U.S. Bureau of the Census
Washington, D.C. 20233-8800
Population Division Working Paper No. 35
Racial-Ethnic and Gender Differences in Returns to Cohabitation and
Marriage: Evidence from the Current Population Survey
by Philip N. Cohen
by Philip N. Cohen
This paper was prepared for presentation at the 1999 Annual Meeting of the Population Association of America (PAA).
This paper reports the research and analysis undertaken by Census Bureau Staff. It has undergone a more limited review than official Census Bureau publications. This report is released to inform interested parties of research and to encourage discussions.
Population Division Working Paper Series
Using data on cohabitation from the 1995-1997 March Current Population Survey, the first three years in which the survey included "unmarried partner" as a relationship category, I measure the relationship between earnings and cohabitation as well as other marital statuses across racial-ethnic groups for men and women. Results show that among 25-54 year-old workers, black women have the largest cohabitation "premium" -- the earnings advantage over never-married workers -- more than three-times the premium for white women. Hispanic women have no cohabitation premium. White men have the largest marriage premium, and each other group except white women also has a significant marriage premium. There is a significant cohabitation benefit for white men, black men, and Hispanic men. Substantial differences in observed effects across groups suggest the need for models that are more complicated than previously used. Research into marital status effects on earnings is misleading when restricted to white men.
A large body of research documents the earnings advantage that married men enjoy over never-married men, the "marriage premium." Marital status is now a control variable in most earnings models, despite disagreements in the literature over whether the source of marital-status effects lies in productivity, selection, discrimination or other factors (Cornwell & Rupert 1997). Some analysts recently have included nonmarital cohabitation in earnings models, generally finding a somewhat smaller but still significant premium to cohabitation (Daniel 1992; Loh 1996). Almost all of this research has examined men's earnings exclusively, and most of it has not examined racial-ethnic groups separately. Using data from the March Current Population Survey, this paper asks the basic question: is there a cohabitation "premium" in wages observed in cross-section in the years 1994-1996, and if so, does this premium differ across gender as well as racial-ethnic groups?PREVIOUS RESEARCH
Historically, married men have earned more than never-married men, controlling for measured differences in social and demographic characteristics. Several explanations have been offered for this marriage premium. First, married men may be more productive at work due to contributions from their wives (childcare or other housework, nurturing, and so on), or they may be more motivated to earn more because of their family obligations, thus seeking out better jobs -- both variants of a productivity hypothesis (Cornwell & Rupert 1997). Second, employers might discriminate in favor of married men. Or third, men who marry might have labor market advantages that are not measured in labor force surveys, in other words a selection effect by which more productive men are at higher risk of union formation or marriage (Loh 1996).
As these competing explanations are being adjudicated, however, the facts are also changing: the "marriage premium" for white men fell substantially from the 1960s to the 1980s (Gray 1997). In Loh's (1996) analysis, the marriage premium for white men peaked at 25% greater than never-married men's earnings in 1969, and then dropped to 11% in 1979. For black men, however, the premium increased steadily from 1939 to 1979, reaching 38%. Because the three possible mechanisms are not mutually exclusive, a change in the amount of the premium could logically result from changes in any of the different mechanisms. Using longitudinal data, Gray concludes that "the marriage premium has ... changed in its origins, from reflecting primarily productivity-enhancing effects of marriage to being exclusively a result of selection into and out of marriage" (1997:495). Loh (1996) does not find support for the productivity hypothesis among white men. Daniel (1992) concludes that about half of the premium is due to selection, and the remainder could be the result of the productivity enhancements of marriage.
Most recently, Cornwell and Rupert (1997) observe from longitudinal data that men who eventually marry earn more than men who do not; that is, "prospective marriage" is a predictor of higher earnings. This undermines the claims that marriage itself enhances men's productivity, either through women's labor contributions or through men's greater commitments or ambitions. They conclude: "We find that the wage premium can be explained largely in terms of unobservable individual characteristics which are positively correlated with marriage and wages. In other words, attributes leading to 'good' (long and stable) marriages are also important in obtaining 'good' (long and stable) jobs and higher wages" (286). Thus they argue that cross-sectional regressions of earnings on marital status will overstate the positive returns to marriage for men.
Cross-sectional studies such as this one may therefore be informed by the theoretical background to these questions, but they are not best suited to testing competing hypotheses in this area. Without the longitudinal data necessary for examining potential effects for individuals over time, then, this study is intended to be more descriptive than explanatory.
Daniel (1992) tests for a cohabitation premium in addition to the marriage premium in his analysis of data from the National Longitudinal Survey of Youth (NLSY). He finds that cohabitation brings returns about half the size of marriage to white men. Loh (1996) also finds a cohabitation premium smaller than the marriage premium (although that analysis is limited to those cohabitors who ended up marrying). Oppenheimer, Kalmijn and Lim (1997) provide evidence supportive of the selection mechanism for cohabitation effects. Their finding that career paths strongly influence entry into marriage is consistent with selection effects in general. And in particular, they argue that partner selection based on economic prospects probably occurs early in the union formation process, so it would presumably influence cohabitation as well as marriage effects. However, selection for cohabiting partners might not be as rigorous as it is for marital partners (Manning & Smock 1995).
Historically, women's careers have been hurt by marriage (Goldin 1995) and childbearing (Leibowitz & Klerman 1995), but this tendency has diminished in recent decades (O'Connell 1990). It is not clear at the outset how current cohabitation effects might differ from marriage effects for women. However, women for whom career success is important are relatively more likely to enter into cohabitation than marriage for their first union, while the same is not true of men (Clarkberg, Stolzenberg & Waite 1995). This suggests that cohabiting women might have higher earnings than married women.
Gender and racial-ethnic differences
The literature on the marriage premium has generally investigated marital effects on men's earnings, reflecting assumptions about both the labor market as well as marriage and family relationships. This is despite the fact that comparing premiums (or penalties) between men and women would be a useful tool for considering competing hypotheses about the mechanisms at work. For example, perhaps the decline in the productivity-enhancing aspects of marriage for white men (Loh 1996) is related to changes in married white women's labor force behavior or housework. Labor market and household behaviors of both partners provide evidence to help sort out these questions (Daniel 1992). I will consider effects for both men and women (but not couples) which might be instructive for future research.
In addition, research on the marriage premium has mostly considered white men. However, available evidence, including that presented by Loh (1996) suggests important differences by race and ethnicity. There has been considerable attention given to variation in marital behavior across race and ethnic groups (e.g., Casper & McLanahan 1995), race differences in the relationship between employment and family transitions (e.g., O'Connell 1990), and important differences in local marriage markets by race (Brien 1997). It seems that attention to wage effects across race and ethnicity might help explain some of the underlying processes in the marriage market or within couples that contribute to these differences. There is also evidence of diverse effects of employment on entry into marriage (South 1996). Manning and Smock (1995) find that employment increases cohabitors' chance of marrying among white men but not among white women, and for both Black men and women.
Evidence from cohabitation studies of other outcomes
Research on non-marital cohabitation has mostly concerned questions of union formation and quality (Brown & Booth 1996), fertility (Bachrach 1987; Manning & Landale 1996), and housework related questions (South & Spitze 1994). However, these studies offer implications for the question of a cohabitation effects on earnings, and for potential gender and racial-ethnic differences.
Connections between earnings and cohabitation are relevant to poor women and families, especially if joblessness is more of a deterrent to marriage than it is to cohabitation. Failing to consider cohabitation among households with lower incomes might lead to misunderstanding the role of income, welfare and union formation (Moffitt, Reville, & Winkler 1998). And if cohabitation effects earnings, this should be taken into account in the consideration of cohabitation's effect on poverty and related issues (Bauman 1997; Manning & Lichter 1996).
Some of the cohabitation research has concerned the extent to which cohabitation is similar to or different from marriage, and the evidence here compels attention to racial-ethnic differences, which has been neglected in much of the cohabitation literature (Brown & Booth 1996; Manning & Landale 1996). In terms of childbearing behavior, cohabitation more closely resembles marriage among Black women than among White women (Loomis & Landale 1994). Manning and Landale (1996) find that, for non-Hispanic White women, cohabitation looks more like a transition to marriage, which is not the case Black women. And for Puerto Rican women, cohabitation is more like marriage itself.1 Most cohabitors do plan to marry, and those that do marry have marriages of quality similar to married couples (Brown & Booth 1996). The question of to what extent cohabiting relationships are like marriages is complicated by the different circumstances of cohabitation. In fact, the extent to which cohabitation on average appears to be somewhere in between dating and marriage on various indicators may be because cohabiting relationships are split between those that are much like marriage, and those that are very little like marriage.DATA AND METHODS
I use cross-sectional data to shed some light on, if not resolve, these questions. The Current Population Survey (CPS) is a large, nationally representative survey, well suited for examination of current cohabitation, marriage and labor force outcomes across racial-ethnic groups. The CPS began including "unmarried partner" as a relationship category in 1995. This paper pools the first three survey years of data from the March survey (1995-1997) with this category to analyze directly-identified cohabitors' individual earnings for the first time with CPS. The partner category is only identified for those who are partners with the household reference person. Cohabitors here are therefore those partners and the household reference people with whom they are living. Marital status is then recoded into exclusive categories including cohabitation.
Given the informal nature of cohabitation arrangements, it is important to note that survey definitions vary more than usual, as perhaps do respondent interpretations of survey questions. Estimates of cohabitation differ sharply depending on whether they are indirectly derived from living arrangements or asked directly, and according to the manner in which they are asked. For a variety of reasons, the cohabitation measure in the March CPS reports lower prevalence rates than do other national surveys with direct measures (for a review, see Casper, Cohen & Simmons 1999). Thus although some cohabitors may be missed by the CPS (including all couples that do not include a household reference person), there is no reason to suspect that those identified as cohabiting are falsely identified.
The sample is selected from White, Black, or Hispanic2 adults ages 25-54. The analysis used is OLS regression, and the dependent variable is the natural log of hourly wages for the previous year. Hourly wages are computed by dividing annual earnings by the product of weeks worked and hours usually worked per week. Hourly wages are log-transformed on methodological grounds to normalize the distribution of the variable, and on theoretical grounds under the assumption that a given dollar increase is more meaningful at lower wage levels. Those who earned less than $2 per hour on average over the year are excluded. The final sample includes 138,499 cases, 2,772 (2.0%) of whom are self-identified as cohabiting.3 Figure 1 shows the marital status distribution for each of the six race-ethnic/gender groups in the analysis.
Wage differences are identified in OLS regressions with dummy variables for five racial-ethnic-gender groups plus white men as the excluded category. With no other variables in the model, the coefficients for these variables will indicate the predicted difference in earnings between each group and white men. Marital status is coded into four dummy variables: married, was married (divorced, separated, or widowed), cohabiting, and never married. Never married is the excluded category. The marital status dummy-variables interact with each of the six racial-ethnic-gender groups (married white man, cohabiting white man, etc.).
When the marital status variables are included in the regression models, the intercept then represents the average logged hourly earnings for never-married white men. And the coefficient for black women, for example, will represent the difference in earnings between never married black women and never married white men. The coefficient for cohabiting black women will represent the average difference in earnings between cohabiting and never-married black women. Thus the marital status interaction variables will allow identification of marital status effects as differences from never-married within each racial-ethnic-gender group. This construction allows measurement of cohabitation or marriage premiums (or penalties) for each group in one equation.
Other variables in the analysis include years of education, hours worked per week last year (logged), potential experience (age-education-6) and its square, and a dummy variable for people with a disability. The presence of children is controlled with three dummy variables: children under age 5 only, children under age 5 and ages 5-17, and children ages 5-17 only (no children is the excluded category). Other controls include dummy variables for the four Census regions of the country, and a dummy variable for residence in a metropolitan area. Table 1 provides a summary of the sample and variables used.
Two sets of the control variables are coded as interactions. The children dummy variables and the potential experience variables are interacted with gender (man * children under age 5 only, etc.). And the education variables are interacted with all six racial-ethnic-gender groups (Hispanic man * education, etc.). These allow the effects of children and work experience to vary across gender, and the returns to education to vary across gender and race-ethnicity.
The regression analysis begins with a baseline model to identify average differences in log wages for each race-ethnic-gender group from white men. A second model introduces the marital status variables, essentially testing the differences in mean wages between marital status groups. A third model adds the controls for children interacted with gender. Because the data are cross sectional, this model is one limited way of identifying the extent to which marital status effects result from the presence of children, which might be the case if marital status effects are substantially reduced with the addition of children variables by gender. The final model adds the rest of the control variables. This analysis allows measurement of marital status effects, and comparison of these effects with and without other control variables.RESULTS
The summary statistics (Table 1) show that, compared to those who are currently married, cohabitors in the sample are on average younger, and (except among Hispanics) less educated. Cohabiting white and black men work fewer hours (as well as weeks) than do married white and black men, and the opposite is true for white and black women. Cohabitors also have fewer children, which are more likely to be younger.
The wages of the gender groups, by marital status, are shown in Figure 2. For white men, black men, black women, and Hispanic men, cohabitors appear to earn wages between the never married and currently married groups, as has been most commonly found in the literature. However, this does not appear to be the case with white or Hispanic women, among whom never married workers appear to earn more. The multivariate analysis will attempt to sort out and test these differences more clearly. One important observation in Figure 2 is that a large portion of white men's wage advantage is correlated with marital status. Only 17 percent of the white men in the sample are never married, but these men do not earn more per hour on average than never married white women, or married black men.
Regression results are presented in Table 2, which shows statistics from the four models. In the first, baseline model, white men earn the most (the intercept translates into $14.50 per hour), followed by black men, white women, Hispanic men, black women, and Hispanic women. This basic model accounts for only 7.5 percent of the variation in wages.
The second model adds marital status variables, and the explained variation is 9.8 percent. Here white men show a significant but not large cohabitation premium of .09 in log wages, compared to a marriage premium of .34 in log wages. Black men and Hispanic men have somewhat larger cohabitation premiums than white men (.11 and .10 respectively), and smaller marriage premiums (.30 and .21).
Model 2 shows much greater differences among women, however, and each group of women shows a different pattern. White women in each group earn lower wages than those who have never been married. (Note that, as seen in Figure 2, there is no significant difference between the wages of never married white women and white men.) For black women the pattern is basically reversed. Black women in all three union statuses have earn more than the never married. Black women's marriage premium (.17) is about half again as much as their cohabitation premium (.11). For Hispanic women, there are no significant differences across marital statuses. Unlike white women, never married Black and Hispanic women earn substantially less than never married white men.
As noted, model 3 is included to see if the marital status effects are accounted for by differences in presence of children across marital status groups. In this model, the intercept represents never-married white men with no children. The group whose marital status effects are most substantially changed is white women. Controlling for the presence of children, married white women earn no less than never-married white women. The effects of cohabiting and being formerly married on white women are reduced almost by half in this model. Thus it appears that for white more than for black or Hispanic women, differences in wages across marital status groups are accounted for by the presence of children. Model 3 accounts for 10.1 percent of the variation in wages.
Finally, the complete set of variables is entered in model 4, which accounts for 26.8 percent of the variation in wages.4 In this final model, black women have the largest cohabitation premium, more than three-times the premium for white women (Hispanic women have no cohabiting premium), and substantially larger than the premium for all three groups of men. With all the control variables entered the cohabitation premium for black women is increased rather than diminished -- to .18 from .13 in the previous model. Comparing this model to model 1, there is some evidence consistent with marital selection effects here, in the decline in marriage premium for white men (about 31%), black men (39%), and Hispanic men (23%) with the inclusion of human capital and other variables. However, except for a decline among Hispanic men, the cohabiting premiums for men are not substantially different in this model.
This last model reflects the much smaller returns to education for Hispanic workers and to a lesser extent black men.5 It also shows somewhat greater returns to potential experience for men, presumably representing their greater "commitment" to the labor force and faster rates of promotion. Wages are higher in each region compared to the South, especially in the Northeast. And disabled workers earn substantially lower wages on average.
The marital status effects from the final model 4 are presented in graphic form in Figure 3, in which each bar represents the predicted difference for a given marital status from never married members of the same race-ethnic-gender group in log wages, net of the control variables. The pattern among men is similar to that reported in previous research: married men enjoy a significant marriage premium compared with never-married men, and cohabiting men's premium is roughly half as large. White men's premiums, especially the marriage premium, appear larger than those for black and Hispanic men. For women, however, the pattern in Figure 3 is much different. Black women have the largest marital status effects, especially their cohabitation premium. For white women the only significant effect is from cohabitation, but this is small.
The differences in the results between black and white women bear emphasis. For white women, the change in direction of the cohabitation effect from negative in model 2 to positive in model 4 for white women is a result of the more favorable distribution of individual characteristics among never-married white women. That is, white women with human capital characteristics conducive to higher wages are more likely to be neither married nor cohabiting. This is consistent with the comparisons of mean values in Table 1. Never married white women in the sample have the highest levels of education, work the longest hours and the most weeks, and are much less likely to have children of any age. Net of all these factors, white women who cohabit earn slightly higher wages than those who are never married.
The same pattern does not hold for black women, among whom cohabitors in the sample work the longest hours and most weeks, and are much closer to never marred women in education and likelihood of having children present. Cohabiting black women earn higher wages than those who are never married. Like white women, however, the inclusion of the full control model increases the positive effects of cohabitation for black women.CONCLUSIONS
This analysis is not structured to concretely test competing hypotheses regarding cohabitation and marriage effects on earnings. Such questions are best addressed by longitudinal studies of the kind reviewed above. However, previous work has not adequately addressed racial-ethnic and gender differences in this area. This study clearly suggests the need for greater attention to these differences in future research.
Consistent with most previous research on men, on average cohabitation appears to fall somewhere between married and never married in most of these results. Unresolved is the extent to which this reflects a truly middle-ground nature to the relationship versus the combination of some relationships that are and some that are not like marriages. The tendency for cohabitation to fall between marriage and never-married raises questions about discrimination as a mechanism for the effect, since being single is usually less of a taboo than living with an unmarried partner.
For white men and women, the effects of cohabitation and marriage in the results here are most consistent with previous research, which is not surprising because most of this research has focused mostly or exclusively on white couples. Although both white and black women have cohabitation premiums net of other factors, black women's effects are much stronger than white women's. Hispanic women, unlike the other two groups, have a marriage premium only. The racial-ethnic differences here demonstrate the need for broader consideration of these effects. Clearly, analysis of whites alone should not be considered generalizable given these results.
The fact that black men and women both have marriage and cohabitation premiums has implications for inequality among black workers, especially given lower black marriage rates. The apparent mutual selection of higher-earning black men and women would contribute to increases in family-level income inequality. That is, at the same time that marriage rates are low, those with higher wages are more likely to be married or cohabiting together. This has implications for inequality among children in different types of families as well. The high cohabitation premium for black women also might suggest that black couples are less likely to marry when women are earning more, creating a hurdle between the formation of cohabiting unions and marriage. That is, black women's earnings might be more of an obstacle to marriage than they are to forming a cohabiting union.
The greater marriage effects for white and black men are consistent with previous research that more successful men are more likely to marry versus remain in cohabiting relationships, although that cannot be tested in this cross-sectional study. For women, only among white women is there evidence here that never married women earn more. With this data we cannot tell if marriage hinders white women's earning potential, or if white men women with higher earning potential or ambition are not marrying in the first place. These differences offer a caution against generalizing across racial-ethnic groups on this question.
2 White and Black are non-Hispanic; Hispanics may be of any race. Respondents who are not members of these groups are excluded because as a whole they are heterogeneous and thus not suited to combining, and because as subgroups their numbers in the CPS are too small to be reliable.
3 It should be noted that cohabitation is often a short-lived relationship (Bumpass & Sweet 1989), which means that those who were cohabiting at the time of the March CPS interview might not have been cohabiting throughout the previous year, and thus these annual earnings data are not necessarily concurrent with the marital status variable.
4 Note that in this model, the intercept now represents never-married white men with no children and zero on all the control variables. While this number is no longer itself relevant, the differences in log wages represented by the other coefficients are still readily interpretable.
5 The positive coefficients for Hispanic men and women dummy variables in this model are an artifact of the education slopes at zero years of education -- outside the range of the data.
Bauman, Kurt. 1997. "Shifting Family Definitions: The Effect of Cohabitation and Other Nonfamily Household Relationships on Measures of Poverty." Poverty Measurement Working Papers, U.S. Bureau of the Census.
Brien, Michael J. 1997. "Racial differences in marriage and the role of marriage markets." Journal of Human Resources 32(4):741-778.
Brown, Susan L. and Alan Booth. 1996. "Cohabitation Versus Marriage: A Comparison of Relationship Quality." Journal of Marriage and the Family 58 (Aug.):668-678.
Bumpass, Larry L. and James A. Sweet. 1989. "National Estimates of Cohabitation." Demography 26(4):615-625.
Casper, Lynne M, Philip N. Cohen and Tavia Simmons. 1999. "How Does POSSLQ Measure Up? Historical Estimates of Cohabitation." Paper presented at the Population Association of America.
Clarkberk, Marin, Ross M. Stolzenberg, and Linda J. Waite. 1995. "Attitudes, Values, and Entrance Into Cohabitational versus Marital Unions." Social Forces 74(2):609-634.
Cornwell, Christopher and Peter Rupert. 1997. "Unobservable Individual Effects, Marriage and the Earnings of Young Men." Economic Inquiry 35(April):285-294.
Daniel, Kermit. 1992. "Does Marriage Make Men More Productive?" Working Paper 92-2. Economics Research Center, NORC and University of Chicago.
Goldin, Claudia. 1990. Understanding the Gender Gap: An Economic History of American Women. New York: Oxford University Press.
Gray, Jeffrey S. 1997. "The Fall in Men's Return to Marriage: Declining Productivity Effects or Changing Selection?" Journal of Human Resources 32(3):481-504.
Leibowitz, Arleen and Jacob Alex Klerman. 1995. "Explaining Changes in Married Mothers' Employment over Time." Demography 32(3):365-378.
Loh, Eng Seng. 1996. "Productivity Differences and the Marriage Wage Premium for White Males." Journal of Human Resources 31(3):566-589.
Loomis, Laura Spencer and Nancy S. Landale. 1994. "Nonmarital Cohabitation and Childbearing Among Black and White American Women." Journal of Marriage and the Family 56(Nov.):949-962.
Manning, Wendy D. and Nancy S. Landale. 1996. "Racial and Ethnic Differences in the Role of Cohabitation in Premarital Childbearing." Journal of Marriage and the Family 58(Feb.):63-77.
Manning, Wendy D. and Pamela J. Smock. 1995. "Why Marry? Race and the Transition to Marriage Among Cohabitors." Demography 32(4):509-520.
Moffitt, Robert A., Robert Reville, and Anne E. Winkler. 1998. "Beyond Single Mothers: Cohabitation and Marriage in the AFDC Program." Demogaphy 35(3):259-278.
O'Connell, Martin. 1990. "Maternity Leave Arrangements: 1961-85," in Work and Family Patterns of American Women, U.S. Bureau of the Census, Current Population Reports, Series P-23, No. 165, U.S. Government Printing Office, Washington, D.C.
Oppenheimer, Valerie Kincade, Matthijs Kalmijn and Nelson Lim. 1997. "Men's Career Development and Marriage Timing During a Period of Rising Inequality." Demography 34(3):311-330.
Oropesa, R. S. 1996. "Normative Beliefs About Marriage and Cohabitation: A Comparison of Non-Latino Whites, Mexican Americans, and Puerto Ricans." Journal of Marriage and the Family 58(Feb.):49-62.
South, Scott J. 1996. "Mate availability and the transition to unwed motherhood: a paradox of population structure." Journal of Marriage and the Family 58(May):265-79.
South, Scott J and Glenna Spitze. 1994. "Housework in marital and nonmarital households." American Sociological Review 59(June):327-47