Dynamic and heterogeneous effects of sibling death on children’s outcomes

Significance

Although childhood mortality rates have plummeted over the past century in developed countries, between 5% and 10% of the US population still experiences a sibling death. While we know a death in the family has strong and durable negative long-term effects on relatives in the long run, little is known about the development of children who experience the death of a sibling in the medium run. This is a key issue, given children’s vulnerability, the malleability of early childhood skills, and their impact on future adult health and socioeconomic outcomes. By analyzing a longitudinal dataset, this study examines how children’s cognitive and socioemotional skills and parental effort change around the time of the death of a child in the family.

Abstract

This paper explores the effects of experiencing the death of a sibling on children’s developmental outcomes. Recent work has shown that experiencing a sibling death is common and long-term effects are large. We extend understanding of these effects by estimating dynamic effects on surviving siblings' cognitive and socioemotional outcomes, as well as emotional and cognitive support by parents. Using the Children of the National Longitudinal Survey of Youth 1979 (CNLSY79), we find large initial effects on cognitive and noncognitive outcomes that decline over time. We also provide evidence that the effects are larger if the surviving child is older and less prominent if the deceased child was either disabled or an infant, suggesting sensitive periods of exposure. Auxiliary results show that parental investments in the emotional support of surviving children decline following the death of their child.

In 2014, almost 40,000 children under the age of 20 y died in the United States, with the highest rates among infants, those between the ages of 15 and 19 y, and boys (1). More worryingly, disadvantaged children are more likely to be affected. For instance, blacks are 50% more likely to lose a sibling by the age of 20 y than whites (2).

Child deaths have large negative effects, including marital disruption, depression, and health problems persisting decades after the child’s death for parents (3, 4); in fact, bereavement following the death of a close relative is ranked among the most severe life events (5). Other relatives, such as surviving siblings, are also influenced by these deaths, both directly through the loss of a sibling and indirectly through parental bereavement. The uniqueness and longevity associated with sibling ties suggest that experiencing a sibling’s death could substantially disrupt the human capabilities of the surviving sibling. However, these experiences go largely unmeasured and are often not targets of interventions and resources, leading surviving siblings to be “forgotten grievers” (6–8). The scope of this omission is large, as US data suggest nearly 8% experience a sibling death before the age of 25 y (9).

Most of the previous literature examining bereaved siblings has examined the effects of adults losing siblings on increased adult all-cause mortality, suicide, and cardiovascular deaths (6, 10, 11). An emerging literature has begun to explore whether losing a sibling during childhood is particularly traumatic. Bolton et al. (12) show large increases in mental disorders for bereaved siblings, especially for those exposed as a teenager. Fletcher et al. (9) find significant reductions in educational attainment and increases in the likelihood of coresiding with parents in adulthood. Yu et al. (13) show increases of over 100% in mortality rates in the first year following the death and larger effects among siblings close in age. Our work extends this set of findings to explore the dynamic impacts of a sibling death using multiple relevant measures of child development and parental investments, and assesses the potential for “sensitive periods” based on children’s age (14, 15).

Insights from multiple disciplines suggest several ways that experiencing a sibling death may compromise adult outcomes. Economists often model siblings as competitors for shared family resources (16), although increases in resources to a surviving child may be counteracted by the effects of bereavement on both parents and children. Moreover, older siblings are role models for younger children, shaping educational and health outcomes (17, 18), suggesting differences in the effects of sibling death depending on their relative age difference. Epidemiologists find it a major stressful life event that leads to increased health risks, including higher rates of mortality in adulthood (11, 13). Psychologists have found that younger children express feelings of grief, sadness, and depression, and pretend play with their deceased sibling, while older children act out more, engage in elevated levels of risk taking, and are more inclined to enter helping professions (19).

We add to the literature by exploring the dynamic effects of exposure to sibling death during childhood. We use data with high-frequency measures to investigate cognitive and socioemotional (human capital) outcomes, in addition to home investments. We find large initial impacts of sibling death on cognitive and socioemotional outcomes that decline over time, suggesting that analyses that focus on only long-term outcomes fail to uncover the entire trajectory of impacts. We also show that the effects are larger if the surviving child is older, suggesting sensitive periods of exposure. One interpretation of these findings is through psychological development vulnerability; for example, older children are more likely to be at a developmental stage where they can understand loss and/or recognize the impacts of the death on their parents (20, 21).

Results

Table 1 reports results for cognitive outcomes: The first column in each panel shows the main effects, the second examines years since sibling death effects, and the third includes whether the deceased sibling was older than the focal child at death. We find negative significant differences in one measure of human capital, reading comprehension, for surviving siblings after versus before the death. We next test for differences in the estimates as the surviving child ages. Results in column 2 of Table 1 strongly support a large but declining effect of experiencing the death of a sibling on all outcomes. For example, for reading, we find a more than nine percentile point immediate impact that is significantly reduced by 0.8 point each year following the death of a sibling, suggesting that, on average, effects fade out ∼11 y after the death. We also find significant reductions of nine points in mathematics, nearly 11 points in reading comprehension, and six points in Peabody Picture Vocabulary Test (PPVT) scores immediately following the death of a sibling, all of which appear to decline over time.

Table 1.

Dynamic effect of sibling's death on cognitive outcomes

	Results
Cognitive outcomes	1	2	3
PIAT-M
Sibling death	−5.441	−8.682**	−9.508*
	(3.262)	(3.106)	(3.840)
Years since death		0.646
		(0.446)
Sibling who passed away was older			6.957+
			(3.550)
n (Wave × Child)	571	571	571
PIAT-Reading
Sibling death	−5.056	−9.078*	−8.575+
	(3.920)	(3.880)	(4.940)
Years since death		0.802
		(0.502)
Sibling who passed away was older			6.039
			(5.157)
n (Wave × Child)	569	569	569
PIAT-Reading Comprehension
Sibling death	−8.415*	−10.880**	−8.374+
	(3.198)	(4.097)	(4.249)
Years since death		0.451
		(0.563)
Sibling who passed away was older			−0.072
			(5.656)
n (Wave × Child)	475	475	475
PPVT
Sibling death	−1.235	−6.479	−8.440
	(4.268)	(5.000)	(5.427)
Years since death		1.172*
		(0.536)
Sibling who passed away was older			11.671*
			(4.945)
n (Wave × Child)	340	340	340

Models weighted to account for oversampling of minorities. SEs in parentheses clustered at the family level. All include the background controls described in SI Appendix. Wave × Child indicates the total observations for all children in all waves. **Significant at 0.01; *significant at 0.05; ⁺significant at 0.1.

Table 2 examines socioemotional outcomes. We find significant evidence of increases (i.e., worsening) of the Behavioral Problems Index (BPI) and significant reductions in the Scholastic Self-Perception Profile for Children (SPPC) in column 1. Results in column 2 show strong immediate effects on the BPI and on the scholastic component of the SPPC.

Table 2.

Dynamic effect of sibling's death on socioemotional outcomes

	Results
Socioemotional outcomes	1	2	3
BPI
Sibling death	8.161+	15.693**	17.650**
	(4.293)	(4.522)	(5.303)
Years since death		−1.535**
		(0.378)
Sibling who passed away was older			−16.544**
			(4.807)
n (Wave × Child)	621	621	621
SPPC-Scholastic competence
Sibling death	−11.425	−16.195+	−19.898+
	(8.039)	(9.475)	(10.631)
Years since death		0.954
		(1.372)
Sibling who passed away was older			17.156
			(14.782)
n (Wave × Child)	266	266	266
SPPC–Self-worth
Sibling death	−1.773	0.488	−6.058
	(6.598)	(8.119)	(8.231)
Years since death		−0.450
		(1.652)
Sibling who passed away was older			8.615
			(16.494)
n (Wave × Child)	265	265	265

Models weighted to account for oversampling of minorities. SEs in parentheses clustered at the family level. All include the background controls described in SI Appendix. Wave × Child indicates the total observations for all children in all waves. **Significant at 0.01; *significant at 0.05; ⁺significant at 0.1.

Next, we examine two aspects of the home environment that might be affected by a child death; these permit us to examine mechanisms of parental human capital investments linking sibling death to cognitive and socioemotional outcomes. We examine the total home environment scale and also divide it into its two components: cognitive and emotional support. The results in Table 3 for the disaggregated components show evidence that the emotional resources in the home environment are significantly lowered following a sibling death, and that this reduction does not diminish over time. In contrast, we find little evidence of changes in the cognitive support in the household. These results are intuitive in that the cognitive measurements include books in the home and other material inputs, while the emotional measurements include the positive and negative ways the parent interacts with the surviving child, which are more likely to be influenced by parental bereavement.

Table 3.

Dynamic effect of sibling's death on HOME scores

	Results
HOME scores	1	2	3
HOME Total
Sibling death	−8.191+	−5.673	−8.717
	(4.236)	(4.367)	(6.156)
Years since death		−0.563
		(0.506)
Sibling who passed away was older			0.847
			(5.816)
n (Wave × Child)	794	794	794
HOME Cognitive support
Sibling death	0.786	2.399	1.174
	(4.374)	(3.980)	(6.016)
Years since death		−0.366
		(0.452)
Sibling who passed away was older			−0.623
			(5.633)
n (Wave × Child)	751	751	751
HOME Emotional support
Sibling death	−13.763**	−12.646*	−15.664**
	(4.177)	(4.985)	(5.145)
Years since death		−0.250
		(0.542)
Sibling who passed away was older			3.112
			(4.612)
n (Wave × Child)	686	686	686

Models weighted to account for oversampling of minorities. SEs in parentheses clustered at the family level. All include the background controls described in SI Appendix. Wave × Child indicates the total observations for all children in all waves. **Significant at 0.01; *significant at 0.05; ⁺significant at 0.1.

Finally, in column 3 of Tables 1–3, we investigate potential heterogeneity in relative age. Including an interaction indicating that the deceased sibling at death was older than the focal child suggests that when the deceased sibling is older, the negative effect is mitigated. (We also explore potential heterogeneity by gender composition of siblings. SI Appendix, Table A.10 shows effects stratified by whether the siblings were of the same gender. While most of the effects are not significantly different than zero, siblings who have the same gender as the deceased sibling seem to be more negatively affected than their counterparts in terms of their perception of their scholastic ability.)

Additional Heterogeneous Effects: 1. Sensitive Period Effects.

In Table 4, we examine whether the influence of a sibling’s death differs by the age of the surviving child when the surviving child experiences the death of his/her sibling (age distribution of siblings at the time of bereavement is shown in SI Appendix, Fig. S1). The results suggest that for all measures of cognitive ability, the negative effect on a child is greater if the child is older when his/her sibling died. Effects are significant for all outcomes except reading and are strongest for the PPVT. To give readers a sense of the difference, we estimated the effect at the ages of 6 and 12 y. For the picture vocabulary tests, the results are −3.6 percentiles at the age of 6 y but −13.8 percentiles at the age of 12 y [the results for the Peabody Individual Achievement Test assessing knowledge and application of mathematical concepts and facts (PIAT-M) are −6.3 and −12.5 percentiles]. For socioemotional outcomes, we find a mixed pattern, with no statistical significance for the SPPC outcomes but strong evidence that behavioral problems are larger if the child was older at the time of the death of his/her sibling (10-unit increase at the age of 6 y, but 22-unit increase at the age of 12 y). This would be consistent with greater opportunity to act out among older children, as well as a stronger reaction among children who spent more years with their sibling or a sibling feeling guilty for his/her sibling’s death.

Table 4.

Effect of age of surviving child on outcomes

Outcomes	Coefficient
PIAT-M
Sibling death	−0.108
	(5.092)
Surviving sibling’s age at death	−1.030+
	(0.581)
n	571
PIAT-Reading
Sibling death	1.514
	(5.694)
Surviving sibling’s age at death	−1.266*
	(0.611)
n	569
PIAT-Reading Comprehension
Sibling death	−4.331
	(4.967)
Surviving sibling’s age at death	−0.749
	(0.620)
n	475
PPVT
Sibling death	6.643
	(4.786)
Surviving sibling’s age at death	−1.699**
	(0.594)
n	340
BPI
Sibling death	−2.267
	(5.326)
Surviving sibling’s age at death	2.051**
	(0.488)
n	621
SPPC-Scholastic competence
Sibling death	−1.525
	(14.729)
Surviving sibling’s age at death	−1.434
	(1.575)
n (Wave × Child)	266
SPPC-Self-worth
Sibling death	−4.274
	(17.890)
Surviving sibling’s age at death	0.364
	(2.019)
n (Wave × Child)	265
HOME Total
Sibling death	−12.643*
	(5.446)
Surviving sibling’s age at death	1.011
	(0.668)
n (Wave × Child)	794
HOME Cognitive support
Sibling death	−3,344
	(5.914)
Surviving sibling’s age at death	0.959
	(0.631)
n (Wave × Child)	751
HOME Emotional support
Sibling death	−17.082**
	(4.518)
Surviving sibling’s age at death	0.788
	(0.703)
n (Wave × Child)	686

Models weighted to account for oversampling of minorities. SEs in parentheses clustered at the family level. All include the background controls described in SI Appendix. Wave × Child indicates the total observations for all children in all waves. **Significant at 0.01; *significant at 0.05; ⁺significant at 0.1.

Additional Heterogeneous Effects: 2. Sibling Death for Families with a Disabled Child.

For some families, the surviving siblings might have higher measured outcomes following the death of a sibling in the instances where the death concludes a long disability that required parental time and attention. Nonetheless, to be detected in our analysis, any rebound effect would need to be net of the bereavement effects of losing a family member for both siblings and parents.

In Table 5, we compare surviving children before and after the death of a sibling for those families with a disabled deceased child at any point in time. The coefficients provide evidence of significant net rebound effects in the home environment in terms of both cognitive and emotional support, suggesting an improvement for surviving children following the death of a disabled sibling. With the exception of mathematics, the results for cognitive and socioemotional assessments suggest no effect on surviving children. Note, however, that the human capital investment on children in families with a disabled sibling is likely to have been disadvantaged relative to their counterparts before the sibling’s death, as their parents might have been focusing their resources on the disadvantaged child and experiencing higher levels of stress than other parents. In addition, the expectation of a possibly imminent death might have led to a longer preparatory bereavement process or anticipatory grief (22). Thus, even though these siblings are affected by the death, the likely partial refocus of parents on the surviving child suggests the possibility of long-term improvements for this subset of children.

Table 5.

Effects on children with deceased disabled siblings

Effects	Coefficient
A. Cognitive assessments
PIAT-M	−22.959*
	(11.273)
n (Wave × Child)	113
PIAT-Reading	5.108
	(6.696)
n (Wave × Child)	111
PIAT-Reading Comprehension	−2.388
	(5.705)
n (Wave × Child)	96
PPVT	−14.307
	(11.454)
	66
B. Noncognitive assessments
BPI	1.426
	(6.202)
n (Wave × Child)	112
SPPC-Scholastic competence	−4.897
	(16.574)
n (Wave × Child)	49
SPPC-Self-worth	13.621
	(20.948)
n (Wave × Child)	49
C. Home Environment
HOME Total	17.627**
	(5.870)
n (Wave × Child)	151
HOME Cognitive support	18.741**
	(6.160)
n (Wave × Child)	151
HOME Emotional support	17.705+
	(9.479)
n (Wave × Child)	129

Models weighted to account for oversampling of minorities. SEs in parentheses clustered at the family level. All include the background controls described in SI Appendix. Wave × Child indicates the total observations for all children in all waves. **Significant at 0.01; *significant at 0.05; ⁺significant at 0.1.

Additional Heterogeneous Effects: 3. Effects of Infant Death.

Most children’s deaths happen in infancy (SI Appendix, Fig. S2). Table 6 includes an interaction term equal to 1 if the deceased child was an infant. For most outcomes, we can see that the negative impact of an infant sibling death is larger than for an older child’s death, but most of the estimates are not significantly different from zero, with the exception of those from reading and the BPI. The most likely mechanism is through parental (and perhaps a child’s own) grief associated with infant death.

Table 6.

Deceased sibling was an infant

Effects	Coefficient
PIAT-M
Sibling death	−4.363
	(3.365)
SD × deceased sibling was an infant	−4.135
	(5.165)
n (Wave × Child)	571
PIAT-Reading
Sibling death	−4.683
	(4.326)
SD × deceased sibling was an infant	−1.426
	(6.929)
n (Wave × Child)	569
PIAT-Reading Comprehension
Sibling death	−7.915*
	(3.695)
SD × deceased sibling was an infant	−1.903
	(5.441)
n (Wave × Child)	475
PPVT
Sibling death	−0.878
	(3.952)
SD × deceased sibling was an infant	−1.441
	(6.748)
n (Wave × Child)	340
BPI
Sibling death	7.217+
	(4.043)
SD × deceased sibling was an infant	3.500
	(6.498)
n (Wave × Child)	621
SPPC-Scholastic competence
Sibling death	−6.339
	(7.667)
SD × deceased sibling was an infant	−18.250
	(14.288)
n (Wave × Child)	266
SPPC–Self-worth
Sibling death	−2.542
	(7.981)
SD × deceased sibling was an infant	2.814
	(12.220)
n (Wave × Child)	265
HOME Total
Sibling death	−8.454+
	(4.313)
SD × deceased sibling was an infant	0.978
	(5.593)
n (Wave × Child)	794
HOME Cognitive support
Sibling death	1.154
	(4.414)
SD × deceased sibling was an infant	−1.370
	(5.733)
n (Wave × Child)	751
HOME Emotional support
Sibling death	−14.585**
	(4.464)
SD × deceased sibling was an infant	3.004
	(6.844)
n (Wave × Child)	686

Models weighted to account for oversampling of minorities. SEs in parentheses clustered at the family level. All include the background controls described in SI Appendix. Wave × Child indicates the total observations for all children in all wave. **Significant at 0.01; *significant at 0.05; ⁺significant at 0.1.

Discussion

This paper seeks to further understand the critical determinants and processes relevant for childhood human capital production. Although siblings are one of the most important and long-lasting relationships, very little is known about the effects of siblings on each other’s development over the life course. We build on new findings (9) that suggest that the experience of a sibling death is both quite common during childhood and has large and profound effects on human capital outcomes as a young adult by seeking to further understand the dynamic processes underlying these results. We examine the paths of achievement on cognitive and socioemotional tests before, during, and after the experience of the death of a sibling. We find large initial impacts on most outcomes that gradually subside over time, although this process often takes a decade to complete. These dynamics suggest that research that focuses on longer term adult outcomes will miss the full impact and channels of experiencing the death of a sibling during childhood. We find some evidence that a potential mechanism of these initial effects is the reduction in emotional support provided by bereaved parents following the death of a child. While children affected by the death of a sibling seem to gradually recover, reductions in the investment in emotional support from bereaved parents are still salient long after the death.

Overall, these findings both increase our understanding of the importance of siblings and families in the well-being and production of children’s human capital and suggest positive spillover effects from policies that reduce the incidence of childhood mortality: Saving one child enhances the human capital trajectories of his/her siblings, and these impacts are rarely captured in evaluations of programs that reduce childhood mortality. This is particularly relevant for the United States, where although child and infant mortality has plummeted in the past century from 3,140 per 100,000 inhabitants and 100 per 1,000 live births to 100 per 100,000 inhabitants and 4 per 1,000 live births, respectively (23–26), the rate of infant mortality is still higher than in other comparable developed countries, particularly among minorities (27). In particular, Umberson et al. (2) have shown that blacks are 50% more likely to lose a sibling by the age of 20 y than are whites. Our results suggest this disparity in exposure to sibling death will produce important divergence in child development and educational achievement by race that is presently unmeasured and a useful subject of future research.

Methods

Data.

We use the publicly available (https://www.bls.gov/nls/nlsy79.htm) National Longitudinal Survey of Youth 1979, a nationally representative sample of 12,686 US individuals who were 14–21 y old in 1979. Starting in 1986, information on every woman’s biological child was collected every 2 y, forming the National Longitudinal Survey of Youth Children survey (CNLSY79). We restrict the sample to mothers who had at least two children and suffered one death to allow comparison within bereaved families (detailed information on further sample selection is available in SI Appendix).

Children’s cognitive assessments measured in the CLNYS79 are the PIAT and the PPVT. The PIAT-M assesses knowledge and application of mathematical concepts and facts, the PIAT-R assesses reading skills, and the reading comprehension test (PIAT-C) measures the child’s ability to derive meaning from sentences that are read silently. Test scores are in age-standardized percentiles.

We also focus on two socioemotional assessments (14): the BPI, which measures the incidence and the severity of behavior problems for children, and the SPPC, which assesses a child’s sense of general self- and self-competence in his/her academic skills. We also include the Home Observation Measurement of the Environment (HOME) scale as a proxy of human capital investment by parents, which was designed by educational experts (28) to measure the quality of parental support in a child’s home environment. The total HOME scale can be divided into two subscores measuring cognitive stimulation and emotional support. Additional details on outcomes measures are provided in SI Appendix and SI Appendix, Tables A.8 and A.9.

Empirical Specification.

To examine the dynamic impacts of sibling death on children’s outcomes, we use multiple observations for each child in the data and compare outcomes before and after the experience of a sibling death among families who have experienced the death of a child. Implicitly, we assume that while the likelihood of exposure to a sibling death may differ across families based on observable and unobservable characteristics of the family, the timing of the sibling death is not related to these characteristics. We show supportive evidence for this assumption in SI Appendix, Tables A.5–A.7. Hence, we compare children’s outcomes before and after experiencing the death of a sibling and across families where the death occurred at different ages to examine the dynamic effects of the process. While pre/post-measures of each child are available, we only have 121 families who experienced a sibling’s death; thus, we face issues of restricted sample size in obtaining precise fixed effects estimates of within-family effects (SI Appendix and SI Appendix, Tables A.1 and A.2).

Thus, we first estimate the pre/post-effects of experiencing the death of a sibling as:

$$y_{ift}={\beta }_0+{\beta }_1{X}_{ift}+{\beta }_{2}Deat{h}_{jft}+{\epsilon }_{ift},$$ [1]

where the outcome (y) of child i in family f at time t is determined by child and family characteristics (X), whether the death of sibling j has occurred (Death), and an idiosyncratic term (ε). X includes background characteristics (detailed in SI Appendix). To examine the dynamic effects, we enhance Eq. 1 as follows:

$$y_{ift}={\beta }_0+{\beta }_1{X}_{ift}+{\beta }_{2}Deat{h}_{jft}+{\beta }_{3}Deat{h}_{jft}*YS{D}_{it}+{\epsilon }_{ift}.$$ [2]

We test for sensitive periods by examining whether the effects of experiencing the death of a sibling vary based on the age of the surviving child at the time of the sibling death:

$$y_{ift}={\beta }_0+{\beta }_1{X}_{ift}+{\beta }_{2}Deat{h}_{jft}+{\beta }_{3}Deat{h}_{jft}*ag{e}_i+{\epsilon }_{ift}.$$ [3]

In all models, we cluster the error terms at the family level to allow for intersibling correlation; in addition, the models are weighted for oversampling of minorities.