University of Southern California Dissertations and Theses

The role of public policy in the decisions of parents and caregivers: an examination of work, fertility, and informal caregiving

(USC Thesis Other)

The role of public policy in the decisions of parents and caregivers: an examination of work, fertility, and informal caregiving

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content The Role of Public Policy in the Decisions of Parents and Caregivers: An Examination of Work, Fertility, and Informal Caregiving by Johanna Thunell A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY PUBLIC POLICY AND MANAGEMENT May 2018 2 Table of Contents INTRODUCTION ........................................................................................................................ 3 Paid Family Leave and Women’s Childbearing Decisions in the United States ..................... 8 INTRODUCTION ...................................................................................................................................9 BACKGROUND ....................................................................................................................................11 EMPIRICAL STRATEGY ...................................................................................................................20 RESULTS ...............................................................................................................................................24 ROBUSTNESS CHECKS .....................................................................................................................31 DISCUSSION .........................................................................................................................................37 CONCLUSION ......................................................................................................................................40 REFERENCES ......................................................................................................................................42 APPENDIX ............................................................................................................................................48 Can Paid Leave Promote Labor Force Attachment Among Parents of Children with Disabilities? .................................................................................................................................. 52 INTRODUCTION .................................................................................................................................53 THEORETICAL BACKGROUND .....................................................................................................56 EMPIRICAL LITERATURE ..............................................................................................................59 DATA AND VARIABLES ....................................................................................................................64 EMPIRICAL STRATEGY ...................................................................................................................67 RESULTS ...............................................................................................................................................70 DISCUSSION .........................................................................................................................................82 REFERNCES .........................................................................................................................................85 APPENDIX ............................................................................................................................................90 Public Support for Long-Term Care of People with Dementia and Their Caregivers: An Analysis of the Patchwork of State and Federal Policies ........................................................ 94 INTRODUCTION .................................................................................................................................95 BACKGROUND ON LONG-TERM SERVICES AND SUPPORTS IN THE U.S. .......................98 DECISIONS ABOUT CAREGIVING AND WORK .......................................................................101 PUBLIC POLICIES FOR CAREGIVERS OF PEOPLE WITH ADRD ......................................108 DISCUSSION AND FUTURE RESEARCH ....................................................................................121 CONCLUSION ....................................................................................................................................124 REFERENCES ....................................................................................................................................126 CONCLUSION ......................................................................................................................... 133 3 INTRODUCTION Modern families in the United States face challenges not encountered in previous generations. Work cultures and family responsibilities rooted in breadwinner and caregiver roles from the past exact a toll on the growing number of families headed by single parents and dual earners. Rather than devoting time to only one domain, today’s workers often divide their time between working and caring for their children or parents. Since the 1980s, the amount of time fathers spend with their children tripled, while working mothers today devote as much time to caring for their children as non-employed mothers in 1975 (Bianchi, 2011). At the same time, the share of mothers in the labor force jumped from 47 to 72 percent. Thus, a majority of children are being raised in families where all parents in the household are working. As the large Baby Boom generation ages, a growing number of workers fall in the so-called “sandwich generation” — caring for both elderly parents and children under eighteen (Miller, 1981). The Pew Center (2013) estimates that 15 percent of adults ages 40 to 59 support both children under 18 and parents over 65. This share is expected to grow as the U.S. population ages in the coming decades, but while the demand for long-term care is expected to increase, changes in family dynamics, such as smaller families and the high rate of maternal labor force participation will reduce the supply of potential caregivers. On an individual level, combining work and childrearing or elder care often leads to undesirable consequences. Studies demonstrate reduced labor supply and earnings among mothers, especially those with chronically ill and disabled children, and caregivers of elderly parents (e.g. Ettner, 1995, 1996; Powers, 2001, 2003). Providing care to aging parents, in particular, can lead to high levels of stress and declining health (e.g. Schulz & Sherwood, 2008; Wolff et al. 2016). From a societal prospective, the tradeoff between work and care shrinks the pool of available 4 workers and/or informal caregivers, which likely results in lower productivity and higher costs associated with long-term care. Public policies in the U.S. evolved over time to address some of these challenges. For example, federal tax deductions provide financial relief for expenditures associated with long-term services and support of a loved one (or dependent) for working caregivers (Internal Revenue Services, 2017). More recently, state and federal governments enacted policies providing services to family caregivers, such as the Lifespan Respite Care Act, in an effort to reduce the stress and burden associated with providing long-term care (Administration for Community Living, 2017). For working parents and caregivers, the federal Family and Medical Leave Act provides twelve weeks of job-protected time off work after childbirth or adoption, or to care for one’s own illness or that of a family member; however, restrictions on eligibility leave about 40 percent of workers without coverage (Klerman, Daley & Pozniak, 2014). A handful of states (including California) have enacted paid family leave programs that replace a portion of workers’ wages when they take time off to care for newborns, children, parents, or themselves, but no national policy guarantees paid time off. Literature from several disciplines evaluates the impact of these leave policies on the outcomes of some beneficiaries, but significant gaps remain in our understanding of their effectiveness throughout different stages in the life cycle and there is little systematic evidence of the impact of the patchwork of policies supporting family caregivers. Research on public policies and work-life balance in the U.S. has focused primarily on family leave and new parents, establishing benefits of unpaid and paid family leave (PFL) policies to new parents (Baum & Ruhm, 2016; Rossin-Slater, Ruhm & Waldfogel, 2013). Evidence from the policy in California suggests PFL increased leave-taking and employment among new mothers and fathers, and improved infant and child health outcomes (e.g. Lichtman-Sadot & Bell, 2017; 5 Rossin, 2011). However, the need for family leave continues throughout workers’ lives, whether they have a chronically ill or disabled child who requires care over a longer period, provide care to their aging parent, or become ill themselves. Indeed, work and care decisions, and subsequent effects on wages, health, and other outcomes, may be influenced by access to PFL for several different groups of people. Caregivers of children with disabilities or elderly parents may struggle to combine work and care in the absence policies that allow ongoing intermittent time off work. As a result, they may reduce their supply of labor, caregiving, or both. On the other end of the life cycle, workers may alter their decisions about childbearing to accommodate work rather than reduce their labor supply. Despite the mounting evidence of paid family leave’s positive impact on new parents’ employment, only five states (California, New Jersey, New York, Rhode Island, and Washington) have paid family leave programs and there is no federal law guaranteeing paid leave to new parents and caregivers. The diffusion of these policies to other states or the federal level entails recognizing that these challenges affect more than new parents and establishing the effect of paid leave on other beneficiaries, such as caregivers of the elderly. In the three studies that follow, I address many of these gaps in the literature and suggest avenues for future research. The first two chapters analyze the impact of California’s PFL policy on fertility outcomes and the labor supply of parents of children with disabilities, using difference- in-differences estimation to compare outcomes in California before and after PFL with a control group. I find the likelihood a woman ages 25 to 40 had a child in the last year increased by about 6.0 percent for all women in California, relative to similar states with no paid leave, and 7.6 percent for women working fulltime. Subsequent analyses indicated heterogeneity in the effect across socioeconomic background and age group. Among parents of children with disabilities, my results 6 suggest access to PFL increased the labor force participation of their mothers and fathers, by 9.4 and 4.3 percent, respectively, when compared to parents in California of children with no disability and parents in the rest of the United States. The final chapter explores policies supporting family caregivers of people with Alzheimer’s disease and related dementias (ADRD). Through a review of the policies and programs providing government funding and supportive services for these caregivers and a critical analysis of the literature related to their effectiveness, I find a general lack of policy-related research on the federal initiatives. There is substantial evidence of the effect small-scale, randomized interventions on people with ADRD and their caregivers. For example, caregiver interventions, such as respite care or counseling, reduced caregiver burden and work productivity loss (e.g. Pinquart & Sorenson, 2006). While results from these targeted interventions generally indicate national-level policies should positively affect ADRD caregivers, the overall impact of these policies is unknown. This final chapter concludes by identifying several avenues for future research, such as estimating the benefits and costs of each federal policy, and their interactions, to determine which are the most effective and cost-reducing programs. Together, these studies contribute to the literature on work and care, by highlighting the broad impact of policies on modern families across the life cycle. While virtually everyone will care for a loved one or be cared for some time in their lives, research on policies affecting work- life balance offers little evidence of the benefits beyond new parents. This research estimates the effect of California’s PFL on new fertility outcomes and a group of long-term caregivers, and explores the availability and effectiveness of policies for family caregivers of people with ADRD. The findings presented here demonstrate the potential of these policies to positively impact caregivers and care recipients throughout the life cycle; however, this work also illustrates the need 7 for future research on other groups of caregivers, alternative measures of labor supply, health and wellbeing outcomes, and federal policies for family caregivers. 8 Paid Family Leave and Women’s Childbearing Decisions in the United States by Johanna Thunell ABSTRACT This study examines how paid leave in the United States shapes working parents’ decisions along the family dimension of work-family interface. Specifically, I use data from the Current Population Survey, 1999 to 2009, and a difference-in-differences approach to estimate the impact of California’s Paid Family Leave (PFL) program on fertility behavior and whether the effect varies across socioeconomic status. Results indicate there is a positive effect of the policy on the probability a 25 to 40-year-old woman had a baby in the last year, which is equivalent to a 6.0 percent increase in birth likelihood after the PFL’s enactment. For women working fulltime, PFL increased birth likelihood by 7.6 percent. There is evidence of heterogeneity of stronger effect among women with earnings in the 25 th to 75 th percentile, high school graduates, and women in their 30s. In combination with previous research highlighting positive effects of PFL on labor market outcomes, these findings suggest the policy helps families combine work and childbearing. 9 INTRODUCTION The rise in mothers’ labor force participation over the last three decades of the 20 th century is well documented in the literature (e.g. Blau & Kahn, 2007). Combined with increasing time devoted to caring for children, modern parents face competing priorities of work and their family lives (Bianchi, 2011; Milkie, Kendig, Nomaguchi & Denny, 2010; Sayer, Bianchi & Robinson, 2004; Williams, 2000). In response, parents may decrease the amount of time devoted to paid work after having a child or couples may adjust the size of their family to accommodate work responsibilities. Indeed, many industrialized countries experienced very low fertility beginning in the 1980s, leading to concerns over the coupling of a shrinking workforce and aging populations (Caldwell & Schindlmayer, 2003; Drago et al, 2009). Unlike other developed countries, fertility in the United States remained stable over the last three decades of the 20 th century (Livingston, 2015). In recent years, however, the fertility rate in the U.S. began dropping to just below replacement level (World Bank, 2015). This trend may reflect a lasting decline in fertility or a delay due to the Great Recession, when many couples postponed childbearing (Cherlin et al. 2013). Governments in most industrialized countries have enacted public policies – such as family leave, subsidized childcare, baby bonuses, etc. – to help parents balance work and their family lives. Parental leave allows mothers and fathers to take time off work around the time of a birth, but policy provisions vary substantially across countries. According to a 2016 Organization for Economic Cooperation and Development (OECD) report, the average amount of paid leave available to new mothers in OECD countries was 54.1 weeks, ranging from 14 weeks in Switzerland to 166 weeks in Estonia. 1 The United States is the only OECD country without a national paid maternity leave program; however, there are several countries, in addition to the U.S., 1 Paid leave in the OECD cited figures includes “paid maternity leave” and “paid parental and home care leave available to mother” (see OECD 2016, p. 1 for definitions). 10 that do not extend paid leave provisions to new fathers (OECD, 2016). At the national level, the 1993 Family and Medical Leave Act (FMLA) in the U.S. provides 12 weeks of unpaid, job- protected leave to some workers and five states (California, New Jersey, New York, Rhode Island, and Washington) have state-level paid leave laws. Debate surrounding national paid family leave intensified in the United States in recent years, when both major parties’ candidates in the 2016 presidential election touted competing paid leave proposals (Sholar, 2016). Much of the literature on paid leave and work in the U.S. focuses on impacts of these polices on parental labor market activity, such as employment and leave-taking. For example, studies find that paid family leave in California increased leave-taking among both mothers (Rossin-Slater, Ruhm & Waldfogel, 2013) and fathers (Bartel et al., 2017; Baum & Ruhm, 2016), as well as women’s labor force attachment (Byker, 2016). Paid leave policies may also influence couples’ fertility behavior, although the effect may be unintended. Empirical evidence on paid leave and fertility has primarily focused outside the U.S. context. Gauthier’s (2007) review highlights several micro-level studies that demonstrate a generally positive (albeit small) relationship between paid leave and fertility, but these studies are primarily descriptive. Although a handful of more recent studies estimate the impact of laws in the U.S. (Cannonier, 2014; Rossin, 2011) and California (Lichtman-Sadot, 2014) on fertility and fertility timing, no study explores the sustained impact of California’s policy on fertility outcomes. The present study builds on the existing literature by estimating longer-term effects of California’s policy and using a direct measure of socioeconomic status (earnings) as well as other proxies, such as education. Specifically, the present study adds to the literature on paid leave policies in the United Sates by asking two distinct questions: 1. Did California’s PFL have a long-term impact on fertility behavior? 2. Are policy effects heterogeneous across socioeconomic background? 11 To answer these questions, I exploit the natural experiment generated by the state-level policy change to estimate the impact of paid family leave on birth likelihood using difference-in- differences (DID) estimation and data from the Current Population Survey (CPS). These are important contributions as they help us understand whether the effects of the policy diminished over time and whether any sustained effects vary with social class. Results indicate there is a positive effect of California’s PFL on birth likelihood for women ages 25 to 40. The data show a 0.40 percentage point increase in the probability a woman has a child younger than 12 months (i.e. a birth in the last year) in California after the policy was enacted, which is equivalent to a 6.0 percent increase over the baseline for the full sample and 7.6 percent for the sample of women working fulltime. These effects vary by social class. In the subsample of women working fulltime, PFL increased birth likelihood among women with annual earnings in the 25 th to 75 th percentile and high school graduates. Differences by women’s age group suggest the policy may have reduced childbirth delay, rather than increased completed fertility. These findings provide evidence that PFL may benefit working mothers not only by improving their labor market outcomes, but by allowing them to have more children or have them earlier in their careers, improving maternal and infant outcomes and reducing healthcare costs. BACKGROUND Demographic Trends The literature documents a decline in fertility in the U.S. and other Western countries beginning in Europe in the late nineteenth century (e.g. Lee, 2003). The demographic transition occurred as countries modernized and both mortality and fertility declined (Kirk, 1996). In the 1970s, the fertility decline in the U.S. was coupled with a rise in women’s labor force participation. 12 Figure 1 displays the decline in total fertility and rise in labor force participation in the United States between 1970 and 2014. Figure 1: U.S. Women’s Labor Force Participation and Total Fertility Rate, 1970 to 2014 Source: Center for Diseases Control, National Vital Statistics Report, Total Fertility Rate, All races, Table 4: Birth rates by age of mother; U.S. Bureau of Labor Statistics, Civilian Labor Force Participation Rate: Women [LNS11300002], retrieved from FRED, Federal Reserve Bank of St. Louis; https://fred.stlouisfed.org/series/LNS11300002. While fertility continued to drop to very low levels in many European countries, overall U.S. fertility remained close to the replacement level (2.1 births per woman) until recently (Hamilton et al., 2014). Some evidence indicates that fertility follows a pro-cyclical pattern and is expected to decline during recessions due to postponement and fertility often recovers along with the economy (Sobotka et al., 2011). Thus, the most recent drop in US fertility may signal permanently lower fertility or a temporary decline due to the Great Recession in 2008 when couples postponed having children (Cherlin et al. 2013). Research also indicates that some public policies aimed at reducing tension between working and raising children can have a positive effect on fertility (Gauthier, 2007). While the U.S. does not have a national policy for paid leave, California provides paid family leave to new parents at the state level. California’s policy has been shown to have a positive effect on the labor market TFR (left scale) LFP (right scale) 40.0 45.0 50.0 55.0 60.0 65.0 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 1970 1974 1978 1982 1986 1990 1994 1998 2002 2006 2010 2014 % of Women In Labor Force Births per Woman Replacement level = 2.1 13 outcomes of new parents (Bartel et al., 2017; Baum and Ruhm, 2016; Rossin-Slater et al., 2013), but it may also influence fertility behavior (Lichtman-Sadot, 2014). Paid Family Leave in California In 2004, California became the first state to offer Paid Family Leave (PFL) for the birth or adoption of a child or to care for a sick family member. Through PFL, eligible workers can take up to six weeks of paid leave for bonding with a child or caregiving, at 55% of their earnings up to a cap of $1,173 per week in 2017 (State of California EDD, 2017). The eligibility requirements are relatively minimal, workers must: be employed, have earned at least $300 with SDI deductions over the past 12 months, and provide evidence of the “serious health condition” or birth/adoption of a child. 2 The leave may be taken all at once or intermittently over a 12-month period. PFL is funded exclusively by an employee-paid tax and run through the State Disability Insurance program. PFL does not protect workers’ jobs while they are on leave. The federal Family and Medical Leave Act (FMLA) and California’s Family Rights Act (CFRA) provide 12 weeks unpaid, job-protected leave for caregiving and bonding (State of California EDD, 2017; US Department of Labor, 2012). Many workers in California use FMLA/CFRA and the PFL leave programs concurrently; however, FMLA and CFRA have more restrictive eligibility requirements than PFL. Only workers who have been with their employer for at least 12 months, worked at least 1,250 hours over the period and whose employer has 50 or more employees qualify for FMLA and CFRA (State of California EDD, 2017; US Department of Labor, 2012). 3 California employers subject to 2 Just under 70% of women ages 25 to 40 in California in the sample report earning $300 or more in the last year. 3 There are slight differences between CFRA and FMLA eligibility that apply to definitions for caregiving leave, rather than bonding leave. 14 FMLA and CFRA can require employees to take the leaves concurrently. In total, only about 55 to 60 percent of US workers are covered by FMLA (Jorgensen & Appelbaum, 2014; Klerman et al., 2012). Thus, although some workers may benefit from PFL, they may be unable to take the leave if their employers are not covered by FMLA and refuse to protect the employee’s job while on leave. This discrepancy in eligibility may affect who uses the policy. Bedard and Rossin-Slater (2016) recently produced a report for the State of California Employment Development Department using administrative data on paid family leave claims and claimants describing leave take-up since the beginning of the program. They find bonding claims comprise the majority (about 90 percent) of all claims. Women are far more likely to take leave than men, but claims for both women and men have increased since the program’s inception. The authors estimate that about 36 percent of new mothers and 4 percent of new fathers who were employed took the leave in 2004. By 2014, the share of leave-takers rose to 45 percent of new mothers and 9 percent of new fathers who were employed. Although the percentage (and number) of leave-takers increased significantly over the 10-year period, the overall share of employed (and likely eligible) new parents accessing PFL remains low, less than 50 percent. Theoretical Framework Economic theories of fertility assume that couples make tradeoffs between children and consumption (Becker, 1981, 1991; Willis, 1974). In a static model, rational couples maximize their utility subject to a budget constraint, which implies that as the price of children increases relative to other goods, demand for children will decrease. Conversely, as wages increase, demand for children (and other goods) increases. A life cycle approach to fertility indicates that changes in income at a given age will impact the couple’s decision to have a child at the age, but will not 15 necessarily influence their completed fertility (Becker, 1981, 1991). When a couple has a child, a new mother may take time out of the labor force. During this time, her general and firm-specific skills may deteriorate, leading to lower lifetime earnings. Thus, a woman with higher wages may postpone birth early in her career during the time of human capital accumulation to lessen the impact on her lifetime earnings, while lower wage women may not delay childbearing (Hotz, Klerman & Willis, 1997). On the other hand, as wages increase, couples may invest more in child quality than child quantity and this effect may be stronger than the income effect (Becker, 1981, 1991; Hotz et al., 1997). The utility-maximization model rests on assumptions of rational decision-making and homogenous preferences (Becker, 1981, 1991; Easterlin, 1975). In the case of fertility, these assumptions may not hold or there may be other underlying variables affecting the decision to have children. Alternative models of childbearing and work consider individual preferences (e.g. Hotz et al., 1997) and societal norms (Easterlin, 1975). For example, a woman’s preferences may influence the number of children she has or the timing of her childbearing, irrespective of the price (or opportunity cost) of having a child (Hotz et al., 1997). If preferences are heterogeneous across women, estimates of fertility responses to public policies may be biased if they cannot account for these differences. Easterlin’s (1975) relative income hypothesis posits young couples’ “material aspirations” are driven by their parents’ generation. If their income is high relative to their parents (e.g. if more jobs are available), they will exceed the standard of living of their youth. This higher relative income allows them to get married and have children at higher rates than their parents, leading to a change in norms and an overall increase in fertility. Paid leave policies can reduce the opportunity cost of having children, by replacing wages during the time surrounding the birth of a child. The life cycle model indicates reducing the 16 opportunity cost of children may lead to an increase in the demand for children at a given age by discouraging motherhood delay (particularly among higher wage women), but it may not impact completed fertility (Becker, 1981, 1991; Hotz et al., 1997; Willis, 1974). However, if couples substitute child quality for child quantity, paid leave may not lead to an increase in fertility levels (Becker, 1981, 1991). On the other hand, if the fertility decision is driven by individual preferences for children, rather than rational decision-making, paid leave may have no effect on their childbearing. Finally, if policies act to change preferences, increase jobs, or otherwise raise the relative income of young adults, we would anticipate higher overall fertility. However, if policies increase material aspirations without increasing relative income, we would anticipate a decline in fertility. Thus, the impact of paid leave on fertility remains an empirical question. Family Policies and Labor Market Outcomes Child care subsidies, parental leave, cash allowances, and other family policies typically aim to reduce conflict between work and family life for parents, especially mothers. Studies explore the impact of these policies on a range of outcomes. Empirical research on maternity and paternity leave generally reveals positive effects of the provision of leave on leave-taking (e.g. Baum & Ruhm, 2016; Rossin-Slater et al., 2013), labor force participation (Jaumotte 2004), women’s wages (Ruhm, 1998; Waldogel, 1998), return to work (Lalive & Zweimuller, 2009; Rønsen & Sundstrom, 2002; Waldfogel, Higuchi & Abe, 1999), infant and mother health (Baker & Milligan, 2010; Rossin, 2011; Stearns, 2015), and longer-term child outcomes (Lichtman-Sadot & Bell, 2017; Ruhm, 2000). However, the effects vary depending on the length of leave. For example, Ruhm (1998) uses policy variation in 9 countries to estimate the labor market effects of parental leave. He finds short leave has a positive effect on the employment-to-population ratio of 17 women, while longer leaves (1 year or more) negatively affect their wages. More recently, Olivetti and Petrongolo (2017) confirm the nonlinear effect of parental leave (whether paid or not) on female employment – with a positive impact up to 50 weeks of leave. Early studies in the US context focused on the labor market effects of state parental leave (e.g. Klerman and Leibowitz, 1997) and FMLA (Waldfogel 1999), finding access to unpaid leave increased leave-taking among new mothers (Klerman & Leibowitz, 1997; Waldfogel 1999) and fathers (Han, Ruhm and Waldfogel, 2009) with little impact on the employment and earnings of mothers (Waldfogel, 1999). In a more recent study, Kerr (2016) finds evidence of variation in the effect of state and federal family leave provisions by income level: while leave-taking increased for all mothers, the gap in leave-taking between low-income and high-income mothers widened after leave reforms. A growing body of literature examines the impact of California’s PFL using difference-in- differences estimation. Research demonstrates that PFL improves leave-taking and subsequent labor market outcomes for both men and women (Baum & Ruhm, 2016; Rossin-Slater et al., 2013). For example, Baum and Ruhm (2016) find that the policy increased fathers’ leave taking by 32 to 48 and increased the likelihood of mothers returning to work 7 to 12 months after giving birth. In contrast to the unpaid federal policy, California’s policy had a stronger effect on the leave-taking of less advantaged populations, e.g. non-college educated and single mothers (Rossin-Slater et al., 2013). Family Policy and Fertility Outcomes While research on parental leave policies in the United States primarily examines the labor market outcomes of new parents, changing fertility patterns and comprehensive family policies 18 abroad sparked greater interest in the role these policies play in couples’ fertility decisions. Though few policies have the explicit goal of raising fertility, apart from baby bonuses in places like Australia (Sinclair, Boymal & DeSilva, 2010), theory indicates they may influence completed fertility or fertility timing. Many early studies exploited cross-national policy variation to estimate the effect of parental leave on fertility levels, using total fertility rate or other similar national measures (for a review of early research see Gauthier, 2007). This line of research produced mixed results on the effect of family policies on fertility, finding positive (e.g. Bjorklund, 2006; Finch & Bradshaw, 2003) or null (Castles, 2003) effects. However, many studies included policies other than leave, such as childcare subsidies or cash allowances. In one such study, Gauthier and Hatzius (1997) find a positive effect of cash allowances on fertility, but no effect of maternity leave. They attribute their null finding to a lack of variation in policies at the time. However, more recent studies reveal a positive relationship between paid leave and total fertility rate in Western countries (Luci- Greulich & Thevenon, 2013; Thevenon & Gauthier, 2011). Individual-level studies using micro-data generally find a small, positive association between parental leave and individual-level fertility (Gauthier 2007). These studies reveal positive effects on both the probability (Lalive & Zweimuller, 2005; Lee, Ogawa & Matsukura, 2009; Rønsen 2004) and timing (Hoem, Prskawetz & Neyer, 2011; Lalive & Zweimuller, 2005) of births in several countries where leave is both paid and longer than in the US; however, the effects often vary by birth parity (e.g. Lee et al., 2009; Rønsen, 2004). Yet, identifying the effect of a public policy on couples’ fertility decisions presents some analytical challenges. Childbearing and work decisions may be jointly determined or influenced by unobserved characteristics. The endogeneity of work in the fertility decision may, in turn, bias estimates of labor policy effects on fertility. 19 Researchers accounting for this endogeneity using panel data methods have found positive effects of access to leave on births in the United States (Averett & Whittington, 2001) and Japan (Lee et al., 2009). Researchers also exploit exogenous changes in parental leave policies to identify a causal impact on fertility, using regression discontinuity (e.g. Lalive & Zweimuller, 2005) and difference- in-differences (DD) estimation (e.g. Cannonier, 2014; Lichtman-Sadot, 2014; Rossin, 2011). For example, Rossin (2011) estimates the impact of the national unpaid FMLA on infant mortality and fertility, finding an increase in first births after the policy; however, Rossin’s repeated cross- sectional data cannot directly account for the endogeneity of the work and childbearing decision, since the data does not follow the same people over time. More recently, Cannonier (2014) reveals a similarly positive relationship between FMLA and birth probability and timing using panel data and DD. FMLA increased the probability of a first birth by 1.8 percent per year and 1.1 percent for a second birth. The policy change also reduced childbearing postponement. To my knowledge, only one study explores the relationship between paid leave and fertility in the US context, using California’s Paid Family Leave program (Lichtman-Sadot, 2014). Lichtman-Sadot (2014) estimates the impact of PFL on pregnancy timing, finding that births increased by 1.7 percent in the second half of 2004 (after policy enactment) and there is evidence of a larger effect among Hispanic and lower educated women. In sum, the literature indicates parental leave can positively affect fertility, particularly in contexts where the leave length and replacement rates are generous. Even in the U.S., where the federal policy is unpaid, research indicates a positive fertility response to changes in family leave policy; however, there are some differences by parity and socioeconomic status. In terms of paid leave in the U.S., Lichtman-Sadot’s (2014) study demonstrates an immediate fertility response to 20 California’s PFL enactment, but the longer-term effects are unknown. Moreover, studies on the effects of parental leave on labor market outcomes in the U.S. reveal significant differences across socioeconomic groups (e.g. Kerr, 2016; Rossin-Slater et al., 2011). The present study adds to the literature by analyzing whether California’s policy had a sustained effect on fertility decisions and whether the effect varies for different socioeconomic groups, using a large dataset with direct measures of individual earnings. EMPIRICAL STRATEGY Data This study uses nationally representative data from the U.S. Census Bureau’s Current Population Survey (CPS) March survey and the Annual Social and Economic (ASEC) supplement. CPS and the supplemental data are made available online by the University of Minnesota through the Integrated Public Use Microdata System (IPUMS) (Flood et al, 2015). The monthly CPS survey includes demographic data, such as gender, age, number of children, and age of youngest child. The ASEC includes information on individuals’ wages and household income, which are crucial to estimating differences across socioeconomic status. Analytical samples for fertility studies commonly include all women of childbearing ages 15 to 44 or 49; however, the fertility response to policy changes is likely quite different among the youngest and oldest mothers who may have less deliberate or controlled fertility. Further, this study aims to estimate the effect of PFL on the fertility of women in their prime working (having completed schooling) and childbearing ages. Thus, I restrict my sample to women 25 to 40. The final sample includes a total 117,031 observations from 1999 to 2009. In robustness checks, I include county-level unemployment statistics from the Local Area Unemployment Statistics (LAUS), as well as the 21 median household income and share of immigrants and Hispanics generated from the full CPS sample of men and women 15 and over. Dependent Variable I construct the dependent variable using the age of the youngest own child in the household. The variable equals one if the youngest child is identified as less than one year old and zero otherwise. Since the sample draws from the March survey, the variable identifies children born between April of the prior year and March of the survey year. 4 For example, a woman who is identified as having a baby in 2004 gave birth (or adopted) her child between April 2003 and March 2004. I also analyze the effect of PFL on higher order births, by creating a variable that equals one if the youngest child is less than one year old and there is more than one child in the household. Independent Variables The two primary independent variables are dummy indicators of whether a woman lives in California and whether the survey takes place in the post-PFL period (after June 2004). There is a three-month overlap in the data in the year 2005. That is, all births in the 2005 survey are included in the post-PFL period. However, the data could include births in April, May, and June 2004; I test the sensitivity of my results to this assumption by excluding 2005. 4 The CPS has a biennial fertility supplemental survey with the precise birth months and years of children in the household, as well as the mother’s age at first birth. While this data could work for the present study, its less frequent occurrence and lack of earnings data make it less suitable for analyzing the impact of SES status. I tested the robustness of my main results by re-analyzing the models using the fertility supplement data. The results are qualitatively and quantitatively similar. 22 Other independent variables account for factors the literature indicates are relevant to the fertility decision — demographic and socioeconomic determinants of fertility. Control variables include: age, age-squared, race/ethnicity, immigrant status, marital status, presence of other children, education level, labor force participation, and annual earnings. Labor force participation identifies whether a woman is in the labor force and immigrant status is determined by whether she was born outside the United States. Specification I exploit the natural experiment created by the enactment of Paid Family Leave in California in 2004 using difference-in-differences (DD). The DD strategy compares outcomes in treatment and control groups before and after a policy change, all else equal. In this case, women ages 25 to 40 living in California comprise the treatment group and observations in California from 2005 to 2009 are “treated” by the policy enactment. The control group consists of women ages 25 to 40 living in control states (described below) between 1999 and 2009. I estimate the effect of the policy change on fertility behavior using a standard DD design with covariates and ordinary least squares: 𝑦 "#$ = 𝛼+𝛽 ) 𝐶𝑎𝑙𝑖𝑓𝑜𝑟𝑛𝑖𝑎 "$ +𝛽 2 𝑃𝑜𝑠𝑡 $ +𝛽 6 𝐶𝑎𝑙𝑖𝑓𝑜𝑟𝑛𝑖𝑎 "$ ∗𝑃𝑜𝑠𝑡 $ +𝛾𝑋 "#$ +𝜀 "#$ Where 𝑦 "#$ = 1 if individual i living in state s has a child under 12 months old in year t. 𝐶𝑎𝑙𝑖𝑓𝑜𝑟𝑛𝑖𝑎 "$ = 1 if individual i lives in California in year t. 𝑃𝑜𝑠𝑡 $ = 1 in years after the 2004 enactment of California’s PFL. 𝑋 "#$ is a vector of individual-level demographic and socioeconomic indicators, as described above. 𝜀 "#$ are individual-level standard errors, which are clustered at the state (treatment) level to account for correlation within states (Angrist & Pishke, 23 2008). 𝛽 6 is the DD estimate of the effect of California’s PFL on birth likelihood among women in California. DD designs hinge on having sufficiently comparable treatment and control groups (Angrist & Pischke, 2008). The treatment group for this study is women ages 25 to 40 in California. Selecting an appropriate control group for a state to satisfy the key identifying assumption of common pre-policy trends presents several challenges. No two states experience the same trends over time and different state-level policies, demographic, and economic trends may confound estimates of the effect of PFL. Baum and Ruhm (2016) developed a procedure to construct a control group for estimating PFL’s effect on leave-taking, using regression analysis to identify states with statistically similar pre-PFL trends in employment. 5 This ensures that any differences in the post-PFL period can be attributed to the policy, rather than trends established before the policy. I adapt Baum and Ruhm’s strategy to determine the control group for the present study. I estimate a separate regression for each state in the US, comparing pre-PFL trends in fertility. Specifically, I estimate the following for each state: 𝑦 "#$ = 𝛼+𝛽 ) 𝑂𝑡ℎ𝑒𝑟 𝑆𝑡𝑎𝑡𝑒 "$ +𝛽 2 𝑝𝑟𝑒_𝑃𝐹𝐿 $ +𝛽 6 𝑂𝑡ℎ𝑒𝑟 𝑆𝑡𝑎𝑡𝑒 "$ ∗𝑝𝑟𝑒_𝑃𝐹𝐿 $ +𝛾𝑋 "#$ +𝜀 "#$ Where 𝑦 "#$ = 1 if individual i living in state s has a child under 12 months old in year t; 𝑂𝑡ℎ𝑒𝑟 𝑆𝑡𝑎𝑡𝑒 "$ = 1 if individual i lives in the non-California state in year t and 0 if he/she lives in California; 𝑝𝑟𝑒_𝑃𝐹𝐿 $ is a linear time trend in the years before PFL (1999 to 2004); and 𝑋 "#$ is a vector of individual controls as described above. 𝛽 6 is the coefficient of interest, which indicates whether there is a difference in fertility trends in California and the other state in the period before 5 Baum and Ruhm (2016) use this method in their sensitivity analysis and describe the procedure and results in Appendix B of their paper. In their sample, pre-PFL trends for all states were similar and they use all states as their control. In the present study, pre-PFL trends in fertility are significantly different in California when compared to all other states combined. Only states with statistically different pre-PFL fertility trends are included in the control group for this study. 24 PFL was enacted. When 𝛽 6 is not statistically significant, the state is considered to have similar pre-PFL fertility trends and is included in the control group. The final control group consists of 16 states (see Appendix A for results from the models). This is conceptually similar to the synthetic control group method developed by Abadie, Diamond and Hainmueller (2010), but the procedure is different and I do not assign different weights to the control states in the regression model. In a sensitivity analysis, I compare my main findings to results obtained using the synthetic control method. I first analyze the effect of PFL on the full sample of working age women. In subsequent analyses, I restrict the sample to women who worked fulltime in the last year to analyze the effect of the policy on working women. For the restricted sample, I stratify the main specification by socioeconomic and demographic categories — earnings, education level, race/ethnicity, and age. I compare DD coefficients across samples to determine whether the policy effects are heterogeneous across population subgroups. RESULTS Descriptive Analysis Figure 1 above showed the decline in the overall fertility rate in the United States since the beginning of Great Recession. The total fertility rate (TFR) in Figure 1 is a hypothetical estimate of fertility based on current age-specific trends, while the current study relies on a measure of actual births. The sample in the present study is restricted in two key ways: (1) women are ages 25 to 40; and (2) the sample covers survey years 1999 to 2009. Thus, the decline in overall fertility during the recession is not included in the sample period. 25 Figure 2: Birth likelihood in California and Control States Before and After PFL Panel A: California (Treatment Group) Panel: B Control Group Source: IPUMS, Current Population Survey (CPS) March and Annual Social and Economic Supplement Surveys, 1999-2009 Notes: Control states – Alaska, Arizona, Colorado, Georgia, Hawaii, Iowa, Kansas, Michigan, Minnesota, New Jersey, New York, Oklahoma, Pennsylvania, Tennessee, Texas, and Virginia. Sample restricted to women ages 25-40 in treatment and control states. All dollar figures are in 2015 dollars. ***p<.001, **p<.01, *<.05, +p<.10 indicates the significance of differences in means before and after PFL for each group. Figure 2 displays trends in the dependent variable for the sample of women ages 25 to 40 in the treatment and control states. Panel A illustrates birth likelihood in California, while Panel B represents the control group. Prior to the 2004 enactment of the policy, birth likelihood in both groups is relatively stable and follows a similar trend, with a slight incline for both. After California’s PFL, there is an immediate and noticeable increase in birth likelihood in California that remains above pre-policy levels, though it declines during the recession (Panel A). In contrast, birth likelihood in the control group appears to continue on an upward trajectory before and after 4.0% 5.0% 6.0% 7.0% 8.0% 9.0% 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Women with birth in last 12 months 4.0% 5.0% 6.0% 7.0% 8.0% 9.0% 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Women with birth in last 12 months 26 2004 (Panel B). The differences in trends in the post-policy period, could be due to several factors, including demographic and economic trends. Multivariate analysis will disentangle changes in the overall population from changes in fertility behavior after PFL was enacted. Table 1: Table of Means, Women ages 25 to 40, Treatment (CA) and Controls Groups (Other States), Before (1999-2004) and After (2005-2009) Paid Family Leave Full Sample Treatment Group (CA) Control Group (non-CA) 1999-2010 1999-2004 2005-2010 1999-2004 2005-2010 Baby in last 12 mos. (%) 7.36 6.62 7.89 *** 6.98 7.93 *** Age (years) 32.73 32.80 32.66 * 32.86 32.57 *** Race/Ethnicity NH White (%) 56.93 42.96 37.85 *** 63.31 60.08 *** NH Black (%) 12.34 6.00 6.43 14.33 14.04 Hispanic (%) 21.85 36.32 40.22 *** 16.08 17.99 *** Asian/PI (%) 7.39 13.61 13.87 5.12 5.91 *** NH Other (%) 1.49 1.11 1.63 ** 1.40 1.39 *** Married (%) 60.78 59.99 60.44 61.84 59.91 *** # of children 1.4 1.43 1.37 ** 1.40 1.39 Education Level Less than HS (%) 12.55 18.37 18.26 10.85 10.74 HS Grad (%) 26.32 22.52 20.54 *** 29.51 25.59 *** Some College (%) 28.38 29.82 27.54 *** 28.39 28.15 BA+ (%) 32.76 29.30 33.85 *** 31.25 35.52 *** Annual earnings ($) 27,486 27,730 27,976 27,294 27,480 Labor force participation (%) 73.79 70.80 70.00 75.34 74.16 *** Employed (%) 70.06 66.32 65.60 71.95 70.50 * Foreign born (%) 23.54 40.20 41.18 17.28 19.69 *** N 117,031 11,974 11,448 50,028 43,581 Source: IPUMS, Current Population Survey (CPS) March and Annual Social and Economic Supplement Surveys, 1999-2009 Notes: Control states – Alaska, Arizona, Colorado, Georgia, Hawaii, Iowa, Kansas, Michigan, Minnesota, New Jersey, New York, Oklahoma, Pennsylvania, Tennessee, Texas, and Virginia. Sample restricted to women ages 25-40 in treatment and control states. All dollar figures are in 2015 dollars. ***p<.001, **p<.01, *<.05, +p<.10 indicates the significance of differences in means before and after PFL for each group. Table 1 compares the mean values in the treatment and control groups for all variables in the analysis, including a test of the statistical significance within a group before and after PFL. 27 The results demonstrate differences between women in California and the control group. Women in California in both periods are more likely to be Hispanic, born outside the United States, and have less than a high school degree, but less likely to participate in the labor force or be employed than women in the control group. These differences between women in the two groups indicate the need for including covariates in the DD model. On the other hand, statistically significant differences before and after the policy within each group highlight the importance of controlling for time-varying individual factors that could be (incorrectly) associated with fertility changes if not included in the model. For example, an increase in the share of Hispanic women in the treatment group could lead to an increase (or decrease) in fertility if their fertility patterns are different from women of other racial/ethnic backgrounds in the treatment group. Multivariate Analysis Table 2 reports the results from linear probability models of the likelihood a woman has a child under 12 months old (i.e. birth likelihood). 6 Model 1 includes only the dummy indicators for California and post-PFL period, along with their interaction. Model 2 adds individual-level covariates as indicated above, while Model 2a is restricted to the sample of women identified as working fulltime. In both models, PFL led to an increase in the likelihood a 25 to 40-year-old woman had a child younger than 12 months in California, as compared to the control states. The effect is larger in magnitude and more significant in Model 2 – birth likelihood increased by 0.40 percentage points among women in California after PFL, which is equivalent to a 6.0 percent increase over the baseline, pre-PFL probability. For women working fulltime, the coefficient 6 I use linear probability models, rather than logit or probit, in the main specifications due to the more straightforward interpretation of interaction terms in the linear model (Ai & Norton, 2003). I tested probit and logit specifications and the interaction term is positive and significant at the .01% level. 28 magnitude is slightly smaller, representing a 0.36 percentage point increase in birth likelihood in California after PFL; however, compared to the lower probability of birth among these women in the pre-PFL period (4.7 percent), the DD estimate is equivalent to a 7.6 percent increase in birth likelihood for these women. Table 2: Linear Probability Models of Birth Likelihood, Women age 25 to 40, Treatment (CA) and Control (non-CA), 1999-2009 Model 1 (Full Sample) Model 2 (Full Sample) Model 2a (Fulltime Sample) California*Post-2004 0.0031 + 0.0040 * 0.0036 * California -0.0036 ** -0.0040 * -0.0026 + Post-2004 0.0095 *** 0.0070 *** 0.0118 *** Age 0.0124 ** 0.0192 *** Age 2 -0.0003 *** -0.0004 *** Race/ethnicity (ref. NH white) NH Black -0.0044 -0.0087 ** Asian/PI -0.0098 + -0.0022 Hispanic -0.0021 -0.0045 + NH Other -0.0063 -0.0062 Married 0.0541 *** 0.0568 *** Presence of other children 0.0641 *** 0.0507 *** Foreign born 0.0027 -0.0018 Education level (ref. <High school) HS Grad 0.0033 0.0048 Some College 0.0133 *** 0.0139 *** BA+ 0.0412 *** 0.0377 *** Annual earnings (log) 0.0023 *** -0.0005 In labor force -0.0731 *** Constant 0.0698 *** -0.0374 -0.2170 ** R 2 0.0004 0.0571 0.0410 N 117,031 117,031 68,726 Source: IPUMS, Current Population Survey (CPS) March and Annual Social and Economic Supplement Surveys, 1999-2009 Notes: Control states – Alaska, Arizona, Colorado, Georgia, Hawaii, Iowa, Kansas, Michigan, Minnesota, New Jersey, New York, Oklahoma, Pennsylvania, Tennessee, Texas, and Virginia. Sample restricted to women ages 25-40 in treatment and control states. All dollar figures are in 2015 dollars. Regressions weighted using individual weights. ***p<.001, **p<.01, *<.05, +p<.10 indicates significance. 29 Coefficients on other variables are as expected based on prior research, with few differences between Models 2 and 2a. Within the sample of all women ages 25 to 40, birth likelihood is positively associated with age (and a slightly negative quadratic term), marriage, presence of other children, education, and log wages. The probability of a woman in the sample giving birth is negatively associated with her labor force participation. Women of all race/ethnicities are less likely to have given birth in the last year compared to non-Hispanic white women, although not all coefficients are statistically significant. Parameter estimates are similar among the sample of women working fulltime, with the exception of log earnings, which is not associated with birth likelihood once non-working women are eliminated from the sample. Theory indicates (and empirical studies demonstrate) the effect of parental leave policies on behavior may vary by a woman’s earnings or other socioeconomic and demographic characteristics. Model 2 reveals a positive effect of PFL on overall birth likelihood among women ages 25 to 40 in California. Unemployed women and those out of the labor force are included in the sample in Model 2, but PFL benefits are conditioned on some paid employment in the last year. While Model 2 provides evidence of an estimation of an overall effect, the effect shown in Model 2a may be more relevant as it indicates changes in behavior among women most likely to face work-life conflict. In Table 3, the sample of women ages 25 to 40 working fulltime is stratified by socioeconomic and demographic categories to estimate PFL’s effect on subgroups of working women. Model 2a is estimated separately for each category of women. The stratified models reveal heterogeneous effects among population subgroups. Specifically, PFL led to larger increases in birth likelihood for women with annual earnings in the 50 th to 75 th percentile (1.1 percentage points) and those with at least a high school diploma, but not a bachelor’s degree (0.8 percentage points). 30 A positive effect among women in the 25 th to 50 th percentile of annual earnings was also larger than the overall effect, 0.65 percentage points. Table 3: Heterogeneity for Subgroups, Women ages 25 to 40, Employed Fulltime, Linear Probability Models of Birth Likelihood, Treatment (CA) and Control (non-CA), 1999-2009 CA*Post N All women ages 25 to 40 employed fulltime (from Table 2) 0.0036 * 68,726 Stratified by: Annual Earnings Below 25 th percentile 0.0029 17,349 25 th to 50 th percentile 0.0065 + 18,844 50 th to 75 th percentile 0.0111 * 15,843 Above 75 th percentile -0.0037 16,690 Education Level Less than HS -0.0046 6,303 HS Grad/Some college 0.0079 * 38,652 BA+ 0.0024 23,771 Race/Ethnicity NH White 0.0093 *** 37,611 NH Black 0.0065 8,218 Hispanic -0.0013 15,677 NH Asian 0.0118 * 5,528 NH Other 0.0087 1,692 Age Group 20-24 -0.0067 13,381 25-29 -0.0055 19,145 30-34 0.0118 * 21,410 35-39 0.0083 * 23,128 40-44 -0.0029 * 24,704 Source: IPUMS, Current Population Survey (CPS) March and Annual Social and Economic Supplement Surveys, 1999-2009 Notes: Control states – Alaska, Arizona, Colorado, Georgia, Hawaii, Iowa, Kansas, Michigan, Minnesota, New Jersey, New York, Oklahoma, Pennsylvania, Tennessee, Texas, and Virginia. Sample restricted to women ages 25-40. All dollar figures are in 2015 dollars. Cutoffs for wages and income based on women ages 25-40 working fulltime before 2004. All specifications include: age, age 2 , marital status, other children, immigrant, and annual earnings. Categorical variables are excluded in stratifications of that variable (e.g. education level). Regressions weighted using individual weights. ***p<.001, **p<.01, *<.05, +p<.10 indicates significance. The demographic stratifications highlight differential effects by race and age group over time. While the policy did not impact fertility among Hispanic or non-Hispanic black women ages 31 25 to 40, birth likelihood significantly increased among non-Hispanic white and Asian women in California after PFL. The life cycle model indicates public policies may not impact completed fertility, but could alter fertility timing. The results of the model stratified by 5-year age groups suggest PFL may have affected fertility timing among women in California. While there is no effect of PFL on women in younger age groups, women in their 30s were significantly more likely to experience a birth after the policy and women in the 40s were less likely. In a model with a three-way interaction between the post-PFL dummy, California, and age group, women ages 40 to 44 were 1.2 percentage points less likely to have a child after PFL in CA relative to women in their 30s. This difference could indicate a reduction in childbearing delay; however, panel data would be needed to determine these differences on an individual level. ROBUSTNESS CHECKS Alternative Dependent Variables Table 4 displays the results for models using alternative dependent variables. Some studies in the U.S. context find the effect of maternity or family leave on fertility is larger (e.g. Averett & Whittington, 2001) or smaller (e.g. Cannonier, 2014) for higher order births. In the main specification, I include the likelihood of any birth in the dependent variable. I also test whether the effect is different for births to mothers with at least one other child. The DD magnitude of the PFL effect on higher order births is slightly smaller than the estimate for any birth, 0.36 percentage points versus 0.40 in the main specification. Following research on labor supply effects of PFL, I estimate the impact of PFL on women’s labor force participation conditional on having a child in the last year. Consistent with the literature, I find a positive effect of PFL, 2.6 percentage points (or about 3.7 percent), on the likelihood a new mother is in the labor force compared to the control group with similar pre-PFL trends in birth likelihood. I also generated a control group that has 32 similar pre-PFL trends in labor force participation of new mothers (see Table A4 in the Appendix). When I estimate Model 2 using this control group, a new mother’s labor force participation in California increased by 4.5 percentage points (or about 9.4 percent) after PFL, compared to the control group. Table 4: Alternative Dependent Variables, Linear Probability Models, Women ages 25 to 40 Dependent Variables CA*Post N Any birth in last 12 months (from Table 2) 0.0040 * 117,031 2 nd or higher order birth 0.0036 * 117,031 Labor force participation, conditional on birth in last 12 months (birth states) 0.0259 * 8,920 Labor force participation, conditional on birth in last 12 months (LFP states) 0.0449 ** 7,747 Source: IPUMS, Current Population Survey (CPS) March and Annual Social and Economic Supplement Surveys, 1999-2009 Notes: Control states (birth sample) – Alaska, Arizona, Colorado, Georgia, Hawaii, Iowa, Kansas, Michigan, Minnesota, New Jersey, New York, Oklahoma, Pennsylvania, Tennessee, Texas, and Virginia. Control states (LFP sample) - Sample restricted to women ages 25-40 in treatment and control states. All dollar figures are in 2015 dollars. Regressions weighted using individual weights. ***p<.001, **p<.01, *<.05, +p<.10 indicates significance. I test the robustness of my findings to the inclusion of county-level trends, state and year fixed effects, a linear trend for years since PFL enactment, and dummy variables for each post- policy year to account for the non-linearity in post-PFL fertility found in the raw data. 7 The results of specifications are displayed in Table 5. All covariates are the same as in Model 2, with adjustments as indicated in each row. The findings are robust to the inclusion of county-level trends, state and year fixed effects, and the linear trend for years since PFL enactment. The county-level variables and linear trend slightly attenuate the effect, but state and linear fixed effects produce the same outcome as the main model. The individual dummy variables for each post-PFL year indicate the positive impact of the policy is concentrated in the first two years after PFL was enacted, after 7 I also tested the sensitivity of the results to the exclusion of the year after the policy (2005) since there is potentially a three-month overlap in births and the exclusion of the year prior to the policy (2004) as the policy announcement may have led to changes in behavior due to anticipation of future benefits. The results are similar: without 2005 (2004), PFL led to a 0.33 (0.34) percentage point increase in birth likelihood. 33 which the effect appears to wear off, suggesting the policy effect may not have been sustained over time. Table 5: Robustness Checks, Linear Probability Models of Birth Likelihood, Women age 25 to 40, Treatment (CA) and Control (non-CA), 1999-2009, Model 2 plus CA*Post SCM Model 2, plus: County-level trends 0.0036 * Linear trend for years since policy 0.0040 * State and year fixed effects 0.0038 * Post-year interactions, ca*year (ref. 1999-2004) 2005 0.0069 + 0.0083 2006 0.0098 *** 0.0030 2007 -0.0036 + -0.0124 2008 0.0048 0.0018 2009 0.0023 -0.0117 Source: IPUMS, Current Population Survey (CPS) March and Annual Social and Economic Supplement Surveys, 1999-2009 Notes: Control states – Alaska, Arizona, Colorado, Georgia, Hawaii, Iowa, Kansas, Michigan, Minnesota, New Jersey, New York, Oklahoma, Pennsylvania, Tennessee, Texas, and Virginia. Sample restricted to women ages 25-40 in treatment and control states. All dollar figures are in 2015 dollars. Regressions weighted using individual weights. ***p<.001, **p<.01, *<.05, +p<.10 indicates significance. Finally, I test the sensitivity of the results to the selection of control states. I estimate models using control groups identified in other work on PFL either in the main specification or as robustness checks, including: all other states (Lichtman-Sadot, 2014); the next three largest states - Florida, New York, and Texas (Byker, 2016; Lichtman-Sadot, 2014; Rossin-Slater et al., 2013); and other states with Temporary Disability Insurance programs - New York, New Jersey, Rhode Island, and Hawaii (Lichtman-Sadot, 2014; Rossin-Slater et al., 2013). The results are qualitatively similar using these alternative control groups, except the group including all other states, where the DD coefficient is not statistically different from zero (see Appendix, Table A3). Synthetic Control Method 34 The primary control group selection based on Baum and Ruhm’s (2016), hereafter BR, method of matching pre-treatment trends by state is conceptually similar to the Synthetic Control Method (SCM) developed by Abadie and Gardeazabal (2003) and Abadie, Diamond, and Hainmueller (2010); however, some key differences make the BR method more suitable for answering the current research questions. Specifically, the BR method has the advantage of exploiting all individual-level variation in covariates before and after PFL implementation, which is important as we can see clear differences in several determinants of fertility pre- and post-PFL in the treatment and control states. Further, the sample can be stratified and the effects compared across subsamples to analyze heterogeneity using the same methods and control group. The equal weighting of states in the control group, however, means the pre-PFL fertility trend for the treatment and control groups may not perfectly align. In contrast, the primary goal of the SCM is to use “data-driven procedures [to] reduce discretion in the choice of the comparison control units” (Abadie et al., 2010). 8 The procedure generates a counterfactual (control group) that approximates the treated unit in the pre-treatment period. 9 In this case, donor states are matched on covariates in the pre-treatment period to most closely resemble California and weights are assigned to individual states that comprise the synthetic control group. Any differences in post-treatment outcomes between the treated unit and the synthetic control are attributed to the treatment. That is, the annual weighted average of the dependent variable (percentage of women giving birth in the last year) for the synthetic control group in the post-treatment period is compared to the observed annual values in the treatment 8 In practice, there are several decision points in the process for the researcher, each of which can impact the selection of states and the weights assigned in the control group. 9 See Abadie and Gardeazabal (2003) and Abadie et al. (2010) for a detailed explanation of the method. 35 group. The key output of the SCM is a visual representation of the results. 10 The method is considered appropriate for any type of comparative analysis for which there exists no single comparison unit. In addition to the benefit of using “data-driven procedures” to select the control group, the SCM only requires aggregate (treatment-level) data, so it is fairly straightforward to implement across contexts. On the other hand, it does not account for changes in covariates in the post-treatment period, which may impact the outcome in both the treatment and control groups. The adapted version of the BR method (above) is preferred due to its ability to take advantage of disaggregated data, control for post-PFL differences in covariates that impact fertility (e.g. changes in the racial composition or educational attainment in the population), and analyze heterogeneous policy responses. I compare my main results (using the BR method) to those generated using the synthetic control method as an additional check on robustness. Figure 3 below compares the output from the synthetic control analysis and a similar graph using the BR method control group. Panel A illustrates the percentage of women ages 25 to 40 giving birth in California and Synthetic California between 1999 and 2009. Panel B is the same graph for California and the BR method control group. Year-to-year fluctuation in birth likelihood (before and after PFL) is seen in both Synthetic California and the preferred control group. The overall pattern in Figure 3 suggests (as above) that California experienced a sharp increase in birth likelihood immediately after the policy, which stayed above pre-PFL levels. On the other hand, birth likelihood Synthetic California also increased after PFL, at perhaps a slightly slower initial rate. The annual fluctuation highlights the importance of controlling for changes in the individual and population characteristics in the post-PFL period that may be driving changes in fertility. Nevertheless, the 10 I use the synth (Abadie et al., 2010) and synth_runner (Galiani and Quistorff, 2016) packages in Stata, version 13.1, to conduct the synthetic control method and subsequent placebo analyses. 36 similarities between Panels A and B indicate the BR method control group approximates a counterfactual at least as well as SCM. Figure 3: Birth Likelihood, Women ages 25 to 40, 1999-2009, California and Controls Panel A: SCM Control Group Panel B: BR Method Control Group Statistical significance (or “inference”) in the SCM is determined using a series of placebo tests, where the treated unit should ultimately appear as an outlier (Abadie et al., 2010). The visual placebo tests for birth likelihood indicate that the SCM findings for PFL may not be unique to California (see Figures A1 and A2 in the Appendix). As an additional check, I conduct a t-test of means in the post-PFL weighted average of birth likelihood in synthetic California and observed average in California to determine if the post-PFL differences in birth likelihood are significant. I find a positive and significant difference between the treatment and control groups in the first two years after the policy was enacted: birth likelihood was 0.54 percentage points (p<.05) higher in California than in Synthetic California. However, the t-statistic loses significance with subsequent years are added to the analysis (2007 to 2009). I also compare the magnitude of the DD effect in each year of the post-period with the annual effect generated using the SCM and report them in Table 5. As indicated in Figure 3, the pattern of effects from the SCM and the preferred control .04 .05 .06 .07 .08 .09 State-level Birth Likelihood 1999 2001 2003 2005 2007 2009 Year California Synthetic California .04 .05 .06 .07 .08 .09 1999 2001 2003 2005 2007 2009 Birth Likelihood California Control States 37 group are similar after PFL, although the magnitude differs somewhat. In both cases, the effect appears to wear off after the first two years. I interpret these findings as indicating the positive results from the preferred control group are generally consistent with those generated using the SCM. DISCUSSION Understanding the role of policies, like California’s PFL, in reducing tension between work and family life is increasingly important as the share of households with all parents in the labor force has grown in recent decades. This study analyzes the impact of the first state-level paid family leave policy in the U.S. on a family dimension of work-life balance, women’s fertility. While prior research has demonstrated a generally positive effect of PFL on new parents’ labor market outcomes, individuals and couples may also adjust their fertility behavior in response to work demands and policy changes, whether the timing of childbearing or completed fertility. I find that California’s PFL led to a 6.0 percent increase in the likelihood a woman 25 to 40 had a child in the last year, which is robust across multiple specifications, and birth likelihood among women working fulltime increased by about 7.6 percent. Stratified regressions using the subsample of women working fulltime reveal larger magnitude effects among non-Hispanic White women, earning income in the 25 th to 75 th percentile, and high school graduates. These findings indicate middle-class working women may be experiencing the strongest response to this policy. My results differ from Lichtman-Sadot (2014) who found stronger immediate effects of PFL on births for Hispanic, lower-educated and single women; however, data used in that study did not allow for inclusion of labor force participation or income and only accounted for the 6 months directly following the policy’s enactment. On the other hand, the results are consistent with Cannonier’s (2014) estimates of the effects of FMLA using the NLSY79 panel over a longer post- 38 FMLA period. Among eligible women in that study, fertility effects were strongest for white women and those with at least some college. He posits these differences could be due to higher marriage rates among that group of women (relative to lower-wage women), which would make taking the unpaid FMLA leave less costly. However, marriage rates do not differ substantially between women in the upper and lower income quartiles in my sample. One explanation for differences in effects by race/ethnicity could be fertility timing among non-Hispanic black and Hispanic women. Sweeney and Raley (2015) show that non-Hispanic black and Hispanic women in the United States are much more likely to have their children younger than their non-Hispanic white counterparts. For example, by age 20, roughly 33 percent of non-Hispanic black and 30 percent of Hispanic have given birth compared to 14 percent of non- Hispanic white women and many of these young women are also poorer and lower educated. Restricting the sample to women 25 to 40 reduces the likelihood of capturing changes in fertility among these younger mothers and may mask any effect for these racial/ethnic groups. Indeed, there was no effect on birth likelihood for women ages 20 to 24. Coupled with the stronger effects for middle-income and high school graduates, these findings suggest that childbearing among younger women with lower socioeconomic backgrounds is unaffected by PFL, many of whom are non-Hispanic black and Hispanic. Other potential explanations for differences in fertility responses to PFL are the price of children (Becker, 1981, 1991) and women’s preferences for children (e.g. Hotz et al., 1997). While PFL may reduce the opportunity cost of having children, it does not reduce the ongoing cost of children (such as childcare, food, etc.). Thus, women with very low earnings may not be able to afford (more) children despite access to PFL. Likewise, those with higher earnings may be unaffected. These women are more likely have access to paid leave through their employers and 39 plan their childbearing accordingly. They can also likely afford childcare for their intended number of children. On the other hand, middle-income women (i.e. those in the 25 th to 75 th percentiles) may not have previously had access to paid leave and, thus, were induced into childbearing or reduced childbearing delay in response to PFL. A challenge in estimating fertility behavior is the endogeneity of fertility and work decisions. Panel data can be used to disentangle the effect of work (and policy changes) on fertility from unobserved factors that may impact both. This study uses repeated cross-sectional data to estimate the overall effect on fertility, but cannot accurately measure changes in timing or fertility preferences. Larger positive DD effects for women in their 30s and a negative effect for those in the 40s suggests the policy may have reduced childbearing delay. In a simple comparison of the average age of first-time mothers in the years just before and after PFL implementation, there was a slight decline in the age of first-time mothers in California between 2004 and 2005, 0.8 years. There was a slight increase (0.1 years) for women in the control group over the same period. Among working women, new mothers and first-time mothers in California experienced a similar decline in average age, 1.0 and 1.3 years, respectively; however, the change occurred a year after the policy was in place between 2005 and 2006. This delayed effect for working women may reflect their need to plan for a child once the policy was enacted or wait for their employers to figure out how to implement the policy. In my sample, selection bias could occur, especially among women working fulltime, if women select into or out of working and/or having children. For example, women with earnings in the 25 th to 75 th percentile could be employed in more family-friendly jobs, especially in the period after PFL, which could be driven by individual preferences; thus, they experience larger effects of PFL on their birth likelihood. On the other hand, an overall increase in fertility among 40 the cohort of women in California post-PFL could be driven by increased material aspirations if their income is higher relative to their parents’ generation (though there is little evidence of higher relative income in the sample). The empirical strategy employed in this study controls for many of the aggregate factors and cohort differences that influence fertility, but individual changes in preferences may still impact the results. My results indicate that PFL may affect the timing of childbearing, rather than completed fertility, but the cross-sectional nature of the data does not allow for direct testing of fertility using a life-cycle model. The results also cannot be generalized to the entire population of childbearing women, as I have limited my sample to women at ages most likely to be building human capital at the same time they are contemplating family. Finally, I identify all children under 1 year as a “birth,” but some of those children could be adopted and it is unclear how that could affect the results depending on where the child was born and the mother’s age, among other factors. CONCLUSION This study aimed to understand the implications of California’s Paid Family Leave (PFL) for decisions women make about childbearing. While the literature established that PFL promotes leave-taking and labor force attachment among new mothers, its impact on the family dimension of work-life balance in the U.S. context is not well understood. In an era of a rapidly aging U.S. society and below-replacement fertility, the potential positive effects of paid leave on fertility behavior could be important. A national paid leave policy could not only bolster labor force attachment among women, it could have a meaningful impact on the future workforce. With consistently high labor force participation among young women and mothers, work-life conflict could be at an all-time high. The finding that PFL increases birth likelihood among groups of 41 women most likely to face this conflict could indicate its ability to reduce the tension between work and family life, allowing women to have more child or begin having children sooner, at ages associated with human capital accumulation. Reduced childbearing postponement also reduces short-term and long-term health risks for the mother and child, which benefits individuals and society. Delayed childbearing carries additional risks for mothers and children and can increase healthcare costs. Women over 40 years old are at increased risk of miscarriage, stillbirth, chromosomal abnormalities, complications during childbirth, and multiple births (e.g. Heffner 2004; Huang et al., 2008; Luke et al., 2007; Tromp et al., 2011). As would-be mothers get older, they become more likely to use Assisted Reproductive Technology (ART), a cost born both by individuals and the healthcare system (Tromp et al., 2011). Moreover, since fecundity decreases significantly with age, delayed childbearing can have psychological repercussions if it reduces the likelihood couples reach their intended fertility or remain involuntarily childless (Billari et al., 2006). The broader implications of family policies in the U.S. on fertility timing present interesting avenues for future inquiry. Further research would also be helpful in exploring PFL’s effect on infant and maternal health and wellbeing, as well as disentangling the average effects from individual preferences and timing effects, using panel data. 42 REFERENCES Abadie, A., Diamond, A., & Hainmueller, J. (2010). Synthetic control methods for comparative case studies: Estimating the effect of California’s tobacco control program. Journal of the American statistical Association, 105(490), 493-505. Abadie, A., & Gardeazabal, J. (2003). The economic costs of conflict: A case study of the Basque Country. The American Economic Review, 93(1), 113-132. Ai, C., & Norton, E. C. (2003). Interaction terms in logit and probit models. Economics letters, 80(1), 123-129. Angrist, J. D., & Pischke, J. S. (2008). Mostly harmless econometrics: An empiricist's companion. Princeton university press. Averett, S. L., & Whittington, L. A. (2001). Does maternity leave induce births? Southern Economic Journal, 403-417. Baker, M., & Milligan, K. (2010). Evidence from maternity leave expansions of the impact of maternal care on early child development. Journal of human Resources, 45(1), 1-32. Bartel, A. P., Rossin-Slater, M., Ruhm, C. J., Stearns, J., & Waldfogel, J. (2018). Paid Family Leave, Fathers’ Leave-Taking, and Leave-Sharing in Dual-Earner Households. Journal of Policy Analysis and Management, 37(1), 10-37. Baum, C.L. & Ruhm, C.J. (2016). The Effects of Paid Family Leave in California on Labor Market Outcomes. Journal of Policy Analysis and Management, 35(2), 333-356. Becker, G.S. (1981, 1991). A Treatise on the Family. Harvard University Press. Bedard, K., Rossin-Slater, M. (2016). The Economic and Social Impacts of Paid Family Leave in California: Report for the California Employment Development Department. California Employment Development Department. Bianchi, S. M. (2011). Family change and time allocation in American families. The ANNALS of the American Academy of Political and Social Science, 638(1), 21-44. Billari, F. C., Liefbroer, A. C., & Philipov, D. (2006). The postponement of childbearing in Europe: Driving forces and implications. Vienna Yearbook of Population Research, 1-17. Björklund, A. (2006). Does family policy affect fertility?. Journal of population economics, 19(1), 3-24. Blau, F. D., & Kahn, L. M. (2005). Changes in the labor supply behavior of married women: 1980-2000 (No. w11230). National Bureau of Economic Research. 43 Byker, T. (2016). Paid Parental Leave Laws in the United States: Does Short-Duration Leave Affect Women's Labor-Force Attachment? American Economic Review, 106(5): 242-46. Caldwell, J. C., & Schindlmayr, T. (2003). Explanations of the fertility crisis in modern societies: A search for commonalities. Population studies, 57(3), 241-263. Cannonier, C. (2014). Does the Family and Medical Leave Act (FMLA) Increase Fertility Behavior?. Journal of Labor Research, 35(2), 105-132. Castles, F. G. (2003). The world turned upside down: below replacement fertility, changing preferences and family-friendly public policy in 21 OECD countries. Journal of European social policy, 13(3), 209-227. Cherlin, A.J., Cumberworth, E., Morgan, S. P. and Wimer, C. (2013). The Effects of the Great Recession on Family Structure and Fertility. Annals of the American Academy of Political and Social Science, 650(1): 214-231. Drago, R., Sawyer, K., Shreffler, K. M., Warren, D., & Wooden, M. (2009). Did Australia's Baby Bonus Increase the Fertility Rate? (No. wp2009n01). Melbourne Institute of Applied Economic and Social Research, The University of Melbourne. Easterlin, R. A. (1975). An economic framework for fertility analysis. Studies in family planning, 6(3), 54-63. Finch, N., & Bradshaw, J. (2003, May). Fertility and supporting the costs of children. In Conference on Recent Fertility Trends in Northern Europe, May 15th-16th, 2003, Oslo. Flood, S., King, M., Ruggles, S., & Warren, J. R. (2015). Integrated public use microdata series, current population survey: Version 4.0. [dataset]. Minneapolis: University of Minnesota, 2015. Gauthier, A. H. (2007). The impact of family policies on fertility in industrialized countries: a review of the literature. Population Research and Policy Review, 26(3), 323-346. Gauthier, A. H., & Hatzius, J. (1997). Family benefits and fertility: An econometric analysis. Population studies, 51(3), 295-306. Hamilton, B.E., Martin, J. A., Osterman M.J.K., et al. (2014) Births: Final data for 2013. National Vital Statistics Reports, 62(9). Hyattsville, MD: National Center for Health Statistics. Retrieved from: https://www.cdc.gov/nchs/data/nvsr/nvsr64/nvsr64_12.pdf Heffner, L. J. (2004). Advanced maternal age—how old is too old?. New England journal of medicine, 351(19), 1927-1929. 44 Hoem, J. M., Prskawetz, A., & Neyer, G. (2001). Autonomy or conservative adjustment? The effect of public policies and educational attainment on third births in Austria, 1975- 96. Population Studies, 55(3), 249-261. Hotz, V. J., Klerman, J. A., & Willis, R. J. (1997). The economics of fertility in developed countries. Handbook of population and family economics, 1(Part A), 275-347. Huang, L., Sauve, R., Birkett, N., Fergusson, D., & van Walraven, C. (2008). Maternal age and risk of stillbirth: a systematic review. Canadian Medical Association Journal, 178(2), 165-172. Jaumotte, F. (2004). Labour force participation of women. OECD Economic studies, 2003(2), 51-108. Jorgensen, H., & Appelbaum, E. (2014). Expanding Federal Family and Medical Leave Coverage: Who Benefits from Changes in Eligibility Requirements? (No. 2014-02). Center for Economic and Policy Research (CEPR). Kerr, S. P. (2016). Parental Leave Legislation and Women's Work: A Story of Unequal Opportunities. Journal of Policy Analysis and Management, 35(1), 117-144. Kirk, D. (1996). Demographic transition theory. Population studies, 50(3), 361-387. Klerman, J. A., Daley, K., & Pozniak, A. (2012). Family and medical leave in 2012: Technical report. Cambridge, MA: Abt Associates Inc. Klerman, J., & Leibowitz, A. (1997). Labor supply effects of state maternity leave legislation. In F. Blau & R. Ehrenberg (Eds.), Gender and family issues in the workplace (pp.65–85). New York: Russell Sage Foundation. Lalive, R., & Zweimüller, J. (2005). Does parental leave affect fertility and return-to-work? Evidence from a true natural experiment. Lalive, R., & Zweimüller, J. (2009). How does parental leave affect fertility and return to work? Evidence from two natural experiments. The Quarterly Journal of Economics, 1363-1402 Lee, R. (2003). The demographic transition: three centuries of fundamental change. The journal of economic perspectives, 17(4), 167-190. Lee, S. H., Ogawa, N., & Matsukura, R. (2009). Is childcare leave effective in raising fertility in Japan?. Asian Population Studies, 5(3), 349-369. Lichtman-Sadot, S. (2014). The Value of Postponing Pregnancy: California’s Paid Family Leave and the Timing of Pregnancies. The BE Journal of Economic Analysis & Policy, 14(4), 1467-1499. 45 Lichtman-Sadot, S., & Bell, N. P. (2017). Child Health in Elementary School Following California's Paid Family Leave Program. Journal of Policy Analysis and Management, 36(4), 790-827. Livingston, G. (4 Feb 2015). Is U.S. fertility at an all-time low? It depends. PEW Center. Retrieved from: http://pewrsr.ch/1JHXDLq Luci-Greulich, A., & Thévenon, O. (2013). The impact of family policies on fertility trends in developed countries. European Journal of Population/Revue européenne de Démographie, 29(4), 387-416. Luke, B., & Brown, M. B. (2007). Contemporary risks of maternal morbidity and adverse outcomes with increasing maternal age and plurality. Fertility and sterility, 88(2), 283- 293. Milkie, M. A., Kendig, S. M., Nomaguchi, K. M., & Denny, K. E. (2010). Time with children, children's well-being, and work-family balance among employed parents. Journal of Marriage and Family,72(5), 1329-1343. OECD (2016). OECD Family Database, TablePF2.1.A Retrieved from: https://www.oecd.org/els/soc/PF2_1_Parental_leave_systems.pdf Olivetti, C., & Petrongolo, B. (2017). The economic consequences of family policies: lessons from a century of legislation in high-income countries. The Journal of Economic Perspectives, 31(1), 205-230. Rønsen, M. (2004). Fertility and public policies-Evidence from Norway and Finland. Demographic Research, 10, 143-170. Rønsen, M., & Sundström, M. (2002). Family policy and after-birth employment among new mothers–A comparison of Finland, Norway and Sweden. European Journal of Population/Revue europeenne de demographie, 18(2), 121-152. Rossin, M. (2011). The effects of maternity leave on children's birth and infant health outcomes in the United States. Journal of health Economics, 30(2), 221-239. Rossin-Slater, M., Ruhm, C. J., & Waldfogel, J. (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(2), 224-245. Ruhm, C. J. (1998). The economic consequences of parental leave mandates: lessons from Europe. Quarterly Journal of Economics, 113(1), 285-317. Ruhm, C. J. (2000). Parental leave and child health. Journal of health economics, 19(6), 931- 960. 46 Sayer, L. C., Bianchi, S. M., & Robinson, J. P. (2004). Are Parents Investing Less in Children? Trends in Mothers’ and Fathers’ Time with Children. American journal of sociology, 110(1), 1-43. Sholar, M. (22 Sept 2016). Donald Trump and Hillary Clinton both support paid family leave. That’s a breakthrough. Washington Post. Retrieved from: https://www.washingtonpost.com/news/monkey-cage/wp/2016/09/22/donald-trump-and- hillary-clinton-both-support-paid-family-leave-thats-a-breakthrough/ Sinclair, S., Boymal, J., & De Silva, A. (2012). A Re-Appraisal of the Fertility Response to the Australian Baby Bonus. Economic Record, 88(s1), 78-87. Sobotka, T., Skirbekk, V., & Philipov, D. (2011). Economic recession and fertility in the developed world. Population and development review, 37(2), 267-306. State of California, Employment Development Department (2017). Paid Family Leave. Retrieved from: http://www.edd.ca.gov/Disability/Paid_Family_Leave.htm Stearns, J. (2015). The effects of paid maternity leave: Evidence from Temporary Disability Insurance. Journal of health economics, 43, 85-102. Sweeney, M. M., & Raley, R. K. (2014). Race, ethnicity, and the changing context of childbearing in the United States. Annual Review of Sociology, 40, 539-558. Tamborini, C. R., Kim, C., & Sakamoto, A. (2015). Education and lifetime earnings in the United States. Demography, 52(4), 1383-1407. Thévenon, O., & Gauthier, A. H. (2011). Family policies in developed countries: A ‘fertility- booster’ with side-effects. Community, Work & Family, 14(2), 197-216. Tromp, M., Ravelli, A. C. J., Reitsma, J. B., Bonsel, G. J., & Mol, B. W. (2010). Increasing maternal age at first pregnancy planning: health outcomes and associated costs. Journal of Epidemiology & Community Health, jech-2009. U.S. Department of Labor, Wage and Hour Division (2012). Fact Sheet #28: The Family and Medical Leave Act. Retrieved from: https://www.dol.gov/whd/regs/compliance/whdfs28.pdf Waldfogel, J. (1998). Understanding the" family gap" in pay for women with children. The Journal of Economic Perspectives, 137-156. Waldfogel, J. (1999). The impact of the family and medical leave act. Journal of Policy Analysis and Management, 281-302. 47 Waldfogel, J., Higuchi, Y., & Abe, M. (1999). Family leave policies and women's retention after childbirth: Evidence from the United States, Britain, and Japan. Journal of Population Economics, 12(4), 523-545. Willis, R. (1974). Economic theory of fertility behavior. In Economics of the family: Marriage, children, and human capital (pp. 25-80). University of Chicago Press. 48 APPENDIX Table A1: California Control State Regressions (Birth Likelihood) State State*pre_trend P-value Control Alabama -0.0057 0.013 Alaska -0.0012 0.178 X Arizona -0.0016 0.060 X Arkansas 0.0014 0.008 Colorado 0.0011 0.141 X Connecticut 0.0022 0.016 Delaware 0.0071 0.004 D.C. 0.0039 0.010 Florida 0.0032 0.002 Georgia 0.0019 0.178 X Hawaii -0.0005 0.183 X Idaho -0.0006 0.014 Illinois 0.0040 0.011 Indiana 0.0054 0.001 Iowa 0.0038 0.073 X Kansas -0.0002 0.110 X Kentucky -0.0027 0.016 Louisiana 0.0045 0.004 Maine 0.0031 0.005 Maryland 0.0024 0.013 Massachusetts 0.0036 0.009 Michigan 0.0011 0.051 X Minnesota -0.0003 0.284 X Mississippi 0.0033 0.009 Missouri 0.0051 0.004 Montana -0.0041 0.007 Nebraska -0.0025 0.001 Nevada 0.0063 0.006 New Hampshire -0.0019 0.008 New Jersey -0.0014 0.081 X New Mexico -0.0019 0.015 New York 0.0006 0.108 X North Carolina 0.0053 0.007 North Dakota 0.0084 0.001 Ohio -0.0033 0.037 49 Oklahoma -0.0001 0.779 X Oregon 0.0021 0.008 Pennsylvania 0.0009 0.128 X Rhode Island 0.0037 0.002 South Carolina 0.0068 0.010 South Dakota 0.0032 0.003 Tennessee -0.0013 0.059 X Texas 0.0015 0.081 X Utah 0.0031 0.042 Vermont -0.0017 0.009 Virginia 0.0054 0.061 X Washington -0.0041 0.027 West Virginia 0.0015 0.008 Wisconsin -0.0049 0.008 Wyoming 0.0007 0.011 50 Table A2: Linear Probability Model of Birth Likelihood, Women age 25 to 40, Treatment (CA) and Control (non-CA), 1999-2009, including county-level measures Model 3 CA*Post 0.0036 * CA -0.0062 *** Post 0.0069 ** Age 0.0124 ** Age^2 -0.0003 *** Race/ethnicity (ref. NH white) NH Black -0.0032 Asian/PI -0.0105 * Hispanic -0.0002 NH Other -0.0053 Married 0.0539 *** Other children 0.0643 *** Education level (ref. <High school) HS Grad 0.0028 Some College 0.0125 *** BA+ 0.0395 *** Foreign born 0.0018 LFP -0.0728 *** Annual earnings (log) 0.0024 *** County Variables Unemployment 0.0001 % Hispanic -0.0042 % Immigrant 0.0036 Med HH Income 0.0000 *** Constant -0.0613 R^2 0.0575 N 117,031 Table A3: Alternative Control Groups Model 2 Model 3 N Next 2 largest (FL, NY, TX) 0.0049 * 0.0048 * 61,102 Other TDI states (HI, 0.0025 0.0045 + 50,199 All states 0.0006 0.0001 254,991 51 Table A4: Synthetic Control Method-Generated Weights State Weight Alabama 0.039 Arizona 0.220 Hawaii 0.089 Iowa 0.002 Kentucky 0.026 Nevada 0.042 New Hampshire 0.071 New York 0.363 Texas 0.149 Figure A1: Synthetic Control Method Placebo tests Figure A2: Synthetic Control Method Placebo tests 0 .05 .1 .15 (mean) baby 1999 2001 2003 2005 2007 2009 Survey year Treated Donors .04 .06 .08 .1 .12 .14 (mean) baby 1999 2001 2003 2005 2007 2009 Survey year Treated Donors 52 Can Paid Leave Promote Labor Force Attachment Among Parents of Children with Disabilities? by Johanna Thunell Abstract This paper explores how California’s Paid Family Leave program (PFL) impacts the labor supply of parents of children with disabilities and whether there are gender differences in the effect. I use nationally-representative panel data from the Survey of Income and Program Participation (SIPP) to analyze whether access to publically-funded family leave influences the work decisions of these parents. Specifically, I estimate triple-difference models that compare the labor supply of parents of children with disabilities in California before and after PFL to parents in California of children with no disability and parents in control states. I find a negative relationship between having a child with a disability and the labor force participation of his/her mother and father. However, access to paid leave increased the labor force participation of mothers and fathers of children with disabilities in California by 9.4 and 4.3 percent, respectively, and reduced the likelihood these parents dropped out of the workforce. 53 INTRODUCTION According to a 2010 U.S. Census Bureau report, about 8 percent of all U.S. children under age 15 have a disability (Brault, 2012). While all children place demands on their parents’ time, children with special health care needs require additional parental care and resources for sustained periods of time. Parents are often called to attend doctor appointments, various forms of therapy, school meetings, or hospitalizations throughout the child’s life. For working parents, these ongoing needs place pressure on the amount of time they can devote to work. Studies on mothers of chronically ill and disabled children consistently demonstrate reduced employment likelihood, working hours, and lost wages (e.g. Earle & Heymann, 2012; Powers, 2003); however, little is known about the labor supply response of their fathers. In modern U.S. families, parents face additional barriers to meeting the needs of their children with disabilities. The number of U.S. families where all parents work (whether dual- earners or single parents) has increased markedly in recent decades, such that most children (over 60 percent) are growing up in these households (U.S. Department of Labor, 2017). Moreover, wages stagnated in recent decades, especially among low and middle-income workers (Mishel et al., 2015). This means today’s families increasingly rely on the income of all parents to maintain their financial wellbeing. When a child is disabled or becomes ill, additional caregiving demands may conflict with work time, and one or both parents may struggle to maintain labor force attachment. Thus, the financial wellbeing of families of children with disabilities may be especially precarious. Paid leave programs that provide time off to care for sick or disabled children can allow these parents to continue working, while caring for their child. In the United States, the Federal Family and Medical Leave Act (FMLA) provides 12 weeks of job-protected leave for workers to bond with a new child, attend to their own illness, or care for an ill loved one (U.S. Department of 54 Labor, 2012). Five states, including California, provide six to twelve weeks paid leave for similar purposes. While the majority of family leave claims come from new parents, caregiving represents a sizable portion of claims, about 18 percent of FLMA claims and 10 percent of California’s Paid Family Leave program (PFL) claims (Bedard & Rossin-Slater, 2016; Klerman, Daley & Pozniak, 2014). 11 In this study, I use nationally-representative data from the Survey of Income and Program Participation (SIPP) to analyze the relationship between labor supply, having a child with a disability, and access to paid family leave and whether there are gender differences in these relationships. Specifically, two research questions are posed: 1. How does having a child with a disability impact the labor supply of their mothers and fathers? 2. Does access to publically-funded family leave (in California) mediate the relationship between childhood disability and parental labor supply? Cross-sectional studies of the relationship between child disability and parental labor supply can suffer from bias due to the endogeneity of the work and care decisions. For example, if a parent provides care to their child with a disability due to a lack of employment options, estimates of the association between the child’s disability and her parent’s labor supply will be biased. I exploit the longitudinal nature of SIPP to estimate models of the relationship between child disability and parental labor supply using static and dynamic measures of labor force participation. I stratify the models based on the parent’s gender to compare the effects for mother and fathers. I find the presence of a child with a disability reduced labor force participation of both mothers and fathers; however, the effect for mothers is larger, 2.9 versus 1.4 percentage points. Using a dynamic measure of labor supply, mothers, but not fathers, were more likely to drop out of the labor force if they had a child with a disability than mothers whose child has no disability. 11 Estimates are based on all caregiving claims, which may include care for spouses or other family members. 55 The previously unknown negative relationship between paternal labor supply and children with disabilities contributes to our understanding of family processes and coping strategies. On the other hand, the relatively small negative effect on paternal labor supply suggests mothers still shoulder a greater caregiving burden in these families. Paid family leave that allows intermittent time off to care for a family member, as in California, could promote labor force attachment among parents of children who require extra time in care. Using a triple-differences model, I find access to paid family leave increased labor force participation among mothers and fathers of children with disabilities in California compared to parents of children with no disability in California and parents in control states, and decreased the likelihood they drop out of the workforce. Again, there are gender differences in the effect, labor force participation increased by 5.7 and 2.9 percentage points for mothers and fathers, respectively, which is equivalent to an increase of 9.4 percent and 4.3 percent in California after PFL over their baseline participation levels. This study adds to the literature on families of children with disabilities and family leave policies in the United States. Empirical research on paid family leave policies in the U.S. has established the effect of paid parental leave in California on a variety of immediate and longer- term labor market outcomes, such as employment, labor force participation, and leave-taking during the period around a child’s birth, finding a positive impact on labor market outcomes of new mothers and fathers (e.g. Baum & Ruhm, 2016; Byker, 2016; Rossin-Slater et al, 2013). The results in the present study suggest paid family leave also promotes labor force attachment among parents of children with disabilities, whose labor supply is especially tenuous. However, evidence of stronger effects of both child disability and paid leave for mothers point to persistent gender differences in work and care. 56 THEORETICAL BACKGROUND The tension between working and childrearing may be especially acute among parents of children with disabilities. Parents are generally expected to devote some amount of goods and time to promote the wellbeing of their children (Browning, 1992), but children with disabilities often require additional parental resources, placing pressure on both the time and budget constraints of their parents. Becker’s (1981, 1985, 1991) theory on human capital and the division of household labor indicates the presence of a child lowers the labor supply of at least one parent in a household by placing demands on the parent’s time, assuming non-labor income (e.g. spouse’s income) is sufficiently high enough to maintain consumption. It follows that an exogenous shock to the child’s health, such as an illness or disability, would augment the labor supply decision through increased demands on the parents’ time, family consumption (e.g. medical expenses and equipment), or both. That is, the child’s disability increases the opportunity cost of working and the cost of the child relative to other goods. Time devoted to the market would be costlier, because substitute child care capable of handling the child’s physical or behavioral demands can be more expensive than traditional child care (Lukemeyer et al., 2000; Parish et al., 2005; Stabile & Allin, 2012). 12 Ongoing child care expenses would place an especially large burden on lower wage workers. On the other hand, non-market time (i.e. time with children) is more costly for higher-wage workers, who may increase their labor supply and substitute private care for their own time. Thus, based on 12 A child’s illness or disability could require additional consumption (such as medical expenses, equipment, etc.) along with the additional time in care. In such cases, we may anticipate an increase in labor supply from his/her parents. I assume the children’s conditions require both time and goods, but the additional time is the primary driver of any changes in parental labor supply among parents of children with disabilities. Thus, I limit my theoretical discussion to the impact of ongoing care requirements. If parents increase their labor supply when a child becomes ill or disabled, any estimated effects of the child’s needs combining time and goods represent a lower bound of the true effect of parents’ additional time in devoted to these children. For a theoretical model and empirical estimate of the effect of time-intensive versus goods-intensive conditions, see Gould (2004). 57 Becker’s framework, the effect of a child’s disability on parental labor supply is ambiguous and depends on whether an income (positive) or substitution (negative) effect prevails. The substitution effect would be particularly strong for married mothers. Among married parents, Becker’s framework suggests that mothers and fathers will specialize in work and home production, making individual investments in one domain. The parent earning a higher wage is expected to remain in the labor market when a child becomes ill or disabled, while the parent investing in home production earns lower (or zero) wages and will devote the additional time needed to care for the child. In practice, mothers earn less than their husbands and women with no children (in part due to this specialization). 13 Thus, we may expect married mothers to assume primary caring responsibilities of a child when he/she is disabled or becomes ill, while married fathers would continue working. Sociological theory provides further indication of gender differences in the expected relationship between labor supply and having a child with a disability. Cultural conceptions of motherhood suggest maternal labor supply would decrease more than paternal labor supply in the presence of a child with a disability. In the U.S., the prevailing ideology of “intensive mothering” compels mothers to invest significant time and devotion to their children, whether they are working or not (Hays, 1998). In the case of a child with disabilities, mothers held accountable for this standard may “do gender” by assuming the primary responsibility of caring for the child (West & Zimmerman, 1987). On the other hand, fathers assuming a “breadwinner” role would increase their labor supply when a child becomes chronically ill or disabled. This framework reinforces the prediction based on the economic model that married mothers will reduce their labor supply more 13 Becker (1985, 1991) proposes mothers may earn less due to specialization in household productions or reduced work effort; empirical studies consistently demonstrate a gender wage gap and a motherhood penalty (e.g. Waldfogel, 1998). 58 than fathers in the presence of a child with a disability; however, if normative expectations of parenting are the main drivers of the relationship, differential wages would have little impact on the labor supply decision. Taken together, these frameworks suggest that parents may increase or decrease their labor supply to accommodate the child’s needs depending on their wages, marital status, and gender. Fathers and higher earners are expected to increase their labor supply, while lower wage workers and mothers are most likely to reduce their labor supply. Empirical studies consistently demonstrate reduced maternal labor supply in the presence of a child with a disability. Thus, I expect mothers of children with disabilities in the present study will have lower labor force participation than those whose children have no disability. For fathers, I expect the effect will be smaller than it is for mothers, but there is not a clear prediction about the direction of the effect. The effect of paid family leave policies, as in California, on the labor supply of parents of children with disabilities is a priori ambiguous under the economic and sociological frameworks. To the extent that access to leave encourages parents (who would have otherwise reduced their labor supply) to continue working while caring for a child with a disability, it should have a positive effect on their labor supply. On the other hand, if the leave increases time away from the workforce for parents who would have otherwise continued working, the policy may decrease their labor supply. In general, short leaves (less than six months) have been shown to have a positive effect on leave-taking and employment of new mothers and fathers (e.g. Ruhm, 2000). California’s PFL allows intermittent leave to care for an ill loved one and could benefit both groups of parents similarly. Thus, I anticipate California’s PFL will have a positive effect on the labor supply of parents of children with disabilities, particularly mothers. Moreover, if the policy signals changing cultural norms about combining motherhood and working, paid family leave may promote a more 59 equitable division of caregiving and work between mothers and fathers; thus, it would increase mothers’ labor supply more substantially than fathers’. EMPIRICAL LITERATURE Empirical evidence generally supports the notion that poor child health and childhood disability exert a negative influence on maternal labor supply; however, less is understood about the effect on fathers. In addition, short paid leaves (such as California’s six-week policy) have been shown to have a positive impact on new parents’ labor market outcomes, but little is known about its effect on caregivers. The following section highlights the extant literature on families of children with illnesses or disabilities and paid parental leave policies in the U.S. Parental Labor Supply The literature consistently finds evidence of reduced labor supply among mothers of children with disabilities (e.g. Powers, 2001, 2003) 14 and long-term, chronic illnesses (e.g. Gould, 2004; Stabile & Allin, 2012), and more severe conditions have larger negative effects on maternal labor supply (e.g. Powers, 2003; Stabile & Allin, 2012; Wasi et al., 2012). While estimates vary across studies, depending on their methods and definition of the child’s disability or illness, research indicates maternal labor force participation decreases about 2 to 19 percentage points, while 15 to 68 percent of mothers reduce work hours (Burton et al., 2014; Stabile and Allin, 2012). Most early studies analyzed cross-sectional data to estimate the association between disability and maternal labor supply, which cannot disentangle the effect of child disability or illness from unobserved characteristics or motivations. For example, survey data typically rely on 14 See Powers (2003), Table A1 for a list of early studies. 60 a parent’s self-report of their child’s health or disability, which could be endogenous if they are trying to rationalize labor supply decisions (Powers, 2001). Related, the work and care decisions may be endogenous if, for instance, a parent provides care due to a lack of employability (Heitmeuller, 2007). A handful of studies try to overcome the endogeneity of child health and the work and care decisions using instrumental variables (Powers, 2001) or longer employment horizons (Kvist et al., 2003; Powers, 2003) finding a negative relationship between maternal labor supply and child disability; however, estimates from these models are typically lower than cross- sectional models. For example, the magnitude of Powers’ (2001) IV estimates are about half the size of the cross-sectional estimates, 6.0 versus 3.0 percentage points for wives. Other research employs panel data techniques to measure the effect of changes in the child’s health status on maternal employment. Hope et al. (2016), used longitudinal data in the UK to disentangle “common causes” (of mother and child) from the relationship between long-term illness and maternal employment. They found childhood illness was negatively associated with employment, after controlling for covariates and individual fixed effects. In their study using Canadian longitudinal data, Burton et al. (2014) estimated dynamic models of the “onset” of a child’s health problem, again finding a negative effect of child health on maternal (but not paternal) employment. Specifically, mothers in their sample were 2.2 percentage points less likely to be working when their child’s health declined. Research also explores heterogeneity in outcomes based on socioeconomic background, demographics, and marital status. Most studies indicate single mothers (or female heads) are more negatively affected by their child’s health or disability status than wives (Powers, 2001, 2003; Wasi et al, 2012); however, there is some evidence that married mothers’ labor supply is more affected by unpredictable illnesses (Gould, 2004) and one study finds no (significant) effect on 61 single mothers’ labor force participation (Breslau et al., 1982). Breslau et al.’s (1982) sample is relatively small and limited to children with specific conditions, whereas more recent research demonstrates the negative impact of childhood illness and disability is larger for parents of children with more severe conditions (Gould, 2004; Salkever, 1982; Stabile & Allin, 2012; Wasi et al., 2012). Larger effects have also been found among low income and black wives using interactions (Breslau et al., 1982). Wasi et al. (2012) took advantage of the large sample in Census 2000 to explicitly test whether impacts are heterogeneous across different types of disabilities, maternal education, and child age using reduced form and structural models of employment likelihood and working hours. Their two-step structural estimation procedure allowed the child’s disability status to influence labor supply indirectly through wages and directly through hours of employment. In the reduced form model, they estimate only the direct effect of the disability. Both estimation methods indicate child disability negatively impacted the likelihood that married and single mothers were employed, which interactions reveal was strongest among women with the lowest education levels. In the structural models, they found both an indirect (reduced wages) and direct effect on labor supply of having a child with a disability. While a growing body of literature examines the effect of child health and disability on maternal labor supply, little is known about the labor force activity of these children’s fathers. Two studies that estimate the effect of child health on fathers’ decisions about work provide some useful insight. Using time diary data in Australia, Brandon (2011) estimates multinomial models of couples’ labor supply finding the odds of not working, working part-time, and working nonstandard schedules are higher for mothers of children with disabilities, but results are not significant for fathers. Similarly, in Canada, Burton et al. (2014) find no effect of child health on 62 the labor force participation or work hours of the fathers in their sample. Both studies estimate the labor supply of couples and indicate specialization occurs within a household - mothers are reducing labor supply and fathers are not. However, this remains an open question. These studies only include married couples and take place outside of the U.S. context, where policies governing healthcare, special education, and other services for these children and their families may substantially alter estimated relationships. Paid Family Leave in the U.S. California became the first state to offer paid family leave through its temporary disability program in 2004 (State of California EDD, 2017). The program provides paid time off for bonding with a new child or caring for one’s own serious health condition or that of a family member (spouse, child, parents, grandparent, parent-in-law, sibling, or domestic partner). Workers are reimbursed 55 percent of their earnings up to a cap of $1,173 in 2017. The leave can be taken all at once or intermittently over a 12-month period. To qualify for the leave, workers much have earned at least $300 with SDI deductions in the last 5 to 18 months. Caregivers must submit evidence of a “serious health condition,” which is certified by a physician. Under the definition provided by California’s Employment Development Department, 15 many children with developmental disabilities, chronic illnesses, or other serious issues would likely qualify for care. PFL does not, however, provide job protection. The federal Family and Medical Leave Act (FMLA) 15 The California Employment Development Department defines serious health condition as: “an illness, injury, impairment, or physical or mental condition of a patient that involves any period of incapacity (e.g., inability to work or perform other regular daily activities) or inpatient care in a hospital, hospice, or residential medical care facility and any subsequent treatment in connection with such inpatient care; or continuing treatment by a physician/practitioner. Unless complications arise, cosmetic treatments, the common cold, influenza, earaches, upset stomach, minor ulcers, and headaches other than migraine, are examples of conditions that do not meet the definition of a serious health condition for purposes of PFL” (State of California EDD, 2017). 63 and California’s Family Rights Act (CFRA) offer job-protected leave for caregiving and many workers in California use FMLA/CFRA and the PFL leave programs concurrently (State of California EDD, 2017; US Department of Labor, 2012); however, FMLA eligibility restrictions leave about 40 percent of workers without access to job-protected leave. Thus, although some workers may benefit from PFL, they may be unable to take the leave if their employers are not covered by FMLA and refuse to protect the employee’s job while on leave. This discrepancy in eligibility may affect who uses the policy. If eligibility criteria are met, PFL is designed such that parents faced with a child’s illness or disability, could qualify for up to six weeks of paid leave on an intermittent basis. Yet, evidence indicates there is a lack of policy awareness among parents or understanding of how the policy applies to their circumstances. Surveys and interviews with workers in California indicate caregivers (as opposed to new parents) were especially likely to have low levels of awareness of the policy and its specific benefits (Chung et al., 2012), and there are differences in awareness by socioeconomic background (Tisinger et al., 2016). Not surprisingly, studies demonstrate that awareness of leave is associated with higher rates of leave-taking (Chung et al., 2007). This suggests awareness could pose a barrier to leave-taking and any secondary effects on labor force participation and employment, especially among parents of children with disabilities. The growing body of literature devoted to California’s Paid Family Leave policy demonstrates positive effects of the policy on a variety of outcomes, all of which employ difference-in-differences estimation, comparing outcomes for new parents in California to a control group before and after PFL. Rossin-Slater et al. (2013) find PFL increased leave-taking among new mothers in California by 3 to 6 weeks. PFL also increased employment among mothers of 1 to 3-year-olds by 10 to 17 percent. Subsequent studies have confirmed the positive effect of 64 PFL on leave-taking of new mothers and fathers (Bartel et al., 2017; Baum and Ruhm, 2016). Evidence also indicates that PFL influenced fertility behavior by altering the timing of births in the year it was enacted (Lichtman-Sadot, 2014). Research has also shown benefits of PFL extend beyond new mothers in California to their children. Huang and Yang (2015) find that breastfeeding increased by 10 to 20 percentage points after PFL was enacted. Lichtman-Sadot et al. (2017) uncover longer-term impacts of mothers’ access to PFL on their children’s health. In their study, kindergarten-age children of mothers in California after PFL had reduced odds of being overweight, or having ADHD, hearing or communications problems, and frequent ear infections. Contribution The literature generally reveals reduced labor supply among mothers of children with illnesses and disabilities, which varies by the intensity of the child’s needs and socioeconomic factors; however, the effect on paternal labor supply (especially in the U.S.) is not well understood and remains an open question. Further, the relationship between labor market outcomes for these parents and paid family leave has not been systematically explored. This study fills these gaps and adds to the literature by estimating the gender differences in the impact of childhood disability on parental labor supply and whether PFL reduces the (presumably negative) effect on their labor supply. DATA AND VARIABLES Data Data for this study come from the 2001, 2004, and 2008 panels of the Survey of Income and Program Participation (SIPP). SIPP consists of short-term panels that are surveyed every four 65 months over the course 3 to 4 years (U.S. Census Bureau, 2008). The primary purpose of the survey is to provide information on participation in government programs, especially those targeting low-income households. In each wave, participants are asked a set of core questions focused on program participation, education, work, and family relationships. These variables are comparable across waves and panels and can be used in longitudinal analyses. In addition, each wave includes a topical module with survey items on special topics such as fertility, taxes, assets and liabilities, among many others. Key to the present study are the series of questions in the Functional Limitations/Disability – Child module. The Functional Limitations/Disability – Child topical module typically occurs once per panel (twice in 2001) and, thus, provides a static measure of the child’s disability. I link data on children’s age, gender, and disability status to the longitudinal data on their parents’ labor force participation, employment, and other characteristics from the core survey. There is a well-known seam bias in the SIPP data, since questions are asked about each of the last four months. It is easier to accurately recall information about wages, employment, etc. in the last month rather than two to four months prior, which can impact estimates of monthly earnings and career trajectories (Moore, 2008). To avoid this issue, I limit my analysis to the reference month for the respondents in each wave. Thus, individuals and households appear in the data up to three times per year (or once per wave). I pool data from all three panels to obtain sufficient sample size before and after the 2004 enactment of PFL. There are 9 waves in the 2001 panel, 12 waves in the 2004 panel, and 10 waves in the 2008 panel, with data spanning every month from February 2001 to December 2011. The analytical sample includes children ages 0 to 17 and working-age parents, 25 to 64. I restrict my sample to children for whom I can identify at least one parent. I exclude parent-child pairs residing in New Jersey, because they passed a similar 66 law in 2009. I also exclude parent-child pairs in the five states that SIPP pools for anonymity (Maine, Vermont, North Dakota, South Dakota, and Wyoming) since I cannot isolate individual state trends and policies over time. The unit of observation is the mother/child or father/child pair; many children appear in models of both their mothers and their fathers. There are more mothers in the sample than fathers, due to the inclusion of single parents. The final sample includes 517,951 mother/child-wave observations and 418,041 father/child-wave observations. Dependent Variables I analyze two dependent variables of mothers’ and fathers’ labor supply. The first is a dummy indicator that equals one if the parent is the labor force in the reference month and zero otherwise. The second is a dynamic measure of labor force participation, which accounts for an individual’s previous labor supply. Conditioning on labor force participation in the first wave the parent appears, the binary variable equals one if the parent is out of the labor force in a subsequent wave. Observations for parents that remain in labor force equal zero. Thus, the measure indicates if a parent previously engaged in the labor force dropped out in subsequent periods, relative to those who remain in the labor force. Models using the dynamic measure of labor supply are restricted to individuals who appear in at least two waves. Independent Variables The key independent variables in this analysis are the child’s disability status, whether the parent/child pair resides in California, and a post-PFL indicator (July 2004 or later). I use a composite measure of disability identified in the Census Bureau’s Americans with Disabilities reports, which classifies severe and mild disability status using questions about limitations on 67 activities of daily living (ADLs) and instrumental activities of daily living (IADLs) (Brault, 2008, 2012). 16 The occurrence of severe disabilities is low (less than 3 percent of the sample); thus, I combine children with mild and severe disabilities into one dichotomous variable equal to one if the child has any disability. I impute the disability status recorded in the topical modules to all other waves where the child appears. 17 Other independent variables control for socioeconomic (education and earnings) and demographic (age, race/ethnicity, marital status, number of children) factors of the parents, the parent’s health (whether fair or poor self-rated health), and the child’s gender and a nonlinear measure of the his/her age. EMPIRICAL STRATEGY Although research demonstrates a negative relationship between childhood disability and maternal work, estimates continue to suffer from the endogeneity of the care and work decisions and the child’s disability itself. Cross-sectional studies cannot disentangle the child’s illness from unobserved factors that may also affect a parents’ work or care decision. Unobserved preferences or capacity for either work or care may further bias cross-sectional estimates. A common empirical strategy for overcoming bias due to confounding variables employs panel data methods, fixed effects, to account for these unobserved differences (Angrist & Pischke, 2009). The fixed effects strategy is not an option with the SIPP data, as the child disability measures do not vary over time. 16 Americans with Disabilities: 2005 and Americans with Disabilities: 2010 provide population estimates using specific information about activities of daily living (Brault, 2008, 2010). The Census Bureau incorporated the composite measure into the 2008 panel and published coding to recreate the variable in the 2001 and 2004 panels. 17 This imputation may introduce some measurement error into my models. I test the sensitivity of my results to the imputation by including in the model only waves when disability questions were asked. The results of the main model are quantitatively similar (see Table 5, Panel A). 68 Instead, I use the dynamic measure of labor supply, as described above, to account for the parent’s past labor supply. I first test the relationship between having a child with a disability and static and dynamic measures of parental labor supply. I estimate linear probability models stratified on the parent’s gender to assess gender differences in the effect. This is an important first step as it indicates the comparability of my results to other studies and establishes the labor response of fathers in the U.S. Model 1 takes the following form: 𝑦 "$ = 𝛼+𝛽 ) 𝐷𝑖𝑠𝑎𝑏𝑖𝑙𝑖𝑡𝑦 " +𝛾 ) 𝑋 "$ F +𝛾 2 𝑋 "$ G +𝑣 $ +𝜂 " +𝜀 "$ (1) where 𝑦 "$ are the static and dynamic measures of labor force participation described above of parent i in time t, 𝐷𝑖𝑠𝑎𝑏𝑖𝑙𝑖𝑡𝑦 " equals one if the child has a disability as described above, 𝑋 "$ F and 𝑋 "$ G are parental and child characteristics, respectively, 𝑣 $ are survey wave fixed effects, 𝜂 " are individual time-invariant characteristics, and 𝜀 "$ is an error term clustered on the parent. 𝛽 ) represents the impact of having a child with a disability on his/her mother’s or father’s labor force participation. This study uses difference-in-difference-in-differences (DDD) estimation to compare parents of children with disabilities in California before and after PFL to those in California with a child with no disability and parents in control states of children with and without disabilities. The identifying assumption in this type of model (DDD) is that the treatment and control groups are equal in expectation prior to treatment (Angrist & Pischke, 2008). That is, the two groups should be similar enough in the pre-treatment (pre-PFL) period that any changes in the outcome observed after the treatment can be attributed to the treatment, rather than underlying differences in the two groups. In particular, it should be established that trends in the dependent variables are the same in both groups before the policy was enacted. Including covariates in the models helps 69 control for other time-varying factors that may influence variation in the dependent variable. If all model assumptions hold, post-treatment changes in the dependent variable for parents of children with disabilities in California can be attributed to the enactment of PFL. As an intermediate step, I estimate pre-PFL (through June 2004) models of parental labor supply in California versus the control states to test the assumption of parallel trends in the two groups. The control states include all U.S. states, except New Jersey, Maine, Vermont, North Dakota, South Dakota, and Wyoming. Model 2 takes the following form: 𝑦 "#$ = 𝛼+𝛽 ) 𝐷𝑖𝑠𝑎𝑏𝑖𝑙𝑖𝑡𝑦 "# +𝛽 2 𝐶𝐴 "$ + 𝛽 6 𝑇𝑟𝑒𝑛𝑑 $ +𝛽 M 𝐷𝑖𝑠𝑎𝑏𝑖𝑙𝑖𝑡𝑦 "# ∗𝐶𝐴 "$ + 𝛽 N 𝐷𝑖𝑠𝑎𝑏𝑖𝑙𝑖𝑡𝑦 "# ∗𝑇𝑟𝑒𝑛𝑑 $ +𝛽 O 𝐶𝐴 "$ ∗𝑇𝑟𝑒𝑛𝑑 $ +𝛽 P 𝐶𝐴 "$ ∗𝐷𝑖𝑠𝑎𝑏𝑖𝑙𝑖𝑡𝑦 "# ∗ 𝑇𝑟𝑒𝑛𝑑 $ + 𝛾 ) 𝑋 "#$ F +𝛾 2 𝑋 "#$ G +𝑣 $ + 𝜂 "# +𝜀 "#$ (2) where all variables are as described in Model 1, 𝐶𝐴 "$ equals one if the parent-child pair live in California and 𝑇𝑟𝑒𝑛𝑑 $ is an annual linear time trend from 2001 to 2004. Standard errors are clustered at the state level. 18 In this model, 𝛽 P measures differences in the annual trends in labor force participation of parents of children with disabilities in California and the control states prior to the enactment of PFL. As indicated above, is important for 𝛽 P not to be statistically significant to consider the control group a suitable comparison for California. The final model employs the DDD strategy to analyze whether labor supply among parents of children with a disability in California after PFL differs significantly from parents in other states and those in California whose children have no disability. Specifically, it reveals whether PFL alters the impact of children with a disability on their parents’ labor supply, by controlling for secular changes in labor supply among parents of children without disabilities in California and 18 Unlike Model 1, which uses individual variation in labor supply, Models 2 and 3 rely on state-level variation in labor supply and policies to estimate the effect of PFL. There are two reasons for this decision. First, it is customary in DD models (including PFL estimates) to cluster errors at the treatment level (e.g. Angrist & Pischke, 2009; Rossin- Slater et al., 2013), when using panel data (Baum & Ruhm, 2016). Second, if the unobserved individual variation captured in the error terms is distributed similarly in each state, state-clustered errors should account for this unobserved individual variation. 70 those of children with and without disabilities in the control states. Model 3 takes the following form: 𝑦 "#$ = 𝛼+𝛽 ) 𝐷𝑖𝑠𝑎𝑏𝑖𝑙𝑖𝑡𝑦 "# +𝛽 2 𝐶𝐴 "$ + 𝛽 6 𝑃𝑜𝑠𝑡 $ +𝛽 M 𝐷𝑖𝑠𝑎𝑏𝑖𝑙𝑖𝑡𝑦 "# ∗𝐶𝐴 "$ + 𝛽 N 𝐷𝑖𝑠𝑎𝑏𝑖𝑙𝑖𝑡𝑦 "# ∗𝑃𝑜𝑠𝑡 $ + 𝛽 O 𝐶𝐴 "$ ∗𝑃𝑜𝑠𝑡 $ + 𝛽 P 𝐷𝑖𝑠𝑎𝑏𝑖𝑙𝑖𝑡𝑦 "# ∗𝐶𝐴 "$ ∗𝑃𝑜𝑠𝑡 $ + 𝛾 ) 𝑋 "#$ F + 𝛾 2 𝑋 "#$ G + 𝑣 $ +𝜂 "# +𝜀 "#$ (3) where all variables are as described in Model 2 and 𝑃𝑜𝑠𝑡 $ equals one after June 2004. 𝛽 P represents the effect of having a child with a disability in California under the paid family leave regime. The parameter estimate, 𝛽 P , indicates whether there are any statistically significant differences in the labor supply between parents of children with disabilities in California after PFL was enacted, as compared to parents of children without disabilities in California, and those with and without disabilities in the control states. RESULTS Childhood Disability and Parental Labor Supply Prior estimates of the prevalence of childhood disability reveal a relatively steady trend over the past several years, with a slight decline in the percentage and number school-age children receiving services in public schools between 2004 and 2011 (Brault, 2012; NCES, 2017). 19 Figure 1 displays the annual percentage of children ages 0 to 17 in the SIPP sample used in this study by severity of the disability. While the percentage of children with severe disabilities increased over the period, it was offset by a decrease in non-severe disabilities. Some of this shift could be due to idiosyncrasies in the SIPP data. For example, the decline between 2004 and 2008 occurs during the 2004 panel and could reflect attrition in follow up surveys of children with non-severe 19 Some studies report an increase in childhood disability rates over the period in the present study (e.g. Houtrow et al., 2014); however, they use different data and definitions of disability and are not directly comparable. 71 disabilities or a change in the interviews for that panel. The overall trend in the percentage of children with any disability, however, remained relatively stable over the period, with only a slight decline from 7.1 percent in 2001 to 7.0 percent in 2011. Figure 1: Annual Trends in Childhood Disability in the US, by Disability Severity, 2001 – 2011 Note: 2001, 2004, and 2008 SIPP. Child disability includes mild (“non-severe”) and severe disabilities as defined by the Census Bureau in the Americans with Disabilities: 2010 report and found in the Functional Limitations Topical Modules. Disability status is imputed across waves and averaged over each survey year. Age of the child is restricted to under 18. Estimates are weighted using child’s individual weight for the wave. Table 1 displays results from Model 1 estimated separately for mothers and fathers using ordinary least squares. 20 The results confirm the negative association found in the literature between maternal labor supply and having a child with a disability. Mothers of children with disabilities are 2.9 percentage points less likely to be in the labor force than mothers of children with no disabilities. The dynamic measure of labor supply suggests, even accounting for previous labor supply, mothers of children with disabilities are less likely to engage in the workforce. That is, conditional on being in the labor force in the first wave observed, mothers of children with 20 Marginal effects from probit specifications are statistically similar to the linear probability models: mother’s labor force participation is -0.031*** and dropping out is 0.021***; fathers labor force participation is -0.010*** and dropping out is 0.002. I use linear probability models due to the more straightforward interpretation of the coefficients and interaction terms (Ai and Norton, 2003). 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 % Children with Disability Mild Severe Any Disability 72 disabilities are 2.0 percentage points more likely to drop out of the labor force in subsequent waves than mothers of children with no disability. Parameter estimates for other covariates are as expected. Mothers are more likely to be in the labor force if they have fewer and older children, and higher education and household income. They are less likely in the labor force if they are married or in fair or poor health. Table 1: Linear Probability Model of Parental Labor Supply, 2001 - 2011 (Model 1) Mothers Fathers In Labor Force Drop out In Labor Force Drop out Child disability -0.029 *** 0.020 *** -0.014 ** 0.002 Child Controls Age group (ref. 0-5) 6-13 0.079 *** -0.008 ** -0.002 0.003 ** 14+ 0.127 *** -0.015 *** -0.003 0.004 * Female -0.005 0.004 0.001 -0.001 Parent Controls Education (ref. <HS) High school grad/some college 0.127 *** -0.039 0.015 ** 0.005 BA Plus 0.139 *** -0.042 0.023 *** 0.008 ** Race/ethnicity (ref. NH White) NH Black 0.081 *** -0.012 * -0.034 *** 0.016 *** Hispanic -0.027 ** 0.010 0.018 *** -0.001 NH Asian/Other -0.036 ** 0.006 -0.024 *** 0.007 * Married -0.161 *** 0.061 *** 0.009 0.001 Number of children -0.045 *** 0.011 *** 0.001 -0.001 Age 0.023 *** -0.012 *** 0.013 *** -0.006 *** Age 2 0.000 *** 0.000 *** 0.000 *** 0.000 *** Household income (log) 0.059 *** -0.049 *** 0.032 *** -0.024 *** Fair/poor health (ref. good+) -0.173 *** 0.078 *** -0.224 *** 0.073 *** Constant -0.113 0.641 *** 0.044 *** 0.295 *** Panel/wave fixed effects X X X X N 517,951 359,279 418,041 389,814 R 2 0.12 0.09 0.13 0.06 Note: 2001, 2004 and 2008 SIPP. Child disability includes mild (“non-severe”) and severe disabilities as defined by the Census Bureau in the Americans with Disabilities: 2010 report and found in the Functional Limitations Topical Modules. Disability status is imputed across waves. Age of the child is restricted to under 18. Parents’ age is restricted to 25 to 64. Missing indicators included for parent health and education. Standard errors are clustered on the parent. Results are weighted using parent’s individual weight for the wave. All US States included, except New Jersey, Maine, Vermont, North Dakota, South Dakota, and Wyoming. ***p<.001, **p<.01, *<.05 73 Results from Model 1 also indicate a negative relationship between paternal labor force participation and having a child with a disability; however, the magnitude of the estimate is much smaller than mothers, 1.4 percentage points, and they are no more likely to drop out of the labor force than fathers whose children have no disability. As with mothers, the coefficients on other variables are as expected. Fathers’ labor force participation is not associated with the age or number of their children or their marital status, higher educated fathers, on the other hand, are more likely to be in the labor force. Childhood Disability and Labor Force Participation in California and Control States To establish whether parents in the control states are a suitable comparison group for California parents, it is important to compare trends in childhood disability and parental labor supply in the two groups. Figure 2 compares annual trends in the percentage of children with any disability in California with the control states. As in Figure 1, there is a slight decline in the percentage of children with disabilities over time for both groups. The figure reveals the percentage of children with a disability in California is lower throughout the period than in the other states. 21 The nearly parallel linear trend lines, however, indicate similar trends over time in California and the control states. 21 It is unclear why California has lower rates of child disability than other states; however, these results are consistent with other estimates of child disability prevalence (e.g. Brault, 2011). 74 Figure 2: Annual Trends (Observed and Linear Trend) in Child Disability in CA vs Control States, 2001 – 2011 Note: 2001, 2004 and 2008 SIPP. Child disability includes mild (“non-severe”) and severe disabilities as defined by the Census Bureau in the Americans with Disabilities: 2010 report and found in the Functional Limitations Topical Modules. Disability status is imputed across waves and averaged over each survey year. Age of the child is restricted to under 18. Estimates are weighted using child’s individual weight for the wave. Control states include all US States, except New Jersey, Maine, Vermont, North Dakota, South Dakota, and Wyoming. The empirical model assumes that the relationship between child disability and labor supply will be different for parents of children with disabilities in California after the treatment relative to parents of children without disabilities in California and parents in other states; however, there should be no difference in the trend of parental labor force participation prior to PFL. Figure 3 displays monthly labor force participation rates for mothers and fathers of children with a disability in California and the control states prior to PFL’s enactment (May 2001 to June 2004), along with linear time trends. Maternal labor force participation is lower than paternal labor force participation throughout the period for both groups. Labor force participation for both parents is generally lower in California; however, the California measures are noisier due to much smaller samples of children with disabilities. For both mothers and fathers of children with disabilities, Figure 3 shows linear time trends that are similar in California and the control states. There is a slight incline in the labor force participation trend for fathers in California; thus, I test the sensitivity of main results to the selection of this comparison group. 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 % Children with a Disability CA Control States 75 Figure 3: Annual Trends (Observed and Linear Trend) in Parental Labor Force Participation for Children with Disabilities, CA vs Control States, May 2001 to June 2004 Note: 2001, 2004 and 2008 SIPP. Child disability includes mild (“non-severe”) and severe disabilities as defined by the Census Bureau in the Americans with Disabilities: 2010 report and found in the Functional Limitations Topical Modules. Disability status is imputed across waves and averaged over each survey year. Age of the child is restricted to under 18. Estimates are weighted using parent’s individual weight for the wave. Control states include all US States, except New Jersey, Maine, Vermont, North Dakota, South Dakota, and Wyoming. The total number of observations for mothers is: 1,143 in California and 11,713 in control states. For fathers, the total number of observations is: 785 in California and 7,908 in control states. Table 2: Linear Probability Model of Parental Labor Force Participation in California versus Control States (pre-PFL only), 1999-2003 (Model 2) Mothers Fathers In Labor Force In Labor Force CA*Child Disability*Time Trend -0.007 -0.006 Child Controls X X Parent Controls X X Panel/Wave Fixed Effects X X N 164,859 133,176 R^2 0.12 0.15 Note: 2001, 2004 and 2008 SIPP. Child disability includes mild (“non-severe”) and severe disabilities as defined by the Census Bureau in the Americans with Disabilities: 2010 report and found in the Functional Limitations Topical Modules. Disability status is imputed across waves. Age of the child is restricted to under 18. Parents’ age is restricted to 25 to 64. Missing indicators included for parent health and education. Standard errors are clustered at the state levelt. Results are weighted using parent’s individual weight for the wave. Control states include all US States, except New Jersey, Maine, Vermont, North Dakota, South Dakota, and Wyoming. ***p<.001, **p<.01, *<.05 Table 2 reports results from models testing the statistical significance of the trends in labor force participation observed in the pre-PFL period in Figure 3, all else equal. The models indicate that labor force participation trends for parents of children with disabilities are statistically similar to trends in parents of children without disabilities in California and parents in the control states. 40.0 50.0 60.0 70.0 80.0 90.0 100.0 May-01 Sep-01 Feb-02 Jun-02 Nov-02 Apr-03 Aug-03 Jan-04 May-04 % in Labor Force Control Moms CA Moms Control Dads CA Dads 76 That is, the three-way interaction between California, the presence of a child with a disability, and a linear time trend through June 2004 is not significant in the labor force participation models for mothers and fathers. Thus, significant changes found in parental labor force participation after PFL can be attributed to the policy. Paid Family Leave and Labor Supply of Parents of Children with Disabilities The results in Tables 1 and 2 above demonstrate that the presence of a child with a disability reduced parental labor supply, but annual trends in parental labor force participation were no different for parents of children with disabilities in California than parents in other states and those in California whose children have no disability. A central question in this study is whether access to paid leave reduces the negative effect on parental labor supply or, put differently, whether labor force participation of these parents increases after they are exposed to paid leave. Since I do not observe whether these parents take leave or changes in the number of hours they work, these models represent estimates of the intent to treat on the parents’ extensive labor supply. Table 3 displays results from the main DDD model testing the effect of California’s PFL on the labor supply of parents whose children have a disability. Results indicate that labor force participation among mothers of children with disabilities in California increased by 5.7 percentage points after PFL relative to mothers of children in other states and those of children with no disability in California. The estimate represents a 9.4 percent increase over the pre-PFL baseline labor force participation for that group. The effect is smaller, but still highly significant, for fathers. Labor force participation among fathers of children with disabilities in California increased 3.9 percentage points after PFL, which is equivalent to a 4.3 percent increase over their pre-PFL baseline. Moreover, the likelihood of dropping out of the labor force decreased in California after 77 PFL for mothers and fathers by 3.6 and 0.9 percentage points, respectively, but only the estimate for mothers is statistically significant. Compared to a relatively low baseline probability, the effect represents a large decline in dropping out for mother and fathers, 30.3 percent and 26.5 percent. The drop out results should be interpreted with caution due to small samples in California, especially for fathers. However, the direction of the effect is consistent with the main results. That is, paid family leave promoted labor force attachment among parents of children with disabilities in California. Table 3: Difference-in-difference-in-differences (DDD) Parental Labor Supply, by Child Disability Status, California versus Control States, Before and After PFL, 2001-2011 (Model 3) Mothers Fathers In Labor Force Drop out In Labor Force Drop out CA*Child Disability*Post 0.057 *** -0.036 *** 0.039 *** -0.009 Child Disability -0.026 * 0.015 -0.007 -0.003 California -0.008 0.009 ** -0.013 *** 0.009 *** Post 0.025 -0.014 * -0.001 -0.001 Child Disability*CA -0.067 *** 0.016 * -0.022 *** 0.001 *** Child Disability*Post 0.000 0.010 -0.010 0.009 California*Post -0.034 *** 0.015 *** 0.009 *** -0.006 *** Child Controls X X X X Parent Controls X X X X Panel/Wave Fixed Effects X X X X N 517,951 359,279 418,041 398,814 R^2 0.12 0.09 0.13 0.06 Note: 2001, 2004 and 2008 SIPP. Child disability includes mild (“non-severe”) and severe disabilities as defined by the Census Bureau in the Americans with Disabilities: 2010 report and found in the Functional Limitations Topical Modules. Disability status is imputed across waves. Age of the child is restricted to under 18. Parents’ age is restricted to 25 to 64. Missing indicators included for parent health and education. Standard errors are clustered at the state level. Results are weighted using parent’s individual weight for the wave. Control states include all US States, except New Jersey, Maine, Vermont, North Dakota, South Dakota, and Wyoming. ***p<.001, **p<.01, *<.05 This positive result does not mean that PFL fully compensated for the negative effect of having a child with a disability. Recall, labor force participation was lower, and the likelihood of dropping out was higher, for parents of children with disabilities than those whose children had no disability, both in California and the control states. Thus, it is helpful to analyze changes in the 78 level of participation before and after PFL. Figure 3 displays the predicted probability of being in the labor force for mothers and fathers of children with and without disabilities in California before and after PFL. For mothers (Panel A), labor force participation was higher among mothers of children with no disability in both periods (the lighter bar); however, the gap in participation between the two groups of California mothers decreased substantially after PFL. The difference between mothers of children with and without disabilities decreased from 9.3 percentage points in the pre-PFL period to only 3.6 after the policy, a 61.3 percent decline in the participation gap. For fathers, there was no difference in labor force participation after the policy. Figure 3: Regression-Adjusted Labor Force Participation of Parents of Children with and without Disabilities in California Before and After PFL, 2001 – 2011 Panel A: Mothers Panel B: Fathers Note: 2001, 2004 and 2008 SIPP. Child disability includes mild (“non-severe”) and severe disabilities as defined by the Census Bureau in the Americans with Disabilities: 2010 report and found in the Functional Limitations Topical Modules. Disability status is imputed across waves. Age of the child is restricted to under 18. Parents’ age is restricted to 25 to 64. Missing indicators included for parent health and education. Standard errors are clustered at the state level. Results are weighted using parent’s individual weight for the wave. Control states include all US States, except New Jersey, Maine, Vermont, North Dakota, South Dakota, and Wyoming. Figure 4 displays the predicted probability of being out of the labor force in a given wave, conditional on participating in the first wave. That is, the figure shows the change in the predicted likelihood of dropping out among parents of children with and without disabilities in California before and after PFL. For mothers of children with disabilities, the likelihood of dropping out of the labor force decreased substantially, while the likelihood of dropping out among other mothers 0 20 40 60 80 100 Pre Post No Disability Disability 0 20 40 60 80 100 Pre Post No Disability Disability 79 in California remained about the same. After PFL, the difference in dropping out between mothers of children with and without disabilities decreased from 3.0 to 0.4 percentage points and was no longer statistically significant. Panel B demonstrates the relatively low probability of dropping out among fathers in general, less than 3.6 percent overall; however, there are differences before and after PFL. The portion of fathers in both groups dropping out of the labor force decreased in the post-PFL period and the difference in the predicted probability of dropping out between fathers of children with and without disabilities decreased after PFL. Overlapping confidence intervals in both periods indicate that fathers of children with disabilities in California were statistically no more likely to leave the workforce than other fathers in either period, which is consistent with the negative (but not significant) parameter estimate in the regression. Figure 4: Regression-Adjusted Conditional Labor Force Participation (Dropping Out) of Parents of Children with and without Disabilities in California Before and After PFL, 2001 – 2011 Panel A: Mothers Panel B: Fathers Note: 2001, 2004 and 2008 SIPP. Child disability includes mild (“non-severe”) and severe disabilities as defined by the Census Bureau in the Americans with Disabilities: 2010 report and found in the Functional Limitations Topical Modules. Disability status is imputed across waves. Age of the child is restricted to under 18. Parents’ age is restricted to 25 to 64. Missing indicators included for parent health and education. Standard errors are clustered at the state level. Results are weighted using parent’s individual weight for the wave. Control states include all US States, except New Jersey, Maine, Vermont, North Dakota, South Dakota, and Wyoming. Sensitivity Analysis 0 2 4 6 8 10 12 14 16 Pre Post No Disability Disability 0 2 4 6 8 10 12 14 16 Pre Post No Disability Disability 80 Table 5 reports estimated DDD coefficients from variations on the main model. Panel A shows the results are consistent across several changes in the model’s assumptions, with some variation in the magnitude of the effect. Specifically, although the magnitude changes slightly, results are robust to the inclusion of state fixed effects and a dummy indicator of SSI receipt. Altering the period of the study (excluding recession years and the year of the policy, 2004), likewise, causes little change in the main effect. Results are also robust to the elimination of a potentially endogenous variable (household income) and the effect of leave for new parents (children under 1 year). The effect size for fathers is particularly robust across specifications, ranging from 3.8 to 4.3 percentage points, while DDD estimates for mothers are slightly more sensitive to changing assumptions. For example, excluding waves with imputed disability status for the child reduces the magnitude of the effect for mothers by 46 percent to 3.1 percentage points, while it increases the effect of fathers by 10 percent to 4.3 percentage points. Panel B in Table 5 displays results from models testing an alternative measure of a child’s need: parent-reported health of the child. As mentioned above, parent-reported measures of the child’s health status could be endogenous if the parent is justifying their decision whether to work or not. Moreover, only a small percentage of children in the sample were reported in fair or poor health, about 2.0 percent. Thus, while it offers some insight into the effect of child health on parental labor supply, results are subject to some bias. Nevertheless, the parameter estimate for the three-way interaction between California, fair or poor child health, and post-PFL is positive and significant for mothers of sick children, but not their fathers. When health is included in a composite measure with child disability (whether sick or disabled), the estimate is positive and significant for mothers and fathers of children with a broader definition of special health care needs. 81 Table 5: Sensitivity Analysis (DDD Coefficient) – All Covariates Included Mothers’ LFP Fathers’ LFP Main Model, Table 4 0.057 *** 0.039 *** Panel A: Main Model Adjusted State fixed effect 0.055 *** 0.039 *** SSI recipient (dummy) 0.053 *** 0.042 *** Exclude recession (2008 and later) 0.039 ** 0.042 *** Exclude policy year (2004) 0.055 *** 0.041 *** Excluding waves with imputed disability 0.031 * 0.043 *** Excluding children under age 1 0.054 *** 0.039 *** Excluding household income 0.049 *** 0.038 *** Panel B: Alternative Definitions of Child Need Fair/Poor Health (parent-rated) 0.087 *** -0.001 Disability OR Fair/Poor Health 0.056 *** 0.020 ** Panel C: Other Control Groups Next 3 largest (FL, NY, TX) 0.068 0.028 Others w/Temp. Disability (NY, RI, HI) 0.057 * 0.068 ** Regression control group 0.045 * 0.044 *** Note: 2001, 2004 and 2008 SIPP. Child disability includes mild (“non-severe”) and severe disabilities as defined by the Census Bureau in the Americans with Disabilities: 2010 report and found in the Functional Limitations Topical Modules. Disability status is imputed across waves. Age of the child is restricted to under 18. Parents’ age is restricted to 25 to 64. Missing indicators included for parent health and education. Standard errors are clustered at the state level. Results are weighted using parent’s individual weight for the wave. Control states include all US States, except New Jersey, Maine, Vermont, North Dakota, South Dakota, and Wyoming. ***p<.001, **p<.01, *<.05 Results from Model 2 above indicate the control states in the main specification are a suitable comparison group for California; however, I also test alternative control groups using groups identified in prior studies and a group determined using regression analysis that tests for individual state-level differences in pre-PFL trends. 22 Panel C displays results from models using alternative control groups: the three next largest states, other states with temporary disability insurance (TDI), and the regression control group. The results for all three groups are qualitatively similar to the main model; however, only the latter two are statistically different than zero. 22 This group was determined by estimating Model 2 to compare pre-PFL trends in California and each of the other states for mothers and fathers, similar to a robustness test conducted by Baum and Ruhm (2016). Only states with statistically similar pre-PFL trends in parental labor force participation are included in the control group. See Appendix Table 4 for included states. 82 DISCUSSION Balancing work and family life may prove especially difficult for parents of children with disabilities in modern U.S. families. Prior evidence of reduced maternal labor supply in the presence of a child with a disability indicates these mothers struggle to maintain labor force attachment. In recent years, as labor force participation in the U.S. declined, parents of children with disabilities may have been especially vulnerable to the consequences of lost wages. A lower- wage mother, for example, without access to employer-sponsored leave would not be able to take time off care for her child, but she may not be able afford substitute child care. In the absence of institutional supports, she may drop out of the labor force and rely instead on government programs like the Temporary Assistance for Needy Families (TANF) and Medicaid to make ends meet. Alternatively, workers with access to employer-sponsored leave can take time off to care for their child, but they may exhaust their leave allowance or face workplace stigma associated with their intermittent absences. Paid family leave policies, as in California, give parents time to tend to the needs of their child with a disability, such as doctor and therapy appointments or hospitalizations, while maintaining their employment and income. These policies also send a positive signal to employers and the workforce, promoting the support of workers’ caregiving responsibilities. Without government-sponsored family leave, employers bear the full cost of their employees’ absences. They not only lose the employee’s time, but they may have to continue their pay or risk the employee leaving the workforce. California’s PFL program provides a portion of the employees’ wages and the cost to the employer is reduced to the segments of time the employee is absent from work. In this study, I confirm previous findings of a negative relationship between maternal labor supply and having a child with a disability, using both static and dynamic measures of labor force 83 participation. I find new evidence of a negative association between paternal labor supply and having a child with a disability. Consistent with expectations, the estimate for fathers is smaller than for mothers, -1.4 versus -2.9 percentage points. Access to paid family leave in California increased labor force participation among mothers and fathers of children with disabilities by 5.7 and 3.9 percentage points, respectively, and decreased the likelihood of dropping out of the workforce. These results are consistent with the hypothesis that the increased demands on families’ resources would reduce labor supply among at least one parent, while paid leave promotes labor force attachment. Similarly, the larger effect of a child with a disability on maternal labor supply is consistent with the expectation that mothers' lower opportunity costs induce them into childcare more than fathers. Likewise, the stronger effect of PFL on maternal labor supply suggests, by providing paid time to care for the child, the policy reduced the opportunity cost of working. Gender differences in the estimated relationship between labor supply and having a child with a disability suggest mothers are more likely to conform to normative gender expectations of caring, by reducing their labor supply. While the large effect of paid leave on maternal labor supply among mothers of children with disabilities could signal a shift in expectations, such an interpretation should be made with caution. Predicted probabilities of maternal labor force participation before and after PFL show mothers of children with disabilities are still much less likely (about 30 percentage points) to be in the labor force than fathers. Thus, there may be ongoing expectations of these mothers both at home and in the workplace, as they attend to their child’s ongoing needs. This study takes an important first step in identifying the how paid leave influences work and care decisions among parents of children with disabilities. It also provides insight into how fathers’ labor force participation responds to having a child with a disability. While this study 84 focuses on the extensive margin of labor supply, mothers have been shown to reduce working hours, as well. Further, data limitations preclude estimating the first order effect of paid leave (i.e. leave-taking) on these parents. An interesting and important extension of this work would estimate whether mothers and fathers of children with disabilities are more likely to be on leave after policy enactment, whether there were changes in the number of hours these parents work, and whether there are differences across socioeconomic background. Also, given the potential role of gender expectations in care, future studies could examine the relationship between attitudes towards gender equality, labor supply, and paid leave among parents of children with disabilities. The implications of this research extend beyond parents caring for children with disabilities to the larger caregiving community. As the population ages, working-age adults care for both their elderly parents and their own children. This caregiving burden can pose difficulties for workers trying to maintain labor force attachment. If paid leave helps these caregivers continue working, there could be broader benefits for economic growth and future taxpayer support. 85 REFERNCES Ai, C., & Norton, E. C. (2003). Interaction terms in logit and probit models. Economics letters, 80(1), 123-129. American Academy of Pediatrics. Initiatives for Children with Special Needs Project Advisory Committee (2004). The medical home. Pediatrics, 113, 1545-1547. Angrist, J. D., & Pischke, J. S. (2008). Mostly harmless econometrics: An empiricist's companion. Princeton university press. Bartel, A. P., Rossin-Slater, M., Ruhm, C. J., Stearns, J., & Waldfogel, J. (2018). Paid Family Leave, Fathers’ Leave-Taking, and Leave-Sharing in Dual-Earner Households. Journal of Policy Analysis and Management, 37(1), 10-37. Baum, C.L. & Ruhm, C.J. (2016). The Effects of Paid Family Leave in California on Labor Market Outcomes. Journal of Policy Analysis and Management, 35(2), 333-356. Becker, G.S. (1981, 1991). A Treatise on the Family. Harvard University Press. Becker, G. S. (1985). Human capital, effort, and the sexual division of labor. Journal of labor economics, 3(1, Part 2), S33-S58. Bedard, K., & Rossin-Slater, M. (2016). The Economic and Social Impacts of Paid Family Leave in California: Report for the California Employment Development Department. Retrieved from: https://www.dol.gov/wb/media/CA_Final_Report_2015_Final_Report.pdf Bianchi, S. M. (2011). Family change and time allocation in American families. The ANNALS of the American Academy of Political and Social Science, 638(1), 21-44. Blau, F. D., & Kahn, L. M. (2005). Changes in the labor supply behavior of married women: 1980-2000 (No. w11230). National Bureau of Economic Research. Brault, M. W. (2008). “Americans With Disabilities: 2005,” Current Population Reports, P70- 117, U.S. Census Bureau, Washington, DC, 2008, available at www.census .gov/prod/2008pubs/p70-117.pdf. Brault, M. W. (2011). School-Aged Children with Disabilities in US Metropolitan Statistical Areas: 2010. American Community Survey Briefs. ACSBR/10-12. US Census Bureau. Brault, M. W. (2012). “Americans With Disabilities: 2010,” Current Population Reports, P70- 1313, U.S. Census Bureau, Washington, DC, 2008, available at < www.census.gov/library/publications/2012/demo/p70-131.pdf. Burton, P., Chen, K., Lethbridge, L., & Phipps, S. (2017). Child health and parental paid work. Review of Economics of the Household, 15(2), 597-620. 86 Byker, T. (2016). Paid Parental Leave Laws in the United States: Does Short-Duration Leave Affect Women's Labor-Force Attachment? American Economic Review, 106(5): 242-46. Chung, P. J., Lui, C. K., Cowgill, B. O., Hoffman, G., Elijah, J., & Schuster, M. A. (2012). Employment, family leave, and parents of newborns or seriously ill children. Academic pediatrics, 12(3), 181-188. Corman, H., Reichman, N. E., & Noonan, K. (2003). Mothers' and Fathers' Labor Supply in Fragile Families: The Role of Child Health (No. w9918). National Bureau of Economic Research. Earle, A., & Heymann, J. (2012). The cost of caregiving: Wage loss among caregivers of elderly and disabled adults and children with special needs. Community, Work & Family, 15(3), 357-375. Gould, E. (2004). Decomposing the effects of children's health on mother's labor supply: is it time or money?. Health economics, 13(6), 525-541. Hartley, S. L., Barker, E. T., Seltzer, M. M., Floyd, F., Greenberg, J., Orsmond, G., & Bolt, D. (2010). The relative risk and timing of divorce in families of children with an autism spectrum disorder. Journal of Family Psychology, 24(4), 449. Hays, S. (1998). The cultural contradictions of motherhood. Yale University Press. Heitmueller, A. (2007). The chicken or the egg?: Endogeneity in labour market participation of informal carers in England. Journal of health economics, 26(3), 536-559. Hope, S., Pearce, A., Whitehead, M., & Law, C. (2016). Effects of child long-term illness on maternal employment: longitudinal findings from the UK Millennium Cohort Study. The European Journal of Public Health, ckw132. Hotz, V. J., Klerman, J. A., & Willis, R. J. (1997). The economics of fertility in developed countries. Handbook of population and family economics, 1(Part A), 275-347. Huang, R., & Yang, M. (2015). Paid maternity leave and breastfeeding practice before and after California's implementation of the nation's first paid family leave program. Economics & Human Biology, 16, 45-59. Jinnah, H. A., & Stoneman, Z. (2008). Parents' experiences in seeking child care for school age children with disabilities—Where does the system break down?. Children and Youth Services Review, 30(8), 967-977. Kerr, S. P. (2016). Parental Leave Legislation and Women's Work: A Story of Unequal Opportunities. Journal of Policy Analysis and Management, 35(1), 117-144. 87 Klerman, J. A., Daley, K., & Pozniak, A. (2012). Family and medical leave in 2012: Technical report. Cambridge, MA: Abt Associates Inc. Kvist, A. P., Nielsen, H. S., & Simonsen, M. (2013). The importance of children's ADHD for parents' relationship stability and labor supply. Social Science & Medicine, 88, 30-38. Lichtman-Sadot, S. (2014). The Value of Postponing Pregnancy: California’s Paid Family Leave and the Timing of Pregnancies. The BE Journal of Economic Analysis & Policy, 14(4), 1467-1499. Lichtman-Sadot, S., & Bell, N. P. (2017) Child Health in Elementary School Following California's Paid Family Leave Program. Journal of Policy Analysis and Management, 36(4), 790-827. Lukemeyer, A., Meyers, M. K., & Smeeding, T. (2000). Expensive children in poor families: out-of-pocket expenditures for the care of disabled and chronically ill children in welfare families. Journal of Marriage and Family, 62(2), 399-415. Milkman, R., & Appelbaum, E. (2013). Unfinished business: Paid family leave in California and the future of US work-family policy. Cornell University Press. Mincer, J. (1962). Labor force participation of married women: A study of labor supply. In Aspects of labor economics (pp. 63-105). Princeton University Press. Mishel, L., Gould, E., & Bivens, J. (2015). Wage stagnation in nine charts. Economic Policy Institute, 6. Moore, J. C. (2008). Seam bias in the 2004 SIPP panel: Much improved, but much bias still remains. US Census Bureau Statistical Research Division Survey Methodology Research Report Series, 3, 2008. Murphy, N. A., Christian, B., Caplin, D. A., & Young, P. C. (2007). The health of caregivers for children with disabilities: caregiver perspectives. Child: care, health and development, 33(2), 180-187. National Survey of Children with Special Health Care Needs. NS-CSHCN 2009/10. Data query from the Child and Adolescent Health Measurement Initiative, Data Resource Center for Child and Adolescent Health website. Retrieved 04/13/2017 from www.childhealthdata.org. Orlando, A. W. (2013). Letter to the One Percent. Lulu. com. Parish, S. L., Seltzer, M. M., Greenberg, J. S., & Floyd, F. (2004). Economic implications of caregiving at midlife: Comparing parents with and without children who have developmental disabilities. Mental retardation, 42(6), 413-426. 88 Parish, S. L., Cloud, J. M., Huh, J., & Henning, A. N. (2005). Child care, disability, and family structure: Use and quality in a population-based sample of low-income preschool children. Children and Youth Services Review, 27(8), 905-919. Powers, E. T. (2001). New estimates of the impact of child disability on maternal employment. The American Economic Review, 91(2), 135-139. Powers, E. T. (2003). Children’s health and maternal work activity estimates under alternative disability definitions. Journal of Human Resources, 38(3), 522-556. Rønsen, M., & Sundström, M. (2002). Family policy and after-birth employment among new mothers–A comparison of Finland, Norway and Sweden. European Journal of Population/Revue europeenne de demographie, 18(2), 121-152. Romley, J. A., Shah, A. K., Chung, P. J., Elliott, M. N., Vestal, K. D., & Schuster, M. A. (2017). Family-Provided Health Care for Children With Special Health Care Needs. Pediatrics, 139(1), e20161287. Rossin-Slater, M., Ruhm, C. J., & Waldfogel, J. (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(2), 224-245. Ruhm, C. J. (1998). The economic consequences of parental leave mandates: Lessons from Europe. The quarterly journal of economics, 113(1), 285-317. Seltzer, M. M., Greenberg, J. S., Floyd, F. J., Pettee, Y., & Hong, J. (2001). Life course impacts of parenting a child with a disability. American journal on mental retardation, 106(3), 265-286. Stabile, M., & Allin, S. (2012). The economic costs of childhood disability. The Future of Children, 22(1), 65-96. State of California Employment Development Department http://www.edd.ca.gov/Disability/Paid_Family_Leave.htm Strickland, B., McPherson, M., Weissman, G., Van Dyck, P., Huang, Z. J., & Newacheck, P. (2004). Access to the medical home: results of the National Survey of Children with Special Health Care Needs. Pediatrics, 113(Supplement 4), 1485-1492. Tamborini, C. R., Kim, C., & Sakamoto, A. (2015). Education and lifetime earnings in the United States. Demography, 52(4), 1383-1407. Tisinger, R., Johnson, M., Hoffman, A., Davis, C. Jean-Baptiste, M., & Tanamore, M. (2016). Understanding Attitudes on Paid Family Leave: Discussions with Parents and Caregivers in California, New Jersey and Rhode Island. L&M Policy Research for Department of Labor, Chief Evaluation Office. Retrieved from: 89 https://www.dol.gov/asp/evaluation/completed- studies/Paid_Leave_AwarenessBenefitsBarriers.pdf U.S. Census Bureau (2010). SIPP Disability Status Recoding Programming Code/Logic. Retrieved from: https://www.census.gov/people/disability/methodology/sippprogramming.html U.S. Census Bureau (2016). Survey of Income and Program Participation, SIPP Introduction and History. Retrieved from: https://www.census.gov/programs-surveys/sipp/about/sipp- introduction-history.html U.S. Centers for Disease Control and Prevention. 2009-2010 National Survey of Children with Special Health Care Needs. Retrieved from: www.cdc.gov/nchs/slaits/cshcn.htm. U.S. Department of Labor, Wage and Hour Division (2012). Fact Sheet #28: The Family and Medical Leave Act. Retrieved from: https://www.dol.gov/whd/regs/compliance/whdfs28.pdf Waldfogel, J. (1998). Understanding the" family gap" in pay for women with children. The Journal of Economic Perspectives, 137-156. Waldfogel, J., Higuchi, Y., & Abe, M. (1999). Family leave policies and women's retention after childbirth: Evidence from the United States, Britain, and Japan. Journal of Population Economics, 12(4), 523-545. Warfield, M. E. (2005). Family and work predictors of parenting role stress among two-earner families of children with disabilities. Infant and Child Development, 14(2), 155-176. Wasi, N., van den Berg, B., & Buchmueller, T. C. (2012). Heterogeneous effects of child disability on maternal labor supply: Evidence from the 2000 US Census. Labour Economics, 19(1), 139-154. Williams, J. (2001). Unbending gender: Why family and work conflict and what to do about it. Oxford University Press. Wolfe, B. L., & Hill, S. C. (1995). The effect of health on the work effort of single mothers. Journal of human resources, 42-62. 90 APPENDIX Appendix Table 1: Covariate Means, Mothers of Children with Disabilities Before and After PFL, California versus Control States California Control States 2/2001 to 6/2004 7/2004 to 12/2011 2/2001 to 6/2004 7/2004 to 12/2011 Mother Variables In Labor Force (%) 60.7 64.3 * 69.1 70.1 * Dropout (%) 11.9 10.1 10.1 11.5 ** Education Less than HS (%) 23.4 17.6 15.1 11.4 HS Grad (%) 56.4 58.8 64.3 65.1 BA+ (%) 20.2 23.6 20.6 23.5 Race/Ethnicity NH White (%) 48.5 48.1 69.6 65.8 NH Black (%) 10.8 6.3 16.1 16.1 Hispanic (%) 31.8 38.0 10.4 12.5 NH Asian/Other (%) 8.9 7.7 3.9 5.6 Married (%) 65.3 66.1 64.4 63.4 * Number of children 2.6 2.4 ** 2.6 2.5 *** Age 38.6 40.2 *** 38.1 38.4 *** Mean household income (2016 dollars) 7,606 7,817 6,250 5,924 *** Health Good+ (%) 77.2 81.0 80.0 80.7 Fair/poor (%) 18.0 12.9 15.1 14.7 Missing (%) 4.8 6.1 4.8 4.7 N 1,143 1,983 11,713 25,731 Note: 2001, 2004 and 2008 SIPP. Child disability includes mild (“non-severe”) and severe disabilities as defined by the Census Bureau in the Americans with Disabilities: 2010 report and found in the Functional Limitations Topical Modules. Disability status is imputed across waves. Age of the child is restricted to under 18. Parents’ age is restricted to 25 to 64. Esitmates are weighted using mother’s individual weight for the wave. Control states include all US States, except New Jersey, Maine, Vermont, North Dakota, South Dakota, and Wyoming. ***p<.001, **p<.01, *<.05 indicates level of significance of difference in means before and after PFL for mothers of children with disabilities in California and the control states. 91 Appendix Table 2: Covariate Means, Fathers of Children with Disabilities Before and After PFL, California versus Control States California Control States 2/2001 to 6/2004 7/2004 to 12/2011 2/2001 to 6/2004 7/2004 to 12/2011 Dad Variables In Labor Force (%) 89.9 91.0 * 92.7 91.9 Dropout (%) 3.4 3.6 2.0 3.0 *** Education Less than HS (%) 30.1 18.4 13.9 12.6 HS Grad (%) 42.8 55.1 59.3 61.0 BA+ (%) 27.1 26.5 26.8 26.4 Race/Ethnicity NH White (%) 57.9 48.6 76.8 70.7 NH Black (%) 5.2 9.2 8.4 9.6 Hispanic (%) 28.4 35.6 11.5 15.1 NH Asian/Other (%) 8.6 6.6 3.3 4.6 Married (%) 93.3 88.7 *** 91.6 89.5 *** Number of children 2.7 2.5 *** 2.6 2.5 *** Age 41.3 42.9 *** 41.3 41.2 Mean Household Income (2016 dollars) 8,630 9,316 4,870 4,518 *** Health Good+ (%) 84.9 73.5 83.8 85.4 Fair/poor (%) 10.5 19.9 11.8 10.0 Missing (%) 4.7 6.6 4.4 4.6 N 785 1,339 7,908 17,432 Note: 2001, 2004 and 2008 SIPP. Child disability includes mild (“non-severe”) and severe disabilities as defined by the Census Bureau in the Americans with Disabilities: 2010 report and found in the Functional Limitations Topical Modules. Disability status is imputed across waves. Age of the child is restricted to under 18. Parents’ age is restricted to 25 to 64. Estimates are weighted using father’s individual weight for the wave. Control states include all US States, except New Jersey, Maine, Vermont, North Dakota, South Dakota, and Wyoming. ***p<.001, **p<.01, *<.05 indicates level of significance of difference in means before and after PFL for fathers of children with disabilities in California and the control states. 92 Appendix Table 3: Covariate Means, All Children in Mother and Father Sample, Before and After PFL, California versus Control States California Control States 2/2001 to 6/2004 7/2004 to 12/2011 2/2001 to 6/2004 7/2004 to 12/2011 Mother Sample Disability (%) 5.4 5.4 7.9 7.7 Fair/poor health (%) 2.2 1.8 *** 2.1 1.9 *** Disability or Fair/Poor health (%) 7.0 6.6 9.2 8.9 Age (years) 8.9 9.1 *** 8.9 8.9 Female (%) 50.4 49.7 49.4 49.2 N 1,143 1,983 143,902 317,731 Father Sample Disability (%) 4.4 4.5 6.7 6.6 Fair/poor health (%) 2.1 1.6 *** 1.6 1.4 *** Disability or Fair/Poor health (%) 6.1 5.7 7.8 7.5 Age (years) 8.6 8.7 ** 8.6 8.6 Female (%) 50.4 50.0 49.0 48.6 N 17,434 30,044 115,743 254,822 Note: 2001, 2004 and 2008 SIPP. Child disability includes mild (“non-severe”) and severe disabilities as defined by the Census Bureau in the Americans with Disabilities: 2010 report and found in the Functional Limitations Topical Modules. Disability status is imputed across waves. Age of the child is restricted to under 18. Parents’ age is restricted to 25 to 64. Estimates are weighted using child’s individual weight for the wave. Control states include all US States, except New Jersey, Maine, Vermont, North Dakota, South Dakota, and Wyoming. ***p<.001, **p<.01, *<.05 indicates level of significance of difference in means before and after PFL for children in California and the control states. 93 Appendix Table 4: State Regressions for Control Group, p-value for DDD Coefficient State Mother’s LFP (p-value) Regression Control Group Father’s LFP (p-value) Regression Control Group Alabama 0.014 No 0.044 No Alaska 0.037 No 0.005 No Arizona 0.034 No 0.079 Yes Arkansas 0.027 No 0.029 No Colorado 0.004 No 0.015 No Connecticut 0.048 No 0.098 Yes Delaware 0.039 No 0.082 Yes D.C. 0.049 No 0.001 No Florida 0.002 No 0.004 No Georgia 0.002 No 0.159 Yes Hawaii 0.018 No 0.067 Yes Idaho 0.019 No 0.291 Yes Illinois 0.068 Yes 0.010 No Indiana 0.058 Yes 0.027 No Iowa 0.014 No 0.185 Yes Kansas 0.016 No 0.009 No Kentucky 0.050 No 0.015 No Louisiana 0.140 Yes 0.170 Yes Maryland 0.068 Yes 0.114 Yes Massachusetts 0.002 No 0.028 No Michigan 0.041 No 0.026 No Minnesota 0.140 Yes 0.061 Yes Mississippi 0.034 No 0.029 No Missouri 0.029 No 0.042 No Montana 0.000 No 0.753 Yes Nebraska 0.051 Yes 0.017 No Nevada 0.027 No 0.061 Yes New Hampshire 0.037 No 0.271 Yes New Mexico 0.877 Yes 0.018 No New York 0.062 Yes 0.305 Yes North Carolina 0.075 Yes 0.057 Yes Ohio 0.831 Yes 0.517 Yes Oklahoma 0.021 No 0.014 No Oregon 0.045 No 0.017 No Pennsylvania 0.232 Yes 0.048 No Rhode Island 0.043 No 0.022 No South Carolina 0.029 No 0.027 No Tennessee 0.173 Yes 0.002 No Texas 0.058 Yes 0.382 Yes Utah 0.040 No 0.079 Yes Virginia 0.004 No 0.187 Yes Washington 0.002 No 0.111 Yes West Virginia 0.039 No 0.021 No Wisconsin 0.026 No 0.095 Yes Note: 2001, 2004 and 2008 SIPP. Child disability includes mild (“non-severe”) and severe disabilities as defined by the Census Bureau in the Americans with Disabilities: 2010 report and found in the Functional Limitations Topical Modules. Disability status is imputed across waves. Age of the child is restricted to under 18. Parents’ age is restricted to 25 to 64. All covariates included. Missing indicators included for parent health and education. Standard errors are clustered at the state level. Results are weighted using parent’s individual weight for the wave. ***p<.001, **p<.01, *<.05 94 Public Support for Long-Term Care of People with Dementia and Their Caregivers: An Analysis of the Patchwork of State and Federal Policies by Johanna Thunell Abstract The growing number of people with Alzheimer’s disease or related dementia (ADRD) in the United States means our reliance on the informal caregivers supporting them is expected to increase in the coming years. Although state and federal policies (e.g. National Family Caregiver Support Program, CARE Acts, and paid family leave) are evolving to support informal caregivers, there is little systematic evidence of their effect on caregivers’ work, care, and health outcomes. In this study, I evaluate the research on interventions and policies supporting caregivers of people with ADRD. Specifically, I analyze the literature on interventions, programs, and policies for ADRD caregivers to determine whether we have a sufficient understanding of their effectiveness and what future research is needed to close the gap in our knowledge. The literature reveals individual caregiver interventions can reduce stress and caregiver burden and delay nursing home entry. Yet, these findings are based on small-scale interventions or state and local implementation of federal programs (e.g. NFCSP). This study reveals a lack of systematic research estimating the overall effects of national initiatives on ADRD caregivers. The study concludes with several recommendations for future research in this policy area. 95 INTRODUCTION Long-term services and support (LTSS) for the United States’ elderly population draws on a cadre of informal, family caregivers, many of whom care for an individual with Alzheimer’s disease and related dementias (ADRD). Some government programs subsidize LTSS for elderly people, both in the community and in nursing homes, but many lack access to paid care. Medicaid, the largest payer of LTSS, only covers about 10 percent of the population ages 65 and older, while the largest health care program for the elderly, Medicare, does not provide LTSS benefits. For those without access to Medicaid, long-term care insurance can fill the gap between subsidized care and unpaid, informal care; however, less than 15 percent of elderly people hold long-term care policies (Brown & Finkelstein, 2011). As a result, the majority of caregiving hours for people with ADRD is provided by family and friends (Center for Disease Control and Prevention, 2016). According to a recent study from the National Academies of Sciences, Engineering, and Medicine (2016), 17.7 million Americans care for people ages 65 and older with limited cognition or physical abilities, while the Alzheimer’s Association (2017) estimates 15.9 million caregivers provide informal care to people with ADRD alone annually. 23 This informal care represents a substantial savings on LTSS for the U.S. economy. In 2016, the total value of informal care for people with ADRD was estimated at $230 billion (Alzheimer’s Association, 2017). Yet, the number of people with Alzheimer’s, their caregiving needs, and the associated costs are expected to rise over the next few decades. Zissimopoulos, Crimmins, and St. Clair (2015) estimate the number of people with Alzheimer’s disease will grow from 3.6 million in 2010 to 9.1 million by 2050, and the associated value of informal care will increase from $126 billion to $361 billion. 23 Estimates of the number of caregivers and their value and the percentage associated with ADRD vary depending on the ages included in the estimate and how care is valued. 96 While the demand for informal care is expected to increase substantially over the coming decades, the supply of family caregivers has declined. Adult daughters, daughters-in-law, and wives traditionally shouldered the bulk of caregiving duties, but trends in family structure and employment altered the availability of these potential caregivers over the past several decades. Women still comprise the majority of caregivers, 62 percent of all ADRD caregivers, but the number of women available for caring and the amount of time they have to devote to caregiving has decreased over the past several decades (Center for Disease Control and Prevention, 2016). Between 1970 and 2014, the percentage of women in the labor force increased from 43 to 57 percent (U.S. Bureau of Labor Statistics, 2016). At the same time, decreasing family size and delayed childbearing also reduced the number of daughters available for care. The total fertility rate (which is equivalent to expected average family size) decreased from 2.48 to 1.86, while the average age of first-time mothers increased from 21.4 to 26.3 and the percentage of births to women ages 35 and older increased substantially from 1.0 to 9.4 percent (Centers for Disease Control and Prevention, 2018). These changes in family structure and women’s employment are especially likely to impact caregiving of the growing number of people with ADRD, as daughters, rather than spouses, most frequently provide their care (Friedman et al., 2015). Thus, the increasing number of women juggling work, children, and caring for an aging parent with ADRD may struggle to balance their competing priorities. Some public policies are in place to address the needs of caregivers, such as family leave and caregiver support programs, but few studies systematically explore the policies and programs available to caregivers and analyze evidence of their effectiveness. Two exceptions are Doty and Spillman (2015) and a recent book from the National Academies of Sciences, Engineering, and Mathematics (2016). Doty and Spillman (2015) provide a detailed overview of the programs 97 providing LTSS and policies supporting caregivers, but their discussion is limited to the policies’ descriptions, rather than academic analyses of their evidence. On the other hand, the NAS book discusses the state of family caregiving, including programs and interventions for family caregivers and some of the studies evaluating the impact of those interventions; however, the book is more geared towards providing recommendations to lawmakers based on current literature, than critically examining studies, determining the shortcomings of existing research, and identifying what additional information is needed. Moreover, the book considers the elder care community generally, rather than the ADRD community specifically. In contrast, the purpose of the present study is to review the literature on policies and programs supporting caregivers of people with ADRD, assess what we can learn from existing studies, and highlight gaps in our knowledge about their effectiveness. This study builds on previous work and contributes to the literature in three key ways. First, I focus explicitly on the unique needs of caregivers of people with ADRD. This is an important contribution, as the literature demonstrates caring for a person with ADRD is quite different than non-dementia caregiving and their policy needs may also differ. Second, I apply a lens from economic theory to help understand how these policies may impact the work and care decisions of potential caregivers, given that a growing number of them confront the tradeoff between work and care. Finally, I analyze the literature on interventions, programs, and policies for ADRD caregivers to determine whether we have a sufficient understanding of their effectiveness and what future research is needed to close the gap in our knowledge. This paper proceeds as follows. I first provide an overview of the patchwork of options to pay for LTSS for people with ADRD to elucidate the institutionalized reliance on informal care in the U.S. Next, I provide a theoretical background on work and care decisions and discuss 98 differences between dementia and non-dementia care. Then, I review the existing programs and policies supporting caregivers of people with ADRD and analyze the literature on their effects. I conclude by summarizing what we know and what research still needs to be done. BACKGROUND ON LONG-TERM SERVICES AND SUPPORTS IN THE U.S. The U.S. Department of Health and Human Services estimates a private room in a nursing home costs over $92,000 annually ($82,000 for a semi-private room), whereas the median cost for formal in-home care is about $48,000; however, most people cared for in their homes receive only informal care (Alzheimer’s Association, 2017; Genworth Financial, Inc., 2017). Paying for (formal) LTSS can be especially costly for people with Alzheimer’s disease and related dementias since they often require care for long periods, e.g. a larger percentage people with ADRD receive care for 5 or more years compared to those with other chronic illnesses (CDC, 2016). Public policy options to pay for LTSS are limited. Whether a person qualifies for Medicaid and/or Medicare or has their own private long-term care insurance often dictates their access to services and the type of services they receive. There are four primary resources for obtaining LTSS: Medicaid, long-term care insurance, private payment, and informal care. Table 1 summarizes the options for paying for formal LTSS, including the population and services covered, specific programs for home- and community-based care, and the estimated percentage of total LTSS spending from that source. 24 Medicaid is, by far, the largest payer of both institutional and home- and community-based LTSS, followed by private payment, long-term care insurance, and Medicare. 24 These estimates were obtained from multiple sources (Brown & Finkelstein, 2011; O’Shaughnessy, 2014; U.S. Centers for Medicare and Medicaid Services, 2017) and, thus, are only estimates. I have found no single resource for estimating LTSS expenditures across these four categories. 99 Table 1: Long-Term Services and Supports Policy/Program Population Covered Services Covered Programs for Home-Based Care % of LTSS Spending Medicaid (Federal and State) Elderly individuals and couples meeting low income requirements; 4.6 million seniors (about 10%) covered Medical, prescriptions long- term services and support (varies by state), home-based and nursing homes Home and Community-Based Waiver Program Community First Choice (currently 5 states) 61% Medicare (Federal) All people 65+; over 95% enrolled Medical, home health aide under some circumstances, short stays in nursing home Parts A and B; medically necessary in-home health care Minimal Long-Term Care Insurance Policy holders; 14% of people ages 60+ hold LTCI policies Long-term services and supports, home based and nursing homes (varies across policies) 90% of policies cover both in-home and nursing homes 4% Private Out-of- Pocket Pay Individual without Medicaid and long- term care insurance N/A N/A 22-33% Sources: AARP Public Policy Institute (2017); Administration for Community Living (2017); Brown & Finkelstein (2011); Doty and Spillman (2016); U.S. Centers for Medicare and Medicaid Services (2017); O’Shaughnessy (2014). Medicaid covers LTSS provided in nursing homes and home and community-based care; however, specific benefits vary across states (U.S. Centers for Medicare and Medicaid Services, 2017). For example, federal Medicaid guidelines require home health care benefits, but the availability and type of services (e.g. adult day care and homemaking) is determined by individual states. As a means-tested program, not all elderly people or those with ADRD, are eligible to receive services under Medicaid. In total, only about 10 percent of all seniors are enrolled in Medicaid. 25 On the other hand, over 95 percent of people 65 and older are enrolled in Medicare, which generally does not cover LTSS. Rather, Parts A and B provide hospice care, short-term nursing home stays after hospitalization, and some types of in-home health care, such as physical 25 Author’s calculations based on number of low-income seniors enrolled in Medicaid, 4.6 million, and American Community Surveys (ACS) estimates of populations 65 and older, 49,215,165, in 2016. 100 or occupational therapy for homebound beneficiaries (U.S. Centers for Medicare and Medicaid Services, 2018). 26 Thus, while the government is the largest payer of formal LTSS, the expenditures are limited to a relatively small portion of the population, the majority of which live in nursing homes. In a well-developed market, long-term care insurance (LTCI) might fill much of the gap between Medicaid and private payment for LTSS, but several flaws in the LTCI market make take- up low. Brown and Finkelstein (2011) argue supply-side and demand-side factors limit the size of the long-term care insurance market. Common market failures, such as asymmetric information and imperfect competition, as well as high loads, 27 reduced demand due to costly premiums, and crowding out by Medicaid are all explanations for the undersized LTCI market. It is likely that a more competitive market would reduce the crowding out caused by Medicaid for all but the poorest seniors; however, making the market function more competitively would require not only Medicaid reform, but alterations to the insurance policies themselves. As a result, only about 14 percent of people ages 60 and older hold LTCI policies and they account for only about 4 percent of overall LTSS spending (Brown & Finkelstein, 2011). Moreover, ownership is skewed towards the wealthiest individuals, likely due to the high cost of premiums. In 2010, the median annual premium (on a policy with comprehensive coverage) for a 55-year-old was $4,637, whereas a 70-year-old would pay $16,901 for the same policy. When comparing the annual cost to the expected benefit, it is clear why people at the lower end of the wealth distribution, and even the middle, are more likely to fall back on Medicaid for their formal LTSS needs. For example, say a 70-year-old man buys a policy. He will pay nearly $170,000 before turning 80. If 26 Some studies include these services in estimations of the percentage of LTSS from each source (e.g. Hagen, 2013); however, Medicare does not provide the type of regular, ongoing services that are central to the present discussion. 27 A “load” is essentially the expected payoff from insurance policies divided by the premium cost, where higher loads are worse for the consumer (Brown & Finkelstein, 2011). 101 he only lives a nursing home for one year, his benefits will pay out about $90,000 (minus a deductible). If he is relatively risk averse or expects to be in a nursing home longer, he may opt to purchase the policy. Thus, the expected payout could be quite low depending on the individual’s age and risk tolerance. In sum, Medicaid eligibility rules, Medicare’s lack of LTSS coverage, and the underdeveloped LTCI market have institutionalized a reliance on private pay (up to one third of spending on LTSS) and informal caregiving, particularly for the middle of the wealth distribution. 28 Indeed, about 55 percent of all LTSS is provided by informal caregivers (Hagen, 2013). The default to informal care is especially salient for people with ADRD as over 80 percent are cared for in their homes (Center for Disease Control and Prevention, 2016). As the number of people with ADRD grows in the coming decades, the rising demand for informal care necessitates support for their caregivers. The remaining sections discuss the decisions caregivers make and how policies may influence those decisions, research on policies and programs for caregivers, and gaps in our understanding of the impact of policies on ADRD caregivers. DECISIONS ABOUT CAREGIVING AND WORK The caregiving decision of family members determines the supply of informal caregivers available to people with ADRD. Since less than a quarter of ADRD caregivers are 65 or older, many family members providing informal care are likely working as well (CDC, 2016). Thus, policies aimed at reducing caregiver burden or stress, or allowing time off work for care, will likely not only improve the caregiver’s health, but also affect their labor market outcomes. Economic 28 People at the low end of the wealth distribution would be covered by Medicaid, while those at the higher end are more likely to have LTCI or afford private payment. 102 theory indicates adult children and spouses of people with dementia allocate their time between caring, work, and leisure activities, subject to budget and time constraints (Becker, 1991, in Johnson & Lo Sasso, 2000). 29 That is, their decision to provide care depends on the amount of time they devote to work and leisure. For working-age spouses and adult children, all else equal, increases in time devoted to caring will decrease their labor supply and vice versa. Yet, several additional factors influence the decision of whether to care (likelihood) and the amount of care (hours), including a caregiver’s own health, their wages, an adult child’s marital status and number of children, the availability of other caregivers, the cost of alternative (paid) care, and non-labor income (Johnson & Lo Sasso, 2000). Among higher-wage workers, the opportunity cost associated with caring is high and they may opt to purchase alternative care services. Similarly, a potential caregiver’s own poor health or that of their spouse or child decreases the likelihood and amount of caregiving provided to the older adult. Adult children with their own families, the so-called “sandwich generation,” already juggle childcare and work and generally have less time to care for their elderly parents (Gans, Katz, & Zissimopoulos, 2013; Miller, 1981). Additionally, if there are other unpaid sources of care (e.g. siblings), a caregiver may devote fewer hours to caregiving and there would be less of an effect of caring on their labor supply. Empirical estimates of the relationship between working and caregiving (and, ultimately, the role of policy) suffer from several potential of sources of endogeneity. The decision of how much time to allocate to work and care is jointly determined and may be affected by unobservable characteristics leading to selection into work or care, e.g. if a family member provides care due to a lack of employability (Heitmueller, 2007; Lilly, LaPorte & Coyote, 2007). Researchers commonly employ instrumental variables estimation (IVE) to overcome this endogeneity bias, 29 See Johnson and Lo Sasso (2000) for a useful extension of Becker’s model. 103 although IVE limits the interpretation of the caregiving effect to those whose care decision is influenced by the selected instruments (Angrist & Pischke, 2008). Plausible instruments include measures of the care recipient’s health (e.g. Heitmeuller, 2007; Van Houtven, Coe & Skira, 2013) and family characteristics, such as the number or proximity of other children (e.g. Ettner, 1995; Spillman & Long, 2009). In general, IV estimates of work and care suggest much larger magnitude effects than OLS estimates. For example, Ettner (1995) instrumented for care hours using the number of siblings and parental education. While less precise, IV estimates of the effect of caring on work were consistently higher: Caring more than 10 hours per week reduced conditional working hours by 73 and 117 hours in the OLS and IV estimates, respectively. Despite these attempts to correct for endogeneity, studies on the labor supply of caregivers reveal mixed results: The negative effect on participation and working hours (especially among female caregivers) spans a wide range but includes estimates lacking statistical significance (Lilly, LaPorte & Coyote, 2007; Johnson & Lo Sasso, 2000; Schmitz & Westphal, 2017). Johnson and Lo Sasso (2000) argue many of the early studies suffered from selection bias, omitted variables, or non-representative samples. More recent work concludes caregiving for elderly adults negatively affects labor supply through its reduction on working hours, rather than participation (Schmitz & Westphal, 2017; Van Houtven, Coe & Skira, 2013). For example, Schmitz and Westphal (2017) use panel data and propensity score matching to model caregiving effects over an eight-year period. They find a reduction in the likelihood female caregivers worked fulltime (by about 35 percent), but caregivers were not statistically less likely to be in the labor force, suggesting the decrease in working hours drove the decline in full-time work. The reduction in work hours can, in turn, lead to long-term wage penalties. Taken together, the literature generally supports the conclusion that caregiving negatively impacts female caregivers’ labor supply in some 104 way, but the evidence on the magnitude of the effect is mixed and there is little evidence on the effect on male employment. A caregiver’s health is another critical component of the work and care decision; however, most studies isolate health from work and care. That is, the majority of research focuses either on work and care (e.g. Ettner, 1995) or caregiver health, stress, and burden (e.g. Schulz & Sherwood, 2008). As indicated above, caregiving can reduce labor supply by limiting the amount of time available for work. In addition, the chronic stress associated with caregiving also takes a significant toll on caregivers’ physical and mental health, which could, in turn, contribute to a reduction in labor supply (Johnson & Lo Sasso, 2000). Similar to the relationship between work and care, estimations of the effect of caregiving on health suffer from several sources of endogeneity. Selection into caregiving may be related to the caregiver’s health or unobserved characteristics, such as the closeness of their relationship with the care recipient or their social engagement more generally. Much of the literature on caregiver health uses small convenience samples in clinical settings or cross-sectional data, which cannot disentangle the effect of caregiving from these factors (see Schulz & Sherwood, 2008 and Vitaliano, Zhang & Scanlan, 2003). Studies of the effect of caregiving on family caregivers generally find negative effects on general health (e.g. self-reported health, depression, medication, chronic conditions) and physiological measures (e.g., stress hormones, blood pressure), and wellbeing (e.g. financial hardship, social activities) (e.g. Schulz & Sherwood, 2008; Wolff et al. 2016); however, differences in samples, methods, and measurement across studies limit the ability to generalize these conclusions. While the research using cross-sectional data is subject to selection bias, some studies use longitudinal data and survival models to attempt to overcome the potential bias. In a landmark study of caregiver mortality, Schulz and Beach (1999) follow a cohort of elderly people 105 for about four years and find an increased risk of mortality among caregivers over the period, which was especially high for caregivers reporting high levels of strain. Meta-analyses have concluded that family caregivers generally experience higher levels depression, anxiety and negative physical health (e.g. Schulz & Sherwood, 2008; Vitaliano et al., 2003). Although effect sizes differ across studies, Vitaliano et al.’s (2003) analysis reveals, across studies, stress hormones were 23 percent higher among caregivers than non-caregivers. More recently, Caputo, Pavalko, and Hardy (2016) use fixed effects to estimate the long-term effects of caregiving on depression, functional limitations, and mortality among middle-aged women. Their approach compares within-person estimates of changes in health and mortality for caregivers and non-caregivers over a 14-year period. They find caregiving had a sustained effect on depression and functional limitations even with the inclusion of several potential confounders (e.g. social engagement and health history), but the risk of mortality was no higher for caregivers relative to non-caregivers. Despite the general consensus across studies that caregivers experience higher levels of stress and reduced health than non-caregivers, some work suggests there are benefits to caregiving, arguing that traditional stress-process models may need to be reconsidered (Roth et al., 2015); however, caregivers may experience strain and positive aspects of caregiving simultaneously and more work is needed to understand the interaction of these effects (e.g. Beach et al., 2000). In addition to the general association between caregiving time and mental and physical health, research highlights several specific factors associated with increased risk of these negative outcomes for caregivers of older adults. Caregiver burden generally increases with age, co- residence, time as a caregiver, financial stress and being female, but decreases with income, educational attainment, and access to social support (Adelman et al., 2014; Pinquart & Sorenson, 2007). The care recipient’s level of need also consistently predicts increased caregiver stress. 106 Caring for a family member with more physical limitations, cognitive impairment, and behavioral problems leads to higher levels of caregiver burden (Pinquart & Sorenson, 2007; Schoenmakers, Buntinx & Delepeleire, 2010), which has been associated with increased likelihood of nursing home placement (Spillman & Long, 2009). Caregivers of people with ADRD are at an even greater risk of caregiver stress and burden, and are more likely to be younger and working for pay, relative to non-dementia caregivers (Kasper et al., 2015; Ory et al., 1999). The nature of caring for people with dementia is different than caring for someone without cognitive impairment; dementia caregivers provide more hours of care for a longer duration, perform more functions, and undertake more difficult tasks than non-dementia caregivers (Ory et al., 1999). For example, caregivers of people with dementia provide an average of 171 hours of informal care per month, compared to 66 hours for those with no cognitive impairment (Friedman et al. 2015). Studies comparing dementia and non-dementia caregivers find dementia status increased physical strain, emotional strain, and financial hardship (Ory et al., 1999). Depression and anxiety symptoms and clinical diagnoses and risk of serious illness are also more prevalent among dementia caregivers than non-dementia caregivers (Schulz et al., 1995; Schulz & Martire, 2004; Shaw et al. 1999). Public policies may reduce caregiver burden and influence their decisions about care and work, yet, the effect of public policies on family members’ care provision is less clear ex ante. Public policy can enter the caregiving decision as supportive services for caregivers or financial benefits for working caregivers and can influence the time they devote to care, work, or both. Policies that allow caregivers to combine work and caregiving can also increase the time they devote to care. For example, paid leave policies that provide time off work and wage replacement during caregiving spells allow workers to maintain attachment to the labor force. Alternatively, a 107 tax deduction for paid formal care would decrease the opportunity cost of work and promote working caregivers to purchase substitute care. If a family member decides to provide at least some care, public policies can also affect their health and wellbeing with services like respite care, counseling, support groups, and other supportive services. 30 This type of caregiver support generally aims to improve their physical and mental health. The model described above indicates that an adult child’s own poor health decreases the amount of time devoted to work, care, or both (Johnson & Lo Sasso, 2000). It follows that programs improving the health of a caregiver will increase the time they devote to working and/or caring. Thus, theory indicates such programs would impact the work and care decisions of adult children and working age spouses (albeit indirectly); however, it is unclear whether improved health would increase labor supply, caregiving, or both. In addition, given that the literature demonstrates a difference between dementia and non- dementia caregiving, one-size-fits-all caregiver programs and policies may influence their decisions differently. As an example, for working caregivers of people with ADRD, paid leave may be helpful for short spells off work, but the protracted nature of the disease may require periodic time off over the course of several years. If the policy only allows for a relatively short amount of time off, doesn’t allow intermittent leave, limits lifetime access to leave, or excludes certain groups of workers, it may not affect the decisions of ADRD caregivers or it may impact ADRD caregivers’ decisions differently. In short, public policies that promote caregiver wellbeing and improve their ability to balance work and care may influence their decisions about caregiving and/or working, but the effect depends on the type of policy, the caregiver’s employment, wages, and health and may also 30 Respite care givers fulltime caregivers a break from caring, either at an adult daycare center, an institutional setting, or through respite care services in the home. 108 depend on whether they are caring for someone with dementia. Empirical estimations of the mediating effect of policies and programs on the work and care decisions must consider these (potentially endogenous) factors. As I discuss in the next section, researchers have used a variety of methods to account for many of these endogenous confounders; however, few studies systematically evaluate the effect of these policies on work and care decisions, whether directly or indirectly. PUBLIC POLICIES FOR CAREGIVERS OF PEOPLE WITH ADRD As discussed above, prior work demonstrates the effect of work on caregiving and vice versa, as well as the effect of caregiving on health. In this section, I analyze the literature on the mediating role of policies and programs for caregivers of people with ADRD. Two caregiver outcomes are particularly salient from a policy perspective and for working caregivers: whether working-age caregivers maintain labor force attachment (and, thus, contribute to overall economic productivity) and whether family members provide informal care (or, put another way, whether costly institutionalization is delayed or forgone). The following discussion pays special attention to whether the extant literature sheds lights on policies’ impact on these outcomes. Policies explicitly targeting caregivers typically offer supportive services - information, counseling, support groups, and respite care - rather than monetary benefits, although a handful of programs offer some financial relief for working caregivers. The outcomes explored in the literature, however, are not confined to caregivers. Indeed, many programs can benefit care recipients, as well, by delaying nursing home entry or reducing care recipient depression. Table 2 provides an overview of the policies and programs for ADRD caregivers, the eligible population, 109 outcomes and findings, whether the studies are specific to the policy versus a similar intervention, and whether the studies are specific to the ADRD caregiver population. Table 2: Government-Funded Supportive Services for Informal Caregivers 31 Policy/Program Covered Population Outcomes and Results Program- Specific Analysis ADRD Caregiver Studies Alzheimer’s Disease Initiative- Specialized Supportive Services (ADI-SSS) Community- dwelling people with ADRD and their caregivers Care recipient: Nursing home entry (-); unmet needs (-); hospital, physician and emergency room visits (-); behavior problems and depression (-) Caregiver: Burden (-), depression (-), and health (+); quality of life (+) [varies by type of service] No Yes Alzheimer’s Disease Supportive Services Program (ADSSP) Community- dwelling people with ADRD and their caregivers Same as ADI-SSS No Yes Caregiver Advise, Record, Enable (CARE) Act Family caregivers of any care recipient after hospitalization (state-level policy) Re-admission rates (-) and average hospital costs (-) No No Family Medical Leave Act (FMLA) of 1993 Working caregivers who worked for same employer at least 1,250 hours over the last 12 months at an employer with at least 50 employees New parents’ labor supply (+) and leave-taking (+) Yes No Lifespan Respite Care Act Caregivers of people with special needs of all ages Caregiver burden (unclear) No Yes National Alzheimer’s Call Center Caregivers of community-dwelling people with ADRD N/A No No 31 There are several smaller programs left off this list, e.g. federal grants to tribal organizations for caregiver support programs. For a list of caregiver support programs see Tables 1-1 and 1-2 in National Academies of Sciences, Engineering, and Medicine (2016). 110 National Family Caregiver Support Program (NFCSP) Caregivers of people 70 and older Service scope and availability (+); caregiver burden (-), caregiving mastery (+), hours of care (+), and service satisfaction (+) [varies by type of service] Yes No Paid Family Leave Most working caregivers in California, New Jersey, New York, Rhode Island, and Washington (eligibility varies by state) Nursing home utilization (-); mental and physical health of caregivers (null) Yes No Tax Deductions Taxpayers providing over 50% support to a co-residing dependent N/A No No Unemployment Insurance Unemployed working caregivers who left their job for “family reasons” (provisions vary by state) N/A No No Sources: AARP Public Policy Institute (2017); Administration for Community Living (2017); Belle et al. (2006); Doty and Spillman (2016); U.S. Centers for Medicare and Medicaid Services, (2017); National Academies of Science, Engineering, and Medicine (2016)); National Association of Area Agencies on Aging (2016). Note: (-) indicates a negative relationship between the program/intervention and the outcome, (+) indicates a positive relationship between the program/intervention and the outcome. ADRD Caregiver Support: ADI-SSS, ADSSP, and National Alzheimer’s Call Center The growing number of people with Alzheimer’s, costs related to their care, and the greater burden faced by their caregivers prompted federal and state governments to develop programs specifically targeting caregivers of people with Alzheimer’s disease and related dementia. The Alzheimer’s Disease Initiative-Specialized Supportive Services (ADI-SSS) and Alzheimer’s Disease Supportive Services Program (ADSSP) provide federal grants to states, local governments, and private service providers to develop and test programs serving people with ADRD living in the community and their caregivers (Administration for Community Living, 2017). These localized programs offer case management and care coordination, respite, counseling, education, 111 and training to people with ADRD and their caregivers. In addition, the Alzheimer’s Association operates the Administration on Aging-funded National Alzheimer’s Call Center, which supports caregivers 24 hours a day with information, referrals, and decision-making assistance (Alzheimer’s Association, 2018). To my knowledge, no research explores the impact of the National Alzheimer’s Call Center (or a similar intervention) on caregiver outcomes and no single, national study systematically analyzes the programs supported by ADI-SSS or ADSSP; however, a significant number of studies have evaluated psychosocial interventions geared towards dementia caregivers. The research reveals benefits of these interventions for caregivers and care recipients, but effects vary depending on the type and amount of services. Differences in the treatments themselves and control groups across many of these studies render direct comparisons difficult. For example, control groups often have access to typical (similar) services, while the treatment group receives slightly different services, or the control group may be placed on a wait list and, thus, partially comprise the intent- to-treat group. Nevertheless, meta-analyses accounting for some of these methodological differences reveal programs with education and training components are most effective at reducing caregiver burden and stress and improving wellbeing, while highly structured, multicomponent interventions have the largest impact on care recipient institutionalization (Gallagher-Thompson & Coon, 2007; Elvish et al., 2013; Parker et al., 2008; Pinquart & Sorenson, 2006). In the early 2000s, the National Institute of Health funded a large, multicomponent randomized controlled trial, Resources for Enhancing Alzheimer’s Caregiver Health (REACH II), with the explicit goal of improving dementia caregivers’ health and wellbeing (Belle et al., 2006). Whereas most previous interventions used small local samples, REACH II took place in five cities and included information, counseling, role playing, social support, education, and other services through 112 telephone support groups. Evaluations of the bundle of services revealed treatment effects similar to other dementia caregiver interventions: reduced caregiver burden and depression and improved self-rated health (Elliot, Burgio & DeCoster, 2010). For a recent study under the Department of Health and Human Services, researchers visited five dementia care programs in different states and reviewed evaluations of their effectiveness (Wiener et al., 2016). 32 The setting for the interventions differed slightly from a medical clinic to in-person or telephone meetings with people in their homes, but services were similar: referrals, education, skills training, assessments, and counseling. In the formal evaluations of these models of dementia care (mostly using pre/post research designs), the most consistent areas of improvement were reduced caregiver depression and burden (Bass et al., 2003; Bass et al., 2013; Bass et al., 2015; Belle et al., 2006; Easom et al., 2013), improved quality of life (Belle et al., 2006; Samus et al., 2014), and reduced unmet needs (Bass et al., 2015; Samus et al., 2014). Care recipients also experienced positive outcomes, including fewer hospital, physician, and emergency room visits (Bass et al., 2015; Clark et al., 2004). Interestingly, studies at two different sites revealed improvement in the ADRD patient’s behavior problems, depression, or dementia (Callahan et al., 2006; Easom, Alston & Coleman, 2013; LaMantia et al., 2015). The literature on the effect of individual interventions on people with ADRD and their caregivers provides promising evidence of the benefits of these types of programs; however, despite some agreement across studies about the potential effect of these programs, the research is limited to mostly randomized controlled trials (RCTs), which lack generalizability. Meta-analyses that calculate average effects across several smaller studies may help assess the potential impact of national or state public policies, as they combine multiple studies and a variety of interventions. 32 Appendix B in Wiener et al. (2016) includes an exhaustive list of dementia care-related programs, with a description of the program, its setting, and any associated research. 113 Several such studies have analyzed the impact of psychosocial interventions on caregivers of people with dementia (e.g. Corbett et al., 2012; Gallagher-Thompson & Coon, 2007; Pinquart & Sorenson, 2006). Specifically, these meta-analyses evaluate the effect of interventions on caregiver burden and depression, demonstrating considerable variation across study, type of intervention, and length of intervention in the magnitude of the interventions’ effect. For example, Pinquart and Sorenson (2006) report standardized effect sizes across studies revealing effects on caregiver burden, depression, subjective wellbeing, and care recipient outcomes that range from 0.10 to 0.42 standard deviation units; however, many estimates are not statistically different from zero. Across interventions, they find an average 0.24 standard deviation unit reduction in depression, which is equivalent to a 12 percentage point post-treatment difference between the treatment and control groups. For caregiver burden, the average effect is -0.12, equivalent to 6 percentage point post- treatment difference between treatment and control groups. In contrast, Corbett et al. (2012), using similar methods, find no significant effect of interventions on caregiver burden. Nuanced differences across meta-analyses and the interventions themselves make drawing conclusions about the effectiveness of a national effort under ADSSP and ADI-SSS difficult. The meta-analyses typically use a similar approach to averaging effects across studies: standardized mean difference between treatment and control groups (e.g. Pinquart & Sorensen, 2006). However, the studies differ significantly in their selection criteria and, thus, vary in the number and type of intervention studies included in their analyses. Some analyses are limited to only RCTs (e.g. Jensen et al., 2015), while others only include interventions related to a specific type of intervention (e.g. education, Jensen et al., 2015). The number of intervention studies included in the reviews ranges from five (Jensen et al., 2015) to 127 (Pinquart & Sorenson, 2006), with most falling below 20 (e.g. Corbett et al., 2012; Gallagher-Thompson & Coon, 2007). Within the intervention studies 114 themselves, there are also significant differences in the samples, such as a lack of gender, age, or racial/ethnic diversity (Gallagher-Thompson & Coon, 2007). Thus, although there is some agreement in the findings of the intervention studies and the meta-analyses averaging their effects, the impact of a national-level policy is unclear. Moreover, the effect of these types of interventions on two salient outcomes from both an individual and policy perspective, labor force attachment among working-age caregivers and informal care hours, is unclear. An important next phase in this area of research would compare caregiver and care recipient outcomes from programs across states, levels of funding, and types of services using representative data to estimate the overall benefits of these programs. Focusing on working caregivers’ ability to maintain labor force attachment and provide informal care would be an especially fruitful avenue of future research that could assist lawmakers in determining whether the benefits outweigh the costs associated with these programs. National Family Caregiver Family Support Program State and local governments administer and coordinate support services covered under the National Family Caregiver Family Support Program (NFCSP), which is the only federal program specifically targeting caregivers of elderly adults (Administration on Community Living, 2017). Since 2000, NFCSP has granted about $145 to $150 million annually to states, with the explicit goals of reducing caregiver burden and delaying institutionalization. Services for caregivers, including support groups, counseling, and respite care, are coordinated through Area Agencies on Aging (AAAs). In a study of the program’s implementation of NFCSP, Feinberg and Newman (2006) survey all 50 states and the District of Columbia and find NFCSP increased service scope and availability, but service options varied both across and within states. The most common 115 services offered were: respite care, information and assistance, education and training, and support groups. The first national evaluation of NFCSP is currently underway to assess whether it improves emotional, physical, and financial outcomes for participants; however, Chen, Hedrick, and Young (2010) survey caregivers participating in one county’s implementation (King County, Washington) of the NFCSP. Their one-time survey reveals some positive associations between specific services and caregiver outcomes. For example, caregivers utilizing referrals, counseling services, and financial support were more satisfied with the services they received than those that used other services (e.g. respite care), while those who received only counseling experienced less caregiver burden. The local, cross-sectional nature of the survey limits generalization of the findings and does not allow a causal interpretation of results. Nevertheless, this study provides some initial evidence that NFCSP reduced caregiver burden. The full-scale evaluation, which runs from 2016 to 2018, plans to use nationally representative samples of NFCSP participants and a similar control group (Westat, Inc., 2017). Assuming care is taken to match treatment and control groups and use appropriate statistical models to estimate the treatment effect on caregiver outcomes (e.g. difference-in-differences or instrumental variables), the national evaluation will shed light on broader policy effects; however, it is unclear whether ADRD caregivers will be analyzed separately in the study. Caregiver Advise, Record, Enable (CARE) Act and Similar Laws Thirty-six states and the District of Columbia have enacted laws requiring hospitals to record the name and contact information of a family caregiver on patients’ medical records, alert caregivers of discharge dates, and provide care instructions to the caregivers (AARP Public Policy 116 Institute, 2017). The law passed under different names in the states, but was deemed the Caregiver Advise, Record, Enable (CARE) Act by AARP during their advocacy efforts and many states maintain that moniker. Provisions vary across states, but the goals remain the same: to reduce readmission and potentially avoidable hospitalizations. The local nature of the laws make direct comparisons and national evaluations difficult; thus, there have been no systematic evaluations of these policies since they began in 2014. Yet, research on hospital-to-home transition interventions indicates CARE-type policies can reduce hospital readmission. In a randomized trial in Colorado, Coleman et al. (2006) find treatment group participants had lower readmission rates at 30 and 90 days after discharge and their average hospital costs were lower compared to the control group. As with other RCTs, the study is limited in its generalizability to other settings, but yields promising results that could be tested in a broader environment using representative data. Guided transitions from the hospital to home environment may be especially important for people with dementia and their caregivers. Daiello et al. (2014) use 2009 Medicare claims data in Rhode Island and multivariate logistic regression to estimate the relationship between dementia and hospital readmissions within 30 days after discharge. They find that, all else equal, people with a dementia diagnosis were more likely to be readmitted within 30 days. The analysis includes all hospitalizations in the Medicare fee-for-service claims data in a one-year period and, thus, represents the population in Rhode Island; however, the generalizability to other states is unclear. A future study using similar data across states would shed additional light on the role of dementia in hospital readmissions. Moreover, the data could be used to exploit between-state variation in CARE Act policies, e.g. policy enactment dates, to estimate their effect on readmissions more generally. 117 Lifespan Respite Care Act The Lifespan Respite Care Act of 2006 provides federal grants to states to coordinate community-based respite care for caregivers of people with special needs of all ages, but the program’s focus is on state coordination, rather than direct services (Administration for Community Living, 2017). The grants fund activities such as establishing state-wide respite plans, training programs, and helplines. Like NFCSP and CARE Act policies, little is known about the effects of the policy on a national scale. Moreover, the effect of respite care itself on caregiver outcomes is unclear in the literature: Meta-analyses of respite interventions on caregiver burden have found no effect (Lee & Cameron, 2004), positive effects (Vandepitte et al., 2016), or only for certain groups (Mason et al., 2007). In the most recent review of the literature on respite care for caregivers of people with dementia, Vandepitte et al. (2016) conclude day care (respite) programs effectively reduced caregiver burden. However, drawing broad conclusions about respite care from their study is limited by the variation in the programs included in the studies and some conflicting findings among them. The literature review examines RCTs, quasi-experimental designs, and studies using secondary data, but does not conduct a meta-analysis and differences between the studies preclude such an analysis. As with the CARE Act, between-state (or county or metropolitan area) comparisons using representative data is needed to fully understand the relationship between respite care and caregiver outcomes, especially among dementia caregivers. Thus, further research is needed to understand whether access to respite care under Lifespan Respite Care Act improves use of services and whether respite improves caregivers’ wellbeing, especially among caregivers of people with ADRD. 118 Family and Medical Leave Act and Paid Family Leave A growing body of literature examines the effect of family leave policies in the U.S. on new parents’ labor market, health, and family outcomes (e.g. Baum & Ruhm, 2016; Rossin-Slater et al., 2013); however, less is known about their effect on caregivers and the aging population. The only national family leave program, Family and Medical Leave Act (FMLA), provides up to 12 weeks of unpaid, job-protected leave, including continued health insurance coverage, to people caring for an ill loved one (U.S. Department of Labor, 2012). FMLA excludes people working for employers with less than 50 employees and those who have not worked for the same employer for 12 months, among other restrictions that disqualify about 40 percent of workers (Klerman, Daley & Pozniak, 2014). Five states, including California, have enacted paid family leave programs that provide wage replacement while on family leave. In California, nearly all workers are eligible to receive 60 to 70 percent of their wages (depending on their income) up to a cap of $1,216 for up to six weeks of leave (State of California Employment Development Department, 2017). To date, California’s PFL has primarily been used as parental leave, but about 10 percent of claims are for caregiving (Bedard & Rossin-Slater, 2016). Although no research estimates the impact of the federal FMLA policy on caregivers, two recent studies explore the effect of California’s Paid Family Leave (PFL) Program on caregivers of the elderly. As is common in the research on new parents, both studies employ a difference-in- differences approach to estimate the effect of PFL on nursing home utilization (Arora & Wolf, 2017) and caregiver health (Gimm & Yang, 2016). Using state-level data from multiple sources, Arora and Wolf (2017) find that elderly nursing home usage in California decreased by 11 percent after PFL, compared to control states. The model allows for a causal interpretation of their results; however, their outcome measure accounts for any nursing home stay of at least one day. Thus, as 119 the authors point out, they may underestimate the true effect of PFL on nursing home usage if it acts to reduce length of stay. Further, although the results suggest PFL can reduce nursing home utilization, the underlying mechanism (or first-order effect) is not explored. That is, the authors assume informal and formal care are substitutes and working caregivers will take time off to care for their aging loved ones. An important next step is to estimate the effect of PFL on leave-taking of caregivers, as well as their labor supply and caregiving decisions. Gimm and Yang (2016), on the other hand, find no impact of PFL on the mental health (depression) and self-reported physical health of caregivers in California using Health and Retirement Study (HRS) data. A few characteristics of the study make the interpretation and generalization of their results uncertain. First, the authors do not isolate working caregivers, who would be eligible for PFL. Thus, their estimates include unemployed and retired caregivers in their analysis of a policy intended for employees and effects may be underestimated. In addition, their data from HRS is skewed towards older caregivers, which means they may miss many of the caregivers most likely to benefit from PFL (i.e. adult children). Finally, they include several measures of health in their estimates of mental and physical (e.g., history of cancer, diabetes, and psychological conditions) that might create bias in their model, but they do not report unadjusted differences between the treatment and control in the post period. Thus, it is difficult to draw general conclusions about the effect of PFL on working caregivers’ health. There is a clear lack of research on family leave policies when it comes to caregiving. Some of the most important outcomes for caregivers have not yet been explored in the literature. Namely, we have no estimates of the effect of these policies on labor supply, caregiving supply, and caregiver burden. Lack of evidence on the last outcome is particularly surprising given the substantial amount research devoted to caregiver burden. As one of the only institutionalized 120 supports for balancing working and caregiving, family leave policies should certainly be systematically examined across a boarder range of outcomes. Moreover, as mentioned above, these types of programs may have differential impacts on ADRD caregivers and any estimate of their impact should isolate this group. Additional Policies and Programs: Tax Deductions and Unemployment Insurance Two other programs have the potential to influence the work and care decisions of family members of people with ADRD, but there is no empirical evidence of the relationship. For working caregivers, the Internal Revenue Service (2017) allows an income tax deduction for medical and long-term care services of a co-residing dependent for whom the taxpayer provides over 50 percent support. To qualify for this deduction, the expenses must exceed 10 percent of the taxpayer’s income, which could be a significant and prohibitive outlay for many working families. To the extent that the tax deduction promotes work over care, economic theory suggests the policy may reduce care hours or likelihood, especially among higher-wage workers with larger opportunity costs of providing care. Specifically, workers would purchase substitute care and receive the tax benefit. Whether this effect would be different for ADRD caregivers is unclear. Unemployment insurance (UI) programs in twenty-four states make exceptions to their voluntary separation exemption for workers who left their jobs for “family reasons” including caregiving; however, the individual must be looking for work (Ben-Ishai, McHugh & Ujvari, 2015). While the unemployment option is not likely to benefit people who choose to provide full- time care, UI can replace lost wages during temporary unemployment spells for some caregivers. Take-up of UI for caregiving purpose appears quite low. In the sample of states (n=10) providing administrative data for the study, Ben-Ishai et al. (2015) calculate less than 3 percent of adjudicated 121 claims and 5 percent of approved voluntary quit claims (in most years) were for caregiving. As with the tax deduction, there is no research on how these policies impact caregivers or care recipients. The effect of this type unemployment insurance on the time devoted to caring and working is theoretically ambiguous and depends on a person’s wages. To the extent that the availability of UI for voluntary quits encourages people to leave their employers who would not have otherwise done so, it would reduce labor supply and likely increase caregiving supply. On the other hand, limits to the level of wage replacement mean higher wage workers would not likely leave a job to care for their loved one. Rather, they may elect to purchase substitute (formal) care. Thus, their effect on caregivers is an empirical question. DISCUSSION AND FUTURE RESEARCH The growing importance of family caregivers in helping contain federal and state governments’ LTSS expenditures requires investment in policies supporting them. Reducing caregiver burden, promoting health, and improving work-life balance for family caregivers may delay or eliminate costly nursing home utilization and improve care recipient outcomes, as well. Policies, like paid leave, can influence work and care decisions by altering the incentives, or opportunity costs, of caring and/or working. On the other hand, supportive services can promote caregiver health and wellbeing and decrease productivity loss, increasing the time family members are available to provide care and freeing up more time to care. Thus, these policies may be equally as important as federal programs paying for LTSS but receive less government funding and less attention in the literature. Caregiving burden is especially high for people caring for loved ones with Alzheimer’s disease and dementia. Research on ADRD caregiver interventions provides evidence of the wide- 122 ranging benefits for people with dementia and their caregivers in targeted programs. Dementia care coordination programs – such as those funded by ADI-SSS or ADSSP grants – appear to improve quality of life for caregivers in addition to reducing costs associated with hospitalization and medical care. Providing resources for dementia caregivers also likely averts or delays nursing home placement. Likewise, transition care programs decrease the likelihood of hospital readmission. Thus, these types of programs may reduce overall costs to society or government providers of LTSS at a relatively low cost. For example, the government granted just under $14 million total under ADI-SSS and ADSSP in FY 2017. National-level studies are necessary to evaluate the benefits and costs of these federal caregiver support initiatives and assess whether they are sustainable in the long term on these programs. Policies like FMLA, California’s Paid Family Leave (PFL), unemployment insurance, and the federal income tax deduction apply to working caregivers, but their effect on the work and care decisions and wellbeing of caregivers is unknown. While FMLA, PFL, and the tax deduction may promote labor force attachment, they may also impact the number of hours devoted to care. The little existing research on paid family leave suggests such policies may reduce nursing home entry, but the first-order effect on labor supply and care has not been established in the literature. Moreover, research on supportive services indicates they improve caregiver burden, wellbeing, and other caregiver and care recipient outcomes, but the effect of policies aimed at working caregivers on their wellbeing is unknown. These three types of policies for working caregivers – paid family leave, tax deductions, and unemployment insurance – likely have interactive effects with effects on potential dementia caregivers working in opposing directions. For example, paid family leave and unemployment insurance could both reduce the opportunity costs of caring and increase time devoted to caregiving; however, unemployment insurance would reduce labor supply, 123 while paid family leave would generally be expected to promote labor force attachment. A tax deduction, on the other hand, would be expected to decrease time devoted to caregiving by reducing the effective cost of alternative care, especially among higher-wage workers that benefit from deductions. Any exploration of the effect of these policies will need to account for potential interactions in their effects. With the exception of the programs specifically designed for caregivers of people with ADRD, there is a general dearth of research analyzing impacts on this growing community of caregivers. The literature also lacks an assessment of the impact of these policies on the work and care decisions of potential and actual caregivers. At a time when we are poised to rely more heavily on informal caregivers to care for people with ADRD, who may also be working, it is critical to understand the impact of public policy on their decisions and wellbeing. The research on dementia caregivers and interventions is limited to RCTs – both large well- designed and smaller, less rigorous studies – and meta-analyses. These studies lack generalizability and paint an incomplete picture from a national policy perspective. Although extant research suggests caregiver interventions reduce burden and depression and paid family leave may reduce institutionalization, a large gap exists in the literature around the effect of the range of public policies on the labor supply, caregiving, and wellbeing outcomes of ADRD caregivers. This gap leaves the door open to an important body of future research systematically exploring these relationships. Well-designed studies could exploit state and local variation in the implementation of federal policies to estimate their effect on a range of outcomes. Such research requires nationally-representative data, but the appropriate data source depends on the research question. Estimating dynamic models of labor and caregiving supply requires longitudinal data (e.g. Health and Retirement Study or Survey of Income and Program Participation); however, large, cross- 124 sectional data sources (e.g. Current Population Survey or American Community Survey) are appropriate for exploring aggregate, contemporaneous responses to policy shocks. Establishing a link between public policy and labor and caregiving supply is an important first step in analyzing the costs and benefits of these policies. With the smattering of similar (possibly redundant) policies currently in place, it may be prudent to direct scarce resources to the programs most effective at promoting both productivity and caregiving. Moreover, the interaction between multiple policies is equally important to understand. For example, if policies paying family members for care decrease their labor supply, they may reduce productivity, but they may also reduce government LTSS expenditures. Which of these outcomes is more desirable and sustainable in the long-run? From a policymaker’s perspective, it is important to consider the overall costs and benefits of these programs and interactions between them. The positive effects of interventions found in the literature and the substantial (and growing) value to society of informal care for people with ADRD should provide impetus for promoting effective, cost-saving programs; however, research has not provided a clear indication of the individual and combined effects of the patchwork of policies for people with ADRD and their caregivers. CONCLUSION Caring for a person with ADRD differs from non-dementia caregiving in intensity, stress level, and duration. With the growing number of people with ADRD, our reliance on informal caregivers is expected to increase in the coming years. The current landscape of policies for caregivers of elderly people may or may not impact ADRD caregivers in the same way. While some research is beginning to elucidate the impact of supportive services on caregivers generally 125 and ADRD caregivers, in particular, there is very little research on the effect of policies like family leave (paid and unpaid), tax deductions, and other policies geared toward working caregivers. Further, the relationship between many of the supportive services and the work and care decisions has not been explored. Finally, there is a glaring gap in the policies themselves that seems to leave out some middle-income working families caring for an elderly person in the community, i.e. the middle- class sandwich generation. Low-income families can rely on Medicaid to fund care services, medical expenses, and equipment. Families above the income threshold, without enough money to afford LTSS on their own, and without long-term care insurance may face difficult decisions about working and caring for their aging loved ones. Can the working mother take unpaid time off from her job (as is granted under FMLA) to take her own mother with dementia to doctor’s appointments or the care coordination center? Will that leave her enough family leave to care for her child if he gets sick? Taking into account the unique needs of current and future generations of family caregivers is a crucial part of a sustainable system of LTSS and an important policy question facing U.S. lawmakers. Results from future research suggested here would provide critical information for evidence-based policy decisions. 126 REFERENCES AARP Public Policy Institute. (2017). Issues: Family Caregiving. Retrieved from: https://www.aarp.org/ppi/issues/caregiving/ Adelman, R. D., Tmanova, L. L., Delgado, D., Dion, S., & Lachs, M. S. (2014). Caregiver burden: a clinical review. Jama, 311(10), 1052-1060. Administration for Community Living. (2017). Programs. Retrieved from: https://www.acl.gov/programs Alzheimer's Association. (2017). 2017 Alzheimer's disease facts and figures. Alzheimer's & Dementia, 13(4), 325-373. Alzheimer's Association. (2018). 24/7 Helpline. Retrieved from: https://www.alz.org/we_can_help_24_7_helpline.asp Angrist, J. D., & Pischke, J. S. (2008). Mostly harmless econometrics: An empiricist's companion. Princeton university press. Arora, K., & Wolf, D. A. (2018). Does Paid Family Leave Reduce Nursing Home Use? The California Experience. Journal of Policy Analysis and Management, 37(1), 38-62. Bass, D. M., Clark, P. A., Looman, W. J., McCarthy, C. A., & Eckert, S. (2003). The Cleveland Alzheimer's managed care demonstration: outcomes after 12 months of implementation. The Gerontologist, 43(1), 73-85. Bass, D., Easom, L., Primetica, B., & Holloway, C. (2015). Reflections on implementing the evidence-based BRI Care Consultation with RCI in Georgia. Generations, 39(4), 49- 56.   Bass, D.M., Judge, K.S., Lynn Snow, A., Wilson, N.L., Morgan, R., Looman, W.J., McCarthy, C.A., Maslow, K., Moye, J.A., Randazzo, R., Garcia-Maldonado, M., Elbein, R., Odenheimer, G., & Kunik, M.E. (2013). Caregiver outcomes of partners in dementia care: Effect of a care coordination program for veterans with dementia and their family members and friends. Journal of the American Geriatrics Society, 61(8), 1377-1386.   Bass, D.M., Judge, K.S., Maslow, K., Wilson, N.L., Morgan, R.O., McCarthy, C. A., Looman, W.J., Snow, A.L., & Kunik, M.E. (2015). Impact of the care coordination program “Partners in Dementia Care” on veterans’ hospital admissions and emergency department visits. Alzheimer’s and Dementia: Translational Research and Clinical Interventions, 1(1), 13-22. Baum, C.L. & Ruhm, C.J. (2016). The Effects of Paid Family Leave in California on Labor Market Outcomes. Journal of Policy Analysis and Management, 35(2), 333-356. 127 Beach, S. R., Schulz, R., Yee, J. L., & Jackson, S. (2000). Negative and positive health effects of caring for a disabled spouse: Longitudinal findings from the caregiver health effects study. Psychology and aging, 15(2), 259. Becker, G.S. (1981, 1991). A Treatise on the Family. Harvard University Press. Bedard, K., & Rossin-Slater, M. (2016). The Economic and Social Impacts of Paid Family Leave in California: Report for the California Employment Development Department. Retrieved from: https://www.dol.gov/wb/media/CA_Final_Report_2015_Final_Report.pdf Belle, S. H., Burgio, L., Burns, R., Coon, D., Czaja, S. J., Gallagher-Thompson, D., ... & Martindale-Adams, J. (2006). Enhancing the Quality of Life of Dementia Caregivers from Different Ethnic or Racial Groups A Randomized, Controlled Trial. Annals of internal medicine, 145(10), 727-738. Ben-Ishai, L., McHugh, R., and Ujvari, K. (2015). Access to Unemployment Benefits for Family Caregivers: An Analysis of State Rules and Practices. AARP Public Policy Institute. Retrieved from: https://www.aarp.org/ppi/info-2015/access-to-unemployment-insurance- benefits-for-family-caregivers.html Brown, J. R., & Finkelstein, A. (2007). Why is the market for long-term care insurance so small?. Journal of Public Economics, 91(10), 1967-1991. Brown, J. R., & Finkelstein, A. (2011). Insuring long-term care in the United States. The Journal of Economic Perspectives, 25(4), 119-141. Callahan, C. M., Arling, G., Tu, W., Rosenman, M. B., Counsell, S. R., Stump, T. E., & Hendrie, H. C. (2012). Transitions in care for older adults with and without dementia. Journal of the American Geriatrics Society, 60(5), 813-820. Caputo, J., Pavalko, E. K., & Hardy, M. A. (2016). The Long-Term Effects of Caregiving on Women's Health and Mortality. Journal of Marriage and Family, 78(5), 1382-1398. Centers for Disease Control and Prevention. (2016). Caregiving for Person with Alzheimer’s Disease or a related Dementia. Retrieved from: https://www.cdc.gov/aging/caregiving/alzheimer.htm Centers for Disease Control and Prevention. (2018). National Vital Statistics System: Birth Data. Retrieved from: https://www.cdc.gov/nchs/nvss/births.htm Chen, Y. M., Hedrick, S. C., & Young, H. M. (2010). A pilot evaluation of the Family Caregiver Support Program. Evaluation and program planning, 33(2), 113-119. Coleman, E. A., Parry, C., Chalmers, S., & Min, S. J. (2006). The care transitions intervention: results of a randomized controlled trial. Archives of internal medicine, 166(17), 1822- 1828. 128 Corbett, A., Stevens, J., Aarsland, D., Day, S., Moniz-Cook, E., Woods, R., ... & Ballard, C. (2012). Systematic review of services providing information and/or advice to people with dementia and/or their caregivers. International Journal of Geriatric Psychiatry, 27(6), 628-636. Daiello, L. A., Gardner, R., Epstein-Lubow, G., Butterfield, K., & Gravenstein, S. (2014). Association of dementia with early rehospitalization among Medicare beneficiaries. Archives of gerontology and geriatrics, 59(1), 162-168. Doty, P., & Spillman, B. (2015). Help for family caregivers available from government programs and policies. Family caregiving in the new normal, 153-190.   Easom, L.R., Alston, G., & Coleman, R. (2013) A rural community translation of a dementia caregiving intervention. Online Journal of Rural Nursing and Health Care, 13(1), 66-91.   Elliott, A. F., Burgio, L. D., & DeCoster, J. (2010). Enhancing caregiver health: findings from the resources for enhancing Alzheimer's caregiver health II intervention. Journal of the American Geriatrics Society, 58(1), 30-37. Elvish, R., Lever, S. J., Johnstone, J., Cawley, R., & Keady, J. (2013). Psychological interventions for carers of people with dementia: A systematic review of quantitative and qualitative evidence. Counselling and Psychotherapy Research, 13(2), 106-125. Ettner, S. L. (1995). The impact of “parent care” on female labor supply decisions. Demography, 32(1), 63-80. Ettner, S. L. (1996). The opportunity costs of elder care. Journal of Human Resources, 189-205. Feinberg, L. F., & Newman, S. L. (2006). Preliminary experiences of the states in implementing the National Family Caregiver Support Program: A 50-state study. Journal of aging & social policy, 18(3-4), 95-113. Friedman, E. M., Shih, R. A., Langa, K. M., & Hurd, M. D. (2015). US prevalence and predictors of informal caregiving for dementia. Health Affairs, 34(10), 1637-1641. Gallagher-Thompson, D., & Coon, D. W. (2007). Evidence-based psychological treatments for distress in family caregivers of older adults. Psychology and aging, 22(1), 37. Gans, D., Lowenstein, A., Katz, R., & Zissimopoulos, J. (2013). Is There a Trade-Off Between Caring for Children and Caring for Parents?. Journal of Comparative Family Studies, 455-471. Genworth Financial, Inc. (2017). Compare Long Term Care Costs Across the United States. Retrieved from: https://www.genworth.com/about-us/industry-expertise/cost-of-care.html 129 Gimm, G., & Yang, Y. T. (2016). The effect of paid leave laws on family caregivers for the elderly. Ageing International, 41(2), 214-226. Hagen, S. A. (2013). Rising demand for long-term services and supports for elderly people. Congressional Budget Office. Heitmueller, A. (2007). The chicken or the egg?: Endogeneity in labour market participation of informal carers in England. Journal of health economics, 26(3), 536-559. Jensen, M., Agbata, I. N., Canavan, M., & McCarthy, G. (2015). Effectiveness of educational interventions for informal caregivers of individuals with dementia residing in the community: systematic review and meta-analysis of randomised controlled trials. International journal of geriatric psychiatry, 30(2), 130-143. Johnson, R. W., & Lo Sasso, A. T. (2000). The trade-off between hours of paid employment and time assistance to elderly parents at midlife. Kasper, J. D., Freedman, V. A., Spillman, B. C., & Wolff, J. L. (2015). The disproportionate impact of dementia on family and unpaid caregiving to older adults. Health Affairs, 34(10), 1642-1649. Klerman, J. A., Daley, K., & Pozniak, A. (2012). Family and medical leave in 2012: Technical report. Cambridge, MA: Abt Associates Inc. LaMantia, M.A., Alder, C.A., Callahan, C.M., Gao, S., French, D.D., Austrom, M.G., Boustany, K., Livin, L., Bynagari, B., & Boustani, M.A. (2015). The aging brain care medical home: Preliminary data. Journal of the American Geriatrics Society, 63(6), 1209- 1213.   Lee, H., & Cameron, M. H. (2004). Respite care for people with dementia and their carers. The Cochrane Library. Lilly, M. B., Laporte, A., & Coyte, P. C. (2007). Labor market work and home care's unpaid caregivers: a systematic review of labor force participation rates, predictors of labor market withdrawal, and hours of work. The Milbank Quarterly, 85(4), 641-690. Mason, A., Weatherly, H., Spilsbury, K., Golder, S., Arksey, H., Adamson, J., & Drummond, M. (2007). The Effectiveness and Cost-Effectiveness of Respite for Caregivers of Frail Older People. Journal of the American Geriatrics Society, 55(2), 290-299. Medicare. (2018). Home health services. Retrieved from: https://www.medicare.gov/coverage/home-health-services.html National Academies of Sciences, Engineering, and Medicine. (2016). Families caring for an aging America. Washington, DC: National Academies Press. 130 National Association of Area Agencies on Aging (2016). About n4a. Retrieved from: https://www.n4a.org/ Ory, M. G., Hoffman III, R. R., Yee, J. L., Tennstedt, S., & Schulz, R. (1999). Prevalence and impact of caregiving: A detailed comparison between dementia and nondementia caregivers. The Gerontologist, 39(2), 177-186. O'Shaughnessy, C. (2014). National spending for long-term services and supports (LTSS), 2012. Pinquart, M., & Sörensen, S. (2006). Helping caregivers of persons with dementia: which interventions work and how large are their effects?. International Psychogeriatrics, 18(4), 577-595. Pinquart, M., & Sörensen, S. (2007). Correlates of physical health of informal caregivers: a meta-analysis. The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 62(2), P126-P137. Rossin-Slater, M., Ruhm, C. J., & Waldfogel, J. (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(2), 224-245. Roth, D. L., Fredman, L., & Haley, W. E. (2015). Informal caregiving and its impact on health: A reappraisal from population-based studies. The Gerontologist, 55(2), 309-319. Samus, Q.M., Johnston, D., Black, B.S., Hess, E., Lyman, C., Vavilikolanu, A., Pollutra, J., Leoutsakos, J.-M., Gitlin, L.N., Rabins, P.V., & Lyketsos, C.G., (2014) A multidimensional home- based care coordination intervention for elders with memory disorders: The Maximizing Independence at Home (MIND) Pilot Randomized Trial. American Journal of Geriatric Psychiatry, 22(4): 398- 414. Schoenmakers, B., Buntinx, F., & Delepeleire, J. (2010). Factors determining the impact of care- giving on caregivers of elderly patients with dementia. A systematic literature review. Maturitas, 66(2), 191-200. Schmitz, H., & Westphal, M. (2017). Informal care and long-term labor market outcomes. Journal of health economics, 56, 1-18. Schulz, R., & Beach, S. R. (1999). Caregiving as a risk factor for mortality: the Caregiver Health Effects Study. Jama, 282(23), 2215-2219. Schulz, R., & Martire, L. M. (2004). Family caregiving of persons with dementia: prevalence, health effects, and support strategies. The American journal of geriatric psychiatry, 12(3), 240-249. 131 Schulz, R., O'Brien, A. T., Bookwala, J., & Fleissner, K. (1995). Psychiatric and physical morbidity effects of dementia caregiving: prevalence, correlates, and causes. The gerontologist, 35(6), 771-791. Schulz, R., & Sherwood, P. R. (2008). Physical and mental health effects of family caregiving. Journal of Social Work Education, 44(sup3), 105-113. Shaw, W. S., Patterson, T. L., Ziegler, M. G., Dimsdale, J. E., Semple, S. J., & Grant, I. (1999). Accelerated risk of hypertensive blood pressure recordings among Alzheimer caregivers. Journal of psychosomatic research, 46(3), 215-227. Spillman, B. C., & Long, S. K. (2009). Does high caregiver stress predict nursing home entry?. INQUIRY: The Journal of Health Care Organization, Provision, and Financing, 46(2), 140-161. Spillman, B. C., & Pezzin, L. E. (2000). Potential and active family caregivers: Changing networks and the ‘sandwich generation’. The Milbank Quarterly, 78(3), 347-374. State of California, Employment Development Department (2017). Paid Family Leave. Retrieved from: http://www.edd.ca.gov/Disability/Paid_Family_Leave.htm U.S. Bureau of Labor Statistics, Civilian Labor Force Participation Rate: Women [LNS11300002], retrieved from FRED, Federal Reserve Bank of St. Louis; https://fred.stlouisfed.org/series/LNS11300002. U.S. Centers for Medicare and Medicaid Services. (2017). Home & Community Based Services. Retrieved from: https://www.medicaid.gov/medicaid/hcbs/index.html U.S. Centers for Medicare and Medicaid Services. (2018). Health care coverage rights and protections. Retrieved from: https://www.healthcare.gov/health-care-law-protections/ U.S. Department of Labor, Wage and Hour Division (2012). Fact Sheet #28: The Family and Medical Leave Act. Retrieved from: https://www.dol.gov/whd/regs/compliance/whdfs28.pdf U.S. Department of the Treasury. Internal Revenue Service. (2017) Publication 502: Medical and Dental Expenses (Cat. No. 15002Q). Retrieved from: https://www.irs.gov/publications/p502 Van Houtven, C. H., Coe, N. B., & Skira, M. M. (2013). The effect of informal care on work and wages. Journal of Health Economics, 32(1), 240-252. Vandepitte, S., Van Den Noortgate, N., Putman, K., Verhaeghe, S., Verdonck, C., & Annemans, L. (2016). Effectiveness of respite care in supporting informal caregivers of persons with dementia: a systematic review. International journal of geriatric psychiatry, 31(12), 1277-1288. 132 Vitaliano, P. P., Zhang, J., & Scanlan, J. M. (2003). Is caregiving hazardous to one's physical health? A meta-analysis. Psychological bulletin, 129(6), 946. Westat, Inc. (2017). Finding how best to support caregivers. Retrieved from: https://www.westat.com/projects/finding-how-best-support-family-caregivers Wiener, J.M., Gould, E., Shuman, S.B., Kaur, R., Ignaczak, M., & Maslow, K. (2016). Examining models of dementia care: Final report. Report prepared for the Office of the Assistant Secretary for Planning and Evaluation. Washington, DC: RTI International. Wolff, J. L., Spillman, B. C., Freedman, V. A., & Kasper, J. D. (2016). A national profile of family and unpaid caregivers who assist older adults with health care activities. JAMA internal medicine, 176(3), 372-379. Zissimopoulos, J., Crimmins, E., & St Clair, P. (2015, January). The value of delaying Alzheimer’s disease onset. In Forum for Health Economics and Policy (Vol. 18, No. 1, pp. 25-39). 133 CONCLUSION This series of studies endeavored to expand our understanding of the role of public policy in helping people balance work and family life, by estimating the effects of these policies on their beneficiaries, both intended and unintended. Modern U.S. families face tradeoffs between work, childbearing and childrearing, and elder care. Consequences of being unable to devote substantial time to all spheres include delayed or forgone childbearing, reduced labor supply, forgone promotions, premature institutionalization of the elderly, and declining caregiver health and wellbeing, among many others. At the very least, the dual responsibilities of working caregivers can lead to difficult decisions for individuals weighing earning a living against caring for loved ones. From a societal perspective, this tradeoff can result in a smaller pool of current and future workers and/or informal caregivers, leading to lower productivity and higher expenditures on long- term services and support. In contrast to other industrialized countries, the United States has few national policies promoting balance among working parents and caregivers, despite mounting evidence establishing the benefits of such programs. For example, while the average number of weeks of paid leave for new mothers in Organisation for Economic Cooperation and Development countries was about 54 weeks in 2016, U.S. workers are guaranteed only 12 weeks of leave with no wage replacement, through the Family and Medical Leave Act (OECD, 2016). Yet, eligibility restrictions leave about 40 percent of workers in the U.S. without access to even job-protected time off to care for a newborn or sick loved one. This disparity in access exacerbates differences between groups from different socioeconomic backgrounds. While higher-wage earners may be able to afford to take unpaid time off, or have access to paid leave through their employers, lower-wage workers are less likely to be eligible for FMLA and may be less likely to take leave even if they are eligible. On 134 the other end of life cycle, United States’ policies provide services supporting family caregivers that are comparable with other industrialized nations (with the exception of paid leave at the national level). For example, the U.S. joins several other OECD countries in funding training and education, respite care, and counseling for caregivers; however, there is significant geographic disparity within the U.S. in access to these services and our understanding of their overall effectiveness is very limited. The present studies considered the ability of public policies to intervene in the decisions people make about work and care. The first two studies explored the effects of an individual state- level policy, California’s Paid Family Leave (PFL) program. While previous work identified positive effects of the policy on the labor supply of new mothers and fathers, the impact on childbearing and caregiving was not established in the literature. These two chapters begin to fill that gap. The first chapter found that PFL increased birth likelihood among women ages 25 to 40, with larger effects for women working full-time; subsequent analyses suggested the increase in fertility was likely driven by reduced childbearing postponement. The second study estimated the effect of PFL on the labor supply of a group of beneficiaries that are especially vulnerable to reduced employment, parents of children with disabilities. Results reveal access to PFL increased labor force participation among mothers and fathers of children with disabilities. The final study explored policies providing support for caregivers of people with Alzheimer’s disease and related dementias (ADRD). In a first attempt to synthesize research on the patchwork of policies and related inventions and their impact on caregiver outcomes, this study revealed a lack of research on overall costs and benefits of the national-level policies and their effect on the wellbeing and decisions of ADRD caregivers. While smaller-scale interventions suggest public policies may improve some care recipient and caregiver outcomes, such as premature nursing home entry and 135 caregiver burden, no nationally-representative studies explore the effect of larger-scale initiatives, the interaction between them, and their benefits and costs to society. Together, these studies provide insight into the role policies can play in the lives of working families in the United States; however, the results point to gaps in both the literature and the policies themselves. The next phase of research on paid family leave should explore policy effects on fertility timing of working mothers and working hours of parents of children with disabilities. The impact of PFL on the labor supply and caregiving hours of ADRD caregivers, and the elderly community more generally, would also be fruitful avenues for future studies. In fact, as beneficiaries of several programs (e.g. National Family Caregiver Support Program) and stakeholders in others (e.g. HCBS waiver programs), it is surprising to find that the impact of national policies on ADRD caregivers has largely been ignored in the literature. Population-level studies using nationally-representative data and econometric models accounting for endogeneity in work, care, and health, that estimate the effects of individual policies are needed to provide insight into benefits of these national-level policies. In addition, a broad analysis of the costs and benefits of the multiple policies would help policymakers assess where to spend limited government funds. While research demonstrates benefits of many of the policies discussed here, there is significant geographic disparity in the availability of nearly all the programs. Currently, only five states (California, New Jersey, New York, Rhode Island, and Washington) have active paid family leave programs. Debates over national paid family leave have been waged for years, with only modest movement towards consensus. Policies for ADRD caregivers, such as federal grants for dementia care programs and Lifespan Respite Care, are distributed unevenly across geographic regions. Thus, gaps remain in the availability of these policies that could benefit workers, 136 caregivers, children, and elderly care recipients. As arbiters of U.S. workers’ and caregivers’ wellbeing, as well as economic stability, federal and state policymakers must carefully consider evidence, such as the findings presented here, on how policies can influence decisions people make about working, having children, and providing care to their loved ones, now and in future generations.

Abstract (if available)

Abstract Modern families in the United States face challenges not encountered in previous generations. Work cultures and family responsibilities rooted in breadwinner and caregiver roles from the past exact a toll on the growing number of families headed by single parents and dual earners. Rather than devoting time to only one domain, today’s workers often divide their time between working and caring for their children or parents. Public policies in the U.S. evolved over time to address some of these challenges. Research in several disciplines evaluates the impact of policies addressing work-life balance on the outcomes of some beneficiaries, but significant gaps remain in our understanding of their effectiveness throughout different stages in the life cycle and there is little systematic evidence of the impact of the patchwork of policies supporting family caregivers. This three-chapter dissertation addresses many of the gaps in the literature and suggests avenues for future research. The first two chapters analyze the impact of California’s paid family leave (PFL) policy on fertility outcomes and the labor supply of parents of children with disabilities, using difference-in-differences estimation to compare outcomes in California before and after PFL with a control group. I find the likelihood a woman ages 25 to 40 had a child in the last year increased by about 6.0 percent for all women in California, relative to similar states with no paid leave, and 7.6 percent for women working fulltime. Among parents of children with disabilities, my results suggest access to PFL increased the labor force participation of their mothers and fathers, by 9.4 and 4.3 percent, respectively, when compared to parents in California of children with no disability and parents in the rest of the United States. The final chapter explores policies supporting family caregivers of people with Alzheimer’s disease and related dementias (ADRD). Through a review of the policies and programs providing government funding and supportive services for these caregivers and a critical analysis of the literature related to their effectiveness, I find a general lack of policy-related research on the federal initiatives. While results from smaller targeted interventions generally indicate national-level policies should positively affect ADRD caregivers, the overall impact of these policies is unknown. This final chapter concludes by identifying several avenues for future research, such as estimating the benefits and costs of each federal policy, and their interactions, to determine which are the most effective and cost-reducing programs. ❧ Together, these studies contribute to the literature on work and care, by highlighting the broad impact of policies on modern families across the life cycle. While virtually everyone will care for a loved one or be cared for some time in their lives, research on policies affecting work-life balance offers little evidence of the benefits beyond new parents. This research estimates the effect of California’s PFL on new fertility outcomes and a group of long-term caregivers, and explores the availability and effectiveness of policies for family caregivers of people with ADRD. The findings presented here demonstrate the potential of these policies to positively impact caregivers and care recipients throughout the life cycle

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Uncovering hidden figures: disparities in dementia burden in an aging America

PDF

Research to develop a manual for parents, caregivers, and volunteers to teach children from birth to five years of age with intellectual disability living in rural communities in the Dominican Re...

PDF

Impacts of caregiving on wellbeing among older adults and their spousal caregivers in the United States

PDF

More knowledge, better plans: a study of heterogeneity in dementia prevalence and mid-life cognitive changes

PDF

Four essays on how policy, the labor market, and age relate to subjective well-being

PDF

The dollars and sense of caregiving: exploring the financial financial realities of providing informal care to an older adult

PDF

The impact of demographic shifts on automobile travel in the United States: three empirical essays

PDF

Changing workplace and inequality in work-family policies

PDF

Essays on work, retirement, and fostering longer working lives

PDF

Adoption and implementation of innovative diagnostic tools for Alzheimer's Disease: challenges and barriers in primary care

PDF

A targeted culturally-informed approach for caregiver stress among Vietnamese caregivers of family members [capstone paper]

PDF

A better method for measuring housing affordability and the role that affordability played in the mobility outcomes of Latino-immigrants following the Great Recession

PDF

Adoption of virtual healthcare, self-sufficient wages and paid neighbors concept will ensure optimal living for vulnerable people and their paid caregivers

PDF

Childhood cancer survivorship: parental factors associated with survivor's follow-up care behavior and mental health

PDF

Redlining, neighborhood change, and individual outcomes: an exploration of how space shapes the landscape of inequality from past to present

PDF

The role of dyadic and triadic factors on psychosocial wellbeing and healthcare interactions among childhood cancer survivors, parents, and medical providers

PDF

Family caregiver well-being: generational differences and coping strategies

PDF

The effects of wisdom-related personality traits on caregivers’ health: an application of the resilience model

PDF

The functions of the middleman: how intermediary nonprofit organizations support the sector and society

PDF

Marital matching in West Africa: an examination of interethnic and interreligious first marriages in Benin

Asset Metadata

Creator Thunell, Johanna A. (author)

Core Title The role of public policy in the decisions of parents and caregivers: an examination of work, fertility, and informal caregiving

School School of Policy, Planning and Development

Degree Doctor of Philosophy

Degree Program Public Policy and Management

Publication Date 04/16/2018

Defense Date 02/06/2018

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Children with disabilities,dementia caregiving,fertility,informal caregiving,OAI-PMH Harvest,paid family leave,Public Policy,working parents

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Zissimopoulos, Julie (committee chair), Hook, Jennifer (committee member), Myers, Dowell (committee member), Painter, Gary (committee member)

Creator Email jthunell@usc.edu,thunell.johanna@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-492514

Unique identifier UC11268312

Identifier etd-ThunellJoh-6254.pdf (filename),usctheses-c40-492514 (legacy record id)

Legacy Identifier etd-ThunellJoh-6254.pdf

Dmrecord 492514

Document Type Dissertation

Format application/pdf (imt)

Rights Thunell, Johanna A.

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA