MALE LOSERS AND WINNERS AFTER SEPARATION/DIVORCE 251 port the instrumental variables estimates us- come after income taxes (including state in- ing the twice-lagged income measure. come taxes and the Earned Income Tax To minimize the impact of outliers,we Credit),payroll taxes (health,unemploy- truncated the dependent variable to the range ment,and retirement),and public transfers. [-2,2],effectively capping losses at 86.5 Public transfers include AFDC,SSI,and percent and capping gains at 639 percent of other“welfare'”programs,Social Security, pre-disruption income or living standards.3 unemployment compensation,worker's com- Our final models include control variables pensation,and the face value of food stamps. for change in job tenure,for age (if age en- POST-GOVERNMENT HOUSEHOLD IN- ters the level model as a quadratic,it remains COME AFTER SUPPORT PAYMENTS.Re- as a linear predictor in the difference model), spondents in all survey years report the total dummy variables for calendar year,and the amount of money paid during the previous indicator for new partner at time t+1.To calendar year toward the support of persons simplify the presentation of results,we re- outside the household,including voluntary port only the coefficient and standard error out-transfers and mandatory support pay- for D.the indicator for union disruption.We ments.We subtracted this total support also report the economic impact of disrup- amount from household post-government in- tion,which equals exp(p)-1. come to create a measure of"net"household income.Beginning in 1985,respondents who reported paying support were then MEASUREMENT probed specifically for child-support and ali- MEASURING INCOME mony payments.We used these data to cre- ate an additional measure of household in- We used the following measures of house- come that subtracts only these compulsory hold income (all adjusted to 1990 dollars us- family-support payments from household ing the Consumer Price Index [CPI-U])as post-government income,thus allowing us to our dependent variables: distinguish these compulsory payments from TOTAL PRIVATE HOUSEHOLD INCOME. any additional support paid on an informal This is our measure of“gross income,”and or voluntary basis for the later subsample equals the combined earned income received Any missing data on support payments were by all household members from all employ- imputed to be zero. ment (including wages,salaries,farm in- Some caution is necessary when dealing come,business income,income from market with men's reports of support payments.Pre- gardening,and income from roomers and vious research has shown that divorced men boarders),assets,and private transfers (in- who pay child support report higher levels cluding alimony and child support)before of transfers than their former wives report taxes and government transfers.It also in- receiving (Braver,Fitzpatrick,and Bay cludes the imputed rental value of housing. 1991;Peters et al.1993;Schaeffer,Seltzer POST-GOVERNMENT HOUSEHOLD IN- and Klawitter 1991),and when third-party CoME.This equals the sum of household in- verification is available (e.g.,when the 3 The untruncated results produce much more 4 The wording of this sequence from the 1993 negative estimates of the average impact of union interview is as follows: dissolution because they are strongly influenced G103.In 1992,did you give any money toward the by those separating men who suffer precipitous support of anyone who was not living with income declines between time t-I and time you at the time? t+1.We experimented with alternative mea- G106.How much money was that altogether in sures that minimize the influence of these outli- 1992? ers,including a measure based on Duncan and G107.Was any of that child support? Hoffman (1985)that placed a floor of S3,000 on G108.How much did that child support amount to nominal income levels.The results for net in- in1992? come using this measure were similar to those G109.Was any of the money you gave in 1992 ali- presented here;however,the income floor ob- mony? scured important nuances in the specific mecha- G110.How much did that alimony amount to in nisms that structured income change. 1992? This content downloaded from 129.96.252.188 on Mon,15 Feb 2016 15:26:54 UTC All use subject to JSTOR Terms and Conditions
MALE LOSERS AND WINNERS AFTER SEPARATION/DIVORCE 251 port the instrumental variables estimates using the twice-lagged income measure. To minimize the impact of outliers, we truncated the dependent variable to the range [-2,2], effectively capping losses at 86.5 percent and capping gains at 639 percent of pre-disruption income or living standards.3 Our final models include control variables for change in job tenure, for age (if age enters the level model as a quadratic, it remains as a linear predictor in the difference model), dummy variables for calendar year, and the indicator for new partner at time t + 1. To simplify the presentation of results, we report only the coefficient and standard error for D, the indicator for union disruption. We also report the economic impact of disruption, which equals exp(yD) - 1. MEASUREMENT MEASURING INCOME We used the following measures of household income (all adjusted to 1990 dollars using the Consumer Price Index [CPI-U]) as our dependent variables: TOTAL PRIVATE HOUSEHOLD INCOME. This is our measure of "gross income," and equals the combined earned income received by all household members from all employment (including wages, salaries, farm income, business income, income from market gardening, and income from roomers and boarders), assets, and private transfers (including alimony and child support) before taxes and government transfers. It also includes the imputed rental value of housing. POST-GOVERNMENT HOUSEHOLD INCOME. This equals the sum of household income after income taxes (including state income taxes and the Earned Income Tax Credit), payroll taxes (health, unemployment, and retirement), and public transfers. Public transfers include AFDC, SSI, and other "welfare" programs, Social Security, unemployment compensation, worker's compensation, and the face value of food stamps. POST-GOVERNMENT HOUSEHOLD INCOME AFTER SUPPORT PAYMENTS. Respondents in all survey years report the total amount of money paid during the previous calendar year toward the support of persons outside the household, including voluntary out-transfers and mandatory support payments. We subtracted this total support amount from household post-government income to create a measure of "net" household income. Beginning in 1985, respondents who reported paying support were then probed specifically for child-support and alimony payments.4 We used these data to create an additional measure of household income that subtracts only these compulsory family-support payments from household post-government income, thus allowing us to distinguish these compulsory payments from any additional support paid on an informal or voluntary basis for the later subsample. Any missing data on support payments were imputed to be zero. Some caution is necessary when dealing with men's reports of support payments. Previous research has shown that divorced men who pay child support report higher levels of transfers than their former wives report receiving (Braver, Fitzpatrick, and Bay 1991; Peters et al. 1993; Schaeffer, Seltzer and Klawitter 1991), and when third-party verification is available (e.g., when the 3 The untruncated results produce much more negative estimates of the average impact of union dissolution because they are strongly influenced by those separating men who suffer precipitous income declines between time t - 1 and time t + 1. We experimented with alternative measures that minimize the influence of these outliers, including a measure based on Duncan and Hoffman (1985) that placed a floor of $3,000 on nominal income levels. The results for net income using this measure were similar to those presented here; however, the income floor obscured important nuances in the specific mechanisms that structured income change. 4 The wording of this sequence from the 1993 interview is as follows: G103. In 1992, did you give any money toward the support of anyone who was not living with you at the time? G106. How much money was that altogether in 1992? G107. Was any of that child support? G108. How much did that child support amount to in 1992? G109. Was any of the money you gave in 1992 alimony? GI 10. How much did that alimony amount to in 1992? This content downloaded from 129.96.252.188 on Mon, 15 Feb 2016 15:26:54 UTC All use subject to JSTOR Terms and Conditions
252 AMERICAN SOCIOLOGICAL REVIEW courts administer the collection and distribu- the amount of support they paid by 50 per- tion of support payments)the wives'reports cent,an amount that exceeds current esti- correspond more closely to the third-party mates of overreporting based on available reports (Schaeffer et al.1991).Part of the evidence).5 We report results based on the discrepancy between ex-partners'reports of adjusted measure of support along with re- money transfers may come about because sults based on men's actual reports. payers and recipients are likely to hold dif- ferent views on what constitutes support. MEASURING STANDARD OF LIVING: Those who pay support may be inclined to EQUIVALENCE SCALES report all expenditures on their children and former partners,including court-ordered tu- Nominal income is an important indicator of ition and health care payments as well as in- the economic status of a household,but an formal transfers,gifts,and loans,while re- imprecise measure of living standards. cipients may be inclined to define support as Economies of scale are important to the esti- compliance with court-ordered cash trans- mation of changes in men's standard of liv- fers (Schaeffer et al.1991).Reporting dis- ing following separation because separation crepancies can also come about when men entails not only a change in income,but also pay support to children from earlier liaisons a change in household composition.Nomi- in addition to the children residing with their nal income change would correspond per- most recent ex-partner.Even taking these fectly with change in living standards only factors into account,men who fail to com- if economies of scale were infinite,that is,if ply fully with court-ordered support have it cost no more to shelter,clothe,and feed a strong incentives to exaggerate support pay- family of four or six than it does to provide ments.Published reports suggest that men for a single-person household.Clearly this overreport by as little as 13.5 percent is not true. (Schaeffer et al.1991)to as much as 27 per- Zero economies of scale,in contrast, cent (Braver et al.1991). would imply that it costs twice as much to Using these data,supplemented with the provide for a couple as for an individual, preliminary PSID data for 1994-1996,we four times as much to provide for a family compared transfers paid by men to transfers of four,and so on.If there were no scale received by women for a small number of economies,living standards could be deter- couples (N =58)in which complete infor- mined by dividing income by the number of mation on alimony and child-support trans- persons in the household,and change in per fers was obtained from both ex-partners at capita income would correspond perfectly the interview following the separation.We with change in living standards.However, also used unmatched data to compare reports the assumption of zero economies of scale is of transfers received by custodial mothers to as implausible as the assumption of infinite reports of transfers paid by noncustodial fa- scale economies. thers in the first year following separation. When plausible assumptions about scale Our findings concur with previous research: economies are used to adjust household in- The proportion of men who claimed to have come by a weight based on the number of paid support was only slightly higher than "scale equivalents"in the household,the re- the proportion of women who claimed that sulting estimates of changes in living stan- they received support,but men's reports of dards are well within these two extremes. the amounts transferred exceeded women's There is no agreement on a single equiva- reports by 15 to 30 percent. lency scale,however,and the choice of scale To provide a reasonable set of bounds for is more consequential for analyses of men's the effects of transfers on living standards. outcomes than for analyses of women's out- we rely on alternative assumptions to con- struct two measures of post-transfer income. 5 We also constructed an adjusted measure that Our unadjusted measure assumes that men randomly assigns one-third of all reported trans- report payments accurately.Our adjusted fer payments to zero,leaving the remaining mea- measure deflates reported payments by one- sures as reported.Results based on this measure third,(i.e.,it assumes that men exaggerate were nearly the same as the results we report. This content downloaded from 129.96.252.188 on Mon,15 Feb 2016 15:26:54 UTC All use subject to JSTOR Terms and Conditions
252 AMERICAN SOCIOLOGICAL REVIEW courts administer the collection and distribution of support payments) the wives' reports correspond more closely to the third-party reports (Schaeffer et al. 1991). Part of the discrepancy between ex-partners' reports of money transfers may come about because payers and recipients are likely to hold different views on what constitutes support. Those who pay support may be inclined to report all expenditures on their children and former partners, including court-ordered tuition and health care payments as well as informal transfers, gifts, and loans, while recipients may be inclined to define support as compliance with court-ordered cash transfers (Schaeffer et al. 1991). Reporting discrepancies can also come about when men pay support to children from earlier liaisons in addition to the children residing with their most recent ex-partner. Even taking these factors into account, men who fail to comply fully with court-ordered support have strong incentives to exaggerate support payments. Published reports suggest that men overreport by as little as 13.5 percent (Schaeffer et al. 1991) to as much as 27 percent (Braver et al. 1991). Using these data, supplemented with the preliminary PSID data for 1994-1996, we compared transfers paid by men to transfers received by women for a small number of couples (N = 58) in which complete information on alimony and child-support transfers was obtained from both ex-partners at the interview following the separation. We also used unmatched data to compare reports of transfers received by custodial mothers to reports of transfers paid by noncustodial fathers in the first year following separation. Our findings concur with previous research: The proportion of men who claimed to have paid support was only slightly higher than the proportion of women who claimed that they received support, but men's reports of the amounts transferred exceeded women's reports by 15 to 30 percent. To provide a reasonable set of bounds for the effects of transfers on living standards, we rely on alternative assumptions to construct two measures of post-transfer income. Our unadjusted measure assumes that men report payments accurately. Our adjusted measure deflates reported payments by onethird, (i.e., it assumes that men exaggerate the amount of support they paid by 50 percent, an amount that exceeds current estimates of overreporting based on available evidence).5 We report results based on the adjusted measure of support along with results based on men's actual reports. MEASURING STANDARD OF LIVING: EQUIVALENCE SCALES Nominal income is an important indicator of the economic status of a household, but an imprecise measure of living standards. Economies of scale are important to the estimation of changes in men's standard of living following separation because separation entails not only a change in income, but also a change in household composition. Nominal income change would correspond perfectly with change in living standards only if economies of scale were infinite, that is, if it cost no more to shelter, clothe, and feed a family of four or six than it does to provide for a single-person household. Clearly this is not true. Zero economies of scale, in contrast, would imply that it costs twice as much to provide for a couple as for an individual, four times as much to provide for a family of four, and so on. If there were no scale economies, living standards could be determined by dividing income by the number of persons in the household, and change in per capita income would correspond perfectly with change in living standards. However, the assumption of zero economies of scale is as implausible as the assumption of infinite scale economies. When plausible assumptions about scale economies are used to adjust household income by a weight based on the number of "scale equivalents" in the household, the resulting estimates of changes in living standards are well within these two extremes. There is no agreement on a single equivalency scale, however, and the choice of scale is more consequential for analyses of men's outcomes than for analyses of women's out- 5 We also constructed an adjusted measure that randomly assigns one-third of all reported transfer payments to zero, leaving the remaining measures as reported. Results based on this measure were nearly the same as the results we report. This content downloaded from 129.96.252.188 on Mon, 15 Feb 2016 15:26:54 UTC All use subject to JSTOR Terms and Conditions
MALE LOSERS AND WINNERS AFTER SEPARATION/DIVORCE 253 comes following separation.This is because for white and black men,the coefficient es- the estimates produced by different scales timates and the estimated percentage impact are more likely to diverge for the financially of separation on total private (gross)house- stronger partner in a union,and the estimates hold income,per capita income,and stan- are also more likely to diverge when there is dard of living.White men who separate lose a large change in household size. roughly one-third of their total private We rely primarily on two equivalency household income,on average,and cohabi- scales in our analyses:One is the"Expended tants lose only slightly more than married Linear Extension System"(ELES)weights. men do.The losses for married African a total consumption-based scale developed American men are similar to those for mar- by Merz et al.(1993)that assumes moderate ried white men,but black cohabitants lose economies of scale for small households and relatively more of their pre-disruption gross large scale economies for large households. income than do whites.In terms of per capita The other is a food-based scale computed income,the disadvantages of losing using U.S.Census poverty thresholds.The partner's income are offset by fewer depen- poverty scale takes the adult/child composi- dents in the new,smaller household.On av- tion into account as well as household size, erage,per capita income increases,espe- and it assumes substantial economies of cially for married men,who tend to be sepa- scale for small families.Empirically,this rating from larger households. means that the estimated advantage of mov- These results summarize two well-known ing from a two-or three-person household facts about men's economic status following to a single-person household is smaller when separation and divorce:Household income the poverty scale is used than when the declines,but income per person increases. ELES scale is used. The equivalency measures are more useful as estimates of change in living standards. MEASURING SPECIALIZATION IN THE By either equivalency measure,the short- term impact of separation on men's gross BREADWINNER ROLE living standards is modest.The impact is We measure specialization in market provi- significant only for African American sion using the male partner's share of total cohabitors,for whom the steep decline in household labor income,or his share of tax- nominal income following separation pro- able income if he had no labor income.We duces a significant decline in gross standard set the male share to zero if there was no tax- of living. able income;in a few cases where business The gross impact of separation is prima- losses resulted in no income or negative in- rily because of changes in labor market in- come we measured market specialization us- come.The net economic impact also in- ing the ratio of the male partner's annual cludes changes in government taxes,public work hours in the year preceding separation transfers,and support payments.The lower to the sum of the hours worked by both part- panel of Table 1 shows that economic de- ners.We grouped men by their share of pre- clines are more pronounced after accounting disruption household income:less than 40 for these nonmarket leveling mechanisms percent of the total,at least 40 percent but White men experience a decline in living less than 60 percent,at least 60 percent but standards of 11 percent to 20 percent,and less than 80 percent,and 80 percent or more unmarried men lose more than married men do.The results for married African Ameri- can men are negative but nonsignificant,and RESULTS again,black cohabitants experience the most serious declines in their living standards fol- EFFECT OF SEPARATION ON MEN'S AVERAGE INCOME AND STANDARD lowing separation.On average,the eco- nomic status of men who separate slips be- OF LIVING low their previous level of material well-be- We begin with estimates of the average im- ing in the immediate aftermath of union dis- pact of separation for married men and co- solution,and both market and nonmarket habitants.The top panel of Table 1 shows, mechanisms contribute to this decline. This content downloaded from 129.96.252.188 on Mon,15 Feb 2016 15:26:54 UTC All use subject to JSTOR Terms and Conditions
MALE LOSERS AND WINNERS AFTER SEPARATION/DIVORCE 253 comes following separation. This is because the estimates produced by different scales are more likely to diverge for the financially stronger partner in a union, and the estimates are also more likely to diverge when there is a large change in household size. We rely primarily on two equivalency scales in our analyses: One is the "Expended Linear Extension System" (ELES) weights, a total consumption-based scale developed by Merz et al. (1993) that assumes moderate economies of scale for small households and large scale economies for large households. The other is a food-based scale computed using U.S. Census poverty thresholds. The poverty scale takes the adult/child composition into account as well as household size, and it assumes substantial economies of scale for small families. Empirically, this means that the estimated advantage of moving from a two- or three-person household to a single-person household is smaller when the poverty scale is used than when the ELES scale is used. MEASURING SPECIALIZATION IN THE BREADWINNER ROLE We measure specialization in market provision using the male partner's share of total household labor income, or his share of taxable income if he had no labor income. We set the male share to zero if there was no taxable income; in a few cases where business losses resulted in no income or negative income we measured market specialization using the ratio of the male partner's annual work hours in the year preceding separation to the sum of the hours worked by both partners. We grouped men by their share of predisruption household income: less than 40 percent of the total, at least 40 percent but less than 60 percent, at least 60 percent but less than 80 percent, and 80 percent or more. RESULTS EFFECT OF SEPARATION ON MEN'S AVERAGE INCOME AND STANDARD OF LIVING We begin with estimates of the average impact of separation for married men and cohabitants. The top panel of Table 1 shows, for white and black men, the coefficient estimates and the estimated percentage impact of separation on total private (gross) household income, per capita income, and standard of living. White men who separate lose roughly one-third of their total private household income, on average, and cohabitants lose only slightly more than married men do. The losses for married African American men are similar to those for married white men, but black cohabitants lose relatively more of their pre-disruption gross income than do whites. In terms of per capita income, the disadvantages of losing partner's income are offset by fewer dependents in the new, smaller household. On average, per capita income increases, especially for married men, who tend to be separating from larger households. These results summarize two well-known facts about men's economic status following separation and divorce: Household income declines, but income per person increases. The equivalency measures are more useful as estimates of change in living standards. By either equivalency measure, the shortterm impact of separation on men's gross living standards is modest. The impact is significant only for African American cohabitors, for whom the steep decline in nominal income following separation produces a significant decline in gross standard of living. The gross impact of separation is primarily because of changes in labor market income. The net economic impact also includes changes in government taxes, public transfers, and support payments. The lower panel of Table 1 shows that economic declines are more pronounced after accounting for these nonmarket leveling mechanisms. White men experience a decline in living standards of 11 percent to 20 percent, and unmarried men lose more than married men do. The results for married African American men are negative but nonsignificant, and again, black cohabitants experience the most serious declines in their living standards following separation. On average, the economic status of men who separate slips below their previous level of material well-being in the immediate aftermath of union dissolution, and both market and nonmarket mechanisms contribute to this decline. This content downloaded from 129.96.252.188 on Mon, 15 Feb 2016 15:26:54 UTC All use subject to JSTOR Terms and Conditions