires, “Seps” separa-tions, and “Recalls” extended leaves of at least one quarter. Col. (1) regressions weighted by new hires. Cols. (2), (3),and (4) regressions weighted by employment.


time, location, and demographic group. Particularly appealing is QWI�s provision of data on thenumber and earnings of stable (not short-term) new hires, margins over which labor marketadjustments are most likely to occur. We believe this type of analysis strengthens and improves ourunderstanding of the impact of CPFL, and provides promise for future labor market policyanalyses.

Although we have emphasized the benefits of the QWI, we acknowledge its limits. First, theQWI contains data on earnings but not hours worked, thus one cannot unambiguously know towhat extent changes in earnings reflect changes in wages versus hours worked (hence our use ofstable new hires employed throughout a measured quarter). Second, while examining flows allowsfor the detection of small changes to labor markets, there may be unmeasured shifts in the compo-sition of the newly hired (and separating) workers (say, with respect to education or occupationalexperience) that occur as a result of the policy. If a policy encourages a different type of worker tojoin or leave the firm, earnings estimates may be biased. Third, the QWI includes establishmentsof all sizes, but does not provide information on size. That is not an issue for the analysis of CPFL,but is a serious impediment for analyses of policies limited to establishments of a given size (e.g.,the FMLA). In short, the analysis performed here comes with a tradeoff. Household data used inprevious studies contain far fewer observations, but are better able to measure and control forindividual-level earnings, hours, attributes, transition patterns, and household information. Whilewe are unable to include controls at the individual or household level, the QWI does contain theuniverse of all private sector worker flows and provides “data cuts” that enable reliance on a largeset of fixed effects to control for demographic and geographic differences.

California�s mandatory paid family leave policy, a first in the U.S., effectively added six weeksof partially paid leave to new mothers (or fathers). Rossin-Slater, Ruhm, and Waldfogel (2013)and other studies indicate that CPFL led to increased time off among mothers with infants. Ouranalysis concludes that CPFL resulted in little change in earnings for young women in California,coupled with increased churn in the form of separations, hires, and possibly recalls (extendedleaves). Although labor market churn has costs as well as benefits, some portion of the increasedchurn is likely the result of reduced job lock and enhanced job matching made possible by univer-sal paid family leave. Blau and Kahn (2013) note the U.S. reversal in female labor force participa-tion, being ranked sixth out of 22 OECD countries in 1990 but 17th of 22 in 2010. They suggestthat a general lack of family-friendly policies in the United States, as compared to other developedeconomies, helps explain this change. Although we do not examine labor force participation, ourresults do suggest that paid family leave expanded options for young women and resulted inenhanced labor market flexibility.

Acknowledgments

We appreciate assistance from Damon Nelson in the Program, Data, and ResearchUnit of California�s Employment Development Department for help in understandingdetails of the CPFL program, the Cornell Social Science Gateway for use of computingresources under NSF grant #1042181, and helpful comments from the editor and refer-ees, Spencer Banzhaf, Chris Ruhm, Ian Schmutte, and participants in seminars at Geor-gia State University, the IZA/SOLE Transatlantic Meeting of Labor Economists, theNBER Summer Institute, the Southern Economic Association Meetings, and the Soci-ety of Labor Economists Meetings.


References

Abowd, J. M., B. E. Stephens, L. Vilhuber, F. Andersson, K. L. McKinney, M. Marc Roemer, and S. Woodcock. 2009.The LEHD infrastructure files and the creation of the Quarterly Workforce Indicators. In Producer dynamics:New evidence from micro data, edited by T. Dunne, J. B. Jensen, and M. J. Roberts. Chicago: University of ChicagoPress, pp. 149–230.

Appelbaum, E., and R. Milkman. 2011. Leaves that pay: Employer and worker experiences with paid family leave inCalifornia. Center for Economic and Policy Research. Available http://www.cepr.net/documents/publications/paid-family-leave-1-2011.pdf.

Baker, M., J. Gruber, and K. Milligan. 2008. Universal childcare, maternal labor supply, and family well-being. Journalof Political Economy 116:709–45.

Baker, M., and K. Milligan. 2008. How does job-protected maternity leave affect mothers� employment? Journal ofLabor Economics 26:655–91.

Baker, M., and K. Milligan. 2010. Evidence from maternity leave expansions of the impact of maternal care on earlychild development. Journal of Human Resources 45:1–32.

Bartel, Ann, Maya Rossin-Slater, Christopher Ruhm, Jenna Stearns, and Jane Waldfogel. 2015. Paid family leave,fathers� leave-taking and leave-sharing in Dual-Earner households. NBER Working Paper No. 21747.

Baum, C. L. 2003. The effects of maternity leave legislation on mothers� labor supply after childbirth. Southern Eco-nomic Journal 69:772–99.

Baum, C. L., and C. J. Ruhm. 2016. The effects of paid family leave in California on labor market outcomes. Journal ofPolicy Analysis and Management 35:333–56.

Bertrand, M., E. Duflo, and S. Mullainathan. 2004. How much should we trust differences-in-differences estimates?Quarterly Journal of Economics 119:249–75.

Blau, F. D., and L. M. Kahn. 2013. Female labor supply: Why is the United States falling behind? American EconomicReview Papers and Proceedings 103:251–6.

Byker, T. 2014. The role of paid parental leave in reducing women�s career interruptions: Evidence from paid leave lawsin California and New Jersey. Unpublished manuscript, University of Michigan, April.

Cameron, C., J. Gelbach, and D. L. Miller. 2008. Bootstrap-based improvements for inference with clustered errors.Review of Economics and Statistics 90:414–427.

Cameron, C., and D. L. Miller. 2015. A practitioner�s guide to cluster-robust inference. Journal of Human Resources 50:317–72.

Card, D. 1992. Regional variation in wages to measure the effects of the federal minimum wage. Industrial and LaborRelations Review 46:22–37.

Carneiro, P., K. V. Løken, and K. G. Salvanes. 2011. A flying start? Maternity leave benefits and long run outcomes ofchildren. IZA Discussion Paper No. 5793.

Council of Economic Advisors. 2015. The economics of family-friendly workplace policies. In Economic report of thePresident. Washington D.C., pp. 157–202.

Dahl, G. B., K. V. Løken, M. Mogstad, and K. G. Salvanes. 2013. What is the case for paid maternity leave? NBERWorking Paper No. 19595.

Das, T., and S. W. Polachek. 2015. Unanticipated effects of California�s paid family leave program, Contemporary Eco-nomic Policy 33:619–35.

Dot-com bubble (n.d.). In Wikipedia. Accessed 22 February 2016. Available https://en.wikipedia.org/wiki/Dot-com_bubble.

Dube, A., T. W. Lester, and M. Reich. 2016. Minimum wage shocks, employment flows, and labor market frictions. Jour-nal of Labor Economics 34:663–704.

Dustmann, C., and U. Sch€onberg. 2012. Expansions in maternity leave coverage and children�s long-term outcomes.American Economic Journal: Applied Economics 4:190–224.

Espinola-Arrendondo, A., and S. Mondal. 2010. The Effect of parental leave on female employment: evidence from statepolicies. Washington State University School of Economic Sciences working paper series WP 2008-15 (revised).

Fass, S. 2009. Paid leave in the states: A critical support for low-wage workers and their families. National Center for Chil-dren in Poverty, Columbia University, NewYork, NY, March.

Gittings, R. K., and I. Schmutte. Forthcoming. Getting handcuffs on an octopus: Minimum wages, employment andturnover, Industrial and Labor Relations Review.

Gruber, J. 1994. The incidence of mandated maternity benefits. American Economic Review 84:622–41.Klerman, J. A., and A. Leibowitz. 1994. The work-employment distinction among new mothers. Journal of Human

Resources 29:277–303.Lalive, R., A. Schlosser, A. Steinhauer, and J. Zweim€uller. 2014. Parental leave and mothers� careers: the relative impor-

tance of job protection and cash benefits. Review of Economic Studies 81:219–65.


http://www.cepr.net/documents/publications/paid-family-leave-1-2011.pdf

http://www.cepr.net/documents/publications/paid-family-leave-1-2011.pdf

http://https://en.wikipedia.org/wiki/Dot-com_bubble

http://https://en.wikipedia.org/wiki/Dot-com_bubble

Lalive, R., and J. Zweim€uller. 2009. How does parental leave affect fertility and return to work? Evidence from two natu-ral experiments. Quarterly Journal of Economics 124:1363–402.

Martin, J. A., B. E. Hamilton, S. J. Ventura, M. J. K. Osterman, S. Kirmeyer, T. J. Mathews, and E. C. Wilson. 2011.Births: Final data for 2009. National Vital Statistics Reports, Division of Vital Statistics, U.S. Department ofHealth and Human Services. Hyattsville, MD, Vol. 60, Nov. 3, 2011.

Obama, B. 2014. “Family-friendly workplace policies are not frills–they�re basic needs.” Huffington Post, June 23. Avail-able http://www.huffingtonpost.com/barack-obama/family-friendly-workplace_b_5521660.html.

Rossin-Slater, M., C. J. Ruhm, and J. Waldfogel. 2013. The effects of California�s paid family leave program on mothers�leave-taking and subsequent labor market outcomes. Journal of Policy Analysis and Management 32:224–45.

Ruhm, C. J. 1998. The economic consequences of parental leave mandates: Lessons from Europe. Quarterly Journal ofEconomics 113:285–316.

Sch€onberg, U., and J. Ludsteck. 2014. Expansions in maternity leave coverage and mothers� labor market outcomes afterchildbirth. Journal of Labor Economics 32:469–505.

Summers, L. H. 1989. Some simple economics of mandated benefits. American Economic Review Papers and Proceedings79:177–183.

Van Giezen, R. W. 2013. Paid leave in private industry over the past 20 years. Beyond the Numbers: Pay and Benefits 2:1–6.


http://www.huffingtonpost.com/barack-obama/family-friendly-workplace_b_5521660.html

Documents

Evaluating Workplace Mandates with Flows Versus …ecobth/Curtis_et_al-Oct2016-SEJ-final.pdf · Evaluating Workplace Mandates with Flows Versus Stocks: An Application to California