Preferences for a sex-mix composition of offspring: A multigenerational approach

Abstract

Parents with two boys or two girls are more likely to have a third child than those with a sex-mix. However, little is known of whether sex-mix preferences extend beyond the nuclear family. This study leverages the random variation of sex at birth to assess whether the sex of nieces and nephews, in combination with own children, matters for fertility choices. Using three-generational data in the US Panel Study of Income Dynamics (PSID), I show that extended families including grandparents, their children, and their grandchildren are collectively more likely to have more than two grandchildren when lacking sex-mix, both when the first two grandchildren are siblings or cousins. I explore the pathways of these offspring sex preferences finding support for a preference for an uninterrupted line of male descendants. This multigenerational approach also contributes a new estimation strategy to causally estimate the effects of family sizes on outcomes beyond fertility.

Introduction

The preference for having at least one boy and one girl is well-documented in American nuclear families. This preference for a sex-mix among offspring is inferred from the higher observed propensity to have a third child if the first two are of the same sex (Ben-Porath and Welch 1976; Angrist and Evans 1998; Lundberg 2005). It is, however, unclear whether the manifestation of this preference pertains exclusively to the childbearing couple or is part of a wider web of preferences and fertility behaviours within the extended family comprised of grandparents, their adult children, and their grandchildren.

Other family members’ fertility behaviours may increase in saliency as fertility decreases and third births become increasingly rare within the same household (De Boca et al. 2005). Moreover, several researchers have found that family members share preferences and fertility behaviours in addition to shared characteristics due to genetic endowments, family culture, and socio-economic environment, thus making the extended family a reasonable place for the manifestation of offspring sex-mix preferences (Axinn et al. 1994; Barber 2000; Bernardi 2003; Murphy 2013; Cools and Hart 2017; Dahlberg and Kolk 2018; Beaujouan and Solaz 2019).

This study proposes a novel methodological approach to this issue by leveraging the random variation of sex at birth in grandchildren and the preference for mixed sex offspring (Ben-Porath and Welch 1976; Angrist and Evans 1998; Andersson et al. 2006) to test the hypothesis that sex-mix preferences also manifest within multigenerational families. More specifically, it explores whether there is a causal effect of the sex composition not only of one’s own children (as has been previously established) but also the combination of one’s own children with one’s nieces/nephews. I then explore some possible pathways looking at drivers of nuclear-family sex-mix preferences, such as symbolic capital and cultural values (Nugent 2013), as well as factors relevant to fertility peer effects, such as age similarity, being sisters, and proximity (Kuziemko 2006).

I use three-generational data from the American Panel Study of Income Dynamics (PSID, 1968-2015) to construct extended family links, defined to include grandparents, adult children in the middle generation (siblings), and grandchildren. I then show that if the first two grandchildren are of the same sex, then the overall extended family is more likely to have more than two grandchildren than if the sex-mix is attained within the first two grandchildren. Crucially, this holds true not only for siblings as expected, but also for cousins, showing that fertility decisions are not only impacted by the sex-mix of one’s own offspring but by that of one’s siblings as well.

Finding that sex-mix preferences do not manifest exclusively within a nuclear family contributes to our understanding of their emergence and actualization in fertility behaviour. For example, the limited number of studies conducted to uncover the reasons behind persistent sex-mix preferences point at sociocultural explanations related to symbolic capital and cultural values, also applicable to multigenerational families (Hank and Kohler 2000; Mills and Begall 2010; Nugent 2013). Parents may value interactions with cousins of opposite sex for their offspring’s developmental growth, as well as be affected by social learning differently according to the sex composition of their nieces and nephews (Raley and Bianchi 2006; Lois and Arránz Becker 2013; Baker and Milligan 2016). This study also provides new evidence supporting the importance of transmitting the family name through an intergenerational line of male descendants and not only in the father-son dyad.

Conditional on the availability of three generational data, this new approach of looking at grandchildren sex-mix, rather than at natural variations within the nuclear family, also contributes a methodology that opens new avenues of research beyond family demography. Indeed, it can identify effects beyond the marginal effect of the third child and is less subject to threats to the exclusion restriction when used in an instrumental variable framework for other outcomes of interest, such as early retirement and labour force participation.

Background

Sex preferences for offspring

Extensive demographic research documents a preference for offspring sex-mix in the United States and Europe. The existence of a preference for at least a son and a daughter relies on the empirical evidence that parents with two boys or two girls are more likely to have a third child than are parents with a boy and a girl (Ben-Porath and Welch 1976; Angrist and Evans 1998; Hank and Kohler 2000; Conley and Glauber 2006; Mills and Begall 2010). This continues to be documented regardless of whether boys are favoured over girls or a girl preference has emerged: when there are two children in a family, a sex-mix is always preferred (Lundberg 2005; Andersson et al. 2006; Dahl and Moretti 2008). Moreover, the strength of the sex-mix preference has not diminished over time, despite a weakening in the 1980s (Pollard and Morgan 2002; Tian and Morgan 2015).

The continuation of childbearing to achieve a preferred sex composition of offspring is often described as the pursuit of a ‘balanced family’ within a nuclear household (Lundberg 2005; Nugent 2013; Larsen Gibby and Thomas 2019). This characterization is a pervasive narrative, from online forums to American fertility clinics marketing their sex selection services (Nugent 2013). However, there is no a priori reason why balancing the sex composition of a family needs to occur solely within a nuclear family. Indeed, the emergence of low fertility reduces the proportion of women in a population with at least two children and transitions to third births are becoming increasingly rare (Del Boca et al. 2005). This puts limitations on the actualization of sex-mix preferences within a nuclear family, as highlighted by the fact that using sibling sex composition as an instrumental variable for fertility replicates better in Mexico and Argentina than in a low fertility context like Greece (Angrist and Evans 1998; Cruces and Galiani 2007; Daouli et al. 2009). Realizing sex-mix preferences in the extended family rather than in a nuclear family would afford greater flexibility in how to achieve a balanced family within the larger pool of grandchildren.

The manifestation of a sex-mix preference across multiple nuclear families within the same extended family presupposes that family members can and do influence each other’s fertility behaviours. This would be in line with previous studies showing that fertility decisions are influenced not only by individual characteristics, but also by the fertility behaviours of other individuals in their sphere of social influence (Coale and Watkins 1986; Bernardi and Klärner 2014). Family members share genetic, environmental, socio-economic, and cultural traits and have higher social proximity, length, and depth of interactions. These factors make the extended family, comprised of grandparents, adult siblings, and grandchildren, a possible locus of the manifestation of family influences on fertility driven by sex-mix preferences (Axinn et al. 1994; Barber 2000; Kuziemko 2006; Lyngstad and Prskawetz 2010; Kotte and Ludwig 2011; Aassve et al. 2012; Buyukkececi et al. 2020).

Extended family fertility influences

Family members are densely connected in a complex web of influences that may facilitate the realization of sex-mix preferences, keeping in mind the overall sex distribution within the family. Participants in qualitative studies implicitly or explicitly acknowledge that their decisions to form a family and have children reflect what other members of the extended family expect and do (Bernardi 2003; Keim et al. 2013). Some recent quantitative studies also lend support to the relevance of more than one source in family influences on fertility, indicating complex extended family influences including preferences and behaviour from prospective parents’ parents, adult siblings, nieces and nephews (Balbo and Mills 2011; Kotte and Ludwig 2011; Aassve et al. 2012; Dahlberg and Kolk 2018; Buyukkececi et al. 2020). However, only a limited number of studies use plausibly causal instrumental variable approaches based on sex-mix composition (Cools and Hart 2017; Hart and Cools 2019; Buyukkececi et al. 2020).

Against the backdrop of multifaceted family influences, the two most studied sources are the vertical one, from parents to their adult children, and the horizontal peer effects of siblings. First, offspring tend to replicate their parents’ family size, an empirical pattern known as intergenerational transmission of fertility (Murphy 2013; Beaujouan and Solaz 2019). Empirical evidence shows how the family of origin influences family size preferences, childbearing intentions, and fertility behaviour (Axinn et al. 1994; Barber 2000; Kotte and Ludwig 2011; Cools and Hart 2017; Dahlberg and Kolk 2018). Second, adult siblings have a positive effect on each other’s fertility, but with significant heterogeneous effects based on parity, gender, and proxies for strength of social ties (Kuziemko 2006; Lyngstad and Prskawetz 2010; Hart and Cools 2019; Buyukkececi et al. 2020). While it is unclear whether cross-sibling influences increase the total number of children or are limited to timing, they tend to be stronger for first births and when the siblings are sisters, close in age, or live in the same state (Kuziemko 2006; Lyngstad and Prskawetz 2010; Balbo and Mills 2011).

Extended family sex-mix preferences

Despite the demonstrated sex-mix preference, there are only a few studies exploring the reasons behind it (Hank and Kohler 2000; Mills and Begall 2010; Nugent 2013). They mostly point at sociocultural factors that can, in large part, be applicable to extended families as well. For example, older generations, in this case the grandparents, may hold stronger preferences for boys and may exercise social pressure on the middle generation differently, according to this preference (Nugent 2013; Bernardi and Klärner 2014). A cultural explanation that gains strength in multigenerational families is the preference for an uninterrupted line of male descendants (Nungent 2013; Larsen Gibby and Thomas 2019). The ‘family name’ is traditionally transmitted through men, and therefore a grandson born of a male child may bear higher importance for the extended family. If this is the case, the presence of a grandson who carries the grandfather’s last name could reduce the likelihood of additional grandchildren because the need for an ‘heir’ is already satisfied.

Another possible explanation for desiring a boy and a girl is the ‘symbolic capital’ associated with a balanced family (Nugent 2013). The saliency of a diverse sex composition may increase if it was not achieved within one generation. For example, if there are two brothers in the middle generation and the first two grandchildren are boys, there may be an additional desire for a third grandchild in the hope to have at least one girl in the family. In this case, (grand)parents that parented only boys could have an additional interest in helping to raise a girl, as it would involve a different set of perceived interests, traits, and skills (Jacobsen et al. 1999; Nugent 2013; Larsen Gibby and Thomas 2019), while reducing social opportunity costs for the parents (Aassve et al. 2012).

Sex-mix preferences in the extended family mean that not only does the sex distribution of one’s own children matter, but also that of nieces and nephews. First, parents may desire that their offspring interact with other children of the opposite sex, be they siblings or cousins, as part of their developmental growth (Kuziemko 2006; Halpern 2011). This would add a behavioural response to the specific sex-mix rather than just to the number of children born of siblings (Hart and Cools 2019). Second, social learning is an important channel for fertility peer effects on first time parents. Observing a sibling raising two boys or two girls with the related economies of scale and joint benefits from sex-specific time investments may positively influence the likelihood of childbearing compared to observing one boy and one girl (Raley and Bianchi 2006; Lois and Arránz Becker 2013; Baker and Milligan 2016; Buyukkececi et al. 2020). Belonging to an emotionally loaded set of relationships within the extended family, with frequent contacts or geographical proximity, enhances the saliency of other family members’ fertility behaviours, although it is hard to quantify precisely (Bernardi 2003; Kuziemko 2006; Lyngstad and Prskawetz 2010; Hart and Cools 2019).

Data

I use the nationally representative sample of American households provided by the Panel Study of Income Dynamics (PSID), which has been conducted since 1968. The key advantage of this dataset is that adult children of the initial respondents are invited to join the survey once they form their own economically independent households. This allows researchers to identify siblings and cousins nested within three generations (PSID 2018). The offspring of the original PSID respondents are now on average older than the mean age at first birth in the United States, although not all may have completed their fertility (NCHS 2018; Lundberg 2020). Following the PSID language, I use ‘dynasty’ to indicate a multigenerational family that comprises at least grandparents, a middle generation, and one or more grandchildren, while I use ‘family’ to indicate a nuclear family within a dynasty.

Using the 1986–2015 family and individual files, I match three generation dynasties where the original 1968 respondents are the grandparents. Figure 1 details the sample selection process. Following previous studies addressing sibling influences (Lyngstad and Prskawetz 2010; Cools and Hart 2019; Buyukkececi et al. 2020), the analyses include only dynasties with two siblings in the middle generation. This restriction significantly simplifies data construction and modelling but reduces the external validity of the study. Conversely, comparing dynasties with only two people in the middle generation increases internal validity by analysing individuals of at least childbearing age who grew up in families of the same size. Dynasties with half-siblings, adopted children, and those with incomplete information on the sex and year of birth of all the members in the middle and youngest generations are not included in the analyses. The resulting analytic sample consists of 906 dynasties with at least two grandchildren.

Figure 1. Sample Selection.

Source: Panel Studies of Income Dynamics (PSID), United States.

Notes: The characteristics of the resulting sample are reported in Table 1 and Table 2. The siblings in the middle generation are both at least of childbearing age (15 years old) to have at least the theoretical possibility of having a child. While not all sibling pairs completed fertility, the median and average age for younger and older siblings are in the late 30s-early 40s, well above the mean age at first birth in the United States, which was 26.6 at the time of last data collection (NCHS 2018).

Figure 2 provides a schematic representation of the types of dynasties, and nuclear families within them, in the analytical sample. Panels (a)-(d) represent four dynasties with different family structures. The grandparents (GP) are the original respondent household sampled in 1968, their two children constitute the middle generation, and are siblings to each other (hence, S1, S2). The youngest generation (grandchildren, GC) is a combination of grandsons (M) and granddaughters (F) and they can be either siblings or cousins, depending on whether they were born of the same parents. The number of (grand)children can differ between nuclear families and dynasties. For example, in panel (a) the dynasty has two grandchildren, while S2’s family is childless.

Figure 2. Schematic representations of families within dynasties by distribution of first two grandchildren.

Notes: Author’s representation. GP: grandparents; S1 and S2: sibling pair; F: granddaughter; M: grandson. This is a schematic representation of a possible family and dynasty structure, and variations are possible in where grandchildren are distributed within the nuclear families and in the sex of grandchildren and adult siblings in the middle generation.

A major limitation of using the PSID is that, by construction, it contains only the descendant of the original respondents. This means that subsequent analyses include the complete universe of grandchildren for the grandparents originally observed in the PSID, but the adult siblings in the middle generation may have additional nieces and nephews through their spouses. I use the available information on these in-laws, such as the prevalence, number, and sex distribution of the spouses’ siblings, to check that they do not differ significantly across observed dynasties with or without sex-mix, but unfortunately the crucial information on the sex distribution of the in-laws’ nieces and nephews is unavailable. As discussed in the next section, this leads to attenuation bias.

Table 1 presents the distribution of grandchildren’s sex at birth within dynasties with at least two grandchildren as well as the prevalence of different dynasty structures. The sex distribution of the grandchildren is slightly in favour of boys, especially for the first birth. Roughly a quarter of the dynasties have only boys or only girls as their first two grandchildren, while the remaining half achieves a sex-mix within the first two grandchildren, as expected. Family structures are unevenly represented, with 67% of the dynasties having the first two grandchildren born within the same nuclear family, i.e. the first two grandchildren are siblings rather than cousins (panels (a) and (b) in Figure 2).

Table 1.

Descriptive statistics for three-generational dynasties with at least two grandchildren: distribution of grandchildren across nuclear families and their sex distribution

	Mean	SD	N
Girl first	0.44	0.496	1,213
Girl second	0.49	0.500	906
Granddaughters – First two GC are girls	0.25	0.435	906
Grandsons – First two GC are boys	0.32	0.465	906
First two GC are of same sex	0.57	0.495	906
First two GC are siblings	0.67	0.471	906

Source: Panel Studies of Income Dynamics (PSID), United States.

Notes: GC stands for grandchildren. Mean coefficients in first column; standard deviations in second column; sample size in third column. Overall, there are 2,287 dynasties with two siblings of childbearing age in the middle generation. Of these, 1,213 (53%) have at least one grandchild, 906 (40%) have two or more grandchildren, and 606 (27%) have more than two grandchildren. Among all the dynasties with two siblings in the middle generation, the mean number of grandchildren is 1.52.

Table 8 in the Appendix reports characteristics of dynasties by number of grandchildren, comparing dynasties with more than two grandchildren to dynasties that are not eligible because they have fewer than two grandchildren.

Table 2 provides additional characteristics for each generation by grandchildren sex distribution. In the grandparents’ generation, the matriarch of the dynasty is born on average at the very end of the 1940s. The grandchildren’s sex distribution and the within family prevalence of cousins mirror the ones just presented in Table 1. In the middle generation, the sex distribution of the adult siblings is what randomness of sex at birth would predict, with a quarter being two brothers, a quarter being two sisters, and the remaining half being a brother and a sister. The first sibling in the middle generation was born in the late 1960s-early 1970s on average, while the second one was born on average three years after. While at least 60% of the adult siblings were married at some time, the first sibling is more likely to be married than the second one. Although information on the families of siblings’ spouses is limited (most notably lacking the number and sex of the other potential nieces and nephews of the couple), the available information suggests that, in the case of the adult siblings who are married, almost all the spouses have siblings and they have on average two.

Table 2.

Balancing tests: dynasties where the first two grandchildren are of the same sex are similar in observable characteristics to those with sex-mix

	(1) First two GC same sex		(2) First two GC different sex
	Mean	SD	Mean	SD	t-stat
Oldest and youngest generation
Grandmother’s birth year	1949.3	12.57	1949.0	11.80	−0.308
Girl first	0.44	0.498	0.45	0.498	0.166
Oldest two GC are siblings	0.65	0.477	0.70	0.463	1.220
Middle generation
Brothers	0.24	0.427	0.20	0.399	−1.471
Sisters	0.25	0.433	0.26	0.440	0.394
Mixed-sex middle generation	0.51	0.500	0.54	0.499	0.877
Sibling 1’s birth year	1969.5	13.64	1970.3	12.78	0.941
Sibling 2’s birth year	1972.9	13.58	1973.4	12.86	0.567
Sibling age difference	3.42	5.22	3.09	4.618	−0.999
Sibling 1 was ever married	0.74	0.441	0.79	0.410	1.768
Sibling 2 was ever married	0.64	0.479	0.61	0.497	−0.865
N	516		390		906
In-laws
Sibling 1’s spouse has siblings	0.95	0.215	0.95	0.223	−0.184
Sibling 2’s spouse has siblings	0.96	0.206	0.96	0.201	0.120
S1’s spouse siblings’ number	2.03	1.742	2.13	1.945	0.522
S2’s spouse siblings’ number	1.95	2.194	2.19	2.096	0.988
N	306		258		564

Source: Panel Studies of Income Dynamics (PSID), United States. Notes: GC stands for grandchildren. S1 is sibling 1 and S2 is sibling 2. Mean coefficients; standard deviations in second column; t-statistics in last column;

^***p<0.01,

^**p<0.05,

^*p<0.1.

Methods

This work identifies sex-mix preferences as manifested in the overall number of grandchildren descending from a set of grandparents by using the sex-mix in the first two grandchildren pooled across nuclear families. Because the sex-mix is randomly assigned, the estimation strategy is straightforward. I use an indicator for whether the first two grandchildren born within a dynasty have the same sex to predict whether the grandchildren pool contains more than two grandchildren. This can be estimated by OLS regression as follows:

$${\text{Y}}_{\text{i}}={\text{β}}_0+{\text{β}}_1{\text{S}}_{\text{i}}+{\text{β}}_2{\text{F}}_{\text{i}}+{\text{ε}}_{\text{i}}$$ (1)

Where the subscript i indicates that the measures are at the dynasty level, Y_i is a dummy variable equal to one if a dynasty has three or more grandchildren, S_i is a dummy variable equal to one if the first two grandchildren born into a dynasty have the same sex, and F_i is an indicator of whether the first two grandchildren are siblings rather than cousins. Note that this method mirrors the first stage of an instrumental variable approach, but the outcome of interest is fertility itself and therefore there is no second stage, nor concerns for an exclusion restriction. Given the randomness of the sex-mix within the first two grandchildren, I do not include additional control variables in the main model. This approach tests the expression of sex-mix preferences in the extended family through additional transitions to more than two grandchildren in dynasties with no sex-mix. Including family structure accounts for differences between previously identified within-nuclear family effects (Ben-Porath and Welch 1976; Angrist and Evans 1998) and those that manifest across the entire grandchildren pool of the observed dynasty.

The key assumption underlying this causal estimation is the randomness with which dynasties achieve a grandchild sex-mix with the first two grandchildren, and which have same-sex grandchildren instead. This holds if the sex at birth is random, is not correlated across siblings, and parents cannot influence the sex of their offspring. These are theoretically reasonable assumptions in developed countries where there is no evidence of severe malnutrition or sex-selective abortion (Almond and Edlund 2008). To corroborate these assumptions empirically, Table 2 compares observable characteristics by generations for dynasties with and without sex-mix. There are no statistically significant differences across sex-mix for observable characteristics at the grandparental level, grandchildren sex composition, middle generation characteristics, and family structure, including in-laws. Therefore, dynasties in the sample are similar along observable characteristics and, thanks to the randomness of the sex at birth, plausibly also on unobservables.

The fact that dynasties with or without a sex-mix in the first two grandchildren display statistically indistinguishable characteristics on average supports the assumptions behind this methodology. However, it does not solve the problem of the lack of information of additional nieces and nephews born from the siblings of the observed adult siblings’ spouses. Table 2 shows that couples with or without a sex-mix do not marry in families of different sizes on average, but the data do not contain information on whether the in-laws have children and whether they achieved a sex-mix within their side of the family. The unobserved in-laws can be seen as a case of non-compliance to (the lack of) sex-mix as some of them may indeed have a boy and a girl. This leads to an attenuation bias towards zero, so the finding of an effect is a harder test of the existence of extended family sex-mix preference than it would be if the in-laws were also observed.

Family structure plays a key role in shaping fertility behaviours within extended families. Pooling across all grandchildren born from the same grandparental line poses the unique challenge of accounting for different transition rates by birth parity. In this setup, a third grandchild could be born within the same nuclear family as the first two, could be a firstborn child with two cousins, or a second child with a cousin. Bringing this back to the schematic representation in Figure 2, grandchildren born in dynasties represented in panels (a) and (b) would be either a third birth for sibling 1, or a firstborn for sibling 2. Those transition rates would be different from families nested in dynasties (c) and (d), where the third grandchild would be a second born, regardless of which family experiences the birth. Therefore, it is important to test the role of family structure by separately controlling for an indicator capturing whether the first two grandchildren are siblings or are born to different parents (i.e. they are cousins). The established sex-mix preference within a nuclear family manifests in the case of a third sibling, but evidence that more grandchildren are born in dynasties with cousins of the same sex than in mixed-sex cousins points to the manifestation of extended family sex-mix preferences and influences.

In addition to identifying an extended family sex-mix effect on fertility beyond previous studies within nuclear families, the three-generation setup allows me to test possible explanations for it. For example, I test whether there are grandson-specific preferences driven by more strongly held preferences in the older generation. Given the cultural importance of the transmission of the family name, I look at whether the grandfather who was an original PSID respondent has an uninterrupted male line through a son and a grandson. Lastly, I test the role of symbolic capital for a sex-balanced three generational dynasty by looking at the effects of lacking descendant heterogeneity (i.e. two brothers with only boys or two sisters with only girls). In the last set of analyses, I introduce characteristics of the middle generation, such as closeness in age and being sisters rather than brothers or a mixed composition, which previous literature on fertility peer effects identified as facilitating social learning. However, it must be noted that this exercise is exploratory and including these additional covariates may introduce endogeneity and thus these last results should be interpreted with caution.

Results

Table 3 presents the main results of estimating the probability of having more than two grandchildren in a dynasty based on the sex-composition of the first two grandchildren. As expected, if the first two grandchildren are of the same sex there is a positive and significant effect on dynasty fertility overall, meaning that the sex-mix preference manifests also in the extended family. At the same time, family structure plays an important role, as dynasties in which the first two grandchildren come from the same nuclear family are less likely to have more than two grandchildren. This indicates that transitions to first and third births combined are less frequent than second births, in line with previous demographic evidence.

Table 3.

Positive effect of no sex-mix in grandchildren pool on dynasty fertility

	Dynasty has more than two grandchildren
	(1)	(2)	(3)	(4)
Two oldest GC same-sex	0.069** (0.0317)		0.062** (0.0312)	0.065 (0.0546)
Two oldest GC same parent		−0.192*** (0.0328)	−0.190*** (0.0328)	−0.187*** (0.0508)
Interaction: same parent x same sex				−0.005 (0.0665)
Observations	906	906	906	906
R-squared	0.005	0.037	0.041	0.041

Source: Panel Studies of Income Dynamics (PSID), United States.

Notes: GC stands for grandchildren. Standard errors in parentheses

^***p<0.01,

^**p<0.05,

^*p<0.1.

What is most noteworthy is that the positive effect of the lack of sex-mix on fertility remains significant and unchanged in sign and magnitude when family structure is controlled for, see Table 3, column (3), indicating a separate effect of this exogenous component. The introduction of an interaction between the indicators for family structure and sex-mix in column (4) covers all schematic possibilities presented in Figure 2. The presence of the interaction term does not change the magnitude of the other coefficients and the model remains jointly significant. However, including the interaction term reduces the statistical power due to a reduction in the degrees of freedom and the interaction is not itself significant. This means that there is no evidence that the main effect of family influences on fertility via the sex composition is driven by a specific nuclear family structure.

The fact that there is no interaction effect between grandchildren being born of the same parents and being of the same sex is relevant because it relates to the central question of whether there is evidence of sex-mix preference that goes beyond the cases where offspring come from the same nuclear family. Table 4 shows the results from the same models but conducted separately on sub-samples of dynasties according to whether the first two grandchildren are siblings or cousins and which adult sibling in the middle generation, if any, has the third grandchild in the dynasty. Although separating dynasty structures means that sample sizes for these models are smaller and the estimates are therefore not statistically significant, this exercise is useful in illuminating how sex-mix preferences and family influences manifest. Table 4 column (1) reports results for sex-mix preferences within the same nuclear family as in this dynasty only one of the siblings in the middle generation has all the grandchildren. The effect size is comparable to those found in previous studies looking at the propensity of having a third child if the first two are of the same sex, confirming the presence of sex-mix preferences within a nuclear family. Columns (2) and (3) use the dynasty structure of the data to estimate sex-mix preferences across cousins, separating fertility sibling peer effects for a first birth (Column 2) and overall preference for sex-mix among cousins (Column 3). Always keeping in consideration the uncertainty around these point estimates, the ordering of the size of effects is in line with theoretical expectation, as within-nuclear family sex-mix preferences are stronger than sex-mix preferences among cousins and fertility peer effects. The cousin results also support the main finding of sex-mix preference manifestation at the extended family as, without coordinated family preferences and influences, there is no reason why when two cousins are of the same sex both observed families are more likely to have a child than if the two cousins were both boys or girls.

Table 4.

Effect of no sex-mix in grandchildren pool on dynasty fertility separating sex-mix preferences within the same nuclear family and across cousins.

		Dynasty has more than two grandchildren
	(1) Two oldest GC siblings: 3rd GC same nuclear family	(2) Two oldest GC siblings: 3rd GC is a cousin	(3) Two oldest GC are cousins
Two oldest GC same-sex	0.067 (0.0473)	0.058 (0.0492)	0.065 (0.0478)
Constant	0.4208*** (0.0351)	0.3500*** (0.0362)	0.752*** (0.0370)
Observations	448	393	301
R-squared	0.004	0.004	0.006

Source: Panel Studies of Income Dynamics (PSID), United States.

Notes: GC stands for grandchildren. Standard errors in parentheses

^***p<0.01,

^**p<0.05,

^*p<0.1.

Column (1) includes dynasties where all the first three grandchildren are born within the same nuclear family, indicating sex-mix preferences within the same family. Column (2) includes dynasties where the first two grandchildren are siblings, but the third one is a cousin born by the previously childless adult sibling in the middle generation, indicating potential for between-sibling peer effects. Column (3) includes dynasties where the first two grandchildren are cousins and therefore the preference for sex-mix among cousins is manifested with a child of parity 2 for either one of the adult siblings in the middle generation.

Table 5 explores the role of family structure more in detail by reporting the distribution of grandchildren for dynasties that go on to have three grandchildren according to their family structures and the sex of the first two grandchildren. Column (1) describes the various birth order possibilities for a third birth, given different combinations of which sibling had the first two grandchildren. Column (2) represents the family structure types described in the first column as events. Event one is realized if sibling 1 has a birth and, conversely, event two if the birth occurs to sibling 2. This representation highlights that symmetrical family structures (e.g. sibling 1 has all three grandchildren or sibling 2 has all three) are grouped together. In line with the descriptive statistics presented in Table 1, the third grandchild is a second birth within a nuclear family in about 30% of cases, making these family structures less prevalent in the sample. However, comparing the prevalence of third births within family structures but across sex-mix in the first two grandchildren (columns (3) and (4)) confirms that dynasties with no sex-mix have a larger share of third grandchildren for all family structures.

Table 5.

Distribution of grandchildren for dynasties with at least three grandchildren by sex-mix and family structure

(1) Family structure of 3rd births	(2) Distribution of who has GC	(3) Different sex two oldest GC	(4) Same sex two oldest GC	(5) Totals
Third GC is a third birth	{1,1,1}, {2,2,2}	95 (38.5%)	152 (61.5%)	247 (41.4%)
Third GC is a first birth	{1,1,2}, {2,2,1}	64 (38.1%)	104 (61.9%)	168 (28.2%)
Third GC is a second birth	{1,2,1}, {2,1,2} {1,2,2}, {2,1,1}	83 (45.8%)	98 (54.1%)	181 (30.4%)
Totals		242 (40.6%)	354 (59.4%)	596 (100%)

Source: Panel Studies of Income Dynamics (PSID), United States.

Notes: GC stands for grandchildren. Column (1) describes the family structure for the first three grandchildren, where in the first and second row the first two grandchildren are siblings, i.e. born of the same parents. Dynasties where the first two children are cousins are in the third row. Column (2) has all the eight possible family structures for the first three grandchildren. 1 indicates that the birth occurred to sibling 1, and 2 indicates a birth to sibling 2. For example, {1,1,2} means that sibling 1 has the first two grandchildren, while sibling 2 has the third grandchild.

Motivations for sex-mix preferences: Across generations sex composition

Table 6 explores possible sources of extended family sex-mix preferences, with a focus on the intergenerational dimension: grandson preferences, symbolic capital in descendent heterogeneity, and cultural value of an uninterrupted male line. The first possible manifestation of intergenerational preferences is through a preference for grandsons driving the results. Indeed, older generations may maintain stronger preferences for boys and may exercise social pressure differentially according to this preference. Substituting the same-sex indicator for one in which the first two grandchildren are boys halves the original coefficient, suggesting that there is no preference for grandsons or granddaughters, but for a sex-mix over either one. Concurrently, it is possible that preferences for sex-mix in the grandchildren pool undergo cohort changes, but a proxy like the grandmother’s year of birth is precisely estimated to be zero, therefore suggesting a limited role of sex-mix preferences changing over time. However, the spread of ages in the grandparental generation may not be wide enough to capture changes in preferences that are slow to change (results available upon request).

Table 6.

Effects on dynasty fertility of sex composition across generations

	Dynasty has more than two grandchildren
	(1)	(2)	(3)	(4)	(5)	(6)
Grandsons	0.0388 (0.0338)	0.0271 (0.0333)
No descendant heterogeneity			−0.0536 (0.0424)	−0.0513 (0.0416)
Male line					−0.0679** (0.0327)	−0.0882*** (0.0322)
Two oldest GC from same parent		−0.191*** (0.0328)		−0.192*** (0.0328)		−0.202*** (0.0328)
Constant	0.650*** (0.0190)	0.781*** (0.0293)	0.671*** (0.0172)	0.799*** (0.0276)	0.687*** (0.0196)	0.829*** (0.0301)
Observations	906	906	906	906	906	906
R-squared	0.001	0.037	0.002	0.038	0.005	0.045

Source: Panel Studies of Income Dynamics (PSID), United States.

Notes: GC stands for grandchildren.

Grandsons is equal to one if the first two grandchildren are both boys.

No descendant heterogeneity is equal to one if the middle generation and the grandchildren are all of the same sex. For example, two brothers with only grandsons or sisters with only granddaughters.

Male line is equal to one if there is at least one grandson in the first two who is born from a man in the middle generation, i.e. he shares the original PSID respondents’ last name.

Standard errors in parentheses

^***p<0.01,

^**p<0.05,

^*p<0.1.

Second, it is possible that there is a broader intergenerational preference for diversity in the sex composition of descendants rather than just for the sex composition in one generation, following the hypothesis that there is symbolic capital associated to a sex-balanced family. Therefore, Table 6 includes a “descendant heterogeneity” term to reflect a possible cross-generation interaction in terms of sex-mix between the middle generation and the grandchildren. There is no descendant heterogeneity when the sex of the middle generation and the grandchildren match and this should increase overall fertility of the dynasty if intergenerational diversity is preferred. Examples of dynasties with no descendant heterogeneity are two brothers in the middle generation and grandsons in the youngest generations, or sisters with granddaughters. The rationale is that, in the absence of sex-mix in the middle generation, seeking the opposite sex in the youngest generation might be even more salient than in a dynasty with a middle generation with a sister and a brother, or in a dynasty where the first grandchild is of the opposite sex than their reference parent. Results in Table 6 columns (3) and (4) do not support the existence of intergenerational sex-mix preferences, as the combination of middle and youngest generation sex is insignificant across all models. However, it is not possible to fully dismiss this hypothesis because of the aforementioned attenuation bias.

Lastly, a cultural explanation combines conceivable preferences for grandsons and the intergenerational dimension to support a preference for an uninterrupted line of male descendants. The “family name” is traditionally transmitted through men and therefore a grandson born of a male child might bear higher importance for the extended family. Coefficients in Table 6 columns (5) and (6) are negative and significant, meaning that the presence of a grandson who carries the observed grandfather’s last name reduces the likelihood of having additional grandchildren because the need for an “heir” is already satisfied. It is worth noting that this is the only case where the magnitude changes when controlling for family structure in column (6). This is due to the de facto restriction to dynasties with at least a brother in the middle generation, which increases the collinearity between the male line and family structure. Therefore, this result should be interpreted with caution and taking into consideration that men tend to have children later in life than women (see also Rupert and Zanella 2018).

Peer fertility effects: Characteristics within the middle generation

Table 7 reports results including characteristics of the middle generation, which are expected to shape fertility based on peer fertility effects, as adult siblings in the middle generations are the ones making the childbearing decisions. Although these characteristics can be informative for exploring mechanisms of sex-mix preference manifestation, it is important to note that including them may introduce bias if they are influenced by offspring sex-mix. Therefore, these results do not necessarily have a causal interpretation. The main takeaway from the table remains that the primary result of the lack of sex-mix in the first two grandchildren positively affecting the number of grandchildren in the overall dynasty remains unchanged.

Table 7.

Effects on dynasty fertility including characteristics of the middle generation

		Dynasty has more than two grandchildren
	(1)	(2)	(3)	(4)	(5)
Two oldest grandchildren have the same sex	0.0667** (0.0310)	0.062** (0.0312)	0.062** (0.0312)	0.060** (0.0300)	0.066** (0.0299)
Two oldest grandchildren from same parent	−0.187*** (0.0326)	−0.188*** (0.0328)	−0.188*** (0.0330)	−0.142*** (0.0320)	−0.140*** (0.0319)
Two men in middle generation (brothers)	−0.125*** (0.0370)				−0.133*** (0.0356)
Two women in middle generation (sisters)		0.045 (0.0354)
Siblings close in age (less than four years apart)			0.012 (0.0325)		−0.003 (0.0311)
Sibling 1 ever married				0.080** (0.0354)	0.085** (0.0352)
Sibling 2 ever married				0.248*** (0.0317)	0.249*** (0.0315)
Constant	0.777*** (0.0331)	0.741*** (0.0341)	0.745*** (0.0406)	0.506*** (0.0465)	0.527*** (0.0512)
Observations	906	906	906	906	906
R-squared	0.053	0.043	0.041	0.115	0.128

Source: Panel Studies of Income Dynamics (PSID), United States.

Notes: Standard errors in parentheses

^***p<0.01,

^**p<0.05,

^*p<0.1

Table 7 column (1) shows that the fact that the two siblings in the middle are both men is negatively associated with the dynasty having more than two grandchildren. This is in line with the fact that men tend to have children later, thus influencing the timing of their parents becoming grandparents (Rupert and Zanella 2018). It is therefore possible that having two sons in the middle generation reduces the probability of having more than two grandchildren because the childbearing time window extends to later ages for men and they may not have completed fertility compared to dynasties with daughters or a mixed-sibling composition. Moreover, we would expect sisters to positively influence each other’s childbearing more (Kumzienko 2006; Lyngstad and Prskawetz 2010), but Table 7 column (2) shows that this is not the case.

Following a similar reasoning, Table 7 column (3) also includes a dummy for whether siblings are less than four years apart, but it does not seem to play a role in overall fertility of the dynasty, contrary to the expectation that social learning occurs more for siblings who are closer in age (Kumzienko 2006). Conversely, as we would expect from positive correlations between fertility and marriage in this sample, marital status of the siblings in the middle generation is positive and significant. The PSID does not include measurements of cohabitation throughout the period, so whether an individual has ever been married has been used as a rough proxy for the middle generation being in a union during their childbearing years. This finding is thus in line with the demographic intuition that being in a union increases the likelihood of childbearing.

Extended family ties

The increase in the propensity to have a third grandchild in a dynasty when the first two are of the same sex indicates a preference for sex-mix, but it does not identify the sources of such preferences. Previous studies suggest that both the middle and the older generation play a role in shaping fertility. Disentangling the two is beyond the scope of this paper, but I did conduct some data exploration on the availability of grandparents and geographical proximity to contextualize my findings. First, having alive grandparents can diminish the social opportunity costs of having an additional child, but not enough grandparental deaths occurred in between the births of different grandchildren to identify discernible patterns between dynasties in which the grandparents were alive or not. However, it is not theoretically clear whether the death of a grandparent makes some theoretical mechanisms, such as the cultural desire for a male heir, more or less salient.

Second, family members who live geographically closer could exert more influence on each other. Members of the observed extended family, excluding in-laws for which this information is not available, tend to live in the same geographical area. At the last survey in 2015, slightly more than 7% of older siblings in this analytic sample did not live in the same state as their mothers and the corresponding figure for younger siblings is slightly less than 7%. At the same time, about 11.4% of siblings do not live in the same state as each other, indicating some overlap in the two groups. This geographical proximity for the majority of the sample suggests that family members had the opportunity to be connected with each other, although the PSID does not contain measures of frequency and depth of contact that could be used to quantify the relationships.

Lastly, Table 10 in the Appendix shows that the sex of the first grandchild alone does not predict additional grandchildren, confirming a sex-mix preference rather than a preference for a specific sex, as previously shown. This addresses the concern that a preference for sons might lead families with a firstborn daughter to have more children overall (Dahl and Moretti 2008). The main results are also robust to alternative specifications of the sample (Table 9 in the Appendix), such as only including dynasties where there are no missing father IDs or excluding cases where cousins were born within a year of each other and therefore the two siblings may have been trying to conceive at around the same time. Unfortunately, limited availability of information on the children born to the siblings of siblings’ spouses does not allow the inclusion of nieces and nephews from the in-laws, but only to follow the descendants of the original respondents for which the full pool of grandchildren is observed.

Discussion and Conclusion

This paper builds on the well-known preference for offspring sex-mix and suggests that it manifests also in the pool of grandchildren descending from common grandparents. By constructing three-generational dynasties including the original PSID respondents, their children, and their grandchildren, I show that dynasties with no sex-mix in the two oldest grandchildren are more likely to have more than two grandchildren than those dynasties in which the first two grandchildren are of different sexes, even if the two types of dynasties are otherwise similar on average. A main contribution of this paper is to look at sex-mix preferences in cases when the first two grandchildren are born of the same parents, when they are cousins, and by the distribution of grandchildren across adult siblings’ nuclear families.

The estimates of the likelihood that a dynasty has more than two grandchildren are remarkably similar for dynasties where the first two grandchildren are siblings and for those where they are cousins. These estimates are comparable to results from previous studies that estimate the propensity to have a third child within the same nuclear family to be about 6 percentage points higher when the two firstborns are of the same sex than when they are of the opposite sex.

There are several factors that could be contributing to this finding. The first one that progression to an additional birth depends on family structure and parity progression rates. While we would expect parents to place more emphasis on the sex composition of their own children than of their nieces and nephews, dynasties where the first two grandchildren are cousins have two adult siblings who have proven to be able and willing to have children. Transitions to a second birth within a nuclear family are easier to achieve than a third birth within the same nuclear family, especially in a context of lower fertility compared to the generations studied for the initial identification of sex-mix preferences (Ben-Porath and Welch 1976).

Yet, transitions to a third grandchild occurred disproportionally in dynasties with no sex-mix in the first two grandchildren in all family structures, highlighting the manifestation of sex-mix preferences despite different parity progression rates. For example, in dynasties where one nuclear family has all three grandchildren, more than 60% of those did not achieve a sex-mix within the first two. This 40-60 gap between those with different sex-mix is similar to those dynasties where the third grandchild is born of the previously childless adult sibling. Therefore, while nuclear family structures within dynasties affect the probability of having more than two grandchildren, there is not a specific distribution of grandchildren driving the results.

The second factor explaining the similarity of the cousins’ and siblings’ estimates are extended family influences per se. Family influences are a complex web coming both from parents and adult siblings and are not empirically distinguishable by source but, rather, interact and reinforce each other (Bernardi 2003; Aassve et al. 2013). For example, it is not necessary to assume that grandparents exercise a strong social pressure to impose their preference for sex-mix for them to have a role. Indeed, they could be acting as facilitators of family interactions between nuclear families (from shared meals and events to joint babysitting), thus increasing the saliency of the sex composition of their nieces and nephews for their childbearing age children. The higher fertility in dynasties with no sex-mix in the first two grandchildren can then suggest a desire for interaction with children of different sexes, be it siblings or cousins, as part of the formative process in support of the view that siblings adopt behaviours that support sharing common experiences for themselves and for similarly aged cousins (Kumzienko 2006). Common measures from the peer effects literature, such as similarity in age or having two sisters in the middle generation, did not play a role in sustaining sex-mix preferences in the dynasty, but it is possible that more detailed data on geographical proximity and frequency of contacts can illuminate this pathway.

There is, instead, support for mechanisms linked to why sex-mix preferences arise. The psychological and sociological origins of offspring sex preferences are still being studied, but I tested the cultural explanation based on patrilinear family name inheritance and the symbolic capital deriving from a balanced family because they may be especially relevant in a multigenerational setting. I find no support for the symbolic capital of a balanced extended family, but my results support the existence of a preference for an uninterrupted line of men from the grandfather to the grandson; those dynasties where this is fulfilled within the first two grandchildren are less likely to have additional grandchildren than if there is no male heir. The two caveats to this result are that men tend to have children later (Rupert and Zanella 2018) and that women in the middle generation may be facing similar pressure from the husband’s family if the in-laws do not have a male heir, but these are not observable with these data.

Lastly, it is difficult to pinpoint exact point estimates and quantify the difference between siblings and cousins estimates because of the small sample size and the attenuation bias due to the lack of information on the in-laws. Indeed, using three generations comes with some limitations. The most evident is the constraint on data choice, the PSID being one of the very few datasets containing three generations, especially in the American context. The PSID follows descendants of the original respondents, so little is known about the other side of the extended family. While access to this type of data would constitute an important addition to the present study, concerns are mitigated by the fact that the non-surveyed side of the extended family is comparable across the sex distribution of the observed grandchildren on the of observable in-laws’ characteristics, such as the number and the sex of the partner’s siblings, and there is no theoretical reason for which the in-laws should be systematically different by observed grandchildren sex-mix. Moreover, the incomplete compliance due to the unobserved sex composition of nieces and nephews in the children of a spouse’s siblings leads to attenuation bias. Therefore, my significant finding of sex-mix preferences in the observed dynasty is, if anything, a lower bound for an even stronger effect.

Another data-related limitation is that, once attrition and childlessness are taken into account, the sample size does not allow for additional stratifications, beside the ones shown, without falling into a small cell size problem. This is particularly evident for the geographical distribution of the extended family members, as most live close to their family of origin. Therefore, separating those adult siblings who live close to their parents, close to each other or far from both, did not yield enough variation. Similarly, grandparental death occurring between the first two grandchildren or after the first two but before the third grandchild amounted to less than one percent each. The small prevalence of grandparental death around the time of childbirth does not make it possible to investigate the role of exogenous reduction of grandparents’ availability to support the middle generation.

Finally, the restriction to dynasties in which the middle generation is comprised of two, and only two, siblings reduces the external validity of this study. There are reasons to believe that, as the number of adult siblings in the middle generation increases, so does the probability that dynasties will have more than two grandchildren. First, because of the larger number of people who could experience childbearing, and second because third births increasingly occur primarily among individuals coming from larger families (Beaujouan and Solaz 2019). However, the case of only children in the middle generation is less clear-cut and could go either way, depending on the strength with which preferences for a sex-mix are held, and on the sources of the family influences on fertility.

In light of these data limitations, future research should make use of registry data from other contexts and with larger sample sizes, where it is increasingly possible to link three generations and different sides of the extended family. The increased power from more observations would allow for stratification beyond the results presented in this study. For example, this type of data could help isolate the effect of grandparents and adult siblings in cases when they have less influence on fertility choices through either geographical distance or death. Moreover, it could be possible to ascertain the role of socio-economic status and whether there is a dominant side of the extended family when it comes to offspring preferences. This could also contribute to the understudied field of why there are sex-mix preferences in the first place.

Lastly, this paper also makes an important methodological contribution that can be used to study outcomes outside fertility as well. Using a natural experiment originally employed only within the nuclear family, it extends the applicability of this method in several ways. It is more applicable to low fertility contexts where higher parity births are becoming less common (Del Boca et al. 2005; Daouli et al. 2009), but there is no clear prevailing norm on the number of grandchildren. Indeed, previous studies using the sex composition of own children as an instrumental variable for additional fertility can only identify the marginal effect of the third child, unlike the case when the third birth is a cousin.

This empirical approach with multigenerational data also opens new estimation avenues for investments in children’s human capital, parental employment, and labour force participation in the grandparents’ generation, an area of growing interest due to population ageing. A natural extension is to use the results so far obtained as the basis for an intergenerational instrumental variable. By limiting the analyses to nuclear families, it is possible that sex-specific returns to scale can make having two children of the same sex directly affect parents’ outcomes, thus threatening the exclusion restriction. Locating the treatment at the level of the extended family rather than the nuclear household, the causal framework proposed in this work can be used as an instrument less subject to such issues. The models presented here pass the heuristic measure for a non-weak instrument when family structure is accounted for and therefore lend themselves to studying the effect of more grandchildren on early retirement or participation in the labour force of grandparents using an alternative instrument less subject to exclusion restriction threats.

Appendices

Table 8.

Descriptive statistics for sex distribution and age by number of grandchildren present in the dynasty

	(1) Dynasties with less than two grandchildren		(2) Dynasties with more than two grandchildren
	Mean	SD	Mean	SD	t-stat
Grandmother’s birth year	1956.3***	10.57	1948.1	12.29	14.492
Sibling 1 birth year	1980.1***	11.32	1968.3	13.22	21.066
Sibling 2 birth year	1982.7***	11.39	1971.6	12.93	19.692
Sibling 1 was ever married	0.264***	0.441	0.794	0.405	−25.932
Sibling 2 was ever married	0.177***	0.382	0.726	0.446	−29.010
Brothers	0.269***	0.444	0.188	0.391	3.989
Sisters	0.214**	0.410	0.267	0.443	−2.677
Mixed-sex middle generation	0.516	0.500	0.545	0.498	−1.191
Sibling age difference	2.554***	4.462	3.358	4.996	−3.680
N	1681		606		2,287

Source: Panel Studies of Income Dynamics (PSID), United States.

Notes: Mean coefficients; standard deviations in second column; t-statistics in last column;

^***p<0.01,

^**p<0.05,

^*p<0.1.

This table reports observable characteristics for dynasties based on the number of grandchildren. As could be expected, dynasties with more than two grandchildren are, on average, older in both in the oldest and middle generation than dynasties that do not meet the inclusion criteria because they have fewer than two grandchildren. Both siblings are also more likely to be married. However, in half of the dynasties the middle generation is composed of a brother and a sister regardless of the number of children.

Table 9.

Positive effect of no sex-mix in grandchildren pool on dynasty fertility for subsample of dynasties with no missing father IDs

	Dynasty has more than two grandchildren
Two oldest GC same-sex	0.0926** (0.0362)		0.0835** (0.0357)	0.0949 (0.0622)
Two oldest GC same parent		−0.188*** (0.0376)	−0.183*** (0.0375)	−0.173*** (0.0578)
Interaction: same parent x same sex				−0.0170 (0.0760)
Constant	0.611*** (0.0272)	0.788*** (0.0306)	0.738*** (0.0372)	0.731*** (0.0481)
Observations	686	686	686	686
R-squared	0.009	0.035	0.043	0.043

Source: Panel Studies of Income Dynamics (PSID), United States.

Notes: GC stands for grandchildren. Standard errors in parentheses

^***p<0.01,

^**p<0.05,

^*p<0.1.

Table 10.

The sex of the first grandchild alone does not predict additional grandchildren

	More than one grandchild
	(7)	(8)	(9)
Grandson	−0.0264 (0.0251)
No descendant heterogeneity		−0.0368 (0.0280)
Male line			−0.0437 (0.0278)
Constant	0.765*** (0.188)	0.760*** (0.0146)	0.762*** (0.0146)
Observations	1,213	1,213	1,213
R-squared	0.001	0.001	0.002

Source: Panel Studies of Income Dynamics (PSID), United States.

Notes: Differently from all other results, this table focusses on the sex of the first grandchild only, therefore the sample includes dynasties with at least one grandchild.

No descendant heterogeneity is equal to one if the middle generation and the grandchild are all of the same sex. For example, brothers with a grandson or sisters with a granddaughter.

Male line is equal to one if the oldest grandchild is a boy and is born from a man in the middle generation, i.e. he shares the original PSID respondents’ last name.

Standard errors in parentheses

^***p<0.01,

^**p<0.05,

^*p<0.1.