1. Introduction

In this thesis we examine the effects of childlessness and genetic relatedness on sibling relationships. In this introduction we briefly outline the structure of this project.

Recently, it has been argued that individualization processes have thoroughly altered family relationships. First, we will very briefly discuss these sociological perspectives on modernization and individualization (2.1). The core idea is that traditional institutions, e.g. marriage, are being replaced by ‘individualized choice’. Moreover according to this perspective, social relationships become centred upon choice. Therefore (traditional) family ties are expected to dissolve. One of the consequences of individualization is the increasing rates of below-replacement fertility and (voluntary) childlessness observed in Western countries. Under 2.2, we will review findings on (voluntary) childlessness and concomitant characteristics.

We will then discuss biological perspectives on kin selection and human behaviour (3). Following this perspective, we derive that childless individuals are likely to maintain stronger relationships with their siblings than ‘parents’. Such high quality relationships could have inclusive fitness consequences. Relationships between full siblings are also predicted to be better than between half-siblings, or relationships where the siblings are unrelated.

Furthermore, we will briefly discuss how social factors affect relationship quality, from a (micro)sociological point of view (4). This is important as we want to examine the effects of childlessness and genetic relatedness on relationship quality, while controlling for other social factors affecting the sibling relationship, e.g. the financial balance of the relationship.

Using a dataset on kinship relations in the Netherlands, the Netherlands Kinship Panel Study, we will then empirically investigate the effects of childlessness and genetic relatedness on sibling relationships. In the first part of the analysis we will assess the effects of childlessness and genetic relatedness on relationship quality (7.1 to 7.3). In the second part of the analysis, we examine the influence of childlessness and genetic relatedness on other relationship aspects, i.e. investment measures (7.4). Finally we will examine differences in sibling relationship characteristics between the voluntary childless and the involuntary childless (7.5).

Due to considerations of length, additional statistical information and coding procedures can be found on the cd-rom. In addition, due to the number and extent of the analyses, we will only discuss the most relevant results of our analyses, i.e. the effects of childlessness and genetic relatedness, rather than all results. Throughout this thesis we will use five percent as significance level.

2. Individualization, childlessness and social relations

2.1 Individualization

“The general diagnosis is that people’s lives are becoming more mobile, more porous, and of course more fragile. In the place of pre-given and often compulsory types of relationship is appearing the ‘until the next thing’ principle, as Bauman calls it, a kind of refusal of lifelong plans, permanent ties, immutable identities. . . . Instead of fixed forms, more individual choices, more beginnings and more farewells.” (Beck-Gernsheim, 2002: 41)

Society has changed drastically over time and rapid modernization has drastically altered contemporary society. According to several social theorists, e.g. Giddens (1991), these modernization processes have led to the breakdown of traditional institutions (e.g. marriage) in late modern societies. Whereas ‘traditional society’ was characterized by institutions and social structures that guided individuals from birth to death, radicalized modern societies have become more and more centred upon ‘individual choice’ (Giddens, 1991; Beck, 1993; Beck & Beck-Gernsheim, 2002). What is left of traditional institutions and norms will further disappear and make way for individualized choice. In late modern societies, individuals have more opportunities for self-expression but also have more uncertainty. Life in late modern societies has become a quest for identity, with all the perils such quests entail.

Individuals are now forced to choose with whom they maintain social relationships as well as to construct their own identities, whereas in traditional society this was governed by traditional norms and social institutions (Giddens, 1991; Beck, 1993; Beck & Beck-Gernsheim, 2002). Rather than relying on traditional institutions as sources for ‘identity scripts’, individuals in radicalized modern societies are confronted with a wide variety of choices for their identity construction. Individual ‘narratives’, life stories, have become more flexible and open-ended. In the absence of institutional solutions, social relationships, including family relationships, have become increasingly open to negotiation and subject to ‘termination’ (Beck, 1993).

2.2 Childlessness and concomitant characteristics

It is ‘individualized choice’, guided by the changed costs and benefits of children (Becker, 1991), that (partially) explains the below-replacement fertility patterns observed in Western countries today (Morgan & King, 2001).[1] The choice to become a parent in a radicalized modern society requires long-lasting financial and emotional investment (van de Kaa, 2004). Individuals can choose not to make such a commitment and not to have a (or another) child. Thus, in radicalized modern societies, because of the availability of contraception, individuals are ‘free’ to choose how many children they have, or whether they have any children at all (McAllister & Clarke, 1998; van de Kaa, 2004).[2] Research has shown that there is an increasing trend of individuals who choose to be childless in radicalized modern societies (DeOllos & Kapinus, 2002; Park, 2005). For the United States, estimates are that approximately seventeen percent of the women born between 1944 and 1955 and up to twenty-two percent of the women born between 1956 and 1972 will remain childless (DeOllos & Kapinus, 2002).[3] The majority of these women are supposedly voluntary childless (Gillespie, 2003; Park, 2005; see McAllister & Clarke, 1998). Similar estimates exist for the UK and the Netherlands, but there is some variation across European countries (NIDI, 1997/2004; McAllister & Clarke, 1998; Ruddock et al., 1998; Pearce et al., 1999). These increasing rates of voluntary childlessness are predominantly a Western phenomenon (Gillespie, 2003). In many other parts of the world contraceptives and reproductive technologies, which facilitate the choice of (voluntary) childlessness, are still largely unavailable. It should also be noted that data on (voluntary) childlessness in men are scarce (McAllister & Clarke, 1998).

Compared with ‘mothers’, voluntary childless women are more likely to be higher educated and employed (McAllister & Clarke, 1998; Bachu, 1999), have a managerial job (McAllister & Clarke, 1998; Bachu, 1999), less religious (Mosher et al., 1992), less traditionally gender orientated (Callan, 1986) and less conventional (Park, 2005). It is unclear, however, whether there is a difference in life satisfaction between ‘parents’ and childfree individuals (Magarick & Brown, 1981; Callan, 1986; Somers, 1993). However, childless couples tend to report higher marital satisfaction than ‘parents’ (Callan, 1987; Somers, 1993). Although, Somers (1993) found no effect of childlessness on marital satisfaction after controlling for religious affiliation and income. While childless couples report higher marital satisfaction than parents, they believe their friends and relatives view them negatively because of their (voluntary) childlessness (Somers, 1993). In late modern societies, childless individuals, especially women, are also often stigmatized (Campbell, 1985; Park, 2002/2005).

According to individualization theory, the increasing rates of below-replacement fertility and (voluntary) childlessness can be explained by ‘choice’ (van de Kaa, 2004). Individuals who choose to remain childless can allocate their time and money in a different way and devote these available resources to self-expression (see Giddens, 1991).

If individuals do not have children, whether they choose to or not, they can allocate their time and money in a different way. Childless individuals can choose to maintain other strong relationships which they perceive as important as having children. Childless individuals do not face the trade-off between ‘investment’ in their children and ‘investment’ in their other social relationships. This could in part explain the finding that childless couples are more satisfied with their marriage than parents (Somers, 1993).

Following individualization theory, whether or not social relationships, including kin relationships, are of high quality has become increasingly dependent on choice. Therefore, whether or not childless individuals maintain good relationships with their siblings is (solely) a matter of individualized choice. If (traditional) family ties dissolve in radicalized modern societies, as individualization theory argues (Giddens, 1991; Beck-Gernsheim, 2002), nothing prohibits childless individuals from maintaining high quality relationships with their partners or friends instead of with their siblings. Childless individuals could invest more time and money in relationships with their partner and/or friends, instead of with their siblings. So, following individualization theory, there is no apparent reason why childless individuals should (continue to) invest in the relationship with their sibling, other than choice. Childless individuals could choose to invest in their relationship with their partner or friends instead. There is possible evidence that childless individuals do so, i.e. their higher marital satisfaction (Callan, 1987; Somers, 1993).

However, from an inclusive fitness point we might expect that childless individuals do maintain ‘strong’ relationships with their (full) siblings, as we will discuss in the following section. These relationships bring along costs in time and money. Maintaining such strong relationships, which require investments in terms of time and money, could be adaptive, at least to a certain extent, and could generate some inclusive benefits for childless individuals. It could be an example of human adaptive decision making, even in radicalized modern societies.

Moreover, if according to individualization theory, family relationships become increasingly centred upon individualized choice rather than (traditional) norms or institutions, we would not expect strong differences in sibling relationships according to genetic relatedness. If choice becomes the crucial determinant of family relationships, genetic relatedness should only play a minor role in sibling relationships and their characteristics. From an inclusive fitness point of view, however, sibling relationships are expected to vary in quality, for instance, according to genetic relatedness.

3. Inclusive fitness and adaptive decision making

Hamilton (1964) developed the concept of inclusive fitness, which can be defined as “the sum of direct and indirect fitness” (Alcock, 1998: G2). Direct fitness refers to genes contributed to the following generation by an individual through personal reproduction, whereas indirect fitness refers to genes contributed in the next generation by helping (non-descendant) relatives. As shown for insect societies (Wilson, 2000[1975]), an individual’s apparent altruistic behaviour can in certain cases be explained by Hamilton’s Rule (r*b>c). For humans, this implies that, under circumstances, it is beneficial to behave altruistically towards (closely) related kin as this might increase their inclusive fitness. There is evidence that kin selection has played a crucial role in shaping human behaviour and influences human behaviour even today.

Within traditional societies, kin often provide (mutual) material and immaterial support. Among the Yanomamö of Venezuela, coalitions for raids are formed based on relatedness (Chagnon & Bugos, 1979). Closer related individuals are more likely to form coalitions in raids, than distantly or unrelated individuals. Similarly, an analysis of data on Norse earldoms, Icelandic families and English royalty, showed that relatedness influences the likelihood of (mutual) support and lethal conflict (Johnson & Johnson, 1991). Support of closely related individuals was common, but fatal conflict between closely related individuals was highly unlikely. An analysis of Viking sagas showed similar results (Dunbar, Clark & Hurst, 1995). In these sagas, closely related kin were found to form more stable, unconditional alliances than distantly related or unrelated individuals.

Besides forming coalitions for conflict, closely related kin often cooperate to overcome what is known as ‘the dilemma of collective action’ (Olson, 1965). For instance, among the Ye’kwana of Venezuela (Hames, 1987) and the K’ekchi of Belize (Berté, 1988), kin help each other with agricultural tasks without requiring a direct return for the given help from their relative. By contrast, help from unrelated individuals appears to be based on direct reciprocity. Among the Inuit, cooperative hunting groups are formed based on relatedness (Morgan, 1979; but see Alvard, 2003). For traditional societies, it has also been shown that closely related kin often share meat as well as other food items (Cashdan, 1985; Betzig & Turke, 1986; Gurven et al., 2000; Gurven, 2004). Similarly, an analysis of worker remittances showed that closely related kin are more likely to receive money than distantly related kin (Bowles & Posel, 2005).

Kinship cues, sharing the same last name, have also been shown to facilitate altruism (Oates & Wilson, 2002). In addition, individuals appear more willing to incur significant ‘costs’ for closely related individuals than for distantly related or unrelated individuals (Burnstein et al., 1994; Barrett et al., 2002: 49-ff). Incurring costs to aid closely related kin, rather than distantly related or unrelated individuals, is also more likely to be deemed rational and ethical (Kruger, 2001).

Kin also make up a significant proportion of one’s social network (Dunbar & Spoors, 1995; Hill & Dunbar, 2003). Furthermore, in traditional societies closer related individuals been shown to interact more often with each other than distantly related kin (Hames, 1979). In (radicalized) modern societies, genetic relatedness is a significant predictor of subjective closeness and social support (Neyer & Lang, 2003). The important role of kin for social support, e.g. Miller and Darlington (2002) and Kana'Iaupuni and colleagues (2005), could lead to significant inclusive benefits, even in modern societies.[4]

 The presence and help of related individuals can even be crucial for survival. For instance, McCullough and Barton (1991) found that relatedness affected survivorship of the Mayflower disaster. In a similar way did an analysis of the Donner Party disaster show that relatedness influenced survival chances and longevity of individuals (Grayson, 1993 in Barrett et al., 2002: 64-65). For traditional societies, it has been shown that the presence of (matrilineal) kin, especially grandmothers, lowers the risk of infant mortality (Hawkes et al., 1997; Sear et al., 2000/2002 but see Adams et al., 2002). The role of kin for survival of children was also clear in the analysis of the Mayflower disaster (McCullough & Barton, 1991). If children had one parent who survived the winter at the Plymouth colony, they (all) survived.

In modern societies, genetic relatedness has been shown to affect the occurrence of child abuse. Rates of child abuse, and fatal child abuse, are significantly lower in households where both parents are related to the child (Daly & Wilson, 1981/1985/1988). Moreover, genetic relatedness influences the occurrence of murder. In modern societies, the occurrence of murder between (closely) related individuals appears to be relatively low (Daly & Wilson, 1982/1988).

(Female) Kin also often provide childcare and their help can increase an individual’s inclusive fitness (Turke, 1988/1989; Bereczkei, 1998; Sear et al., 2000/2002). Such ‘helper’ behaviour between related individuals has been shown in several non-human species (Kurland, 1980; Emlen, 1995). In humans, kin can also serve as ‘helper’s at the nest’ and individuals of low reproductive value, e.g. grandmothers, often help kin (Hawkes et al., 1997; Sear et al. 2000). The presence of (matrilineal) kin has been shown to have significant beneficial effects on infant survival and (inclusive) fitness (Hawkes et al., 1997; Sear et al., 2000/2002 but see Adams et al., 2002). In traditional societies, older siblings also often fulfil a ‘helper role’ (Weisner & Gallimore, 1977; but see Crognier et al., 2001). Older siblings can take care of younger siblings, while the mother is occupied with domestic labour and/or provisioning. For instance, among the Toba of Argentina, girl helpers were shown to significantly reduce the workload of the mother by helping and caretaking (Bove et al., 2002; but see Hames & Draper, 2004). Caretaking by elder siblings could have beneficial effects on younger siblings’ their development (Sigman et al., 1988; see Belsky et al., 1991). In modern societies, childless individuals could fulfil helper roles, as suggested by Essock-Vitale and McGuire (1985).

The absence of helping kin has been cited as a critical factor for the demographic transition by Turke (1989). Turke (1989) argued that the emergence of the nuclear family and the breakdown of extended kin networks as key factors for the demographic transition. These processes have shifted the burden of children to increasingly smaller families. Therefore, in the face of increasing costs and low benefits, individuals nowadays choose to have less children or have no children at all (Morgan & King, 2001; van de Kaa, 2004). As we have discussed above, individuals who do not have children can allocate the time and money, which they would otherwise invest in children, differently.

Although, helping behaviour between kin is common, as described above, several studies have shown that evolution can promote intense competition between kin as well, especially sibling rivalry (Mock & Parker, 1997/1998). Sibling rivalry occurs in several animal taxa (birds: O’Connor, 1978; insects: Grbic et al., 1992; mammals: Frank et al., 1991; Cockburn, 1994) and even in plants (Shaanker et al., 1988).

For humans, it has been shown that if births are closely spaced, survival rates tend to decline for children (Hobcraft et al., 1985; Curtis et al., 1993; Alam, 1995). This could be attributed to sibling rivalry for parental investment. For example, Muhuri and Preston (1991) showed that girls in Bangladesh have a higher mortality risk, if they have an elder sister. Boys were found to have a higher mortality risk, if they have two older brothers (Muhuri & Menken, 1997). There was also an increase in mortality risk, indifferent of sex, if the first born was a son. Apart from affecting infant mortality, sibling rivalry has been argued to have significant effects on psychological development (Sulloway, 1996/2001).

Besides the occurrence of strong competition between kin, it should also be taken into account that decisions on whether or not to support kin are contingent upon several other factors. As with all individual decision making in a social context (Parsons, 1937; Coleman, 1990), an individual’s decision to ‘invest’ in kin relations depends upon a wide variety of factors. For instance, Wang (1996) found that besides genetic relatedness, other factors, e.g. age, influenced risk-sensitive decision making. In his experiment subjects were asked to choose between a deterministic option and a probabilistic option to ‘save’ related individuals. Besides relatedness, decision making was found to be influenced by age, gender and social context. Likewise, Dunbar, Clark and Hurst (1995), in their study of Viking sagas, found that besides relatedness, other factors, such as perceived dangerousness, influenced decisions whether or not to avenge a murder of a related individual or settle for blood money.

Of special importance for kin-based decision making is the factor ‘reproductive value’ (Hughes, 1988). Reproductive value refers to the probable number of offspring an individual will have during the rest of his or her lifetime (Low, 2000: 64; Barrett et al., 2002: 52). This concept, first developed by Fisher (1958 [1930]), has been utilized for a wide variety of demographical issues, e.g. Keyfitz (1977), but also has implications for the analysis of human kin-based decision making. Wang’s (1996) experiment shows that respondents take into account reproductive value while making risk-sensitive decisions. If the reproductive value of the related individuals was low, respondents were more likely to choose the probabilistic option over the deterministic option. If the reproductive value of respondents was high, respondents were much more likely to choose a deterministic option. Respondents were thus more likely to ‘gamble’ with decisions affecting individuals with low reproductive value than with individuals who have high reproductive value. Bowles and Posel (2005) also found that reproductive value influenced kin-based decision making. Relatedness weighted for reproductive value, explained more of the variance in migrant workers’ remittances to their family than genetic relatedness in itself. When deciding how to allocate their remittances migrant workers take into account reproductive value, as well as other factors, e.g. economic need, besides genetic relatedness.

The literature on kin selection and human decision making readily shows that, under circumstances, individuals help their (closely related) kin, leading to a higher inclusive fitness. In our opinion, childless individuals also make this type of adaptive decisions, at least to a certain extent. Childless individuals can choose to invest their time and money in relations with their friends, partner(s) or kin. As they do not ‘invest’ in offspring they are able to allocate their time and money differently, towards their relationship(s) with their partner, kin and friends. As already discussed, based on individualization theory, there is no apparent reason why childless individuals should choose to maintain a stronger relationship with their siblings than parents.

However, from an inclusive fitness point of view, childlessness, whether by choice or not, should affect the relationship with an individual’s kin. As childless individuals do not have to trade off investments between their relationship with kin and with their offspring, they are predicted to have ‘better’ relationships with their kin than ‘parents’, all else being equal. Such stronger relationships could have inclusive fitness consequences. As we will discuss in the following section, high quality relationships are likely to go together with investments.

There is some evidence that childless individuals invest more in relationships with their kin. Essock-Vitale & McGuire (1985) found that childless women were significantly more likely than ‘parents’ to give help to their nieces and nephews. Yet, childless women were also found more likely to receive help from their nieces and nephews. Nelson (1998) found that childless women were more likely than parents to support nieces and nephews and siblings. However, the effects of childlessness on sibling relationships have not been thoroughly studied from an evolutionary perspective, while controlling for other factors that influence relationship characteristics.

Besides childlessness, relatedness is of importance for this study. Closer related individuals should generally have stronger relationships than distantly related or unrelated individuals. Therefore relationships between full siblings (r=0,5) are predicted to be stronger than between half-siblings (r= 0,25) or relationships where a sibling is adopted (r= 0). Moreover, relatedness is predicted to influence social and financial investment in sibling relationships. Relationships between closer related individuals are predicted to show more investment than those of distantly related individuals, all else being equal.

4. Social factors and quality of relationship

‘Birds of a feather flock together’

As our analyses mainly focus on quality of social relationships, we will briefly discuss social factors affecting quality of relationship using insights from (micro)sociology.

According to the exchange paradigm (Homans, 1951/1958; Blau, 1964; for a review: Cook & Whitmeyer, 1992), social relationships can be seen as forms of immaterial and/or material exchange. Gift or money relationships will, to a certain extent, reflect the strength of a dyadic relationship. Symmetrical money or gift-giving relationships, rather than asymmetrical relationships, for instance, are likely to be a reflection of a high quality relationship. Similarly, relationships in which individuals show equal social investment, e.g. contacting each other, are predicted to be of a higher quality. In general, ‘symmetrical’ relationships are predicted to be more stable and of a higher quality. Power asymmetries are likely to be unstable and lead to a negative evaluation of a dyadic relationship (Blau, 1964 but see Molm, 1997).

Structural sociological theory predicts that social relations between individuals are structured by the characteristics they share, also referred to as ‘homophily’ (Blau, 1977/1980; McPherson et al., 2001). Within a certain dyad, the degree of ‘homophily’ is a strong predictor of the strength and quality of a dyadic relationship. Therefore, for example, if age differences between siblings are large their relationship is expected to be of a poorer quality.

Research on social support has shown that women rely more on their relatives than men for emotional and social support (Turner & Marino, 1994; Phillipson, 1997). In contrast, men tend to primarily rely on their partner for social and emotional support (Turner & Turner, 1999; Mirowsky & Ross, 2002). Therefore, we expect that women have better relationships with their siblings than men.

From a social network perspective (McPherson et al., 2001), it is important to bear in mind that high quality social relationships, strong ties, often require ‘high (mutual) investments’ in terms of time and money.[5] Humans, like other primates, appear inherently limited in the number of these strong, high quality relationships they have (Dunbar, 2004 [1996]). These high quality relationships are part of a vertically structured network, with distinct groupings, e.g. ‘support cliques’ and ‘sympathy groups’ (Hill & Dunbar, 2003; Zhou et al., 2005).

Support cliques can be described as the set of individuals with whom one maintains ‘strong’ relationships as well as turns to for advice or support while under financial or emotional distress (Dunbar & Spoors, 1995; Hill & Dunbar, 2003). Support cliques generally consist of three to five individuals. The support clique size appears to be fairly constant across cultures and has been replicated in several studies (Marsden, 1987; Dunbar & Spoors, 1995; Dunbar, 1998). Whereas support cliques seem to typically consist of about three to five individuals, (human) sympathy groups consist of approximately twelve individuals (Buys & Larsen, 1979; Dunbar, 1993/1998/2004; Dunbar & Spoors, 1995; Hill & Dunbar, 2003). These individuals are typically contacted once a month. Based on extrapolation of primate data on group size and relative neocortex size, the overall size of an individual’s social network is predicted to be limited to approximately 150 individuals (Dunbar, 1993). Cross-cultural evidence, sociological and organizational studies, e.g. Coon (1946), support this prediction (Dunbar, 1993/2004).

 Given the findings of a fairly constant size of support cliques and human sympathy groups, the number of high quality relationships is inherently limited. This implies that the relationship quality of one relationship is traded off against other relationships. As childless individuals do not have to trade off high quality relationships with relationships with their children, they are predicted to have higher quality relationships with their siblings than parents.

5. Research questions and aims of the study

The main research question of this project is whether or not childless individuals maintain stronger relationships with their sibling than ‘parents’ (7.1). Genetic relatedness is also expected to significantly influence relationship quality. If we would find differences according to these variables this leads to suspect that kin selection might play a role, contrary to the claim that family relationships have become centralized (solely) upon ‘individualized choice’. Furthermore, we will investigate whether these high quality relationships require ‘symmetry’ and ‘investment’. By investment we refer to investment in terms of time (showing interest, showing initiative to contact the other) and money (donating or lending).

This study will investigate the effects of childlessness and genetic relatedness on relationship quality with a sibling whilst controlling for ‘investments’ as well as other aspects influencing the relationship, e.g. differences in age between the siblings. Therefore we will follow a multivariate design, which allows examining the independent effects of genetic relatedness and childlessness on relationship quality while controlling for other variables.

The effects of genetic relatedness and childlessness on other relationship aspects besides quality, i.e. ‘investments’, will then be investigated separately (7.4). For instance, childless individuals are expected to show more interest in the personal life of their sibling, but not necessary receive more interest from their sibling, than parents.

The relationship aspects which we will analyze are giving and receiving interest (7.4.1 to 7.4.3), initiative of contact (7.4.4), financial balance (7.4.5) and giving and receiving financial help (7.4.6 to 7.4.8). The effects of childlessness and genetic relatedness will be examined on these relationship aspects while controlling for the other relationships aspects. Childless individuals are predicted to invest more than parents, although they could also have symmerical relations, as symmetry is characteristic for high quality relationships. A higher degree of relatedness is predicted to lead to more investment in sibling relationships.[6]

In addition, we will investigate differences between the voluntary and the involuntary childless (7.5). This allows to examine the claim by individualization theory that the higher educated are likely to be childless by choice. Of interest is also whether involuntary childless individuals maintain stronger relationships with their sibling than ‘childfree’ individuals. Also, we will look at differences between involuntary childless and childfree individuals in terms of ‘investment’. Involuntary childless individuals are predicted to invest more in their sibling relationship than ‘childfree individuals’. If differences between the ‘childfree’ and the involuntary childless in investment or relationship quality occur, this indicates that ‘choice of childlessness’ plays a role for sibling relationships.

In conclusion, first we will investigate the effects of childlessness and genetic relatedness on the quality of the relationship, while attempting to control for other factors influencing the sibling relationship. Then we will examine whether childlessness and genetic relatedness affect other dimensions of the relationship, e.g. whether or not money or valuables are given. Finally, we will compare the childfree and involuntary childless individuals for relationship quality and other dimensions of sibling relationships.[7]

6. NKPS-Dataset and methodology

6.1 The NKPS dataset and the selected variables

The first wave of the Netherlands Kinship Panel Study (NKPS) dataset was obtained through the Netherlands Interdisciplinary Demographic Institute (NIDI). The NKPS is a large scale longitudinal study, designed to investigate family and kin relations in the Netherlands (Dykstra et al., 2004). The first wave was completed mid-2004; we used this version (version of 4-7-04; main sample). The sampling procedure, representativity as well as the survey method is described in detail by Dykstra and colleagues (2004).

In order to ensure we correctly coded childlessness, we selected all individuals of forty years or older. We assume that forty years is more or less the age at which the reproductive phase is completed. Individuals were then coded as ‘childless’, if they did not have any children by any means (including adoption), and stated that they had no intentions of having children in the future.[8] As we selected only individuals over forty, all analyses presented here are unweighted for the global Dutch population structure. Several variables were recoded, the procedure of recoding can be found on the cd-rom. In the NKPS-survey questions were asked about ‘sibling a’, a randomly selected sibling of the respondent. Correspondent characteristics, e.g. gender of sibling a, were matched, as described on the cd-rom.

The variables selected for analyses are presented in the table below (Table 1). The correspondent questions and additional information can be found in the NKPS codebook (Dykstra et al., 2004), except for the constructed variables (age difference, sibling a childless; education; sex sibling a; total living sibs; genetic status sibling a) for which additional information, can be found on the cd-rom. Table 2 contains additional information on the questions used for certain variables.
 

Table 1 : Variables and categories

Table 2 : Formulation of questions for certain variables (Dykstra et al., 2004)

(Source: Dykstra et al. 2004)

6.2 Preliminary analyses (HOMALS)

Before we examine the effects of childlessness and genetic relatedness on quality of relationship in a multinomial logistic design, we present descriptive statistics and some exploratory analyses.

The relationship between genetic status and quality of relationship is significant (χ² = 120,24; df= 6; p= 1,45*10^-23; Figure 1).[11] As there are only a small number of cases where sibling a was adopted, we merged the categories ‘half-sibling’ and ‘adopted sibling’ (χ ² = 119,1; df= 3; p= 1,21*10^-25; see cd-rom).

Figure 1 : Genetic relatedness and quality of relationship.

In order to visualize the relationship between aspects of social relationships and genetic status, we performed a homogeneity analysis by alternating least squares (HOMALS). This technique, also known as multiple correspondence analysis, allows the descriptive analysis as well as the visual presentation of several variables (see Clausen, 1998). It can be described as a form of principal component analysis for nominal data. No assumptions are made about the distribution of the data. For a further description of the technique we refer to Michailidis and de Leeuw (1998) or the SPSS manual (SPSS, 2001). The common representation of HOMALS is by (just) a two-dimensional graph.[12] Points lying close to each other indicate ‘similarity’, whereas points lying far away from each other indicate categories that discriminate. HOMALS is used here merely for exploring and describing relationships between certain variables, which will then be explored by multinomial logistic regression. More preliminary analyses by HOMALS can be found on the cd-rom.

For the first HOMALS we included interest given to sibling a; interest received from sibling a; genetic status sibling a and quality of relation with sibling a.

Figure 2 : Homogeneity analysis by alternating least squares for genetic status,
 interest received and given and quality of a sibling relationship

The first dimension distinguishes between a ‘not great’ and a ‘very good relationship’ and between ‘own siblings’ and ‘adopted or half-siblings’.[13] The second dimension distinguishes between ‘good’ and ‘very good’ relationship, and between ‘own siblings’ and ‘adopted or half-siblings’. The HOMALS indicates that ‘interest given and received’ and quality of relationship correspond quite well. Furthermore, respondents appear much more likely to have a good relationship, in which interest is given and received, with their full sibling than with an adopted or half-sibling. On the contrary, if their sibling is a half-sibling or adopted sibling, respondents appear to have a ‘not great’ relationship with their sibling.

If we replace genetic status of sibling a with childlessness in HOMALS, the following graph shows (Figure 3). The first dimension appears to discriminate between ‘very good’ and ‘not great’ relationships. The second dimension appears to discriminate according to ‘childlessness’ and between ‘good’ and ‘very good relationships’. When examining the second dimension, there is some indication that childless individuals are more likely than ‘parents’ to have a very good instead of a good relationship. The very good relationship is characterized by (strong) ‘mutual interest’.

Figure 3 : Homogeneity analysis by alternating least squares for childlessness,
interest received and given and quality of a sibling relationship

6.3 Multinomial logistic regression (MLR)

In this section we briefly discuss (multinomial) logistic regression (MLR). Logistic regression is a technique that allows hypothesis testing. The technique is described by Hosmer and Lemeshow (1989), Menard (1995) and Pampel (2000). Logistic regression is relatively free of assumptions and statistically robust. It assumes that there is a linear relation between the independent variable and the logit. If an independent variable has an independent significant effect on the (logit of the) dependent variable, it will be selected in the iterative procedure by which the model is constructed. Whereas estimates in an ordinary least squares (OLS) regression seek to minimize the sum of squared ‘distances’ to the regression line, the parameters in the MLR-model are estimated by maximum likelihood. There are several methods for parameter selection procedure. The method used here is ‘forward stepwise’, yet estimates for correspondent ‘backward stepwise’ models can be found on the cd-rom. In general, the parameter selection procedure did not affect the models presented. In cases where differences between models do occur due to parameter selection procedure, these were only very small differences in terms of model fit and explained variance.

Similar to the equation under multivariate (OLS) regression a multinomial logistic regression equation can be written:

Ln[P(Y=k_m/k_n)/1-P(Y=k_m/k_n)]= a + Σ_i λ_i*X_i

With Ln[P(Y=k_m/k_n)/1-P(Y=k_m/k_n)]= the logit for P(Y=k_m/k_n) (the probability that Y is in category m (versus reference category n)); a is the intercept; X_i= independent variable and λ_i= correspondent parameter for X_i .

Also, similar to OLS regression, a (pseudo-)R² can be calculated. There is a variety of R² measures for logistic regression; we will report Nagelkerke’s (1991) R², as it is most commonly used. For a more in depth discussion of logistic regression we refer to the references cited above.

7. Results from analyses of NKPS

7.1 MLR (main effects) for quality of relationship

The dependent variable for the analysis is ‘quality of relationship with sibling a’.

The (proposed) independent variables and predicted relationships are presented in the table below (Table 3). Most of the predictions can be derived from the theories presented above (2-4). Education is expected to lead to a worse relationship, as higher educated individuals are more ‘individualized’ than lower educated individuals. As the number of (full) living siblings increases the relationship with a random ‘sibling a’ is expected to be worse. Individuals with more siblings have to maintain more relationships. Therefore, it is likely that the quality of a relationship with a randomly assigned sibling is lower.

Although many variables proposed for the model are ‘not biological’, they are necessary for the construction of a model for quality of a relationship with a sibling. Furthermore, we are interested in the effects of ‘biological variables’ while controlling for social variables affecting the quality of a relationship.

Table 3 : Independent variables for the multinomial logistic model

As the second dimension in the second HOMALS indicates that the difference between the childless and individuals with children lies between having a good or a very good relationship with sibling a, the reference category for the MLR is set as ‘good’ (Figure 3).

The independent variables presented above were entered in a forward stepwise multinomial logistic regression with as dependent variable quality of relationship. The variables that were selected are presented in the table below (Table 4). The (independent effects of the) variables sex sibling a, sibling a childless, genetic status, age difference did not significantly improve the model (at α= 0,05).

Table 4 : Variables selected for the multinomial logistic model (Step summary)

 Method: Forward stepwise
 

The final model has a -2LL of 8165,18 (χ²= 2145,87; df= 69; p<0,001). It has a Nagelkerke R² of 0,43. This is (very) good according to standards in the social sciences. The likelihood ratio tests for the variables are presented in table 5. The χ² tests refer to the difference between -2LL of the final model and the reduced model (Table 5). Reduced model refers to the model without the given variable.

Table 5 : Likelihood ratio tests for the variables in the model

There appears to be no indication that the parameter selection procedure, forward stepwise, affected the chosen variables.[14] The parameter estimates for each category are presented in separate tables, although they are part of the same analysis. It is important to bear in mind that the comparison of interest is ‘very good’ versus ‘good’ (see: Figure 3). Therefore, we will not (extensively) discuss the other comparisons. The parameter estimates for having a ‘not great’ relationship (versus a good relationship) are presented in table 6. The categories with λ equal to zero are reference categories. Dichotomous variables are treated as interval variables (see: Table 1). The Wald statistic in the tables follows the χ² distribution and allows determining whether or not an individual parameter is statistically significant (Pampel, 2000).

Table 6 : Parameters for ‘not great’ (versus ‘good’) relationship with sibling

Parameters are most easily interpreted in term of odds ratios (Exp(λ); e^λ) (Pampel, 2000). For example, women are 1,47 times more likely than men to state that their relationship with their sibling is not great instead of good. This while controlling for the other variables in the model. Another example, individuals who did not receive any interest in their personal life at all are 29,31 times more likely than individuals who received interest several times to have a ‘not great’ relationship instead of a good relationship, while controlling for the other variables.

Perhaps the most interesting, yet intuitive, finding is that a lack of interest (given and received) and the presence conflicts are very strong predictors of having a ‘not great relationship’.

The parameters for a reasonable versus a good relationship (Table 7) and for a very good relationship versus a good relationship (Table 8) are presented below.

Table 7 : Parameters for ‘reasonable’ (versus ‘good’) relationship with sibling

Table 8 : Parameters for ‘very good’ (versus ‘good’) relationship with sibling

7.2 Discussion of MLR for relationship quality

7.2.1 Variable selection

As expected most variables were selected as predictors for ‘quality of relationship with sibling a’. Social factors, e.g. interest, were highly significant predictors of relationship quality. However, genetic status, age difference, sibling a childless and sex sibling a were not found to be significant (independent) predictors of relationship quality.

Especially, for genetic status, this is puzzling. A possible explanation for the non-selection of genetic status as predictor is that other variables in the model are far stronger predictors of quality of the relationship. For instance, ‘interest received’ and ‘interest given’ are very strong predictors. However, if genetic status is a significant predictor of ‘interest’ variables, it is possible that it influences the quality of a relationship in an indirect way.[15] As can be seen in the figure below, genetic status does significantly relate to ‘interest received’ (χ²= 13,81; df= 2; p= 0,001) but not to ‘interest given’ (χ²= 5,60; df= 2; p= 0,061).[16] Respondents are significantly more likely to receive interest from a full sibling than from an adopted or half-sibling.[17] Under 7.4.1 and 7.4.2, we will further analyze the effect of genetic relatedness on interest given and interest received.

Genetic status also varies according to attained level of education (χ²= 24,14; df= 7; p=0,001; Table 9). In addition, ‘genetic status of sibling a’ obviously relates to ‘total full living siblings’ (r= 0,183; p= 1,02*10^-36; n = 4710). Effects of genetic status on these variables could explain why it is not selected as predictor for quality of relationship with sibling a. Similar analyses for genetic status and other parameters in our model, such as ‘initiative of contact’ of ‘giving financial help’, were not significant and can be found on the cd-rom (α = 0,05).

Figure 4 : Genetic relatedness and interest [18]

Table 9 : Frequency table for education by genetic status

Omitting ‘interest received’, ‘total living own siblings’ and ‘education’, leads to a model with a Nagelkerke R² of 0,35 (final model: -2LL= 8169,62; χ²= 1644,82; df= 39; p<0,001). This is still (very) good according to standards in the social sciences. It is important to note that we assigned ‘not great’ as the reference category (see: Figure 1). In this model, genetic status is a significant predictor at five percent significance level (Table 10).[19] Here we present only the parameters for the comparison between ‘not great’ and ‘good’ and between ‘not great’ and ‘very good’. The parameter estimates for ‘not great’ versus ‘reasonable’ can be found on the cd-rom.

Table 10 : Step summary for model with genetic status

 Method: Forward stepwise
 

Table 11 : Likelihood ratio tests for model with genetic interest

If the respondent’s sibling is a full sibling, he or she is approximately two and half times more likely to have a good, instead of a ‘not great’ relationship with them, than when the sibling is an adopted or half-sibling (Table 12). When comparing a very good versus a not great relationship, full siblings are 2,75 more likely than not fully related siblings to have a very good relationship versus a not great relationship (Table 13). Both effects of genetic relatedness are found while controlling for other variables in the model.[20] However, if we control for ‘interest received’, we find no significant (independent) effect of genetic relatedness on relationship quality (see cd-rom).

Table 12 : Parameter estimates for ‘good’ versus ‘not great’ relationship

Table 13 : Parameter estimates for ‘very good’ versus ‘not great’ relationship

The direct and indirect effects of genetic status on relationship quality could be further explored by logistic path models (Eshima et al., 2001) or structural equation modelling (Loehlin, 2004; see cd-rom for example).

Similar to genetic status, childlessness of the sibling does not (independently) appear to affect the quality of the relationship (in our initial model). However, we should take into account that the question about relationship quality was asked to the anchor and not to the sibling.[21] Sex of the sibling did not (independently) affect the quality of the relationship, contrary to the expectation that women would have better relationships with our kin. Yet, this could be due to the fact that the question was asked to the anchor and not to the sibling. Age difference between the siblings did not (independently) affect the quality of the relationship. This could be due to the selection of respondents aged over forty; (relative) age difference might not be that relevant as individuals grow older. Moreover, we should bear in mind that age difference, the difference between birth years, is not a completely accurate representation of age difference between the siblings.

7.2.2 Comparison good versus very good relationship

In general, the direction of the variables that were selected for the initial model was in the expected direction (Table 14). Although, some parameter estimates were not significant in the comparison of a good versus very good relationship, the direction of the parameter estimates in the comparisons of ‘not great’ or ‘reasonable’ versus ‘good’ was generally consistent with the predicted relationships for most variables. A model with reference category set as ‘not great’, which can be found on the cd-rom, more clearly shows that (nearly all) the parameter estimates are in the predicted direction.

‘Giving financial help’ and ‘giving and receiving interest’ are (highly) significant predictors of having a very good relationship (Table 8). This supports the idea that high quality relationships go together with (mutual) investments in terms of time and money. As predicted childless individuals were also significantly, nearly one and a half time, more likely than parents to have a very good relationship, instead of a good relationship, while controlling for other variables (Table 8).

Table 14 : Results compared with predictions for comparison of ‘good’ versus ‘very good’ relationship.

7.3 MLR with interaction effects for relationship quality

Although, the base model with just the main effects is perfectly valid (see, Peng et al., 2002), the model could be further improved by incorporating interaction effects. We explored all two-way interaction effects.[22] Only the interaction effects between sex of the respondent and financial help received and between age and financial help received were found significant predictors of relationship quality (α = 0,05). The model including interaction terms is only marginally better than the previous model (Nagelkerke R²= 0,44; -2LL= 8144,48; χ²= 2166,57; df=78; p<0,001). The statistics and parameters estimates for this model can therefore be found on the cd-rom.

Interestingly, none of the interaction effects between childlessness and other independent variables were found significant (α = 0,05). There were also no significant interaction effects between genetic relatedness and other variables (α = 0,05).^[23] Also of interest is that there is no interaction effect for interest given and interest received or for financial help given and received.

Moreover, there was no significant interaction effect for ‘childlessness’ and ‘sibling a childless’. We would expect that the relationship would be better if the anchor is childless but the sibling has children. It is possible that childless individuals are closer to their sibling regardless of whether they have children are not. Also, the absence of an interaction effect could be due to the fact that the respondents are over forty years, therefore their nieces and nephews are probably self-sufficient. Although, an analysis for childless individuals appears to indicate that the relationship is in fact better if the sibling has children (Somers’ D_yx = -0,071; one-tailed p= 0,051; n= 818; Figure 5).

Figure 5 : Sibling childlessness and quality of relationship for childless respondents

7.4 MLR for ‘investment measures’ of relationships

Although we have shown that childless individuals have ‘better’ relationships with their siblings than parents, and that these ‘better’ relationships go together with some ‘investments’ (giving interest and giving financial help), we have not yet shown that childless individuals actually ‘invest’ in these relationships.

For this section, we are mainly interested in how childlessness and genetic relatedness influences ‘investment measures’ while controlling for other relation aspects. Therefore we will not report and discuss all models in detail. All the models and their statistics can be found in full on the cd-rom. ‘Forward stepwise’ was used as method for parameter selection. The parameter selection procedure generally did not affect the selected variables. However in some cases, the backwards stepwise procedure leads to (marginally) different results, which can be found on the cd-rom. As predictor variables we select the other previously used variables, excluding ‘quality of relationship’ (Table 1). These variables are included in our models, as we want to investigate the (independent) effects of childlessness and genetic relatedness, while controlling for other (social) factors influencing the sibling relationship. For these models, we will only investigate the main effects.

By ‘investment’ measures, we refer to ‘giving interest’, ‘giving financial help’, ‘financial balance’ (giving more than receiving) and ‘initiative of contact’ (showing initiative). Although the cost of such investments is not always large, these variables do represent a cost in time or money and an opportunity cost, i.e. investment in other relationships. Childless individuals are predicted to ‘invest’ more than ‘parents’. If siblings are closer related they are also predicted to invest more.

7.4.1 Giving interest

The model shows no significant (independent) effect for childlessness and genetic status (Table 15). There were also no significant (independent) effects for sibling a childless, sex sibling a, age difference, age, financial help given/received and initiative of contact. The model has a Nagelkerke R² of 0,48 (-2LL= 6892,94; χ²= 2400,55; df=30; p<0,001). The parameter estimates as well as the step summary for this model can be found on the cd-rom.

Table 15 : Likelihood ratio tests for model for ‘interest given’

However, if we omit ‘interest received’, which is a very strong predictor[24] of ‘interest given’, from the model we do find effects for childlessness and genetic relatedness (Table 16). This alternative model has a Nagelkerke R² of 0,11, which is reasonable for the social sciences (-2LL= 8850,78; χ²= 442,71; df=38; p<0,001). Age, financial balance and childlessness of sibling a were not significant (independent) predictors of ‘giving interest’.

Table 16 : Likelihood ratio tests for model for ‘interest given’ (omission of interest received)

Childless individuals are 1,36 times more likely than ‘parents’ to have given interest to their sibling several times over the past three months, instead of not at all (Table 17). This while controlling for the other variables in the model. Respondents are 2,04 times more likely to have given interest several times, instead of not at all, to a full sibling than to an adopted or half-sibling and this while controlling for the other variables in the model. The parameter estimates for the comparison between interest given ‘once or twice’ versus ‘not at all’ can be found on the cd-rom.

Table 17 : Parameter estimates for ‘interest given several times’ (versus not at all)

Yet, if we control for interest received, we find no significant (independent) effects for childlessness and genetic relatedness.

7.4.2 Receiving interest

The model has a Nagelkerke R² of 0,51 (-2LL= 6787,49; χ²= 2557,36; df=28; p<0,001). Childlessness, education, age, sex and financial help received were not significant predictors of receiving interest (Table 18). Childlessness does not appear to independently affect the personal interest received from a sibling.

Table 18 : Likelihood ratio tests for model for ‘interest received’