1When conducting sociodemographic surveys, considerable efforts are required to establish a representative sample, to reach the selected persons and convince them to take part, and to obtain exact – or at least consistent – answers to the questions asked. This last requirement is fulfilled if two conditions are met, i.e. the questions are understood and accepted in the same way by everyone, and the interview conditions are as similar as possible for all respondents. The French version of the European Generations and Gender Survey was conducted by INED in 2005, 2008 and 2011 using face-to-face interviews, with experienced, professional interviewers from the French statistical office (INSEE). Using data from the first survey wave, Patrick Festy, Joëlle Gaymu and Marc Thévenin show that among persons in a union, men’s and women’s answers on financial wellbeing are generally similar, but that the partner’s presence at the interview considerably influences the answers given, and in opposite directions for men and women. This article provides a timely reminder that analyses of survey data must take account of data collection effects which may modify respondent’s responses in complex ways.
2The quality of answers to a questionnaire depends on two key parameters: the persons who responds, and the conditions in which the interview takes place. This article aims to measure the impact of certain interview conditions on men’s and women’s answers to a question on the financial situation of their household.
3Research on interview conditions has focused heavily on the mode of questionnaire administration (one-to-one interview, phone or internet, self-administered questionnaires, paper or electronic formats, etc.), often in association with analyses of the suitability of different techniques for asking questions on highly personal, intimate or difficult subjects (Riandey and Firdion, 1993). In this field of study, an important question concerns the environment in which the interview takes place. Is someone else present in the room other than the interviewer and the respondent, and does this presence affect the quality of information obtained? Among persons liable to be present, we know that children, parents and parents-in-law have little impact, unlike the partner, who is more often an influential presence (Aldeghi, 1998). Some studies (although few in the field of demography) have measured the effect of the partner’s presence in surveys involving factual and subjective questions on marriage, family and income.
4For William S. Aquilino (1991), when the information is factual – even if the subject is sensitive (in this case, a question on conjugal life with a previous partner) – the respondent tends to give more accurate answers if the partner is familiar with the situation, but not if he/she knows little about it. In this latter case, the respondent tends to give an answer designed to reassure the partner present at the interview if the information is emotionally charged for him or her, but not if the partner is indifferent to it. The same applies to subjective questions: the respondent’s answers are coloured by their potential effect on the partner who is present.
5Aquilino’s hypotheses were confirmed by his research on illicit drugs. François Héran also showed that the respondent’s partner influenced answers to questions on voluntary activities (1988) and help with children’s school work (1994). Arnaud Régnier-Loilier (2007, 2009) returned to this topic more recently, with the French version of the Generations and Gender Survey (ERFI, Étude des relations familiales et intergénérationnelles), to study the effect of the partner’s presence on answers to factual questions. He concludes that when respondents are asked questions on the division of household chores, the partner’s presence serves as a “lie detector”, reducing the tendency – among men especially – to give stereotypical answers in line with the egalitarian norm, thereby prompting more accurate responses than in a one-to-one interview. In contrast, the partner’s presence leads to under-reporting of the respondent’s previous unions, notably by men, revealing that this is a sensitive topic, difficult to broach when the current partner is present.
6These results can be interpreted through the prism of gender relations within the couple. For sociologists, the social habitus pushes men and women into predetermined roles characterized by male domination (Pierre Bourdieu, 1998), while for economists, the gender division of roles is an effective means to maximize wellbeing (Becker,1991). To establish some common ground, we should perhaps, like Michel Bozon (1991), apply the notion of “accepted domination”, or, like John F. Zipp and colleagues (2004), that of a “hidden power in marriage”, i.e. a female propensity to value joint commitment to the relationship, and a male preference for independence.
7In this field of research, there has been strong focus on questions liable to generate conflicts of interest between partners, especially those where the answer might place one or other in an uncomfortable position. We thought it would be interesting to approach this question from a different angle, using an apparently more neutral question on the household’s relative success or failure in making ends meet. This question is on the borderline between factual observation and subjective assessment, and provides a window into the range of economic, cultural and social influences that may potentially differ between the sexes.
8There is abundant literature on the question of inequality between male and female roles within the couple at working ages (Hakim, 2003; Sümer, 2009; Esping-Andersen, 2009), but studies among older couples are more scarce. For this reason, we will concentrate on persons aged 50-79, who are all the more interesting for being survivors of the generations in which the traditional “male breadwinner” model prevailed and which pioneered a more egalitarian approach to the gender division of roles. Moreover, at these ages, when many respondents are already retired, the partner is more often present at the survey interview, so it is especially important to study his or her influence. Other studies have shown that the determinants of male and female subjective well-being are not necessarily the same at these ages (Pinquart and Sörensen, 2000; Gaymu et Springer, 2010), so another aim of this article is to verify whether men and women have differing perceptions of their financial situation, and whether data collection conditions have any gender-specific effect on older men’s and women’s reported financial wellbeing.
Number of respondents by sex and age group
Number of respondents by sex and age groupNote: The numbers in italics indicate the number of respondents for whom there are data for all variables of our models.
I – Data, variables and method
9The data are taken from the first wave (2005) of the French version of the Generations and Gender Survey (GGS), known as ERFI (Étude des relations familiales et intergénérationnelles) conducted jointly by INED and INSEE. The sample includes 2,471 cohabiting couples  with one randomly selected respondent aged 50-79 (Table 1). In 52% of households, the respondent is male.
Measure of perceived financial hardship or wellbeing
10The analysis is based on a question about the respondent’s perception of the household’s financial situation. Questions of this type are widely used in surveys on household living conditions, and underpin “subjective” approaches to poverty, either as a direct measure or as part of an overall score of financial hardship combining answers to several questions. These questions may ask respondents whether the household is forced to dip into its savings to pay regular bills or outstanding debts, whether their income is below the minimum needed to make ends meet, or, more rarely, whether they see themselves as poor. These questions are sometimes criticized for being too subjective because they merely seek to verify whether the needs expressed by the household are greater than its disposable income (Lollivier and Verger, 1997; Verger, 2005). While level of income or monthly household expenses are indeed objective and systematically robust determinants, financial difficulties may also arise from overspending in households with an otherwise comfortable standard of living.
11In the ERFI survey, the exact wording of our main question is “Thinking of your household’s total monthly income, is your household able to make ends meet…” The respondents gave one of the six possible answers in the following proportions: with great difficulty (3%), with difficulty (9%), with some difficulty (18%), fairly easily (38%), easily (27%), very easily (5%).
12To better appreciate the financial situation of those who make ends meet “with difficulty”, we combined this answer with those given to another set of questions designed to assess the household’s material situation, such as whether or not they were able to save, to afford certain items or activities, or to pay off debts as scheduled. As one in five of these respondents were not unable to save, pay rent, etc., we excluded them from the group of persons reporting a difficult situation and placed them in the “fairly easily” group. The share of respondents who made ends meet with difficulty (including “with great difficulty” and “with some difficulty”) thus fell from 30% to 27%. Conversely, 31% were able to make ends meet “easily” (including “very easily”). The intermediate situation (“fairly easily”) is the most common, accounting for 42% of respondents.
13Opinion about making ends meet is an ordered variable grouped into three modalities – “with difficulty” (including “with great difficulty”, “with difficulty” and “with some difficulty”); “fairly easily”; and “easily” (including “easily” and “very easily”) – so generalized ordered logistic models were used.  Here, the parallel odds assumption must be considered when analysing the effects of an independent variable. When this assumption is accepted, the effect measured by the regression coefficients or their corresponding odds ratios (OR) shows no significant difference, whether comparison is between households making ends meet with difficulty (versus fairly easily or easily) or between those making ends meet easily (versus with difficulty or fairly easily). The determinants of financial wellbeing and hardship can thus be considered interchangeably. If the assumption is rejected, the coefficients are more complex to interpret, especially if their signs are different. This situation was observed only once, however.
Interview conditions and other explanatory variables
14In scientific surveys, interviewers are routinely asked at the end of the questionnaire to provide information about the interview conditions, notably the presence of one or more other people and their degree of involvement. The ERFI survey is original in that it provides information on interview conditions at five key moments (“clocking points”) of the interview (Régnier-Loilier, 2006): at the beginning, just before questions on the division of childcare tasks, on division of domestic tasks, on partnership quality, and finally at the end of the interview. The question on making ends meet is in the last part of the questionnaire, but certain questions relating to the household’s perceived financial situation are asked earlier, and are also liable to be affected by the presence of a third party. We therefore chose three indicators:
15P1: partner clocked as present before the questions on the division of domestic tasks
16P2: partner clocked as present at the end of the interview
17P1P2: partner present at one or both clocking points.
19Other variables were introduced into the models (Table 2), most of which are systematically used in this type of analysis,  i.e. income, health (disability), level of education, home occupancy status, labour market situation (working person in the household), and household composition (children and type of co-residence). Alongside these standard variables, we added other additional potential determinants.
Sense of financial wellbeing or hardship, and its explanatory dimensions
Sense of financial wellbeing or hardship, and its explanatory dimensions
20Given that a large share of the households are elderly, they may receive informal support from family, friends or neighbours for care, household tasks or paperwork. This informal support must be taken into account as it may reduce the need for paid help, thereby lowering the household’s expenditure. In the ERFI survey, this support is identified in a rather subjective manner by asking respondents whether there are “plenty of people that they can lean on in case of trouble”. This broad wording may cover situations which affect the household’s material situation either directly or indirectly.
21Moreover, the randomly selected survey respondent is not necessarily the person most qualified to assess the household’s ability to make ends meet. The ERFI survey is able to control for this via one of the questions on the division of domestic tasks in which the respondent is asked to say who looks after the household’s finances (paying bills and keeping financial records). Likewise, and in order to characterize the couples, not the respondents, other variables were considered – i.e. differences in age and educational level between partners, women’s contribution to household income – in order to illustrate the inequalities of position, and hence perhaps of power, between the two partners.
22To measure the partners’ relative economic position, we calculated the female-to-male ratio of income within each couple. The link between this ratio and the household’s standard of living is by no means systematic. A high ratio may indicate that the male partner’s income is very low, due to unemployment, illness or disability for example. Moreover, in couples reaching the age of 60, the two partners very rarely claim a retirement pension in the same year. Compared with men, women more frequently have to wait until beyond the legal retirement age to receive a full pension. 
II – Results
23Before examining how the presence of another person affects respondents’ opinions about their household’s financial situation, we will show how interview conditions vary by the respondent’s sex. The aim is to verify whether the persons interviewed in the presence of their partner are different from those interviewed alone. So that we can judge the impact on the comparison between respondents’ perceived financial situations, our preliminary analysis uses the same independent variables as those described above.
Different interview conditions for men and women
24Whatever the clocking point, the partner was present during the interview in around 40% of cases, and in more than 45% of cases the partner was present at one or both points (Table 3). Women were more frequently interviewed alone than men, with their partner being present in 37% of cases, versus 53% for men. This is probably because women are more frequently at home than men, even at ages when men are no longer working: after retirement, gender disparities in the division of household chores among these generations continue to reflect the situation that existed at younger ages (Solaz, 2009; Anxo et al., 2011), with women more frequently staying in the home and men preferring to get out and about.
Percentage of interviews conducted in the partner’s presence
Percentage of interviews conducted in the partner’s presenceNote: The percentages only concern respondents for whom there are data for all variables.
25Moreover, for both men and women, respondents interviewed alone have different characteristics from those interviewed in the presence of their partner (Table 4). The latter are older (among men, 32% are aged 70-79, versus 19% of those interviewed alone), more often have disabilities (29% versus 21% for both sexes interviewed alone), have a lower level or education or income (for example, a quarter of the men with a partner present report an income of below 1,500 euros, versus 15% of those interviewed alone), are more often in an inactive couple and less frequently report managing the household finances on their own. In addition, when women are interviewed in the presence of their partner, there is a larger average age difference between partners (2.6 years versus less than 1.9 years in one-to-one interviews).
Distribution of respondents or households by the modalities of various characteristics, by respondent’s sex and partner’s presence or absence during the interview
Distribution of respondents or households by the modalities of various characteristics, by respondent’s sex and partner’s presence or absence during the interviewNote: The percentages only concern respondents for whom there are data for all variables.
26All other things being equal, men are present at their partner’s interview less often than women (odds ratio (OR) = 0.53), and the presence of a partner corresponds to specific types of couple (Table 5). Certain determinants are common to both male and female respondents: being older increases the likelihood of a partner’s presence, while a high level of education (OR = 0.46)  reduces it. When the respondent is a man, one other factor has an influence: his partner is more often present if neither partner is in employment (OR = 1.54). When the respondent is a woman, having no-one to lean on, or being solely responsible for the household finances reduces the probability of the partner being present (OR = 0.71). Last, the younger the woman with respect to her partner, the greater the likelihood that he will be present during the interview.
27So the couples where both partners are present at the interview are selected, partly on the basis of criteria which differ for men and for women. In addition to the constraints of life (for example, working partner unavailable), as a general rule, situations of greatest sociodemographic vulnerability (being older, low level of education, etc.) favour the partner’s presence, but these characteristics are more numerous among women. This emphasis on the respondent’s vulnerability suggests that factors of male domination or protection play a role, partly offsetting the mechanical effect of women’s more frequent presence in the home. It is difficult, however, to determine the extent to which this reflects a desire to provide mutual support (help with memory recall, knowledge of the subject, for example) or rather a desire to control and dominate. True, the partner may arrive uninvited during the interview, but the respondent may also have asked him or her to attend, or the two may have decided jointly; and the partners of women in the most fragile situations may be more involved in the domestic sphere (Glaude and de Singly, 1986). The variables used to illustrate the degree of inequality between partners do not all work in the same direction, so it is very difficult to interpret the partners presence at the interview in terms of gender roles.
28The link between a large age difference between partners and women’s greater risk of being interviewed in the partner’s presence may result from an effect of the partner’s older age (and hence greater likelihood of being at home) or from a classic example of women’s “accepted domination” (Bozon, 1991), but no such link is observed for differences in educational level. Regarding contribution to household income, a strong contribution by the woman goes hand in hand with a more frequent presence of the partner at the interview.
29To what extent do these different interview conditions affect men’s and women’s perceptions of their ability to make ends meet? The factors that increase the likelihood of the partner’s presence do not necessarily all affect the sense of financial hardship or wellbeing in the same way. For example, persons interviewed alone are over-represented among the youngest age groups and in households with the highest incomes (Table 5); yet we will see that the former see themselves as less well-off than the average, and the latter the reverse. These selection effects may shape perceptions of financial situation differently for men and women.
Determinants of partner’s presence during the interview, by sex (dichotomous logistic regression)
Determinants of partner’s presence during the interview, by sex (dichotomous logistic regression)Note: The table gives the exponentials of parameters (odds ratios), except for the constant. For each variable, the odds ratio for the reference modality is 1. The significance coefficient associated with the (p>|z|) test is given in brackets.
Interview conditions and reported financial wellbeing: a strong gender effect
30Around 27% of men and women report that it is difficult to make ends meet, and 31% report that it is easy (Table 6). This similarity of men’s and women’s responses is observed at all ages, although respondents in their 50s more often report difficulties than the oldest age group: one-third of the former versus just over 20% of the latter (25% of men and 22% of women). Measures of financial wellbeing give differences in the reverse direction.
Percentages of respondents who report that it is difficult or easy to make ends meet, by age, sex and partner’s presence at the interview*, °
Percentages of respondents who report that it is difficult or easy to make ends meet, by age, sex and partner’s presence at the interview*, °Notes: The percentages only concern respondents for whom there are data for all variables.
*: Significant difference (at 5%) between partner’s presence or absence.
°: Significant difference (at 5%) between men and women.
31The interview conditions give rise to a clear gender effect, with the partner’s presence influencing men’s and women’s responses differently. Whatever the interview conditions, men’s answers are identical: Around 30% aged 50-59 and 24% aged 60-79 find it difficult to make ends meet, while 24% and 34%, respectively, have no financial problems (Table 6). By contrast, women are much more negative in their partner’s presence than when interviewed alone. They more often report difficulty making ends meet (39% versus 31% at ages 50-59, 26% versus 18% at ages 60-79) and less often the opposite (21% versus 31% at ages 50-59, 31% versus 38% at ages 60-79). So the interview conditions appear to modify men’s and women’s perceived financial situation in different ways. Are women more influenced than men by the presence of their partner? Or does this reflect differences by respondent’s sex in the selection of couples where the partner attends the interview? A regression analysis can be performed to eliminate these structural effects.
32A logistic model estimates the odds ratio (OR) of reporting financial wellbeing rather than hardship. An OR below 1 indicates less wellbeing than the reference category, and an OR above 1 indicates more wellbeing.
33For the sample as a whole, all other things being equal, there is no difference between men’s and women’s responses (OR = 0.97, Table 7), or those in the partner’s presence or absence (OR = 1.02). But the partner’s presence strongly influences men’s and women’s responses in opposite directions. If the partner is present, women are less likely than men to report financial wellbeing (OR = 0.72), but if interviewed alone, they are more likely to do so (OR = 1.24).
Effects of partner’s presence on reporting of financial wellbeing by respondent’s sex (ordinal logistic regression)
Effects of partner’s presence on reporting of financial wellbeing by respondent’s sex (ordinal logistic regression)Notes: The table gives the exponentials of parameters (odds ratios). The significance coefficient associated with the (p>|z|) test is given in brackets. The independent variables are the same as those of the models presented in Table 8. The detailed regression results are given in the Appendix Table.
34Having eliminated the differences in household structure linked to the interview conditions, we can see that respondents’ answers are indeed affected by the interview conditions: in their partner’s presence, women give more negative answers than men, but more positive ones when interviewed alone. The strength of this effect of different interview conditions on men’s and women’s answers is also illustrated by the finding that women not only reply more negatively than men when in the presence of their partner (OR = 0,72), but also more negatively than women interviewed alone (OR = 0.76). Conversely, men interviewed in their partner’s presence differ from those interviewed alone (and contrary to the descriptive results) by a sharp increase in the probability of reporting financial wellbeing (OR = 1.31).
35In a predictable role play in front of their partner, respondents tend to say what they think the other expects to hear, or what they consider to be socially correct, most men in these cohorts being the breadwinners and most women the housekeepers.
Interview conditions and determinants of financial wellbeing: few differences between men and women
36This gender effect is especially remarkable given that it is one of the rare factors influenced by the interview conditions: whether men and women are interviewed alone or in the presence of their partner, the other determinants of their capacity to make ends meet are very similar. This is notably the case for all variables describing the household’s socioeconomic situation which, in line with the literature (Senik, 2002), have a major impact on perceived financial situation (Table 8). For example, the higher the respondents’ income, the greater their tendency to not report any difficulty in making ends meet: the odds ratios range from 0.25 to 0.42 for persons with a low income, up to 2.64-3.53 for those with the highest incomes. Being a homebuyer (and often in debt) or a tenant reduces the likelihood of reporting financial wellbeing (OR from 0.28 to 0.50). These are the key determinants of financial hardship. We note also that, whatever the interview conditions, and for men and women alike, the presence of a working person in the household is the only factor with no effect on financial difficulties.
37Health problems negatively affect perceived capacity to make ends meet (OR of 0.45 to 0.68), except for men interviewed alone. The role of the other factors also varies according to interview conditions, sometimes even in reverse directions for men and women. This is true for the family situation. Having at least one cohabiting child reduces the likelihood of reporting financial wellbeing for men (0R of 0.43 in the partner’s presence and 0.32 alone), and only for women interviewed alone (OR of 0.39). Likewise, it is only in this survey context that the lack of people to lean on in case of trouble is mentioned as a source of difficulties (OR of 0.58 among men and 0.48 among women). In their partner’s presence, do respondents – women especially – tend to censor certain factors that might reflect negatively on their family circle?
38Other factors, i.e. a low educational level for women (OR = 0.58) and handling household finances alone for men (OR = 0.63), only have a negative impact on perceived financial wellbeing when respondents are interviewed alone. Last, the variables illustrating the degree of inequality between partners do not affect men’s perception of their financial situation. Among women, however, differences in age and in educational level are associated with perceived financial difficulties only for those interviewed alone, while inequality in economic power affects responses only in the partner’s presence (OR = 0.52). On this latter point, some have suggested that differences in men’s and women’s preferences in terms of income equality – the former preferring the traditional breadwinner model and the latter wishing for greater equality – could be a source of greater dissatisfaction for women in cases of very inegalitarian income distribution between partners (Bonke, 2008). Our results follow this pattern, but women with high economic power do not mention lower financial wellbeing except in the presence of their partner. In other words, in these cohorts, being in a union where both members depart from their traditional roles gives rise to dissatisfaction for women, only when the partner is listening in.
Determinants of the household’s financial wellbeing by partner’s presence or absence during the interview (ordinal logistic regression)
Determinants of the household’s financial wellbeing by partner’s presence or absence during the interview (ordinal logistic regression)Notes: The table gives the exponentials of parameters (odds ratios). The test value (p>|z|) is given in brackets. For 77 variables out of 80, the parallel slopes assumption is accepted at the 5% level (the three exceptions are shown in italics).
39Although interviewers are generally instructed to talk to respondents alone, a large share of interviews are conducted in the partner’s presence: 40% in the ERFI survey when the respondent is aged 50 or over; around 50% in the SHARE survey (Survey of Health, Ageing and Retirement in Europe), with a proportion ranging from 25% in Greece to more than 60% in Belgium and Italy. Our analyses show that couples where the partner attends the interview do not have the same characteristics as those where the respondent is interviewed alone, and the differences between the two groups are more marked when the respondent is a woman. In addition, the effect of interview conditions on reported financial difficulties is the opposite for men and women. While there is no difference between men’s and women’s responses over the sample as a whole, men more often report having financial difficulties than women when interviewed alone, while the reverse is true when the partner is present, in which case they more frequently report financial wellbeing than women.
40So the partner’s presence appears to alter men’s and women’s perceptions of reality, or at least their reporting of it. The economic dimension has already been explored in two studies to measure the impact of the partner’s presence during the interview (Aquilino, 1991; Cantillon and Newman, 2005). In both cases, men interviewed with their wife tended to under-estimate their income (or over-estimate their hardship) compared with those interviewed alone, and the reverse was true for women. A gender contrast was thus observed, but its direction was reversed with respect to the present research. Both studies concerned objective indicators of economic wellbeing; the situation appears to be very different when a subjective indicator is used. This confirms the importance of these two objective and subjective dimensions in measures of well-being and poverty (Verger, 2005), but it also serves as a warning, since the perceptions studied here may relate to a reality which is itself affected by reporting bias. The regression analyses allow us to examine the question without the risk of ignoring certain factors that contribute differently to women’s and men’s reported financial well-being or of neglecting household selection biases linked to interview conditions. Last, this result perhaps also reflects differences in men’s and women’s capacity to express negative feelings (Simon and Nath, 2004; Mirowsky and Ross, 1995).
41Some authors have shown that the presence of other people improves the quality of information obtained, notably by helping the respondent with memory recall (Auriat, 1996) or by preventing exaggeration: “The example of the division of household tasks suggests rather that the partner’s presence causes respondents to “moderate” their responses, while his or her absence encourages them to overstate their role and report a more balanced division of tasks, in line with a certain egalitarian norm” (Régnier-Loilier, 2007). Others argue in favour of one-to-one interviews, since the presence of third persons, the partner especially, tends to result in more stereotypical responses (Aquilino, 1991).
42It is impossible to decide one or way the other, but our results show the importance of collecting detailed information on interview conditions. Even if the factors associated with financial hardship are identical for men and women, their responses are not the same when the partner is present or absent, and the respondents’ characteristics may differ according to whether or not they are interviewed alone. With the growing importance of quantitative survey data in measuring behaviours and attitudes, further methodological research is needed in this area.
Appendix Table. Determinants of the household’s financial wellbeing (ordinal logistic regression)
Appendix Table. Determinants of the household’s financial wellbeing (ordinal logistic regression)Note: The table gives the exponentials of parameters (odds ratios), except for the constant. For each variable, the odds ratio for the reference modality is 1.
Institut national d’études démographiques.
Correspondence: Patrick Festy, Institut national d’études démographiques, 133 Boulevard Davout, 75980 Paris Cedex 20, tel.: +33 (0)1 56 06 22 01, email: email@example.com
Households with a person other than a child living in the home were excluded from the sample to ensure a homogeneous study population.
The ordered logistic model is estimated with Stata using the OLOGIT command, with the associated parallel odds tests: Wald test for each variable and Brant test to verify the hypothesis on the complete model.
In Table 3, the partner’s presence appears to be quite similar with the three indicators, and the analysis results are largely unaffected by the choice between P1, P2 and P1P2.
Two issues of the journal Économie et statistique are devoted to poverty measures: in 1997, issues 308-309-310 “Mesurer la pauvreté aujourd’hui” (Herpin and Verger, 1997); in 2005, issues 383-384-385, “Les approches de la pauvreté à l’épreuve des comparaisons internationales” (Verger, 2005).
In the 1934 cohort, a third of women claimed their pension at age 65 versus less then 4% of men, most of whom (86%) retired at age 60 (Bonnet et al., 2004).
Likewise if the partner is highly educated. This variable is tested on another model, not presented here.