Remaining childless: examining the different patterns of expectations that lead to a childless life for men

Early life fertility expectations fail to explain why a substantial proportion of men remain childless. To understand which roads lead to childlessness, it is important to investigate the patterning of fertility expectations across men’s life courses. Therefore, the current study focuses on pathways into childlessness for men, by differentiating groups based on whether, and if so when, changes in fertility expectations occurred. In addition, we examine how these patterns of fertility expectations were linked to men’s labour force participation (LFP) and marital relationships. We use data from the American National Longitudinal Survey of Youth and applied sequence, clustering, and multinomial logistic regression methods to analyze stability and change in childless men’s life course between the ages of 22 and 47 (N = 679). Based on their patterns of fertility expectations, we identified four groups of childless men. (1) Early switchers (29%), (2) mid-thirty switchers (29%), (3) late or never switchers (25%), and men with unstable expectations (18%). Early switchers often got married early, while late or never switchers were more likely those who got married at a later age. In contrast, interrupted LFP was only weakly related to patterns of fertility expectations. Our longitudinal focus revealed, firstly, that different roads of fertility expectations lead to childlessness for men. Second, changes in men’s partnership status often precede changes in their fertility expectations. This study is the first to use a prospective design to investigate childless men’s complete reproductive history, revealing that different trajectories of fertility expectations lead to childlessness.


Introduction
In many Western countries, a substantial portion of men remain childless even though they expected to become a father. In the US for example, at age 23, 95% of men indicated that they expected a child someday, while only 78% of them actually made the transition to fatherhood (Berrington and Pattaro 2014). These findings show that early articulated fertility expectations do not suffice when one wants to understand why men remain childless. This may not come as a surprise, given that during a man's life course, behavior and circumstance might change that could influence his fertility expectations and subsequently his childbearing behavior. During one's life, situations may change, and things may happen, that lead to adaptations of one's fertility expectations; Long-term romantic relationships may end, and financial situations may fluctuate, for example. However, and we perceive this to be quite startling, most research investigating the antecedents of childlessness among men, does not take a longitudinal perspective when investigating childlessness among men. In the current paper, therefore, we move beyond earlier research on the role of men's fertility expectations in explaining childlessness among men, by investigating, first, if and when childless men change their fertility expectations during their life course, and second, how these patterns of fertility expectations can be explained.
For women, in contrast, scholars have shown the importance of investigating changes in fertility expectations to understand which individuals remain childless. Already the pioneering work of Houseknecht (1979) and Veevers (1973) distinguished between early articulators, women who decided to remain childless early in life, and postponers, women who realized at a later stage in their life course that they would not make the transition to motherhood and adjusted their fertility expectations accordingly. Also more recent studies that focused on patterns of fertility expectations over the life course of childless women showed that there are different pathways into childlessness (Gemmill 2018;Rybińska and Morgan 2018). In addition to two groups already distinguished by Houseknecht and Veevers, they found a third group of women who adjusted their fertility expectations multiple times throughout their life course.
There is no previous research on patterns of fertility expectations preceding childlessness among men. This would not have been a problem if we were able to extrapolate the findings from women to men. Unfortunately, however, that is not possible, given that women have much stricter biological and social age deadlines related to family formation than men (Billari et al. 2011;Harris et al. 2011), and given the fact that for women the presence of a stable relationship is a much less strong prerequisite to becoming a parent than it is for men. Therefore, we need to investigate the patterning of fertility expectations explicitly for childless men as well.
Besides mapping the patterning of men's fertility expectations, we set out to explain these patterns as well. Little is known about the factors that relate to different patterns of fertility expectations for men. Previous studies that focus on factors that explain men's fertility expectations are cross-sectional (Fiori et al. 2017;Miettinen 2010;Miettinen and Szalma 2014) or only focus on a part of men's life course, and do not include information about eventual childlessness (Berrington and Pattaro 2014;Buhr and Huinink 2017;Gray et al. 2013;Heaton et al. 1999). Nevertheless, from these studies we do know that partnership characteristics and labour force participation (LFP) are important predictors of fertility expectations. In the current study, we therefore investigate how timing concerning their marriage and instability in LFP are linked with different patterns of fertility expectations.
Up to the best of our knowledge, the current study is the first to prospectively investigate men's fertility expectations throughout their entire reproductive span. Our first aim is to gain a nuanced insight into the patterns of change and stability in men's fertility expectations that lead up to permanent childlessness. Our second aim is to examine how these different patterns of fertility expectations are related to men's LFP and romantic relationships.

Different patterns of fertility expectations
In the first part of this paper, we will examine to what extent different patterns of stability versus change in men's fertility expectations can be distinguished among men who remained childless. Our starting point is the existing knowledge that we have on these trajectories for women, and we couple this knowledge with specific information on men's reproductive span. As abovementioned, when scholars have made recent typologies of childless women based on their desires and expectations, they often made a distinction between early articulators, postponers, and women with unstable fertility expectations (Gemmill 2018;Rybińska and Morgan 2018).
For men, we expect to identify four different groups based on the fertility expectations they expressed during their fertile years. First, similar to what is found for women, we expect to be able to identify early articulators; a group of men who never expressed the expectation of having a child, or who very early on in their life course switched from expecting to become a father to no longer having this expectation. Previous cross-sectional research on men's fertility expectations showed that there is a group of men who already at quite young ages (18-24) do not expect to become a parent (Miettinen and Paajanen 2005). Furthermore, an exploratory study that used interviews with 30 voluntary childfree men also described a group of early articulators, who already early in their life were sure they wanted to remain childless (Lunneborg 1999).
Second, for men we expect to identify a group of childless men who keep expecting to become a parent up to their thirties but eventually expect to remain childless-we label this group mid-thirty switchers. One of the possible reasons could be that these childless men compare themselves to their peers who are becoming fathers. In the US in the period we study, the average age of becoming a father is around 25 years (Stykes 2011), and the vast majority of men who make the transition to fatherhood do so before the age of 35 (Pudrovska and Carr 2009). In the 1st years after their peers became fathers, childless men will likely still expect to become a father, but after a while, they might realize or perceive that they have missed a social deadline to start a family. We therefore expect to identify a group of men in our sample who will adjust their fertility expectations around age 35.
Contrary to women, we expect that there is also a group of men who keep expecting to become a father until they are in their late forties, this group we label as late or never switchers. We base this expectation on differences in biological and social age deadlines between men and women. Although the biological ability to have children decreases with age for men, this is much less steeply the case compared to women (Harris et al. 2011) and men generally have their first child a bit later than women (Khandwala et al. 2017). Also the social age deadlines -the age after which it is socially less accepted to become a parent-are less strict for men than they are for women (Billari et al. 2011). In addition, even in their forties, men could start a relationship with much younger women, which may provide them with a second chance to start a family, allowing them to keep expecting to become a father until relatively old ages. Previous research on men and their fertility expectations showed that at the beginning of their forties still quite some childless men expect to become a father in the future (Berrington 2004;Kessler et al. 2013). A qualitative study among childless also men described a group of men who kept expecting to become a father up to high ages (Lunneborg 1999).
Furthermore, we expect to identify a group of men who change their expectations multiple times in their life course, which we will label as men with unstable expectations. We formulate this expectation because men in particular are dependent on the presence of a partner to have children and the presence of a partner may change during men's life course. Furthermore, given the importance that many men and women attach to having the financial means to raise a child, and in particular the extent to which men can fulfill their role of breadwinner, we expect that changes in men's LFP may also lead to adjustments of men's fertility expectations. Therefore, we expect that men are likely to change their fertility expectations based on the absence or presence of a partner and a job. Heaton et al. (1999) found that in a time frame of 6 years already 19% of the men change their desire to become a father once. Therefore, it is likely that during the complete reproductive life span a part of the men changes their mind multiple times.
Finally, we expect that there will also be a group of men who remain childless because of their own (or their partner's) biological infertility. In previous studies, estimates of the size of the group who remain childless due to infertility range from 3 to 26% (Kreyenfeld and Konietzka 2017;Tanturri and Mencarini 2008). Since the data that we use to answer our research questions does not contain information on infertility, we cannot identify men who remain childless because of infertility, which is a limitation of our study. This implies that our inability to identify these men may somewhat blur our categorization of groups of men based on their patterns of fertility expectations throughout their fertile years. We return to this issue in the "Discussion and conclusion" section.
To summarize, we expect a group of (1) early articulators, men who either never expressed the expectation to have children or already at a young age switched from expecting to have children to not expecting this, we expect a group of (2) mid-thirty switchers, men who in their thirties change from expecting to have children to not having this expectation, (3) late or never switchers, men who change their mind when they are in their forties or over, and (4) we expect to find a group of men who changed their expectations several times during their reproductive span.

Labour force participation and marriage throughout their reproductive span
Although no previous research has specifically focused on different patterns of fertility expectations that precede childlessness among men, several studies have focused on explaining the expectation or desire to remain childless, and on explanations for childlessness among men. In the following sections, we draw on expectations based on these empirical findings, in combination with different theories that can be linked to fertility expectations, to form hypotheses on how the timing of marriage relates to the hypothesized four patterns of fertility expectations (early articulators, mid-thirty switchers, late or never switchers and those with unstable expectations).

Marriage and the timing of marriage
Based on the life course perspective, in particular the concept of cumulative contingencies, which indicates that not only the current situation but also all of the previous experiences shape behavior in the life course (Elder et al. 2003), it can be argued that the timing of romantic relationships is important for understanding stability and change in men's fertility expectations. It is not only important for predicting patterns of fertility expectations if a man currently has a partner, but also when he started a romantic relationship.
First, there is reason to assume that men who are single for a large part of their life will keep expecting to become a father for a long time, which will make them late or never switchers. This might sound counterintuitive at first since men often consider having a stable romantic relationship and having a suitable partner as a requirement for starting a family (Roberts et al. 2011;Testa 2006). However, when considering theories of goal adjustment there is reason to assume that men who are single for a large part of their life keep expecting to become a father. According to the life-span theory of control (Heckhausen and Schulz 1995), people try to achieve their goals first by trying to influence their circumstances (primary control), in this case finding a partner. Secondary control, which is the adjustment of goals, is assumed to happen when people approach their developmental age deadline (Heckhausen et al. 2001). However, due to the long reproductive life span of men (many men become fathers when they are over the age of 40 (Harris et al. 2011)), it is possible that men keep expecting to find a partner and become a father up to high ages, and do not apply secondary control. Therefore, we expect that men who never got married or never got partnered (as opposed to all other men) are most likely to be late or never switchers (as opposed to all other patterns) (hypothesis 1a).
There are reasons to assume that men who got married early will be most likely mid-thirty switchers. The first reason for this expectation is related to the age of the partner. For women, age is a very important predictor of the expectation to become a parent and the expectation of women decreases more steeply than for men (Berrington 2004). Because partners very often have similar fertility expectations (Berrington 2004), men who have a partner likely adjust their fertility expectations downwards more steeply than single men. A second reason is that since being in a romantic relationship is such a strong predictor of having children, it might be the case that couples who are together for a longer period, but who do not have children yet are a select group. For example, one or both of the partners may not have the desire to make the transition to parenthood. Alternatively, the couple could be experiencing difficulties in conceiving a child, and at some point give up on their expectations to have a child. Thirdly, it has been argued that having a stable romantic relationship is an alternative to having children for life fulfillment among men (Buhr and Huinink 2017). Several studies support the idea that men who are in a romantic relationship decrease their fertility expectations more strongly than single men. For example, Liefbroer (2009) shows that (taking into account the number of children people have), being married increases the expected number of children. However, this effect decreases with age; at age 28 being married positively associates with fertility expectations while from age 40 onwards the effect becomes negative. Also Kessler and colleagues (2013) show that at somewhat higher ages (between 30 and 44 years), men who are living with a spouse or partner are less likely to desire to become a father than single men. This would lead to an expectation that men who got married early (as opposed to all other men) are most likely to be mid-thirty switchers (as opposed to all other patterns) (hypothesis 1b).
For men who found a partner at later ages (for example when they are in their late thirties or their forties), we anticipate that fertility expectations will be adjusted not too long after they start a relationship. This is for the same reason that is mentioned above, namely that men adjust their fertility expectations in sync with their partners. Because men generally get married to women who are quite close to them in age (on average 2-3 years younger than they are) (England and Mcclintock 2019) we expect the men in this group to switch from expecting children to not expecting children not too long after they started their romantic relationship. This will make them fall into the category of late or never switchers, although we specifically expect them to be late switchers and not never switchers. This will lead to the expectation that men who got married late (as opposed to all other men) are most likely to be late or never switchers (as opposed to all other patterns) (hypothesis 1c).

Stability in labour force participation
According to most men, having a stable career and having financial security are seen as requirements for starting a family (Roberts et al. 2011;Testa 2006). The importance of financial security is understandable, since in the cohort of men born around 1960 (which we look at in this study), one of the main roles of the father was still being the breadwinner who provides financially for his family (Lamb 1987). Not only current employment but especially being stably employed over a longer period likely increases the financial security that men experience. Given the cumulative character of occupation careers, men who did not have a stable job for long periods are less likely to find a stable job in the future. Therefore, men who are jobless for long periods in their life likely realize they are not in the required position to start a family and will adjust their fertility expectations. Another reason why long-term unemployed men would be more likely to adjust their fertility expectations downwards, is marital conflict. The reasoning is that financial insecurity is related to more conflict and lower relationship quality, which in turn also reduces the expectation to have children (Berninger et al. 2011). Finally, long-term unemployed men might also adjust their fertility expectations downward given that these men are considered unattractive marriage material (Fieder and Huber 2007) due to which they are more likely to remain single and without children. This would lead to the expectation that men who have interrupted employment early in life are most likely to be early or mid-thirty switchers (as opposed to all other patterns) (hypothesis 2a). Furthermore, men who have interrupted employment later in life are most likely to be late switchers (as opposed to all other patterns) (hypothesis 2b). Furthermore, in line with these hypotheses, previous cross-sectional studies showed that men with interrupted occupational careers were more likely to remain childless (Keizer et al. 2008). In addition, previous cross-sectional studies showed that unemployed men are more likely to intend to remain childless (Berrington and Pattaro 2014), are more likely to change from expecting children to not expecting children (Buhr and Huinink 2017) or to adjust their fertility desires downwards (Gray et al. 2013).

The importance of taking fertility desires into account
A suggestion from the literature is that some men are avoiding a relationship to be able to postpone or refrain from parenthood (Jensen 2016). They are a select group who does not want to settle down and start a family. For these men, there would not be a causal link from the romantic relationship to fertility expectations but rather there is selection at play: men who do not want to have children, are both more likely to remain without a partner or only have short relationships, and are more likely to express that they have no intentions to have a child.
This selection issue may also apply to men's LFP. It could be the case that men who do not wish to start a family, do not see the need for stable income as much as men who desire to become a father. Men who do not desire to have children could both be more likely to have unstable LFP and be more likely to express that they intend to remain childless. To obtain an accurate understanding of linkages between men's patterning of fertility expectations and their patterning of changes in partnership characteristics and LFP, in the current study, we control for the desire to have children as expressed early in the life course, which is measured between age 18 and 22.

Current study
In the current study, we examined the patterning of fertility expectations of men who have remained childless and examined how LFP and romantic relationships relate to these patterns of fertility expectations. We use the National Longitudinal Study of Youth (NLSY79) from the United States (Bureau of Labor Statistics U.S. Department of Labor 2019), which measured fertility expectations, romantic relationship status, and LFP yearly and later biyearly, from men's early twenties to their fifties. We use sequence and clustering techniques to categorize stability and change in fertility expectations. In examining the relationship between these factors, we control for static characteristics of men (and when applicable their partner), such as their highest level of education, ethnicity, and religion, as well as the desire to have children.

Data
In our study, we used the National Longitudinal Survey of Youth 1979 (NLSY79), which follows the lives of 12,686 individuals (6403 men) born between 1957 and 1964 (Bureau of Labor Statistics U.S. Department of Labor 2019). Respondents were first interviewed in 1979 when their ages varied between 14 and 22. Respondents have been interviewed subsequently every year until 1994, and every 2 years after that. The last round of interviews that we used took place in 2016 when the respondents were between 51 and 60 years old. For the current study, we only selected men over the age of 50 at the time of their last interview (not all men completed interviews up to 2016), because previous research shows that only 0.6% of the children are born to fathers after the age of 50 (Martin et al. 2011). For this reason, childlessness at the age of 50 years and older will be very similar to lifetime childlessness. This selection resulted in a reduction of the sample size from 6403 to 3837 men with an average age of 55 during their last interview. We selected the childless men in our sample, a total of 806 men (21%). The sample dropped to 737 because 69 men had missing information on their fertility expectations (see section "Measurements"). The sample was further reduced with 58 cases due to missingness on desired childlessness between age 18 and 22, and missing information on characteristics of the partner, yielding a final sample of 679 respondents. Due to the sampling procedure, the NLSY79 includes an overrepresentation of poor black and Hispanic individuals. Because of this selection of men over 50, and to adjust for the sampling procedure of the NLSY we applied custom weights from the NLSY that gives weights according to the probability to be followed up to the age of 50 as well as the initial sampling procedure.

Measurements
Fertility expectations Using sequence analyses with Optimal Matching and Ward cluster analysis we categorized different sequences of fertility expectations (Mills 2011). These analyses were conducted in the TraMineR package and were assessed with the WeightedCluster packages in R (Gabadinho et al. 2011;Studer 2013). We based these sequences on answers the respondents gave on all the waves between age 22 and age 47. We used information from the question that was asked every wave about how many children the respondents expected to have. We made the distinction between men who expected to have no children (expected childless) and men who expected one or more children (expected children). The insertion and deletion costs for the cluster analyses are 1, and the substitution costs are based on the transition rates between states in the data, resulting in a cost of 1.66 from expecting a child to not expecting a child and vice versa, and a cost of 2 from one of these two states to missing. Using sequence analysis, a five-cluster solution was chosen because it allowed meaningful interpretation (see Fig. 1). This five clusters could be described as (1) early switchers, men who started off expecting to have children, but switched their expectations before 30 years of age, (2) later switchers, who started off expecting to have children, but switched their expectations halfway through their thirties, (3) late or never switchers, who expected to become a father well into their forties and only late in their fertile years or men who never changed their expectations, (4) men with unstable expectations, who changed their expectations multiple times throughout their fertile years, and (5) men who had a lot of missing information on their expectation (and who were removed for further analysis). The average silhouette width (ASW) of this cluster solution was relatively low, 0.30, while a high ASW indicates high distances between groups and strong homogeneity within groups and an ASW is 0.5 or higher is recommended (Hennig and Liao 2013). When looking at the groups in detail, it appears that especially the ASW of cluster 5 (which was removed in the analyses) and of cluster 4 (the group with unstable preferences) are low (respectively 0.14 and − 0.09 for the missing and the unstable expectations). This means that in the group of men with unstable expectations there is a lot of variability in sequences. In the other clusters the ASW is higher (respectively 0.4, 0.38, and 0.48 for the early switchers, mid-thirty switchers, and late or never switchers). Comparing the ASW of this 5-cluster solution to any cluster solution ranging from 2 to 7 separated clusters shows that only the 2-cluster solution has a higher ASW of 0.33. However, this 2-cluster solution only distinguishes between those already expect to remain childless around age 30 (the early switchers) and all other men, which is theoretically less relevant. Therefore, we stick to the five-cluster solution, in which we removed the last cluster of men with many missing information (n = 69).
Regarding marriage and the timing of marriage, we use the information on the age of their first marriage, which is categorized into before 21, between 21 and 25, between 26 and 30, between 31 and 35, and 36 and over. On top of that, there are separate categories for men who never married nor partnered, men who were in a relationship but never married, and men who were married but the age at which they entered their first marriage is unknown.
For unstable labour force participation (LFP), in every wave, respondents were asked how many weeks they worked in the past year. Responses were categorized into 0, 1-13, 14-26, 27-39, 40-48, 49-51, and 52 weeks. We counted the number of waves between age 20 and age 48 in which respondents indicated they worked 0 weeks. Given that weeks spent out of the labour force (keeping house, going to school, unable to work, other) and weeks spent in the armed forces were counted as weeks not worked, we adjusted our variable based on information on whether our respondents were (a) in full-time education and (b) in the armed forced. Unfortunately, however, in the years 2000-2004 and from 2008 onwards the questions on being in full-time education and being in the armed forces were not asked and information on these two activities is therefore missing. In the years 2000-2004 and from 2008 onwards, we therefore could not rely on information about enrollment in full-time education for creating our interrupted LFP measure. Consequently, in these years, we might have slightly overestimated unemployment if respondents were actually enrolled in full-time education. However, our inability to correct might not pose a major limitation, given that the men in our sample almost all completed their educational attainment by 2000. For handling missings on being in the armed forces, we decided on the following strategy: When a respondent indicated in 1998 that he was in the armed forces, and he reported to have worked zero hours in 2000-2004, we have assigned him a 0 for unemployment in 2000-2004, assuming that this respondent was still in the military. Similarly, when a respondent indicated in 2006 that he was in the military, and he reported to have worked zero hours in 2008+, we have assigned him 0 for unemployment in 2008+, assuming again that this respondent was still in the military. Consequently, given this operationalization, we might have slightly underestimated unemployment in these years. We recoded unstable labour force participation into four categories, (1) men who never had a year in which they did not work between the ages of 20 and 48, (2) men who indicated at least 1 year in which they did not work between the ages of 20 and 34, but who worked continuously between the ages of 35 and 48 (3) men who indicated at least 1 year in which they did not work between the ages of 35 and 48, but who worked continuously between the ages of 20 and 34 and (4) men who indicated that there was at least 1 year in which they did not work between the ages of 20 and 34 and at least 1 year in which they did not work between the ages of 35 and 48.
Controls In our models we control for the desire to remain childless, which was asked in 1979 and 1982. To reduce the age difference between respondents on when they reported their desires, we used information from the wave in which they were between 18 and 22 years old (would we only use the 1979 wave, respondents would be between 14 and 22, and only using the 1982 wave would result in an age range from 17 to 25). We use the highest completed Educational level for which we distinguish between "not finished high school," "high school," "some college," "BSc," and " > BSc." If respondents ever cohabited with a partner or were married, we include the educational level of their partner and age difference with their partner (if respondents answered this question about multiple partners or in multiple waves, we used the answer they gave when they were 30 years of age, or the closest wave to 30 years in which they answered this question, because this is the age at which men are most "at-risk" of becoming a father). Robustness checks revealed that similar results were obtained when we incorporated characteristics of the partner at the last available waves. We also include religion (distinguishing between Protestant, Roman Catholic, no religion, or other) and ethnicity (white, black, and Hispanic).

Statistical techniques
To examine which life course characteristics were associated with which patterns of fertility expectations, we applied multinomial logistic regression analysis (Kwak and Clayton-Matthews 2002). We used the early switchers as the reference group. We obtained the explained variance of our models by calculating the reduction in the log likelihood of the model compared to the empty model (McFadden pseudo Rsquare). To assess the relative importance of each of the predictors in our models we looked at the explained variance of the model when only that specific variable was included in the model. To be able to compare different groups of fertility sequences (e.g., not only compare early switchers to late or never switchers, but also those with unstable expectations to mid-thirty switchers) as well as to compare the association with timing of marriage and LFP, we graphically displayed predicted probabilities and calculated post hoc tests for differences between predicted probabilities [as one cannot draw conclusions about the significance of differences between groups by only visually inspecting confidence intervals (Cumming and Finch 2005)].
We based our predicted probabilities and post hoc tests on the complete multinomial logistic regression model that includes all control variables. The predicted probabilities for the timing of marriage and interrupted LFP that are displayed in the figures and used for the post hoc tests are marginalized over all other levels of factors in the model. They are set fixed at the reference groups for religion and ethnicity, because otherwise the number of combinations of factor levels was too large. For these marginalized predicted probabilities, we used weights in proportion to how frequent the factor combinations were present in our data. We used the nnet package for multinomial logistic regression (Ripley and Venables 2022) and we used the emmeans package (Lenth et al. 2023) to calculate predicted probabilities, with 95% confidence intervals around these predictions and post hoc tests.

Descriptives
In the introduction, we anticipated finding four groups of men. Firstly, we expected to find early articulators, who already early in their life course expected to remain childless. Largely in line with this expectation, however not completely similar, we found a group of early switchers, who switched from expecting to become a father in their twenties to expecting to remain childless before 30 years of age (see Fig. 1). We label these men early switchers because they all made the switch from expecting to not expecting to become a father. Contrary to our expectations, almost all of these men did have the expectation to become a father in their early twenties-only 6% of them never reported the expectation to become a father between age 22 and 47 (see Online Resource Table A1). Around 29% of the men in our sample fall in this category of early switchers. Secondly, we expected to find a group of mid-thirty switchers and we indeed found that around 29% of the men could be considered mid-thirty switchers. Thirdly, we expected to find late or never switchers, and we indeed found a group of men who expected to become a father until they were well into their forties and who only very late in their fertile years or never switched to a childless expectation. This group comprises of 26% of the childless men. Of this group of late or never switchers, almost a third expected to become a father in every wave (see Online Resource Table A1). These findings are in line with the idea that men keep expecting to become a parent longer than women do. Finally, 18% of the men in our sample had unstable expectations, and these men on average changed their expectations about having children three times throughout their fertile years (see Online Resource Table A1).
Considering the characteristics of these groups of men (see Table 1), our results revealed that the early switchers were much more often married early in comparison to the other groups, desired to remain childless more often, were a bit lower educated, and more often had a partner with the same age or older than them. In comparison to the other groups, the mid-thirty switchers often found a partner or got married a bit later in life, were more often higher educated and had a younger partner. The late or never switchers remained single most often, were a bit higher educated and had most often uninterrupted LFP. Those men with unstable expectations did not clearly stand out from the other groups of men in any aspect.

Hypothesis testing
The role of marriage and timing of marriage Using multinomial logistic regression, we examined how never getting married and the timing of marriage relate to patterns of fertility expectations. Results are displayed in Table 2 and Fig. 2. We added this figure to facilitate comparisons between the subgroups of men, and we furthermore calculate post hoc tests for differences between predicted probabilities, which are displayed in Online Resource Table A2a. This Figure and Table show that those men who got married early (before 21) are significantly more likely to be early switchers than late to never switchers, but do not significantly differ from mid-thirty switchers and men with unstable expectations. Men who married between 21 and 25 are significantly more likely to be early switchers than late to never switchers and they are also significantly more likely to be early switchers than men with unstable expectations. They do not significantly differ from mid-thirty switchers. The same pattern is found for men who married between 26 and 30; They are significantly more likely to be early switchers in comparison to late to never switchers and men with unstable expectations. They do not significantly differ from mid-thirdly switchers. Furthermore, men who got married later (36+) were significantly more likely to be mid-thirty switchers, although only in comparison to men with unstable expectations. Men who married between the ages of 31 and 35, men who never married but did enter a romantic partnership (only partnered) and men who married at an unknown age did not stand out in terms of a specific fertility expectations profile. Comparing our findings across marital age categories, the post hoc tests displayed in Online Resource Table A2b furthermore show than men who got married before 21, between 21 and 25, or between 26 and 30 are more likely to be early switchers than men who got married at age 36 and over and compared to men who only partnered. In turn, men who got married before age 21 are less likely to be late or never switchers than men who got married between 31 and 35, men who got married at age 36 and over and men who only partnered.
To summarize, the predicted probabilities (Fig. 2) reveal a clear pattern of fertility expectations depending on age of marriage. As the age of marriage increases, the predicted probability of being an early switcher steadily decreases while the predicted probability of being a late or never switcher increases. The turning point, where the probability of being a late or never switcher surpasses the predicted probability of being an early switcher happens in the group of men that married between 31 and 35 years of age. There is no clear pattern apparent for mid-thirty switchers or men with unstable expectations.
Linking these findings to our initially formulated hypothesis, firstly, we do not find support for hypothesis 1a, that men who never got married/partnered are most The total number of men in the sample is not equal to the sum of the 4 clusters represented, because a fifth cluster with men who had a lot of missing information is not presented in this table likely to be late or never switchers. That said, although never married men do not have the relatively highest predicted probability to be late to never switchers, they are more likely to be late or never switchers compared to men who married before the age of 21. Secondly, we expected that men who entered marriage early would be most likely to be mid-thirty switchers (hypothesis 1b), but instead, we found that men who got married before the age of 21 are actually more likely to be early switchers, with a predicted probability of almost 60%, albeit only in comparison to being late or never switchers. Thirdly, we expected that men who started a relationship later (at age 36 and over) would be late or never switchers (hypothesis 1c). Our findings do not provide support for this hypothesis.
To try to obtain a better understanding of cause and effect in the association between (timing of) marriage and childbearing intentions, we graphically displayed the association between age at first romantic relationship and age at first expectation to remain childless in Fig. 3. We draw a diagonal line, and dots on this line would represent men that stop expecting to become a father at the same age as when they start a relationship. All dots above this line represent men who stop expecting to become a father after they start a relationship, and dots below this line represent men who stop expecting to become a father before they start a relationship. This Figure shows that many men switched to expecting childlessness in the years after they started a romantic relationship, while very few already did so before they started a romantic relationship. Figure 3 also adds some nuance to the interpretation of our findings in relation to hypothesis 1b. Our findings might actually not be too different from our hypothesis given that the men in our sample do not change their expectations prior to the beginning of a relationship. Instead, they change their expectation rather quickly after the beginning of their romantic relationship, while we expected there to be a larger period between the start of a relationship and changing expectations.

The role of labour force participation throughout the life course
In our multinomial logistic regression model, we also examined how interrupted LFP in different periods in life relates to patterns of fertility expectations. Results are displayed in Table 2 and Fig. 4, and the results of the post hoc tests are displayed in Online Resource Table A2c. These results show that interrupted LFP (either early -between the ages of 20 and 34-, or late -between the ages of 35 and 48-, or both) is only weakly linked with our four patterns of men's fertility expectations. Our results show only a few associations with our patterns of men's fertility expectations; men who had an interruption in their labour force participation later in life (between age 35 and 48) are more likely to be mid-thirty The explained variance is calculated as the reduction in the log likelihood compared to the empty model (McFadden's pseudo R 2 ). The proportion correctly classified in the early switchers group (reference group) is 0.563. *p<0.05, **p<0.01, ***p<0.001

Fig. 2
Patterns of fertility expectations by the timing of marriage. The error bars represent 95% confidence intervals. Estimates are based on the multinomial logistic regression model displayed in Table 2 Fig . 3 The relationship between the age at which men started their first relationship, and the age at which men for the first time in two consecutive waves did not expect to become a father. If men for the first time at age 22-23 and age 24-25 indicated they did not expect to remain childless, this is categorized as 22. The colors indicate the different groups of childlessness expectation sequences. All dots below the dashed horizontal line represent men who did not expect to become a father before they started their first romantic relationship, while all dots above the dotted horizontal line represent men who did not expect to become a father after they started their first romantic relationship. The late or never switchers are rarely present in this graph because very few of them had a partner and very few of them expected to remain childless in two consecutive waves switchers than late or never switchers and men with unstable expectations. Furthermore, men who never had an interruption in their LFP are more likely to be early switchers or mid-thirty switchers in comparison to men who have unstable expectations. Comparing our results across the different LFP categories, our findings furthermore show that men who have had an interrupted LFP early in life are less likely to be mid-thirty switchers and more likely to be late or never switchers in comparison to men who had an interruption in their LFP later in life. These findings do not lend support for hypothesis 2a, that men who had interrupted LFP early in life are most often mid-thirty switchers and they also do not lend support for hypothesis 2b, that those who had interrupted LFP later in life are most often later switchers.

The role of our control variables
Findings for our control variables show four things. First, men who expressed a desire to remain childless have higher odds of being early switchers, especially compared to men who were late or never switchers. Second, men who did not obtain their high school qualification have significantly lower odds of being mid-thirty switchers than being early switchers compared to men with a high school qualification. Men who finished some college, on the other hand, have significantly higher odds of falling into any  Table 2 other category of fertility expectations compared to early switchers. There are no statistically significant differences for men with a BSc or higher education level. Thirdly, the characteristics of the partner play a role, as the relative odds of being a late or never switcher versus an early switcher increases significantly if having a partner with any education level other than high school (or not having a partner at all) compared to having a partner with a high school degree. This difference is especially pronounced for men with highly educated partners (> BSc), who have 8.1 times the estimated odds of being late or never switchers rather than an early switcher compared to men with a partner who only finished high school. The relative odds of being a late or never switcher or having unstable childlessness expectations increases significantly if having a younger partner rather than having a partner the same age. On the other hand, having an older partner significantly decreases the odds of being a mid-thirty or late or never switcher compared to being an early switcher. There are no significant differences by ethnicity or religion.

Relative importance
When added separately to the model, the timing of marriage variable explained 2.9% of the variance, compared to only 1.2% when we would add a simple measure that indicates if a man ever had a partner. The unstable LFP variable explained only 0.06% of the variance. Next to the explained variance by our variables of interest, the characteristics of the partner also explained a substantial part of the variance; the age difference with the partner explained 4.8%, and the educational level of the partner 4.4%. To obtain the latter two estimates, we examined the increase in model fit when we already had included the variable on the timing of marriage. This is needed because otherwise the abovementioned estimates would be inflated due to information on the presence of a marital partner. Furthermore, the difference in the desire to remain childless explained 4.9% of the variance. Education, religion, and ethnicity each explained approximately 1.5%. These findings underscore that the romantic relationship and characteristics of the partner are the most important predictors of the patterns of fertility expectations. In the full model, the explained variance is 20%.

Discussion and conclusion
Previous research showed that many men remain childless despite the intention to become a father early in their lives. Therefore, in this study we used a longitudinal approach to examining if, when, and why men change their expectations. The aim of this study was twofold, the first aim was to gain insight into patterns of fertility expectations that precede childlessness among men, and the second aim was to gain an understanding of how the timing of marriage and LFP relate to these patterns of expectations.
Regarding our first aim, we identified four different groups of childless men. A group of early switchers, a group of mid-thirty switchers, a group of late or never switchers, and a group of men with unstable expectations. Based on previous research on categories of women and on fertility desires and expectations of men, we also expected to find a small group of men who never had the expectation to become a father (Gemmill 2018). In our sample, we only found a negligible number of men (who fall in our group of early switchers) who did never expect to become a father. This indicates that never expecting to become a father is even rarer than was expected. Future research should find out if in more recent cohorts, due to the increasing acceptance of childlessness (Koropeckyj-cox and Pendell 2007), more men express to be intentionally childless already at young ages.
Furthermore, in the group of late or never switchers, there were quite some men who at the end of their forties still expected to become a father. This group of late or never switchers is not previously found among women, indicating that the larger time span in which men can have children influences their expectations. Our finding that many men who remain childless do so despite expecting to have children for a large part of their life is worrisome. Previous research showed that the consequences of childlessness on health and well-being depend on among others socio-economic status, marital status, and race (Umberson et al. 2010). No previous studies examined how the pathway of expectations preceding childlessness relates to well-being. However, for those men who keep expecting to become a father the consequences for well-being are likely more detrimental.
Regarding our second aim, we found strong evidence that the timing of marriage is related to different patterns of fertility expectations. Our results show that finding a partner is often followed shortly by switching to the expectation to remain childless. This finding is in line with studies that show that at higher ages having a partner is related to lower fertility expectations (Kessler et al. 2013;Liefbroer 2009). Our findings complement insights derived from cross-sectional studies, showing that, with samples that included men who still might make the transition to parenthood, having a partner increases the expectation to become a father (Gray et al. 2013;Heaton et al. 1999). Taken together, these results suggest that different mechanisms might be at play for different groups of men: among men who will have children, starting a relationship is often a predecessor to having children. Among the select group of men who will remain childless, starting a relationship is instead related to reducing fertility expectations.
There could be different possible explanations for why partnered men who remain childless shift their expectations to remaining childless earlier than single men. A first explanation is that men with a partner are constrained by the age of their partner, and either due to biological age deadlines experience infertility or due to social age deadlines consider it too late to have children (Billari et al. 2011;Harris et al. 2011). This explanation is supported by the findings that men with an older partner are more likely to be early switchers. Single men might presume that they can have children up to very old ages. An alternative explanation might be that the preferences of the partner are highly influential on men's fertility expectations. Finally, it might be the case that men may only find out about their own or their partner's infertility soon after starting to make family plans with a partner.
Altogether, our findings suggest that the partner, either directly or indirectly, has a strong influence on the change in men's fertility expectations, which is in line with previous research showing that men are very responsive to the fertility desires of their partner (Bauer and Kneip 2014;Testa 2012). We encourage future research to investigate the mechanisms that underlie the link between the entry into a partnership and men's change in fertility expectations.
Although many previous studies showed that men with an unstable occupational career remain childless more often (Keizer et al. 2008; Nettle and Pollet 2008), we did not find strong evidence that these men differed from men with stable careers in their patterns of fertility expectations. Nevertheless, we did find that men who had an interruption in their labour force participation later in life (between age 35 and 48) were more likely to be mid-thirty switchers compared to being late to never switchers. This finding suggests that men's labour force participation might be a consequence rather than a precursor of their fertility expectations. It might be the case that when men no longer expect to have children, they also see less necessity for obtaining financial stability. Previous studies found that jobless men have lower fertility desires, or that these men adjust their fertility desires down (Berrington and Pattaro 2014;Buhr and Huinink 2017;Fiori et al. 2017;Gray et al. 2013). Based on our findings and also some of the limitations of our interrupted labour force participation measure, we encourage scholars to replicate our findings and to scrutinize the causal mechanisms that underlie the link between men's interrupted labor force participation and change in fertility expectations.
Although the focus of this paper was on changes in the domains of romantic relationships and LFP, we find some interesting results regarding our control variables, and in particular for men's own and their partner's educational attainment. Among women, previous research shows that higher educated women often had the expectation to have children for a large part of their life, but due to postponement eventually remained childless (Verweij et al. 2021). Our findings do not show a similar pattern for men's own educational attainment, as there were no statistically significant differences for men with a BSc or higher education level. That said, our findings did reveal that men on the other end of the educational attainment spectrum, those with the lowest level of educational attainment, have significantly lower odds of being mid-thirty switchers compared to early switchers. Finally, our findings for men with a highly educated partner fully resemble the patterns found in the literature for highly educated childless women.
In interpreting these findings, several limitations should be kept in mind. A shortcoming is that we do not have information on infertility. Theoretically, men who experience infertility could fall into most groups of childless men that we distinguish. They could find out early in life and therefore be early switchers, but they could also find out later in their life, for example after they found a partner and attempted to have children for a (long) while, and therefore could be mid-thirty switchers. Also, the late or never switchers could be infertile but be unaware of this. Integrating medical information in social science surveys or integrating questions on desires and expectations in medical surveys could potentially provide more insights in future research. Such research designs would also allow for investigating whether the witnessed changes in fertility expectations after entry into a partnership are caused by infertility. Another shortcoming of the current study is the lack of information on the fertility expectations or desires of the partners of the men in our sample. A combination of information on infertility and fertility expectations of both partners could provide more insight into why men who enter a new partnership start expecting to remain childless.
Finally, our results are limited by the fact that we reduced variability between the men in our sample by creating four different patterns of fertility expectations throughout men's life courses. Although this variability is relatively low in the groups of early switchers, mid-thirty switchers, and late or never switchers, variability is quite substantial in the group of men with unstable expectations. While some men in this group change their expectations almost every few years, others have especially unstable expectations in their thirties or only in their forties. Therefore, results regarding this group should be interpreted with caution.
To conclude, this study is the first, to the best of our knowledge, that used a prospective design to assess the complete history of men's fertility expectations and to derive different patterns of fertility expectations. In addition, we used information on the timing of marriage and interruptions in LFP to explain the patterning of men's fertility expectations. This work highlights that there isn't one single road that leads to a childless life. Furthermore, we find evidence that men change their fertility expectations downwards after entry into a romantic relationship, which underscores that women's behavior, desires, and expectations have a strong influence on men's fertility behavior. Finally, our findings highlight that a focus on 'the' childless man does not do justice to reality; there are very different roads that lead to a childless life for men.