Introduction
In 1989, Nancy Datan's ‘Aging Women: The Silent Majority’ was published. In this pamphlet, published post-mortem, Datan rails against the lack of visibility of older women in public life, in science and in policy. According to her, older women are cast aside and forgotten in society. And because of this neglect, older women are prone to destitution. Today, more than 20 years later, the question of poverty among elderly women is at the centre of the public and scientific debate on ageing. Nevertheless, in-depth insight into the exact mechanisms leading to old-age poverty still remains largely absent. Granted, social policy researchers have succeeded in clearly acknowledging the importance of pension regulations in protecting against old-age poverty (e.g. Smeeding and Williamson Reference Smeeding and Williamson2001; Zaidi et al. Reference Zaidi, Makovec, Fuchs, Lipszyc, Lelkes, Rummel, Marin and De Vos2006); yet, they have failed to investigate the importance of sequence, timing and interaction of family and career events. Conversely, this interaction is a prime research theme for demographers and family sociologists alike. In fact, it is central in the abundant literature on the economic consequences of divorce (e.g. Brown and Lin Reference Brown and Lin2012; Cooke Reference Cooke2004; Wilmoth and Koso Reference Wilmoth and Koso2002). Nonetheless, in this strand of study, the elderly are largely overlooked, and interactions between these events and social policy are rarely touched upon.
In response to this dearth of knowledge concerning why the financial situation of many elderly women remains so precarious, it is essential to investigate how family histories, career histories and social policy interact. Research studying the income packages of elderly women in industrialised countries has shown that their largest source of income consists of pension benefits (Choi Reference Choi2006; Gornick, Sierminska and Smeeding Reference Gornick, Sierminska and Smeeding2009). Therefore, one could assume that the level of pension benefits largely determines poverty among elderly women. Consequently, the impact of detrimental lifecourse events can be mitigated depending on the pension regulations in place. In this article, we examine how a combined focus on lifecourse histories and pension regulations strengthens our understanding of old-age poverty. To that end, we draw from register data on some 9,000 women aged 65–71, living in Belgium in 2008. Data on the poverty risk of these women is linked to over 45 years of career and family data.
The approach as described in this paper has two main advantages. First, it facilitates an understanding how previous lifecourse events determine poverty among elderly women. Especially relevant in this respect is the impact of divorce on old-age poverty, as the generation under study is the first to have had a substantial divorce rate. In Belgium, one-fifth of all marriages that took place in 1965 have, so far, ended in divorce. For those who entered into marriage only one decade earlier, divorce was only half as frequent (Matthijs Reference Matthijs2009). As such, this constitutes the first generation of women for whom the effects of divorce on old-age poverty can adequately be studied. Second, this paper advances the on-going debate on pension reform. Although the prime focus in this debate remains on the financial sustainability of pension systems (e.g. European Council 2011), this last decade has seen more and more research focusing on the distribution of pension protection. By further understanding where public policies succeed, or fail, in preventing old-age poverty, this paper advances the debate on adequate pension protection by proposing specific policy actions that can prevent old-age poverty for the most vulnerable groups in society.
Poverty among elderly women
International studies have consistently shown that the poverty risk among elderly women is higher than among elderly men, independent of the type of indicator, the data used or the country surveyed (European Commission 2006; Organisation for Economic Co-operation and Development (OECD) 2008a ; Smeeding and Williamson Reference Smeeding and Williamson2001). To account for variation among women, such studies most often correlate poverty risk to marital status, living situation and number of children – factors that, clearly, are strongly intertwined. In terms of marital status, without exception poverty risk has been found to be lowest among married women. Most research has concluded that the highest poverty risk is found among divorced, separated and never married women, while being somewhat lower among widows (Haider, Jacknowitz and Schoeni Reference Haider, Jacknowitz and Schoeni2003; Harrington Meyer Reference Harrington Meyer1990; Harrington Meyer and Herd Reference Harrington Meyer and Herd2007). Furthermore, all studies agree that, in terms of living situation, elderly women living alone have higher poverty risk than their counterparts that share a household (Choi Reference Choi2006; OECD 2008a ; Smeeding and Williamson Reference Smeeding and Williamson2001). Finally, several studies have found a positive correlation between number of children and old-age poverty risk among women (Choudhury and Leonesio Reference Choudhury and Leonesio1997; Marier and Skinner Reference Marier and Skinner2008).
In order to account for the above findings, most of these studies conclude that such results are the products of the male breadwinner paradigm, whereby the husband performs the paid labour outside the home, while the wife (and even more so the mother) is responsible for the unpaid domestic labour associated with family life. If women do work, they often do so in the so-called secondary labour market, characterised by low wages, part-time and/or temporary jobs, and less occupational pension build-up (especially important in most Liberal welfare states; Ghilarducci Reference Ghilarducci2008; Ginn Reference Ginn2003). Accordingly, at old age women have lower individual incomes than men. Of course, as long as the wife lives with the breadwinner, this lower individual income is not necessarily cause for concern. However, if, for whatever reason (be it death, separation or divorce), the wife ends up living alone, the high risk of falling into poverty becomes apparent as the wife then must rely on her own lower income and can no longer benefit from the economies of scale associated with sharing a household with the breadwinner. To account for the fact that some groups of women living alone (e.g. widows) are better off, financially, than other groups (e.g. divorced or separated women), some studies cite the differential pension regulations based on marital status as the cause for these disparities (Choi Reference Choi2006; Haider, Jacknowitz and Schoeni Reference Haider, Jacknowitz and Schoeni2003; Harrington Meyer Reference Harrington Meyer1990; O'Rand and Henretta Reference O'Rand and Henretta1999; Sandell and Iams Reference Sandell and Iams1997).
Unfortunately, most of the above studies share some defining limitations. For one, detailed lifecourse histories are seldom considered. Thus, complex lifecourses are mostly reduced to a few simple socio-demographic variables, such as marital status. As a result, several groups with very distinct characteristics are inaccurately bunched together (e.g. divorced and widowed women, see Dewilde Reference Dewilde2012). Additionally, when lifecourse events are used to explain old-age poverty risk they are either not tested (e.g. Harrington Meyer Reference Harrington Meyer1990) or the intermediary mechanisms linking lifecourse events to old-age poverty (e.g. labour market participation) are not explicitly investigated (e.g. Choudhury and Leonesio Reference Choudhury and Leonesio1997). Generally, in such studies the exact way in which lifecourses influence old-age poverty remains vague, with researchers content to merely conclude that lifecourses are important (e.g. Vartanian and McNamara Reference Vartanian and McNamara2002) or that the ways in which lifecourses influence later life are too complex to develop specific hypotheses (Sefton et al. Reference Sefton, Evandrou, Falkingham and Vlachantoni2011). A final limitation of most existing research in this area is that it focuses on Liberal welfare states. However, central to most social policy research after Esping-Andersen's (Reference Esping-Andersen1990) seminal contribution is that social policy institutions differ manifestly across regime type.
In response to these limitations of previous research on poverty risk among elderly women, this paper demonstrates how a specific focus on the interaction between lifecourses and pension regulations can provide greater insight into the mechanisms that produce poverty among elderly women in a Continental European welfare state, such as Belgium.
The Belgian research context
Family and career from the late 1950s
The cohort under study was coming into their late teens and early twenties at around 1960, a time when adherence to the male breadwinner model was at its historical peak in most Western countries, including in Belgium (Lewis Reference Lewis2001; Vanhaute Reference Vanhaute and Jensen2002). The institution of marriage was sacrosanct, divorce exceptional. Once married, women were expected to raise children and run the household. While this male breadwinner model represented an often unrealistic ideal, figures show that female labour market participation was historically low during this period (Vanhaute Reference Vanhaute and Jensen2002). Unsurprisingly, various studies from that era testify that the great majority of women considered home-making to be their vocation, while full-time or part-time work outside the home was only of secondary interest (Bruntz Reference Bruntz1962; Rostow Reference Rostow1964).
However, the large number of sociological studies on working wives emerging from the late 1950s onwards (Myrdal and Klein [Reference Myrdal and Klein1956] 2003; Yudkin and Holme Reference Yudkin and Holme1963) indicated the beginning of a slow process of change. Mainly as a result of a booming economy and the growth of the service sector, more and more (married) women began working in the formal economy. At the same time, important changes were occurring in marital life – that other central tenet of the male breadwinner model. Gradually, from the early 1960s and manifestly from the mid-1970s onwards, divorce figures increased spectacularly. Of all marriages entered into in 1955, 8 per cent were dissolved after 25 years; of marriages entered into in 1970 the same percentage of divorce was reached after only ten years of marriage (Matthijs Reference Matthijs2009). From the 1980s on, the number of new marriages declined while the practice of sharing a household outside the institution of marriage grew (Deboosere et al. Reference Deboosere, Lesthaeghe, Surkyn, Willaert, Boulanger, Lambert and Lohlé-Tart2009).
Pension regulations
The Belgian pension system gradually came into existence in the first half of the twentieth century, culminating in the defining pension acts of the 1950s and 1960s (Berghman Reference Berghman1981), which are still central to understanding Belgian pension regulations. Born in a male-breadwinner context, the pension system's dual eligibility structure is unsurprising: pensioners are either entitled to a pension on the basis of their own employment history (retirement pensions – mostly for men) or on that of their (former) spouse (derived pensions – mostly for women).Footnote 1 Retirement pensions are based on one's own career. Most importantly, for every year worked, 1/45th of a full retirement pension is built up, equalling 60 per cent of previous wages. Moreover, some periods of inactivity or absence from work, such as those due to disability, unemployment or maternity/paternity leave, are also assimilated into periods of employment. The pension rights received from these ‘assimilated periods’ (also known as career or pension credits) are (largely) identical to those built up in the previously worked period. While the existence of assimilated periods is not unique to Belgium (OECD 2003), the extent of their usage likely is: around one-quarter of male and one-third of female pension accrual consisted of such periods (Peeters and Larmuseau Reference Peeters and Larmuseau2005).
Derived pensions, in contrast, are calculated on the basis of the career of the (previous) breadwinner (usually the husband). The calculation of derived pensions varies by marital status, reflecting the social reality as it was when these pension regulations were first constructed. Generally speaking, married women are assumed to benefit from a male earned family income, while divorced and widowed women are assumed to be single, whereby the state replaces the former breadwinner as the supplier of income. Unmarried women (i.e. women who have never been married) only benefit from individual retirement pensions. The manner in which derived rights are calculated is rather complex, as illustrated in Figure 1. In this figure, the derived pension is presented as a percentage of the (previous) breadwinner's retirement pension on the y-axis, while the woman's retirement pension is displayed on the x-axis as a percentage of that of the (previous) breadwinner.Footnote 2 The sum of the percentages given on both axes depicts the total pension of the wife as a percentage of that of the (former) husband.
Figure 1. Women's derived pensions by marital status: a comparison of married, divorced and widowed women.
Let us first consider the situation of married women. As it is assumed that a married woman is covered by the breadwinner's income, she does not receive derived pension benefits. Her husband, on the other hand, receives additional spousal benefits equalling 25 per cent of his own retirement pension to provide for his wife. These benefits, however, are only granted on the condition that the wife's retirement pension does not exceed 25 per cent of the husband's. In this case, the wife's retirement pension is not paid out: rather, her own pension rights (if any) are transferred to her husband (hence ‘negative derived rights’). As a consequence of spousal benefits, the wife does not contribute to total family income as long as her own pension rights do not exceed the mentioned 25 per cent.
In contrast to married women, widows are entitled to very advantageous derived pensions. On the death of the husband, a widow is entitled to a survivor's pension equalling 100 per cent of the deceased spouse's retirement pension (60 per cent of previous wages). This derived pension can be retained as long as the wife's retirement pension is 10 per cent or less than that of the deceased husband. If the wife's retirement pension exceeds this 10 per cent, the survivor's pension is diminished proportionally, which means that whether the widow has built up her own rights amounting to 10 per cent or up to 110 per cent of her husband's retirement pension, or anything in between, makes no difference for her total pension (add-up percentages on the x- and y-axes in Figure 1).
Finally, the current divorce pension, established in 1975, entitles an ex-wife to a maximum of 62.5 per cent of the former husband's retirement pension. This maximum is only granted on the condition that the ex-wife has no retirement pension of her own and has been married to her ex-husband for the duration of his 45-year career (cf. Figure 1; divorced after 45 years of marriage). However, if the ex-wife is entitled to her own retirement pension, the derived pension is diminished proportionally: no derived benefits are granted if the wife's retirement pension exceeds 62.5 per cent of the husband's retirement pension (see x-axis). Alternatively, if the couple had been married for less than 45 years, only a fraction of the divorce pension is granted to the ex-wife. Since in the cohort under study women divorced on average after just 22 years of marriage, only 22/45th of a full divorce pension is granted, equalling 30 per cent of the ex-husband's retirement pension (cf. Figure 1; divorced after 22 years of marriage). Paradoxically, in this case, if the ex-wife has a retirement pension of her own that has been built up outside marriage, the derived rights granted are not reduced accordingly and can, therefore, be higher than in the case of a 45-year marriage.Footnote 3
The above figure assumes that women were married, divorced or widowed only once in their lifetimes. Of course, in reality many women have more elaborate marital histories. However, this does not render the figure less accurate, as with Belgian pension regulations only current marital status matters. Thus, divorced women who remarry lose their divorce pension. Similarly, widows who remarry lose their survivor's pension. There is one exception to this rule. If subsequent marriages dissolve, the woman may reclaim the divorce and/or survivor's pension derived from her previous marriage(s).Footnote 4
Hypotheses
In line with the general focus of this paper, we develop hypotheses on how pension regulations can explain the impact of detailed lifecourses on old-age poverty. We first examine the direct impact of marital history on poverty risk for elderly women. We then consider the indirect impact of family history (marital and parenthood histories) on poverty risk, through the way in which family history affects women's work histories and consequent retirement pension build-up.
Direct impact of marital history on old-age poverty risk
Controlling for whether or not one lives alone and, if applicable, for the income of the other household member(s), as well as for work history (see following paragraph), we expect that women who receive the highest derived pension benefits will have the lowest poverty risk. Accordingly, based on derived pension regulations (cf. Figure 1), we therefore expect that widows will have the lowest poverty risk, and unmarried and married women the highest. The poverty risk of divorced women, we believe, will lie somewhere in between (Hypothesis 1).Footnote 5 Also, and contrary to what is commonly held in lifecourse studies, based on derived pension regulations we predict that only current marital status is important in determining poverty among elderly women (Hypothesis 2). We expect there to be only one exception to this general rule; namely that among divorcees, previously widowed women will have the lowest poverty risk because they can benefit from beneficial survivor's pensions (see previous section).
Indirect impact of marital and parenthood histories on old-age poverty risk
While the previous section considered the direct impact of marital history on elderly women's poverty risk, most existing studies on poverty among elderly women focus on the indirect impact of marital and parenthood history. The implicit assumption in most studies is that marital and parenthood histories influence work history, that work history influences the individual pension rights that are accumulated which, in turn, determine old-age poverty risk (e.g. Quadagno Reference Quadagno, Ginn, Street and Arber2001: vii). Forming hypotheses on this indirect relationship, we first examine the impact of work history on old-age poverty risk through its expected effect on retirement pension build-up. We then go on to formulate hypotheses on the indirect impact of marital and parenthood histories on elderly women's poverty risk.
In so-called Bismarckian welfare states, the number of years worked and wages earned influence the level of the retirement pension (Bonoli and Palier Reference Bonoli and Palier2007; Peeters et al. Reference Peeters, Debels, Verschraegen and Berghman2008). In the Belgian context, retirement pensions are also influenced by the number of assimilated years. Following Belgian pension regulations, one could expect to see a negative effect of number of years worked, wages earned and the number of assimilated years on old-age poverty risk. However, as was previously discussed, a higher number of career years, assimilated years or higher wages earned do not necessarily translate into a substantial increase in total pension protection. Drawing directly from Figure 1, we expect that the greater the number of years worked and assimilated, as well as the higher the wages earned, (a) the lower the poverty risk will be, particularly for both unmarried and married women, (b) less so for divorced women, and (c) least of all for widows (Hypothesis 3).
For the generation under study, childbirth usually signified a retreat from the labour market (Girard Reference Girard1958; Guilbert and Isambert-Jamati Reference Guilbert and Isambert-Jamati1958; Klein Reference Klein1959; Mincer and Polachek Reference Mincer and Polachek1974), or for those mothers who did remain active in the workforce, it signified having a low-paying job (Poloma and Garland Reference Poloma, Garland and Michel1971; Rostow Reference Rostow1964). More specifically, earlier research showed that being younger at the time of first childbirth, having a higher number of children (Girard Reference Girard1958) and undergoing a longer period of having children who are under the schooling age (Klein Reference Klein1959) would cause women to exit the labour market or to work in low-paying jobs for a longer period of time. As it was still not universally accepted for women of this generation to return to the labour market after their children were grown up (Dumon Reference Dumon1973), we expect a negative effect of the woman's age at first childbirth on old-age poverty risk and a positive effect of number of children and years with children under the schooling age on old-age poverty risk.
Irrespective of the influence of childbirth for the generation under study, it was not unusual for women to already scale down labour market participation after marriage (Guilbert and Isambart-Jamati Reference Guilbert and Isambert-Jamati1958). In fact, in Belgium up until the late 1950s, in certain sectors (e.g. education), women were forced to give up their jobs once they married (Dumon Reference Dumon1973). Hence, we expect that the older the woman at the time of marriage, the lower the risk of poverty in old age. Similarly, we expect that the older the woman at the time of divorce, the higher the poverty risk after retirement age.
As the importance of paid labour and the consequent accumulation of retirement entitlements for old-age poverty risk depends on marital status (see ‘Pension regulations’ section), we expect that the impact of marital and parenthood histories on old-age poverty risk also depends on marital status. More specifically, on the basis of the information presented in Figure 1, we expect marital and parenthood histories (a) to be most important for unmarried and married women, (b) to be less important for divorced women, and (c) to be least important for widows (Hypothesis 4).Footnote 6
Data, operationalisation and method
Data
The data used are drawn from three national register datasets that are linked at the individual level in the Belgian Datawarehouse Labour Market and Social Protection. The Pension Register contains all pension payments, including receipt of old-age social assistance benefits; the National Register contains data on household composition and a complete account of marital and parenthood history; and the Sigedis dataset includes career data and data on pension build-up from 1955 onwards. The research population consists of women aged 65–71 and living in Belgium in 2008. Women receiving pension benefits from the self-employed and civil servants schemes are not included in the research population (excluding 36 per cent of all Belgian women in that age group), as these pension regulations are manifestly different from those of the employee pension scheme. Similarly, first-generation immigrant women as well as women having built up pension benefits abroad are not included (accounting for another 9 per cent of this population of women), as both categories have (potentially) built up pension rights in unknown pension schemes. From the research population, a 5 per cent random sample of 9,382 elderly women is drawn; to this sample an additional 921 women with less common marital histories (i.e. widowed twice; divorced twice; divorced and widowed) are added, totalling 10,303 women. After removing 217 severely handicapped or institutionalised women, 142 cases with inaccurate marital, parenthood or career histories, and 73 cases with other administrative errors, a sample remains of 9,871 elderly women.Footnote 7 Of these women in 2008, 269 were unmarried, 6,881 married, 103 separated, 666 divorced and 1,134 widowed. In weighted percentages this amounts to 3 per cent unmarried, 1 per cent separated, 76 per cent married, 7 per cent divorced and 13 per cent widowed. Unfortunately, separated women cannot be analysed further due to their insubstantial number.
By using register data, we avoid several difficulties characteristic of the use of survey data for lifecourse research (Wallgren and Wallgren Reference Wallgren and Wallgren2007). For instance, due to the relatively small number of respondents in existing surveys, lifecourse researchers often have no other choice but to reduce complex lifecourse histories to simple socio-demographic variables (e.g. number of children or current marital status, see Choi Reference Choi1995), or to group women with distinct lifecourse histories into one single category (e.g. combining never married and divorced women, see Dewilde Reference Dewilde2012; Vartanian and McNamara Reference Vartanian and McNamara2002). Additionally, studies using survey data to reconstruct extended lifecourses do so on the basis of panel studies (e.g. Vartanian and McNamara Reference Vartanian and McNamara2002) or on the basis of retrospective modules (Avendano and Mackenbach Reference Avendano, Mackenbach, Börsch-Supan, Brandt, Schröder and Hank2011). The former studies are plagued by selective drop-out in between waves (Robine, Thinggaard and Christensen Reference Robine, Thinggaard, Christensen, Börsch-Supan, Brugiavini, Jürges, Kapteyn, Ackenbach, Siegrist, Weber and Andersen-Ranberg2008; Schröder Reference Schröder, Börsch-Supan, Brugiavini, Jürges, Kapteyn, Ackenbach, Siegrist, Weber and Andersen-Ranberg2008), while the latter are prone to substantial and stratified memory bias (Drasch and Matthes Reference Drasch and Matthes2011). An ultimate problem facing researchers using survey data is non-response. Even surveys with the most meticulous design have to contend with high non-response rates that cannot be assumed as random (Bethlehem, Cobben and Schouten Reference Bethlehem, Cobben and Schouten2011; Mitchell Reference Mitchell2010). For all these reasons register data, when available, are preferable to survey data for lifecourse research.
Register data have their own limitations, one disadvantage being that not all relevant information is registered. Especially important for this research is that no information is available on alimony payments, on (income from) assets (e.g. investments or house ownership) and on educational level. However, with regard to the latter, some secondary information is available. Based on the second wave (2007) of the Survey of Health and Retirement in Europe (SHARE), educational levels by current marital status can be ascertained. Among the respondents aged 60–75, we found the highest educational levels among unmarried and divorced women (36 and 33 per cent, respectively, had higher education), somewhat lower levels among married women (25 per cent with higher education) and the lowest educational levels among widows (20 per cent with higher education).
Operationalisation
Poverty risk is measured by means of a dummy variable indicating receipt of means-tested benefits in old age. In 2008, pensioners were entitled to means-tested old-age benefits only if their monthly total income did not exceed €811 (individual living alone) or €541 (individual sharing a household with another adult). These amounts are lower than the 60 per cent of median income threshold, which were €899 and €674, respectively (Eurostat 2013). Moreover, in contrast to the much-used relative poverty threshold, calculated solely on the basis of income, the means test also takes into account the assets owned by the household members (e.g. home ownership). Therefore, it can be assumed that pensioners who do receive a means-tested benefit also fall under the previously discussed relative poverty threshold. Although non-take up of social assistance benefits is possible, for the age group under study entitlement to social assistance is automatically investigated by the pension administration, making non-take up less likely (Ooms Reference Ooms2008).
Family histories are distinguished on the basis of marital and parenthood history. Marital history is based on current (and previous) marital status. Accordingly, the following marital statuses and marital status sequences are identified: Unmarried (U), Married (M), Married and Divorced (MD), Married and Widowed (MW), MDM, MWM, MDMD, MDMW, MWMD and MWMW. Regarding the timing of changes in marital status (and indirectly the duration of marital periods), ‘age of marriage’ refers to the age of the woman at her first marriage, in the case of more than one marriage. ‘Age of divorce’ refers to age of the woman at the (first) dissolution of marriage through divorce. For the assessment of the impact of parenthood, the following variables are included: ‘number of children’, ‘duration of child rearing’ (number of years living with children under the age of six) and ‘age of the mother at first childbirth’.
Work history is modelled based on the ‘number of years worked’, ‘the number of years assimilated’ and the ‘average daily wage’. The number of years worked refers to full-time equivalents: if someone performed two years of half-time work, this is compressed into one year of full-time employment. Likewise, the number of years assimilated refers to full-time equivalents. The average daily wage refers to the average wage over all periods of employment. This is calculated by summing all wages earned during one's career (adapted to the 2008 consumer price index) and then dividing the result by the total full-time equivalent days ever worked.
Three control variables are included: ‘age’, ‘sharing a household’ (i.e. whether someone else is registered at the same address) and the ‘average income of the other household members’. This income is calculated by dividing the household income of the individual(s) with whom one shares a household by the number of individuals living within the research unit. A similar process is followed by the pension administration when performing the means test that is at the basis of determining social assistance entitlement. The household income includes all income from work, from social security (pensions, invalidity benefits, etc.) and from social assistance. Unfortunately, information on (income from) total assets is not available in the register datasets.
Method
Due to the fact that the dependent variable (i.e. poverty risk) is dichotomous, logistic regression is applied. Different models are used to distinguish between the direct impact of marital history on old-age poverty risk and the indirect impact of marital and parenthood history on poverty risk through their effects on work history and consequent retirement pension build-up. We first test the hypotheses on the direct impact of marital histories on poverty risk in old age. Two models are constructed for the full sample. Model 1 contains both marital history and the previously mentioned control variables. A second model also includes total pension income as an independent variable. If the different poverty risk by marital status histories is indeed the consequence of differential pension regulations, these differences should cancel out after controlling for total pension income. We then go on to investigate the indirect impact of marital and parenthood histories on old-age poverty risk. Here, we perform the analyses separately for each of the four sub-populations divided by current marital status, as they are subject to different pension regulations. For every sub-population we present one model containing the family variables and control variables, and a second model where employment histories are also included. If family histories indirectly influence old-age poverty through their impact on work histories (and the consequent influence on pension rights), the effects of the family variables should cancel out after controlling for work histories.Footnote 8
A complication arises because not all variables apply to all research units. Since not all women have children, the inclusion of the variable ‘age of mother at first childbirth’ as well as ‘duration of child rearing’ poses an obstacle. Similar problems arise with regard to ‘years assimilated’ and ‘daily wage’ (only applicable to women with careers) and ‘average income of the household members’ (as not all women share households). Therefore, to be able to generate one model covering the entire sample, we follow Mirowsky's (Reference Mirowsky, Aneshensel, Phelan and Bierman2013) approach for ‘internal moderators’. Consequently, to account for both women with and without children, we first centre the variables ‘age of mother at first childbirth’ and ‘duration of child rearing’. Childless women are awarded a value of 0 for both variables. Next, a dummy is introduced representing whether a woman has children. Finally, this dummy is multiplied by the variables ‘number of children’, ‘age of mother at first childbirth’ and ‘duration of child rearing’. As a result of this procedure, the interaction terms compare the importance of parenthood variables among mothers; the dummy compares childless women with average mothers. Similar procedures are followed with regard to ‘work history’ (a dummy ‘ever worked’ compares women who have never worked with women with average careers; the (interaction) variables ‘years assimilated’ and ‘average daily wage’ only apply to those having ever worked) and ‘average income of household members’ (the dummy ‘sharing a household’ compares women living alone with the average women sharing a household; the variable ‘average income household members’ only applies to those sharing a household).
A second complication is inherent when comparing logit coefficients over different populations and models. As highlighted by several authors (Allison Reference Allison1999; Mood Reference Mood2010; Williams Reference Williams2009), logit coefficients not only reflect the effect of independent variables, but also the size of unobserved heterogeneity. As this heterogeneity differs across populations and models, comparisons of logit coefficients are not straightforward. Various solutions have been proposed, but a standard method remains absent. Therefore, to compare coefficients across populations we employ three approaches (Mood Reference Mood2010): an early approach advocated by Allison (Reference Allison1999), an approach developed by Williams (Reference Williams2009) and linear probability modelling (Hellevik Reference Hellevik2009). Differences between coefficients across populations will only be mentioned if all three approaches produce significant differences at an α=0.05 level. To compare coefficients across nested models, linear probability modelling is used, applying the method advocated by Clogg, Petkova and Haritou (Reference Clogg, Petkova and Haritou1995). The models used to compare coefficients across populations and models are not presented here, but are available from the authors.
When testing the indirect impact of marital and parenthood histories on old-age poverty risk, our first model (not containing career variables) will likely overestimate the influence of the family variables on poverty risk. This potential over-estimation is the result of the difficulty in establishing causality in work–family research, where not only family events influence careers, but educational and employment backgrounds simultaneously play a role in determining specific family trajectories (Brewster and Rindfuss Reference Brewster and Rindfuss2000). Indeed, in this strand of research, analysis of the time-component inherent to causality is not always evident. For instance, do women refrain from entering the labour market in anticipation of bearing children in the future, or is it that women who are not gainfully employed are more likely to bear children? For our research population, we expect that this issue will lead to an over-estimation of the initial family variable coefficients as those women not working or who have low-paid employment, and who are, thus, more vulnerable later in life, are likely to marry earlier (Glick and Carter Reference Glick and Carter1958; Henryon and Lambrechts Reference Henryon and Lambrechts1968), tend to have more children (Pauwels et al. Reference Pauwels, De Wachter, Deschamps and Van Dongen1984) and tend to have children at a younger age (Kravdal Reference Kravdal1994; Liefbroer and Corijn Reference Liefbroer and Corijn1999).
Descriptive statistics
Table 1 presents weighted descriptive statistics. The table first presents the percentage of women living in poverty. In general, 4 per cent of women live in poverty. However, this general percentage belies considerable differences by current marital status. While up to one in four divorced and one in six unmarried women experience poverty, this is rarely seen among married and widowed women. This finding is largely consistent with previous studies, albeit that the advantageous situation of widows is specific to the Belgian context (Haider, Jacknowitz and Schoeni Reference Haider, Jacknowitz and Schoeni2003; Harrington Meyer Reference Harrington Meyer1990; Harrington Meyer and Herd Reference Harrington Meyer and Herd2007). In fact, with virtually no poor widows to speak of (only six out of more than 1,100), it becomes methodologically impossible, as well as intrinsically unnecessary, to further analyse poverty among this group of women.
Table 1. Weighted descriptive statistics for the entire population and by sub-population
Notes: 1. Population: women previously divorced. 2. Population: women previously widowed. 3. Population: mothers. 4. Population: women that have worked at least one day. 5. Population: women sharing a household.
Source: Authors' calculations based on the Belgian Datawarehouse Labour Market and Social Protection.
Regarding the independent variables, and more specifically marital and parenthood histories, the high proportion of previously divorced widows and divorcees is apparent, as well as the different profile of unmarried women with regard to the child variables. Concerning work histories, we find married women to have worked the least, followed by widows and divorcees. Unmarried women have worked the most and earned the highest wages. Assimilated periods prove to be important for all marital groups. Comparing the number of years assimilated to the number of years worked, we find assimilated periods to equal one-quarter of personal pension build-up among unmarried women and widows, and as much as one-third among married and divorced women.
Considering the control variables, we find that around two-thirds of unmarried, divorced and widowed women live alone; married women by definition share a household (otherwise they would be ‘separated’; see Note 2). The average income of the other household members is highest among married women (some €1,600), while being lower among the other sub-populations (€1,300–1,400).
Results
The direct impact of marital history on old-age poverty risk
We now investigate whether the findings for the direct impact of marital history on poverty risk are in line with what was expected, based on the differential derived pension rights available for the different groups of women (Hypothesis 1). Model 1 in Table 2 shows that, when controlling for other variables and considering only current marital status (see italic status), unmarried and divorced women have the highest poverty risk, married women and widows having the lowest poverty risk. Based on other analyses (not shown here), we also know that widows have significantly lower poverty risk than married women. These results are largely expected, as they reflect the differential derived pension benefits granted. The notable exception concerns married women. Based on derived pension regulations, one would expect that when controlling for background variables, such as whether or not one shares a household and the income of household member(s), married women would have higher poverty risk than divorced women.
Table 2. The importance of current marital status: determinants of poverty risk among women aged 65–71 in 2008 (logit coefficients)
Notes: 1. Current marital status is italic. AIC: Akaike information criterion.
Significance levels: * p<0.05, ** p<0.01, *** p<0.001. All differences between Models 1 and 2 with regard to marital history are significant at an α=0.001 level.
Source: Authors' calculations based on the Belgian Datawarehouse Labour Market and Social Protection.
Model 2 includes total pension income as an independent variable, along with marital history. If one's pension were all that mattered in determining poverty risk, we would expect that after controlling for pension income, marital history would have no remaining effect on poverty risk. We do indeed find that certain effects of marital history are lower in the second model, particularly that the difference between widows and divorced women has decreased, suggesting that widows benefit from the rather generous survivor's pensions. The coefficients of married women, on the other hand, are larger in the second model, suggesting that their beneficial situation is due to factors other than individual pension income and other variables controlled for in the model. One such external factor is wealth. As previously mentioned, the means-test determining eligibility for social assistance, and hence our dependent variable, also includes assets. Married women may, therefore, very well be protected against poverty due to joint accumulated assets. This supposition coincides with previous research showing that marriage facilitates the accumulation of wealth (Schmidt and Sevak Reference Schmidt and Sevak2006; Wilmoth and Koso Reference Wilmoth and Koso2002).
In line with Hypothesis 2, we find that previous marital statuses are of minimal import in determining old-age poverty risk among women (see Model 1). The poverty risk of those women still married to their first husband does not significantly differ from that of previously divorced women or widows who are currently re-married. Also, women widowed once share similar poverty risk to those widowed twice. We expected there to be one exception to the hypothesis that only current marital status matters. Belgian pension regulations stipulate that women who widow first and then divorce their second marriage partner can again take up the survivor's pension from the first marriage. Hence, we expected previously widowed divorcees to have lower poverty risk than other divorcees. Our results confirm this expectation. However, contrary to expectations, women divorcing twice have higher poverty risk than women divorcing once, albeit that this result is only significant at the 0.05 level.
The indirect impact of marital and parenthood history on old-age poverty risk
This section explores the direct impact of work history on old-age poverty risk as well as the indirect impact of marital and parenthood history on poverty risk through their (expected) effects on the career and consequently on retirement benefits (Table 3). Due to differences in derived pension regulations, these impacts are investigated separately for each of the four sub-populations distinguished by current marital status. For each sub-population, Model 1 first presents only family variables and control variables. Model 2 also includes work histories. This second model has a dual purpose. It provides information on the importance of work histories (and hence retirement pension build-up) in accounting for old-age poverty risk. It also provides insight into precisely how marital and parenthood histories influence old-age poverty risk. If these histories influence old-age poverty risk indirectly through their impact on labour market participation, the effects of family variables should cancel out after controlling for women's work history. Table 3 does not present any further information regarding the impact of family and work history for widows, as in line with Hypotheses 3 and 4, Table 1 has established that poverty among widows is negligible.
Table 3. Determinants of poverty risk among married, separated and divorced women aged 65–71 in 2008 (logit coefficients)
Note: AIC: Akaike information criterion.
Significance levels: * p<0.05, ** p<0.01, *** p<0.001. For unmarried and divorced women, differences between Models 1 and 2 regarding marital history are significant at an α=0.001 level, the sole exception being ‘duration of child rearing’. For married women, the difference between Models 1 and 2 with regard to ‘ever had children’ and ‘number of children’ is significant at an α=0.001 level.
Source: Authors' calculations based on the Belgian Datawarehouse Labour Market and Social Protection.
First, we consider the importance of marital and parenthood histories for previously and currently married women. For divorced women we find that the older the woman at the time of divorce, the higher the risk of poverty in old age. For re-married women, in contrast, we find no such relationship. The parenthood variables, investigating the so-called ‘child penalty’, have very little effect on poverty in old age, again with the exception of divorced women. For these women, in line with our expectations, the number of children is positively correlated with poverty risk. However, contrary to expectations, divorced women who have children at a later age have higher poverty risk. Finally, no effect was found on poverty risk among any of the marital groups based on the number of years sharing a household with a child younger than six.
Overall, as a consequence of derived pension rights (see Figure 1), we expected family history to be most important for unmarried and married women, while less important for divorced women in determining poverty risk (Hypothesis 4). However, this hypothesis was not corroborated. On the contrary, the number of children and age at first childbirth is more a determinant of poverty risk for divorced women than for married women.Footnote 9 Regarding the importance of work history for poverty risk among elderly women, as expected poverty risk among women who have never worked is higher than among women with an average career. Poverty risk also decreases the greater the number of years worked or assimilated. Assimilated periods, therefore, do help women in vulnerable situations to remain above the poverty threshold. Regarding wage level, the expected negative effect on old-age poverty risk was not corroborated, despite finding a bivariate correlation (not presented here). This is possibly the result of the increase of daily wages with work experience. The hypothesis that work history influences poverty risk most strongly for both unmarried and married women and less so for divorced and separated women, cannot be confirmed (Hypothesis 3), as the number of years worked determines poverty risk more among divorced women than among married women.
As discussed previously, the models including work history not only provide insight into the importance of work-related variables, but also into the extent to which family history influences old-age poverty risk through its effect on labour market participation and the resulting impact on retirement pension build-up. In this way, if work (and hence retirement pension) determines poverty risk, then the effects of family history should cancel out after controlling for work history.Footnote 10 Our findings suggest that the effect of age at the time of divorce among divorced women does, indeed, cancel out. Accordingly, the older the woman at the time of divorce, the greater the risk of old-age poverty due to her diminished labour market participation. Also, the effect of number of children among divorced mothers can be explained by work history, while the effect of age at (first) childbirth cannot.
Summary and discussion
Poverty among elderly women is a pressing issue, though by no means a new one. It has been studied from many different angles and various initiatives have been taken to increase awareness of the problem among policy makers. However, despite such efforts, old-age poverty among women persists. And, even though our understanding of this complex reality has improved, we still lack in-depth understanding of the multiple institutions producing it, as well as the way in which they interact. The goal of this paper was to advance our understanding of this state of affairs by investigating how an explicit focus on pension regulations might shine light on the way in which family life and employment history influence poverty risk among elderly women. The hypotheses developed in this article were tested using a unique administrative dataset encompassing the lifespan of a sample of women aged 65–71 and residing in Belgium in 2008. Poverty risk was measured by means of a dichotomous variable indicating receipt of means-tested benefits in old age.
We expected old-age poverty risk among women to be dependent on very specific configurations of marital status, parenthood and work histories. These factors were expected to differentially influence pension entitlements and, consequently, poverty risk. Marital status was expected to directly affect old-age poverty risk through the derived pension regulations that they determine. Contrary to what is common in lifecourse studies, we hypothesised that marital status histories have no influence on poverty risk apart from current marital status. Our results confirm this hypothesis. Whether or not a woman was previously divorced is not indicative of her poverty risk, whereas her current marital status is. Accordingly, our results show that while one in six unmarried women and a quarter of divorced women live in poverty after retirement, poverty is low among (re-)married women and virtually non-existent among widows. Married women and widows generally having the lowest poverty risk. This finding warrants special attention, as bivariate analyses have shown that widows in this age category generally have had the least education, the shortest careers and the lowest wages (see also Zick and Smith Reference Zick and Smith1991).
We also expected family history to have an indirect effect on poverty risk through its impact on work history which, in turn, would affect retirement pension build-up. Strikingly, family history was inconsequential in determining poverty risk among widows. Largely thanks to beneficial derived rights, merely being widowed is sufficient to protect women against poverty. For divorced women, on the contrary, family history does influence old-age poverty indirectly through its impact on work history and consequential retirement pension build-up. Thus, here there was a positive correlation between number of children and old-age poverty. Also of particular interest is the finding that women who divorce at an older age are more likely than those divorcing earlier on in life to end up living in poverty. This is cause for concern, because research shows that divorce rates among elderly women have increased substantially in the last decade (Brown and Lin Reference Brown and Lin2012; Corijn Reference Corijn2011). Fortunately for all sub-groups of women, Belgium has an advantageous system of assimilated periods or pension crediting, so that many periods of inactivity or absence from work are counted as periods of employment for pension build-up. Our results show the importance of these periods in mitigating the effects of family variables on old-age poverty risk. Consequently, we empirically support the arguments of various authors that highlight the benefits of such systems for women (e.g. Bonnet and Hourriez Reference Bonnet and Hourriez2012; Jankowski Reference Jankowski2011).
Our study also revealed various unexpected results whereby poverty risk was not determined by pension benefits. Most importantly, the results concerning married women require further research. Even after controlling for total pension income, as well as the incomes of other household members (including the husband), married women have one of the lowest poverty risks, lower than that of unmarried and divorced women. This finding suggests that other factors unrelated to pension regulations prevent married women from falling into poverty, particularly wealth. Research in the United States of America has shown marriage to be an ideal platform for wealth accumulation (Schmidt and Sevak Reference Schmidt and Sevak2006; Wilmoth and Koso Reference Wilmoth and Koso2002). Therefore, the fact that Belgium boasts an extremely high home-ownership rate, over 80 per cent of the elderly being home-owners (De Decker and Dewilde Reference De Decker and Dewilde2010), and that married couples make up a disproportionate percentage of home-owners (Dewilde Reference Dewilde2012), are probably key in diminishing poverty risk among elderly married women.
It is tempting to use the presented results to inform recent discussions on pension reform. However, policy implications are far from straightforward. In fact, while a focus on poverty prevention is important in devising adequate pension schemes, it is far from sufficient. Here, we would like to point out two issues that have so far remained largely ignored despite requiring further investigation. First, an exclusive focus on poverty overlooks an additionally important and often neglected function of pension policy: the maintenance of living standards, particularly after the death of a partner. Such a focus is especially important for women, since their higher life expectancy means that most elderly women are confronted with the transition from a two-person household with economies of scale to a one-person household – a transition which jeopardises maintaining a comparable standard of living (Peeters, Verschraegen and Debels Reference Peeters, Verschraegen and Debels2014). Second, an exclusive focus on poverty prevention also overlooks women's financial dependency on men. One consequence of Belgian pension policy is that many women (unlike most men) have to take their marriage vow, ‘Till death us do part’, quite literally. Though currently married women might not be poor, nor will be upon the death of their husband, they run a high risk of falling into poverty if their marriage dissolves before one of them dies. This reality highlights a less visible component of wellbeing that is not measured by typical poverty indicators, including our own, namely women's lack of financial autonomy or independence (e.g. O'Connor, Orloff and Shaver Reference O'Connor, Orloff and Shaver1999).
Though an in-depth policy proposal focusing on adjusting existing pension regulations in Belgium, and in other countries sharing similar pension regulations, is outside the scope of this paper, we can make some provisional policy suggestions. First, the current calculation of retirement pensions in Belgium is too strongly founded on the stereotypical male career model (45 full-time years worked for a full pension). A large proportion of women who cannot rely on the pension benefits that are built up by a husband end up in poverty due to an insufficient number of years worked. This is cause for concern as the number of career years needed for a full pension is increasing in many OECD countries (OECD 2011). Secondly, a pension split should be established for divorced women. In Belgium, at the time of divorce, the breadwinner retains all of his pension rights while the dependent spouse receives only 62.5 per cent thereof. By maintaining the same financing composition, a ‘pension split’ would be possible where both partners would receive 81 per cent of the husband's pension build-up, instead of the current 100:62.5 per cent distribution. This provision would make both partners equally protected, rather than having the financial consequences of divorce rest solely on the dependent spouse. Moreover, such a regulation is consistent with the logic of the equal pooling of resources acquired during marriage, with pensions being a ‘postponed wage’. Even though, for some time, ‘splitting’ has been proposed in many countries (for an early statement, see Burkhauser Reference Burkhauser, Burkhauser and Holden1982), so far it only exists in a few of countries, where it is only partially enacted and is often on a voluntary basis (Bonnet and Hourriez Reference Bonnet and Hourriez2012). This policy suggestion of splitting might be even more relevant for countries other than Belgium, given the Belgian divorce pension's qualification as a ‘best practice’ (Horstmann and Hüllsman Reference Horstmann and Hüllsman2009), despite the higher prevalence of poverty among divorced women after retirement.
Many of the problems raised in this paper are likely to dwindle as women's careers increasingly resemble those of men. However, it should be noted that many of the women we have investigated are currently poor and will remain so for the next couple of decades if no action is taken. Moreover, even though young women today do increasingly resemble men in their labour market participation, more so in some countries than in others, their participation is still not identical. Even in the Scandinavian countries with impressive records on gender equality, part-time work is still much more common among young women than among men (OECD 2012). Therefore, simply waiting for the situation to equalise itself is unwise. Furthermore, the gains made by women in the labour market might very well be offset by increased instability in the family sphere as more and more marriages end in divorce and fewer and fewer individuals get married. Therefore, the consequences of both contrasting tendencies for poverty among future elderly women are far from clear. For these reasons, addressing the issues affecting current elderly women now may, in fact, prove relevant and beneficial for future generations in protecting against some of the pitfalls of growing old in modern western societies.
