Introduction

Europe has experienced a very serious problem of population ageing that deteriorates growth and welfare prospects for decades. People live longer and simultaneously fewer children are born. The total fertility per woman ensuring generation replacement (i.e. 2.1) is now unachievable for European countries. Only Georgia is quite close to this level, reporting a fertility rate over 2.0 in the past few years; however, this shows a decreasing trend. In 2020, it was 1.98 according to Eurostat data.Footnote 1 Across the most developed countries, France stands out in these terms with a fertility rate exceeding 1.8 in 2020. Spain and Malta experience low fertility, mostly with rates below 1.2. Data show that the problem is vital and its solution or at least mitigation is receiving special attention and concern from demographers and other social researchers as well as policy makers. This finds confirmation in numerous empirical analyses focused on various factors of fertility decisions. At the general institutional level, the development of the welfare state over recent decades is said to be one of factors hampering fertility. The state as a generous caregiver reduces the perception of the role of children as their parents’ investment. Relatively high public pension benefits have undermined the necessity of offspring providing benefits in cash or in kind (Billari and Galasso 2009). This is reinforced by difficulties that women experience trying to combine work and motherhood; that is why formal childcare and, according to evolutionary theory, kin ease this dilemma (Kaptijn et al. 2010). Therefore, grandparental support is mentioned as one of the crucial factors affecting household or extended family fertility decisions depending on the more selfish or altruistic behaviour of individuals (Altonji et al. 1992). Grandparents are considered to be a supportive power in childcare, affecting fertility positively in this way (Hoem 2008; Kaptijn et al. 2010; Wang and Zhao 2022; Tanskanen and Rotkirch 2014). However, there is also research demonstrating a restrictive role of grandparental support for fertility. This may have, inter alia, ideological grounds. For example, as indicated by Tan (2023), in China the geographical proximity of a wife’s parents may impact fertility negatively, as the husband’s family is expected to provide support for his wife, including childcare. The results obtained for Spanish households suggest that the impact of expected grandparental support positively affects only women’s fertility intensions, not men’s intentions (Rutigliano and Lozano 2022). The findings of a study conducted for Germany imply that grandparents’ time resources released as a consequence of retirement affects the fertility timing of their adult children, but does not affect the total number of offspring they have (Eibich and Siedler 2020). As indicated by Kochhar and Cohn (2011), US data show that a growing number of people living in multigenerational households (an increase from about 13% of the US population in 1970 to 17% in 2009) does not translate into increasing fertility (2.48 in 1970, 2.00 in 2009).Footnote 2

As we demonstrate in the literature review (Sect. "A multigenerational context"), a large amount of research often focuses on the impact of grandparents supporting their adult children with childcare (Amorim 2019; Kaptijn et al. 2010; Rutigliano and Lozano 2022; Tanskanen and Rotkirch 2014; Okun and Stecklov 2021). The impact of the multigenerational character of a household has received less attention, however. Addressing this gap, we hypothesise that it does matter where grandparents live, i.e. whether they live in the same household with their adult children or not. We draw from the above-mentioned research, indicating that grandparents as childcare providers support fertility on the one hand, but they do not necessarily stimulate fertility as members of multigenerational households on the other. We investigate whether a household is multigenerational or not through the inclusion of a binary variable explaining whether or not a household includes any older members (i.e. aged 65 years or older) in models describing the number of young people (children or youths) living in the same home. Consequently, we account for two possible circumstances that may negatively affect fertility: the number of household members and potential overcrowding problems, as well as co-residence with parents as a socio-economic barrier (not a favourable condition) to having children (or more children). Thus, not questioning previous literature confirming the positive impact of grandparents’ support on fertility, we make a step forward and examine whether their co-residence with adult children enforces or weakens this support. Finally, we pose the following research question: does a multigenerational household (the co-residence of grandparents and their adult children) facilitate the formation of multichild families?

In testing this nexus, we use data from the Luxembourg Wealth Study (LWS). Our data set covers the following countries: Austria, Germany, Estonia, Spain, Finland, Greece, Italy, Luxembourg, Norway, Slovenia and Slovakia. The findings do not support the view that intergenerational households, in which grandparents live, include more offspring aged 13 or under, or 17 or under. Thus, co-residence of two adult generations of pensioner parents and their working-age children does not seem to stimulate growth in the number of multichild families. The results deliver some initial arguments that what can matter is gender equality policy facilitating solving of the motherhood–work conflict for families and women.

The paper contributes to the strand of literature addressing households’ fertility decisions and the forces driving these, particularly centring on the family constellation and referring to multigenerational families. Although the majority of models used in previous studies explain the fertility decision as such (to have or not have a child), we focus on the number of children, particularly interpreting the results with reference to multichild families. What makes our investigation novel is the use of the co-residence of two adult generations in the same home as the main predictor, whereas other research focuses on grandparents’ support and rather disregards the co-residence issue. Using microdata from LWS for 11 countries, we conduct a study based on cross-national microdata while single-country analyses dominate in the literature. Additionally, we relate our study to family and gender equality components of public policy, which allows for the development of some initial and cautious policy implications of the research.

The remainder of the paper is structured as follows. In the literature review section, we discuss theoretical concepts of children as consumption and investment goods, refer to the theoretical grounds of family policies and summarise the vast body of empirical studies on the driving forces of fertility with special attention focused on grandparents’ support. Subsequently, we characterise data used and present and justify our methodological approach. The two last sections include presentation and discussion of the results, ending with final conclusions.

Determinants of fertility decisions: a literature review

Children as a source of utility

The fundamental question behind any household-level studies on fertility regards the motivation making people want to have children. In 1997, Robinson, referring to the vast body of demographic literature and noticing an “intellectual triumph” of the “economic model” as “the most widely used framework” for studying fertility, argued that there is much to explain about fertility around the world. There are two main theoretical approaches to fertility mentioned in the economic literature. One approach is based on the assumption that a child is a consumption good; the other assumes a child to be an investment good. According to the consumption theory of fertility developed by Leibenstein (1957), a couple weighs up utilities and disutilities when making a decision about having (or not having) the first or subsequent child, obviously maximising their lifetime utility. Becker (1965) then completed this model with neoclassical assumptions (including fixed preferences and maximising behaviour) and placed fertility in the broad context of household economics. An important development in the fertility model included an altruistic motive. Becker’s (1974) “altruistic model” or collective choice assumes an altruistic parent and egoistic but rational children. In the “consumption” motive of fertility, parents call on demand for services that children are able to provide. Consumption of these services generates pleasure for parents, i.e. increases their lifetime utility. The altruistic model adds an additional source of utility, which is altruism. This factor matters for a household head who takes pleasure in their own altruistic behaviour towards children.

In another theory, a child is perceived as an investment good and a tool for smoothing consumption over the life cycle. Expenditures on children’s education require earlier saving, and saving is perceived as delayed consumption. That is why this theory of fertility is linked to the retirement income motive, as parents want to have children to ensure their own income (from children) when they retire and no longer work. Parents, to combine present and future income optimally, choose an appropriate number of children. The theory of children as an investment good has a long tradition and dates back at least to Mill; however, it was also later mentioned, for example, by Friedman or Neher (Cochrane 1975). There is some empirical research that aims to verify whether parents perceive children more as consumption or investment goods, taking also the generosity of public social security into account, and it seems that the latter view finds more support in the existing literature (see for review: Billari and Galasso 2009; Danzer and Zyska 2020; Fenge and Scheubel 2014; Rossi and Godard 2019; Shen et al. 2020).

Mother–worker conflict and state involvement

Not only micro- but also macroeconomic factors matter for fertility, including governmental family policy in particular. The fundamental premise behind this is the conflict between women as mothers and women as workers, which potential parents face when making a fertility decision. The desire for or need of labour participation acts as a constraint on fertility. On theoretical grounds, Rindfuss and Brewster (1996) expect that any easing of this conflict should result in a rise in fertility, providing that other factors remain unchanged (ceteris paribus rule). This hypothesis is based on rational expectations of parents who are assumed to make their fertility decisions aiming at maximisation of their lifetime utility. Thereby, parents account for both the possible costs and benefits of childbearing and behave in line with the microeconomic theories mentioned above, i.e. perceiving children as consumption or investment goods. As Gauthier (2007) notes, referring to Becker (1981) and Cigno (1991), neoclassical economic theory predicts that any support from the government resulting in the reduction of childbearing costs or income growth should cause an increase in fertility. State family policy may be more familistic or defamilistic in its nature. The former is based on the idea of the involvement of family members in caring for children, which is identified with the “freedom of the family”. The latter eases this involvement and supports the idea of “freedom from the family” (Leitner 2003; Lohmann and Zagel 2016). As women usually deliver care services within a family, the notion of defamilisation is sometimes narrowed to the notion of degenderisation. Some scholars even postulate the replacement of defamilisation with the notion of degenderisation (Saxonberg 2013), but some others convincingly argue that they are irreplaceable (Kurowska 2018). Nevertheless, in the context of family policy aimed at fertility stimulation, women should be placed at the central point as they personally face the motherhood–labour participation conflict. However, this conflict affects the male partners as well, since its solution determines the distribution of childbearing duties among parents. Familising (or genderising) family policy embraces means to encourage parents to organise the childcare within the family, usually involving mothers as caregivers which results in a decrease in the labour supply. Defamilising (or degenderising) policy engages the state in offering public services that reduce family dependencies and facilitate the parents’ (mainly mothers’) return to the labour market. Familising or defamilising policies may also refer to the involvement of grandparents in taking care of their grandchildren, which is important in the context of our study. Familising policy supports this involvement whereas defamilising policy does not. As a result, the former may act in a degenderising manner towards young mothers (replaced by grandparents in their childcare responsibilities), whereas the later, contrarily, may enhance genderising trends within a family.

Going beyond pure economic motives of fertility decisions

Further developments of theoretical explanations in this field exceed the bounds of economics and incorporate social and psychological contexts to demonstrate how complex fertility decisions are. These are, for example, life course cube concept (Bernardi et al. 2019) or the theory of planned behaviour (Ajzen 1991). The former accounts for the distribution and interplay between resources investment in various life domains, i.e. a professional career may affect the timing of fertility. Succeeding in professional life may be a substitute for the satisfaction or utility derived from family life on the one hand, yet on the other stable and lucrative employment may be a precondition for starting a family and having children (Huinink and Kohli 2014). The later one highlights the individual’s intention as the main driving force of behaviour reducing the role of rationality simultaneously. Intentions reflect motivational factors that cause people to take various actions (Ajzen 1991), and as some empirical studies demonstrate, it relates to fertility as well (see for review: Caplescu 2014). However, large disjunctures between intentions and behaviours in this life domain can be also observed (Morgan and Bachrach 2011).

A multigenerational context

A multigenerational context can be also very important for fertility analyses at least for two reasons. The first one is an intergenerational continuity in fertility that can be explained by “transmission of values and preferred family size” or “continuities in education and socioeconomic status” (Kolk 2014). The other one, more important in the context of our paper, is the support from grandparents with childcare that may affect fertility decisions. Both intrafamily relations and macroeconomic policy may moderate this nexus. Nevertheless, in light of neoclassical economic theory, if any reduction in the cost of children or any increase in income due to having children increases fertility ceteris paribus, living in a multigenerational household with one or more older people acting as grandparents should result in an increase in fertility. The reason is that grandparents can provide a household with free care services, which, if bought on the market, are costly. Even if the state provides families with such public services for free or at a reduced cost, or subsidises such services delivered by the private sector, not all families are covered by this state support. Thus, intuition suggests that the support of grandparents facilitates parents in making fertility decisions. However, on the other hand, the multigenerational nature of a household may be the effect of some budget constraints of adult children, who cannot afford to rent or buy a house. Then, living with their aged parents is not a question of choice but necessity, caused by their economic situation. The question is whether such circumstances encourage or discourage couples to have children. The scenario in which the economic dependence or overcrowding problem may reduce the positive effect of living with aged parents, or even reverse it, should not be rejected a priori.

Do grandparents matter when their adult children make fertility decisions? The vast body of empirical literature delivers a positive answer to this question. Multigenerational co-residence, i.e. the generation of adult parents, their parents and their children living in one home, decreases childcare spending and increases spending on education (as compared to two-generational households). However, these differences vary across types of households and are moderated by parents’ income and relationship status (Amorim 2019). Kaptijn et al. (2010) refer to “the cooperative breeding hypothesis” which predicts that a wider set of kin stimulates female reproduction, which is strongly embedded in our evolutionary history. In line with this view, they argue that grandparents increase the chance of having children. Based on data gathered among Dutch grandparents, they ask about the frequency of taking care of their grandchildren and find that grandparents delivering childcare services enhance fertility; however, this is alongside the accessibility of formal childcare. They finally state that fertility needs the compatibility between motherhood and work. This conflict, mentioned above, requires more gender equity at the institutional level, which seems to be obvious (e.g. access to work, equal salaries), but also, which is less obvious, within a family (McDonald 2000). The grandparents’ engagement in childcare supports this equity, acting as a degenderising factor (with reference to young mothers) enabling them to return sooner to the labour market. This confirms some other research which demonstrate a positive impact of grandparents acting as caregivers for their grandchildren directly (Rutigliano and Lozano 2022; Tanskanen and Rotkirch 2014) or indirectly (Okun and Stecklov 2021).

Although these are in the minority, there are also some studies suggesting more caution regarding the positive impact of co-residence with older parents on their adult children’s fertility. Using Bulgarian data, Ghodsee and Bernardi (2012) demonstrate that in the case of some countries—especially post-Soviet states—the nexus under investigation can be more complex and moderated by circumstances not observed in Western Europe. What causes Bulgarian factors behind low fertility to be different are the socio-economic side effects of the collapse of communism around 1990 as experienced by Central and Eastern European countries. The public childcare system then experienced a deep deterioration and families were, and very often still are, forced to buy such services in the market, from private institutions or private persons. Many young couples cannot afford it due to income constraints. Therefore, they have to cohabitate with their parents after the birth of their first child. For the same economic reason, they often refrain from having a second child, as they regard moving out and buying or renting their own home as a priority. This finding justifies our question of whether a multigenerational household always supports having more children. As housing conditions matter for fertility decisions (Atalay et al. 2021; Kulu and Vikat 2007; Stoenchev and Hrischeva 2023), young couples may weigh up the advantages of the grandparents’ support in childcare on the one hand, and disadvantages of co-residence and sharing the limited living space on the other.

Data and research procedure

Data

We test the nexus between the multigenerational character of a household and the number of children living there, controlling for other socio-economic household attributes. We employ data from Wave X of the Luxembourg Wealth Study (LWS 2022) comprising the years 2016–2017. Our data set covers the following countries: Austria (data for 2017), Germany (2017), Estonia (2017), Spain (2017), Finland (2017), Italy (2016), Norway (2016), Slovenia (2017), Slovakia (2017), Luxembourg and Greece (2018—Wave XI, as data from Wave X for these two countries are not available in accordance with LIS Cross-National Data Center terminology, see at: https://www.lisdatacenter.org/our-data/lws-database/).

Our sample consists of 117,140 households with a head aged 20–45 years old, reflecting approximately women’s reproductive years. As multigenerational households in our sample are classified those in cases, where (1) a household head (the person with the highest income) is aged 20–45 years old, and (2) at least one household member is aged 65 years or older and is being paid pension benefits.

Models and variables

To test the relationship between the multigenerational character of a household and the number of children, we employ a multinomial logistic regression. Using this model, one can estimate the probability a household with given characteristic counts a given number of children expressed as one of the category of dependent variable (Y = 1, 2 or 3) as compared to the referential category (Y = 0). If a dependent variable has K categories (j = 1, 2, …, K), K − 1 binomial logistic regressions are estimated with the r category as a referential one for each model:

$$\text{ln}\frac{P(Y=j)}{P(y=r)}={\beta }_{j0}+{\sum }_{i=1}^{m}{\beta }_{ji}\bullet {x}_{i}$$

Models are estimated independently for two variables: Mem13- and Mem17-, which reflect the number of household members aged 13 years or under and 17 years or under, respectively. Naturally, data on the number of children aged 17 or less include children aged 13 or less. Thus, there can be a kind of redundancy in terms of the information covered by the explained variable. However, these two variables are treated independently in the estimation process, so they do not interact in our empirical procedure. On the other hand, the comparison of the estimates obtained for the two dependent variables is a kind of robustness check for the stability of the parameters (mainly their sign and statistical significance). Although some difference in the estimates between models for Mem13- and Mem17- are naturally expected as the former age interval is wider, which should be reflected in the differences in probability estimates (and is consistent with respective frequencies, see Figs. 1 and 2), possible substantive discrepancies in this respect should be carefully investigated. They could suggest an important change in the relationship studied, caused by a significant impact of the number of children aged 14–17 on the results.

Fig. 1
figure 1

Source: LWS database

Distribution of households (by country) in terms of number of children aged 13 or under.

Full size image
Fig. 2
figure 2

Source: LWS database

Distribution of households (by country) in terms of number of children aged 17 or under.

Full size image

Although the number of children is a measurement on the ratio scale, we transform it into an ordinal one: 0 for no children, 1 for one child, 2 for two children and 3 for three or more children. There are two reasons for this approach. First, it allows us to estimate multinomial logistic regression models, as they are not suitable for ratio scale-dependent variables. Second, the distribution of the number of children clearly shows that across the countries studied, households with three or more children are definitely less frequent than households with zero, one or two children. (The only exception is Slovenia.) Thus, using a ratio scale would result in a very low frequencies for the household number with three, four or more children as treated separately.

The tested factor reflecting the multigenerational character of a household is the residence of at least one person aged 65 + who is not a household head (binary variable: 0 for no members aged 65 +  and 1 for one or more such members). The set of control variables includes the demographic and socio-economic characteristics of a household or its head: age, gender, marital and labour force status, homeownership, income and public or private transfers, country of residence and education. The predictor and control variables are characterised in Table 1.

Table 1 Explanatory variable specifications.
Full size table

The survey country samples vary noticeably in LWS. Hence, to ensure their applicability to cross-country analyses, two sets of weights are employed: normalised household weights (hwgt) and household weights (hpopwgt). hwgt weights are household-level, cross-sectional weights that ensure the normalisation to 10,000 by country. As a result, each country’s weight is the same in a cross-sectional sample. hpopwgt weights are population household cross-sectional weights which are used to reflect in a sample the cross-national variation in population sizes of countries covered by a data set. Models estimated with the use of hwgt weights are our main models, as we aim to ensure cross-country comparability, not to reflect the cross-national variation in population sizes in our sample. However, we estimate models with hpopwgt weights to enrich our results by the comparison of two different estimate procedures as well. If in the case of two types of weights the parameters estimated are similar, one can conclude that the change in the sample structure, i.e. the replacement of samples of similar sizes (10,000 by country) by those reflecting population sizes, does not change the overall picture of the nexus investigated. This would suggest that an increase (decrease) in the role of a given country in the data set does not affect the results since countries are quite similar in terms of the analysed relationship. Consequently, the family or other public policies or some cultural or social factors probably do not differentiate households from different countries in terms of the translation of a multigenerational household into the number of children. Different estimates obtained for these two types of weights would suggest that such factors are of varying significance across countries. Despite both dependent variables measured on an ordinal scale, we refrain from the use of an ordinal logistic regression as the hypothesis of parallel regressions (i.e. similar parameters for the regressions for all the categories of the dependent variable) was rejected by the Chi-squared test of parallel lines. Consequently, we finally estimate multinomial regression models.

Results

The distribution of fertility

Figure 1 shows that although distributions of the number of children aged 13 or under in households across European countries is similar, some variation can be observed. This variation matters. Spain and Slovakia are countries with a relatively low (44% and 40% respectively) proportion of households without children. Simultaneously, in their cases, the highest proportions of households with one (29% and 32%) and two children (22% and 23%) are reported. Another country slightly outlying from the whole set is Slovenia with the lowest proportion of households with two children (11%) and with the highest share of households with three or more children (14%). In other countries constituting a great majority of those studied here, households without children are predominant (over 50%, in the cases of Finland and Norway, which even has over 60%), with similar proportions of about 20% of one-child and of two-child households. All countries, with the exception of Slovenia mentioned above, have a proportion of households with three or more children lower than 7%. In the case of the distribution of households in terms of the number of children aged 17 or under (Fig. 2), two similar outliers can be identified—Spain and Slovakia. Slovakia, in the case of this distribution, reports more households with one child than without any children (34–29%). In the case of Spain, similarly to Slovakia, a significant difference in the ratio of households with one and two children is observed, whereas other countries report quite similar proportions for these two variants. Slovenia, as previously stated, dominates in terms of households with three or more children. The distribution of the number of children (mainly Fig. 2) confirms that “a two-child ideal” still holds (Sobotka and Beaujouan 2014).

Descriptive statistics of explanatory variables

We also present some descriptive statistics for the predictors and controls; means for quantitative variables and frequencies for qualitative variables (see Table 2). The data show that multigenerational households including grandparents and their adult children are infrequent, but this varies across the countries studied from less than 1% to over 9%. A household head is aged 32–38 in average terms and in Austria, Germany, Greece and Slovenia it is slightly more often a woman than a man. Norway is the only country where living without a partner is more common than with a partner. The variation in homeownership rates across the country samples is noticeable (between 26% in Germany and 82% in Slovakia) which is a consequence of very different housing regimes, from those with a predominance of renting to those mainly based on homeownership. Only 4.5% of Italian households in the sample receives any public transfers, whereas in Scandinavian countries representing the socio-democratic welfare state regime this ratio exceeds 70% (75% in Norway and almost 80% in Finland). Additionally, Finnish households receive support in the form of private transfers the most frequently.

Table 2 Descriptive statistics for explanatory variables across countries.
Full size table

Estimation results

Generally, the models estimated (see Tables 3 and 4) do not support the view that intergenerational households, in which grandparents (members aged 65 +) live, include more children aged 13 or under, or 17 or under. In the case of models for the number of children aged 13 or under (Mem13-), in both cases of weights used—normalised household weights (hwgt) and household weights (hpopwgt)—a household member aged 65 or older decreases the probability of having three or more children as compared to the referential category which is having no children. In the case of the main models, i.e. incorporating hwgt weights, for both dependent variables (Mem13- and Mem17-) in households with elderly residents, having a child or two children is not more, or even less likely than not having children at all (odds ratio > 1 only for two children aged 13 or under, however, for p-value = 0.226, so the parameter is statistically insignificant). In the case of models with hpopwgt weights, an older person (or people) living in a household increases the probability of having one (for Mem13-) or two children (Mem13- and Mem17-) as compared to the referential category. The reason behind this is that countries such as Spain or Italy significantly increase their share in an overall sample due to hpopwgt weights as compared to hwgt weights having a relatively high ratio of households with one or two children. The parameter estimates next to the variable pension confirm a negative relationship between a multigenerational household and the number of children living in it. Namely, in households with income from a pension system, having one, two, three or more children is less likely as compared to the referential category, which is having no children.

Table 3 Multinomial logistic regression results for Mem13-.
Full size table
Table 4 Multinomial logistic regression results for Mem17-.
Full size table

The differences between models estimated with two types of weights but also the odds ratios for country-specific variables that account for some unobservable factors undoubtedly suggest that country-specific conditions, including family policy or social and cultural factors, affect and differentiate fertility. Mediterranean countries like Slovenia or Italy see an enormous increase in the probability of having two, three or more children compared with other countries. To this set, Luxembourg and Norway should also be added.

The estimates by other variables deliver some interesting findings. All the models clearly indicate that a woman as a head of household increases the probability of having children. Models with hwgt weights demonstrate that the probability of having one child or having two children is over 80% (for Mem13-) and almost 130% (for Mem17-) greater as compared to the referential category for households with woman as the head (which in some cases can be assigned to the household member reporting the highest incomeFootnote 3). For the variant of dependent variable “three or more children”, this probability is over 50% and over 120% greater, respectively. This finding indicates that it is very important to support families when seeking the remedy for the motherhood–work conflict as women succeeding in professional life, which is reflected in their household position, among others, greatly enhances the chance for having offspring. What also increases the chance of having children is a partner, which is much more important than marital status. This indicates that a real foundation for establishing a family, i.e. being a couple, matters more than whether this relationship is formalised. Moreover, the positive impact of a partnership increases alongside the number of children in a household. The probability of having three or more children when having a partner is over 13 (15) times greater than not having a child at all in the case of the model for Mem13- (Mem17-). Employment of the household head also increases the chance of having one, two, three or more children. Similarly, homeownership matters. The probability of having more children increases alongside households’ disposable net worth but not with equivalised income. This all means that the economic situation of a household is a driving force for having children; however, it is not the present income but (accumulated) wealth that matters. Not only having a job, but also owning residential property impacts positively the chance of having any number of children. Both public and private transfers also matter and support fertility. However, the stimulative force of the former is incomparably stronger. Public transfers increase the probability of having three or more children over 17 times in the case of dependent variable Mem13- and over 36 times in the case of Mem17- variables in comparison to the referential category. This implicitly shows the difference in overall public support addressed to households without and with children. As for education, the models report an ambiguous impact on the probability of having children. The main (i.e. hwgt) model for Mem17- suggests that low education increases and medium education decreases the chance of having one or more children as compared to high education. These interpretations of the models estimated are generally consistent with cross-tabs for combinations of dependent and respective independent variables.

Robustness of the results

When estimating the models using the same sample and hwgt weights, we have used different combinations of variables. The set of explanatory variables which was fixed covered Mem65 + , Age, Sex, Merital_status, Home_own, Income, Worth, Publ_transfers and Priv_transfers, Country and Edu. Nevertheless, we estimated the models in several combinations, which included (or excluded) the following additional explanatory variables: Emp, Partner, Pensions as well as some variables not used in the final models presented in Tables 3 and 4, such as Nearn (number of household members with labour income) or Year (year of data). We have additionally reduced the database only to households where the head lived with a partner and estimated models with Mem65 + , Age, Sex, Emp, Income, Publ_transfers, Private_transfers, Country and Edu variables. In all the models, the odds ratio for Mem13- equal to 3 or more and for Mem17- equal to three or more were lower than one and statistically significant for p-value < 0.1 (excluding two cases with p-value < 0.124). For the number of children equal to one or two, the consistence was weaker, however, still satisfactory. The comparison between final models estimated with hwgt and hpopwgt weights as well as between models for Mem13- and Mem17- (see Tables 3 and 4) also indicates some consistence, especially with reference to odds ratios for the number of children no less than three. As for divergence between these models, we refer to it when discussing the results and drawing final conclusions.

Discussion and conclusions

We posed the research question whether a multigenerational household (i.e. co-residence of grandparents and their adult children) facilitates the formation of multichild families. Focusing on the households with three or more children, our models do not support this view. None of the four models estimated delivers arguments that grandparents living in the same home increase the probability of having at least three children. In the case of models estimated with hwgt weights, a multigenerational family does not support fertility at all. When we change the sample structure and reflect the cross-country diversity in terms of population sizes (hpopwgt weights), a multigenerational family facilitates having one or two children as compared to childless households, but not having three or more children. Simultaneously, the estimates demonstrate that the country of a household’s residence matters, which may suggest that cultural or social factors of fertility or family policy play a role. In our models, Slovenia reports the highest odds ratios for the highest category of the dependent variable. In the case of this country, the probability of having three or more children is from over three to almost six times greater (depending on the model) than in the case of the referential category which is Slovakia. Figures 1 and 2 confirm the highest share of households with the number of children being three or more in the case of this country. When comparing these findings with the results obtained by Chybalski and Marcinkiewicz (2021) when studying European welfare regimes with the incorporation of pro-family and pro-female components, one can discover an interesting coincidence. Namely, Slovenia is the country with the highest value of a synthetic measure of a pro-female component among the countries investigated in the present paper (with a medium level of pro-family component). A pro-female component includes such characteristics as for instance, the gender pay gap or employment rate, whereas a pro-family component accounts among other things for the length of various leave allowances dedicated to parents after their child’s birth or children’s enrolment in early childhood education. The correlation analysis of country-level data used by Chybalski and Marcinkiewicz (2021) with the estimates obtained in this paper allows us to associate our results to both family policy and gender equality in the labour market across the countries studied. This comparison shows the following. Firstly, the case of Slovenia suggests that the number of children equal to three or more is most frequently observed in the country with the highest level of gender equality in the labour market. Thus, the country that supports women in solving the motherhood–work dilemma seems to benefit from it as it reports more multichild families, which is crucial for generation replacement. Greece is on the other end of the spectrum and reports households with three or more children the least frequently, having the lowest value of pro-family synthetic indicator (and a medium level of pro-female component) in the study by Chybalski and Marcinkiewicz (2021). Further, a more in-depth analysis of the results obtained in the two papers mentioned indicates that countries with the higher ratio of households counting no fewer than three children also have higher participation rates of children aged 6–11 years in centre-based out-of-school-hours care services, a lower employment rate gap and higher maternal employment rates (the Pearson correlation coefficient is statistically significant for p-value = 0.10). Moreover, they report greater pro-female component synthetic indicator values.

The mentioned coexistence is consistent with the findings by Eibich and Siedler (2020) as it suggests that what matters for stimulating growth in the number of multichild families is not encouraging older people to exit the labour market to stay in multigenerational homes and take care of grandchildren. It also does not mean that grandparents’ support is not important for fertility stimulation. Hence, we do not undermine the vast amount of research that demonstrates that grandparents’ assistance works as a driving force for fertility (Okun and Stecklov 2021; Rutigliano and Lozano 2022; Tanskanen and Rotkirch 2014). Our study suggests only that co-residence in a multigenerational household can be a barrier to having more than two children as it may cause, for example, a problem of overcrowding. Co-residence may also not be a question of choice, but a necessity caused by economic constraints (young couples unable to afford buying or renting a home). Therefore, the hypothesis that multigenerational households are a panacea for the shortage of multichild families does not find support in our empirical investigation. The results deliver some initial arguments in favour of the view mentioned in the literature that for fertility stimulation, assisting families and women in coping with the motherhood–work conflict is substantial (Kaptijn et al. 2010; McDonald 2000; Rindfuss and Brewster 1996; Rutigliano 2023). This includes the assistance of grandparents as well. However, this kind of support does not mean co-residence as assistance can be delivered by grandparents forming a separate household. Thus, among various actions taken in order to increase fertility, any support for women to facilitate the combination of both motherhood and labour market participation is useful.

Although the goal of our study was not to solve the child as investment versus child as a consumption good dilemma, our results can be also related implicitly to this issue. We compared the obtained odds ratios for country variables with the present aggregate replacement ratios as calculated by EurostatFootnote 4 and with projected gross pension replacement rates as calculated by OECDFootnote 5 from 2017 (to ensure time consistency with the data used in our study). The Pearson’s correlation coefficients for the mentioned cross sections are modest and positive for both of the used replacement rates and odds ratios in case of the number of children equal to one or two. For three or more children, the correlation is definitely weaker. Thus, for countries with higher pension replacement rates (both present and projected), the odds ratios of having one or two children as compared to the referential category (Slovakia) are also higher. This is hardly observed for the number of children equal to three or more. So, in case of our data set, more generous pension system does coexist with a greater chance of having children. The relationship is reversed as one could expect in the world where parents perceive children as an investment good and have them to build economic security for old age.

In our models, we focus on the number of children aged 13 or under and 17 or under and we do not know whether grandparents were living in the same household when a child or children were born. We know that they share the same home at the moment of data collection. This can be perceived as a conceptual limitation of our study as we do not link co-residence directly with the fertility decision. On the other hand, however, grandparents’ support may be useful for the young parents not only directly after a child is born but also later, in the preschool or school period when grandparents may provide transport, meals, educational support or other care services. Thus, our view is wider as we generally account for grandparents’ support, no matter at which stage of childhood (or youth) and for what number of children. Our study accounts for co-residence, which is a binary phenomenon. However, sometimes grandparents can live in the closest vicinity, or they can live quite far from their adult children, even in a different country or continent. Geographical proximity can matter here, and our variable does not capture it. Thus, this can be regarded as a limitation of the study, which results from data availability constraints. Our study also has some methodological limitations. First, the models estimated do not allow direct inference in causal terms. This is why we formulate our conclusions with some caution and attempt to refer them to appropriate theoretical grounds or results of previous empirical investigations. Second, there is a risk of omitted variables that is common not only in case of regression analyses but generally in any empirical studies on various relationships. To mitigate this possible bias and its consequences for the endogeneity problem (omitted variables may cause endogeneity), we not only tried to include a wide range of control variables, but also accounted for possible unobservable country-level factors through the inclusion of country variables.

The results of our study may inform policy in a few aspects and indicate, similarly, for example, to Hoem (2008), that actions taken to support fertility or the formation of multichild families should be multidimensional. The estimates obtained demonstrate an importance of homeownership or public and private transfers in this regard. The support for women to solve the motherhood–worker conflict is crucial. The family policy implemented by the government may be very helpful in this regard as degenderising and defamilising components of such policy as well as any labour protection guarantees or incentives to return to the job after childbirth enhance the perception of economic security not only of women but of their partners as well. This can support the mitigation of negative consequences of demographic crisis on two fronts as it influences both fertility and labour supply.

One of the results we have obtained, i.e. the positive relationship between the role of a woman in the household (i.e. being a head which may reflect an income status across household members) and the probability of having more children, highlights the need for further studies on the issue of degenderising policies and their impact on fertility. In the present paper, we address this finding only as a “side effect” of testing the main research hypothesis. However, we perceive it as a very interesting and potentially important from policy perspective.