Source: Map 2.0, Census Atlas of the 2007 Population and Housing Census of Swaziland
5. Indicators 304. There are a number of conventional indicators of household composition that may be relevant for different kinds of gender analysis. Several of these go back to the household classification schedule introduced in the previous section. Rather than making comparisons in terms of male and female-headed households in the aggregate, which mix all of these distinct categories, it is much better to make comparisons between equivalent household composition categories. The comparison can be made both in terms of how much more common the female-headed households are represented in the relative incidence of poverty or vulnerability. Most census reports, however, still limit themselves to the conventional division of households into male-headed and female-headed, despite the known limitations of these categories. Figure 8 below illustrates the kind of maps that can be constructed with this information, based on the 2007 census of Swaziland.
305. Another household composition indicator, with more conventional applications, is the so-called headship ratio, i.e. the percentage of men or women of a given age that are heads of their households. It is useful primarily for household projections, as the population by age and sex at any given time, multiplied by the respective projected headship ratios, yields the number of households. Obviously, it also allows the comparison of the percentage of men and women of any particular age that are heads of their households. Again, as discussed earlier, the use of the sex of the head of household has some serious limitations for analytical purposes. Results obtained by comparing male/female heads of household should be backed up by more detailed analysis in which type of household is brought in as an additional explanatory variable.
306. The concept of headship ratio can be extended to other categories of household status, allowing the construction of descriptive indicators such as the percentage of women of a certain age who may be the following:
a) Living in their parental household;
b) Living by themselves in a one person household;
c) Living with a husband or partner, without children;
d) Living with a husband or partner, with children;
e) Living without a husband or partner, but with children; or
f) Living with unrelated household members, as a domestic servant.
Comparing the evolution of these indicators between successive censuses can provide interesting information about changes in gender relations, e.g. an increase in the percentage of women aged 20-24 who are living alone or with a husband or partner, without children, which might indicate that young women are having greater opportunities to extend their education or to work, prior to forming families, than in the past. Obviously, these comparisons only make sense if the concept of head of household, and the operationalization of this concept (e.g. the response category, or who decides to designate the head) have not changed over time. Many countries have adapted their strategies to appoint the head of household.
307. In section 2, co-residing domestic servants were mentioned as a particularly vulnerable group. This suggests compiling data on the following indicators:
a) Percentage of households (possibly divided by type, such as one-person, couple without children, or couple with children –preferably by broad age category) that have co-residing domestic servants, by sex of the domestic servant; and
b) Percentage of working women that make their living as domestic servants (co-residing and not).
308. In recent decades many countries have seen a steady growth in the number of one-person households. Especially in Europe the rate of one-person households increased rapidly between 1970 and 1990. Sweden had the highest rate of one-person households (39.6 per cent) in 1990. The growth of one-person households has some important consequences for policy making. Research showed that for instance, energy consumption per person is higher for one-person households than for households with more persons (Hanssen; Scherg and Christensen, 2009). Living costs (per person) for one-person households are generally higher than for multi-person households. Also, a single person may be more vulnerable, as there is no back-up in case of unemployment or other mishaps. The ‘Worlds Women 2012 (United Nations, 2010 a) showed that women run a higher risk than men when living in a one-person household. Therefore, ample attention should be given to this category when studying gender issues in poverty research.
309. An interesting perspective on household composition is that of the children. Here one may compute the percentage of children under age 15 (or another relevant age limit) that live:
1. With both parents and no other adult household members;
2. With the mother only;
3. With the father only;
4. Without their parents, cared for by grandparent(s) or other family members;
5. With both parents and a grandmother and/or domestic servant; and
6. With the mother and a grandmother and/or domestic servant.
310. Another, potential source of useful information has to do with the physical characteristics of the housing unit, in addition to the characteristics of the household as such. These may provide useful information on the level of comfort and housekeeping chores. For example, the proportion of girls/women living in a dwelling without easy access to water or which uses firewood for cooking is a good indicator of the burden they assume to fetch water and cooking fuel for the household, which may have adverse consequences for girls’ school attendance and women’s labour market participation. The proportion of women living in dwellings without a proper kitchen, especially if cooking is done using firewood or charcoal, is an indicator of the health conditions to which women are exposed in preparing food. The size of the household may be used as a further qualifier of the amount of work involved in these household chores.
311. The following indicators from the Minimum Gender Indicator Set approved by the UN Statistical Commission in February of 2012 can whenever the data are available be computed from census data:
1. Proportion of population using the internet, by sex (asked, for example, in the 2010 census of Qatar, in addition to computer use);
2. Proportion of population using mobile/cellular telephones, by sex; and
3. Access to mass media and information and communications technology.
6. Multivariate and further gender analyses 312. Household composition, in combination with other explanatory variables, can be an important explanatory variable for various types of multivariate analyses. As was already indicated in other parts of this document, the concept of household headship, as implemented in most censuses, is riddled with ambiguities. It is possible, however, to gain a better understanding of what factors determine household headship, by decomposing conventional headship rates by age and sex (particularly female headship rates by age) in terms of some potential determining factors. Handa (1996) used these methods to look at the determinants of female household headship in Jamaica. Joshi (2004) used Matlab data to analyse household headship of widows and married women in rural Bangladesh; her results cannot be easily replicated with census data.
313. In fact, this kind of analysis involves answering two different questions simultaneously, namely why the composition of the household is as it is and why, given this composition, a given person has been selected as head. Therefore, it is best to do this kind of analysis by type of household. In order to explain the incidence of female headship in one person households, for example, the question is really why this particular person is living alone, rather than with others, and the relevant characteristics those of the household head him/herself and those of the geographical setting (e.g. urban/rural residence). In households with more than one adult member, on the other hand, the characteristics of the other household members may be at least as important as those of the household head him/herself.
314. One way to carry out the analysis is as a logistic regression, in which the dependent variable is female headship and the independent variables are household characteristics such as the number of adult male and female household members, the number of children, the number of economically active male and female household members, per capita household income, etc. A more refined, but more complex procedure is to use a multinomial regression model with explanatory variables such as the age, sex, marital status, economic activity, level of education and (if available) personal income of each adult household member. The outcome of such a model is a headship probability for each individual (adult) household member, rather than just the probability that the head of household will be female.
315. In addition to its greater complexity, one of the problems of the multinomial approach is that not all of the relevant variables can be easily determined in many censuses. For example, one of the likely determinants of household headship is how many children a given candidate for headship has living in the household. But given the way family relationships are determined in most censuses, this can be hard to establish. It is easier in the case of women because women over age 15 or between the ages of 15 and 50 are usually asked for their number of surviving children. Some censuses, like the one of Vanuatu (2009), even ask whether each person's mother lives in the household. But in the case of fathers, such questions are less common and only the children of the actual head of household can be easily determined.
316. To avoid such problems, the following analysis of headship in Vanuatu takes a more limited approach, by looking only at nuclear families with children and trying to determine which characteristics of both of the partners determine whether the choice fell upon the woman, rather than the man. Nuclear households without children were not considered, because the intention was to look at the effect of the number of children (both sons and daughters) on the probability that the mother would be chosen as head of household.
317. Instead of separate regression equations for males and females, based on household records, with the characteristics of both partners. The sex of the head was used as dependent variable. Only explanatory variables which describe the combined characteristics of both spouses were included. The results of this analysis are shown below, in Table 26.
Table 26: Vanuatu (2009) - Logistic regression to predict the choice of a female head of household in nuclear households with children, depending on the characteristics of the couple
Source: Population and Housing Census of Vanuatu (2009)
318. An even simpler way to analyse the same data is a logistic model in which the characteristics of both sexes are analysed separately. Although the previous analysis is preferable from a theoretical viewpoint, the results are actually quite similar. In this second approach, a logistic regression model was used in which the dependent variable was ‘0’ if the husband was chosen as head of the household and ‘1’ if the wife was selected. In Vanuatu, only 6.5 per cent of nuclear households with children had a female head. The regression was run separately for males and females. This was done to see if the impact of each of the explanatory variables was different for males and females. A number of models were tested, next to characteristics of the respondents; also some predictors of both spouses combined were included. For instance, an explanatory variable was created which incorporated the educational attainment of both spouses. In this variable, three educational attainment categories of the respondent (less than primary, primary and more than primary) were linked to the same categories for the respondent’s spouse. The same strategy was followed for the variable ‘worked/did not work’. The values in the exp(B) column present the odds ratio for women in that particular category to be selected as head compared to the reference group.
Table 27: Vanuatu (2009) - Logistic regression to predict the men or women with certain individual characteristics will be chosen as heads of household in nuclear households with children
Partners both less than primary
Primary, partner less than primary
More than primary, partner less than idem
Primary, partner less than primary
Both partners primary
Primary, partner more than primary
More than primary, partner less than idem
More than primary, partner primary
Both partners more than primary
Age 10 year age-groups
15 - 24 yrs.
24 - 34 yrs.
35 - 44 yrs.
45 - 54 yrs.
55 - 64 yrs.
65 - 74 yrs.
75 - 84 yrs.
No. of sons in hhold
No. of daughters in hhold
Church of Christ
Assemblies of God
Neil Thomas Minsitry
Refuse to answer
Vanuatu by birth
Vanuatu by naturalisation
Age husband - age wife
Working status of couple
Both partners worked
Worked, partner did not work
Did not work, partner worked
Both partners did not work
Source: Population and Housing Census of Vanuatu (2009)
319. Table 27 presents the results of this logistic regression. Compared to couples who have both less than primary education, all other educational combination categories score significantly higher, i.e. women in these educational groups are much more likely to be heads of household. For instance, if a woman has more than primary education and her husband has less than primary, her odds to be head of the household are 2.3 times higher than in the case where both spouses have less than primary education. The largest effect of all variables in the equation is urban/rural. In rural areas the odds for a male to be selected as head of the household are 3.6 times higher than for a female to be selected, after controling for other intervening factors. For women, each extra son increases her likelihood slightly to be selected as head (odds ratio 1.05), while each additional daughter diminishes her changes (odds ratio .975). It should not come as a surprise that women who practice a traditional religion have a much lower likelihood of becoming head of household (odds ratio 0.378). Women who belong to a traditional religion will most probably live in households that are more conservative in terms of the position of women within the family. ‘No religion’ also scores very low (.425). However, it is possible that interviewers interpreted traditional beliefs as ‘no religion’. Finally, it is interesting to see that men who worked, but whose wife did not, have much lower odds of being selected as head of the household. This may show that enumerators did not always use strict rules to assign a head but perhaps made a person head who was readily available (or that women selected themselves as head when an enumerator showed up when their husbands were out to work).
320. Based on data from the 1982, 1990, 2000 and 2005 Chinese censuses, Lin and Zhao (2010) investigated the effect of the sex of the first-born child on his/her chances of living with the mother, the father or both. They decomposed this probability by the different events that affect it: whether the parents got married, whether either of the parents migrated, whether the couple divorced/separated and, if so, who got child custody. Parental mortality was not considered as it is hard to imagine how the sex of the first-born child might affect it. They found a small but significant tendency for first-born boys to live with either or both parents more often than first-born girls. The decompositions indicated that: 1) When unmarried, having a first-born son increases the probability of subsequent marriage; 2) When married, having a first-born son decreases the probability of parent migration and reduces the probability of divorce; 3) When divorced, having a first-born son increases the probability of custody by the father. They also found that all these effects have become stronger over time, particularly in the 2005 census. Again, a likely explanation is the preference by parents of sons over daughters, but other, more subtle possibilities cannot be entirely ruled out. It may be, for example, that parents, knowing that a first-born girl objectively decreases their chances for sustenance in old age, regardless of their own preferences in the matter, take certain compensatory actions (additional children, migration, job changes) that affect their subsequent chances of staying together as a couple.
321. Type of household is an important predictor for most of the topics covered in this manual. Below, some examples from the literature are presented in which the type of household is used as an explanatory variable in the gender study of fertility, mortality and education. As mentioned in the box above, Chu, Xie and Yu (2007) showed in the case of Taiwan that there is a positive relationship between the proportion of girls in the household and the total number of children. This suggests that in countries with marked son-preference parents continue to have children until they have at least one boy. In such countries, computing this correlation coefficient is recommended as a standard practice, together with the sex ratios at birth by parity, to be discussed in the next sub-chapter. Note, however, that this makes sense only for one-family households. Other ways in which household composition data can be used to study fertility behaviour include the Own Children Method, which was briefly mentioned in the sub-chapter on fertility
322. Households may vary in terms of poverty, health status or school attendance of the children, depending on whether certain kinds of household members are present or not. There is a lot of public debate on whether the absence of a father, a mother or both parents has a negative impact on the development of the children. One-parent families, particularly families consisting of a mother without husband and several dependent children, are thought to pose greater risks for the health of children. In fact, the health indicators of such families are often more unfavourable, but so are their socioeconomic characteristics, so that it is not clear if it is the latter or the former that increases the risk. Blakely et al. (2003) have carried out multivariate analyses on data from the 1991 census of New Zealand, which were linked to mortality records in order to control for the socioeconomic determinants. Their conclusion is that there does not appear to be notable variation in relative risk terms of socioeconomic differences in child mortality by age or cause of death and that any association of one-parent families with child mortality is due to associated low socioeconomic position. It may be appropriate to replicate this kind of study in other contexts, to see if similar results are obtained in developing countries.
323. A similar issue that has come up in the literature is whether children are better cared for in female-headed households than in households with male headship. The argument is that mothers will usually make household decisions based on the best interests of their children, but may not be in a position to do so if they are subject to the authority of a male head of household (e.g. Castle, 1993). However, female-headed households are different from male-headed households in a variety of ways and many factors have to be controlled for in order to conclude that this is indeed the determining factor. One study that attempted to do this, for example, is the one by Adhikari and Podhisita (2010), on household headship and child deaths in Nepal, based on the 2006 Demographic and Health Survey of that country. Using a binary logistic regression model which contains the age at first marriage, children ever born, place of residence, ecological zone, literacy status, religion, wealth status, use of family planning methods, visits to a health facility, and antenatal care for the last pregnancy as predictors. Controlling such factors, the authors found that deaths among children born during the last five years were 31 per cent less common in female-headed households than in male-headed households. Basing studies of this kind on DHS, rather than census data has certain advantages because the information on dates of birth and certain potential determinants of child health is more extensive in the DHS than in the census. Nevertheless, to the extent that the census contains the basic data on children ever born and surviving, it should be possible to carry out similar analyses using census data.
324. In some national contexts, the number of siblings and the presence of a grandmother have been identified as having an impact on girls’ school attendance. Parker (2005) used 16,000 aged 6-14 sampled by the 2001 Lesotho Demographic Survey to look at the relationship between residence with a grandmother and current school enrolment for children, ages 6-14, in Lesotho. Logistic regression was used to establish whether having a grandmother living in the household was associated with school attendance. The results showed this association to be positive. Taking this analysis, which is equally feasible with census data as with DHS data, one step further, one might differentiate between 16 categories, depending on the presence of the father, the mother, the maternal grandmother and the paternal grandmother, controlling perhaps for urban/rural residence and the level of education of the head of household to investigate school attendance of boys and girls by age in each category.
325. The importance of studying one-person households was mentioned earlier. The basic question is what type of persons live in such households. The 2009 Vanuatu Population and Housing Census was used to investigate the factors that determine whether a person lives in a one-person household or not. A logistic regression was set up in which the dependent variable was whether the person was head of a one-person household or living in another type of household (with multiple persons). The results of this analysis are presented in Table 28. To show the importance of setting up the right multivariate equation, two models were constructed. In model I, next to sex the following predictors were included: Age, Age2, Citizenship, Urban/rural residence, Education (3 categories), Residence 5 years ago and whether the person worked the week before the census. This model shows that the two categories with the highest odds for a person to be found in a one person household are: ‘being born outside Vanuatu’ and ‘female’. A person who is born outside Vanuatu and who is not naturalized is about 4 times more likely to live in a one-person household than a person born in Vanuatu. Women have a 2.2 larger odds ratio to live alone than men. This finding confirms the results obtained by using a simple cross tabulation between sex and one person/multiple person household. In model II, Marital status is introduced. The results show that this now becomes the most discriminatory variable in the equation. Compared to never married persons, all other marital statuses have much lower probabilities of living alone. Somebody who is in the married state has an odds ratio of almost 0.034 of living alone compared to a never married person. However, the effect of sex becomes relatively minor and is even less than 1 (0.886), which means that, after controling for marital status, the likelihood that a woman lives in a one-person household is less than that of a man.
Table 28: Vanuatu (2009) - Logistic regression for whether the head of household conforms a one-person household or not, by selected explanatory variables
Vanuatu by birth
Vanuatu by naturalisation
Education (3 categories)
Less than primary
More than primary
Worked week before census
Did not work
Residence 5 years ago
Sample place as census
Source: Population and Housing Census of Vanuatu (2009)
326. There is a good chance that the likelihood to live alone is less for men and women who have children with whom they could go and live at an older age. The census provides data on the number of children ever born (CEB) and on the number of children who have deceased, although only for women. To measure the effect of having children on the likelihood of living alone, the number of surviving children of woman was calculated. Then two logistic regressions were carried out – one for each sex - with the same set of explanatory variables as before, but including the number of surviving children in the regression for women (see last two columns of Table 28). The results clearly show that having surviving children, the likelihood to live alone becomes considerably smaller. With each additional surviving child, the odds ratio of living in a one-person household decreases by about 18 per cent.
7. Interpretation, Policy and Advocacy 327. Advocacy efforts to reduce gender inequalities regarding household and family composition should include some or all of the following elements.
a) Addressing diverse living arrangements
As the ICPD Programme of Action notes (Par. 5.1), when policies and programmes that affect the family ignore the existing diversity of family forms, or are insufficiently sensitive to the needs and rights of women and children, parents may face great difficulties in reconciling work and family responsibilities. The Beijing Platform for Action (Par. 46) also stresses that since “many women encounter specific obstacles related to their family status, particularly as single parents,” policies must pay special attention to address their needs and to support family stability.
Special attention must also be paid to the needs of widows and orphans (ICPD Programme of Action, Par. 5.13).
b) Addressing the burden of household and family activities
Promotion of changes in traditional notions of gender-based division of parental and domestic functions, in order to reduce the burden of care and household activities that women and girls frequently face and that jeopardize their educational and professional opportunities.
Along with this, governments should promote quality and comprehensive sexual and reproductive health services in order to ensure women and men the opportunity to balance the size of their families with their needs, desires, and goals. Affordable and physically accessible care facilities (e.g. child-care facilities, kindergartens, care services for those who are ill, disable, elder, etc.) must also be provided to support the different types of families in their efforts to reconcile productive and reproductive roles. Another important action to be taken is the promotion of family-friendly work environments, including the right to flexible working hours and schedules, paid parental and maternal leave, maternal protection, health insurance and social security (Beijing Platform for Action, several articles).
Regarding education, increase enrolment and retention rates of girls may also be supported “by allocating appropriate budgetary resources and by enlisting the support of parents and the community, as well as through campaigns, flexible school schedules, incentives, scholarships and other means to minimize the costs of girls’ education to their families and to facilitate parents’ ability to choose education for the girl child” (Beijing Platform for Action, Par. 80 f.).
The ICPD Programme of Action (Par. 5.2.b) emphasizes the importance of establishing social security measures that address the social, cultural and economic factors behind the increasing costs of child-rearing.
c) Addressing family poverty risk
According to ICPD Programme of Action (Par. 5.4), “when formulating socio-economic development policies, special consideration should be given to increasing the earning power of all adult members of economically deprived families, including the elderly and women who work in the home, and to enabling children to be educated rather than compelled to work”. The text also adds that “particular attention should be paid to needy single parents, especially those who are responsible wholly or in part for the support of children and other dependants, through ensuring payment of at least minimum wages and allowances, credit, education, funding for women’s self-help groups and stronger legal enforcement of male parental financial responsibilities.”