Married Women s Labor Force Participation and The Role of Human Capital Evidence from the United States

Transcription

1 C L M. E C O N O M Í A Nº 17 MUJER Y ECONOMÍA Married Women s Labor Force Participation and The Role of Human Capital Evidence from the United States Joseph S. Falzone Peirce College Philadelphia, Pennsylvania USA Abstract The single most important trend in labor market activity during the last half century has been the increase in the labor force participation of married women. Married women differ from both men and unmarried women in that their decision to work inside or outside the home depends upon the earnings of other family members. The overall rise in the labor force participation rate of married women can be explained by both changes in characteristics, especially human capital investment, and by changes in behavior The objective of this paper is to examine the decision on the part of prime age married women to participate in the labor force. Employing a probit model, I decompose the effects of these distinct changes in order to analyze the relative importance of each in the labor force participation decision of married women. Key Words: married women, labor market, labor force, probit model. JEL Classification: J21, J22 1

2 2

3 1. Introduction The single most important trend in labor market activity during the last half century has been the increase in the labor force participation of married women. While the growth in women's labor force participation rates has its origins in the late 19th century 1, by 1970 married women accounted for almost one half of the total increase in labor force participation growing from 30.5 percent of the total labor force in 1960 to 40.8 percent. [Feilds:1976] Throughout the 1970s we witness an increase in the number of weeks worked by married women and a decline in the percentage of married women supplying zero hours of labor to the market. This was true for both married women with children and those without children. Monthly data from the Bureau of Labor Statistics show a monotonic rise in the labor force participation rates for all women over the age of 20 from 1948 until the late 1990s. [Figure 1] The focus on married women is important because their participation in the labor force provides an important means of economic growth especially given the decline in male labor force participation rates since the mid-20 th century. 2 In his pioneering study, Mincer [1960] placed married women's labor supply within the context of the family, specifically highlighting family income. Married women differ from both men and unmarried women in that their decision to work inside or outside the home depends upon the earnings of other family members. Unlike men, married women's labor force participation ought not to be viewed solely in terms of the demand for leisure; rather, it should be considered within the context of the demand for household production. Thus increases in the earnings of one family member, the husband, can result in more hours of labor supplied by him since the opportunity cost of leisure is higher. At the same time this can result in greater household production for the wife. 3 Smith and Ward [1985] consider factors that led to the vast increases in female labor force participation in the 20 th century looking at the simultaneous effects of wages on fertility. The authors find increasing real wages of women explain about 60 percent of the growth in female labor force participation rates and that 50 percent of this was the effect that higher real wages played on declining fertility. Pencavel [1998] investigates why the labor force behavior of married women has become more like that of men and unmarried women. The author concludes that market work has become more hospitable and household activity less attractive to married women and that their greater labor force participation cannot be explained solely by higher real wages. Hotchkiss [2006] considers the changes in female labor force participation which have occurred over the past 30 years. Her findings for all prime age women indicate that the decline in 3

4 female labor force participation between 2000 and 2005 was not due solely to macroeconomic or demographic changes; rather it was due to changes in behavior. The overall rise in the labor force participation rate of women in general and married women in particular can be explained by both changes in characteristics, especially human capital investment, and by changes in behavior The objective of this paper is to examine the decision on the part of prime age married women to participate in the labor force. Employing a probit model, I decompose the effects of these distinct changes in order to analyze the relative importance of each in the labor force participation decision of married women. Part II presents a model of married women s labor force participation, Part III gives an overview of the data, Part IV discusses the results, and Part V offers a summary and conclusion 2. Model Given that the individual is endowed with 24 hours per day, assume that she may engage in only two mutually exclusive activities, labor and leisure. "Labor" is defined as working in the market for a wage while "leisure" is defined as any other non-market activity. Thus, she will choose that combination of activities (labor/leisure tradeoff) that maximizes utility from some consumption bundle of goods and leisure. Hours of labor are, of course, constrained by physical endurance with maximum hours worked approaching 65 per week. We find however a corner solution, particularly among married women, of 24 hours of leisure per day. The decision to participate in the labor market can be viewed as the opportunity cost of leisure; that is the individual will work only if the utility of the an hour's worth of labor (or the wage rate) exceeds the utility of an additional hour of leisure. Thus, the decision to work can be modeled as [1] W m i - W r i H=0 > 0 LFP = 1 W m i - W r i H=0 0 LFP = 0 where W m i is the market wage for individual i or the value of time for the working individual given at the margin and is a function of her human capital investment. W r i H=0 is the marginal value of time for non-worker i or her reservation wage. This is the highest wage at which the individual will not work. Labor Force Participation (LFP) is thus a binary choice variable equal to 1 if the individual participates in the labor market and equal to zero otherwise. 4

5 The decision to participate in the labor market can be expressed using a probit model. We assume that the difference between individual i's market and reservation wage can be modeled as a linear function of both observable characteristics and an unobservable random disturbance term in [1] above. The model may be estimated by [2] I * t = W m i - W r i H=0 = β 0 + β ' 1 X m,i + β ' X r,i + ε i where X m,i is a vector of observable characteristics which determine the individual's market wage. Since investment in human capital represents the most important determinant of the market wage, years of education is included in X m,i. Schooling not only increases the expected market wage but is also the source of jobs with higher social status. We can view additional years of schooling as signifying both the individual s natural aptitude as well as her taste for market work over household production. Age is included here as a proxy for labor market experience. Also included is the individual's health status. Better health is hypothesized to lower her reservation wage since the opportunity cost of non-participation falls while better health increases her market wage associated with the greater marginal productivity. Since labor market conditions affect the labor/leisure decision as well as the market wage, the unemployment rate in the individual's state of residence is included to capture the macroeconomic effects of labor market conditions. Similarly I hypothesize that urban versus rural areas of residence will affect wages as well as job opportunities and have included a dummy variable = 1 if the individual resides in an area with a population density greater than 250 million. The vector X r,i determines the individual's marginal value of time outside of the labor market. This includes the age of a woman's youngest child with the probability of married women participation a rising function of her youngest child's age. 4 I also include the natural log of her husband's wage since married women's decision to participate in the labor market is a function of family income and her participation decision is likely to be inversely related to her husband's earnings. Her husband's health status is included to capture the effects of poor health on her decision to participate. A dummy variable for race = 1 if the woman is white is included to capture the effects associated with labor market discrimination or social or cultural differences associated with race. The random error term ε i is assumed to be normally distributed with mean zero and variance = 1. The parameter coefficients in [2] are estimated using a maximum likelihood probit model. 5

6 3. Data Data are extracted from Waves 29, 31, 32, and 35 of The Panel Study of Income Dynamics corresponding to years 1991, 1996, 2001, and PSID is a longitudinal survey of a representative sample of U.S. individuals and their families. The survey has been ongoing since The reference period for the PSID for each wave is the preceding year. The sample used here consists of more than 18,850 married women who were between the ages of 25 and 55 years (inclusive) during the sampling year. 5 [Table 1] Labor force participation is defined here as working, seeking employment, or temporarily laid off. Within the sample labor force participation averaged percent over the sample years. The remaining 25 percent were retired, permanently disabled, keeping house, or student. Turning first to observable characteristics which determine the woman's market wage, average age (a proxy or labor market experience) increased from a low of years in 1991 to years in 2007, an increase of 2 percent. Similarly we see a rise in average years of education over the period rising from in 1991 to in 2007, an increase of more than 8 percent. Health status remained in the range of 2.2 and 2.3 with "2.0" denoting Very Good health and "3.0" Good Health. Average state unemployment rates peaked at 5.78 percent in 1991, falling through 1996 and 2001 then rising to 4.73 percent in I hypothesize that variables affecting a married woman's marginal value of time will also affect her labor force participation. These variables include the age of her youngest child with labor force participation expected to vary directly with youngest child's age. This variable rises from a low of 6.6 years in 1991 to 11 years in Husband's wage is expected to vary inversely with wives' labor force participation. The log of husband's wage remains within the range of 2.42 to 2.86 during the 4 sample years. Husband's health status remained fairly constant as well within the range of 2.22 to A dummy variable denoting race (white) remained between 64 and 71 percent. 4. Results Probit coefficient estimates and marginal effects for married women aged 25 through 55 are presented in Table 2. Turning first to the vector of observable characteristics which determine the market wage, the estimated coefficient for Age (a proxy for labor market experience) is 6

7 significant during all sample years but negative. As a proxy for labor market experience it was hypothesized that age would be positive. The age variable may be capturing the withdrawal from the labor force of married women in the upper tail of the age distribution. The negative coefficient may reflect married women's retirement decisions as a function of the retirement decisions of their older husbands. A dummy variable indicating residence in an area with a population of 250,000 or more likewise displays different signs and levels of significance. The estimated coefficient on married women s health status is negative and significant for the years 1991, 2001, and As hypothesized poor health is likely to decrease the probability that a married women participates in the labor market. The effect of the unemployment rate within the state of residence is significant and negative only in 1996 and in Turning next to the vector determining the marginal value of time outside of the labor market, the coefficient associated with the age of her youngest child is positive and highly significant during all four years. This implies that as the age of her youngest child rises her reservation wage falls and the probability of a married woman participating in the labor market rises. Since her labor market decision is made within the context of family the natural log of husband s average wage is included to capture the effect of family income. The estimated coefficient on this variable is negative and significant during all four years suggesting that rising family income increases the reservation wage of the wife thereby reducing the probability she participates in the labor market. The effect of husband s health is, as hypothesized, positive, suggesting poor health increases the probability that his wife works but significant only for 2001 and A dummy variable denoting white race is negative and significant for three of the four sample years suggesting differential socio-cultural factors decrease the probability of labor force participation of white married women. Hotchkiss [2006] discusses changes associated with labor force participation which arise from three sources. One such source is a change in a married woman s characteristics. These are changes in variables. Thus for example, a greater investment in years of education would tend to raise her market wage or an increase in the age of her youngest child would lower her reservation wage thereby raising the probability that she participates in the labor market. Another source of the change in labor force participation arises from changes in behavior; that is, the means by which characteristics work their way through labor market participation decisions. These behavioral changes are captured by the estimated parameter coefficients associated with a given set of characteristics. More specifically, estimated parameter coefficients capture the marginal utility 7

8 associated with a given characteristic with larger parameter coefficients indicating a greater propensity to participate in the labor market. Parameter coefficients may therefore be viewed as a means through which characteristics are translated into behaviors. 6 The final source of change in labor force participation is the change in unobservable factors such as a change in tastes for labor versus leisure which are captured by the intercept term. One such unobservable factor important in married women's employment decisions is the tastes for home-oriented work and predilection for "mothering activities". 7 Since the focus in this paper is on the role of human capital on a married woman's decision to participate in the labor market we note first that between 1991 and 2007 observed probability of labor force participation increased by 7 percent. At the same time, years of education increased by 8.3 percent. Thus, ceteris paribus greater investment in human capital would contribute to an increase in labor force participation. We see too that the estimated coefficient on a married woman s years of education; that is her responsiveness to education is positive and significant at 1 percent or less for all sample years. This coefficient rose between 1991 and 1996, fell in 2001 and increased to its highest level in Consequently both a rise in years of education and a greater responsiveness to human capital investment contributed to an increase in both observed and predicted labor force participation of prime age married women. [Table 3] Figure 2 plots the "Predicted Probability" based on probit estimates evaluated at the mean of their covariates. "Simulated Probability" plots the probability of labor force participation based on women's behaviors over the sample years (the estimated β coefficients) while holding characteristics (the X vectors) constant for each sample year. The vertical distance between Predicted and Simulated participation rates indicates the differences in labor force participation due to changes in characteristics (for example years of education) holding behaviors constant. Thus, for the period 1991 to 2007 both observed and predicted labor force participation rates increased. Holding behavior constant at 1991 levels (the Beta91 in Fig. 2), we find that labor force participation rates increased more than 5.5 percent during the same period. Holding behavior constant at 2007 levels (the Beta07 in Fig. 2) labor force participation rates increased by about 4.0 percent. Yet holding characteristics constant at 1991 levels we find that participation rates increased from to percent, an almost 15 percent increase. Holding characteristics constant at 2007 levels we find participation rates increased by more than 13 percent. While both changes in behaviors and changes in characteristics reinforced changing participation rates, it was changes in behavior (the response to characteristic) of married women that had the greater effect 8

9 on the rise in married women's labor force participation rates from 1991 to Recall that estimated parameter coefficients capture the marginal utility associated with a given characteristic. Focusing on the marginal effects of years of education we see that in 1991 an additional year of education increased the probability of a married women's participation in the labor market by while in 2007 that change was During that same time, years of education increased by more than 8 percent. Thus, while the response to human capital remained relatively constant, perhaps reflecting diminishing returns, the behavioral response was enhanced by an increase in years of education. 5. Summary and Conclusions The labor force participation rate is an important indication of labor market conditions providing significant insight into evaluating an economy's growth potential. During the past half century the most salient feature of labor market trends has been the entry of married women into the labor market. BLS data since 1948 chronicle this rise showing a leveling off only in the late 1990s at around 60 percent. The changes reflect the interaction of both characteristics of married women as well as their behavioral response to those characteristics, especially human capital investment. The causes for changes in both characteristics and behavior are complex and causation may run in both directions. Nonetheless, the results reported here indicate that years of education and the age of her youngest child consistently exert the strongest positive influence on a married women's decision to enter the labor market while her husband s earnings exert the strongest negative influence on that decision. An analysis of the decomposition of characteristics and behaviors; that is, the difference between Predicted and Simulated participation rates reveals that changes in behaviors fueled the major portion of changes in married women's labor force participation during From a policy perspective the results reported here indicate that further gains in the labor force participation rates of married women appear to arise from inter-generational changes in behaviors rather than in characteristics, most especially in years of schooling, the number and ages of children, and husband's earnings. We take it as given that individuals respond to economic incentives. That being said we must acknowledge the presence and influence of unobservable factors in the decision criteria and the family context within which those decisions are made. The intrinsic randomness of decision making captured by the error term may not be "well behaved" as 9

10 postulated in our model. There are of course elements in human decision making which cannot be explained with our model. More to the point, the impressive and vast increases in the labor force participation rates of married women we witness over the last century may have reached a natural upper limit. Changes in technology, both in the workplace with diminished emphasis on physical prowess and in the home, with the ability of women to plan the number and spacing of children have greatly changed the decision matrix for work outside of the home. This, coupled with greater opportunities in the acquisition of human capital and the legal and social mandates associated with women in the workplace may too have reached an upper limit in their ability to draw more married women into the workplace. As women's labor force participation rates begin to mirror those of men, further increases in married women's labor force participation rates may result from intergenerational changes in perspectives on the part of both women and men. 10

11 Figure 1 Labor Force Participation Rate Women 20 yrs. & over Table 1 Sample Means: Married Women 25 to 55 Yrs of Age For Years 1991, 1996, 2001, and 2007 (Standard Deviation) In LF 0.71 (0.45) 0.73 (0.44) 0.73 (0.44) 0.78 (0.41) Age (7.81) (7.57) (7.84) 39.83(9.05) Race White 0.71(0.46) 0.64 (0.48) 0.68 (0.36) 0.70 (0.46) AgeYgKid 6.62 (4.85) (6.44) 9.77 (6.26) (6.78) YrsEd (6.62) (2.38) (2.76) (2.49) LgMetArea 0.71 (0.45) (8.25) 0.41 (0.49) 0.66 (0.47) LogWageHusb 2.42 (0.67) 2.69 (0.73) 2.86 (0.74) 2.79 (0.489) HlthStaHusb (1.07) 2.25 (1.04) 2.22 (1.01) 2.23 (0.99) HlthStaWf 2.32 (1.02) 2.22 (0.98) 2.30 (0.99) 2.33 (0.99) StateUnpRt 5.78 (2.07) 4.97 (1.23) 4.43 (0.78) 4.73 (0.78) Number 2,659 2,260 10,012 2,502 11

12 Table 2 Maximum Likelihood Probit Estimates: Prime Age Married Women 1991, 1996, 2001, and 2007 ( Marginal Effects) Variable Coefficient... Constant 5.91E-01** 8.36E *** 1.08** (0.017) (0.975) (0.000) (0.013) Age -1.06E-02** -1.84E-02*** -6.19E-03*** -7.65E-03* (0.033) (0.000) (0.007) (0.077) 3.68-E E E E-02 AgeYgKid 5.21E-02*** 4.38E-03*** 4.37E-02*** 5.49E-02*** (0.000) (0.000) (0.000) (0.000) 1.80E E E E-02 YrsEd 6.86E-02*** 1.05E-01*** 8.47E-02*** 1.11E-01*** (0.000) (0.000) (0.000) (0.000) 2.34E E E E-02 LAvgWgHus -8.51E-02** -9.46E-02** -2.81E-01*** -1.53E-01*** (0.044) (0.028) (0.000) (0.000) -2.59E E E E-02 LgMetArea -1.33E-01** -5.51E E E-02 (0.026) (0.889) (0.000) (0.267) -4.54E E E E-02 HlthHusb 2.41E E E-01* 6.89E-02** (0.432) (0.146) (0.075) (0.054) 8.38E E E E-02 HlthWf 1.54E-01*** 5.72E E-02*** -2.29E-02*** (0.000) (0.871) (0.000) (0.000) -5.34E E E E-02 StUnpRt -1.67E E-02** -1.07E-01*** 3.37E-02 (0.191) (0.050) (0.000) (0.247) -5.81E E E E-02 RaceWt -2.68E-01*** -1.16E-01* 5.39E E-01* (0.000) (0.067) (0.664) (0.105) -9.04E E E E-02 Pseudo R Number 2,659 2,260 10,012 2,502 Binary dependent variable: In Labor Force = 1, zero otherwise. *** Significant at 1% or less ** Significant at 5% or less * Significant at 10% or less P > z in parentheses Marginal effects are the change in the probability of labor force participation being limited for a one-unit change in the corresponding independent variable. These are computed at the mean of the independent variables. 12

13 Table 3 Probabilities Married Women Labor Force Participation Rates 10 Beta91 Beta96 Beta01 Beta X 69.78% 59.46% 75.95% 84.41% 1996X 74.88% 65.89% 80.22% 88.27% 2001X 74.35% 64.93% 78.81% 87.73% 2007X 75.39% 66.55% 80.30% 88.58% Predicted 70.13% 74.70% 76.10% 81.70% Observed 71.00% 73.00% 73.00% 78.00% Figure

14 References Bowen, William G. & Finegan, Aldrick T. (1969) The Economics of Labor Force Participation, Princeton University Press. Bremmer, Dale & Kesselring, Randy "Divorce and Female Labor Force Participation: Evidence from Time- Series Data and Cointegration", (2004) Atlantic Economic Journal 32(3) pp Brusentsev, Vera (2006) Evaluation of Female Labor Force Participation in the United States: , International Advances in Economic Research 12 pp Feilds, Judith M. (1976) A Comparison of Intercity Differences In The Labor Force Participation Rates of Married Women In 1970 with 1940, 1950, and 1960, The Journal of Human Resources XI, (4). Hotchkiss, Julie L. (2006) Changes in Behavioral and Characteristic Determination of Female Labor Force Participation, , Economic Review, (Second Quarter:2006) pp Mincer, Jacob Labor Force Participation of Married Women, Aspects of Labor Economics, (1960). Nakamura, Alice & Nakamura, Nasao, (1994) Predicting Female Labor Supply Effects of Children and Recent Work Experience, The Journal of Human Resources XXIX, (2), Spring Smith, James P & Ward, Michael P., (1985) Time Series Growth in the Female Labor Force, Journal of Labor Economics, Vol 3, #1, pp U.S. Department of Labor, Bureau of Labor Statistics, Endnotes 1 From 1900 to 1940 female participation rates increased fivefold. Participation rates actually decline for women under 35 years of age between 1940 and The decline in male labor force participation is due in large part to the increase in years of schooling, postponing younger males entry into the labor force, and to the tendency of older males to retire earlier than their fathers. 3 Mincer notes that the demand for household production is not likely to be negative. 4 The effects of children on married women's labor force participation may be decomposed into three categories. More children increase the amount of work done in the home, thereby reducing labor force participation. At the same time more children increases the need for family income while older children may provide a source of assistance with younger children as well as other household tasks. 5 Only "prime age" married women are included here since under the age of 25 the issue of school and work arises. Retirement decisions play a role in participation decisions after the age of 55. Prime age workers may also be less affected by the business cycle due to their relative greater attachment to the labor market. 6 The question as to whether behaviors drive characteristics or vice versa is a thorny one indeed. We assume here that changes in characteristics are exogenous. 7 Nakamura and Nakamura note that in addition to capturing the effect of children on labor supply; i.e. their time demands, the children variable also captures a taste for household production. 8 Data on city or SMSA were not included in the 1996 Wave. State population in millions was used as a proxy. 9 Health status for both husbands and wives are given the designation 1 through 5 with "1" excellent health and "5" the poor health. 10 Beta(year) refers to behaviors (the estimated beta coefficients) during a given year. (Year)X refers to characteristics during a given year. Thus reading across rows indicates constant characteristics and changing behaviors. Reading down rows refers to constant behaviors and changing characteristics. 11 Unfortunately the correspondence between 1996X and 2007X makes the two simulated curves over lap so closely that they are difficult to distinguish in this graph. 14