«Chinhui Juhn University of Houston and NBER cjuhn Kristin McCue U.S. Census Bureau kristin.mccue September 2011 Abstract: Using ...»
When we take into account husbands’ earnings, the level of inequality across education groups is even higher although the increase is not as dramatic as when considering women’s own earnings. We find that the most dramatic increase in inequality is across education and marital status. While in older cohorts they had a modest advantage, more recent cohorts of collegeeducated women with long marriage spells have cumulative earnings about 5 times those of less educated women who spent little or no time in marriage. These developments suggest that going forward, less educated women will face considerable challenges in terms of retirement security.
On one hand, these women have spent fewer years in marriage relative to their more educated counterparts and are less likely to qualify for spousal benefits. On the other hand, their own earnings have not kept up to compensate for the loss of spousal benefits.
Our sample of individuals is drawn from respondents to the 1990-1993, 1996, 2001, and 2004 SIPP panels who provided the information needed to validate matches to Social Security Administration (SSA) earnings records. Individuals had to be at least 15 years old at the time of their second SIPP interview to be eligible for inclusion in the matched data.4 For matched individuals, we have annual earnings for 1951-2006 based on annual summaries of earnings on jobs recorded in SSA’s Master Earnings File. The primary source of the earnings information is W-2 records, but self-employment earnings are also included. We include employees’ contributions to deferred compensation plans as part of our earnings measure. We obtain marital histories, educational attainment, and women’s fertility histories from the SIPP. Age and gender are based on combined information from the SIPP and SSA sources, with the administrative data used to fill in missing values.
We use these data to look at cohorts born between 1936 and 1970, following their earnings over the period 1954-2006 during years in which they were aged 18-60. To determine marital status at a point in time, we use the marital history information collected in the relevant SIPP panel with some additional updates from changes in later waves of that panel. This gives us the information we need on marital status for years leading up to or during the SIPP panel, but not for the years after the panel is over.
We would like to project women’s own earnings and any earnings that spouses contribute to household income out to age 60. Doing so requires that we impute information for some ages for The SIPP is a series of short panel surveys in which respondents are surveyed every 4 months to collect detailed information on household members’ income, employment and program participation over the previous months. The surveys also periodically collect detailed information on the demographic characteristics and relationships of
household members. Panels have ranged in length from about 2 to 4 years. More detail on the SIPP is available at:
many sample members. We have developed methods to do so here, but plan to investigate the sensitivity of our results to alternative methods in future drafts. We describe our methods of imputing earnings and marital status below.
A complication in examining the earnings of spouses is that we only have information on both members of couples identified during their SIPP panel. For a woman who divorced before the start of the SIPP panel, we have information on which prior years she was married, but we cannot, for example, look at spousal characteristics in those earlier marriages because her previous spouse is not in the sample. For these women, we impute previous spouses’ earnings as we describe below. For women with a linked spouse, but with earlier marriages to other men, we use the current husband’s earnings as a proxy for earlier spouses’ earnings.
For the years 1951-1977, the Social Security earnings data are capped at the taxable ceiling.
Over that period, the overall share of workers with earnings above the taxable ceiling ranged from 15 to 36%.5 Since our sample consists of women and married men, the share affected by this topcoding is a good bit higher than the overall levels for men, but much lower for women.
For years 1951-1977, we impute earnings above the taxable ceiling in the following manner. We identify the percentile of the income distribution affected by the topcode for each group (defined by gender and potential experience) for each year. We use 1978-1980 data to calculate the ratio of mean earnings above each of those percentiles to the value of income at that percentile of the 1978-1980 distribution, and then apply this ratio to all top-coded earnings observations.
B. Imputing Marital Status for Women Beyond the SIPP Samples See Table 1, CRS (2006).
While we observe women’s complete marital histories up to the end of their SIPP panel, we have no information on their subsequent marital status. For some cohorts of women, it is possible to estimate marital status using women in the same birth cohort observed in later SIPP surveys. For younger cohorts, however, we have to extrapolate marriage exit and entry rates using information on previous birth cohorts. Our basic methodology is the following. For women for whom we observe actual marital status, we use that information. For women for whom we do not observe actual marital status we predict marital status in the years we do not observe them in the following manner. For women who are married when last observed, we estimate parametric survival models for each education group, assuming a gamma distribution and using cohort dummies, age of marriage, and dummies for whether this is a first, second or 3+ marriage as regressors. We then predict the probability they remain married in the following years, conditioning on marital tenure at the point their marital status is last observed.
Analogously, for women who are not married in the last period we observe them, we estimate similar survival models for the non-marriage spell. We estimate models separately for our three education groups, using cohort dummies and more detailed education dummies as covariates. We estimate these models separately for never married women, and treat those spells as dating from age 18 For women with previous marriage spells, we also include age at which the current non-marriage spell started and dummies for whether they had two marriages or 3+.
As we did for those married when last observed, we condition on the length of the non-marriage spell at their last observation date in predicting the probability that they remain married.
C. Predicted Lifetime Earnings at age 60 For younger cohorts of women, we do not observe marital status or earnings beyond the SIPP survey. We described above how we project marital status for these younger cohorts. Here we describe how we predict cumulative earnings at age 60 for younger cohorts. We estimate the
following log earnings equation:
log Yi ( X ) = β1 X + β 2Cohort + β3Years _ Married + β 4 Kids + ε where Y(X) refers to the woman’s cumulative earnings up to potential experience X, and we include as regressors a linear spline in potential experience with 9 break points, 5-year cohort dummies, cohort dummies interacted with experience, fraction of potential years married, cohort dummies interacted with fraction of years married, fraction of years with children, and cohort dummies interacted with children variables, as well as individual fixed effects. We estimate the above equation separately for three education groups and predict out women’s cumulative earnings. For years in which we have actual earnings, we use that cumulated value. For years after that, we use the growth rate implied by the parameters of the regression model to increment earnings growth from the last observed year for all years up to age 60.
Imputation of Missing Spouses’ Earnings For women who are currently married and living with a spouse, we use the spouses’ earnings for the years the couple is married. For previously married women, we use current spouses’ earnings as a proxy for earlier spouses. For women who are currently divorced, separated or widowed, we randomly assign a spouse from women in the same birth cohort and education group who married around the same age.
Blundell, Richard, Luigi Pistaferri, and Ian Preston. 2008. "Consumption Inequality and Partial Insurance." American Economic Review, 98(5): 1887–1921.
Blau, Francine and Lawrence Kahn. 2007. "Changes in the Labor Supply Behavior of Married Women: 1980-2000" ahn) Journal of Labor Economics 25 (3): 393-438.
Cancian, Maria, Danziger, Sheldon, and Peter Gottschalk. 1993. “Working Wives and the Distribution of Family Income.” In Rising Tides: Rising Inequality in America. New York: Russell Sage Foundation.
Congressional Research Service. 2006 “Social Security: Raising or Eliminating the Taxable Earnings Base” CRS Report for Congress.
Coronado, Julia L., Fullterton, Don, and Thomas Glass. 2000.“The Progressivity of Social Security.” NBER Working Paper No. 7520.
Gustman, Alan L., and Thomas Steinmeier. 2001.“How Effective is Redistribution Under the Social Security Benefit Formula?” Journal of Public Economics, 82: p.1-28.
Gustman, Alan L., Steinmeier. Thomas, and Nahid Tabatabai. 2011. “The Effects of Changes in Women’s Labor Market Attachment on Redistribution Under the Social Security Benefit Formula.” NBER Working Paper 17439.
Isen, Adam and Betsey Stevenson. 2010. "Women's Education and Family Behavior: Trends in Marriage, Divorce and Fertility," NBER Working Paper No. 15725.
Juhn, Chinhui and Kristin McCue. 2010. “Selection and Specialization in the Evolution of Couples’ Earnings.” NBER Papers on Retirement Research Center.
Juhn, Chinhui, and Kevin M. Murphy. 1997. “Wage Inequality and Family Labor Supply.” Journal of Labor Economics 15, pt. 1: 72–97.
Liebman, Jeffrey B. 2002. “Redistribution in the Current U.S. Social Security System.” In The Distribution Aspects of social Security and Social Security. Chicago: University of Chicago Press.
Stevenson, Betsey and Justin Wolfers. 2007. “Marriage and Divorce: Changes and Their Driving Forces.” Journal of Economic Perspectives, 21(2): 27-52.
Married $14,713 $18,034 $27,339 $35,986 $40,002 Separated $29,748 $28,035 $33,718 $34,363 $45,011 Widowed $28,391 $25,305 $29,062 $35,188 $41,500 Divorced $32,959 $31,783 $40,152 $44,451 $46,735 Never Married $32,769 $35,027 $41,119 $45,445 $45,807 Notes: Based on authors’ calculations from March Current Population Survey public use files, 1968-2010.
Table 6: Mean earnings including husband's earnings ($2005) A. High school or less
$0 1936‐40 1941‐45 1946‐50 1951‐55 1956‐60 1961‐65 1966‐70 Notes: Age 60 projections of household earnings (woman’s earnings plus those of any spouse).
Based on SIPP/SSA data and authors’ projections. Measured in thousands of 2006 dollars.
Figure 2: Growth of own and household earnings by marital history A. = High school graduate $1,200 $900 $600 $300 $0 1941‐45 1946‐50 1951‐55 1956‐60 1961‐65 1966‐70 B. Some college $2,000 $1,600 $1,200 $800 $400 $0 1941‐45 1946‐50 1951‐55 1956‐60 1961‐65 1966‐70 C. College graduates $4,000 $3,200 $2,400 $1,600 $800 $0 1941‐45 1946‐50 1951‐55 1956‐60 1961‐65 1966‐70 Own: 10 yr marriage Household earnings: 10 yr marriage Own: No 10 yr marriage Notes: Earnings projections to age 60, based on SIPP/SSA data and authors’ projections.