«Chinhui Juhn University of Houston and NBER cjuhn Kristin McCue U.S. Census Bureau kristin.mccue September 2011 Abstract: Using ...»
Turning to women with a high school degree or less in the top panel, we see that contrary to the pattern observed in earlier decades, never-married women earn significantly less than married women in 2010. The reversal is less pronounced among the more educated groups. Among college women, “never married” women still earn more than married women, $45,807 vs.
$40,002, although there has been considerable catch-up over the four decades. Examining inequality of women’s earnings across demographic groups, we see that the earnings gap across marital status has essentially closed by 2010. On the other hand, the earnings gap between educated and less educated women has widened across all marital status categories.
Table 6 contrasts total earnings where we have included husband’s earnings for married women. For never married single women, total earnings are simply their own earnings. Again We re-code those women who report negative self-employment income to zero earnings in this exercise.
we examine women 30-60 years old to minimize the age differences between single and married women. Table 6 highlights the extent of inequality that exists across marital status and education categories. For example, while married college-educated women were about on par with “never married” less educated women (high school grad or dropout) in terms of their own earnings in 1970, when we add in husband’s earnings, total family earnings of college-educated married women were roughly 5.2 times (74.9/14.5) that of less educated “never married” women.
Moreover, the inequality across these two groups has increased. In 2010, the ratio of family earnings of college-educated married women was roughly 8.8 times (113.2/12.9) that of less educated “never married” women.
IV. Women’s Lifetime Earned Income In the previous section we investigated readily available cross-sectional data (CPS) to describe changes in women’s marriage and earnings. In this section, we use the Survey of Income and Program Participation matched to Social Security earnings records (SIPP-SSA) to construct long-term measures of earnings. The main advantage of these data is that they provide a long panel data set of women’s earnings (from Social Security records) and marital status (from SIPP marital histories). There are multiple advantages to panel data for our purposes.
First, while we can estimate average lifetime earnings by birth cohort using repeated crosssections, we cannot examine how much it varies across people. While the cross-sectional variance observed at a particularly point, say age 40, may be a proxy for the variance of lifetime earnings, it is likely to be poor proxy since transitory earnings variance is a significant component of the overall variance for women.
We use 1951-2006 data on earnings from this panel to examine patterns in the value of accumulated earnings over women’s lifetimes. The estimates discount future earnings at a rate of 2 percent, and the value of earnings is deflated by the Personal Consumption Expenditures index to put them in 2006 dollars. For women who did not attend any college, we accumulate earnings starting at age 18, while we use 20 as a starting age for those with some college, and 22 for those with at least a college degree.
For younger cohorts who are likely to still be in the labor force for years after 2006, we base projections past 2006 on the shape of earnings profiles at older ages for earlier cohorts. We estimate these profiles by running education-group specific fixed-effects regressions of the log of the discounted value of accumulated earnings on controls for experience, marital status, presence of children, birth cohort, and interactions among these variables. Details of the methodology are described in the appendix. In the current draft we do not include measures of the uncertainty associated with our estimates. To do so we need to capture both the uncertainty arising through the use of a sample and that added on by our imputation of marital status and earnings in years that we do not observe. We plan to use multiple imputation methods to estimate standard errors and confidence intervals in our next draft.
Clearly, projections to age 60 for recent cohorts are at best suggestive of what might happen, so we present estimates at age 45 as well. For the age 45 projections, the information on earnings comes only from actual earnings for the first five birth cohorts, and comes primarily from observed earnings for the sixth. Members of the most recent birth cohort were ages 36-40 in 2006, so even for the youngest and most educated group, only the last third of their earnings are projected, and the shape of that projection is based on earnings profiles for the 1961-1965 cohort. These forecasts for younger cohorts clearly may differ substantially from eventual outcomes, but we think they are useful as an illustration of what one might expect to see if current trends continue.
Table 7 presents the age 60 estimates of women’s own present discounted value of earnings. In all cases, accumulated earnings are greater for more educated women, with collegeeducated women earning roughly 60% more than those with some college and 150% more than those with a high school degree or less in the earliest cohorts. Based on these projections, the education differential between college grads and those with some college grew modestly across these cohorts, but both groups had faster growth than did those with no college.
Table 8 presents analogous results using the age 45 projections. Comparing Table 8 to Table 7, the substantial growth across birth cohorts is evident even using earnings to age 45. But the pattern of much faster growth in lifetime earnings to age 60 among more educated women is clearly quite dependent on the accuracy of our projections, as we see very modest divergence between education groups as of age 45.
We now turn to measures of household earnings, which we define as women’s own earnings, plus the earnings of their spouse when they are married. We use SIPP marital histories to determine marital status up to the end of the SIPP panel that a woman was in, and then assign a probability of being married for years after that. The probability of being married is based on hazard models of the risk of a marriage ending for women who are married when we last see them. For those not married when last observed, we instead use hazard models of the risk of a spell of non-marriage ending through marriage. These estimates condition on age at marriage, the number of times a woman has been married, and on education and birth cohort. Details are in the appendix.
For the oldest cohorts, roughly 13% of years to age 60 have a forecast of whether a woman is married rather than actual marital status, but that rises steadily across cohorts and reaches roughly 70% for the 1966-1970 cohort. The age 45 profiles have almost no imputed marital status for the first three cohorts, while the share grows from about 5% for the 1951-1955 cohort to about 30% for the most recent cohort.
Table 9 gives age 60 projections of household earnings. We see that the discounted value of household earnings has also grown across birth cohorts, and has grown more for more educated women. Figure 1 presents these numbers in graphical form. The figure (along with Table 9) shows the extent of rising inequality across education categories, taking into account both changes in own earnings, marriage probability, and spouse earnings. For the 1936-40 cohort, college-educated women had discounted lifetime household earnings approximately 2.2 times those of women with high school degree or less. For the 1966-70 cohort, college-educated women had 4.3 times the household earnings of less educated women. Interestingly, the gap has not grown as fast in percentage terms as when we consider only their own earnings, reflecting both less time spent married among more recent cohorts and faster growth in earnings for married women than for married men. These same general patterns appear in Table 10, which gives the projections to age 45.
In Table 11, we split our sample into women with at least one marriage that lasts at least 10 years versus those who never marry or are married only for short periods. We are interested in investigating differences between women for whom spousal earnings are an important source of income versus those who rely primarily on their own earnings. Using 10 years with one spouse as a cutoff is in part motivated by current social security benefit rules under which women may receive spousal benefits based on the income of a current or former spouse, provided the marriage lasted at least 10 years.
The first two columns of Table 11 give own and household lifetime earnings for women with at least one 10 year marriage. The third column gives the share of women who have lifetime earnings that are at least half their spouse’s lifetime earnings, to try to capture how important married women’s own earnings are likely to be relative to husband’s earnings. In all cases, women with no college are less likely to have discounted lifetime earnings that are at least half of their husbands’ earnings, and for this group there is no clear trend in this share. Those with some college had lower shares than women with college degrees in the earlier cohorts but have very similar probabilities to those with a college degree in the most recent cohorts. This pattern also shows up in the estimates in Table 12 estimates based on the age 45 projections.
Note that the same group of women are categorized as having a 10+ year marriage spell in Tables 11 and 12, as it is based on any marriages (or projected marriages) to age 60. The difference between the two tables is in how many years of earnings are used.
The some-college group has the greatest increase over time and the probabilities look very similar to those for college graduates in the last two cohorts. The age 45 projections show lower shares of women having earnings over half of their husbands, likely because years in which their household included young children would account for a larger share of years to age 45 than to age 60.
The fourth column of both tables gives own earnings for women who either did not marry or who had only relatively short marriage spells. For all three education groups, this group of women has higher earnings in the early birth cohorts, but the differential narrows. For those without a college degree, the difference in lifetime earnings associated with marital status is largely gone for the most recent cohort. For college grads, a modest differential remains. These trends are illustrated in Figure 2.
As we found in annual measures in Table 5, there is substantial and growing inequality between the lifetime resources available to married, well-educated women and single women with no college. For example, in the age 45 projections (Table 12) the 1941-1955 birth cohort of college-educated women with at least one 10-year marriage spell had about 1.6 times as much in lifetime earnings as did women with no college and no long marriage spells. But differences between these two groups in household lifetime earnings were much more dramatic. For that cohort, the households of college-educated/10+ year marriage spell women had 4.6 times as much household earnings as those of less educated/no-10-year-spell women. Moreover, the inequality across these two groups has increased so that for the most recent cohort, that ratio has grown to 6.4. The differences based on the age 60 projections are more dramatic (though less reliable)—according to these projections, the ratio grew from about 5.5 to almost 12 across these cohorts.
Over the past four decades the model of a traditional family with a working husband and nonworking wife has become less common. Married women’s employment rates have increased dramatically at the same time that marriage rates have fallen, especially among the less educated. In this paper we examine the impact of these developments on women’s lifetime earnings in terms of both their own earnings as well as household earnings including spouses’ earnings. While married and single women may have always been similar in terms of “potential” earnings, we find that the married women have caught up in terms of actual earnings. While the earnings gap across marital status has narrowed, inequality across education groups has increased, reflecting the general trend towards rising inequality across skill groups. Collegeeducated women as well as women with some college education experienced more growth in lifetime earnings compared to women with high school degree or less.