Examining the wage differential for married and cohabiting men.

Stratton, Leslie S.

Leslie S. Stratton (*)

I. INTRODUCTION

Wage analyses almost universally indicate that married men earn more than do single men, even after controlling for observable human capital characteristics. The same appears to be true for cohabiting men. Research has failed as yet to reach a consensus regarding the nature of these differentials. The evidence is consistent with a number of alternative explanations. First, marriage could increase men's market productivity because economies of scale and increased specialization within a multiperson household give them more time and energy to devote to market-related activities. If this is the mechanism driving the marital wage differential, then cohabiting men may also experience a wage boost, albeit a smaller one, as cohabitation is a less stable relationship and less likely to engender specialization. Wages could jump at the start of a relationship, but an increased focus on market-related activities is more likely to cause wages to rise at a faster pace. Alternatively, men who marry or cohabit may be inher ently different from men who do not. If men are selected into relationships based on their earnings ability, then cross-section wage analyses will indicate a wage differential for married and cohabiting men, but difference estimation will not. The goal of this analysis is to shed light on the nature of the marital and cohabitation wage premiums for men by estimating wage models that permit both differential wage growth and selection effects.

II. LITERATURE REVIEW

Evidence of a marital wage premium for men abounds. A thorough literature review is beyond the scope of this article, but incorporation of a marital dummy in wage specifications for men is fairly standard. Hill (1979) provides one much-cited work. Empirical estimates in the range of 10% to 30% are typical.

A number of researchers have explored the nature of this wage differential. One explanation focuses on the marital decision itself, the argument being that men who are inherently more productive are more sought after marriage partners and hence more likely to marry. This is the selection model. Attempts, such as that by Nakosteen and Zimmer (1987), to estimate a model in which marital status and wages are endogenously determined have yielded inconclusive results and are sensitive to identification restrictions. An alternative approach, less subject to specification error, is to estimate a fixed effects specification that eliminates all individual-specific time-invariant characteristics. (1) Researchers Korenman and Neumark (1991), Bartlett and Callahan (1984), Daniel (1991), and Gray (1997) have employed this technique and continue to observe a significant marital wage differential, indicating that selection effects do not explain the entire differential. Gray (1997) finds some evidence that the premium has declined over time, as he does not find a significant marital wage differential using data from the National Longitudinal Survey of Youth (NLSY), the most recent cohort of data tested. However, Daniel (1991) also uses data from the NLSY, and he finds a significant marital wage differential of about the same magnitude as that obtained from older cohorts. Generally speaking, a comparison of cross-section and panel results suggests that less than 20% of the marital wage differential is attributable to individual specific components or selectivity.

If selection alone does not explain the differential, then wages must increase following marriage. Wages could be higher because they jump or because they rise more rapidly following marriage. Empirical estimates reported by Kenny (1983) using the Coleman-Rossi Retrospective Life Histories Study and by Korenman and Neumark (1991), Loh (1996), and Gray (1997) using the National Longitudinal Study of Young Men suggest that the growth rate of wages increases on marriage. Once again, results from the NLSY are mixed with Daniel (1991) finding faster wage growth following marriage and Gray (1997) finding no marital wage differential.

The theory most frequently cited to explain increased wage growth following marriage derives from Becker's (1981, 1985) model of the household. According to this theory, multiperson households are able to take advantage of economies of scale in home production and of gains from specialization both in the allocation of time and in the accumulation of human capital within the household. Economies of scale arise in home production because it takes about as long to cook a meal for two as it does to cook a meal for one. Specialization within the household occurs when one member of the household takes on a greater share of the home production tasks and another takes on a greater share of the market work. Historically men have specialized in market production, women in home production. With less time and energy devoted to home production both because of economies of scale and because of specialization, men who marry should be free to invest more in market specific human capital and hence have a steeper experience-ea rnings profile than men who do not marry. (2)

Research involving cohabiting men is considerably more limited. Both Loh (1996) and Daniel (1991) theorize that if the marital wage differential is due to the economies of scale and specialization possible within a joint household, then cohabiting men should also experience a differential. Loh (1996), Daniel (1991), and Cohen (1999) all report finding that men currently cohabiting receive significantly higher wages than other never married, not cohabiting men. Loh further postulates that men whose current marriage began with cohabitation should receive a wage differential but finds no evidence of such. None of these studies, however, controls for time cohabiting, and only Daniel controls for years married or for selection effects. Thus, it is not clear from this work whether cohabitation influences wages in the same manner as marriage. The analysis proposed here will include controls for both time married and time cohabiting and will formally test whether marriage and cohabitation have the same impact on wage s.

Of course, cohabiting relationships are different from marriages. Legally, cohabiting couples are less responsible for supporting one another than are married couples. Cohabiting relationships are also less stable than marriages. Research by Bumpass and Sweet (1989) indicates that 17% of all first cohabitations end within one year, 33% within three years. Comparable figures for first marriages indicate that 5% end within one year and 13% within three years. Finally, cohabiting relationships are of a much more heterogeneous nature than marital relationships. The degree of commitment to the relationship can vary widely. Each of these differences could cause the relation between marriage and wages to be different from the relation between cohabitation and wages.

The greater legal responsibilities imposed by marriage could have a greater "settling effect" on men than cohabitation. Married men may adopt a more serious attitude toward employment, increasing their efforts and hence productivity on the job. Wives may be more inclined to provide both spousal pressure and spousal support for market work than cohabiters, who have less of a stake in their partner's career.

The stability of the relationship could also be important. (3) Following Becker's model of the household, if there are any costs associated with changing one's schedule of activities, then individuals in more stable relationships are more likely to specialize in market/household production and to reap the resulting benefits. Alternatively, a more stable relationship may bring about a greater change in nonproduction-oriented activities. There may be less bar hopping and fewer nights on the town to disrupt on-the-job performance. Both the differences in stability and the differences in legal responsibilities suggest that the cohabiting wage effect may be smaller than the marital wage effect.

The heterogeneous nature of cohabiting relationships suggests a more complicated comparison. Willis and Michael (1994) suggest that couples might move in together to test a relationship about which they are uncertain. Schoen and Weinick (1993) suggest that cohabitation is an alternative to marriage. Rindfuss and VandenHeuvel (1990) suggest it is an alternative to being single. Finally, there are those who report cohabiting to save money for a wedding. With so many alternative motivations, it is difficult to justify treating all cohabiters the same. Indeed Winkler (1997) has demonstrated that as a group cohabiters do not pool their income in making labor supply decisions, but some longer-term cohabiters and those with biological children do. The level of commitment to the relationship (not the legal status) may be the key factor. Fixed effects estimates will control for the heterogeneity in cohabiting relationships arising from fixed, individual specific effects, much like the selection effects in the marriage literature.

III. EMPIRICAL SPECIFICATION

The analysis begins with a standard, cross-section log wage regression augmented by controls for marital status:

(1) log([W.sub.i]) = [alpha][X.sub.i] + [tau]MS[T.sub.i] + [[epsilon].sub.i],

where W is the wage, X is a vector of observable characteristics, MST stands for marital status, and the subscript i denotes the individual. Following the literature, two marital status indicators are included in specification (la): one to indicate those who are married and another to indicate those who have been married but are now divorced, separated, or widowed. It is this specification that models wages jumping with changes in marital status. In a second specification (1b), variables reflecting years married and years divorced/separated/widowed are also added. These duration measures are nonoverlapping, so for those who have remarried, time divorced/separated/widowed measures only time between marriages. These measures are also cumulative, including time in both current and previous spells. This specification permits the growth rate of wages to differ by marital status. (4)

As discussed in the literature, selection or heterogeneity concerns may bias cross-sectional analysis. Difference or panel analysis provides a means of controlling for bias that is individual specific and time invariant or at least slow to change. The difference model is

(2) log([W.sub.it]) - log([W.sub.is]) = [alpha]'[DELTA][X.sub.i] + [tau]'[DELTA]MS[T.sub.i] + [DELTA][[epsilon]'.sub.i].

If [tau] and [tau]' are similar in sign and magnitude, observed marital differences are likely attributable to productivity not selection effects. However, if [tau]' is close to zero and statistically insignificant, then the significance of [tau] is likely driven by selection effects. (5) Once again, two alternative specifications will be estimated: one that controls only for marital status (2a) and the other that controls as well for marital duration (2b).

The relation between cohabitation and wages is explored in much the same manner. Dummy variables for current and past cohabitation are added to the model first, followed by measures of the time spent cohabiting (again a cumulative measure including both current and past cohabitations). Note that the coding of current and past cohabitation status is exclusive. One is either currently cohabiting, has cohabited in the past, or has never cohabited. Overlaps are possible between past cohabitation status and marital status. Thus those currently married who cohabited with their current spouse prior to marriage are coded both as currently married and as having cohabited in the past.

If the duration of the relationship is key, then relationships that are expected to be short-lived, whether cohabitations or marriages, could have different effects than those that are expected to be long-lived. Expected duration is unknown, but actual duration is often observed. This possibility is explored by performing a variety of sensitivity tests using measures of actual duration for marriages and cohabitations and by exploiting further information on the outcome of the cohabitations (specifically whether or not they ended in marriage).

IV. DATA

The data used in this study come from the National Survey of Families and Households (NSFH). This survey consists of a national random sample of individuals plus a double sample of cohabiting couples, single-parent families, families with stepchildren, recently married couples, and those of nonwhite or Hispanic descent. Particular care was taken in designing this survey to elicit honest, detailed responses regarding marital status and cohabitation, making this information unusually reliable as surveys go. (6) Respondents were interviewed first in 1987-1988 and again in 1992-1994.

Two different samples are employed in the analysis that follows: one for the crosssection and one for the panel analysis. In all cases, the analysis is restricted to white, non-Hispanic men under the age of 65. Of the 13,007 (10,005) individuals interviewed in the first (second) survey, only 3,281 (2,515) meet these criteria.

The data are pooled over time for the cross-section analysis. (7) Standard screening to exclude those enrolled full-time in school, those in the military, and those with missing or questionable education, job experience, or job tenure reduces the pooled cross-section sample by 10.1%. Of this sample, 10.3% were not working at the time of the interview, and 11.0% failed to report a valid wage. (8) Finally, the sample is restricted to those individuals for whom complete marital and cohabitation histories are available. Less than 3% of the sample is missing marital information, but approximately a quarter of those men reporting cohabitations (11.3% of the sample) fail to fully report the dates, and the more cohabitations reported, the more likely the individual is to have an incomplete cohabitation history. Fortunately, sensitivity analysis (discussed later) suggests that the results are not sensitive to this sample selection criterion. The final cross-section sample consists of 3,583 observations: 2,044 from th e first wave and 1,539 from the second wave of the survey.

A panel data set is constructed similarly. Of the 2,515 white, non-Hispanic men under the age of 65 at the time of both surveys, 1,358 men provide the information necessary for panel analysis. This sample is smaller than the second wave cross-section sample because some of this sample were enrolled in school or in the military or failed to provide employment or wage information at the time of the first interview. However, over 100 respondents, who did not cohabit between interviews but were excluded from the cross-section analysis because of incomplete cohabitation histories, are included in the panel data set because their activities prior to the first survey will difference out from the analysis. (9) As difference estimates of the marital and cohabitation wage effect are obtained based only on individuals whose marital or cohabitation status changes between interviews, it is important to note that 20% of the panel sample changes marital status and 14% change cohabitation status during the observation perio d (further details available on request).

Table 1 provides weighted sample statistics by survey wave for both the cross-section and the panel data sets. Most of the differences between surveys are driven by age differences, a few by data definition. Tenure at the time of the first survey is, for example, overstated for some respondents in the cross-section sample. (10) Residence in a standard metropolitan statistical area (SMSA) and union activity are also defined differently between surveys. (11) As compared to the data used in previous research on the marital wage differential, the NSFH sample exhibits greater heterogeneity. The respondents range in age from 18 to 64, rather than from the mid-twenties to only the mid-thirties. The more heterogeneous sample employed here has the distinct advantage of being less subject to multicollinearity problems when controlling for job experience, tenure, and years married. (12)

A comparison of wages by marital and cohabitation status (not shown) using the weighted NSFH sample sheds some light on the phenomenon to be explained. On average, married men earn 56% and previously married men 29% more than never married men. Of course, on average married men in this sample have over ten years more experience than never married men and so would be expected to earn more. Cohabiting men earn 29% more than those who are neither currently married nor cohabiting, but 14% less than those currently married. They are on average six years younger than married men with commensurately less experience and have on average a half year less education. Still, cohabiting men are a diverse group, and it is difficult to make broad comparisons. Regression analysis will control for observable differences in education, experience, and tenure.

V. RESULTS

Estimates of the impact marriage and cohabitation have on wages are presented in Table 2. The first two columns present results from the pooled cross-section analysis (specification 1); the last two columns present results from the panel analysis (specification 2). All models include a linear measure of education and time not working; (13) quadratic controls for experience and tenure; and dummy variables for residence in an SMSA and in the South, for union activity, and for the presence of children. The cross-section analysis also includes dummy variables for 11 industries, 7 occupations, and observations from the second wave of the survey. Industry and occupation dummies are excluded from the panel analysis because of the sensitivity of panel analysis to coding error, but the results are not sensitive to their treatment. Variables interacting the survey dummy with the SMSA, union activity, and (for the cross-section models) tenure measures are incorporated to account for possible time differences in the mea surement of these variables. Parameter estimates for these variables are comparable to those reported elsewhere in the literature and are provided in Appendix A. All regressions are weighted to account for the double sampling of cohabiting and recently married persons but are estimated with standard errors robust to the fact that each observation represents only one respondent. The same conclusions are supported by unweighted estimates.

The top part of Table 2 presents estimates controlling for marital effects alone. These results are comparable to those reported elsewhere in the literature. Cross-section analysis including only dummy variables for marital status (specification [1a]) indicates that married men receive wages that are substantially higher than never-married men (on the order of 18.4%), with some evidence of a smaller premium (5.9%) for men who are not now but have been married (those divorced/separated/widowed). The marital wage differential is significant at the 1% level; the effect of past marriages is significant only at the 25% level using a two-tailed test. This regression-based estimate of the marital wage differential is about one-third the size of the raw differential (56%), suggesting that differences in education, experience, and the other explanatory variables explain about two-thirds of the gross differential.

Column two reports the parameter estimates for the specification that controls for years married and years divorced/separated/widowed (1b). Though previous researchers have uniformly controlled for a quadratic in years married and a linear or quadratic term in years postmarriage, the specification employed here uses modified log duration measures. Both parameterizations imply that the return to years married increases at a decreasing rate; a quadratic specification further implies that the return to years married will eventually decline.

Previous research focused on 20-35-year-olds, individuals so young that few if any had experienced a marriage long enough to have reached the point of decline. The NSFH data employed in this study include men of all ages. When the NSFH sample is restricted to those less than age 36 and a quadratic duration measure is employed, the resulting parameter estimates match those reported in the previous literature quite closely. This sample does not appear unusual except in its age distribution.

However there is no evidence of a declining return to marriage even after 20 years of marriage and the modified log duration measures provide a fit that is more robust across survey waves. The modification consists of adding one to the marital duration measure before taking the log. As the natural log of zero is undefined and the natural log of one is zero, this modification ensures that the log duration measures are all well defined.

The results of specification (1b) indicate that the entire observed marital wage differential is attributable to faster wage growth during marriage. When marital duration measures are added to the specification, the impact of marital status alone becomes negative and statistically insignificant, whereas the impact of years married is significant at the 1% level. Evaluated at sample means, wages rise 6% during the first year of marriage, 4% in the second year, and 3% in the third year. By the fifth year, the total return to time married is 16%; by the tenth year it is 24%. Similar results were obtained when dummy variables for different marital durations were used in lieu of the log duration measure. Any wage differential observed between never married and previously married men appears to be attributable to differences in years married as neither the indicator variable for those previously married nor the measure of years divorced/separated/widowed are statistically significant in the cross-sectional analysi s. Overall, these results are similar to (if not stronger than) those reported elsewhere in the literature.

Panel estimates remove any individual specific selection effects that may bias the cross-section results. These estimates are reported in the last two columns of Table 2, top. Results controlling only for marital status (specification [2a]) indicate that the effect of marriage is approximately equal to the effect of divorce, separation, or widowhood. [14] What this means is that wages rise on first marriage but do not fall on the dissolution of a marriage. If it is faster wage growth during marriage that drives the marital wage differential, this is the result one would expect. Controlling for years married and years divorced/separated/widowed (specification [2b]) reduces the statistical significance of the marital status variables considerably. They are no longer statistically significant individually or jointly at even the twenty percent level. The duration measures are, however, jointly significant at the 10% level. Indeed, in estimates (not shown here) excluding the marital status measures, the coefficie nt to the measure of years married is significant at the 2% level and, at 0.0933, is of approximately the same magnitude as that observed in the cross-section analysis. Given the high degree of collinearity between experience, tenure, and marital duration in these estimates, it is surprising that separate parameters can be estimated with even this much precision. Thus, the panel estimates lend further support to the argument that the marital wage differential is attributable to faster wage growth during marriage.

As a sort of sensitivity test, similar specifications were estimated using a sample that includes those individuals excluded here because of missing information on cohabitation duration. Results from these larger samples (4,039 cross-section and 1,427 panel observations) were quite similar (see Appendix B). Sample selection based on the availability of cohabitation information does not appear to bias the marital wage effects.

The lower part of Table 2 reports coefficient estimates for specifications controlling for both marital and cohabitation effects. The first two columns provide cross-section estimates that may be subject to selection bias; the last two columns provide fixed effects estimates that should control for simple selection bias. In each case, estimates for the specifications that include only indicator variables (1a and 2a) as well as the specification that controls for duration measures (1b and 2b) are presented.

Estimates of specification (1a) are comparable to those reported in Cohen (1999). Both the marital and cohabiting wage differentials are statistically significant with currently married men earning about 22% more and currently cohabiting men earning about 13% more than the base group of men who have never married or cohabited. Neither past marriage nor past cohabitation appears to significantly influence wages. Considering both current and past status measures, although cohabitation appears to have a smaller impact on wages than marriage, the difference is not statistically significant (p-value 0.31).

Adding controls for time spent in each type of relationship (specification [1b]) yields substantially different results. Years married once again proves to be the driving force behind the marital wage differential, but wage growth is actually slower, though not significantly so, during cohabitations. Thus, years married and cohabitation status are the key factors in the cross-sectional analysis. One can in this case reject the hypothesis that marriage and cohabitation have the same impact on wages (p-value of 0.05) but not reject the hypothesis that cohabitation has no impact on wages (p-value of 0.22).

Controlling for individual specific fixed effects in the panel analysis causes the cohabitation effect to become statistically insignificant. The point estimates of cohabitation status become negative, though not significantly so, in both specifications (2a) and (2b), whereas years cohabited enters now with a positive coefficient of approximately the same magnitude as the coefficient to years married. (15) Unfortunately due to large standard errors, one cannot reject either the hypothesis that cohabitation has no impact on wages (p-value of 0.60) or the hypothesis that cohabitation has the same impact on wages as marriage (p-value of 0.40). The substantial differences observed between the cross-section and panel results do suggest, however, that selection plays a key role in the cohabitation but not the marital wage effect.

Several alternative specifications were estimated to gauge the robustness of these results. First, estimates of specifications (1a) and (2a) were obtained using the larger sample (for which time cohabiting is not always available) to determine if the results are sensitive to the sample selection controls. These results (see Appendix B) are virtually identical to those reported in Table 2. Likewise, results using the larger sample and controlling for both cohabitation status and duration, with the addition of dummy variables to denote observations with missing or incomplete cohabitation duration, yield similar results. Restricting the sample to those who report complete cohabitation histories again does not appear to bias the results.

Second, the panel estimates presented in Table 2 force those entering a cohabitation (or marital) relationship to have a wage change that is of the same magnitude but the opposite sign as those leaving such a relationship. Such symmetry may not hold. Results from a less rigid specification, however, support the assumption of symmetry. (16)

Third, several tests were conducted to determine the sensitivity of the results to the duration of the marriage, cohabitation, or relationship in general. Excluding marriages that lasted less than six months, less than one year, or less than two years had no impact on the fit of the model. This makes sense if what is really desired is information on the expected duration of the relationship and few marriages are initiated on the expectation that they will be short-lived.

The analysis of cohabitations supports the results of Winkler (1997) that not all cohabitations are the same; heterogeneity is a problem not entirely addressed in the fixed-effects model. Although long-lasting cohabitations are the exception, not the rule, long-lasting cohabitations are clearly more similar to marriages in their impact on wages. Excluding cohabitations that last for less than one year has no impact on the fit of the model. Excluding cohabitations that last for less than three years increases the [R.sup.2] slightly from 0.4150 to 0.4154 in the cross-sectional specification (1b) and more significantly from 0.0924 to 0.0945 in the panel specification (2b). Treating cohabitations that last more than three years as if they were marriages yields similar results. Indeed, after some point, long-lasting cohabitations become common-law marriages, and some individuals in such relationships replied that they were married when asked their marital status.

The finding in the panel analysis that the impact of years cohabited is approximately equal in magnitude to the impact of years married may also be attributed to those in long-term cohabiting relationships. In the panel analysis, the coefficient to years cohabited is based off the change in years cohabiting between interviews. While 41% of those cohabiting between interviews cohabit for less than one year, these individuals account for only 13% of the weighted total years cohabited. The 19% involved in long-term cohabitations (longer than three years) by comparison account for about 52% of the time cohabited.

Finally, distinguishing between cohabitations that did and did not end in marriage did not improve the fit of the cross-section model but did improve the fit of the panel model, with time spent in cohabitations that did not end in marriage having an effect of approximately the same magnitude as time spent in marriages, possibly because individuals who maintained long-term cohabitating relationships use cohabitation as an alternative to marriage. Overall, it would appear to be the case that men who use cohabitation as a long-term alternative to marriage experience wage growth comparable to that experienced by married men, and those for whom cohabitation is a short-lived experience do not.

VI. CONCLUSION

The purpose of this analysis has been to examine the nature of reported marital and cohabiting wage differentials for men. Estimates of marital and cohabiting wage differentials for white men are obtained using a relatively new data set (the NSFH) and found to match those reported elsewhere in the literature. The observed marital wage differential is robust to fixed individual specific selection effects and appears primarily attributable to faster wage growth during marriage. The observed wage differential for cohabiting men, however, disappears in the face of individual specific selection effects. Only men in long-term cohabiting relationships appear to experience any substantial wage gains, and these gains appear to match those of married men quite closely, with wage growth not selection explaining the differential.

These findings suggest that men do not receive a wage benefit from all joint household operations, but instead that marriage has a nearly unique effect on men's productivity. As all joint households should gain from economies of scale in household production, the evidence does not support this explanation of the marital wage differential. An explanation based on specialization within the household may still be worth exploring, given the differences between marital and cohabiting relationships. Evidence from a study by South and Spitze (1994) suggests that specialization is much greater in married households, as married women spend substantially more time than cohabiting women on housework. However, work by Hersch and Stratton (2000) finds that controlling for time spent on housework has no influence on the male marital wage differential. Thus, an analysis of how marriage but not short-term cohabitation influences productivity in others ways--such as effort on the job or job training opportunities and choices --might be of even greater merit in identifying the mechanism underlying the marital wage differential.

TABLE 1

Sample Statistics (NSFH)

 Cross-Section Data Panel Data
 1st Wave 2nd Wave 1st Wave
 Weighted Weighted Weighted
Variable Name Mean Mean Mean

Real wage (1992$/hour) $15.82 $16.92 $15.85
Log wage 2.60 2.65 2.62
Married 70.4% 76.9% 73.2%
Divorced/separated/widowed 8.1% 10.8% 8.0%
Never married 21.5% 12.2% 18.8%
Cohabiting 3.9% 5.6% 3.3%
Has cohabited 21.6% 29.4% 25.6%
Years married 12.94 15.51 11.70
Years divorced/ 0.88 1.33 0.99
separated/widowed
Years cohabited 0.46 0.75 0.51
Age 37.55 41.32 36.28
Education 13.70 14.08 13.88
Years of experience 17.55 21.10 16.21
Years of tenure 10.43 9.44 8.19
Years not employed 1.06 0.96 0.97
Residence in SMSA 76.2% 80.0% 75.1%
Residence in South 30.9% 31.1% 31.4%
Active in union 11.9% 18.6% 12.1%
Children present 53.6% 50.8% 60.7%
Number of observations 2,044 1,539 1,358

 Panel Data
 2nd Wave
 Weighted
Variable Name Mean

Real wage (1992$/hour) $16.63
Log wage 2.65
Married 77.5%
Divorced/separated/widowed 11.9%
Never married 10.6%
Cohabiting 5.2%
Has cohabited 32.2%
Years married 16.03
Years divorced/ 1.59
separated/widowed
Years cohabited 0.77
Age 42.02
Education 13.92
Years of experience 21.85
Years of tenure 9.76
Years not employed 1.02
Residence in SMSA 80.1%
Residence in South 31.6%
Active in union 17.5%
Children present 52.4%
Number of observations 1,358
TABLE 2

Impact of Marriage and Cohabitation on Log Wages (NSFH)

 Cross-Section Analysis
Variable (1a) (1b)

Marital Effects Alone

 Married 0.1696 (***) -0.0036
 (0.0432) (0.0652)
 Divorced/Separated/Widowed 0.0572 -0.0571
 (0.0503) (0.0763)
 Log(Years Married +1) 0.0912 (***)
 (0.0261)
 Log(Years Divorced/ -0.0198
 Separated/Widowed +1) (0.0233)
 [R.sup.2] 0.4077 0.4129

Marital and Cohabitation
 Effects

 Married 0.2031 (***) 0.0285
 (0.0436) (0.0654)
 Divorced/Separated/Widowed 0.0541 -0.0627
 (0.0489) (0.0760)
 Log(Years Married +1) 0.0881 (***)
 (0.0267)
 Log(Years Divorced/ -0.0208
 Separated/Widowed +1) (0.0249)
 Cohabiting 0.1235 (*) 0.1489 (*)
 (0.0640) (0.0818)
 Has Cohabited -0.0253 0.0161
 (0.0261) (0.0423)
 Log(Years Cohabited +1) -0.0142
 (0.0362)
 [R.sup.2] 0.4102 0.4150

 Panel Analysis
Variable (2a) (2b)

Marital Effects Alone

 Married 0.0909 0.0358
 (0.0614) (0.0823)
 Divorced/Separated/Widowed 0.1389 (*) 0.1119
 (0.0707) (0.0921)
 Log(Years Married +1) 0.0699
 (0.0543)
 Log(Years Divorced/ -0.0693
 Separated/Widowed +1) (0.0491)
 [R.sup.2] 0.0840 0.0904

Marital and Cohabitation
 Effects

 Married 0.1147 (*) 0.0294
 (0.0614) (0.0770)
 Divorced/Separated/Widowed 0.1540 (**) 0.1067
 (0.0740) (0.0899)
 Log(Years Married +1) 0.0864 (*)
 (0.0517)
 Log(Years Divorced/ -0.0761
 Separated/Widowed +1) (0.0525)
 Cohabiting -0.0276 -0.0416
 (0.0864) (0.0865)
 Has Cohabited -0.0629 -0.0594
 (0.0704) (0.0741)
 Log(Years Cohabited +1) 0.0874
 (0.0665)
 [R.sup.2] 0.0851 0.0924

Notes: Standard errors are reported in parentheses below coefficient
values. Asterisks indicate the level of statistical significance using a
two- tailed test: (***)for 1%, (**)for 5%, (*)for 10%. Each equation
includes quadratic controls for experience and tenure, a linear measure
of education and time spent not working, and indicators for residence in
the South and in an SMSA, for union activity, and for the presence of
children in the household. The cross-section analysis also includes an
intercept, 7 occupation dummies, 11 industry dummies, a dummy variable
for observations from the second wave of the survey, and variables
interacting the survey dummy with SMSA, union activity, and tenure. The
panel analysis permits residence in an SMSA and union activity to have a
differential effect in each wave. All results are weighted to account
for the double sampling of cohabiting and recently married persons.

(*) I would like to thank seminar participants at the 1996 ASSA meetings for comments on an early draft, as well as more recent seminar participants at Virginia Commonwealth University and William and Mary and the anonymous referees. Any remaining errors are my own.

(1.) This approach controls for selection based on wage level. Researchers Korenman and Neumark (1991) and Gray (1997) have also explored the possibility of selection based on wage growth, but to no avail.

(2.) Unmarried men who expect to marry should also have a steeper experience-earnings profile, but they may face more significant time constraints because of the need to perform extra household production activities.

(3.) The theoretical model of Becker (1981, 1985) points out the importance of the duration of the relationship. Willis and Michael (1994) also contribute to the discussion.

(4.) Marriages are assumed to end on the date of separation rather than divorce. If married men earn more than single men because of specialization within the household, it is the date at which specialization is no longer possible that is relevant.

(5.) Measurement error could also cause parameters that are significant in cross-section analysis to be insignificant in panel analysis. The variables of particular interest here, marital and cohabitation status, are however believed to be measured fairly accurately, particularly during the relatively short period between interviews. Interviewers followed up in the event of inconsistencies with respect to date sequences and spouse/partner interviews could be used to double check information.

(6.) Recognizing that cohabitations would likely be underreported, the NSFH was designed to "maximize the completeness of reporting" (Bumpass and Sweet, 1989, 617). It is difficult to gauge precisely the degree to which the survey is successful, as cohabiting couples were not distinguished from roommates or boarders in most national surveys until quite recently-1990 for the census, 1995 for the Current Population Survey (CPS). However, the NSFH reports higher current cohabitation rates than the CPS (5.6% for the 1992-1994 wave of the NSFH as compared with 2.3% using the 1995-1997 March CPS-similarly restricted to white men of working age who provide wage information) and a slower growth rate in cohabitations over time than the census/CPS reports, suggesting that the retrospective histories are reasonably inclusive.

(7.) A Chow test failed to reject at a 10% significance level, pooling of the data for the specifications reported in Table 2.

(8.) All dollar figures are converted to constant 1992 dollars using monthly consumer price index measures. Individuals reporting real wage values below $2.50 or above $100 an hour were excluded from the analysis. Sample attrition for reasons of missing wage in the NSFH is not unusual. Data from the March 1988 CPS indicate that 13.8% of white men between the ages of 18 and 64 who were not in the military or primarily engaged in school were not employed, and 13.8% of those that were queried about their wages either failed to provide the information or reported a real wage below $2.50 or above $100 an hour in 1992 dollars.

(9.) Further information on both samples is available on request from the author.

(10.) In the second wave and the panel data samples, tenure is measured in the usual manner--as time with current employer. However, for almost 25% of the first wave cross-section, tenure information per-se is unavailable and the duration of the most recent employment spell is used as a proxy. For at least a quarter of this sample, multiple spells are reported. Often, the date at which an employment spell is reported to end is identical to the date at which an employment spell is reported to begin, indicating that the respondent was providing information on job changes (as desired for setting tenure) rather than simply information on joblessness. However, for some first survey respondents the tenure measure is an overestimate.

(11.) SMSA status is determined using 1980 census definitions in the first wave and 1990 census definitions in the second wave. Examination of individuals who did not move between surveys suggests substantial variation in the coding of this variable. Union activity in the first wave is an indicator of attendance at union functions. In the second wave, the question was amended to identify activity in any professional organization and the fraction reporting such activity jumped 50%. Industry and occupation codes technically also differ between the two interviews, but little change was observed in practice for those who did not change employer.

(12.) Korenman and Neumark (1991) controlled for experience, and Loh (1996) controlled for tenure. Inclusion of both variables may have been precluded due to multicollinearity problems. Failure to control for tenure could produce an upwardly biased estimate of the return to years married if tenure and years married are positively correlated.

(13.) Calculated as the maximum of zero and age minus education minus experience minus seven.

(14.) A test of the hypothesis that marriage and divorce/separation/widowhood have the same impact on wages yields a p-value of 0.34.

(15.) Further analysis suggests that it is primarily the duration of the current marriage that matters. Controls for thc duration of the current cohabitation were statistically insignificant.

(16.) A test for symmetry for both marital status and cohabitation status in specification (2a) yields a p-value of 0.13. A joint test in specification (2b) yields a p-value of 0.60.

REFERENCES

Bartlett, R. L., and C. Callahan III. "Wage Determination and Marital Status: Another Look." Industrial Relations, 23(1), 1984, 90-96.

Becker, G. S. A Treatise on the Family. Cambridge, MA: Harvard University Press, 1981, rpt. 1991.

-----. "Human Capital, Effort, and the Sexual Division of Labor." Journal of Labor Economics, 3(1, Part 2), 1985, S33-S58.

Bumpass, L. L., and J. A. Sweet. "National Estimates of Cohabitation." Demography, 26(4), 1989, 615-25.

Cohen, P. N. "Racial-Ethnic and Gender Differences in Returns to Cohabitation and Marriage: Evidence from the Current Population Survey." U.S. Bureau of the Census, Washington, DC. Population Division Working Paper no. 35, 1999.

Daniel, K. "Does Marriage Make Men More Productive?" Photocopy, University of Chicago, 1991.

Gray, J. S. "The Fall in Men's Return to Marriage: Declining Productivity Effects or Changing Selection?" Journal of Human Resources, 32(3), 1997, 481-504.

Hersch, J., and L. S. Stratton. "Household Specialization and the Male Marriage Wage Premium." Industrial and Labor Relations Review, 54(1), 2000, 78-94.

Hill, M. S. "The Wage Effects of Marital Status and Children." Journal of Human Resources, 14(4), 1979, 579-94.

Kenny, L. W. "The Accumulation of Human Capital during Marriage by Males." Economic Inquiry, 21(2), 1983, 223-31.

Korenman, S., and D. Neumark. "Does Marriage Really Make Men More Productive?" Journal of Human Resources, 26(2), 1991, 282-307.

Loh, E. S. "Productivity Differences and the Marriage Wage Premium for White Males." Journal of Human Resources, 31(3), 1996, 566-89.

Nakosteen, R. A., and M. A. Zimmer. "Marital Status and Earnings of Young Men." Journal of Human Resources, 22(2), 1987, 248-68.

Rindfuss, R. R., and A. VandenHeuvel. "Cohabitation: A Precursor to Marriage or an Alternative to Being

Single?" Population and Development Review, 16(4), 1990, 703-25.

Schoen, R., and R. M. Weinick. "Partner Choice in Marriages and Cohabitations." Journal of Marriage and the Family, 55(2), 1993, 408-14.

South, S. J., and G. Spitze. "Housework in Marital and Nonmarital Households." American Sociological Review, 59(3), 1994, 327-47.

Willis, R. J., and R. T. Michael. "Innovation in Family Formation: Evidence on Cohabitation in the United States," in The Family, the Market and the State in Ageing Societies, edited by J. Ermisch and N. Ogawa Oxford: Clarendon Press, 1994, 9-45.

Winkler, A. E. "Economic Decision-Making by Cohabitors: Findings Regarding Income Pooling." Applied Economics, 29(8), 1997, 1079-90.

RELATED ARTICLE: ABBREVIATIONS

CPS: Current Population Survey

NLSY: National Longitudinal Survey of Youth

NSFH: National Survey of Families and Households

SMSA: Standard Metropolitan Statistical Area

APPENDIX TABLE A1

Other Parameter Estimates from Log Wage Equations (NSFH),
Cross-Sectional Analysis


 Marital Effects Alone
Variable (1a) (1b)

Intercept 0.9774 (***) 1.0817 (***)
 (0.0977) (0.0995)
Education 0.0746 (***) 0.0716 (***)
 (0.0070) (0.0069)
Experience 0.0259 (***) 0.0184 (***)
 (0.0055) (0.0057)
Experience squared -0.0004 (***) -0.0004 (***)
 (0.0001) (0.0001)
Tenure 0.0181 (***) 0.0195 (***)
 (0.0044) (0.0044)
Tenure squared -0.0003 (**) -0.0003 (**)
 (0.0001) (0.0001)
Tenure * Wave 2 0.0076 0.0063
 (0.0070) (0.0071)
Tenure Squared * Wave 2 -0.0002 -0.0002
 (0.0002) (0.0002)
Years Not Employed -0.0090 (*) -0.0140 (***)
 (0.0051) (0.0053)
Residence in SMSA 0.2058 (***) 0.2070 (***)
 (0.0271) (0.0270)
SMSA * Wave 2 -0.1004 (**) -0.0924 (**)
 (0.0391) (0.0390)
Residence in South -0.0583 (**) -0.0619 (**)
 (0.0272) (0.0270)
Active in Union 0.0967 (***) 0.0960 (***)
 (0.0297) (0.0300)
Union * Wave 2 0.0164 0.0143
 (0.0471) (0.0472)
Children Present 0.0044 -0.0152
 (0.0302) (0.0308)
Wave 2 Dummy -0.0200 -0.0096
 (0.0506) (0.0507)

Industries

 Agriculture, forestry, -0.1051 -0.1121
 fisheries (0.1012) (0.0981)
 Mining 0.2613 (***) 0.2468 (***)
 (0.0936) (0.0916)
 Construction 0.0799 0.0802
 (0.0510) (0.0509)
 Public utilities, 0.0348 0.0334
 transportation, and (0.0445) (0.0438)
 communication
 Wholesale trade -0.1158 (*) -0.1168 (*)
 (0.0601) (0.0598)
 Retail trade -0.2218 (***) -0.2179 (***)
 (0.0443) (0.0440)
 Finance, insurance, 0.0634 0.0736
 real estate (0.0740) (0.0735)
 Business and professional -0.0640 -0.0638 (*)
 services (0.0389) (0.0387)
 Personal and -0.1954 (**) -0.1762 (**)
 entertainment services (0.0819) (0.0830)
 Public administration 0.0003 0.0036
 (0.0399) (0.0398)
 Other industries 0.2092 (*) 0.2317 (*)
 (0.1222) (0.1304)
Occupations
 Managerial/professional 0.0734 (*) 0.0775 (*)
 (0.0417) (0.0417)
 Technical/sales 0.0338 0.0342
 (0.0484) (0.0481)
 Administrative support -0.1252 (***) -0.1214 (**)
 (0.0480) (0.0482)
 Service -0.2064 (***) -0.1954 (***)
 (0.0495) (0.0490)
 Farming, forestry, fishing -0.1632 -0.1568
 (0.1305) (0.1264)
 Operators, laborers -0.1572 (***) -0.1479 (***)
 (0.0338) (0.0334)
 Other occupations -0.1072 -0.1193
 (0.1569) (0.1620)
 [R.sup.2] 0.4077 0.4129

 Marital and
 Cohabitation Effects
Variable (1a) (1b)

Intercept 0.9665 (***) 1.0556 (***)
 (0.0990) (0.1006)
Education 0.0742 (***) 0.0714 (***)
 (0.0069) (0.0069)
Experience 0.0261 (***) 0.0193 (***)
 (0.0055) (0.0059)
Experience squared -0.0004 (***) -0.0004 (***)
 (0.0001) (0.0001)
Tenure 0.0177 (***) 0.0189 (***)
 (0.0044) (0.0044)
Tenure squared -0.0003 (**) -0.0003 (**)
 (0.0001) (0.0001)
Tenure * Wave 2 -0.0002 -0.0002
 (0.0002) (0.0002)
Tenure Squared * Wave 2 -0.0002 -0.0002
 (0.0002) (0.0002)
Years Not Employed -0.0095 (*) -0.0137 (***)
 (0.0051) (0.0053)
Residence in SMSA 0.2066 (***) 0.2081 (***)
 (0.0271) (0.0270)
SMSA * Wave 2 -0.1030 (***) -0.0945 (**)
 (0.0391) (0.0389)
Residence in South -0.0587 (**) -0.0604 (**)
 (0.0271) (0.0270)
Active in Union 0.1004 (***) 0.1003 (***)
 (0.0301) (0.0302)
Union * Wave 2 0.0128 0.0110
 (0.0473) (0.0473)
Children Present -0.0032 -0.0201
 (0.0301) (0.0308)
Wave 2 Dummy -0.0219 -0.0161
 (0.0508) (0.0510)

Industries

 Agriculture, forestry, -0.1110 -0.1170
 fisheries (0.1004) (0.0989)
 Mining 0.2532 (***) 0.2451 (***)
 (0.0949) (0.0931)
 Construction 0.0761 0.0751
 (0.0501) (0.0504)
 Public utilities, 0.0342 0.0312
 transportation, and (0.0438) (0.0436)
 communication
 Wholesale trade -0.1125 (*) -0.1161 (*)
 (0.0607) (0.0605)
 Retail trade -0.2215 (***) -0.2197 (***)
 (0.0436) (0.0432)
 Finance, insurance, 0.0624 0.0743
 real estate (0.0734) (0.0729)
 Business and professional -0.0645 (*) -0.0647 (*)
 services (0.0389) (0.0386)
 Personal and -0.1850 (**) -0.1723 (**)
 entertainment services (0.0820) (0.0831)
 Public administration -0.0017 0.0013
 (0.0399) (0.0397)
 Other industries 0.2240 (*) 0.2477 (*)
 (0.1290) (0.1357)
Occupations
 Managerial/professional 0.0762 (*) 0.0793 (*)
 (0.0414) (0.0415)
 Technical/sales 0.0361 0.0359
 (0.0480) (0.0478)
 Administrative support -0.1197 (**) -0.1169 (**)
 (0.0478) (0.0479)
 Service -0.1965 (***) -0.1874 (***)
 (0.0489) (0.0487)
 Farming, forestry, fishing -0.1567 -0.1507
 (0.1298) (0.1274)
 Operators, laborers -0.1546 (***) -0.1465 (***)
 (0.0335) (0.0333)
 Other occupations -0.1179 -0.1297
 (0.1611) (0.1657)
 [R.sup.2] 0.4102 0.4150

Notes: Standard errors are reported in parentheses below coefficient
values. The excluded industry is manufacturing. The excluded occupation
is craft workers. Asterisks indicate the level of statistical
significance using a two-tailed test:

(***)for 1%

(**)for 5%

(*)for 10%.
APPENDIX TABLE A2

Other Parameter Estimates from Log Wage Equations (NSFH), Panel Analysis


 Marital Effects Alone
Variable (2a) (2b)

Education 0.0516 0.0547
 (0.0438) (0.0458)
Experience 0.0343 (***) 0.0293 (**)
 (0.0096) (0.0124)
Experience Squared -0.0005 (***) -0.0005 (**)
 (0.0002) (0.0002)
Tenure 0.0133 (**) 0.0134
 (0.0002) (0.0002)
Tenure Squared -0.0003 -0.0003
 (0.0002) (0.0002)
Years Not Employed -0.0744 (*) -0.0747 (*)
 (0.0380) (0.0385)
Residence in SMSA 0.0259 0.0228
 (0.0513) (0.0515)
SMSA * Wave 2 -0.0499 -0.0516
 (0.0514) (0.0517)
Residence in South -0.0804 -0.0837
 (0.0754) (0.0740)
Active in Union 0.0771 (*) 0.0747 (**)
 (0.0399) (0.0375)
Union * Wave 2 0.0788 (**) 0.0807 (**)
 (0.0340) (0.0342)
Children Present -0.0248 -0.0377
 (0.0321) (0.0315)
[R.sup.2] 0.0840 0.0904

 Marital and
 Cohabitation Effects
Variable (2a) (2b)

Education 0.0488 0.0556
 (0.0436) (0.0464)
Experience 0.0357 (***) 0.0258 (**)
 (0.0095) (0.0122)
Experience Squared -0.0006 (***) -0.0005 (**)
 (0.0002) (0.0002)
Tenure 0.0133 (**) 0.0133 (**)
 (0.0062) (0.0062)
Tenure Squared -0.0003 -0.0003
 (0.0002) (0.0002)
Years Not Employed -0.0744 (*) -0.0761 (**)
 (0.0379) (0.0384)
Residence in SMSA 0.0243 0.0232
 (0.0513) (0.0516)
SMSA * Wave 2 -0.0516 -0.0497
 (0.0515) (0.0517)
Residence in South -0.0812 -0.0911
 (0.0756) (0.0747)
Active in Union 0.0748 (**) 0.0716 (**)
 (0.0387) (0.0363)
Union * Wave 2 0.0759 (**) 0.0805
 (0.0339) (0.0341)
Children Present -0.0292 -0.0430
 (0.0312) (0.0304)
[R.sup.2] 0.0851 0.0924

Notes: Standard errors are reported in parentheses below coefficient
values. Asterisks indicate the level of statistical significance using a
two-tailed test:

(***)for 1%

(**)for 5%

(*)for 10%.
APPENDIX TABLE B1

Impact of Marriage and Cohabitation on Log Wages (NSFH)

 Cross-Section Analysis
Variable (1a)

Marital Effects Alone
 Married 0.1586 (**)
 (0.0400)
 Divorced/Separated/Widowed 0.0688
 (0.0454)
 Log(Years Married +1)

 Log(Years Divorced/Separated/
 Widowed +1)
 [R.sup.2] 0.4102

Marital and Cohabitation Effects
 Married 0.1840 (***)
 (0.0406)
 Divorced/Separated/Widowed 0.0670
 (0.0449)
 Cohabiting 0.0959 (*)
 (0.0569)
 Has Cohabited -0.0170
 (0.0239)
[R.sup.2] 0.4118

 Cross-Section Panel Analysis
 Analysis
Variable (1b) (2a)

Marital Effects Alone
 Married 0.0228 0.1130 (**)
 (0.0583) (0.0569)
 Divorced/Separated/Widowed -0.0095 0.1556 (**)
 (0.0657) (0.0644)
 Log(Ycars Married +1) 0.0738 (***)
 (0.0230)
 Log(Ycars Divorced/Separated/ -0.0229
 Widowed +1) (0.0189)
 [R.sup.2] 0.4145 0.0885

Marital and Cohabitation Effects
 Married 0.1233 (**)
 (0.0562)
 Divorced/Separated/Widowed 0.1650 (**)
 (0.0669)
 Cohabiting -0.0305
 (0.0797)
 Has Cohabited -0.0356
 (0.0670)
[R.sup.2] 0.0889

 Panel
 Analysis
Variable (2b)

Marital Effects Alone
 Married 0.0488
 (0.0738)
 Divorced/Separated/Widowed 0.1138
 (0.0805)
 Log(Ycars Married +1) 0.0792
 (0.0505)
 Log(Ycars Divorced/Separated/ -0.0615
 Widowed +1) (0.0440)
 [R.sup.2] 0.0951

Marital and Cohabitation Effects
 Married

 Divorced/Separated/Widowed

 Cohabiting

 Has Cohabited

[R.sup.2]

Notes: Standard errors are reported in parentheses below coefficient
values. Asterisks indicate the level of statistical significance using a
two- tailed test:

(***)for 1%,

(**)for 5%,

(*)for 10%.

The cross-section sample consists of 4,039 idividuals, the panel sample
of 1,427 individuals. Each differs from the sample employed in the main
text in that it includes those with missing cohabitation histories. Each
equation includes quadratic controls for experience and tenure, a
linear measure of education and time spent not working, and indicators
for residence in the South and in an SMSA, for union activity, and for
the presence of children in the household. The cross- section analysis
also includes an intercept, 7 occupation dummies, 11 industry dummies, a
dummy variable for observations from the second wave of the survey, and
variables interacting the survey dummy with SMSA, union activity, and
tenure. The panel analysis permits residence in an SMSA and union
activity to have a differential effect in each wave. All results are
weighted to account for the double sampling of cohabiting and recently
married persons.

Stratton: Associate Professor, Department of Economics, P.O. Box 844000, 1015 Floyd Ave., Virginia Commonwealth University, Richmond, VA 232844000. Phone 1-804-828-7141, Fax 1-804-828-1719, E-mail Isstratt@vcu.edu