DO NOT CITE WITHOUT PERMISSION
Timing of susceptibility to parental absences depends on life history outcome and gender Cristina Moya, Anna Goodman, Ilona Koupil, Rebecca Sear Abstract: Many researchers have proposed that children have sensitive periods of development during which environmental instabilities like parental absences have particularly marked effects on their life history strategies. These developmental windows are often justified by reference to physiological constraints, but their relevance to other reproductive and socio-economic measures is less clear. Furthermore, the literature has focused primarily on the effect of father absences and therefore has not fully explored functional differences depending on the sex of the parent-offspring dyad. Here we use a large longitudinal database of individuals born in mid-20th-century Sweden to investigate the extent to which mother and father deaths at different ages have long-term consequences for children’s reproductive and economic strategies. We find some evidence that children experience sensitive periods during which maternal deaths have larger effects on their age at first birth, fertility, education and income. On the other hand, fathers' deaths during childhood and adolescence only have pronounced effects on daughters' ages at first birth. We further find that most of these effects are likely to be causal given that socio-economic covariates act as negative confounders, meaning that these parental absence effects are likely to be even larger if we could control for more unmeasured variance in family background. However, the pattern of sensitive periods that we find does not support a critical cut-off before age 7 for girls' life history strategy development, changes across cohorts and defies simple interpretations across domains and sex of parent-offspring dyads.
INTRODUCTION Premature parental losses have important consequences for a range of outcomes during development and adulthood (Cas, Frankenberg, Suriastini, & Thomas, 2014; Case & Ardington, 2006; Smith, Hanson, Norton, Hollingshaus, & Mineau, 2014; Tennant, 1988; Willführ, 2009). This is not surprising given the importance of parental care in a species such as humans with long periods of childhood and adolescent dependence (Gurven, Kaplan, & Gutierrez, 2006; Hill, 1993). More specifically, evolutionary social scientists have argued that reduced parental investments or parental availability can change a developing child's optimal life history strategy – i.e. how they allocate resources between reproduction (e.g. earlier maturity) and embodied capital (e.g. somatic growth, immune functioning, education) Several researchers have further argued that parental absences have particularly strong effects on life history outcomes if they happen during specific critical periods of development in childhood (Alvergne, Faurie, & Raymond, 2008; Draper & Harpending, 1982; Sheppard & Sear, 2012). One proposed mechanism for early childhood sensitive periods of development is the fact that endocrinological changes related to puberty begin around age 7 (Ellis, 2004). This developmental milestone therefore constrains a child's ability to markedly change energetic allocations in later adolescence. Such a developmental constraint makes sense for tightly linked physiological outcomes, but it is less clear why they should necessarily affect the development of other behavioral strategies that may be more plastic. Many models of parental absence effects on children's life history development assume sensitive periods in early childhood during which individuals assess their local socio-ecology (Belsky, Steinberg, & Draper, 1991; Chisholm, 1993; Draper & Harpending, 1982; Ellis, 2004) or their internal state and available resources (Ellis, 2004; Rickard, Frankenhuis, & Nettle, 2014), and adapt their reproductive and investment strategies accordingly. However, very few of these models, or of other 1
DO NOT CITE WITHOUT PERMISSION accounts of parental absence effects on life history strategy (Matchock & Susman, 2006; Moya & Sear, 2014; Surbey, 1990), really hinge on specific critical periods of development. Although recent theoretical work has made important contributions to our understanding of how evolution should design sensitive periods of development, (Frankenhuis & Panchanathan, 2011; Giudice & Belsky, 2011), it is still difficult to derive clear adaptationist predictions about the domains in which we should expect sensitive periods and about their specific ranges. Therefore, we argue for a partly datadriven approach to examining critical periods of development in the life history domain. Gender of parent-offspring dyads Most of the literature on life history development has focused on effects due to father, rather than mother, absences. This is likely due to the greater prevalence of paternal absences due to both death and divorce, and because the consequences of mother’s death during childhood are known to have negative and immediate consequences (Sear & Mace, 2008). Some studies comparing fathers and mothers directly have suggested stronger developmental acceleration effects of father absence (Bogaert, 2005; Sheppard, Garcia, & Sear, 2014). Similarly, the literature has primarily focused on female adolescents’ development for a series of methodological, policy, and theoretical reasons – menarche is easier to measure than adrenarche, female teen pregnancies are seen as the larger social problem (Card & Wise, 1978), and the tradeoffs between early reproduction, and somatic and educational investments seem starker for females (Ellis, 2004). However, direct comparisons are rare and generally there seem to be few clear patterns regarding the interaction of parent and child sex on developmental outcomes (Russell & Saebel, 1997). Some models of parental absence effects on life history strategy may differentiate between maternal and paternal effects. For example, models focusing on inbreeding avoidance should only predict paternal absence effects for daughters, and perhaps maternal absence effects for sons maturation (Matchock & Susman, 2006). Models focusing on the direct role of parental investments in modifying optimal resource allocation should predict larger mother absence effects given the larger role of women in raising children in most societies (Ellis, 2004; Rickard et al., 2014). However, most models that focus on father absence effects are ambiguous about why the sex of the absent parent should matter (Belsky et al., 1991; Draper & Harpending, 1982; Moya & Sear, 2014). Furthermore, none of the models explicitly predict different sensitive periods for different kinds of parental absence. Life history outcomes Life history approaches to human behavior often treat several behaviors like early maturation, high fertility, promiscuity, impulsivity, reduced educational investment as if they were components of a single adaptive strategy (Belsky et al., 1991; Simpson et al., 2012). However, the theoretical grounds for assuming that these are a set of correlated responses to the same socio-ecology are often ambiguous. If parental losses are indications that investments in embodied capital are not worth it (either because of poor resources availability in the natal home or because of high mortality rates), these should most clearly affect reproductive tempo and educational attainment (Ellis, 2004). However, it remains an open question whether total fertility and other economic outcomes are also affected by parental losses during childhood. Causal role of parental loss The fact that the empirical tests for these sensitive periods are correlational means that they suffer from the possibility that unmeasured variance confounds the associations between parental absences and life history development. We know that in post-demographic transition countries socioeconomic status (SES) is positively associated with lifespan, with later reproduction (Burström, Johannesson, & Diderichsen, 2005; Geronimus, Bound, & Waidmann, 1999), and often with lower fertility, especially in women (A. Goodman & Koupil, 2009; Huber, Bookstein, & Fieder, 2010). This may lead to a spurious association between premature parental deaths, early ages at first birth (AFB) and high fertility due to correlated socio-ecologies between the generations. If the association between 2
DO NOT CITE WITHOUT PERMISSION early parental deaths and reproductive strategy resulted from such confounding we should conclude that direct parental care is not causally relevant to how children and adolescents develop their reproductive and economic strategies. Nor would such a pattern be consistent with the idea that parental presence specifically is serving as a cue to children about the environmental context in which they are developing. On the other hand, reproductive strategies are also intergenerationally transmitted meaning that young parents tend to have offspring who reproduce early (Stanfors & Scott, 2013). This means that young parents are less likely to die during their offspring's childhood, and more likely to have children who reproduce early. Such confounding may result in our underestimating the extent to which parental deaths expedite reproduction. In fact, several studies have used clever methods to show that parental absences have a causal role in expediting reproduction or maturation. These methods include controlling for family-level effects (Ermisch, Francesconi, & Pevalin, 2004; Tither & Ellis, 2008), examining the effect of being adopted by wealthy, but not biologically related parents (Teilmann, Pedersen, Skakkebaek, & Jensen, 2006) and analyzing the effect of parental absences resulting from more extrinsic circumstances such as war or natural disasters (Cas et al., 2014; Pesonen et al., 2008). Article goals and overview In this paper we address several of the gaps or problems in the literature that we have discussed above to examine the possibility of sensitive periods in the first two decades of life during which parental deaths may have particularly marked effects on life history strategy. First, we use both a data-driven and theory-driven approach to detecting critical periods of development. This means that we compare the plausibility of all possible critical periods of even length, and let the data suggest the most likely cut-offs. This provides a generous test of the presence of critical periods of development without having to pre-specify arbitrary or controversial age categories. We also directly test whether early and late childhood are sensitive periods during which parental losses have large effects, using age cut-offs that others have theorized are likely to be important. Second, we examine all possible parent-offspring dyads (i.e. maternal and paternal death effects on sons and daughters). This allows us to test whether father and mother absence effects have categorically different results, and whether sons and daughters are particularly responsive to these stressors. Third, we examine effects on four life course outcomes. This allows us to check whether the effects of parental absences on various behavioral outcomes occur during similar sensitive periods, thus suggesting a coherent developmental pathway. These measures include two reproductive ones (age at first birth and total fertility), and two socio-economic ones (university attendance and adult household income). AFB and university attendance correspond to early adulthood decisions that trade-off with one another, while fertility and income represent later decisions throughout adulthood. Fourth, while our study is also correlational, we examine the possibility of residual unmeasured variance being responsible for our results in two ways. We examine the effect of early parental deaths later in adulthood on earlier life events. For example, we examine whether people whose parents die after the index individual is 45 years old systematically differ in terms of university attendance and AFB, from those whose parents are still alive. If we find such associations even after controlling for SES covariates, this suggests that residual unexplained variance is plausibly partly to blame for earlier childhood parental absence effects. Additionally, we examine the extent to which adding covariates reduces the size of the parental effects to assess the extent to which unmeasured variance in family background may be a negative or positive confounder of the effects. Furthermore, we use a dataset that reduces the likelihood that unmeasured residual variation in intergenerational correlations drives parental effects. We investigate sensitive periods using data from the Uppsala Birth Cohort Multigeneration Study (UBCoS), an intergenerational and longitudinal dataset spanning most of the 20th century in Sweden. This dataset allows us to reduce the risk of unmeasured variance in two ways. It includes individual’s complete reproductive histories and several socio-economic indicators across two generations, allowing us to control for intergenerational reproductive and economic correlations to a much greater extent than most studies. Additionally, 20th 3
DO NOT CITE WITHOUT PERMISSION century Sweden represents a social context where both educational and reproductive goals are heavily subsidized by the state (Hoem & Hoem, 1997), meaning that parental deaths are likely to entail somewhat smaller disadvantages than they would in a society with a less well-developed social safety net. The cultural context provides a strong test of the hypothesis that parental presence causally explains variation in reproductive and socio-economic outcomes, given that the Swedish state from the early to mid 20th century onwards has provided much social assistance for individuals to attain their educational and reproductive goals. On the other hand, the context might represent one with relatively high rates of paternal care compared to most societies, making it more likely that we detect father absence effects. Finally, the intergenerational dataset allows us to examine critical periods in two cohorts so that we can assess the extent to which these sensitive periods of development are canalized and impervious to historical changes.
METHOD Participants We focus our analysis on the second generation (20,727 individuals) of the Uppsala Birth Cohort Study (UBCoS). We will refer to this as the index generation. These are children of the parent generation, a cohort of individuals born in Uppsala, Sweden between 1915 and 1929, who were followed throughout their lives to measure social and biological data of relevance to several health outcomes. The index generation of this multigenerational, longitudinal dataset represents the only cohort for which we have complete information about parental death dates and nearly complete information about their full reproductive lives. The index individuals were born between 1932 and 1990, but the bulk of them had reached 45 years of age by the last time they had been observed (only 7% had not). To test whether the effects of early parental deaths have changed in recent history, we analyze the life histories of the child generation (the children of the index generation, n=37,118) when possible. However all analyses refer to the index cohort unless otherwise specified.
Measures Outcomes: Reproductive We examine family background effects on two reproductive variables; age at first birth and total fertility (see Table 1 for distributions). To test hypotheses regarding reproductive pace we fit linear models predicting age at first birth (AFB). We include the all members of the index generation in the analysis since the vast majority would have progressed to this event (Figure S1). However, for the historical comparsion of the index and child generations we use event history analyses to deal with the greater likelihood that participants' reproduction is censored (i.e. they have yet to reproduce) in the more recent cohort (Figure S1 shows survival curves). To examine completed fertility outcomes we use zero-inflated poisson models to predict the number of children born given that they fit better than poisson and negative binomial models when fitting simple models predicting fertility from parental presence and basic covariates. The zeroinflation was driven by men, who were more likely than women to be childless, but to facilitate comparison we use zero-inflated poisson models for both sexes. Using data from the older members of the index generation who had completed their reproduction, we estimate that 99% of women and 95% of men had finished reproducing by age 45. Therefore, we restrict our analyses of completed fertility to index generation members who were at least 45 the last time their data was updated (the earlier of death, extended emigration outside the country or the end 2009).
Outcomes: Socio-economic We also examine two socio-economic outcomes. First, given that college education most directly trades off with onset of reproduction in post-industrial societies, and years of schooling is very left skewed, we use a binary variable of having ever attended college and fit logistic regressions. 4
DO NOT CITE WITHOUT PERMISSION Second, we consider household income between the ages of 21 and 64. Income is age and sex standardized by year and these z-scores are averaged from data collected between 1970 - 2008. We use the full index generation when examining educational and economic outcomes.
Predictors: Parental absences We only examine parental absences due to death as we have the best temporal resolution for these data. We use two strategies stacked in favor of finding sensitive periods to examine whether deaths during specific critical ages have particularly large effects on reproductive and socio-economic outcomes. First, we use a data-driven approach comparing models of all possible critical periods of even lengths within the first 20 years of life. That is, we create categorical variables of parental death happening between ages 0-2, 0-4, 0-6 ..., 2-4, 2-6, 2-8…,16-18,16-20,18-20; resulting in 55 possible critical periods (see Table 1 for distribution of age at parental death by sex). We run models predicting each outcome from each of the resulting binary variable of having experience a parental death during the given critical period. We also consider the possibility that parental absences are better modeled as having a linear effect within the first two decades of life. This would reflect the possibility that a child experiences an additional hardship for each extra year of parental absence that he experiences (i.e. for each year earlier that one's parent dies). In addition to this data-driven empirical search for sensitive periods, we examine specific theoretically relevant age-ranges during which parental deaths occur. We consider deaths by age 7, 715, 16-24, 25-45, after 45, and compare them to parents still alive. We choose these cut offs to correspond to several important life history events: developmental changes associated with puberty start around age 7 (Ellis, 2004), major decisions regarding university attendance start around age 16, the median age of first births is 25 for women in this cohort (and just 2 years later for men), and most adults have completed their reproduction by age 45.
Covariates One of the strengths of this dataset is that it allows us to control for several parental and family background reproductive and socio-economic variables that are known to be intergenerationally transmitted (Borgerhoff Mulder et al., 2009; Anna Goodman, Gisselmann, & Koupil, 2010; Murphy & Knudsen, 2002) and are possible confounds affecting both parental availability and ego’s reproductive outcomes. We include the mean of parents’ disposable household income metric (age, year and sex standardized), the maximum educational level of parents, and grandparental socioeconomic status as measured by occupational status in models (this last variable is only available for the index generation). For models of father presence we include father’s age at first birth and total fertility as predictors, and for mother presence and parental separation models we include the corresponding mother’s reproductive variables. This allows us to control for potential genetic confounds as well given that such reproductive outcomes are partly heritable (Pettay, Kruuk, Jokela, & Lummaa, 2005). Additionally, to control for historical changes within each cohort we control for birth bands of roughly 5-year intervals. Finally, given the non-independence of observations within a family we use robust standard errors, clustered by shared mother, or shared father if mother identity was missing.
Analyses The models we run vary by dependent variable as described above, and include all co-variates unless otherwise stated. For ease of interpretation we run models on female and male participants separately given their systematic difference in timing of first birth and educational attainment. The fact that mean fertility is lower for males than females in the sample (Table 1) suggests that UBCoS women were more likely to reproduce with people outside the index cohort than men were during this time period or that fathers are slightly more likely to be underreported on birth certificates (A. Goodman & Koupil, 2009) . We also run mother and father availability models separately as some individuals had data on one parent but not the other. 5
DO NOT CITE WITHOUT PERMISSION
Model comparisons of critical periods Our first goal is to test the more general hypothesis about the presence of sensitive periods of vulnerability to parental deaths. As such, we use analyses that are most favorable to detecting critical periods. Instead of pre-specifying age cutoffs where we would expect critical periods, we use a datadriven approach of model comparison. This allows us to determine which, if any, of the 55 critical periods of even length that we devised (described above) fits the data best. For each of these critical periods we ran models predicting each outcome from all the covariates and a binary variable of parental death during the corresponding age interval and calculate the model's AIC weight – i.e. the relative probability that the model is the best one out of all the ones being compared, given the data. AIC scores penalize models that have too many parameters relative to sample size. This means that there may be some critical periods during which parental deaths are estimated to have large effects, but that have low AIC weights because there are two few parental deaths observed in that age range. Because there are few parental deaths in our sample for some of the shorter sensitive periods tested (see Table 2) it is worth noting that there is no clear relationship between number of parental deaths within a critical period and its resulting AIC weight during the model selection process (See Supplementary Materials Table S1 and Section 2 for more details about the model selection process). Once we find the best fitting of all the critical period models, we compare it to a linear model of parental presence within the first 20 years of life (each year of parental presence has an additive effect), and a null model with all the same covariates but no parental presence indicator (Figure 1).
Estimating effect sizes We report unstandardized effect sizes of experiencing a parent's death during specific moments in childhood relative to having a parent alive. Figures 2-5 show these effect sizes for each possible 2-year age band from 0-20 during which a parent could have died, corresponding to the data-driven analysis. Table 3 shows these effect sizes derived from a single model with a categorical age at parental death predictor using theoretically motivated age cut-offs. Unmeasured variance as confound? Although the UBCoS dataset has a wealth of family background variables that we use as controls in our models, there is still a real possibility that any correlational results are not causal. We try to assess whether other unmeasured variables affecting both the likelihood of premature parental death and a child's later reproductive and socio-economic outcomes may confound the relationship between the two. We do this in two ways. First we compare the outcomes of individuals whose parents died after they were 45 years of age to those whose parents were still alive. If we find differences between these two categories we might surmise that they are partly due to residual unmeasured variance that confounds the relationships since these parental deaths are likely happening after most people have finished reproducing and attending university. Second we compare the effect of parental deaths in a full model with covariates to a simple model without covariates. If we find that adding family background covariates reduces the effect size of experiencing a parental death we may surmise that unmeasured variance in family background is partly responsible for any associations we do see in the full model. On the other hand, if there is no difference in effect sizes of parental deaths between the full and simple models we may more safely conclude that parental deaths are indeed having a causal effect on the child's life history. Finally, if we find the effect sizes increase when we add family background covariates, this suggests that we might even be underestimating the size of the effect of parental deaths in childhood because of unmeasured negative confounds. 6
DO NOT CITE WITHOUT PERMISSION
Historical changes Finally, to examine whether the critical periods we detect are robust and canalized, we run the same analysis on a more recent cohort from the UBCoS dataset. We can most readily examine progression to first birth as an outcome for this purpose given it's a relatively early life history event. We use discrete-time event history analyses to model the probability of progressing to a first birth any given year from age 16-45 given that one had not already done so in the index and child cohorts separately. Interactions between time, time2 and parental death are included in all models to allow the progression rate to vary by time in ego's life. Table 1
Distribution of outcome measures, by sex. Age at first birth is abbreviated AFB, and university attendance is a binary outcome. Income is not shown as it was age, sex and year standardized, but this data was available for 10,383 sons and 9891 daughters. n mean SD min max Sons fertility 10292 1.83 1.30 0 11 AFB 8102 27.32 5.52 15 50 university 10384 0.31 Daughters fertility 9749 1.95 1.19 0 14 AFB 8320 24.75 5.31 14 48 university 9,889 0.37 Table 2 Number of parental deaths in each 2-year age category in AFB models / Education models. Total sample sizes after excluding individuals with missing data in the full models are reported below. AFB and education represent the outcomes with the smallest and largest sample sizes. Sons Daughters Age at death Mother deaths Father deaths Mother deaths Father deaths 0-2 4/6 6 / 12 2/2 9 / 10 2-4 3/5 15 / 20 6/6 11 / 15 4-6 8 / 10 17 / 20 8 / 12 14 / 21 6-8 7/9 23 / 34 17 / 19 15 / 17 8-10 9 / 15 30 / 37 18 / 19 28 / 34 10-12 19 / 23 39 / 48 15 / 22 28 / 37 12-14 26 / 33 44 / 59 19 / 23 32 / 40 14-16 11 / 22 60 / 75 27 / 30 35 / 50 16-18 21 / 27 53 / 63 28 / 33 54 / 69 18-20 30 / 40 53 / 76 31 / 41 72 / 84 Total sample 7902 / 9859 7766 / 9689 8141 / 9411 8007 / 9264
7
DO NOT CITE WITHOUT PERMISSION
RESULTS Model comparison results When we compare the best fitting critical period model to linear parental presence models and null models with no parental absence predictor, we find stronger evidence that children experience critical periods of vulnerability to maternal losses than to paternal losses. This is evidenced by the larger the best fit critical period models having higher AIC weights than the null or linear models when examining maternal losses (Figure 1). In fact, we find relatively little evidence that paternal absences have any effect on children's outcomes (the null models fit best) except insofar as fathers' deaths affect daughters' ages at first birth. Figure 1 also shows that the effect of maternal deaths on children's education and sons' ages at first birth can be modeled just as well as a linear effect, wherein the younger a boy loses his mother the larger the effect. It is also worth noting that even when a critical period model fits the data best, these may capture relatively long ranges of susceptibility. For example, the model predicting a daughter's AFB from a father's death at any time between 4 and 18 years of age fits the data best.
Estimating effect sizes While the model comparison alerts us to the best way of modeling parental absences, Figures 2-5 show the strength of these effects at different ages of parental loss and allows us to qualitatively examine the developmental patterns. The earlier a son loses his mother the earlier his age at first birth (Figure 2). For each year earlier that a boy's mother dies within his first 20 years of life, the son's predicted age at first birth decreases by 0.13 years (SE=.05). On the other hand, father's deaths later in childhood and adolescence expedite daughter's first births the most. These effect sizes are fairly similar regardless of when fathers' deaths occur between ages 4 and 18. Figure 2 shows that the effect of the same sex parent's death on their childrens' AFBs are less clear. Fathers have little effect on sons first births, and while daughters AFBs are lowest if they experience a maternal loss between 16-18, these data are not clearly patterned. Figure 3 shows that the developmental effects of parental loss on completed fertility are weaker. Although sons' and daughters' fertility seem to be particularly affected if their mothers die between 4-6 or 10-12 respectively, these may be spurious given that maternal deaths on either end of these short 2-year intervals do not have similar effects. Furthermore, these maternal deaths have opposite effects on sons and daughters and do not correspond to other developmental trajectories revealed by our analysis. Father's deaths have little effect on total fertility, though we can detect a small linear effect of later age at father's death (within first 20 years of life) on fertility for sons fertility (B=.009, SE=.004). There is little evidence that children experience sensitive periods of parental death that affect their probability of attending university later in life (Figure 4). Age at maternal death within the first two decades of life can be captured as a positive linear effect for both sons (log odds=.043, SE=.023) and daughters (log odds=.041, SE=.023), meaning that the earlier in one's childhood a mother died the less likely one is to go to college. Again we find little evidence that fathers' deaths affect children's likelihood of attending college, but if anything earlier paternal deaths are associated with a greater likelihood that daughters go to university. Maternal deaths anytime from late childhood through adolescence are associated with similar decrements in her sons' adult household income, while daughters' period of susceptibility seems to be more constrained to earlier childhood as concerns her household income (Figure 5). Father's deaths in childhood have no clear effects on this metric. Table 3 also shows the effects of losing parents in childhood using larger, theoreticallymotivated age categories. This analysis shows similar effects as the empirically driven analysis. However, it lets us see that a proposed sensitive period of vulnerability before age 7 is only borne out in models predicting sons' reproductive outcomes from maternal deaths. Fathers' deaths are associated with earlier ages at first birth for daughters, but only if they occur later in childhood and adolescence. The analysis in Table 3 also shows that people whose parents die after they are 45 years of age also 8
DO NOT CITE WITHOUT PERMISSION have younger ages at first birth, suggesting that parental deaths might not be causing the variation in reproductive and socio-economic outcomes. In the next section we consider whether residual unmeasured variance in family and socio-economic background may be confounding the effects of parental deaths in childhood that we have detected. Unmeasured variance as confound? Table 3 compares the effect of parental deaths in the full models with all available socioeconomic and family background covariates as controls, to the effect of parental deaths in a simple model with only birth year band as a covariate. It shows that all the significant differences in AFB between people whose parents died after age 45 and those whose parents were still alive were heavily confounded by family and socio-economic background variables – i.e. the effect size was heavily reduced by adding these variables as covariates. This is to be expected since these parental deaths are likely happening after the index individuals' first births and therefore cannot be on the causal pathway. On the other hand, all associations between experiences of parental death in childhood and adolescence with earlier ages at first birth were negatively confounded by the available covariates – i.e. adding covariates increased the effect size. This suggests that any unmeasured variance in family background is likely making us underestimate the size of the effect of parental deaths in childhood on ages at first birth. Similarly the effect of experiencing a maternal death in childhood on one's probability of attending university and on son's fertility is robust to the addition of family background and socioeconomic covariates. Again the evidence suggests that if anything adding covariates is increasing the size of the effect and therefore that were we to measure these confounds perfectly we would see an even larger effect of such early maternal deaths. On the other hand, any evidence of parental deaths affecting later household income may be driven by confounding unmeasured variance in family background. Historical changes We find robust effects of maternal deaths in childhood on AFB, and of paternal deaths on daughters' AFB within the index generation. However, it is worth noting that these effects may be quite labile, changing according to historical and cultural context. Figure 6 shows that men whose mothers died in their first seven years of life are less likely to remain childless in both the index and child generation. However, the pace at which they progress to their first birth is not otherwise faster in the child generation, although it was significantly faster in the index generation. Figure 7 shows that the effect of paternal deaths on progression to a first birth may be increasing for both sons and daughters. For example, experiencing a paternal death by age 7 is associated with a faster progression to a first birth for daughters (and sons up to age 30), but only in the more recent child generation.
9
DO NOT CITE WITHOUT PERMISSION
Figure 1 Comparing the best-fitting critical period, linear parental presence, and null models. The top row a) shows the AIC weights from the different models predicting son’s outcomes from parental deaths, while b) shows the same model comparisons for daughters. Numbers in the red bars refer to the age range for the bestfitting critical periods. 1.00 0.80 0.60
Fertility
AFB
Mothers
16-18
Income
8-18 Income
2-6
8-12 Education
a) Sons
AFB
0.00
Education
4-6 Fertility
0.20
0-10
0.40
Fathers
1.00 0.80 0.60
0-8
4-18 Education
4-18
10-12 Fertility
4-8
16-18
0.20
AFB
0.40
Mothers
Income
Education
Fertility
AFB
Income
AIC Weight
b) Daughters
0.00
Fathers
best critical period
linear until 20
10
null
DO NOT CITE WITHOUT PERMISSION
Daughters Effect on AFB (in years)
Sons
Figure 2 Effect size of parental death on AFB by age at time of death. Regression coefficients are from models comparing individuals whose parents died within a given 2 year to those whose parents did not die within the interval. 95% CI shown. Mother Father 6
6
4
4
2
2
0
0
-2
-2
-4
-4
-6
-6
-8
-8 0 2 4 6 8 10 12 14 16 18 20
0 2 4 6 8 10 12 14 16 18 20
6
6
4
4
2
2
0
0
-2
-2
-4
-4
-6
-6
-8
-8 0 2 4 6 8 10 12 14 16 18 20
Age at time of parental death
11
0 2 4 6 8 10 12 14 16 18 20
DO NOT CITE WITHOUT PERMISSION
Daughters Effect on number of births
Sons
Figure 3 Effect size of parental death on fertility by age at time of death. Regression coefficients are from models comparing individuals whose parents died within a given 2 year to those whose parents did not die within the interval. 95% CI shown. Mother Father 1
1
0.75
0.75
0.5
0.5
0.25
0.25
0
0
-0.25
-0.25
-0.5
-0.5
-0.75
-0.75
-1
-1 0 2 4 6 8 10 12 14 16 18 20
0 2 4 6 8 10 12 14 16 18 20
1
1
0.75
0.75
0.5
0.5
0.25
0.25
0
0
-0.25
-0.25
-0.5
-0.5
-0.75
-0.75
-1
-1 0 2 4 6 8 10 12 14 16 18 20
Age at time of parental death
12
0 2 4 6 8 10 12 14 16 18 20
DO NOT CITE WITHOUT PERMISSION
Daughters Sons Effect on log odds of university attendance
Figure 4 Effect size of parental death on education by age at time of death. Regression coefficients are from models comparing individuals whose parents died within a given 2 year to those whose parents did not die within the interval. 95% CI shown. Mother Father 2
2
1.5
1.5
1
1
0.5
0.5
0
0
-0.5
-0.5
-1
-1
-1.5
-1.5
-2
-2 0 2 4 6 8 10 12 14 16 18 20
0 2 4 6 8 10 12 14 16 18 20
2
2
1.5
1.5
1
1
0.5
0.5
0
0
-0.5
-0.5
-1
-1
-1.5
-1.5
-2
-2 0 2 4 6 8 10 12 14 16 18 20
Age at time of parental death
13
0 2 4 6 8 10 12 14 16 18 20
DO NOT CITE WITHOUT PERMISSION
Daughters Sons Effect on sex-age standardized hh income
Figure 5 Effect size of parental death on household income by age at time of death. Regression coefficients are from models comparing individuals whose parents died within a given 2 year to those whose parents did not die within the interval. 95% CI shown. Mother Father 1
1
0.75
0.75
0.5
0.5
0.25
0.25
0
0
-0.25
-0.25
-0.5
-0.5
-0.75
-0.75
-1
-1 0 2 4 6 8 10 12 14 16 18 20
0 2 4 6 8 10 12 14 16 18 20
1
1
0.75
0.75
0.5
0.5
0.25
0.25
0
0
-0.25
-0.25
-0.5
-0.5
-0.75
-0.75
-1
-1 0 2 4 6 8 10 12 14 16 18 20
Age at time of parental death
14
0 2 4 6 8 10 12 14 16 18 20
DO NOT CITE WITHOUT PERMISSION
Figure 6
Child Generation Proportion of population remaining childfree
Index Generation
Mother effects on AFB across cohorts. Predicted probability of progressing to a first birth by participant age and age at which participant’s mother died. Results are from an Event History Analysis model allowing for linear and squared interactions of parental deaths with time. Predicted values are given at mean parental income, parents having a high school education, parental fertility of 2, and parents age at first birth of 24, and a birthband of 1955-60 or 1970-75 for each cohort. Sons Daughters 1
1
0.75
0.75
0.5
0.5
0.25
0.25
0
0 15
20
25
30
35
40
45
1
1
0.75
0.75
0.5
0.5
0.25
0.25
0
0 15
20
25
30
35
40
45
Age in years dead by 7
dead 7-16
alive at 16
15
15
20
25
30
35
40
45
15
20
25
30
35
40
45
DO NOT CITE WITHOUT PERMISSION
Figure 7
Child Generation Proportion of population remaining childfree
Index Generation
Father effects on AFB across cohorts. Predicted probability of progressing to a first birth by participant age and age at which participant’s father died. Results are from an Event History Analysis model allowing for linear and squared interactions of parental deaths with time. All other control variables were the same. Sons Daughters 1
1
0.75
0.75
0.5
0.5
0.25
0.25
0
0 15
20
25
30
35
40
45
1
1
0.75
0.75
0.5
0.5
0.25
0.25
0
15
20
25
30
35
40
45
15
20
25
30
35
40
45
0 15
20
25
30
35
40
45
Age in years dead by 7
dead 7-16
alive at 16
16
DO NOT CITE WITHOUT PERMISSION
hh. Income
University
Fertility
AFB
Metric
Table 3. Comparison of Effect Sizes in Simple vs. Full models with Covariates. Coefficients are the effect of having a parent die in the given age bracket relative to the reference category of having a parent still alive. The right-most proportion columns indicate the fraction of the simple model coefficient, that the difference between model coefficient represents. A proportion>0 means that adding co-variates decreased the effect size –i.e. that the effective was negatively confounded by other socio-economic factors. A proportion less than 0 indicates that the co-variates were negative confounders. Effects that were marginally significant in the full model are in bold, and additionally italicized if the co-variates were positive confounders. Effects that were qualitatively confounded –i.e. adding co-variates changed the direction of the effect are denoted q. Sons Daughters age when Mother Father Mother Father parent simple full prop. simple full prop. simple full prop. simple full prop. died ref:(alive) by 7 7-15 16-24 25-44 after 45 ref:(alive) by 7 7-15 16-24 25-44 after 45 ref:(alive) by 7 7-15 16-24 25-44 after 45 ref:(alive) by 7 7-15 16-24 25-44 after 45
linear B (SE) -2.25 (0.99) -0.79 (0.66) -0.61 (0.45) -0.59 (0.2) -1.25 (0.15) poisson (SE) 0.17 (0.12) -0.1 (0.08) 0.01 (0.05) -0.04 (0.02) 0.01 (0.01) logit (SE) -0.09 (0.41) -0.55 (0.27) -0.31 (0.16) -0.29 (0.07) -0.29 (0.06) linear B (SE) -0.1 (0.09) -0.18 (0.05) -0.13 (0.03) -0.08 (0.02) -0.02 (0.01)
-2.78 (0.88) -1.01 (0.62) -0.53 (0.43) -0.51 (0.19) -0.23 (0.15)
-0.23 -0.27 0.13 0.14 0.81
-0.95 (0.84) -0.96 (0.45) -0.98 (0.33) -1.09 (0.19) -1.84 (0.18)
-0.91 (0.82) -0.5 (0.43) -0.67 (0.3) -0.58 (0.18) -0.69 (0.18)
0.05 0.48 0.32 0.47 0.62
-0.52 (0.93) 0.27 (0.61) -1.04 (0.42) -0.47 (0.2) -1.63 (0.14)
-0.09 (0.78) -0.08 (0.56) -1.17 (0.39) -0.28 (0.18) -0.33 (0.14)
0.83 q -0.12 0.41 0.8
-0.78 (0.78) -1.27 (0.51) -0.97 (0.32) -1.05 (0.18) -1.7 (0.17)
-0.33 (0.66) -1.35 (0.46) -0.38 (0.3) -0.36 (0.16) -0.08 (0.16)
0.57 -0.1 0.61 0.66 0.96
0.26 (0.1) -0.03 (0.09) 0.02 (0.05) -0.04 (0.02) -0.03 (0.02)
-0.52 0.71 -0.42 0.02 q
-0.11 (0.09) -0.08 (0.05) -0.03 (0.03) 0.02 (0.02) 0.09 (0.02)
-0.09 (0.09) -0.02 (0.05) 0.01 (0.04) 0.02 (0.02) 0.06 (0.02)
0.19 0.72 q 0.08 0.36
-0.04 (0.12) -0.09 (0.06) 0.02 (0.04) -0.02 (0.02) 0.002 (0.01)
0.04 (0.12) -0.09 (0.07) 0.05 (0.05) -0.02 (0.02) -0.03 (0.02)
q 0.05 -1.14 -0.22 q
-0.13 (0.11) -0.04 (0.06) -0.02 (0.03) -0.01 (0.02) 0.04 (0.02)
-0.05 (0.11) -0.02 (0.06) -0.02 (0.03) -0.02 (0.02) 0.01 (0.02)
0.65 0.52 -0.1 -0.9 0.85
-0.4 (0.51) -0.56 (0.28) -0.13 (0.19) -0.19 (0.08) -0.09 (0.06)
-3.43 -0.02 0.58 0.35 0.68
-0.84 (0.31) -0.67 (0.17) -0.62 (0.12) -0.54 (0.07) -0.41 (0.06)
-0.57 (0.36) -0.28 (0.18) -0.37 (0.13) -0.31 (0.07) -0.18 (0.07)
0.32 0.58 0.41 0.41 0.56
-0.83 (0.49) -0.3 (0.22) -0.29 (0.16) -0.4 (0.07) -0.29 (0.05)
-0.6 (0.5) -0.4 (0.26) -0.22 (0.19) -0.29 (0.08) -0.03 (0.06)
0.28 -0.32 0.24 0.27 0.91
0.05 (0.31) -0.27 (0.18) -0.33 (0.11) -0.32 (0.07) -0.37 (0.06)
0.45 (0.32) -0.11 (0.19) -0.06 (0.12) -0.12 (0.07) -0.1 (0.07)
-8.4 0.6 0.8 0.62 0.74
-0.06 (0.1) -0.14 (0.05) -0.07 (0.03) -0.05 (0.02) 0.003 (0.01)
0.34 0.24 0.43 0.43 q
-0.1 (0.06) -0.14 (0.03) -0.11 (0.03) -0.09 (0.02) -0.03 (0.02)
-0.003 (0.06) -0.05 (0.03) -0.04 (0.03) -0.05 (0.02) -0.01 (0.02) 17
0.97 0.68 0.65 0.47 0.63
-0.18 (0.06) -0.03 (0.06) -0.06 (0.04) -0.05 (0.02) -0.03 (0.01)
-0.13 (0.06) -0.04 (0.06) -0.04 (0.04) -0.03 (0.02) -0.03 (0.01)
0.29 -0.34 0.36 0.47 0.19
-0.08 (0.07) -0.07 (0.05) -0.06 (0.03) -0.06 (0.02) -0.02 (0.02)
-0.03 (0.07) 0.005 (0.05) -0.004 (0.03) -0.03 (0.02) -0.003 (0.02)
0.64 q 0.93 0.55 0.82
DO NOT CITE WITHOUT PERMISSION
18
DO NOT CITE WITHOUT PERMISSION
DISCUSSION Our results indicate some support for the existence of sensitive periods during which parental absences can shape a child’s life history strategy. However, the nature of these critical periods varies by the sex of the parent-offspring dyad and by the outcome being measured. This suggests that the outcomes do not constitute a correlated set of traits indicative of a singular life history strategy as others have argued (Belsky et al., 1991; Simpson et al., 2012). Furthermore, these complex associations may indicate that diverse mechanisms are responsible for the associations between parental absences and reproductive and socio-economic outcomes. Reproductive outcomes Maternal deaths before a son turns 7 expedite his first birth and increase his fertility. However, for daughters parental deaths in later childhood or adolescence have stronger effects, though only on timing of first birth and not on fertility. This is consistent with results from an American sample showing that environmental unpredictability in the first 5 years of life expedited males', but not females' sexual debut (Simpson et al., 2012). Previous research also suggests that females may be more vulnerable to stressors more generally later in childhood, while males are more susceptible to stressors earlier in childhood (Masten, Best, & Garmezy, 1990; Werner, 2010). This pattern implies that changes to timing of pubertal development – which commences around age 7 (Ellis, 2004) – can only account for the effect of mothers on sons, but not the effects of parental deaths on daughters first births. The later developmental effects of parental deaths on daughters' age at first birth suggest a flexible behavioral response rather than one constrained by physiological changes earlier in her development. The data are also more robust for parental deaths expediting first births rather than increasing total fertility per se. Effects on reproductive timing rather than fertility are both theoretically more cogent and straight forward predictions from life history theory (Ellis, 2004; R. Quinlan, 2007), and possibly more likely to be detected in a population with low overall fertility norms where people do not take advantage of their full reproductive life span. Socio-economic outcomes We find much weaker evidence that parental deaths during critical periods in development affect socio-economic outcomes. This is despite the tradeoffs between early reproduction and university attendance experienced in many post-industrial countries such as Sweden. The effect of a mother's presence in the first 20 years of life on the probability of going to college can be parsimoniously modeled as additive, with each extra year of life increasing equally the log odds of her daughters and sons attending university. Fathers; deaths did not have similarly important consequences for college attendance. Similar differences between the effect of maternal and paternal losses have been documented in South Africa (Case & Ardington, 2006), suggesting that it is not just because of generous state subsidies that we fail to find a father absence effect in Sweden. In fact it is worth noting that if anything earlier paternal deaths were associated with increased chances of daughters attending college. Parental deaths in early in life were only weakly associated with adult household income, and we find evidence that these may be due to unmeasured variance in family background that confounds the relationship – i.e. unmeasured variance that causes both premature parental deaths and lower socio-economic status. 19
DO NOT CITE WITHOUT PERMISSION
Caution in interpreting current literature We generally find that maternal deaths have larger effects on children's life history outcomes than do paternal deaths. This suggests that the literature's emphasis on father absence (Bogaert, 2005; Hoier, 2003; R. J. Quinlan, 2003; Shenk, Starkweather, Kress, & Alam, 2013; Sheppard & Sear, 2012; Sheppard, Snopkowski, & Sear, 2014) may be motivated by the higher prevalence of father absent households making statistical anlaysis easier. The popularity of analysis should not be confused for stronger evidence of an important effect. Furthermore, we see particularly large effects of early maternal absences on sons' reproductive development and outcomes, meaning that the emphasis on daughter's puberty and sexual behavior (e.g. Chisholm, Quinlivan, & Petersen, 2005; Milne & Judge, 2011; Tither & Ellis, 2008) may not be theoretically or empirically warranted. The results regarding critical windows of vulnerability to parental loss illustrate the importance of combining a data-driven approach to documenting developmental trajectories with a theory-driven one. Psychologists and biologists understanding of developmental sensitive windows is inchoate for many phenomena. The literature on sensitive windows of life history development uses diverse age cut-offs, making comparisons across papers difficult. Furthermore, testing for critical windows suggested in the literature may mask more accurate representations of the data. We recommend that exploratory model comparisons of critical windows, such as the one presented here be included in publications to facilitate future meta-analyses of developmental processes. Our historical analysis also intimates that caution is necessary in drawing generalizations about human biology and psychology from patterns in a few sites (Henrich, Heine, & Norenzayan, 2010). For example, we find evidence that the effect of experiencing a father's death early in childhood on reproductive timing is larger in the more recent generation. This is particularly surprising given that the two generations represent populations only a couple decades apart in the same country. This historical change may reflect the increasing importance of fathering in post-demographic transition countries. This means that much of the research focusing on contemporary effects of paternal absences in post-industrial contexts exaggerate the strength and importance of these effects for human societies more generally. On the other hand, the fact that we do find long-term consequences of maternal deaths on reproductive and educational outcomes in a society with strong social safety nets, even when we control for family background covariates, strengthens the evidence for the causal role of parental investments in influencing decisions later in life. More crosscultural and trans-historical evidence is needed to determine how canalized such sensitive periods are. ACKNOWLEDGEMENTS We would like to thank Amy Heshmati for help running scripts, the Evolutionary Demography Group at LSHTM for feedback on earlier versions of the manuscript, and Bret Beheim for statistical advice. This work was funded by the European Research Council Starting Grant #263760. BIBLIOGRAPHY Alvergne, A., Faurie, C., & Raymond, M. (2008). Developmental plasticity of human reproductive development: effects of early family environment in modern-day France. Physiology & Behavior, 95(5), 625–32. doi:10.1016/j.physbeh.2008.09.005 20
DO NOT CITE WITHOUT PERMISSION
Belsky, J., Steinberg, L., & Draper, P. (1991). Childhood experience, interpersonal development, and reproductive strategy: An evolutionary theory of socialization. Child Development, 62(4), 647–670. Retrieved from http://onlinelibrary.wiley.com/doi/10.1111/j.1467-8624.1991.tb01558.x/abstract Bogaert, A. F. (2005). Age at puberty and father absence in a national probability sample. Journal of Adolescence, 28(4), 541–6. doi:10.1016/j.adolescence.2004.10.008 Borgerhoff Mulder, M., Bowles, S., Hertz, T., Bell, A., Beise, J., Clark, G., … Wiessner, P. (2009). Intergenerational wealth transmission and the dynamics of inequality in smallscale societies. Science, 326(5953), 682–8. doi:10.1126/science.1178336 Burström, K., Johannesson, M., & Diderichsen, F. (2005). Increasing socio‐ economic inequalities in life expectancy and QALYs in Sweden 1980–1997. Health Economics, 14(8), 831–850. Retrieved from http://onlinelibrary.wiley.com/doi/10.1002/hec.977/abstract Card, J., & Wise, L. (1978). Teenage mothers and teenage fathers: The impact of early childbearing on the parents’ personal and professional lives. Family Planning Perspectives. Retrieved from http://www.jstor.org/stable/10.2307/2134267 Cas, A. G., Frankenberg, E., Suriastini, W., & Thomas, D. (2014). The Impact of Parental Death on Child Well-being: Evidence From the Indian Ocean Tsunami. Demography, 51(2), 437–457. doi:10.1007/s13524-014-0279-8 Case, A., & Ardington, C. (2006). The impact of parental death on school outcomes: Longitudinal evidence from South Africa. Demography. Retrieved from http://link.springer.com/article/10.1353/dem.2006.0022 Chisholm, J. (1993). Death, Hope, and Sex: Life-History Theory and the Development of Reproductive Strategies. Current Anthropology, 34(1), 1–24. Retrieved from http://www.jstor.org/stable/10.2307/2743728 Chisholm, J., Quinlivan, J., & Petersen, R. (2005). Early stress predicts age at menarche and first birth, adult attachment, and expected lifespan. Human Nature, 16(3), 233–265. Retrieved from http://www.springerlink.com/index/C3TQ19BEEDNDXEJ6.pdf Draper, P., & Harpending, H. (1982). Father absence and reproductive strategy: An evolutionary perspective. Journal of Anthropological Research, 38(3), 255–273. Retrieved from http://www.jstor.org/stable/10.2307/3629848 Ellis, B. J. (2004). Timing of pubertal maturation in girls: an integrated life history approach. Psychological Bulletin, 130(6), 920–58. doi:10.1037/0033-2909.130.6.920 Ermisch, J., Francesconi, M., & Pevalin, D. J. (2004). Parental partnership and joblessness in childhood and their influence on young people’s outcomes. Journal of the Royal Statistical Society: Series A (Statistics in Society), 167(1), 69–101. doi:10.1111/j.1467985X.2004.00292.x
21
DO NOT CITE WITHOUT PERMISSION
Frankenhuis, W. E., & Panchanathan, K. (2011). Balancing sampling and specialization: an adaptationist model of incremental development. Proceedings. Biological Sciences / The Royal Society, 278(1724), 3558–3565. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/21490018 Geronimus, A., Bound, J., & Waidmann, T. (1999). Health inequality and population variation in fertility-timing. Social Science & Medicine. Retrieved from http://www.sciencedirect.com/science/article/pii/S0277953699002464 Giudice, M. Del, & Belsky, J. (2011). The development of life history strategies: Toward a multi-stage theory. Development. Retrieved from http://bsb-lab.org/site/wpcontent/uploads/DelGiudice_Belsky_2011_development_LH-strategies_chapter.pdf Goodman, A., Gisselmann, M., & Koupil, I. (2010). Birth characteristics and early-life social characteristics predict unequal educational outcomes across the life course and across generations. Longitudinal and Life Course Studies, 1(4), 317–338. Retrieved from http://researchonline.lshtm.ac.uk/20821/1/sa2010Goodman_Birth_characteristics_and_e arly-life_social_characteristics_predict_unequal_educational.pdf Goodman, A., & Koupil, I. (2009). Social and biological determinants of reproductive success in Swedish males and females born 1915–1929. Evolution and Human Behavior, 30(5), 329–341. doi:10.1016/j.evolhumbehav.2009.03.007 Gurven, M., Kaplan, H., & Gutierrez, M. (2006). How long does it take to become a proficient hunter? Implications for the evolution of extended development and long life span. Journal of Human Evolution, 51(5), 454–470. Retrieved from http://www.sciencedirect.com/science/article/pii/S0047248406000960 Henrich, J., Heine, S., & Norenzayan, A. (2010). The Weirdest People in the World? Behavioral and Brain Sciences, 33, 61–135. doi:10.1017/S0140525X0999152X Hill, K. (1993). Life history theory and evolutionary anthropology. Evolutionary Anthropology, 2(3), 78–88. Retrieved from http://onlinelibrary.wiley.com/doi/10.1002/evan.1360020303/full Hoem, B., & Hoem, J. (1997). Sweden’s family policies and roller-coaster fertility. Retrieved from http://websv.ipss.go.jp/syoushika/bunken/data/pdf/16896401.pdf Hoier, S. (2003). Father absence and age at menarche. Human Nature, 14(3), 2003. Retrieved from http://link.springer.com/article/10.1007/s12110-003-1004-2 Huber, S., Bookstein, F. L., & Fieder, M. (2010). Socioeconomic status, education, and reproduction in modern women: an evolutionary perspective. American Journal of Human Biology, 22(5), 578–87. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/20737603 Masten, A. S., Best, K. M., & Garmezy, N. (1990). Resilience and development: Contributions from the study of children who overcome adversity. Development and Psychopathology, 2, 425–444. doi:10.1017/S0954579400005812
22
DO NOT CITE WITHOUT PERMISSION
Matchock, R. L., & Susman, E. J. (2006). Family composition and menarcheal age: antiinbreeding strategies. American Journal of Human Biology : The Official Journal of the Human Biology Council, 18(4), 481–91. doi:10.1002/ajhb.20508 Milne, F. H., & Judge, D. S. (2011). Brothers delay menarche and the onset of sexual activity in their sisters. Proceedings. Biological Sciences / The Royal Society, 278(1704), 417– 23. doi:10.1098/rspb.2010.1377 Moya, C., & Sear, R. (2014). Intergenerational conflicts may help explain parental absence effects on reproductive timing: a model of age at first birth in humans. PeerJ. doi:10.7717/peerj.512 Murphy, M., & Knudsen, L. (2002). The intergenerational transmission of fertility in contemporary Denmark: The effects of number of siblings (full and half), birth order, and whether male or female. Population Studies, 56(3), 235–248. Retrieved from http://www.tandfonline.com/doi/abs/10.1080/00324720215937 Pesonen, A.-K., Räikkönen, K., Heinonen, K., Kajantie, E., Forsén, T., & Eriksson, J. G. (2008). Reproductive traits following a parent-child separation trauma during childhood: a natural experiment during World War II. American Journal of Human Biology : The Official Journal of the Human Biology Council, 20(3), 345–51. doi:10.1002/ajhb.20735 Pettay, J. E., Kruuk, L. E. B., Jokela, J., & Lummaa, V. (2005). Heritability and genetic constraints of life-history trait evolution in preindustrial humans. Proceedings of the National Academy of Sciences of the United States of America, 102(8), 2838–43. doi:10.1073/pnas.0406709102 Quinlan, R. (2007). Human parental effort and environmental risk. Proceedings of the Royal Society B-Biological Sciences, 274(1606), 121–125. Quinlan, R. J. (2003). Father absence, parental care, and female reproductive development. Evolution and Human Behavior, 24(6), 376–390. doi:10.1016/S1090-5138(03)00039-4 Rickard, I., Frankenhuis, W., & Nettle, D. (2014). Why Are Childhood Family Factors Associated With Timing of Maturation? A Role for Internal Prediction. Perspectives on Psychological Science, 9(1), 3–15. doi:10.1177/1745691613513467 Russell, A., & Saebel, J. (1997). Mother–Son, Mother–Daughter, Father–Son, and Father– Daughter: Are They Distinct Relationships? Developmental Review, 17(2), 111–147. doi:10.1006/drev.1996.0431 Sear, R., & Mace, R. (2008). Who keeps children alive? A review of the effects of kin on child survival. Evolution and Human Behavior, 29(1), 1–18. doi:10.1016/j.evolhumbehav.2007.10.001 Shenk, M. K., Starkweather, K., Kress, H. C., & Alam, N. (2013). Does absence matter? : a comparison of three types of father absence in rural bangladesh. Human Nature, 24(1), 76–110. doi:10.1007/s12110-013-9160-5
23
DO NOT CITE WITHOUT PERMISSION
Sheppard, P., Garcia, J., & Sear, R. (2014). A not-so-grim tale: how childhood family structure influences reproductive and risk-taking outcomes in a historical u.s. Population. PloS One, 9(3), e89539. doi:10.1371/journal.pone.0089539 Sheppard, P., & Sear, R. (2012). Father absence predicts age at sexual maturity and reproductive timing in British men. Biology Letters, 8(2), 237–40. doi:10.1098/rsbl.2011.0747 Sheppard, P., Snopkowski, K., & Sear, R. (2014). Father absence and reproduction-related outcomes in Malaysia, a transitional fertility population. Human Nature, 25(2), 213–34. Retrieved from http://link.springer.com/article/10.1007/s12110-014-9195-2 Simpson, J., Griskevicius, V., Kuo, S., Kuo, C., Sung, S., & Collins, W. (2012). Evolution, Stress, and Sensitive Periods: The Influence of Unpredictability in Early Versus Late Childhood on Sex and Risky Behavior. Developmental Psychology, 48(3), 674–686. Retrieved from http://www.researchgate.net/publication/221827671_Evolution_stress_and_sensitive_pe riods_the_influence_of_unpredictability_in_early_versus_late_childhood_on_sex_and_r isky_behavior/file/9fcfd505ba945616c2.pdf Smith, K. R., Hanson, H. A., Norton, M. C., Hollingshaus, M. S., & Mineau, G. P. (2014). Survival of Offspring who Experience Early Parental Death: Early Life Conditions and Later-Life Mortality. Social Science & Medicine. Retrieved from http://www.sciencedirect.com/science/article/pii/S0277953613006667 Stanfors, M., & Scott, K. (2013). Intergenerational transmission of young motherhood. Evidence from Sweden, 1986–2009. The History of the Family, 18(2), 187–208. doi:10.1080/1081602X.2013.817348 Surbey, M. (1990). Family composition, stress, and the timing of human menarche. In T. Ziegler & F. Bercovitch (Eds.), Socioendocrinology of primate reproduction (pp. 11– 32). New York: Wiley-Liss. Retrieved from http://psycnet.apa.org/psycinfo/199098040-001 Teilmann, G., Pedersen, C. B., Skakkebaek, N. E., & Jensen, T. K. (2006). Increased risk of precocious puberty in internationally adopted children in Denmark. Pediatrics, 118(2), e391–9. doi:10.1542/peds.2005-2939 Tennant, C. (1988). Parental loss in childhood: its effect in adult life. Archives of General Psychiatry. Retrieved from http://archpsyc.ama-assn.org/cgi/reprint/45/11/1045.pdf Tither, J., & Ellis, B. (2008). Impact of fathers on daughters’ age at menarche: a genetically and environmentally controlled sibling study. Developmental Psychology, 44(5), 1409– 20. Retrieved from http://psycnet.apa.org/journals/dev/44/5/1409/ Werner, E. E. (2010). HIGH-RISK CHILDREN IN YOUNG ADULTHOOD: A Longitudinal Study from Birth to 32 Years. American Journal of Orthopsychiatry, 59(1), 72–81. doi:10.1111/j.1939-0025.1989.tb01636.x
24
DO NOT CITE WITHOUT PERMISSION
Willführ, K. P. (2009). Short- and long-term consequences of early parental loss in the historical population of the Krummhörn (18th and 19th century). American Journal of Human Biology : The Official Journal of the Human Biology Council, 21(4), 488–500. doi:10.1002/ajhb.20909
25
DO NOT CITE WITHOUT PERMISSION
SUPPLEMENTARY ONLINE MATERIALS Section 1 – Descriptive Statistics Figure S1
birth
1.00
given
0.80
having
Survival curves for progression to 1st birth. Individuals who were censored for nonreproductive reasons were taken out of the analysis. Open markers denote the index (2nd ) generation – the primary sample for the paper. The closed circles are the child (3 rd ) generation used for historical comparison.
0.60
Males G2 Females G2 Males G3 Females G3
Proportion
not
0.40
0.20
0.00 15
25
35
45
Age
Section 2 – Model selection process We try to assess the relative likelihood of various possible critical periods during which children may be more sensitive to parental deaths. To do so we compare the AIC weights of 55 models that differ only in the age cut offs during which an individual is coded as having experienced a parental loss. These 55 models consist of every possible critical period of even length within the first two decades of life. In other words the critical periods can start at any even age between 0 and 18, and end at any even age between 2 and 20. The AIC weights for these critical periods are plotted in Figures S2-S5. For example, looking at Figure S2 a model predicting boys' ages at first birth from a maternal death between ages 14 and 18 is very unlikely given the data (and a maternal loss during that age span has an effect of B=.2 on age at first birth for sons). On the other hand, a regression model predicting sons' age at first birth with a binary predictor of having lost a mother between ages 0 and 10 fits the data relatively better. This latter critical period is therefore our best estimate of where a critical period of susceptibility to maternal loss would be for boys insofar as it affects their ages at first birth – maternal losses during this age span expedites sons' first births by about 2 years (B=-2.1). Table S1 shows that there is no clear correlation between the number of individuals experiencing a parental death during a critical period and the AIC weight of the corresponding model.
26
DO NOT CITE WITHOUT PERMISSION
Figure S2. Best fit critical period models for age at first birth. The horizontal axis represents the starting age of the critical period window, and the vertical axis represents the ending age. The AIC weight for models with parental deaths in the respective window are represented by the color of the cell. Darker cells represent critical periods of parental death that fit the data better. The regression coefficient of having a parent die within that critical window are reported numerically in each cell. The letter in the box denotes whether the best fitting model is a critical period one (CP), a linear effect of length of parental presence (L), or a null model with no parental presence predictor (N). The critical window with white text has an AIC weight of 0.26, beyond the scale. Mothers Fathers
N
CP
CP
Daughters
Sons
L
27
DO NOT CITE WITHOUT PERMISSION
Figure S3
Daughters
Sons
Best fit critical period models for total fertility. The horizontal axis represents the starting age of the critical period window, and the vertical axis represents the ending age. The AIC weight for models with parental deaths in the respective window are represented by the color of the cell. Darker cells represent critical periods of parental death that fit the data better. The from zero-inflated poisson model regression coefficient of having a parent die within that critical window are reported numerically in each cell. The letter in the box denotes with the best fitting model is a critical period one (CP), a linear effect of length of parental presence (L), or a null model with no parental presence predictor (N). Mothers Fathers
CP
N
CP
N
28
DO NOT CITE WITHOUT PERMISSION
Figure S4
Daughters
Sons
Best fit critical period models for college education. The horizontal axis represents the starting age of the critical period window, and the vertical axis represents the ending age. The AIC weight for models with parental deaths in the respective window are represented by the color of the cell. Darker cells represent critical periods of parental death that fit the data better. The log odds regression coefficient of having a parent die within that critical window are reported numerically in each cell. The letter in the box denotes with the best fitting model is a critical period one (CP), a linear effect of length of parental presence (L), or a null model with no parental presence predictor (N). Note: models for two short critical periods could not be fit for mother effects on daughters and are therefore blank. Mothers Fathers
CP
N
CP
N
29
DO NOT CITE WITHOUT PERMISSION
Figure S5 Best fit critical period models for age-sex-year adjusted income. The horizontal axis represents the starting age of the critical period window, and the vertical axis represents the ending age. The AIC weight for models with parental deaths in the respective window are represented by the color of the cell. Darker cells represent critical periods of parental death that fit the data better. The regression coefficient of having a parent die within that critical window are reported numerically in each cell. The letter in the box denotes with the best fitting model is a critical period one (CP), a linear effect of length of parental presence (L), or a null model with no parental presence predictor (N). Mothers Fathers
N
CP
N
Daughters
Sons
CP
Table S1. Correlations (r) between AIC weights and number of individuals experiencing death of a parent in a given critical period. *p<0.05, **p<0.01, ***p<0.001 Mothers Fathers AFB Fertility Education Income AFB Fertility Education Income Sons -0.09 -0.3* 0.11 0.68*** 0.01 -0.12 0.04 -0.12 Daughters -0.05 -0.17 -0.12 -0.21 0.57*** -0.27* -0.3* -0.33*
30
