Journal of Gerontology: SOCIAL SCIENCES1999, Vol. MB, No. 4, S2O7-S218

Copyright 1999 by The Gerontological Society of America

The Retiree Identity: Gender and Race Differences

Maximiliane E. Szinovacz1 and Stanley DeViney2

'The Glennan Center for Geriatrics and Gerontology, Eastern Virginia Medical School,department of Social Sciences, University of Maryland Eastern Shore.

Objectives. This study's focus on the retiree identity complements research on institutionalized retirement criteria. We testwhether tensions between individuals' own life circumstances and institutionalized retirement criteria manifest themselves inthe extent to which individuals assume a retiree identity and in the overlap between self-definitions and attainment of institu-tionalized retirement criteria.

Methods. The analyses rely on logistic regressions and are based on data from the National Survey of Families andHouseholds (N=l,633).

Results. Labor force participation serves as the main basis for the retiree identity, but other life experiences (work history,disability, spouse's retirement, economic status, family history) also contribute to individuals' self-identification as retiree andthe "fit" between self-identification and attainment of institutionalized retirement criteria. Men's retirement identity seems moreclosely tied to their attainment of institutionalized retirement criteria and a continuous and successful work career, whereas avariety of life circumstances impinge on women's retiree definitions. Moderately disabled African Americans are more likely toself-identify as partly or fully retired.

Discussion. While employment constitutes the main reference for individuals' self-definition as retiree, other life circum-stances also have some influence. Full understanding of retirement transitions will require more attention to the meaning con-text of retirement among divergent population groups.

RETIREMENT has become an institutionalized and ac-. cepted life course transition in postindustrial societies

(Kohli, 1986; Quadagno & Hardy, 1996). Consequently, the re-tirement transition process, as well as its precursors, are guidedby a set of expectations that specify the meaning of retirement,its timing, and, to some extent, appropriate role behaviors(Ekerdt, 1986). Though essentially derived from economic andpolitical initiatives (employer pensions, Social Security), theseexpectations also constitute a basis for individuals' identifica-tion with the retiree role. For example, anticipation of retire-ment leads to psychological disengagement from the work role(Ekerdt & DeViney, 1993). However, some persons or groupsmay not assume a retiree identity when they leave the laborforce or become beneficiaries.

Retirement research has emphasized institutionalized, or ob-jective, retirement criteria such as labor supply or pension re-ceipt, but paid relatively little tribute to individuals' own defini-tions or their subjective retiree identity. For example,investigations documenting the complexity and dynamics of re-tirement transition processes typically center on such events aspartial retirement (Burr, Massagli, Mutchler, & Pienta, 1996;Gustman & Steinmeier, 1984a, 1984b, 1986; Honig, 1985;Honig & Hanoch, 1985), bridge jobs (Quinn & Kozy, 1996),and work after retirement (Hay ward & Grady, 1990; Morrow-Howell & Leon, 1988) or after the receipt of pension benefits(Gendell & Siegel, 1996; Holden, 1988; lams, 1986; Wray,1996). Concerns about the definition of retirement (Palmore,1965) stimulated studies exploring the extent of overlap amongdivergent retirement criteria (Ekerdt & DeViney, 1990; Parnes& Less, 1985; Zsembik & Singer, 1990). However, these inves-tigations tend to view disparity among retirement criteria moreas a validity rather than as an identity issue. Others explored

specific subgroups of retirees who do not fit well into tradi-tional retirement categories; for instance, available nonworkers(Hardy, 1991), disabled nonworkers (Gibson, 1987, 1991,1993), or retired housewives (Adelmann, Antonucci, &Jackson, 1993; Heyman, 1970). These latter studies providesome initial evidence that individuals' retiree identity is notonly influenced by institutionalized criteria (labor force with-drawal, pensions) but also reflects complex life course pro-cesses such as work, health, or family history (Belgrave, 1988,1989; Gibson, 1987, 1991, 1993; Hardy, 1991; Hayward &Grady, 1990; Zsembik & Singer, 1990).

While it is certainly essential to explore the precedents andconsequences of labor force exit or benefit receipt, full under-standing of retirement processes also requires examination ofpersons' identification with the retiree role. Whether and at whattime during the objectively and institutionally defined retirementprocess individuals assume the retiree identity may have impor-tant consequences for their behaviors, attitudes, and, ultimately,their well being. For example, studies show that early retireesare more prone to reenter the labor market (Hayward, Hardy, &Liu, 1994), that perceptions of untimely retirement hinder retire-ment adjustment (Szinovacz, 1989), or that low identificationwith one's work role can undermine job satisfaction and feelingsof self-worth (Mutran, Reitzes, Bratton, & Fernandez, 1997).Similar outcomes may characterize individuals who experiencedissonance between their objective retirement status and theirsubjective identification with the retiree role.

As retirement transitions become increasingly blurred(Hayward, Crimmins, & Wray, 1994; Henretta, 1997;Quadagno & Hardy, 1996), disparity between objective retire-ment criteria and individuals' serf-definitions as retiree may be-come more common. This could mean that larger population

S207

S208 SZINOVACZAND DEVINEY

groups will be less susceptible to program and policy incentivesgeared toward retirees. For example, the "retiree busy ethic"(Ekerdt, 1986) secures a supply of volunteers and leisure con-sumers, but such activities may not appeal to individuals whoreject the retiree role. It could also imply that more retirees willpursue activities (e.g., reentry into the labor market) that are in-consistent with societal expectations and policies. Indeed,blurred retirement transitions could be viewed as manifesta-tions of clashes between institutional expectations and personalpursuits. Thus, a focus on retirement identity and its sourcescan contribute to a more in-depth understanding of the reasonsfor such blurred boundaries.

Distinction between retirement as defined by institutional-ized criteria (e.g., labor force withdrawal) and the retiree iden-tity draws further attention to tensions between institutionalizedprograms or policies and subgroup experiences (Kohli, 1986).Such tensions are evident from research demonstrating that dis-parity between institutionalized and subjective retirement defi-nitions prevails among selected societal groups, such as womenand minorities (Ekerdt & DeViney, 1990; Gibson, 1993). Forexample, rejection of the retiree identity among older AfricanAmericans who have exited the labor force can be traced to theconsequences of racial discrimination, both in the labor marketand in the health care system (Gibson, 1991; Hay ward,Friedman, & Chen, 1996), and ambiguities in women's retire-ment definitions reflect conflicts between the work require-ments and family wage assumptions implied in retirement poli-cies and expectations about women's family and gender roles(Belgrave, 1989). Analyses of gender or race differences in ob-jective retirement processes can identify the sources (e.g, dis-ability) and economic consequences (e.g., reduced benefits) ofatypical retirement transitions among societal subgroups(Hayward et al., 1996; Meyer, 1990), but not their more per-sonal ramifications (e.g., acceptance of the retiree role and itsimplications for well-being). Consideration of subjective retire-ment definitions would, therefore, seem particularly importantto further understanding of retirement processes and their con-sequences among women and minorities.

In this article, we aim to enhance understanding of individu-als' retiree identity by exploring the conditions under which in-dividuals define themselves as retired and the conditions thatcontribute to disparity between objective and subjective retire-ment definitions. Examination of the precursors of retirementself-definitions can offer insights into potential tensions be-tween institutionalized retirement processes and subjective ex-periences by pinpointing which life course experiences andconditions, other than institutionalized criteria, contribute to theretiree identity. Also, investigation of discrepancies betweensubjective and objective retirement definitions can show underwhich conditions objective retirement transitions promote orinhibit assumption of a retiree identity. Race and gender effectson retirement self-definitions and on disparity between objectiveand subjective retirement definitions, as well as race and genderdifferences in the conditions leading to such self-definitions anddisparity, can reveal whether observed subgroup differences de-rive from exposure to specific life course experiences.

We proceed from the assumption that the retiree identityevolves from interpretative processes that link institutionalizedretirement criteria to individuals' current and past experiences,that is, what individuals make of the retiree role (Turner, 1962).

Individuals will usually define themselves as retiree if their lifecircumstances fit institutionalized retirement criteria, that is,when there is little tension between institutional programs andsubjective construction of the life course (Kohli, 1986). Indeed,previous research demonstrates considerable congruity betweenobjective and subjective retirement definitions (Ekerdt &DeViney, 1990; Parnes & Less, 1985). However, specific lifecircumstances impede the applicability of societal retirementcriteria. Lack of job opportunities or disability can bar access tothe labor force, and discontinuous work histories may renderthe exact time of retirement ambiguous (Hardy, 1991; Gibson,1991, 1993). Also, work obligations outside the labor force(e.g., family or volunteer work) may clash with expectations ofthe retiree role, and housewives especially may find that soci-etal retirement criteria offer few guidelines for their situation(Adelmann et al., 1993; Heyman, 1970). Under such condi-tions, interpretative processes grounded in individuals' lifecourse experiences are likely to play a paramount role in retire-ment definitions. Thus, the more individuals' cultural and lifecourse experiences depart from institutionalized patterns, themore their retirement definition processes will shift from mereroletaking to rolemaking.

Based on this argument, four sets of factors can be expectedto influence retirement definitions: (a) attainments linked to in-stitutionalized retirement criteria; (b) life course experiencespertaining to individuals' work history as well as to the impor-tance of the work career in their lives; (c) socioeconomic status;and (d) membership in cultural or subcultural groups that de-fines the meaning context of work and retirement, influencesaccess to work and retirement benefits, and, to some extent,structures individuals' life course experiences.

The institutionalization of retirement in modern societies canbe traced to the implementation of Social Security and em-ployer pensions (Henretta, 1994; Kohli, 1994). Retirement, asan institution, delineates the rights and duties associated withthe retiree status, as well as the timing of the transition toretirement. At least in the United States (Sainsbury, 1996;Williamson & Pampel, 1993), retirement entitlements both inthe public (Social Security) and private sector (employer pen-sions) are tied to individuals' work careers (years in the laborforce or firm, earnings) and serve as a definitional basis for re-tirement status (Ekerdt, Deviney, & Kosloski, 1994). Becauseof their regulatory function (Kohli, 1986), these entitlements re-quire at least partial withdrawal from the labor force.Individuals who fail to meet this condition are penalized (re-duced benefits). In addition, retirement as an institution estab-lishes age expectations for the retirement transition (what Kohli,1986, calls chronologization of the life course). Despite theabolishment of mandatory retirement, such expectations persistthrough age constraints in the Social Security program. Thus,institutionalized retirement criteria combine attainment of enti-tlements (receipt of Social Security or pensions), fulfillment ofobligations (continuous employment, withdrawal from thelabor force at age 62 or 65), and achievement of retirement age.However, premises of the male breadwinner role and the familywage that are inherent in many Social Security and pensionschemes (Sainsbury, 1996) sever, to some extent, the direct linkbetween individuals' own work careers and their retirement en-titlements. Consequently, selected features of individuals' mari-tal history (length of marriages, widowhood), as well as their

RETIREMENT DEFINITIONS S209

spouses' work careers and benefit entitlements, also becomepart of institutionalized retirement criteria.

Inherent in institutionalized retirement is the presumption ofa life course which is characterized by fairly uninterrupted workcareers, withdrawal from the labor force at a certain age, and afamily wage. Whenever these criteria are not met, individualseither may not qualify for benefits at all or may receive some-times substantially reduced benefits (DeViney & Solomon,1995; Meyer, 1990; O'Rand, 1988). These penalties, as well asdiscrepancies between the "normative life course" (as reflectedin institutionalized criteria) and individuals' own life course ex-periences, may give rise to subjective retirement definitions thatdiffer from institutionalized criteria. Several conditions accountfor such discrepancies. First, institutionalized public and privatecriteria may diverge. Employer pensions and employer retire-ment incentives sometimes induce individuals to leave theirjobs before they are eligible for Social Security benefits(Applebaum & Gregory, 1990; Hanks, 1990; Quinn & Kozy,1996). In such cases, persons may either retire early, that is,leave the labor force prior to the age criteria set by SocialSecurity, or take on a bridge job until they qualify for SocialSecurity benefits. In this situation, assumption of the retireeidentity is ambiguous—it may occur either at the time of with-drawal from the main career or at the time of final withdrawalfrom the labor force.

Second, retirement transitions diverge increasingly from thetraditional abrupt and complete labor force withdrawal at theinstitutionalized (Social Security) retirement age. Some indi-viduals may continue employment (perhaps at reduced hours orin bridge jobs) past the usual retirement age, either becausethey enjoy their careers or because of financial need. Onceagain this may create discrepancies between institutionalizedretirement criteria (especially age and benefit receipt) and aperson's self-definition as retiree. In addition, the self-employedtend toward a more gradual and blurred withdrawal from thelabor force than those employed by others (Mutchler, Burr,Pienta, & Massagli, 1997), which is likely to render a clear de-marcation between work and retirement more difficult. Othersmay be forced out of the labor force (either for health reasonsor for lack of job opportunities) before selected institutionalizedretirement criteria are met. These individuals may meet the re-tirement criterion of withdrawal from the labor force but rejecta retiree identity because of the involuntary nature of such with-drawal (Gibson, 1993; Hardy, 1991).

In addition, the retiree identity is likely to reflect the impor-tance and continuity of the work role in individuals' lives.Persons with discontinuous work histories may find it difficultto demarcate the exact time of retirement or view themselves aspartially retired (Honig, 1985; Honig & Hanoch, 1985),whereas those with marginal attachments to the labor force butheavy involvement in nonpaid labor (e.g., housewives) may re-ject either the retiree identity altogether, or base it on other lifecircumstances such as the retirement of the spouse (Adelman etal., 1993; Heyman, 1970).

Life circumstances that do not fit the patterns implied in institu-tionalized retirement programs predominate among selected popu-lation groups, especially women and minorities. This is evidentfrom race and gender differences in the meaning of and reasons forretirement (Gibson, 1993; Hanson & Wapner, 1994), as well as inthe prevalence of inconsistency among retirement criteria

(Belgrave, 1988,1989; Ekerdt & DeViney, 1990; Gibson, 1987,1991,1993; Hatch, 1990; Zsembik & Singer, 1990). One factorcontributing to ambiguities in the retirement definition of women istheir heavy involvement in family roles and its consequences.Women's involvement in family roles competes with their role asemployee and often results in discontinuous work histories (Pienta,Burr, & Mutchler, 1994) that render distinctions between work in-terruptions and retirement more difficult (Belgrave, 1989; Sherrell& Halleen, 1996; Zsembik & Singer, 1990). Such discontinuouswork is likely to prevail among women with early and/or multiplechildbearing histories. Probably because of continuing family workobligations, women also attach a different meaning to retirementthan men. Whereas men view retirement as a distinct life transition(new beginning, imposed disruption, transition to old age), womenstress continuity in their pre- and post-retirement lives (Hanson &Wapner, 1994). Discontinuities in their work histories, as well assex discrimination in the labor market, often result in low retire-ment incomes among women. Consequently, divorced womenwho do not qualify for husbands' benefits may have to postponelabor force withdrawal past the institutionalized retirement age tosecure adequate retirement incomes (Morgan, 1992; DeViney,1995; Appelbaum & Gregory, 1990; DeViney & Solomon, 1995).

Assumption of the retiree identity may be adapted subjec-tively to fit societal prescriptions about gender roles in the mari-tal system. Because husbands continue to be viewed as mainproviders, they may reject the retiree role or postpone retirementbeyond institutionalized retirement age as long as their wives arestill employed. Research indicates, for example, that a spouse'slabor force participation serves as a strong incentive to remain inthe labor force (Ruhm, 1996; Szinovacz, 1989). Conversely, re-tired husbands exerting pressure on their wives to retire as wellmay contribute to inconsistency in retirement definitions, ifwives leave the labor force before they reach pension or SocialSecurity eligibility (Skirboll & Silverman, 1992; Szinovacz,1989). Wives may self-define as retirees on the basis of theirhusbands' retirement status (Belgrave, 1989), and widows maybase retirement self-definitions on the receipt of widow pen-sions. Thus, spouse's retirement and widowhood can promoteself-definitions as retiree and disparity between objective andsubjective retirement definitions. Because men's breadwinnerrole places primary importance on their work achievements, weexpect family and marital life experiences to have a more pro-found influence on women's than on men's retiree identity.

Discrimination in the labor market, as well as disability, havebeen linked to early labor force withdrawal and/or unemploy-ment in later life among minorities (Burr et al., 1996; Hayward &Grady, 1990; Hayward et al., 1996). Because such adverse cir-cumstances promote involuntary exits from paid employment,minorities may reject the retiree identity. For example, AfricanAmericans define themselves more in terms of a sick role than asretirees (Gibson, 1987,1991,1993; Hardy, 1991; Hayward &Grady, 1990) and frequently exit the labor force because of dis-abling health conditions (Hayward et al., 1996). Similarly,marginal attachment to the labor force among minorities through-out the life course may render the retirement transition (i.e., thetransition from intermittent employment or unemployment tofinal labor force withdrawal) blurred, if not meaningless.

In summary, we contend that the retiree identity emergesfrom the complex interplay between institutionalized retirementcriteria, individuals' life course experiences, and cultural con-

S210 SZINOVACZAND DEVINEY

text. To the extent that individuals' life course experiences donot fit the life course implied in institutionalized retirement pro-grams, clear demarcations of the retirement transition becomeambiguous. Consequently, such life course experiences (andsociocultural contexts that sustain and bolster exposure to theselife course experiences) will promote rejection of the retireeidentity when selected institutionalized retirement criteria aremet, or assumption of the retiree identity when institutionalizedcriteria are not met, thus contributing to disparity between se-lected institutionalized retirement criteria and individuals' re-tiree identity. Based on these premises, we hypothesize:

1. Women and African Americans will be less likely than menand Whites to define themselves as retirees and more likelyto encounter disparity between subjective retirement and ob-jective retirement criteria. These race and gender differenceswill be reduced (though not necessarily eliminated) when in-stitutionalized criteria and selected life course experiences(disability, work and family histories) are controlled.

2. Selected characteristics of individuals' life course experi-ences will influence the retiree identity, as well as disparitybetween self-identification as retiree and selected institution-alized criteria, beyond the effect of (other) institutionalizedcriteria (i.e., when such criteria are controlled). Specifically,rejection of the retiree identity and disparity between subjec-tive retirement and objective retirement criteria will be morecommon among individuals who have relatively low in-comes, are self-employed, experienced more and/or longerwork disruptions, were seeking employment after their lastjob, receive public assistance, and are disabled.

3. Attainment of institutionalized retirement criteria (laborforce withdrawal, pensions, Social Security benefits, institu-tionalized retirement age, spouse's retirement) will promoteself-identification as retiree.

4. Disparity between self-definition as retiree and selected insti-tutionalized retirement criteria (e.g., self-definition as retireeamong individuals who are still in the labor force) will be re-duced when other institutionalized criteria (benefits, retire-ment age) are met. However, receipt of employer pensions maylead to self-identification as retiree among current workers.

5. The impact of institutionalized retirement criteria and selectedlife course experiences on the retiree identity, and on disparitybetween self-identification as retiree and selected institutional-ized criteria, will vary by race and gender. Spouse's retirement,the experience of divorce or widowhood, number of children,and timing of births will be more influential for women thanfor men. Disability is expected to exert greater influence on theretiree identity and on disparity between objective and subjec-tive retirement definitions of Blacks than of Whites.

METHODS

Sample.—This study relies on secondary analyses of theNational Survey of Families and Households (NSFH). TheNSFH used a multistage area probability sample that was aug-mented by oversampling of minorities, one-parent families,families with stepchildren, cohabitors, and recently married in-dividuals. The original sample (including the oversample) con-sisted of 33,869 addresses. These addresses were screened first,and successfully screened households were approached for in-

terviews. First wave interview response rates for the success-fully screened households were 73.5% for the main sample,and 76.8% for the oversample, resulting in a total sample of13,008 respondents. (For a detailed description of the study, seeSweet, Bumpass, & Call, 1988). The Wave 2 interviews werecompleted by 10,008 respondents, representing an attrition rateof 23.2%.

The two waves of data were collected in 1987/1988, and in1992 to 1994. Data were collected from randomly selectedhousehold members aged 19 years or older, their spouses and,in some cases, tertiary household members. In both waves, re-spondents were interviewed and also completed a self-adminis-tered questionnaire. Interviews lasted typically between 70 min-utes and two hours.

The subsample used in the present analyses consists of respon-dents aged 50 to 75 years at Time 2 who reported some employ-ment during their life and responded to the retirement self-defini-tion questions. Reliance on Time 2 responses was necessarybecause only this wave contained questions on respondents' self-definition as retiree. Approximately 14% of respondents failed toanswer questions about their self-perception as retiree. Becauseimputation of missing data for this dependent variable may leadto overestimates of relationships with the independent variables,missing cases were excluded. This exclusion may compromisegeneralizability slightly. Comparisons between respondents whodid and those who did not answer the retirement self-definitionquestions reveal that missing cases are somewhat overrepre-sented among individuals with lower incomes and health prob-lems. However, the differences are quite small (r < .10), andthere are no significant differences between respondents andnonrespondents by gender, race, marital status, receipt of pen-sions, or unemployment. Also excluded were minorities otherthan African Americans. Subgroups of these minorities(Hispanic, Asian, etc.) are too small to allow separate analysesand the combined group would represent populations with quitedivergent employment histories and socioeconomic back-grounds. The resulting subsample size is 1,633, unweighted.

Measures.—The dependent variables are individuals' self-definition as retiree as well as two measures of disparity be-tween subjective retirement definition and objective retirementcriteria, namely, labor force participation and benefit receipt.Self-reported retirement was derived from responses to thequestion: "At this time, do you consider yourself partly retired,fully retired, or not retired at all?" Labor force participationwas derived from respondents' reported work hours at Time 2.Those who reported some work hours during the past weekand/or indicated that they usually engage in employment activi-ties were considered employed, those with 0 work hours re-tired. Receipt of benefits was based on questions about pensionreceipt and income from diverse sources. Respondents who in-dicated that they currently received a pension, or reported in-come from private pensions, were coded 1. Also coded 1 wereindividuals who reported Social Security income and who wereeligible for Social Security (married, divorced, or never marriedrespondents aged 62 and older, and widows aged 60 and older).

Computations of disparity between labor force participationand self-reported retirement rely on employment status andwork hours. Respondents were deemed to be congruent if they:(a) reported to be employed and indicated they were "not at all

RETIREMENT DEFINITIONS S211

retired;" (b) classified themselves as "partly retired" andworked less than 40 hours per week; and (c) reported no em-ployment and considered themselves "fully retired." There aretwo groups whose subjective and objective retirement defini-tions are incongruent. The first group contains individuals whoindicate that they work 40 or more hours per week, but con-sider themselves partly or fully retired, as well as those whowork (regardless of work hours) and report full retirement. Thesecond group consists of respondents who are not employedbut identify themselves as not retired or only partly retired.Frequencies for this variable revealed that only very few per-sons fell into the first group (N=41 or 2.4%). These individuals,as well as respondents with inconsistent employment informa-tion (employed with 0 work hours, N=31) were excluded fromthe final analyses. Thus, Ate for this dependent variable aresmaller than Ns for the other dependent variables.

Disparity between benefits and self-reported retirement wascoded congruent if respondents reported either no benefits andnot being retired, or if they considered themselves retired(partly or fully) and received benefits. Those receiving benefitsbut classifying themselves as "not retired at all" constitute oneincongruent group (benefits, not retired), whereas individualswithout benefits but self-reported partial or full retirement makeup the second incongruent group (retired, no benefits). The dis-tribution of the dependent variables is shown in Table 1.

Independent variables.—Three sets of independent variableswere used in the analyses: background variables, institutional-ized retirement criteria, and selected life course experiences andcircumstances.

Background characteristics were gender, race, education, re-spondent's and spouse's income, and receipt of public assis-tance. Education was coded in years, with a range from 0 to 18.Income was based on constructed measures provided in theNSFH. The income variable used here is respondent's total in-come from all sources. Income was coded in $100 units andlogged. Spouse's income was dichotomized ($30,000 and over= 1, other = 0). Information on public assistance was based onreported income sources. A code of 1 was assigned if respon-dents reported income from public assistance (welfare, AFDC,general assistance, food stamps, or SSI). Even though incomeand public assistance overlap, we include both variables be-cause receipt of public assistance may have different ramifica-tions for the retiree identity than low income per se.

The life experiences and conditions included in the analyseswere disability, marital status, fertility, and work history. Workdisability was based on a question concerning limitations in ac-tivities of daily living. Working for pay was one of the listed ac-tivities. Respondents indicated whether they were limited "a lit-tle" or "a lot." These responses were used as dummy variables,and respondents with no limitations are the omitted category.Because past experiences of divorce or widowhood may affectthe retiree identity, we conducted initial analyses that consid-ered the influence of length of time married, divorced, or wid-owed, as well as whether currently married persons had beendivorced or widowed. These analyses revealed that only pastexperiences of widowhood had significant effects on any of thedependent variables. Thus, marital status was captured throughfour dummy variables: currently widowed, currently divorced,never married, and currently married after widowhood. Married

individuals without experiences of widowhood are the refer-ence group. Fertility was measured with number of childrenborn and the timing of the first birth (early = age 21 or under,late = age 30 or over, reference = 22 to 29). An additional set ofdummy variables capturing the time interval between first andlast births did not contribute to the models once number andtiming of births was considered. The following work historyvariables were used in the analyses: total number of years in the

Table 1. Distribution of Dependent and Independent Variables

Independent VariablesBackground

Female

Black

Age 55-59

Age 60-61Age 62-64

Age 70-75

Life circumstances

Mild work disability

Severe work disability

Divorced

Not married

Widowed

Remarried after widowed

Education

Had first child late

Had first child early

Number of children born

Number of years in labor force

2+labor force exits

Had work disruptions since Wave 1

Self-employed Wave 1

Institutionalized criteria

Gets Social Security

Gets pension

Spouse receives pension

Full-time employed

Part-time employed

Looking for work

Spouse fully retired

Economic status

Income (logged)

Receives public income

Spouse has high income

Dependent VariablesRetirement self-definition

Fully retired

Partly retired

Not retired

Employment/self-definition disparity

Congruent

Not in labor force and self-defined as partly retired or not retired

In labor force and self-defined as partly retired or not retired

Benefit/self-definition disparity

CongruentNo benefits and self-defined as fully retired

Gets benefits and self-defined as not retired

M o r % SD

61.8% —

17.5% —

26.0% —

9.9% —

14.5% —

26.3% —

10.8% —

18.1% —

17.5% —

4.2% —

23.2% —

2.6% —

11.97 2.87

10.2% —

28.9% —

.28 2.04

30.21 13.69824.4% —

8.8% —

9.2% —

54.1% —

39.1% —

17.8% —

27.4% —

8.7% —

2.7% —

19.2% —

209.17 256.356

5.0% —

12.8% —

51.4% —

15.2% —

33.4% —

84.4% —

11.2% —

2.4% —

88.0% —

5.0% —

7.0% —

Note: N= 1633, unweighted.

S212 SZINOVACZAND DEVINEY

labor force, number of work interruptions, occurrence of workinterruptions since Wave 1, and self-employment at Time 1.Number of years in the labor force is a continuous variable,which was constructed from work histories collected at bothwaves. Number of work interruptions was also based on workhistories from both waves. Because three or more interruptionswere quite rare, we created a dummy variable (1 = two or moreinterruptions, 0-1 interruption = reference). Recent work dis-ruptions may be indicative of blurred retirement transitions.This variable was derived from the Wave 2 work histories and

coded 1 = yes, 0 = no. Self-employment (at Time 1) was de-rived from reported self-employment income. Those with suchincome were coded 1, all others 0. Time 1 self-employmentwas used to avoid overlap with current labor force participation.

Institutionalized retirement criteria consist of age, benefits, andlabor force participation. To reflect distance from institutionalizedretirement age, the age variable was divided into five categories:55-59,60-61,62-64,65-69, and 70-75. Respondents who hadachieved retirement age (65-69) were used as reference, the otherage groups were treated as dummy variables. Information on pen-

Table 2. Multinomial Logistic Regressions Predicting Retirement Self-Definition (Relative Risk Ratios)c

PR vs NRd

BackgroundFemale .61**Black .62Age 55-59 .05**Age 60-61 .14**Age 62-64 .29**Age 70-75 .71

Life circumstancesMild work disability —Severe work disability —Divorced —Not married —Widowed —Remarried after widowed —Education —Had first child late —Had first child early —Number of children born —Number of years in labor force —2+ labor force exits —Had work disruptions since Wave 1 —Self-employed Wave 1 —

Institutionalized criteriaGets Social Security —Gets pension —Full-time employed —Part-time employed —Looking for work —Spouse fully retired —

Economic statusIncome (logged) —Receives public assistance —Spouse has high income —

N —

WaldF —

Model 1

FRvsNRd

.821.03

.03**

.08**

.30**1.29

—————————————

—

——————

———

—

—

PR vs FRe

.74

.611.581.74.99.55*

—————————————

—

——————

———

1633

29.68**

PR vs NRd

.73

.59*

.04**

.14**30**b

.85

1.24a

7.37**.50*.58.61.18*.99.38**

1.36.90

1.021.221.72*b

.94

——————

———

—

—

Model 2

FRvsNRd

.56**

.83

.02**

.07**

.30**1.31

1.40°17.35**

.48**

.51

.75

.35*"1.00

44**1.23.87**.99".92.48*".21**

——————

———

—

—

PR vs FRe

1.31.71

2.00*2.02

.99

.65

.89

.43**1.041.14.82.52.99.86

1.101.031.03**"1.333.60**4.39**

——————

———

1629

20.20**

PR vs NRd

.63

.5720**.53.37*.56

1.13*3.33**

.66

.50

.48*

.42

.95

.52——

1.05**—.90

1.32

5.00**3.19**b

.09**1.851.37.64

.92

.41"1.07

—

—

Model 3

FRvsNRd

.71

.81

.12**b

.22**

.651.17

1.58a

6.28**.78.46.85.87".97.48——

1.04**—.41.39

3.67**8.88**"

.00**

.02**"

.542.47**

1.031.97".96

—

—

PR vs FRe

.88

.701.62"2.38*

.57

.47*

.71

.53

.841.10".57*.49.98

1.08——1.01—

2.203.38**

1.36.36**

35.40**82.52**2.53*

.26**

.90

.21**1.11

1630

48.80**

Notes: PR = partly retired; NR = not retired; FR = fully retired.*p<.05;**p<.01."Race interaction p < .05.•"Gender interaction p < .05.cModel 1 tests race, gender, and age effects; Model 2 adds selected life experiences and conditions; Model 3 adds institutionalized retirement criteria and socioeco-

nomic background but excludes life circumstances that do not contribute to model fit. The analyses rely on two separate equations, one with not retired and the otherwith fully retired as base categories. The exponentiated value of a coefficient eb is the relative risk ratio for a one unit change in the corresponding variable; risk isbeing measured as the risk of the category relative to the base category (Stata, 1997, Vol. 2, p. 542). Relative risk ratios over 1.00 indicate a positive and those under1.00 a negative effect. All comparisons are based on the same equations, but the reference groups were altered (either not retired-first two columns for each model, orfully retired-last column for each model; see also Analysis section).

••Not retired is base category.eFully retired is base category.

RETIREMENT DEFINITIONS S213

sion receipt was derived from a direct question: "Are you now re-ceiving benefits from a pension or retirement plan?" Respondentswere instructed not to include Social Security in their answers. Inaddition, data on pension and (separately) Social Security incomewere obtained during the interviews. Receipt of pensions andSocial Security were coded as dummy variables (1 = reports apension and/or pension income; 1 = receives Social Security in-come). The same procedure was used for spouses' pensions.Because of multicollinearity with the marital status variables,spouses' receipt of Social Security was not included in the analy-ses. Labor force participation was coded into three dummy vari-ables: employed full-time (30 hours or more), employed part-time(1-29 hours), and job seekers (individuals who are currently notworking but had looked for work since their last employment).Those not employed and not looking for work serve as reference.For spouses, we included subjective retirement (is fully retired =1) rather than employment status. The two variables are highlycorrelated and spouse's retirement identity proved to be the betterpredictor in preliminary analyses.

Analyses.—The analyses consist of multinomial or dichoto-mous logistic regressions. We first tested models (Model 1 inthe tables) including only race, gender, and age to assess the in-fluence of these variables independent of the other independentvariables. Race X gender interactions proved to be not signifi-cant for all dependent variables. In the second model, lifecourse experiences (education, disability, marital status, fertil-ity, and work history) were added. We then tested models thatincluded race, gender, selected life course experiences, otherbackground variables, and the appropriate institutionalized re-tirement criteria. Criteria implied in the dependent variables(labor force participation and job seeking for analyses of dis-parity between labor force participation and retiree identity andbenefits, except spouse's pension, for analyses of disparity be-tween benefits and retiree identity) were excluded as indepen-dent variables. Based on these analyses, the final models(Model 3 in the tables) contain gender, race, age, disability, andmarital status, as well as those life course experiences and insti-tutionalized criteria that contributed to model fit.

For the analyses pertaining to retiree serf-definitions (Table 2),we are interested in contrasts between those who self-define asfully or partly, compared with those who self-define as not re-tired, as well as contrasts between self-definitions as full and partretirees. Consequently, the same models were estimated twice,once with "not retired" and once with "fully retired" as base cate-gory. Model fit for these two models is, of course, the same.

Due to small Ns, race interactions were tested only for dis-ability. Predicted gender interactions were tested both in the lifeexperience models and in the final models. In addition, we alsosystematically tested for gender interactions with all institution-alized retirement criteria and with the work history variables.Continuous variables used in interaction terms were centered(Aiken& West, 1991).

All analyses rely on weighted data. Sample selection for theNSFH relied on a complex survey design where selection intothe sample is not independent of other respondents. This can leadto an underestimation of error variances and covariances. Our useof sampling weights, strata, and PSUs in the analysis is intendedto reduce this problem. Both the logistic and multinominal logis-tic regressions were estimated using the complex survey design

methods provided by Stata (Stata, 1997). Specifically, Stata usespseudomaximum likelihood estimates for error variances. Foroverall model fit, Wald-F is produced and reported. We also esti-mated unweighted equations (not shown) and found the errorvariances to be smaller in these equations. We decided to followthe more conservative method and used the weights.

RESULTSIn view of the number of dependent variables and the com-

plexity of findings, we discuss results in conjunction with eachhypothesis (and across the tables) rather than for individual de-pendent variables. Gender and race interaction terms are notshown in the tables, but reported throughout the results section.

Race/gender.—Hypothesis 1 posits direct race and gendereffects on retiree self-definitions and on disparity between serf-definitions and attainment of institutionalized retirement crite-ria. The results do not confirm the hypothesized race effect, butprovide some support for gender effects. There are no signifi-cant race X gender interactions. Women are less likely than

Table 3. Logistic Regressions Predicting Disparity Between LaborForce Participation and Retirement Self-Definition (Odds Ratios)

BackgroundFemaleBlackAge 55-59Age 60-61Age 62-64Age 70-75

Life circumstancesMild work disabilitySevere work disabilityDivorcedNot marriedWidowedRemarried after widowedEducationHad first child lateHad first child earlyNumber of children bornNumber of years in labor force2+ labor force exitsWork disruptions since Wave 1Self-employed Wave 1

Institutionalized criteriaSpouse fully retired

Economic statusIncome (logged)Receives public assistance

N

WaldF

Not in labor force and self-defined as partlyor not retired vs congruent1'

Model 1

1.69*1.061.241.54.68.52*

——————————————

—

—

—

1564

3.53**

Model 2

.831.33.99

1.37.61.53*

1.07.97.68

1.65.55*.53.97.80".90.97.95**

1.191.75a

1.60

—

—

—

1560

3.42**

Model 3

.651.11.93

1.12.63.52*

.98

.79

.871.57"

.72——.97——.96**—

2.41**1.69

.59s

.66**

.86

1564

7.99**

*p<.05;**p<.01."Gender interaction p <.05.bBase category is congruent; for model descriptions see Table 2. Odds ratios

over 1.00 indicate a positive and odds ratios under 1.00 indicate a negative effect.

S214 SZINOVACZAND DEVINEY

men to define themselves as partly, rather than not, retired(Table 2). Gender also influences disparity between objectiveretirement criteria and self-definitions. Women are more pronethan men to reject an identity as full retiree when they are not inthe labor force (Table 3) and to provide serf-definitions that areincongruent with benefits receipt (Table 4). None of these gen-der effects is maintained when various life course experiencesand, especially, institutionalized criteria are included in themodels. This suggests that gender differences in the retireeidentity can be attributed to divergent life course experiencesand attainment of institutionalized retirement criteria.

Age.—Age exerts a significant influence on retirement self-definitions. Persons under institutionalized retirement age tend to

self-define as not retired, whereas those aged 70 and older favorself-definitions as fully, rather than partly, retired. However, theseeffects are only partially maintained when other institutionalizedretirement criteria are controlled (Model 3, Table 2). The age ef-fects vary somewhat by gender. Specifically, men under age 60are more prone than women to reject a retiree identity (b=l.79,p<.0l for fully vs not retired; and £=-1.85, p<.0l for partly ver-sus fully retired). The data also confirm age effects on disparitybetween self-definitions and institutionalized criteria. Disparitybetween self-definition and employment is particularly rareamong those aged 70 and older. However, individuals in this agegroup are more likely to reject a retiree identity after benefit re-ceipt, a finding that might indicate persistent rejection of the re-tiree identity among some individuals. Nevertheless, most indi-

Table 4. Multinomial Logistic Regressions Predicting Disparity Between Benefit Receipt(Pensions or Social Security) and Retirement Self-Definition (Relative Risk Ratios)b

BackgroundFemaleBlackAge 55-59Age 60-61Age 62-64Age 70-75

Life circumstancesMild work disabilitySevere work disabilityDivorcedNot marriedWidowedRemarried after widowedEducationHad first child lateHad first child earlyNumber of children bornNumber of years in labor force2+ labor force exitsWork disruptions since Wave 1Self-employed Wave 1

Institutionalized criteriaSpouse receives pensionFull-time employedPart-time employedLooking for workSpouse fully retired

Economic statusIncome (logged)Receives public assistanceSpouse has high income

N

WaldF

Model 1

No benefits and selfdefined as fully

retired vscongruent

2.57**1.451.201.751.271.27

——————————————

—————

———

—

—

- Gets benefits andself-defined as not

retired vscongruent

1.62*.68

2.41*2.62*2.88**2.26*

——————————————

—————

———

1633

2.30*

Model 2

No benefits and self-defined as fully

retired vscongruent

2.23*1.311.29a

1.831.051.03

1.472.58**

.50

.07*

.31**2.01

.931.281.541.02.98*

1.38.28.93

————

—

———

—

—

Gets benefits andself-defined as not

retired vscongruent

1.30.67

2.51*2.302.83*2.07*

.58

.19**1.61.33

2.95**6.62**1.082.15*1.091.19**a

.991.462.181.65

————

—

———

1629

9.61**

Model 3

No benefits and self-defined as fully

retired vscongruent

1.881.322.241.641.731.09

1.402.02*1.60.04**.74

3.11—1.56—1.021.011.51.46—

.41*

.09*

.422.54a

3.31**

.46**a

2.51.56

—

—

Gets benefits andself-defined as not

retired vscongruent

1.30.69.87.96

1.792.03*

.68

.31*1.07.33

2.28*5.25**—

2.19*—1.12'.97**

1.402.07—

1.317.75**2.855.00*

.89

1.12.96a

.63

1630

4.09

*/?<.05; **/?<.01."Gender interaction p <. 05.bBase category is congruent; for model descriptions and explanation of relative risk ratios see Table 2.

RETIREMENT DEFINITIONS S215

viduals (and men more than women) appear to apply institution-alized age criteria in their self-definitions.

Life course experiences.—According to Hypothesis 2, lifecourse experiences should affect the retiree identity, even wheninstitutionalized retirement criteria are controlled. In addition,selected life circumstances were predicted to have a greater im-pact on women than on men (Hypothesis 5). To assess the ef-fects of life circumstances, we first present a model that ex-cludes institutionalized retirement criteria and economicbackground (Model 2 in all tables), and then discuss the fullmodels (Model 3 in all tables). Note that selected life circum-stances with no effect on the dependent variable were droppedfrom some final models.

The data leave little doubt that disability contributes to retireeself-definitions, as well as to disparity between self-definitionsand institutionalized retirement criteria. Generally, severely dis-abled persons tend to assume a retiree identity (they are morelikely to define themselves as partly or fully rather than not re-tired, and as fully, rather than only partly, retired). Except for thecomparison between part and full retirement, these effects aremaintained when institutionalized retirement criteria and othercontrols are included in the model. Severe disability also pro-motes self-definitions as retiree among individuals who cur-rently receive no benefits, and it discourages self-definitions asnot retired among beneficiaries. However, disparity betweenself-definition as retiree and employment status is not contingenton severe disability. The effects of mild disability differ by raceand gender. Mild disability tends to promote self-definitions asretired, especially among Blacks (£=1 .82, p<.05 for partly vsnot retired; and b=1.88, p<.05 for full vs not retired) and amongmen (b=-l.lS,p<.05). Disparity between self-definitions andinstitutionalized criteria is not affected by mild disability.

Marital status also exerts some influence on the retiree iden-tity. Because several of these effects are confounded with bene-fit receipt, age, and employment status (as indicated by differ-ences between Models 2 and 3), we focus on relationships inModel 3. Widowhood is associated with serf-definitions of not,rather than partly, or fully, rather than partly, retired, and remar-ried widows (but not remarried widowers, £=3.89, p<.0l) tendto self-define as fully, rather than not, retired. In addition, bothwidowhood and remarriage after widowhood encourage self-definitions as not retired among beneficiaries. Never marriedindividuals are particularly unlikely to self-define as retiree be-fore benefit receipt, and never married women favor serf-defini-tions as part, rather than full, retirees (£=2.24, p<.05).

The data provide limited support for effects of fertility historyon retirement definitions. Self-definitions as retiree are less com-mon among those who delayed childbearing and those with morechildren, but these effects are eliminated once institutionalized re-tirement criteria are included. Delay of childbearing, as well ashaving more children, encourages rejection of a retiree identityamong individuals who receive benefits. Number of children bornis more important for rejection of the retiree identity among fe-male beneficiaries (£=.26,/K.01), and women who delayed child-bearing are also less likely to self-define as retirees when they arenot employed (£=1.33,/?<.05). This latter relationship is elimi-nated once institutionalized retirement criteria are controlled.

There is considerable evidence that past work history is re-flected in the retiree identity. Number of years in the labor force

contributes to the assumption of a retiree identity (partly orfully, rather than not, retired in Model 3). The significant gen-der interactions in Model 2 (£=-.06, p<.05 for partly vs fullyretired, and £=.06, p<.0l for fully vs not retired) become non-significant once the institutionalized retirement criteria are inthe equation. Furthermore, those with longer work historiestend to self-define as retirees when they are no longer in thelabor force and when they receive benefits.

Recent work disruptions (since Wave 1 of the survey) tend toencourage self-definitions as partly retired and to discourage anidentity as full retiree. The former effect (partly vs not retired) isstronger for men (£=-1.34, p<.05), whereas the latter effect (fullvs not retired) is stronger among women (£=-1.27, p<.05).Neither gender interaction is maintained in Model 3. In addition,those who experienced work disruptions between waves are lesslikely to self-define as retirees when they are not in the labor force.This effect is more pronounced for men in Model 2 (£=-1.74,/K.01), though the gender interaction is not significant in Model 3.

Former self-employment has divergent effects on the retireeidentity in Models 2 and 3. When institutionalized criteria are notcontrolled (Model 2), self-employment contributes to identifica-tions as part, rather than full, and as full, rather than not, retired.Once institutionalized criteria are included, only the former effectremains significant. Self-employment does not impinge on dis-parity between self-definitions and attainment of institutionalizedretirement criteria. Education and number of labor force exitshave no influence on self-definitions or disparity.

Overall, these findings substantiate that life circumstancesother than institutionalized retirement criteria impinge on theretiree identity. Individuals whose life experiences deviate frominstitutionalized patterns tend to reject the retiree identity. Theinfluence of fertility and, to some extent, marital status on theretiree identity seems to be tied to their relationships to attain-ment of institutionalized criteria, whereas work history effectsin general and disability effects among African Americans ap-pear to be relatively independent of such attainments. Family-related influences (fertility, marital status) play a somewhatgreater role among women, whereas disability and work dis-ruptions have more of an impact on men's retiree identity.

Institutionalized retirement criteria.—According to Hypothe-ses 3 and 4, attainment of institutionalized retirement criteria isexpected to enhance self-definitions as retiree and to encouragecongruence between self-definitions and institutionalized retire-ment criteria. The data lend some support to these hypotheses.

Receipt of Social Security and employer pensions signifi-cantly contributes to self-definitions as partly or fully retired,compared with being not retired, and pension receipt also dis-courages self-definitions of partly, rather than fully, retired. Theeffect of pensions varies significantly by gender (£=-1.67,/?<.01 for partly vs not retired; and £=-1.84,/?<.01 for fully vsnot retired). Receipt of pensions contributes more to men's thanto women's retiree identity. In contrast to Hypothesis 4, pen-sions did not influence disparity between employment and re-tiree self-definitions and were dropped from Model 3.

Current labor force participation exerts a strong influence onself-definitions. Full-time employed individuals tend to assumean identity as not retired, and they are more likely to see them-selves as only partly, rather than fully, retired. As might be ex-pected, part-time employment enhances self-definitions as

S216 SZINOVACZAND DEVINEY

partly, rather than fully, retired, and it also contributes to self-definitions as not retired, rather than fully retired. The latter ef-fect is considerably stronger for men than for women (b=-2.60,p<.05). In addition, employment status impinges on disparitybetween benefit receipt and self-definitions. Full-time employ-ees tend to assume an identity that corresponds with their em-ployment status rather than their benefit receipt. Specifically,they are less likely to self-define as retiree when they receive nobenefits and more likely to self-define as partly or not retired,even when they receive benefits.

Those who lack employment opportunities (job seeking) favordefinitions of part over full retirement, and they are also more likelyto self-define as not retired after receiving benefits. However,women (b=4.14, p<.0l) who looked for work especially tend toself-define as retirees before reaching benefit eligibility.

Our results also confirm the notion that spouse's employ-ment is reflected in the retiree identity. Spouse's retirement in-creases the probability of self-defining as fully, rather thanpartly or not, retired. Furthermore, spouse's retirement con-tributes to self-definition as retiree among persons who do notreceive benefits. Contrary to our expectations, these relation-ships do not vary by gender. On the other hand, spouse's retire-ment reduces the likelihood of disparity between self-definitionand labor force participation. Individuals with retired spousestend to adopt a retiree identity when they are not in the laborforce, and this trend is more pronounced for women than men(b=-l.lS,p<.01). While spouse's retirement has relatively con-sistent effects on retirement definitions, spouse's pensions haveonly a marginal impact. Those whose spouses receive pensionsare particularly unlikely to self-define as retirees without re-ceiving benefits themselves.

Generally, these findings indicate that institutionalized retire-ment criteria have been internalized and serve as strong anchorpoints for individuals' self-definition as retiree. Moreover, theeffects of institutionalized criteria are additive, that is, attain-ment of specific institutionalized retirement criteria contributesto the retiree identity, even when fulfillment of other criteria iscontrolled. Pensions and attainment of institutionalized retire-ment age contribute more to men's than women's retiree iden-tity, whereas lacking job opportunities and spouse's retirementtend to further disparity between women's self-definitions andtheir attainment of institutionalized criteria.

Economic status.—Variables considered under economicbackground are respondent's income, spouse's income, and re-ceipt of public assistance. Neither income nor spouse's incomeinfluence self-definitions as retiree, though respondents' in-come has some effect on disparity. Respondents with higher in-comes are more likely to assume a retiree identity when theyleave the labor force. In addition, self-definitions as retireeamong individuals who are not beneficiaries become less com-mon as income increases. This latter effect applies foremost tomen (gender interaction b=33,p<.05).

Current receipt of public assistance relates significantly to re-tiree self-definitions, as well as to disparity between self-defini-tions and selected institutionalized retirement criteria.Recipients of public assistance are less likely to self-define aspartly, rather than either not retired at all or fully retired. Self-definition as partly, rather than not, retired is particularly un-common among men (b=2.92, p<.05), whereas female public

assistance recipients are more likely than their male counter-parts to identify as fully, rather than not, retired (b=2.96, p<.0\).In addition, women who receive public assistance are less pronethan men to describe themselves as not retired once they re-ceive Social Security or pensions (b=-3.14,p<.05).

Overall, economic status plays a relatively minor role in retire-ment self-definitions. Among men, economic achievements ap-pear to reinforce self-identification in line with institutionalizedcriteria, whereas women on public assistance favor a retireeidentity.

CONCLUSIONThis study's focus on the retiree identity complements earlier

research that centered foremost on objective or institutionalizedretirement criteria. We conceptualize retirement self-identifica-tion as a complex process which reflects potential tensions be-tween individuals' own life circumstances and experiences onthe one hand, and the expectations derived from institutional-ized retirement programs and policies on the other hand (Kohli,1986). Such tensions were expected to derive from differencesbetween public and private retirement criteria (Social Securityvs employer pensions), from blurred or nonvoluntary laborforce exits, and from experiences that diverge from the lifecourse implied in institutionalized retirement programs and pre-dominate among minorities and women. They should manifestthemselves in the extent to which individuals assume a retireeidentity and in the overlap between self-definitions and attain-ment of institutionalized retirement criteria.

The findings confirm that institutionalized retirement criteriaserve as foundation for individuals' self-definition as retiree(Ekerdt et al., 1994; Setterstein & Hagestad, 1996). Eventhough all criteria (employment status, pension receipt, SocialSecurity income, and attainment of institutionalized retirementage) independently contribute to the retiree identity, employ-ment status apparently takes precedence over other criteria. Notonly is the association between employment and self-identifica-tion as retiree stronger than the relationships between the retireeidentity and other institutionalized criteria, but employment sta-tus also promotes self-definitions as retiree when other retire-ment criteria are not met. Moreover, the extent of individuals'employment is clearly reflected in their self-identifications.Because institutionalized programs imply a family wage, wefurther expected that spouses' retirement or pensions wouldcontribute to the retiree identity. The results confirm especiallythe effect of spouses' retirement on self-identifications (whichcovary strongly with spouses' employment status), suggestingperhaps a "couple retiree identity" among married individuals.

Incongruence between public and private criteria seems toplay only a minor role in retirement self-identification pro-cesses. More important for the retiree identity are labor forceexits that are not clearly demarcated or involuntary. The self-employed, as well as "discouraged" workers (those looking forjobs after their last employment), tend to describe themselvesas partly, rather than fully, retired. Financial attainments tend toencourage self-definitions as retiree foremost among men.

There is consistent evidence that disability influences the retireeidentity. Contrary to our expectations, disability tends to promoteself-definitions as retiree. For mild disability, this tendency pre-vails among Blacks and men. The race differences especially con-tradict earlier research. Closer comparison between previous re-

RETIREMENT DEFINITIONS S217

search (Gibson, 1987,1993) and our findings suggests that ques-tion wording may be responsible for the divergent results. TheNSFH question asks respondents whether they consider them-selves retired and provides a forced choice among full, partly, ornot retired with no differentiation within the "not retired" cate-gory, whereas in the National Survey of Black Americans used byGibson (1987) "respondents were asked to explain why they werenot working. The eight fixed choices were homemaker, physicallydisabled, welfare, student, retired, transportation problems, cannotfind a job, and do not wish a job" (p. 694). It is conceivable thatthe latter question evokes responses that refer to reasons for retire-ment rather than to the retiree identity itself. On the other hand,the NSFH may encourage self-definitions as retiree because itdoes not provide the option to identify as disabled. BecauseBlacks are more likely to retire for health reasons (Hayward et al.,1996), they may be more likely than Whites to indicate "disabled"as their main reason for not working, rather than choosing "re-tired," which may be associated with an expected and voluntaryretirement transition. We suggest that identification with the re-tiree role among disabled workers carries greater psychologicalbenefits than identification with a sick role. Retirement is a so-cially sanctioned role, and early retirement has become an ac-cepted life choice, whereas a long-term sick role is not sociallysanctioned and may have particularly negative connotationsamong older persons ("old and sick"). Within this context it is alsounderstandable that mildly disabled women are less likely to iden-tify with a retiree role than men with similar health limitations—the role of housewife (not retired and not in the labor force) maybe seen as more acceptable than that of early and involuntary re-tiree. Given the small frequencies, especially for minorities, theseconclusions clearly warrant further substantiation.

Past employment experiences exert a significant influence on theretiree identity. Work histories that fit the expectation of life-longwork with few disruptions apparently render the retirement transi-tion less ambiguous and promote acceptance of the retiree role.

We further hypothesized that life experiences other than workhistory and spouses' retirement or benefit status would influencethe retiree identity, especially among women. Support for thishypothesis is very weak. While marital status and fertility his-tory exert some influence on selected dependent variables, theseeffects are often inconsistent, and in some cases more pro-nounced, for men than women. The only finding in line with theexpected gender differences is that high fertility enhances rejec-tion of the retiree identity among women beneficiaries. Theoverall trend implied in the gender differences is that men's,compared with women's, retiree identity is more strongly tied totheir own attainments of institutionalized retirement criteria andof a continuous and successful work career. This would suggestmore flexibility in women's retiree identity according to a vari-ety of circumstances; a finding which is consistent with the in-tent, though not the specific predictions of, Hypothesis 2.

Our data have implications both for measurement and theoret-ical approaches to retirement research. They support the empha-sis on labor force participation, rather than age or benefits, as themain criterion used in retirement research. However, some incon-gruence between self-definitions and institutionalized retirementcriteria occur among part-time and self-employed individuals,and both conditions may have somewhat different connotationsfor women and men. Thus, future research needs to explore underwhat circumstances part-time and self-employed workers should

be classified as workers or as retirees, or for whom part- and self-employment constitutes a "bridge job." In addition, researcherswill need to keep in mind that institutionalized retirement criteriacontain somewhat of a White, middle-class, male bias. The fit be-tween institutionalized criteria and self-definitions is best for menwith continuous and successful work careers. Our results indicatethat the influence of selected life circumstances differs by genderand race, and that some of these gender and race interactions aremaintained when institutionalized retirement criteria, work histo-ries, and economic status are in the equations. It is conceivablethat gender and race affect the importance of the work role in in-dividuals' lives, but further research on the reasons for race andgender differences in retirement definitions is clearly indicated(see also Honig, 1996; Quinn & Kozy, 1996; Santiago &Muschkin, 1996). For example, future research should addressquestions such as: Does rejection of the retiree identity amongthose who meet selected institutionalized retirement criteria (orself-definition as retiree among those who do not meet these cri-teria) have divergent consequences for the self-image, attitudes,or behaviors of men and women, or of African Americans andWhites? Can retirement programs and policies appeal to or besuccessful among individuals who do not follow the life courseimplied in traditional programs and policies?

Future research is also essential to determine how questionand answer categories affect findings concerning retirement sta-tus and retiree identity. Our results on disability suggest that re-tirement reasons and retirement self-identifications may di-verge, and our overall findings suggest that the retiree identityis not always synonymous with attainment of institutionalizedretirement criteria. Whether and to what extent these differ-ences influence results concerning the predictors and conse-quences of retirement remains unclear. For example, predictorsand consequences of labor force exits may diverge amongwomen who define their labor force withdrawal as a transitionfrom employment to housewife, and those who perceive it as atransition from employment to retiree.

More generally, our research underlines the need to couchretirement transition processes within a theoretical perspectivethat is sensitive to differences, not only in life experiences them-selves, but also in their consequences for and interpretation bydivergent societal subgroups. Focusing more research on the re-tiree identity may help clarify the meaning context of retire-ment and, thereby, contribute to enhanced concept validity andtheoretical advancement in studies of retirement transitions.

ACKNOWLEDGMENTS

This research was funded by a grant from the National Institute of Aging(1RO1AG1318O), Maximiliane E. Szinovacz, principal investigator. TheNational Survey of Families and Households was funded by a grant (HD21009)from the Center for Population Research at the National Institute of ChildHealth and Human Development. The survey was designed and carried out atthe Center for Demography and Ecology at the University of Wisconsin-Madison. The field work was done by the Institute for Survey Research atTemple University. Address correspondence to Maximiliane E. Szinovacz,Eastern Virginia Medical School, The Glennan Center for Geriatrics andGerontology, Hofheimer Hall, Room 201, 825 Fairfax Avenue, Norfolk, VA23507-1912. E-mail: maxres@visi.net

REFERENCES

Adelmann, P. K., Antonucci, T. C, & Jackson, J. S. (1993). Retired or homemaker.How older women define their roles. Journal of Women and Aging, 5, 67-68.

Aiken, L. S., & West, S. G. (1991). Multiple regression: Testing and interpret-ing interactions. Newbury Park, CA: Sage.

S218 SZINOVACZAND DEVINEY

Appelbaum, E., & Gregory, J. (1990). Flexible employment: Union perspec-tives. In P. B. Deringer (Ed.), Bridges to retirement, (pp. 130-144). Ithaca,NY: ILR Press, Cornell University.

Belgrave, L. L. (1988). The effects of race differences in work history, work at-titudes, economic resources, and health on women's retirement. Researchon Aging, 10, 383-388.

Belgrave, L. L. (1989). Understanding women's retirement. Generations, 13,49-52.

Burr, J. A., Massagli, M.P., Mutchler, J. E., & Pienta, A. M. (1996). Labor forcetransition among older African American and White men. Social Forces,74, 963-982.

DeViney, S. (1995). Life course, private pension, and financial well being.American Behavioral Scientist, 39, 172-185.

DeViney, S., & Solomon, J. C. (1995). Gender differences in retirement in-come: A comparison of theoretical explanations. Journal of Women andAging, 7, 83-100.

Ekerdt, D. J. (1986). The busy ethic: Moral continuity between work and retire-ment. The Gewntologist, 26, 239-244.

Ekerdt, D. J., & DeViney, S. (1990). On defining persons as retired. Journal ofAging Studies, 4, 211-229.

Ekerdt, D. J., & DeViney, S. (1993). Evidence for a preretirement processamong older male workers. Journal of Gerontology, 48, S35-S43.

Ekerdt, D., DeViney, S., & Kosloski, K. (1994, November). The heterogeneityof work and retirement plans. Paper presented at the Annual Meeting of theGerontological Society of America, Atlanta, GA.

Gendell, M., & Siegel, J. S. (1996). Trends in retirement age in the United States,1955-1993, by sex and race. Journal of Gerontology: Social Sciences, 51B,S132-S139.

Gibson, R. C. (1987). Reconceptualizing retirement for Black Americans. TheGewntologist, 27, 691-698.

Gibson, R. C. (1991). The subjective retirement of Black Americans. Journal ofGerontology: Social Sciences, 46, S204-S209.

Gibson, R. C. (1993). Retirement and work: The Black American retirementexperience. In J. S. Jackson, L. M. Chatters, & R. J. Taylor (Eds.), Aging inBlack America (pp. 277-297). Newbury Park, CA: Sage.

Gustman, A. L., & Steinmeier, T. L. (1984a). Modeling the retirement processfor policy evaluation and research. Monthly Labor Review, 107, 26-33.

Gustman, A. L., & Steinmeier, T. L. (1984b). Partial retirement and the analysisof retirement behavior. Industrial and Labor Relations Review, 37, 403-415.

Gustman, A. L., & Steinmeier, T. L. (1986). A structural retirement model.Econometrica, 54, 555-584.

Hanks, R. S. (1990). The impact of early retirement incentives on retirees andtheir familes. Journal of Family Issues, 4, 424-437.

Hanson, K., & Wapner, S. (1994), Transition to retirement. Gender differences.International Journal of Aging and Human Development, 39, 189-208.

Hardy, M. A. (1991). Employment after retirement. Research on Aging, 13,267-288.

Hatch, L. R. (1990). Effects of work and family on women's later-life resources.Research on Aging, 12, 311-338.

Hayward, M. D., Crimmins, E. M., & Wray, L. A. (1994). The relationship be-tween retirement life cycle changes and older men's labor force participa-tion rates. Journal of Gerontology, 49, S219-S230.

Hayward, M. D., Friedman, S., & Chen, H. (1996). Race inequities in men's re-tirement. Journal of Gerontology: Social Sciences, 5IB, S1-S10.

Hayward, M. D., & Grady, W. R., (1990). Work and retirement among a cohortof older men in the United States. Demography, 27, 337-356.

Hayward, M. D., Hardy, M. A., & Liu, M. (1994). Work after retirement: The expe-riences of older men in the United States. Social Science Research, 23, 82-107.

Henretta, J. C. (1994). Social structure and age-based careers. In M. W. Riley,R. L. Kahn, and A. Foner (Eds.), Age and structural lag, (pp. 57-59). NewYork: Wiley.

Henretta, J. C. (1997). Guest editorial: Changing perspectives on retirement.Journal of Gerontology: Social Sciences, 52B, S1—S3.

Heyman, D. K. (1970). Does a wife retire? The Gewntologist, 10, 54-56.Holden, K. C. (1988). Physically demanding occupations, health, and work

after retirement: Findings from the New Beneficiary Survey. Social SecurityBulletin, 57(11), 3-15.

Honig, M. (1985). Partial retirement among women. The Journal of HumanResources, 20, 613-621.

Honig, M. (1996). Retirement expectations: Differences by race, ethnicity, andgender. The Gewntologist, 36, 373-382.

Honig, M., & Hanoch, G. (1985). Partial retirement as a separate mode of re-tirement behavior. The Journal of Human Resources, 20, 21-46.

lams, H. M. (1986). Employment of retired-worker women. Social SecurityBulletin, 49(3), 5-13.

Kohli, M. (1986). The world we forgot: A historical review of the life course. InV. W. Marshall (Ed.), Later life, (pp. 271-304). Beverly Hills, CA: Sage.

Kohli, M. (1994). Work and retirement: A comparative perspective. In M. W.Riley, R. L. Kahn, & A. Foner (Eds.), Age and structural lag, (pp. 80-106).New York: Wiley.

Meyer, M. H. (1990). Family status and poverty among older women: The gen-dered distribution of retirement income in the United States. SocialPwblems, 37, 551-563.

Morgan, L. A. (1992). Marital status and retirement plans: Do widowhood anddivorce make a difference? In M. Szinovacz, D. J. Ekerdt, & B. H. Vinick(Eds.), Families and Retirement, (pp. 114-126). Newbury Park, CA: Sage.

Morrow-Howell, N., & Leon, J. (1988). Life-span determinants of work in re-tirement years. International Journal Aging and Human Development, 27,125-140.

Mutchler, J. E., Burr, J. A., Pienta, A. M., & Massagli, M. P. (1997). Pathwaysto labor force exit: Work transitions and work instability. Journal ofGewntology: Social Sciences, 52B, S4-S12.

Mutran, E., Reitzes, D. J., Bratton, K. A., & Fernandez, M. E. (1997). Self-esteemand subjective responses to work among mature workers: Similarities and dif-ferences by gender. Journal of Gewntology: Social Sciences, 52B, S89-S96.

O'Rand, A. M. (1988). Convergence, institutionalization, and bifurcation:Gender and the pension acquisition process. Annual Review of Gewntologyand Geriatrics, 8, 132-155.

Palmore, E. B. (1965). Differences in the retirement patterns of men andwomen. The Gewntologist, 5, 4-8.

Parnes, H. S., & Less, L. J. (1985). The volume and pattern of retirement,1966-1981. In H. S. Parnes, J. E. Crowley, R. J. Haurin, L. J. Less, W. R.Morgan, F. L. Mott, & G. Nestel (Eds.), Retirement among American men,(pp. 57-78). Lexington, MA: Lexington Books.

Pienta, A. M., Burr J. A., & Mutchler, J. E. (1994). Women's labor force partici-pation in later life: The effects of early work and family experiences.Journal of Gewntology: Social Sciences, 49, S231-S239.

Quadagno, J., & Hardy M. (1996). Work and retirement. In R. H. Binstock, &L. K. George (Eds.), Handbook of aging and the social sciences (4th ed.,pp. 326-345). New York: Academic Press.

Quinn, J. F, & Kozy, M. (1996). The role of bridge jobs in the retirement tran-sition: Gender, race, and ethnicity. The Gewntologist, 36, 363-373.

Ruhm, C. J. (1996). Gender differences in employment behavior during latemiddle age. Journal of Gewntology: Social Sciences, 5 IB, SI 1-S17.

Sainsbury, D. (1996). Gender equality and welfare states. Cambridge, UK:Cambridge University Press.

Santiago, A. M., & Muschkin, C. G. (1996). Disentangling the effects of dis-ability status and gender on the labor supply of Anglo, Black, and Latinoolder workers. The Gewntologist, 36, 299-310.

Setterstein, R. A., & Hagestad, G. O. (1996). What's the latest? n. Cultural agedeadlines for educational and work transitions. The Gewntologist, 36, 602-613.

Sherrell, K. E., & Halleen, J. M. (1996, November). Study of ambiguity in thequestions, "What is retirement?", when conducting research on adjustmentto retirement. Paper presented at the Annual Meeting of the GerontologicalSociety of America, Washington, D. C.

Skirboll, E., & Silverman, M. (1992). Women's retirement: A case study ap-proach. Journal of Women and Aging, 4, 77—90.

Stata (1997). Stata user's guide (Release 5) [Computer Software]. College Station,TX: Stata Corporation.

Sweet, J., Bumpass, L., & Call, V. (1988). The design and content of theNational Survey of Families and Households. Madison, WI: Center forDemography and Ecology.

Szinovacz, M. (1989). Decisionmaking on retirement timing. In D. Brinberg & J.Jaccard (Eds.), Dyadic decision making (pp. 286-310). New York: Springer.

Turner, R. H. (1962). Roletaking: Process versus conformity. In A. Rose (Ed.),Human behavior and social pwcess (pp. 20-40). Boston: Houghton Mifflin.

Williamson, J. B., & Pampel, F. C. (1993). Old-age security in comparative per-spective. New York: Oxford University Press.

Wray, L. A. (1996). The role of ethnicity in the disability and work experienceof preretirement-age Americans. The Gewntologist, 36, 287-298.

Zsembik, B. A., & Singer, A. (1990). The problem of defining retirement amongminorities: The Mexican Americans. The Gewntologist, 30, 749-757.

Received May 1, 1998Accepted February 24,1999