Joint Committee on Taxation

Transcription

1 bs_bs_banner Review of Income and Wealh Series 60, Number 4, December 2014 DOI: /roiw ACCOUNTING FOR UNITED STATES HOUSEHOLD INCOME INEQUALITY TRENDS: THE CHANGING IMPORTANCE OF HOUSEHOLD STRUCTURE AND MALE AND FEMALE LABOR EARNINGS INEQUALITY by Jeff Larrimore* Join Commiee on Taxaion Using a shif-share analysis on March CPS daa, his paper esimaes he degree o which changes in labor earnings, employmen, and marriage paerns accoun for household income inequaliy growh in he Unied Saes since The facors conribuing o he rapid rise in income inequaliy in he 1980s differ subsanially from hose conribuing o is slower increase since ha ime. Unlike findings for he 1980s when changes in he correlaion of spouses earnings accouned for income inequaliy growh, his facor is no longer a major conribuor oward is coninued increase. Addiionally, he 2000s business cycle is he firs full business cycle in a leas 30 years where changes in earnings of male household heads accouned for declines in income inequaliy. Insead, he coninued growh in income inequaliy in he 2000s was accouned for primarily by increases in female earnings inequaliy and declines in boh male and female employmen. JEL Codes: D31, J12 Keywords: earnings correlaions, income inequaliy decomposiion 1. Inroducion Since researchers firs documened he increases in U.S. income inequaliy and labor earnings inequaliy ha began in he lae 1970s, here has been a srong desire o undersand hese rends. Much of he research has focused on wha accouns for rising labor earnings inequaliy among full-ime workers (Juhn e al., 1993; DiNardo e al., 1996; Lemieux, 2006; Auor e al., 2008). Equally imporan, however, is he relaed quesion of wha facors accoun for household income inequaliy rends. 1 Household income inequaliy would be equivalen o labor earnings inequaliy if each household conained only one worker, ha worker worked full-ime, and here were no non-labor income or governmen ransfers. Bu since his is no he case, household income inequaliy Noe: The auhor hanks Richard Burkhauser, Sephen Jenkins, John Cawley, Don Kenkel, David Ribar, he Edior of his journal, and he wo anonymous referees for heir helpful commens. All opinions expressed in his paper are hose of he auhor and should no be aribued o he Join Commiee on Taxaion or any Member of Congress. *Correspondence o: Jeff Larrimore, Join Commiee on Taxaion, 1625 Longworh Building, Washingon DC 20515, USA ( jeff.larrimore@jc.gov). 1 The erms earnings and labor earnings are used inerchangeably in his paper o refer o earnings from wages and salaries, self-employmen, or farm-employmen. The erms income and household income refer o all income wihin a household from any income source. 683

2 and labor earnings inequaliy rends may diverge. Using a shif-share analysis, his paper explores he facors ha accoun for he growh in household income inequaliy since There are four primary possible conribuors o household inequaliy rends considered in he previous lieraure: increases in male and female labor earnings inequaliy, shifs in male and female employmen raes, increases in he correlaion beween earnings of household members, and he decline in households headed by married couples. Earlier research considering facors ha accoun for income inequaliy changes primarily focused on he period of rapid household income inequaliy growh in he 1980s. This research ypically separaes employmen from and earnings of he head of household and his or her spouse (who will boh be referred o as household heads in his paper) from ha of secondary household members, wih he consensus finding being ha increases in labor earnings inequaliy among male household heads was he primary driver of he 1980s rise in household income inequaliy (Karoly and Burless, 1995; Daly and Vallea, 2006). However, while male earnings inequaliy growh is crucial for undersanding rising household income inequaliy in he 1980s, researchers also agree ha here is no a one-o-one relaionship beween male labor earnings inequaliy and household income inequaliy, as oher facors play a smaller, bu imporan, role. In paricular, a subsanial porion of income inequaliy growh during his period has also been aribued o a rise in he correlaion of spouses earnings (Karoly and Burless, 1995) and a decline in households headed by married couples (Bishop e al., 1997; Daly and Vallea, 2006). In his paper, I build on pas research o provide an updaed assessmen of he facors affecing inequaliy in U.S. household income, exending he analysis ino he recen business cycle which ended wih he recession beginning in lae As his paper will demonsrae, here are subsanial differences in facors accouning for inequaliy rends in he 1980s business cycle ( ) when compared o subsequen periods. Mos noably, in he firs business cycle of he weny-firs cenury ( ), male head labor earnings inequaliy fell, accouning for a reducion in household income inequaliy. Thus, income inequaliy increases in his period came compleely from oher sources. This is in conras o he 1980s business cycle and he 1990s business cycle ( ) when male head earnings inequaliy growh was he mos imporan facor accouning for household income inequaliy growh. 2 Similarly, he role of spouses earnings correlaions have changed over he pas 30 years. In conras o he 1980s, when rising correlaions of spouses earnings accouned for rising inequaliy, in he early 2000s changes in spouses earnings 2 The saring and ending years of business cycles are considered he peaks in median size-adjused household income of persons (Karoly and Burless, 1995; Daly and Vallea, 2006; Burkhauser e al., 2012b; each use similar definiions). These years ofen correspond o he las full year of macroeconomic growh as defined by he NBER. However, his is no always a perfec alignmen, as is he case in he business cycle where he macroeconomic decline began in December 2007, which was also he peak year of median earnings. If he 2000s business cycle were o be defined as , inequaliy growh during his business cycle would have been greaer han ha observed here since inequaliy declined in This decline is largely aribuable o declines in male and female earnings inequaliy. 684

3 correlaions acually accouned for income inequaliy declines. Thus, par of he slowdown in income inequaliy ha has occurred can be aribued o his changing rend in spouses earnings correlaions. I is imporan o noe ha all analyses in his paper describe facors accouning for inequaliy changes, bu do no necessarily indicae a causal relaionship. This maches he approach of earlier income inequaliy decomposiion research. However, while recognizing his limiaion, his paper demonsraes how facors accouning for income inequaliy rends have evolved over ime and no longer reflec hose observed in previous research exploring he 1980s. 2. Daa This analysis is based on public use March CPS daa supplemened wih he cell-mean daa from Larrimore e al. (2008) o correc for Census opcoding of high incomes. The March CPS is a naionally represenaive survey adminisered by he U.S. Census Bureau ha inquires abou he oal pre-ax, pos-ransfer cash income of households from boh labor and non-labor income sources, excluding capial gains (see Appendix Table A1 for a deailed lis of sources capured in he survey). 3 Income is aggregaed o he household level and hen, o accoun for economies of scale in consumpion, size-adjused household income is compued by dividing income by he square roo of household size. This adjusmen is common for U.S. and cross-naional sudies of inequaliy (see, e.g., Goschalk and Smeeding, 1997; Akinson and Brandolini, 2001; Burkhauser e al., 2011) and closely maches he adjusmens for household size implied by he official U.S. povery hresholds (Ruggles, 1990). The sizeadjusmen is performed a he income source level when decomposing income inequaliy rends. For example, all individuals in a household are assigned he same male head labor earnings equal o he earnings of he male household head divided by he square roo of household size. A known limiaion of he March CPS for observing long-erm income inequaliy rends is an arificial inequaliy increase beween 1992 and 1993 ha resuled from changes in Census daa collecion procedures (Ryscavage, 1995; Jones and Weinberg, 2000; Burkhauser e al., 2012a). To remove his rend break, a procedure similar o ha used by Akinson e al. (2011) and Burkhauser e al. (2012a) is implemened where all inequaliy series are adjused upwards prior o 1993 such ha no income inequaliy growh occurred from 1992 o Of course, his procedure canno disinguish real inequaliy changes ha occurred in his year from hose due o he redesign, and by necessiy suppresses boh such changes. This approach is used boh for he overall sample resuls and for resuls for each sep of he decomposiion such ha no change from any characerisics 3 While his paper focuses on he sandard cash-income measure, some researchers have recenly advocaed using broader income measures ha include excluded income sources such as non-cash ransfers and employee benefis, ax credis, and ax liabiliies (Reynolds, 2006; Burkhauser e al., 2012b; Congressional Budge Office, 2012) or measuring consumpion inequaliy insead (Krueger and Perri, 2006; Meyer and Sullivan, 2009). Decomposing inequaliy rends using hese alernae measures would be a valuable avenue for furher research. 685

4 are capured beween 1992 and All resuls of he decomposiion, herefore, also exclude any changes from his one year of he business cycle. All individuals in he March CPS are included in he analysis regardless of age, excep hose in group quarers or miliary households. While his lack of an age resricion is common in he inequaliy lieraure (see, e.g., Ryscavage, 1995; Bishop e al., 1997; Pikey and Saez, 2003; Burkhauser e al., 2009; DeNavas-Wal e al., 2011; Congressional Budge Office, 2012), some researchers, including Blank (2011), Daly and Vallea (2006), and Karoly and Burless (1995) resric heir samples o households or families wih working-age heads. This may resul in modes differences from he findings of Daly and Vallea (2006) and Karoly and Burless (1995). However, i also allows for a more complee picure of populaion-wide income inequaliy rends. Furher, including individuals of all ages allows he analysis o accoun for public ransfers programs such as Social Securiy, which canno be fully included when analyzing only working-age individuals. 3. Mehod of Decomposing he Increase in Household Income Inequaliy To decompose he change in household income inequaliy ino ha aribuable o is componen sources, a shif-share analysis is used ha sars wih income daa from 1979 (from he 1980 CPS). To obain he imporance of each facor, each is added sequenially and he resuling income inequaliy rend is compared o ha which would have occurred had he specified facor remained unchanged. Poenial facors are divided ino hree caegories: changes o he prevalence of populaion characerisics, changes o he disribuion of source-level incomes, and changes o he correlaions of income across income sources. The mehods for capuring each caegory s relaionship wih household income inequaliy changes are described in greaer deail below. Unlike he DiNardo e al. (1996) decomposiion mehod, which is used by Daly and Vallea (2006), his approach does no iniially condiion on deailed background characerisics such as age, locaion, and race when evaluaing rends. The DiNardo e al. (1996) approach is valuable for observing he relaionship beween individual characerisics and income inequaliy rends. However, Cowell and Fiorio (2011) noe ha he daa inensiy of heir mehod prevens i from being suiable for all decomposiions of ineres. In paricular, i is limied in is abiliy o observe how a range of income sources inerac o accoun for changing inequaliy, which is he primary focus of his paper. This is exemplified by Daly and Vallea (2006) who use DiNardo e al. s (1996) approach bu hen mus analyze source-level income changes uncondiional on oher facors. In conras, he shif-share approach used here sars wih a small se of caegorical characerisics (marial saus and employmen sauses of he household heads) bu allows for a more horough undersanding of he relaionship beween changes o income sources and household income inequaliy. 4 4 While demographic facors are no included and he analysis here sars wih marial saus, i was considered o sar he decomposiion wih caegorical age caegories so all resuls would be condiional on age. This inclusion produced largely similar resuls and was herefore omied. However, he aging of he populaion accouned for sligh increases in inequaliy in he 1990s and 2000s and including his facor slighly reduced he inequaliy changes accouned for by he male labor earnings disribuion during his ime. 686

5 Changes in he Prevalence of Populaion Characerisic The firs decomposiion approach addresses changes in caegorical characerisics, including he marial and employmen saus of he household heads. For example, i considers how an increase in full-ime workers will change he overall income disribuion, holding he income disribuion of full-ime, par-ime, and non-workers consan. The porion of household income inequaliy changes aribuable o he prevalence of such populaion groups is esimaed using Akinson s (1998) and Burless s (1999) approach where wihin-group income disribuions are held consan and, hrough reweighing observaions, he relaive size of each group is allowed o change. This esimaion is performed by reweighing observaions from he base year,, such ha he fracion of he populaion in each group maches ha in fuure years,. By increasing he weigh of individuals wih characerisics ha are more prevalen in year han in year, he impac of changing he prevalence of characerisics is esimaed wihou alering he underlying income disribuions wihin each group. For evaluaing long-erm income inequaliy rends over business cycles, as is done here, he year is he following business cycle peak. Changes in Source-Level Income Disribuions Wihin Populaion Groups The second se of facors recognizes ha he income disribuions wihin each of he aforemenioned groups have changed along wih each group s prevalence. These changes can resul from any income source including male head labor earnings, female head labor earnings, non-head earnings, or non-labor income. The effec of changes o source-level income disribuions are analyzed using a rankpreserving income exchange derived from Burless (1999) and Daly and Vallea (2006). In doing so, noe ha each individual s income, Y ik can be represened as he sum of heir incomes from each income source, f1 ik hrough f Nik : (1) Yik = f1 ik + f2ik + + fnik. Individuals are assigned a percenile rank, p fik, for each source based on he rank of heir source-level income wihin heir populaion group k. For now, rank-correlaions across income sources wihin each populaion group are assumed o be consan as hese correlaion changes are considered separaely. Thus, o esimae he impac ha changes o he disribuion of source f 1 have on income inequaliy, each individual s income from he source f 1 in year is replaced wih he income of he individual a he same percenile rank of he source f 1 income disribuion in year : (2) ˆ Y ( p ) = f ( p ) + f + + f. ik 1ik 1ik 1ik 2ik This preserves he condiional earnings rank of each individual from source f 1 and he rank-correlaion of earnings from source f 1 wih oher income sources, bu Nik 687

6 capures changes in he source-level income disribuion of source f 1 wihin each populaion group. Since his procedure combines income across years, all income is adjused using he CPI-U-RS prior o he analysis. Changes in Income-Source Rank Correlaions Wihin Populaion Groups The previous mehods held he rank correlaion of income sources consan, such ha if he male and female heads a percenile-ranks p 1ik and p 2ik in heir condiional earnings disribuions are married o each oher one year, hen he same rank pairing is assumed o exis in all fuure years. Thus, by performing rank-preserving income exchanges for sources f 1 and f 2 separaely, he impacs of he separae earnings disribuions are analyzed wihou impacing he correlaion beween he wo: (3) ˆ Y ( p, p ) = f ( p ) + f ( p ) + f + + f. ik 1ik 2ik 1ik 1ik 2k 2k 3ik In order o updae he correlaion beween sources f 1 and f 2, raher han dividing income ino N separae sources i is divided ino (N - 1) sources such ha g 1 = f 1 + f 2 and f 3 hrough f N are unchanged. The rank-correlaion change of sources f 1 and f 2 is capured by combining hese sources prior o he rankpreserving income exchange raher han afer. Thus, calling each individual s percenile-rank in he g 1 disribuion q fik, esimaed incomes can be calculaed as: (4) ˆ Y ( q ) = g ( q ) + f + + f, ik 1ik 1ik 1ik 3ik which updaes he correlaion beween sources f 1 and f 2 along wih heir income disribuions. By comparing he resuls when only heir separae income disribuions change (equaion (3)) wih ha when heir join disribuion changes (equaion (4)), he impac of he changing correlaion beween sources f 1 and f 2 is capured in he analysis. Nik Nik 4. Decomposiion Resuls In order o focus on rends devoid of business cycle variaions, we proceed by considering income inequaliy rends across peak years of median income in each business cycle. The firs row of Table 1 presens he acual average annual percenage change in Gini coefficiens for each business cycle since Household income inequaliy, measured using he Gini coefficien, increased a a rapid rae of 0.97 percen per year in he 1980s business cycle ( ). While i coninued o rise in subsequen business cycles, his increase was small compared o ha seen in he 1980s. In he 1990s business cycle ( ) i increased by an average of jus 0.08 percen per year, and in he 2000s business cycle ( ) i increased by an average of 0.10 percen per year. This mirrors he household income inequaliy rends observed by Burkhauser e al. (2011) and Larrimore e al. (2008). The remainder of Table 1, and he discussion ha follows, considers he facors accouning for hese rends. 688

7 TABLE 1 Esimaed Average Annual Percenage Change in he Size-Adjused Household Income Gini Coefficien Aribuable o Facor Componens by Business Cycle (1) Acual Gini average annual percenage change Average annual percenage change accouned for by: (2) Marriage raes (3) Male head employmen (4) Male head earnings disribuion (5) Female head employmen (6) Female head earnings disribuion (7) Spouses earnings correlaion (8) Non-head labor earnings disribuion (9) Non-head labor earnings correlaion (10) Privae non-labor income disribuion (11) Privae non-labor income correlaion (12) Public ransfers disribuion (13) Public ransfers correlaion Noes: References o male and female head employmen and earnings refer o hose of boh he household head and his or her spouse. Due o changes in he March CPS daa collecion procedures beween 1992 and 1993 ha limi comparabiliy beween hose years, inequaliy changes from 1992 o 1993 are suppressed using he procedure from Akinson e al. (2011) and described in he main ex. Source: Auhor s calculaions using March CPS daa. Changes o Male Employmen Saus and Earnings Disribuions The firs major facors considered in he analysis are earnings and employmen changes among male head of households. 5 Before addressing hese key facors, he analysis firs conrols for he underlying demographic shif from declining marriage raes in he Unied Saes since he 1970s. While his shif in marriage paerns provides an underpinning for inequaliy growh, maching Karoly and Burless s (1995) and Daly and Vallea s (2006) findings for he 1980s, i alone canno explain subsanial inequaliy growh. From Row 2 of Table 1 we see ha i accouns for approximaely 0.09 percen per year of inequaliy growh over he hree business cycles and has been relaively sable over ime. Thus, we urn our aenion o hose facors, such as employmen and earnings, which are generally viewed as more imporan for household income inequaliy rends. Over he pas 30 years here has been a general decline in work among male heads of households (Panel A of Table 2), alhough his decline is primarily concenraed among par-ime workers (hose working less han 35 hours per week or less han 50 weeks of he year) raher han among full-ime workers. Using he reweighing echnique described previously, he relaionship beween hese employmen declines and inequaliy growh can be observed. When doing so, we 5 Since household heads are analyzed separaely from oher household members, all references o male and female earnings refer o he size-adjused labor earnings of he male and female household heads and heir spouses. Addiionally, since household heads and spouses are considered equivalen for his paper, references o household heads refer o boh he head and his or her spouse. Togeher, household heads and spouses made up 79 percen of all individuals over he age of 18 in 2007 and 89 percen of all labor earnings in he Unied Saes in ha year, so his represens he vas majoriy of all earnings received. 689

8 Year TABLE 2 Percenage of Male and Female Household Heads and Spouses Working Full-Time, Par-Time, and No Working Percen Working Full-Time Gini for Full- Time Earnings Percen Working Par-Time Gini for Par- Time Earnings Percen No Working Panel A: Employmen saus and Gini coefficien for earnings by employmen saus for male household heads and male spouses of household heads Panel B: Employmen saus and Gini coefficien for earnings by employmen saus for of female household heads and female spouses of household heads Source: Auhor s calculaions using March CPS daa. see ha he male employmen changes had only a small impac on income inequaliy (Row 3 of Table 1). In each full business cycle since 1979, male employmen declines accouned for less han a 0.05 percen-per-year change in income inequaliy. This is no o say ha employmen effecs never have an effec, as Burkhauser and Larrimore (2011) illusrae he imporance of employmen changes during periods of economic declines. Bu i is clear ha male head employmen paerns are no primary drivers of long-erm income inequaliy growh. In conras o male employmen changes, shifs in he male earnings disribuion condiional on employmen and marial saus were exremely imporan for undersanding income inequaliy growh in boh he 1980s and 1990s business cycles. During his ime, male earnings inequaliy changes accouned for more of he income inequaliy growh han any oher single facor. In he 1980s business cycle when income inequaliy grew mos rapidly, male earnings changes accouned for income inequaliy growh of 0.65 percen-per-year. This represens over wo hirds of he ne increase in income inequaliy during his period, which is in line wih esimaes previously observed for he 1980s by Daly and Vallea (2006). In he 1990s business cycle, male earnings changes accouned for a smaller 0.36 average annual percenage increase in income inequaliy. Bu he conribuion of male earnings did no slow nearly as much as household income inequaliy growh, which slowed o 0.08 percen growh per year. As a resul, in he 1990s business cycle rising male earnings inequaliy accouned for over four imes he increase in household income inequaliy. This suggess ha i was he slowing inequaliy growh or faser inequaliy declines from oher facors which accoun for why household income inequaliy growh slowed o he exen ha i did in he 1990s. In he mos recen business cycle, however, he conribuion of male earnings changes shifed dramaically. This shif is noable given ha male earnings are a primary focus of many inequaliy discussions. In he 2000s business cycle i 690

9 accouned for a 0.35 average annual percenage poin reducion in income inequaliy, raher han accouning for growh as was previously he case. Thus, while male earnings changes over-accouned for inequaliy growh in he 1990s business cycle, his rend reversed in he 2000s as i accouned for inequaliy declines. This paern is broadly consisen wih he paern of inequaliy in male full-ime earnings seen in Column 2 of Table 2, which presens he Gini coefficien for annual male labor earnings among household heads working full-ime. This column illusraes ha male head earnings inequaliy rose for hese men during he 1980s and 1990s business cycle before falling during he 2000s business cycle. While he focus here is on male household heads, his is also rue among all men working full-ime and is broadly consisen wih Kopczuk e al. (2010) who use Social Securiy Records o observe ha male inequaliy in he las year of heir sample, 2004, was virually he same as in This decline in earnings inequaliy among full-ime workers in he early 2000s maches he previous observaion ha male labor earnings inequaliy no longer conribued o household income inequaliy growh during his period. A furher robusness check of hese resuls is o consider heir sensiiviy o he choice of inequaliy measure. Table 3 does jus ha, replicaing he decomposiion approach from Table 1 for seven addiional inequaliy measures ha are sensiive o differen areas of he income disribuion: hree Generalized Enropy (GE) measures, he coefficien of variaion (CV), and hree percenile raios. The Gini, GE(1), and CV are all relaively middle-sensiive; he GE(0) is relaively boomsensiive; he GE(2) is relaively op-sensiive; and he P90/P10, P90/P50, and P50/P10 are sensiive o changes around he specified percenile poins in each series. As wih Table 1, he firs row of each panel presens he acual increase observed for he inequaliy measure and he subsequen rows provide he increase accouned for by each facor. From Table 3, we see ha he paern describing he changes accouned for by male earnings is largely robus o he inequaliy measure chosen. For all inequaliy series, male earnings were he dominan facor for household income inequaliy growh in he 1980s. In he 1990s, is conribuion slowed bu no o he same exen as household income inequaliy growh slowed so male earnings now over-accoun for oal income inequaliy growh. And in he 2000s, he conribuion of male head earnings urned negaive under every measure. While here are differences in he magniude of hese rends, based on where in he disribuion inequaliy measures are sensiive, i is clear ha male earnings did no coninue o accoun for income inequaliy growh in he early 2000s as i did in earlier decades. Changes o Female Employmen Saus and Earnings Disribuions A key shif in household incomes in he pas 30 years came from he dramaic increase in female employmen. As seen in Row 5 of Table 1, hese increases slowed household income inequaliy growh over boh he 1980s and 1990s. Specifically, increases in female head employmen accouned for slowing inequaliy growh by 0.15 and 0.16 percen-per-year, respecively, in hese wo business cycles. In he 2000s business cycle, however, female employmen declined from 691

10 TABLE 3 Esimaed Average Annual Percenage Change in Various Inequaliy Measures Accouned for by Facor Componens Gini GE(0) GE(1) GE(2) CV P90/P10 P90/P50 P50/P10 Panel A: Inequaliy changes from 1979 o 1989 Acual (2) Marriage raes (3) Male head employmen (4) Male head earnings disribuion (5) Female head employmen (6) Female head earnings disribuion (7) Spouses earnings correlaion (8) Non-head labor earnings disribuion (9) Non-head labor earnings correlaion (10) Privae non-labor income disribuion (11) Privae non-labor income correlaion (12) Public ransfers disribuion (13) Public ransfers correlaion Panel B: Inequaliy changes from 1989 o 2000 Acual (2) Marriage raes (3) Male head employmen (4) Male head earnings disribuion (5) Female head employmen (6) Female head earnings disribuion (7) Spouses earnings correlaion (8) Non-head labor earnings disribuion (9) Non-head labor earnings correlaion (10) Privae non-labor income disribuion (11) Privae non-labor income correlaion (12) Public ransfers disribuion (13) Public ransfers correlaion Panel C: Inequaliy changes from 2000 o 2007 Acual (2) Marriage raes (3) Male head employmen (4) Male head earnings disribuion (5) Female head employmen (6) Female head earnings disribuion (7) Spouses earnings correlaion (8) Non-head labor earnings disribuion (9) Non-head labor earnings correlaion (10) Privae non-labor income disribuion (11) Privae non-labor income correlaion (12) Public ransfers disribuion (13) Public ransfers correlaion Panel D: Inequaliy changes from 1979 o 2007 Acual (2) Marriage Raes (3) Male Employmen Raes (4) Male Earnings Disribuion (5) Female Employmen Raes (6) Female Earnings Disribuion (7) Spouses Earnings Correlaion (8) Non-head Labor Earnings Disribuion (9) Non-head Labor Earnings Correlaion (10) Privae Non-labor income Disribuion (11) Privae Non-labor income Correlaion (12) Public Transfers Disribuion (13) Public Transfers Correlaion Noes: See Noes o Table 1. Since he Generalized Enropy inequaliy merics are undefined for negaive incomes, negaive income values are boom coded o $0.01 when calculaing inequaliy for hese hree measures. Source: Auhor s calculaions using March CPS daa. 692

11 peak-o-peak of a business cycle for he firs ime since a leas he 1970s (Panel B of Table 2). This, in urn, reversed he conribuions o inequaliy from female employmen changes seen in earlier business cycles. In he business cycle, female employmen acually accouned for an increase in income inequaliy. This reversal helps accoun for why moderae levels of income inequaliy growh coninued in he early 2000s despie he earlier observaion ha male earnings changes accouned for inequaliy declines. Of course, in addiion o changes in female employmen, we observe changes in he earnings disribuion of female heads of household, condiional on marial saus and wheher hey works full- or par-ime. The impac of hese changes in female head earnings can be seen in Row 6 of Table 1 where hey parially offse he household income inequaliy declines ha resuled from he increase in female employmen. However, unil he 2000s business cycle, he ne increase in inequaliy is sill smaller han i would have been had boh female employmen and female earnings inequaliy remained unchanged a he 1979 levels. I is only in his early 2000s business cycle ha he changes in female head employmen and earnings disribuion boh worked in he same direcion. In he early 2000s, hese facors combined o accoun for a 0.25 percen-per-year increase in he household income Gini coefficien. Thus, raher han miigaing income inequaliy growh, he shifs in female employmen and earnings accouned for sizable income inequaliy increases in he mos recen business cycle. Addiionally, his paern across business cycles for female employmen can be seen for each of he major inequaliy merics in Table 3. In paricular, female employmen growh accouned for inequaliy declines for all bu one meric in he 1980s and 1990s and female employmen declines accouned for inequaliy growh across all merics in he 2000s. Changes o Spouses Earnings Correlaions Thus far he analysis has assumed ha he rank correlaion across income sources remains unchanged. However, his has no been he case as spouses labor earnings have become more correlaed since he 1970s. This increase in rank correlaions has, in urn, accouned for increased household income inequaliy by concenraing wealh ino a smaller number of households (Row 7 of Table 1). Closely maching Burless s (1999) findings, his increase in correlaion accouned for a 0.14 percen-per-year increase in income inequaliy during he 1980s business cycle. However, his was he high-poin of correlaion accouning for inequaliy growh. By he 1990s business cycle, we observe ha earnings correlaion s imporance fell and only accouned for a smaller 0.02 percen-per-year of inequaliy growh. By he 2000s business cycle he effec reversed direcions and acually accouned for inequaliy declines. Thus, a shif in correlaion rends sheds ligh on he slowdown in household income inequaliy growh ha persised hrough he 1990s and 2000s. Bu wha led o he differen paerns of correlaion changes in he 1980s compared o he 1990s and 2000s? Correlaion changes can come eiher from shifs in he correlaion of earnings among duel-earner couples or from changes in where 693

12 Figure 1. Percenage Change in Wives Employmen Rae by Decile of Husbands Earnings Among Working-Age (22 62 Years) Couples Noe: Includes head-of-household couples where boh individuals are of working-age (22 62). Source: Auhor s calculaions using March CPS daa. in he income disribuion women are enering he labor marke and men are leaving he labor marke. I appears ha he shifs in where enry and exis from he labor marke are occurring are driving hese resuls. Previous research has observed an increase in he correlaion beween a man s earnings and his wife s probabiliy of working since he lae 1960s, wih more wives of high earning men working now han was he case in he pas (Schwarz, 2010). Bu while his is rue over he long-run, by he early 1990s he wives of high earning men were no longer enering he labor marke a a faser pace han hose women married o lower earning men. This can be seen in Figure 1, which illusraes he change in female employmen raes among married female household heads in each full business cycle since 1979 based on he decile of heir husband s labor earnings. Since mos labor earnings belong o individuals of working age, he sample is resriced here o couples where boh members are beween he ages of 22 and 62. Addiionally, based on he crosssecional naure of he daa, i does no rack he same individuals over he period bu insead considers female employmen raes in he firs year of he business cycle based on heir husbands earnings in ha year and hen considers female employmen raes in he las year of he business cycle based on heir husbands earnings in his final year. 694

13 When considering employmen rends in his way, i is apparen ha in he 1980s he mos rapid rise in female employmen occurred among women married o high earning men. Among working-age women married o men in he op decile of he male earnings disribuion, employmen increased by over 13 percenage poins. However, among women married o non-working men, employmen increased by less han 2 percenage poins. This difference is valuable for undersanding why correlaions rose during ha ime, since he increase in employmen among women married o high earning men necessarily increases spouses earnings correlaions. Looking a he 1990s and 2000s business cycles when he correlaion s influence on income inequaliy rends slowed, he paern is quie differen. During he 1990s business cycle, women married o non-working men increased heir employmen by 11 percenage poins while women married o working men in any decile saw slower employmen growh ha was less han 5 percenage poins in all deciles. This reduced earnings correlaions as zeros (non-earners) are no longer paired as he woman eners he labor marke. Similarly, in he 2000s business cycle women married o non-working men increased heir employmen rae faser han women married o men in six of he earnings deciles. So while he increased income inequaliy from rising correlaions in he 1980s came from high-earning households increasing heir incomes as spouses enered he labor marke, in he 1990s and 2000s earnings correlaions fell a he boom of he disribuion hrough he decline in no-earner households. One poenial explanaion for he decline in no-earner households in he 1990s is he increase in public policies encouraging employmen among families wih children. Significan expansions of he Earned Income Tax Credi beween 1993 and 1996 incenivized low income households wih children o have a leas one working household member. Addiionally, welfare changes in he Welfare Reform Ac of 1996 made i more difficul for households wih children o survive on governmen benefis alone. Thus, if he increase of female employmen in families where he man is no working occur primarily in families wih children, i could be aribued o hese public policy changes. Figure 2 provides evidence ha his is he case. Among married working-age women wih a non-working husband, employmen increased over he 1990s for boh hose wih and wihou children, bu he growh is faser among mohers, who would be mos impaced by he policy shifs. In 1996, prior o he implemenaion of he Welfare Reform Ac in 1997, he employmen rae for mohers wih a non-working husband was 5.3 percenage poins greaer han ha for non-mohers wih a non-working husband. By 2000, his gap had grown o 14.6 percenage poins. Among unmarried women wih children, who are similarly affeced by he policy, we also see a disproporionae rise in employmen when compared o unmarried women wihou children. In conras, no similar dynamic occurred over his period among women wih working husbands, who would be less affeced by changing welfare rules. In 1996, women wih working husbands wih children were 5.7 percenage poins less likely han women wih working husbands wihou children o work. In 2000, his gap remained a similar 6.4 percenage poins. Therefore, while i is likely ha oher 695

14 Figure 2. Wives Employmen Rae by Husbands Employmen Saus and Presence of Children Among Working-Age (22 62 Years) Couples Noe: Includes head-of-household couples where boh individuals are of working-age (22 62). Unmarried women include head-of-household working-age (22 62) women. Children include hose living in he household under he age of 18. Source: Auhor s calculaions using March CPS daa. broad-based facors also conribued o he growh in employmen among women married o non-working men in he 1990s, i is probable ha a leas par of his rend and he relaed slower growh in earnings correlaions beween spouses can be aribued o he labor response o hese public policies. Oher Income While declining marriage raes, changing employmen sauses, increased labor earnings inequaliy of household heads, and increased correlaions of spouses earnings accouns for much of he increase in inequaliy, shifs in hese facors alone undersae he rapid rise in inequaliy in he 1980s and 2000s and oversae i in he 1990s. This is because non-labor income and he earnings of non-heads have no ye been included in he analysis. During he rapid income inequaliy increases in he 1980s business cycle, he combinaion of facors considered hus far accouned for 0.89 percen-per-year of he 0.97 percen-per-year increase in he Gini coefficien ha acually occurred. In paricular, reducions in he amoun of public ransfers and reducions in heir concenraion among he lowes income individuals (Rows of Table 1) accouned for furher income inequaliy growh and explain mos of his gap beween inequaliy growh already accouned for and he 0.97 percen-per-year of 696

15 acual inequaliy growh. This increase accouned for by public ransfers should no be aken o say ha public ransfers increased inequaliy bu raher ha hey became less effecive a reducing inequaliy. In he 1990s, on he oher hand, public ransfers increased and became more correlaed wih low incomes which helped accoun for he slowdown in income inequaliy as measured using he Gini coefficien. Similarly, in he 1990s when he facors discussed so far over-accoun for inequaliy growh, i is eviden ha shifs in he labor earnings disribuion of non-heads miigaed income inequaliy increases. This is in conras o he 1980s where changes in he labor earnings disribuion of non-heads slighly increased income inequaliy. Thus, along wih he oher facors which accouned for slower inequaliy growh from he 1980s o he 1990s, he reversing impac of non-head income is an imporan facor in accouning for he slowdown in household income inequaliy growh afer is rapid increases in he 1980s. A furher consideraion of paricular relevance o hese facors is he choice of inequaliy measure. Since public ransfers end o be concenraed among lowincome individuals and privae non-labor income end o be concenraed among hose wih high-incomes, we may expec o see differences for alernae inequaliy measures. This is, in fac, he case. Mos noably, he boom-sensiive GE(0) observes he influence of public ransfers more han any oher inequaliy measure. For he overall 28-year period (Panel D of Table 3), changes in he level and disribuion of public ransfers accouned for almos 60 percen of he growh in he GE(0) inequaliy meric. This is in conras o ha seen for he oher measures where i never accouns for more han 10 percen of income inequaliy growh. Even when using he P50/P10 raio his relaionship is no as srong, suggesing ha i ruly is he poores of he poor, below he 10h percenile of he disribuion, where he changes capured by he GE(0) are occurring. Furhermore, he imporance of public ransfers o he GE(0) exiss across all business cycles, as i accouns for 32 percen of GE(0) inequaliy growh in he 1980s, 68 percen in he 1990s, and 86 percen in he 2000s. The declining abiliy for public ransfers o miigae income inequaliy sands ou as he primary facor for he acceleraion of inequaliy growh in he 2000s when measured using he GE(0) series. This is in conras o ha seen earlier for he Gini, where inequaliy growh slowed and public ransfer income was relaively unimporan for income inequaliy rends in he 1990s and 2000s. I is also among hese boom-sensiive inequaliy measures where non-head earnings represen he bigges miigaing facor agains income inequaliy growh. Previously, we observed ha increases in earnings of hese secondary household members accouned for relaively modes inequaliy declines paricularly since Consisen wih low-income houses joining ogeher o pool resources, when focusing on he boom-sensiive GE(0) and P50/P10 raio we can see ha increases in secondary wage earnings had a more subsanial impac when focusing on he lower porion of he disribuion. Finally, a he oher end of he income scale, privae non-labor income such as ineres and dividends are more prevalen. In he 1980s, changes o he disribuion for hese income sources resuled in increases in he op-sensiive GE(2) inequaliy measure more han ha seen for oher measures. However, hese 697

16 increases were no susained in he fuure business cycles, suggesing ha ineres and dividends did no accoun for increases in inequaliy among he upper half of he disribuion in he mos recen business cycles. This does no, however, rule ou he possibiliy of capial gains leading o income inequaliy growh in recen decades since capial gains are no capured in he CPS and are no included in hese inequaliy calculaions. Robusness Checks In addiion o esing he robusness of resuls o differen inequaliy merics, wo furher checks of he robusness of resuls were employed. The firs addresses a known limiaion of shif-share analyses, where resuls may be sensiive o he order in which he componen facors are analyzed (Jenkins, 1995; Fournier, 2001; Daly and Vallea, 2006). A common approach o address his possibiliy is analyzing facor componens in reverse order (see, e.g., Daly and Vallea, 2006). The resuls of reversing he order of analysis are provided in Appendix Table A2. When doing so, he general paerns of facors accouning for income inequaliy growh in each business cycle remain largely consisen wih hose discussed above. In mos insances, he change accouned for by each facor is wihin 0.03 percen-per-year regardless of which order is used. However, his is no universally he case as here are some differences in he magniude of resuls. The mos noable are ha declines in marriage and male labor earnings changes accoun for less income inequaliy growh in he 1990s when using he reverse decomposiion, while privae non-labor income and labor earnings of non-heads or heir spouses accouned for more. Thus, when using he reverse-decomposiion, he exen o which male earnings inequaliy over-accouns for income inequaliy growh in he 1990s is less subsanial. While some such differences exis, he consisency of he main resuls o hese reverse-order decomposiion resuls suggess ha he order of analysis is no driving he resuls. A second robusness check addresses wheher he imporance of marial changes and non-head labor earnings may be oversaed. Over he pas several decades here has been an increase in cohabiaion, which may resul in couples acing like married couples bu no being couned as married in CPS daa. These long-erm cohabiaion are esimaed by assuming ha all pairs of unrelaed similarly aged (wihin 10 years of each oher) unmarried aduls of opposie gender who live ogeher wihou oher aduls presen are a cohabiing couple (Congressional Budge Office (1997) and Marin (2006) employ similar definiions). Appendix Table A3 reruns he analysis from Table 1, reaing hese cohabiing couples as married. When doing so, he income inequaliy growh accouned for by marriage paerns is slower han ha repored in Table 1 since his reduces he observed decline in marriage. I similarly reduces he miigaion of income inequaliy growh from changes in he non-head labor earnings disribuion. Thus, for hese wo variables i is eviden ha shifs o cohabiaion played a role, paricularly in he 1990s. However, he magniude of differences beween hese resuls and hose in he main resuls in Table 1 remains small and never differs by more han 0.04 percen-per-year for any facor. Similarly, he inequaliy changes accouned for by he correlaion of earnings beween spouses and cohabiing couples is virually unchanged. Thus, he conribuions of marriage paerns and 698

17 non-head earnings o income inequaliy rends appear o go beyond recen increases in cohabiaion. 5. Conclusions Numerous sudies have documened he increases in male earnings inequaliy and household income inequaliy ha occurred since he lae 1970s. Despie close relaionships beween hese series, here are numerous oher facors ha mus also be considered o undersand household income inequaliy rends. When disaggregaing he increase in household income inequaliy ino is componen sources, during he 1980s business cycle when household income inequaliy growh was mos pronounced, he increase came from an alignmen of numerous facors all driving income inequaliy higher. During his period, male earnings changes accouned for wo-hirds of he rise in household income inequaliy. However, declines in marriage raes and an increased correlaion beween spouses earnings furher conribued o increasing inequaliy, leading o he rapid overall income inequaliy growh. During he 1990s business cycle when income inequaliy growh was he slowes, while he conribuion of male earnings changes slowed i did no slow nearly as much as household income inequaliy growh did. However, he conribuion o rising income inequaliy of marriage rae declines, spouses earnings correlaion changes, and changes o he non-head earnings disribuions and public-ransfers disribuions all declined or reversed by he business cycle, largely accouning for he slower inequaliy growh acually observed. In conras, he mos recen business cycle from 2000 o 2007 is noable because for he firs ime since a leas 1979, male earnings inequaliy growh could no accoun for any increases in household income inequaliy over a business cycle. Furhermore, while female employmen changes miigaed income inequaliy growh for each of he previous wo business cycles, during his business cycle i conribued o rising inequaliy. As his paper illusraes, he facors accouning for rising household income inequaliy have changed over ime and he relaive imporance of hese facors is more differen oday han i was 30 years ago. I also highlighs a subsanial challenge in reducing inequaliy from is curren elevaed levels, as he primary facor miigaing inequaliy growh in he 1980s and 1990s female employmen growh no longer accouns for income inequaliy declines. If female employmen has reached a plaeau and is unable o be increased furher, oher facors will have o drive any effor o reduce inequaliy going forward. In furher considering how o reverse he upward rend in inequaliy, i may be fruiful o exend upon hese facors furher and consider how axes and in-kind ransfers have influenced inequaliy rends. Many of he policy discussions regarding inequaliy incorporae some elemen of ax policy, so deermining he influence of axes on inequaliy rends in a pos-ax environmen could help o guide policy decisions. Such facors canno be fully evaluaed using he pre-ax, cash-income definiions ha are common in he income inequaliy lieraure, alhough incorporaing hese facors in a pos-ax analysis will be a valuable exension of his area of research. 699