Parental Money Help to Children and Stepchildren

Abstract

Divorce and remarriage have reshaped the American family giving rise to questions about the place of stepchildren in remarried families. In this article, we examine money transfers from a couple to each of their children. We introduce characteristics of the family and estimate the role of shared family membership affecting all children in the family as well as the difference that stepchild status and other individual characteristics make in transfer flows. Data are from the Health and Retirement Study. There are two central results in the analysis. Overall, provision of financial help from parents to children is a family phenomenon. While help to a particular child is episodic, differences between families in provision of help were much greater than the differences in helping one child versus another within families. Second, stepchild status does differentiate one child from another within a family. Stepchildren are disadvantaged, particularly stepchildren of the wife.

Changing patterns of parenthood and domestic partnership have reshaped the American family (Cherlin, 2004; Thornton & Young-DeMarco, 2001). In the past decade, families in the United States have experienced high levels of partnership turnover (Cherlin, 2009) with three-quarters of men (78%) and two-thirds of women (69%) remarrying following divorce (Schoen & Standish, 2001). Remarriage is an “incomplete institution”, particularly when one or both partners bring children from a previous marriage into the newly formed family (Cherlin, 1978). Blended families formed after a divorce or widowhood may include joint biological children as well as stepchildren of one or the other spouse. Nearly one in five children live with a stepparent at some point during childhood (Björklund, Ginther, & Sundström, 2007; McLanahan & Percheski, 2008). In blended families, remarried or cohabiting couples and their offspring must overcome ambiguities and uncertainties about relationships (Ganong & Coleman, 2004), appropriate kinship terms (Koenig Kellis, LeClair-Underberg, & Lamb Normand, 2008), and family membership boundaries (Boss & Greenberg, 1984; Stewart, 2005).

The diversity of relationships within a family raises questions about the meaning of family membership and possible differential treatment of children. This article focuses on one aspect of family relationships by examining financial flows from parents to young adult children. We examine both the shared aspects of family membership which lead to similar treatment of all children and diversity of parental relationships which may lead to differential treatment of children in the same family.

Parent to child financial help is important for two reasons. First, providing financial assistance to adult children may help to ease their transition to adulthood (Avery, Goldscheider, & Speare 1992; Semyonov & Lewin-Epstein 2001) and assist less economically-successful children (McGarry & Schoeni 1995; 1997; Schoeni 1997). Children not receiving assistance may be handicapped compared to those who do. Second, intergenerational transfers are indicators of the nature and boundaries of the family group (Schwartz 1967). The degree to which all children are treated equally provides insight into the meaning and significance of family. Conversely, evidence that children with certain characteristics are excluded from the family transfer network may indicate their marginality to the family.

Previous Research on Stepfamily Relations

Stepchildren -- particularly stepchildren of the wife -- receive less financial support from their parents than do biological children (Killian 2004; Berry 2008; Clark & Kenney 2010). The implications of stepchild status may depend on the child’s age at the time of the remarriage. The period of elevated stress that coincides with the formation of a stepfamily tends to be more pronounced and longer in duration when stepfamilies are complex and comprised of half- or stepsibling relationships (Ganong & Coleman, 2004; Hetherington & Kelly, 2002). In contrast, the extent of destabilization tends to be shorter when children are relatively young at the time of remarriage (Hetherington & Kelly, 2002).

A possible explanation for lower investments in and transfers to stepchildren of a mother is the evolutionary argument that mothers have greater investment in their own biological children and hence will favor them (Anderson, Kaplan, Lam, & Lancaster 1999; Biblarz & Raftery 1999). Yet, research focusing on cognitive outcomes -- which reflect investments of parental time and effort (Astone & McLanahan 1991) – challenges this explanation, finding that joint children in blended families perform similarly to stepchildren (Gennetian, 2005; Ginther & Pollak 2004; Halpern-Meekin &Tach, 2008; Tillman, 2008), and that parental investments in adopted children are equal to those in two biological-parent families (Hamilton, Cheng, and Powell 2007). These latter findings suggest the possibility that family functioning in blended families or selection of families into blended or step family status may be responsible for the observed poorer performance of stepchildren.

Though there is evidence of lower parent-to-child transfers in families with stepchildren (Eggebeen 1992; Berry 2008; Clark & Kenney 2010), the distinction between being a stepchild and being in a blended family has received limited attention. The most direct evidence comes from Berry’s (2008) fixed-effects estimates indicating that stepchildren, compared to their siblings, receive substantially lower money transfers. On the other hand, indirect evidence suggests that the overall family effect, which leads to equal treatment of all children in a family, is also strong. In addition to the finding noted above that joint children in blended families have cognitive performance similar to their half-siblings who are stepchildren, substantial evidence suggests that transfers of time or money have a strong family component (Hogan, Eggebeen, & Clogg 1993; Grundy & Henretta 2006; Zissimopoulos & Smith 2010) as well as varying between children in a family. Hence there is evidence for both general family characteristics that produce similar treatment for all children in a family and within-family differences that differentiate treatment between children in a family. The analysis presented here estimates the relative magnitude of these two tendencies and identifies some of the sources of both phenomena.

Research Design

Building upon previous studies on transfers to children (e.g., Killian 2004; Berry 2008), our analysis examines the extent to which parental and individual characteristics affect money transfers. We extend this approach in three important ways. First, we develop a fuller characterization of family structure to examine whether the disadvantage of stepchildren stems from the individual child’s stepchild status or affects all children in a family, even those who are not stepchildren. We examine this issue by estimating the effect of being in a blended family – that is, one including both joint biological children and stepchildren of one or the other spouse – as well as estimating the effect of being a stepchild. The most direct previous research on transfers leads us to expect that stepchild status affects transfers over and above the effects of being in a blended family. Second, we extend the measurement of stepchild status by measuring the child’s age at the time a stepparent entered the household. Existing research suggests stress is lower when children enter blended families at younger ages. Extending this reasoning to relationships with adult children, we expect that stepchildren who were young at the time of the marriage are more likely to be treated like biological children.

Third, we consider the more general family context of transfers by partitioning the variance in provision of help into three components: variance due to family membership, variance unique to each child within a family, and variation over time in the help given to a particular child. Hence we are able to estimate the degree to which transfers are a family phenomenon and to what degree they are specific to individual children within a family. As an adjunct to this goal, we estimate the extent to which financial transfers to children in a family are correlated. That is, to what extent are children in a family treated the same? In addition, we examine the degree to which the measured characteristics of parents, children, and families can account for the correlation in transfers within a family. While previous research suggests a strong family component, there is not adequate evidence to hypothesize the relative size of the shared family variance compared to the variance due to individual children.

Data and Measures

The analysis presented here is based on the Health and Retirement Study (HRS) birth cohort of 1931–1941. These respondents constitute the parental generation, and we examine their relations with their children and stepchildren. The 1931–1941 U.S. birth cohort was the first affected by changing marital patterns (US Census Bureau 1992) that have produced an increased number of blended families.

We analyze panel data over four intervals: 1996 to 1998; 1998 to 2000; 2000 to 2002; and 2002 to 2004. Use of four intervals allows for the episodic nature of transfers from parents to children. Covariates are measured either at the beginning of the interval or within it. The outcome measure, provision of $500 or more to a child in the last two years, is assessed at the end of each interval; thus, we use each interval’s ending year as its label. Measurement of the outcome varied before 1998, and therefore these earlier waves are not utilized. The analysis presented below is restricted to sample member families in which the husband and wife were either married or coupled in 1998, and data are included for intervals in which they remained coupled at the end of the interval. To minimize the possible effects of attrition, the analysis utilizes all available data on any dyad. Therefore, data on families that drop out are used until the point that they drop out. Households with more available intervals have greater influence on the analysis because they appear in the data a larger number of times.

There are 14,544 resident and non-resident children in 3812 sample member households who are age 18 and older in the sample, producing 47,891 child observations over the four intervals. After deletion for missing data, the final sample consists of 40,763 observations on 12,334 children in 3447 respondent households. Over one-third (36%) of missing observations result from missing marriage dates for 188 households (affecting 895 individual children and 2591 child observations over the four intervals). These data are required to calculate a child’s age at the time of the marriage.

Table 1 presents means and percentages for the variables used in the analysis. Data are weighted using the household sampling weight. The response variable is a binary indicator of whether the parents report giving the child or the child’s children a total of $500 or more in financial help over the previous two years (excluding shared housing or shared food).¹ If respondents asked for a definition of financial help, they were told that it included “giving money, helping pay bills, or covering specific types of costs such as those for medical care or insurance, schooling, down payment for a home, rent, etc. The financial help can be considered support, a gift or a loan.” For nearly 18% of the observations, the respondent couple had given money to the child in that dyad.

Table 1.

Means and Percentages, Child Observations from Health and Retirement Study (N = 40,763)

Variable	Percent	Mean	S.D.
Child received money	17.6%
Respondents' Characteristics
Year of observation
1998	26.6%
2000	26.4%
2002	24.2%
2004	22.8%
Older respondent’s age
under 57	4.0%
57–61	26.3%
62–66	34.1%
67–71	24.5%
72+	11.1%
Ethnicity (vs. white)
White	85.2%
Black	5.8%
Hispanic	7.4%
Other	1.6%
Female health
excellent	14.2%
very good	35.1%
good	29.8%
fair	14.8%
poor	6.1%
Male health
excellent	12.2%
very good	30.6%
good	32.6%
fair	17.3%
poor	7.4%
Assets
negative	1.9%
zero	0.6%
LT 25K	7.2%
25–29K	6.7%
50–99K	13.7%
100–249K	27.5%
250K+	42.5%
Log income - male		10.0	1.8
Log income -female		7.7	3.7
Log other household income		5.7	4.0
Family Characteristics
N. male children		2.25	1.56
N. female children		2.20	1.62
Blended (joint biological and stepchildren)	13.0%
Children's aggregate characteristics.
within family
% college	54.5%
% male	51.2%
% married	63.6%
% have children	66.6%
Male-family received help	12.6%
Female-family received help	12.7%
Individual Child Characteristics
Child's characteristics
Male	51.2%
Married	63.7%
Has children	66.7%
Attend college	54.6%
Child's age at respondents' marriage and relationship to female
own child	83.0%
step-ages 0–9	3.1%
step- ages 10–17	5.2%
step-age 18 and over	8.7%
Child's age at respondents' marriage and relationship to male
own child	84.2%
step-ages 0–9	3.1%
step- ages 10–17	4.7%
step-age 18 and over	8.1%
Child's age
18–24	6.2%
25–34	37.2%
35–49	55.0%
50–64	1.6%
Child was born in a marriage that ended
divorce	23.7%
widowhood	4.7%
not classified	4.4%
born in current marriage	67.2%

Respondents’ characteristics

Characteristics of the respondent couple are included because they may affect the willingness or ability of the parents to provide financial help. They include: year of observation; the age of the older respondent in the couple, coded in five-year categories with under 57 as the reference category; ethnicity measured in four categories: non-Hispanic white, non-Hispanic black, Hispanic, and other. Non-Hispanic whites are the reference category. Two categorical variables measure wife’s and husband’s self-rated health, with five categories: excellent (the reference category), very good, good, fair, and poor. Net worth measures housing and financial assets of the respondent couple, coded in seven categories: negative; zero; $1–24,999; $25,000–49,999; $50,000–99,999; $100,000–249,999 (the reference category); and over $250,000. Three income measures: log husband’s earned income, log wife’s earned income, and log other household income measure income of the household, excluding children’s income. Unlike other variables that are measured at the beginning of the interval, income is measured within the interval. The measure refers to the calendar year preceding each interview. To address negative values for other household income, we logged the absolute value of the negative values, multiplied them by negative 1 (−1) and then combined them with the positive values. Zero values were given a log value of zero. The result is a metric in which changes in both large negative and large positive values have declining effects on the probability of providing transfers to children.

Family context measures

Characteristics of the family context are included because it is expected that the characteristics of the family as a group will affect transfers to individual children. They include: whether the family is a blended family including both joint biological children and stepchildren of one or the other spouse; the size of the family measured as the number of male children and number of female children listed by the respondent household; six aggregate characteristics of the children in the family measured as the proportion who attended college, are male, married; and two indicator variables of whether the female or male respondents’ families were helped by relatives while they were growing up. The survey question used is: “Before age 16, was there a time when you or your family received help from relatives because of financial difficulties?” We conceptualize these respondent reports of their family of orientation as measures of the family culture the respondents propagate in their family of procreation, making them more than simply individual characteristics.

Individual child measures

Characteristics of each individual child that may affect parents’ willingness to help financially include the child’s sex; marital status; parental status; college attendance; stepchild status; and age. Stepchild status is measured as the child’s age at the time of the parents’ current marriage based on the wife’s report of the marriage date. The biological child of the respondent is the reference category and is contrasted with under age 10, ages 10–17, and age 18 or older. Children’s characteristics are measured at the beginning of each interval, and both resident and non-resident children of the household are included. We also include an indicator, child status, to measure whether the marriage in which the child was born ended in widowhood or divorce or whether it was not possible to identify the marriage. This variable allows for a difference in parental help to children who were orphaned by the death of a biological parent or who are children of divorce. The reference category is divorce. Of those who were not born in the current marriage, over 70% are children of divorce. We code unknown as an additional category on the child status variable. HRS asks respondents how each marriage ended, and it is possible to identify the marriage into which a particular child was born. HRS collects child’s age, not year of birth. Hence there is some ambiguity in imputing an age at birth, and we allowed births one year before or after a reported marriage to count as born in that marriage. We include college attendance as a measure of the child’s economic status. HRS also asks child’s income, but we have not included it because a number of characteristics make it difficult to include in the analysis.²

Table 2 provides information on the structure of families in the sample. While Table 1 includes observations on individual children in each year, Table 2 includes respondent households once in each year, and the data are weighted using the household weight. The upper panel cross-classifies the number of children brought to the marriage by the husband and by the wife. Slightly over two-thirds of families do not include stepchildren and in nearly 9% each spouse brought two children to the marriage. The bottom panel indicates the proportion of marriages that have biological children of both spouses in addition to stepchildren. All families included in the analysis have at least one child; thus, couples who did not bring children to the marriage all have children born in the marriage. The bottom panel indicates that the probability of having a child in the marriage declines as the number of stepchildren brought to the marriage increases. There is little association between which parent brought the children to the marriage and whether children have been born in the marriage.

Table 2.

Proportion of Families with Stepchildren and With Children Born in Current Marriage^a

		Number of Children Wife
		Brings to Marriage
		0	1	2 or more
Number	0	69.6%	2.0%	5.4%
Husband	1	2.2%	1.0%	1.7%
Brings	2 or more	6.6%	3.0%	8.7%
Number of Children		Proportion with Child
Brought to Marriage by		Born in Current Marriage
Husband	Wife
0	0		100%b
0	1		74.0%
1	0		70.9%
1	1		47.8%
0	2 or more		41.4%
2 or more	0		42.4%
1	2 or more		10.2%
2 or more	1		22.2%
2 or more	2 or more		7.7%

Notes:

^aHouseholds are observed once at each wave

^bOnly households with children are included in the analysis. Hence households with no stepchildren must have at least one child in the marriage.

Model Specification

Parental money transfers to adult children may involve unequal distributions to different children within the family as well as help to any one child that varies over time. It is also likely that there is variation from one family to the next in their overall tendency to help their children. We estimate a multilevel random intercept model to allow all three types of variability. The model allows estimates of differences between children within the family and differences between families in provision of support. In addition, the estimates within the family are conditional on the specific family context estimated as a family effect (Rabe-Hesketh & Skrondal, 2005, pp. 120 – 124). The approach also accounts for the clustering created by multiple observations within a family and on any one child (Goldstein, Rasbash, Browne, Woodhouse, & Poulain, 2000; Goldstein, Bourne, & Rasbash, 2002).

There are three levels to the data: multiple observations across waves on each child’s receipt of financial help, multiple children in a family, and many families. The baseline variance components model separates variability in the binary response Y into these three levels defined as:

1. $${\mathrm{\pi }}_{\text{ijk}}=\mathrm{P}({\mathrm{Y}}_{\text{ijk}}=1)\text{ with logit }{\mathrm{\pi }}_{\text{ijk}}={\mathrm{B}}_{\text{jk}}$$ Eq. 1a
2. $${\mathrm{B}}_{\text{jk}}={\mathrm{\delta }}_{\mathrm{k}}+{\mathrm{u}}_{\text{jk}}$$ Eq. 1b
3. $${\mathrm{\delta }}_{\mathrm{k}}={\mathrm{\varphi }}_0+{\mathrm{v}}_{\mathrm{k}}$$ Eq. 1c

The subscript i refers to repeated observations of each child’s receipt of money, j indexes children in a family, and k indexes families. Level one (Eq. 1a) concerns within-child variability, i.e. the probability a child receives money, with the random coefficient (B_jk) producing a correlation among observations on a child across waves. Level two (Eq. 1b) concerns child-to-child variability within a family. The child effect, B_jk, equals a family effect (δ_k) plus an error term for each child (u_jk). The family effect, which is shared by all children in a family, produces the correlation between children in a family. In the model shown, the within-child correlation and the correlation between children in the same family are assumed to be positive. Level three (Eq. 1c) describes variation between families equal to a constant (ϕ₀) plus an individual term for each family (v_k). Combining equations 1a–c, the full model for the distribution of the binary response Y can be expressed as:

$$\text{logit }{\mathrm{\pi }}_{\text{ijk}}={\mathrm{\varphi }}_0+({\mathrm{u}}_{\text{jk}}+{\mathrm{v}}_{\mathrm{k}}).$$ Eq. 2

When the model is elaborated by adding covariates measured at each level, money transfers to a child are conditional on the parents’, child’s, and family’s characteristics, and unique elements for each child (u_jk), and each family (v_k) not captured by covariates. Because the behavior of an individual child within a family depends on the family context as well as the child’s own unique element, this is a subject-specific model (Rabe-Hesketh & Skrondal, 2005, p. 120).

Estimation results for multilevel models with a binary outcome vary by estimation method (Guo & Zhao, 2000). The models presented in this paper are maximum likelihood estimates and were derived using the GLLAMM adaptive quadrature procedure in STATA (Rabe-Hesketh, Skrondal, & Pickles, 2004, 2005). Guo and Zhao (2000) treat maximum likelihood estimates as the standard for comparing approximation methods that are computationally more efficient. The implications of weighting in multilevel models are unsettled (Rabe-Hesketh & Skrondal 2006; Brumback et al. 2010), and therefore we present unweighted results.

Results

We began by estimating the intercept-only variance components model in Table 3. Family membership accounted for 47.3% of the variance in transfers.³ This statistic is the same as the residual intra-class correlation (Snijders & Bosker, 1999, p. 224) which measures the degree of similarity between siblings – i.e., the family effect. The result indicates that family transfer behavior differs substantially across families such that differences between families accounts for nearly half of total variance. In contrast, 13.3% of the variance is between children in families and the remainder results from variation in what an individual child receives from one wave to the next, reflecting the episodic nature of transfers. Hence, the overall picture is that family membership is the predominant factor in transfers – suggesting it is a family phenomenon – and, while help to an individual child is episodic, differences between children in a family (including an individual child’s stepchild status) are relatively small contributors to total variance.

Table 3.

Variance Components Model for Money Transfers from Parent to Child Health and Retirement Study Birth Cohort 1931–1941 Observed 1996–2004

	Coef.	Std. Err.
Constant	−2.457	0.049	**
Variance within families	1.106	0.081
Variance between families	3.943	0.117
Between family as % of variance	47.30%
Within family as % of variance	13.30%

Table 4 elaborates the model by adding covariates to the multilevel logistic regression models for money transfers from parent to child. Model 1 adds parents’ characteristics to the intercept-only equation reported above. Year of observation indicates that transfers were more likely in the 2002–2004 period (the reference category) than in earlier periods and that transfers were least likely during the 2000–2002 period. Interpretation based on macroeconomic trends is speculative, particularly because economic conditions may affect the older and younger family generations differently. It is tempting to suggest, however, that the low point in 2000–2002 may have interrupted the secular trend toward greater transfers because of the US recession between March and November 2001 (National Bureau of Economic Research, 2010) which would have affected the transfers reported in 2002. Older age of the older member of the parent couple is associated with lower transfers. However, the reduction in the size of the age contrasts in Model 3, which includes child’s age, indicates that most of the parents’ age effect results from younger parents having younger adult children.

Table 4.

Multilevel Logistic Regression Model for Money Transfers from Parent to Child Health and Retirement Study Birth Cohort 1931–1941 Observed 1996–2004

	Model 1			Model 2			Model 3
	Coef.	Std. Err.		Coef.	Std. Err.		Coef.	Std. Err.
Year of observation (vs. 2004)
1998	−0.189	0.065	**	−0.175	0.063	**	−0.276	0.063	**
2000	−0.155	0.056	**	−0.143	0.055	**	−0.205	0.056	**
2002	−0.296	0.052	**	−0.290	0.052	**	−0.314	0.052	**
Older respondent’s age (vs under 57)
57–61	−0.389	0.102	**	−0.250	0.101	*	−0.094	0.101
62–66	−0.612	0.115	**	−0.331	0.113	**	−0.055	0.115
67–71	−0.969	0.133	**	−0.597	0.129	**	−0.255	0.131
72+	−1.335	0.162	**	−0.904	0.156	**	−0.475	0.158	**
Ethnicity (vs. white)
Black	−0.137	0.133		−0.010	0.122		−0.065	0.121
Hispanic	−0.859	0.156	**	−0.572	0.142	**	−0.790	0.141	**
Other	0.176	0.307		−0.014	0.279		−0.092	0.274
Female health (vs. excellent)
very good	−0.015	0.07		0.031	0.068		0.035	0.068
good	−0.165	0.078	*	−0.104	0.076		−0.092	0.076
fair	−0.294	0.095	**	−0.196	0.093	*	−0.178	0.092
poor	−0.273	0.132	*	−0.129	0.128		−0.109	0.126
Male health (vs. excellent)
very good	−0.032	0.073		−0.028	0.071		−0.026	0.071
good	0.018	0.078		0.047	0.076		0.053	0.076
fair	−0.108	0.092		−0.036	0.089		−0.042	0.089
poor	−0.132	0.124		−0.014	0.120		−0.040	0.120
Assets (vs. 100–249K)
negative	−0.768	0.188	**	−0.643	0.184	**	−0.689	0.185	**
zero	−0.742	0.393		−0.680	0.390		−0.637	0.387
LT 25K	−0.872	0.122	**	−0.697	0.120	**	−0.724	0.119	**
25–29K	−0.318	0.107	**	−0.180	0.104		−0.213	0.104	*
50–99K	−0.154	0.075	*	−0.064	0.074		−0.074	0.074
250K+	0.340	0.061	**	0.303	0.060	**	0.319	0.059	**
Log income - male	0.076	0.013	**	0.074	0.013	**	0.075	0.013	**
Log income -female	0.027	0.007	**	0.022	0.007	**	0.026	0.007	**
Log other household incomea	0.039	0.007	**	0.032	0.007	**	0.034	0.007	**
Male-family received help	0.291	0.115	*	0.293	0.103	**	0.281	0.101	**
Female-family received help	0.240	0.117	*	0.301	0.105	**	0.291	0.103	**
Blended family (joint and step)	−0.031	0.112		0.314	0.105	**	0.569	0.123	**
Children's aggregate characteristics within family
% college				0.470	0.090	**	0.519	0.108	**
% male				0.191	0.240		0.279	0.241
% married				−0.686	0.097	**	−0.036	0.111
% have children				−0.400	0.105	**	−0.521	0.120	**
N. male children				−0.331	0.042	**	−0.273	0.042	**
N. female children				−0.330	0.043	**	−0.301	0.043	**
Child's characteristics
Attend college							−0.120	0.061	*
Male							−0.159	0.049	**
Married							−0.541	0.055	**
Has children							0.394	0.060	**
Child's age (vs. 18–24)
25–34							−0.981	0.078	**
35–49							−1.440	0.090	**
50–64							−1.857	0.249	**
Child's age at respondents' marriage and relationship to female (vs. own child)
step-ages 0–9							−0.792	0.222	**
step- ages 10–17							−0.934	0.154	**
step-age 18 and over							−1.046	0.136	**
Child's age at respondents' marriage and relationship to male (vs. own child)
step-ages 0–9							0.430	0.195	*
step- ages 10–17							−0.243	0.150
step-age 18 and over							−0.510	0.139	**
Child was born in a marriage that ended in: (vs. divorce)
widowhood							0.210	0.176
not classifiable							0.107	0.151
Blended family interaction with:
Child's age at respondents' marriage and relationship to female (vs. own child)
step-ages 0–9							−1.047	0.316	**
step- ages 10–17							−0.842	0.302	**
step-age 18 and over							−0.502	0.342
Child's age at respondents' marriage and relationship to male (vs. own child)
step-ages 0–9							−1.285	0.289	**
step- ages 10–17							−0.349	0.306
step-age 18 and over							0.101	0.405
Constant	−2.608	0.223	**	−1.378	0.269	**	−0.598	0.271
Variance within families	1.132	0.083		1.165	0.084		0.735	0.069
Variance between families	3.021	0.155		2.168	0.12		2.205	0.117
Between family as % of variance	40.6%			32.7%			35.4%
Within family as % of variance	15.2%			17.6%			11.8%

Notes:

^aOther household income has negative values. We logged the absolute value of negative incomes. The logged absolute value was then multiplied by negative 1 (−1) and combined with the positive values. Zero values were given a log value of zero.

Hispanics are less likely to provide money to children. Poor health for the mother is associated with a reduced probability of helping children, but this relationship is non-significant once children and family characteristics are added to the equation. Higher wealth levels are associated with a higher probability of help to children. These relationships are attenuated somewhat once child and family characteristics are added to the equation, indicating that higher asset levels are associated with family characteristics such as the number of children and proportion who have attended college. Income level is measured separately for men, women, and other shared household income (excluding children’s income). Income of the male has a substantially larger effect than either female income or other income. The associations of these variables with transfers are not substantially changed by the addition of family characteristics.

The parent’s experience while growing up also is associated with transfer behavior. If either parent’s family received help from a family member when he or she was growing up, the family is more likely to help their children. The blended family variable is not significant in Model 1, indicating that, net of parents’ characteristics there is no overall difference in transfers between blended families that include both joint children and stepchildren and other families. Blended status is significant and positive in Model 2 which includes the aggregate children’s characteristics. It is also significant in Model 3, but Model 3 includes interactions involving this variable. We delay discussion of the blended results until later so that the implications of the interactions are included. The inclusion of parental characteristic in Model 1 reduces the proportion of residual variance due to family membership to 40.6% of total variance compared to 47.3% in the intercept-only model.

Model 2 adds children’s aggregate characteristics that also measure the composition of the sibship. Measures of the aggregate characteristics of the children in the family indicate that each child is more likely to receive help in smaller families. Coefficients for number of male children and number of female children are nearly equal. Having a higher proportion of children who attended college – possibly an indicator of parental willingness to pay for a college education for children – is associated with a higher probability of transfers. In Model 2, having a higher proportion of children who are married is associated with a lower probability of transfers. However, this relationship is not significant in Model 3, which suggests that much of the effect of the proportion of married children is attributable to the age or marital status of individual children, as opposed to aggregate family characteristics. Having a higher proportion of children who have produced grandchildren is associated with a lower probability of transfers. The bottom of the column indicates that the residual variance due to family membership is reduced to 32.7% in this model. Hence the composition of the sibship is an important reason why families differ from each other.

Model 3 adds the individual child’s characteristics, including an interaction of blended family status with the individual child’s stepchild status. Examining individual child characteristics, the model indicates that younger, female, and unmarried children, as well as those with lower levels of education, are more likely to receive help. Holding these characteristics constant, children who have produced grandchildren receive higher levels of financial help. In combination with the negative coefficient for the proportion of children who have produced grandchildren, the positive coefficient for the individual child’s having produced grandchildren suggests that the value of grandchildren as attractors of gifts declines as the characteristic becomes more common in the sibship. The negative coefficient for an individual child’s having attended college, controlling for the proportion of children in the family who attended college, indicates a slight advantage in the family for children who have not attended college.

We measure stepchild status with two variables. For each spouse we created a categorical measure of whether the child is a biological child or a stepchild who was aged 0–9, 10–17, or 18 and over at the time of the marriage. The effect of these variables differs between blended and non-blended families so we also include the interaction of each variable with blended status.

Because interaction results are difficult to interpret by simple inspection of coefficients, Table 5 presents these interactions as predicted probabilities of a transfer to a child depending on the child’s stepchild status. The table is for an unmarried female child in a four child family.⁴ The top half of the Table shows the predicted probability of a transfer to a joint child in a four child family, comparing children in step families to those in non-stepfamilies. A transfer to a joint child has a greater probability of occurring in families with step children. Joint biological children in families without stepchildren have a predicted probability of transfer receipt of 21.8%. In contrast, a joint child in a family with stepchildren has a predicted probability of 33%.

Table 5.

Predicted Probability of a Money Transfer from Parents to a Female Child in a Four Child Family When Random Effects Equal Zero

	Predicted probability of a transfer to a joint child in a family
	with stepchildren	without stepchildren
	33.0%	21.8%
	Predicted probability of a transfer to a stepchild in a family
	with joint children	without joint children
Step of male
Child's age at marriage
0–9	17.3%	30.0%
10–17	21.4%	17.9%
18 and older	24.6%	14.3%
Step of female
Child's age at marriage
0–9	7.2%	11.2%
10–17	7.7%	9.8%
18 and older	9.5%	8.9%

Stepchildren in a family with joint children have a substantially lower probability of receiving transfers compared to a joint child. Stepchildren of the male respondent have predicted probabilities of receipt ranging from 17.3% to 24.6%. Surprisingly, these probabilities are higher among children who were older at the time of the marriage. Stepchildren of the female have probabilities of transfer receipt that are less than half of those of the male’s stepchildren. Probability of receipt for the female’s stepchildren are also slightly higher for those who were older at the time of the parental marriage.

In families without joint children, the probability of receiving help is higher than in families with joint children but the stepchildren of the female remain disadvantaged. In addition, the association with age at the parental marriage reverses direction. Those who are younger at the time of the marriage are more likely to receive help.

Returning to Table 4, the model with children’s characteristics and interactions indicates that 35.4% of the residual variance is due to family membership, 11.8% is due to variation between one child and another in the family, and nearly 53% results from the episodic nature of help to any one child. The estimate of variance due to family membership is slightly higher than the preceding model as can happen in maximum likelihood estimation.

Discussion

There are two central results in the analysis. First, provision of financial help from parents to children is a family phenomenon. While help to a particular child is episodic, differences between families in provision of help were much greater than the differences in helping one child versus another within families. As a result, research on which child in a family receives help addresses only one part of the phenomenon. Overall, nearly half of the variance in helping children is accounted for by family membership and only 13.3% results from differences among siblings within a family. Even after adjusting for parental characteristics, including income, the family history of giving in both spouses’ families, and the aggregate characteristics of children, about one-third of the residual variance still resulted from family membership. Hence, this family effect reflects more than measured characteristics. Some parents give and others do not, a phenomenon that has been observed in other studies as well (Henretta, Soldo, & Van Voorhis, 2011; Henretta, Wolf, Van Voorhis, & Soldo, 2012; Seltzer et al., 2005). An important question for future research is whether the lack of money transfers in some families is associated with lower levels of solidarity measured in other dimensions.

Second, stepchildren are disadvantaged within families, particularly stepchildren of the wife. One possible explanation is that this pattern reflects the predominance of mothers as custodial parents after divorce (Kreider 2005). If a child lives with her mother following the biological parents’ divorce, the biological father and stepmother will have relatively less contact with her and might be less likely to provide money help after the child reaches adulthood. The husband’s children from a previous marriage are disadvantaged after remarriage compared to children born in the present marriage or children the wife brings to the marriage. Moreover, stepchildren who are younger at the time of the new marriage – who are the ones who may lose contact if they live with the other biological parent – are less likely to receive help than children who were older at the time of the marriage. An alternative explanation for the finding that blended families are more likely to aid stepchildren who were older at the time of the marriage may lie in the comparisons parents make. Younger stepchildren may be relatively disadvantaged because they are more likely to be compared to joint children in the new family and may be in more direct competition with them because they are closer in age. This explanation has the advantage of being consistent with the finding that in families without joint children, younger stepchildren at the time of the marriage are more likely to receive money help than similarly-aged children in families with joint children.

Both the relatively privileged position of stepchildren of the male and the tendency toward coresidence with the mother after a divorce are consistent with the genetic argument that women are more concerned for their biological children – that is, they are more altruistic toward them – than are men. Overall, the disadvantaged position of the wife’s stepchildren suggests their marginal position in the family.

In addition to these central findings, the analysis provides additional insight into between-family differences as well as within-family differences. While much of the family effect is unexplained by covariates, the family histories of both husband and wife are important. Those who report that their families received help while they were growing up are more likely to assist children. This measure may be considered a direct measure of family culture. Family differences appear to persist across generations, and future data collection might well devote more attention to measuring this culture.

The research has several limitations. As noted in the earlier definition of the variables used, our measure of child’s need is child’s education, a limited measure. We have not utilized child’s income because it was measured inconsistently over the waves used and has extensive missing data. In addition, we have used a binary indicator of money help to parents, and have not utilized the full information on amount because we do not have suitable instruments to identify a selection equation to address the large majority of respondents who do not provide any money help. It is not clear what the effect of these two data issues on the results might be. Finally, the analysis is limited to the households of the birth cohort of 1931–41 during the period 1996–2002. Hence there may be both cohort and period effects. This cohort, as parents of the post-war baby boom, had higher fertility than subsequent cohorts and therefore have more children whom they might assist. The period observed was a relatively prosperous one economically with the exception of a short recession in 2001, and therefore these results might not be duplicated in a different period.

Based on the two main findings, the results of this analysis suggest an important agenda for family research. First, the results suggest a need to refocus research on intergenerational transfers to examine family differences instead of differences between children in a family. Most of the variance in providing help to either an ascending (Henretta, Soldo, & Van Voorhis 2011) or descending generation occurs between families, but we currently have only a limited understanding of the family differences that produce this variance. Second, there are important within-family differences that result from stepchild status. These differences in relationship continue after children become adults and differ depending on whether the remarriage produced joint children. Hence there is an interaction between the family characteristic of a blended family and the individual child’s status, suggesting a need to consider the diversity of family structures in order to develop a fuller understanding of the implications of being a stepchild.