Mediators of the Relation of Family Income with Adolescent Behavior Problems and Cognitive Achievement: Material Hardship, Parent Distress and Parent Support

Abstract

Structural equation modeling (SEM) with longitudinal survey data was used to test a proposed developmental model of the association of family income (with children aged 6–9) to parent behaviors (for children at 10 years of age) and adolescent cognitive achievement and behavior problems (at age 15). Data from the Child Development Supplement (CDS) and Panel Study of Income Dynamics (PSID) provided a representative US sample (n = 953). The SEM measurement model of parent behaviors showed two robust latent variables representing parent distress (based on two measures) and parent support (composed of four measures of parent investment, cognitive stimulation, emotional warmth, and educational expectations for the child). The SEM structural model indicated that the relation between average family income between 1998 and 2001 for young children (ages 6–9) and adolescent cognitive achievement and behavior problems in 2007 (age 15) was almost entirely mediated by parent distress, parent support and material hardship, all measured in 2002. Results suggested that the structural model was strongest (RMSEA = .08) when all three mediating variables were included. These results provide a clearer picture of the developmental mechanisms by which family income becomes associated with adolescent cognitive achievement and behavior problems over time.

Many studies have examined the link between children’s household income growing up and development of child skills and behaviors (Duncan, Magnuson, & Votruba-Drzal, 2015), but they disagree on what mechanisms account for this link. Researchers have suggested three mechanisms: (a) low income and resulting material hardship (e.g., food insecurity or housing instability) increase parent distress or strain (e.g., marital conflict, depression), which leads to negative parenting behaviors (e.g., spanking) that reduce child cognitive achievement and increase behavior problems; (b) high income supports parent investment of time and money in child enrichment (e.g., procuring educational materials and activities and attending school meetings), which stimulates child cognitive achievement and reduces behavior problems; and (c) high income, social class, and cultural beliefs are correlated with more supportive parenting behaviors (e.g., cognitive stimulation, emotional warmth, and parent expectations for high child educational achievement) according to cultural theories of parent behaviors and values.

There is also a gap in the literature on family income and child behaviors in terms of the ages of children studied. A recent review summarized results of extensive past research about the relation of early family income to child behaviors, but few studies explored the relation of early family income and adolescent behaviors (Duncan, Magnuson, & Votruba-Drzal, 2015). One researcher investigated the relation of early family income to pre-adolescent child behaviors (Votruba-Drzal, 2006), and a review of economic studies on the effects of family income identified only one longitudinal study that examined the relation between early family income and adolescent cognitive achievement (Mayer, 2010); its author concluded that adolescent cognitive test scores were lower if the family had lived in poverty during the period of early childhood to early adolescence or during early adolescence alone (Guo, 1998). Thus there appears to be an important gap in our understanding of the relation of early family income to parent behaviors and to adolescent cognitive achievement and behavior problems.

In this developmental study, I explore these gaps using a large random sample of US families with longitudinal data. Prior to other variables, average yearly family income was calculated for the period of 1998–2001, when children were ages 6–9. Parent behaviors were measured in 2002 when children were 10 years on average. Adolescent cognitive achievement and behavior problems were measured in 2007, when the children were 15 years old, on average. The availability of longitudinal data allows comparison of results from multiple models of the developmental influence of early family income and later parent behaviors on adolescent behaviors.

Prior research shows that the mechanisms of parent distress, parent investment, and parent support are individually related to child cognitive achievement and behavior problems, but has less frequently tested them in conjunction with one another (Gershoff, Aber, Raver, & Lennon, 2007; Raver, Gershoff, & Aber, 2007). A joint test of the mechanisms is useful for two reasons. First, the parent behaviors involved in these three explanations appear interrelated and complementary, so it is helpful to test their association with income at the same time in order to assess relative strength of relationships. Second, the three mechanisms represent variables that theoretically mediate the relationship of income to child behaviors, but the possibility of mediation should be tested for all three mechanisms at the same time in order to assess their orthogonality.

My hypotheses are based on two assumptions implicit in earlier theory and which are tested here through confirmatory analyses. First, the three theories assume that the three parent behaviors are separate, independent constructs; I test this assumption by comparing alternative factor models examining interrelations among parent behaviors. Second, the hypotheses assume that family income is related to subsequent parent behaviors and that they in turn are related to later adolescent behaviors; assessment of family income, parent behaviors and child behaviors at different time periods does not ensure that the associations are causally determined but it provides greater confidence of sequential effects than would be the case with cross-sectional data. This permits examination of potential mediation effects between two sets of variables through a third set of variables, all measured at different points in time, so the sequence is known. I examine this expected mediation model through consideration of a structural model that assesses the relative strength of relations of early family income (averaged for ages 6–9) with material hardship and parent variables (age 10), and with adolescent cognitive achievement and behavior problems (age 15). Further, I compare model goodness of fit when different mediating variables are included or excluded from the overall model.

1. Theoretical Issues

As noted above, there are three theoretical frameworks about parent behaviors that may mediate the relation of income to adolescent behaviors: parent distress (or strain) theory, parent investment theory, and cultural theories of parent behavior. Based on these three frameworks, I propose the mediation model shown in Figure 1, leading to the hypotheses discussed below. My primary purpose is to propose and test this integrated model, which combines and extends hypotheses about the mediating effects of family and parent behaviors derived from the three theoretical frameworks about the relation of early family income to adolescent cognitive achievement and behavior problems.

Figure 1.

Proposed Model

1.1. Parent Distress Theory

The first framework, parent distress theory, posits that low income causes parent distress; empirical results show that low income parents express less warmth and support, do not engage in cognitive stimulation, or hold high educational expectations in relation to the child, as compared to high income parents (Conger et al., 1992; Conger, Conger, & Martin, 2010; Conger, Ge, Elder, Lorenz, & Simons, 1994; Elder, Conger, Foster, & Ardelt, 1992; McLeod & Shanahan, 1993). Findings also show that these parent behaviors are associated with child behavior problems (e.g., aggression or depression). Results from two major studies have extended the model to also consider material hardship experienced by the family; hardship mediated the relations of income to parent behaviors and child behavior problems and cognitive skills, albeit in children that were slightly younger than those studied here (Gershoff et al., 2007; Raver,et al., 2007). Income is a fluid asset that is readily available for use and easily measured in currency; it influences parent behavior and can be influenced through policy interventions that increase family income (e.g., the Earned Income Program of the US federal government). Nonetheless, a family’s financial situation cannot be reduced to income alone. Measures of material hardship (e.g., specific financial difficulties, such as eviction or difficulty paying bills) capture important additional information about a family’s financial situation, because family income likely has a nonlinear “threshold effect” on the amount and kinds of financial troubles the family experiences, that vary with regional costs of living. Material hardship or economic strain may also be directly associated with child behavior problems or cognitive achievement, independent of their association with parent distress; children may be unaware of actual income levels but are often cognizant of specific sources of material hardship, such as home instability or food insecurity, that then influence their behavior. To summarize, research from Gershoff, Raver and colleagues suggests that low income and material hardship are related to child behavior problems; further, low income and material hardship are related to parent distress, which is related to child achievement (inversely) and child problems (positively).

Parent distress theory can also be extended by considering relations to parent investment theory and cultural theories. Parents in a family with low income or high levels of material hardship are less likely to engage in parent involvement or in supportive parent behaviors, given the time and financial constraints that confront them. Analyses reported below examine the interrelations among material hardship, parent distress, parent investment and parent support, consistent with the model shown in Figure 1; past research has rarely considered all of these variables and their corresponding theories simultaneously (with the exception of the two studies by Gershoff, Raver and colleagues).

1.2. Parent Investment Theory

Second, parent investment theorists argue that high income permits parents to invest more money and time in materials (e.g., books or musical instruments) or activities (e.g., attendance at cultural events) that provide cognitive stimulation to the child and lead to development of cognitive skills (Mayer, 1997). Empirically, there are significant differences between low- and high-income families in terms of the level of investments made in materials available and activities provided to the child (Kaushal, Magnuson, & Waldfogel, 2011). Research has also shown differences in parent school involvement and attendance at activities because high income parents may have more discretionary time to engage in such activities (Dumais, Kessinger, & Ghosh, 2012; Lee & Bowen, 2006; Sui-Chu & Willms, 1996). For example, Lee and Bowen (2006) found that parent investment, measured as frequency of physical attendance at school events, was related both to family socioeconomic status and to child academic achievement. Even if school events do not involve the child, they embed the parent and family in the community, providing social capital and knowledge that may help the child’s overall cognitive and behavior development (Coleman, 1988). In this way, family income is associated with parent investment, which in turn is related to child cognitive achievement and behavior problems.

Researchers of parent investment theory have not directly linked their work to that of researchers on parent distress theory, although it seems likely that distressed parents are less likely to attend school events or take their children to cultural events. Engaging in events outside the home may simply require too much energy for people who are already feeling psychological distress. In simplistic terms, parent distress theory suggests that low income parents are more likely to be distressed parents who therefore lack capacity to fully support their children because they lack time and other resources (e.g., transportation) needed to fully engage in activities at the child’s school or in cultural venues (e.g., museums). Neither theory claims to provide the sole explanation of the relation of income to child cognitive achievement or behavior problems, which is why variables from both perspectives are grouped together in Figure 1.

1.3. Cultural Theory

Finally, cultural theories of parenting focus on parents’ ideas, beliefs, and assumptions, and how they form particular “clusters” by social class. In the Bourdieusian tradition they are integrated and overarching ways of doing things that are “taken for granted” or expected, referred to as “habitus” (Bourdieu, 1977; Lewis, 1966), that confer advantages through exchangeable forms of capital (e.g., economic resources, social ties, and cultural repertoire). Other studies treat these clusters as more explicit values (Kohn, 1959, 1963; Kohn & Schooler, 1969) or discourses of parenting (Lareau, 2011) that are associated with (but not wholly determined by) economic class. Lareau ([2003], 2011) found that parent behavior which she described as “concerted cultivation” was related to child behavior and cognitive achievement (Weininger, Lareau, & Conley, 2015). Results from several studies suggest that the relation between income and cognitive skills is partially mediated by habitus, often measured as the parent’s expectations of the child’s college aspirations (Bodovski & Farkas, 2008; Dumais, 2002; Gaddis, 2013; Irwin & Elley, 2011), or as concerted cultivation of the child’s development (Cheadle, 2008, 2009; Cheadle & Amato, 2011; Guo & Harris, 2000; Redford, Johnson, & Honnold, 2009). Cultural theory differs from the parent distress model by arguing that cultural or ideological differences among parents—rather than their distress—lead to differences in their behaviors, which then are related to child behaviors. Cultural theory also extends the parent investment model by including investment as only one of many supportive parent behaviors pertaining to habitus or concerted cultivation. As suggested by Figure 1, the predictions of cultural theory are not inconsistent with parent distress theory or parent investment theory, but instead focus on a supplemental explanation of child behavior problems and cognitive achievement.

1.4. Hypotheses

In all three theories, average family income of young children is predicted to be associated with material hardship and parent behaviors (distress, investment, and concerted cultivation). They, in turn, are expected to be related to adolescent cognitive achievement and behavior problems. The relation of family income to adolescent behavior is expected to be largely indirect, as mediated by material hardship and parent behaviors in the pre-adolescent years.

Hypothesis 1a. Family income is inversely related to material hardship.

Hypothesis 1b. Family income is inversely related to parent distress.

Hypothesis 1c. Family income is positively related to parent investment in children.

Hypothesis 1d. Family income is positively related to parent behaviors of cognitive stimulation, warmth, and education expectation expressed toward children.

My second set of hypotheses focuses on material hardship, parent distress, and parent support as mediators of the relation of family income to adolescent behaviors. Hypotheses 2 and 3 are based on parent distress theory and its argument about how parent distress and material hardships mediate the relation of family income to child behavior problems. Hypotheses 4 and 5 are based on investment theory and cultural theory notions about how parent behaviors mediate the relation of the independent variable, family income, with the dependent variables of child cognitive achievement and behavior problems. In order for mediation to occur, the dependent variables must be associated with material hardship, parent distress, parent investment, and other supportive parent behaviors.

Hypothesis 2. Material hardship is positively related to adolescent behavior problems and inversely related to adolescent cognitive achievement.

Hypothesis 3. Parent distress is positively related to adolescent behavior problems and inversely related to adolescent cognitive achievement.

Hypothesis 4. Parent investment is positively related to adolescent cognitive achievement and inversely related to adolescent behavior problems.

Hypothesis 5. Parent behaviors of cognitive stimulation, warmth, and education expectation are positively related to adolescent cognitive achievement and inversely related to adolescent behavior problems.

2. Method and Materials

In this study I postulate that the relations of family income to adolescent behaviors are mediated by two sets of variables: material hardship and parent behavior (parent distress, parent investment, and concerted cultivation). First I use structural equation modeling (SEM) to test a measurement model to assess whether the parent behaviors represent separate and independent constructs. Second I use SEM to test a structural model to determine whether material hardship and the parent behaviors partially or fully mediate the relationship between family income of young children and their behaviors as adolescents. Finally I examine goodness of fit of the structural model relative to simpler alternative models that exclude the mediating variables (to show the comparative importance of the different mediators).

2.1. Participant Population and Sample

This study was based on secondary analysis of the Panel Study of Income Dynamics (PSID), a longitudinal study that began with a representative sample of U.S. families in 1968, followed them and their descendants to the present in yearly or biyearly waves, and added a refresher sample in 1997 to make it representative of the contemporary US (PSID, 1968). The 2001 wave of the PSID provided baseline demographic information about the family in that year, and waves of the PSID from 1999, 2001, and 2003 provided total household income data for the years 1998 through 2001. Data about children were taken from the Child Development Supplement (CDS), a supplemental longitudinal study that provides additional information on PSID respondents’ children (PSID, 1968). The CDS began in 1997 with a subsample of 3,563 children aged 0–12 in 2,394 PSID households and has since collected two additional waves of information from that subsample, in 2002 and 2007.

Several factors reduced the size of the usable sample. First, attrition of 649 children between the 1997 and 2002 waves, “graduation” of 1,413 children of age 18+ to the Transition into Adulthood survey in 2007 (which did not collect the same measures and therefore could not be used in this study), and 97 non-responses to the 2007 child cognitive tests reduced the sample to 1,397 cases with data in all three waves. Following Davis-Kean (2005) I randomly removed 389 siblings to preserve independence of cases; the CDS included more than one child from these families, which may have led to biased results. I also removed 55 cases because they were part of very small ethnic categories (those not classified as White, African American, or Hispanic) for which results might be unreliable (Davis-Kean, 2005). Because all of these cases were missing systematically, it was inappropriate or impossible to use them even with full-information maximum likelihood (FIML) estimation, which I used to address missing data. (See section 2.3, Analytic Approach, for more information on FIML.) The final sample (n = 953) represented 27% of the original subsample in the first wave of data, but still had sufficient power for the analyses completed. The CDS provided a data set that was ideal for testing the hypotheses because it provided longitudinal data from a sample constructed to be representative of US families, from 1998–2007, when (on average) children were aged 6–15 years.

On average, children in the sample were 10 years old and had 1.28 siblings in 2002. Approximately half the children were male (51%) and white (48%). Median family income in the multi-year period of 1998–2001, adjusted for inflation, was $45,100.

In summary, results from this study should be broadly generalizable to the population of US children with average ages of 6–15, in the time period of 1998–2007; the original sampling frame was random and stratified. This time period was economically stable, coming before the Great Recession that started in 2008, so the results should not be affected by atypical economic conditions. Analyses were limited to White, Black, and Hispanic children, because the representation of other ethnicities was too small to provide generalizable results. All analyses controlled the effects of demographic variables so they should not have influenced results regarding family income in relation to parent behaviors or adolescent behaviors.

2.2. Measures

The PSID and CDS provide extensive measures of variables specified in the hypotheses, using standard scales or carefully developed new scales, as well as thorough measurement of demographic variables. Data were collected with multiple methods to increase reliability of measurement, including interviews with the parent or primary caregiver (PCG); interviews with the child; observation of PCG and child interactions in the home environment; and cognitive testing of the PCG and the child.

Adolescent behaviors, including cognitive achievement tests and measures of behavior problems, were assessed in 2007, when the sample mean was 15 years of age. Interviews with PCGs in 2002 (mean child age of 10) provided measures of material hardship and parent behaviors. I calculated family income as an average score based on data about the years 1998 through 2001 (when mean child ages were 6–9 years) from PSID waves in 1999, 2001, and 2003.

2.2.1. Adolescent Behaviors, 2007, Age 15

2.2.1.1. Cognitive Achievement.

Adolescents completed the Woodcock-Johnson Revised Tests of Achievement (Woodcock, 1977), measured according to the commonly used Woodcock-Johnson Psycho-Educational Battery-Revised. These tests compare test-takers’ results with those of national averages for the child’s age. The test scores were count variables of number of items answered correctly on three separate tests: Letter-Word Identification, Passage Comprehension, and Applied Problems.

2.2.1.2. Behavior Problems.

I used two subscales of the Behavior Problems Index (BPI) that were developed (Peterson & Zill, 1986) for the National Longitudinal Survey of Youth and used in the CDS: externalizing behavior (α = .86), measured with 16 items reflecting aggressive and antisocial behavior toward others (e.g., bullying, disobedience, impulsivity); and internalizing behavior (α = .83), measured with 13 items indicating levels of depression, anxiety, and loneliness (e.g., fearful, withdrawn, unhappy, worries often). PCGs rated each item for the child, producing two scores: number of externalizing behaviors exhibited and number of internalizing behaviors exhibited. Scale items were first recoded as binary variables (0 = few or no behavior problems of the type and 1 = problems that were sometimes or often true) and then summed to create the scores. These two scores are widely used in studies of behavior problems (e.g., in Gershoff et al., 2007) because they represent conceptually distinct scores.

2.2.2. Parent Distress, 2002, Child Age 10

I used two measures of number and severity of stressors experienced by PCGs. Interviewers asked PCGs about their frequency of emotional distress (e.g., nervousness, hopelessness, or worthlessness) experienced in the past month using six items from the K-6 Non-Specific Psychological Distress Scale as a measure of depression and anxiety (Kessler et al., 2003). The scoring procedure for this scale is to rate each item on a Likert scale of frequency of occurrence (1 = never to 5 = often) and then average the ratings. The survey also constructed an Aggravation in Parenting Scale with seven items (e.g., being a parent is constraining, exhausting, difficult, or frustrating) which were rated on a Likert scale (1 = not at all true to 5 = completely true) and averaged.

2.2.3. Parent Behavior, 2002, Child Age 10

I used four measures of parent behavior, including two scales from the Home Observation for Measurement of the Environment-Short Form (HOME-SF): the Cognitive Stimulation subscale and the Warmth and Support subscale (Caldwell & Bradley, 1984), following coding procedures used in the National Longitudinal Survey of Youth (Baker, Keck, Mott & Quinlan, 1993). Both scales have been widely used and have high internal reliability and validity (Mott, 2004; Smith, Brooks-Gunn, & Klebanov, 1997), so scores on the subscales are used in the model (rather than the individual items) to facilitate comparison with other studies. I also used a count variable of Parent Involvement and a recoded measure of Parent Education Expectation of the child’s achievement.

2.2.3.1. Cognitive Stimulation.

This 15-item subscale from the HOME-SF concerned the child’s intellectual environment, including (a) access to cognitively stimulating materials (e.g., books, magazines, newspapers); (b) organized activities (e.g., mother provides toys, family encourages hobbies, child is taken to theatre and museums frequently); and (c) physical environment (i.e., home environment is dark, monotonous, cluttered, clean, safe). Interviewers observed the home or asked questions about these characteristics and reported results in ordinal indicators (e.g., athletic or sports team participation was coded from 1, “less than once a month,” to 6, “every day”). The items were then recoded as binary variables (e.g., cognitively stimulating materials present versus not, participated in activities versus did not, and safe or clean environment versus not) and added up. This helped to ensure that items were comparable and reliable, and allowed the subscale to take into account different items for different age ranges (e.g., being read to for age 3–5 children versus reading for older children). The resulting scale ranged from 0 to 15, depending on the number of subscale items that were applicable.

2.2.3.2. Warmth and Support.

Ordinal items in this subscale from the HOME-SF were recoded as binary variables and summed. I then rescaled the subscale to a standard range of 1–5 because different numbers of questions were asked of different child age categories (6–9 and 10+ in the 2002 wave of data collection). Seven items measured interaction with the child during the interview based on interviewer observation (e.g., PCG talked with, hugged, or spanked child). Eleven items asked PCGs to rate frequency of child interactions with family and friends, explain expectations concerning child behaviors (e.g., in completing chores), and describe typical disciplinary actions. High scores represented behaviors that indicated greater emotional warmth toward the child and less corporal discipline. The total score showed the number of supportive PCG behaviors toward the child, as rated by the interviewer and the PCG.

2.2.3.3. Involvement.

PCGs indicated their attendance at the child’s school activities (e.g., PTA meetings, volunteering to help in school, and school events). I counted the events attended. Other studies of parent involvement have also counted number of events that parents physically attended (e.g., Lee & Bowen, 2006). The total score on event attendance provides a measure of PCG’s time investment in activities at the child’s school.

2.2.3.4. Education Expectation.

PCGs were asked the number of years of education that they thought the child was likely to attain. I recoded the responses by subtracting 12 years to get the expected number of years of education beyond high school (12 years), bottom-coding at 0 years (which represents high school completion or less). Thus, 2 years represents an associate’s degree, 4 years represents a bachelor’s degree, and higher values represent advanced degrees. Parents’ expectations of the child’s college aspirations are often used to measure habitus (e.g., Bodovski & Farkas, 2008; Dumais, 2002; Gaddis, 2013; Irwin & Elley, 2011), meaning the cultural views that parents hold about their educational goals for their children. These views may then influence parent behavior to reach those goals (e.g., seeking cognitive stimulation and pushing children to succeed academically).

2.2.4. Material Hardship, 2002, Child Age 10

I used the Economic Strain Scale (Conger & Elder, 1994) to measure material hardship (i.e., the level of economic difficulties and adjustments made in response to those difficulties). It was constructed as the count of 15 yes/no items that asked whether the family had particular economic problems (e.g., filed for bankruptcy) or made changes to cope with financial difficulties (e.g., had foregone large purchases) over the past year. Scores could range from 0–15 and represented the total number of difficulties experienced.

2.2.5. Average Family Income, 1998–2001, Child Ages 6–9

PCGs reported their family income for the past year and the year before that in the 1999, 2001, and 2003 waves of the PSID. Average log household income (based on annual income measures adjusted for inflation to 2002 dollars) was calculated from multiple years, a measure recommended by Mayer (1997) to represent “permanent income.” Models using other specifications of income (e.g., spline function or non-log-transformed income) produced similar results, and other measures of socioeconomic status added to the analyses (e.g., wealth in 2001) had no effect net of average income; therefore, I used this measure of the average of log household income from 1998–2001, when child age was 6–9 years (on average).

2.2.6. Control Variables

I controlled for demographic background of the PCG and child in the same time frame as the independent variable (i.e., family income); this ensured the temporal primacy of the controls and the independent variable because they were measured before the mediating variables and dependent variables. The control variables were selected because they might have been confounded with income. Controlling for these variables contemporaneous with income meant that the statistical associations of income could be examined without contamination from the control variables. For example, parents’ highest level of education was a categorical variable constructed from the 2001 wave of the PSID as the household head’s education or the education of the spouse or partner (if present), whichever was higher; it was coded with categories of less than high school, high school diploma or GED (reference category), college (bachelor’s or associate) degree, and advanced degree. It was controlled in 2001 so that the statistical effects of family income measured in the same time period could be assessed, independent of parents’ education. Categorical measures were used in order to test for possible non-linear relations. I measured two additional control variables based on 2001 data: number of siblings, coded as a count variable (top-coded at eight siblings); and whether both biological parents were present in the household (with both present as the reference category). The presence of more siblings and fewer parents usually affects income level so they were controlled to allow a more accurate estimation of income’s effects. Child’s age, in years, was measured as of 2002. Three time-invariant control variables were used in the structural model: child’s ethnicity, with categories of White (reference), Black, or any Hispanic ethnicity; parent’s cognitive achievement, measured as raw Woodcock-Johnson passage comprehension score in 1997; and child’s gender, with male as the reference category.

2.3. Analytic Approach

I used structural equation modeling (SEM) to analyze the data for three reasons. First, it calculated a measurement model for latent variables to assess whether the measures represented coherent constructs that could be measured based on their relations to the indicator variables. Second, the structural models produced by SEM were used to test the hypotheses and to assess the relative strengths of income and the four hypothesized mediators, by examining goodness of fit for the alternative models. Third, SEM was ideal for performing formal tests of the mediating relations proposed in the hypotheses, including classification of the relations of income to adolescent behaviors into direct and indirect effects—that is, how much of each total association (e.g., between income and child cognitive achievement) is accounted for by the intervening variables. I used bootstrapping with 1,500 replications to provide reliable assessment of relative direct, indirect and total effects of key variables on the dependent variables (Preacher & Hayes, 2004, 2008).

To address missing data, I used Stata’s “MLMV” (maximum likelihood with missing values) option for SEM, which estimates the model using all available non-missing information from cases without using listwise deletion. This is also referred to as full information maximum likelihood (FIML) and provides less biased parameter estimates in SEM than listwise deletion (Enders & Bandalos, 2001) where data are missing at random (MAR) or missing completely at random (MCAR), as is true for these data. FIML estimates a likelihood function for each individual case based on non-missing data on the variables so that all the available data are used. Further, the use of MLMV rather than multiple imputation for SEM appears to produce less bias with longitudinal latent growth models when using small samples with intermittent missing data (i.e., at least MAR) and non-normality (Shin, Davison, & Long, 2017), as is the case with these data. For these reasons I used MLMV (or FIML) rather than listwise deletion of data or multiple imputation.

As noted above, family income was assessed as log average income for 1998–2001. Material hardship and other parent variables were measured in 2002. Adolescent behavior problems and cognitive achievement were scored in 2007.

Calculation of sample mean scores on income and parent variables from 2002 to 2007 showed very little change and there was high auto-correlation between variables from the two time periods. Analyses of other data sets have also shown little change in socioeconomic status and parent variables (Votruba-Drzal, 2003, 2006), so this is not surprising. I did not control for the 2007 measures of family income and parent behaviors because their high correlation with measures from 2002 would have led to collinearity.

3. Results

Descriptive information is provided in Table 1. Results from the SEM measurement model are presented in Table 2 and goodness of fit tests in Table 3. Results from the structural models permitted specification of paths for income and the mediating variables which were used to test the hypotheses (Tables 4 and 5) and calculate indirect and total effects of income and mediating variables (Table 6). I subjected results of the structural models to examination of goodness of fit tests (Table 7) to assess the strength of the mediation model.

Table 1.

Descriptive Information for Variables (N = 953)

	Mean or %	Standard Deviation
Control Variables
Single biological parent in home 2001	42%
Number of siblings 2001	1.28	1.04
Highest education level of either parent 2001
Less than high school	18%
High school (reference)	53%
College degree	23%
Advanced degree	6%
Child’s ethnicity
White (reference)	48%
Black	43%
Hispanic	8%
Child’s gender: female	49%	0.50
Child’s age 2002	9.49	2.19
Parent’s cognitive achievement 1997	30.59	5.46
Independent Variables
Log average income 1998–2001a	10.48	1.04
Parent Behavior Variables (2002)
Family material hardship	1.67	1.98
Parent support c	0.00	0.88
Parent’s cognitive stimulation of child	6.36	1.70
Parent’s warmth toward child	4.04	0.60
Parent attendance at school events	9.73	18.21
Parent’s expected educational attainment for childb	3.15	2.39
Parent distress c	0.00	0.43
Aggravation in parenting	1.29	0.81
Parent psychological distress	4.13	3.64
Adolescent Behavior Variables (2007)
Cognitive achievement c	0.00	12.38
Letter-word score 2007	100.92	16.24
Passage comprehension score 2007	97.64	14.46
Applied problems score 2007	102.39	15.89
Behavior problems
Externalizing	5.26	4.12
Internalizing	2.94	3.21

^aAdjusted for inflation to 2002 dollars. Mean family income = $54,900 (SD = $50,420).

^bCoded as years of education beyond high school (12 years), bottom-coded at 0 years (high school equivalent).

^cLatent variable estimated from variables below in measurement model (see Table 2).

Table 2.

Loadings for Two-Factor Model of Parent Variables and One-Factor Model of Child Cognitive Achievement

	b	β	SE
Two-Factor Model of Parent Variables
Parent support 2002
Cognitive stimulation of child	1.00	.65
Warmth toward child	0.18***	.33	.02
Attendance at school events	5.59***	.32	.74
Years of expected educational attainment beyond high school	1.13***	.52	.13
Parent distress 2002
Aggravation in parenting scale	1.00	.66
Psychological distress scale	4.55***	.68	.78
One-Factor Model of Child Cognitive Achievement
Child cognitive achievement 2007
Letter-word score 2007	1.00	.82
Passage comprehension score	0.91***	.87	.03
Applied problems score	0.87***	.76	.04

Note. chi-squared (24df) = 50.05; RMSEA = 0.03; AIC = 47368.59; BIC = 47514.38; CFI = 0.99; TLI = 0.98; CD = 0.97.

^***p < .001

Table 3.

Comparison of Alternate Measurement Models of Parent Support and Parent Distress Latent Variables, Listing Fit Statistics when the Measure is Included or Excluded and Difference from Full Model

		Fit Statistics
Model		RMSEA	CFI	AIC	BIC
1	Two-factor model of parent distress and parent support separated (full model)	0.03	0.99	47372.81	47518.60
2	Parent support factor excluding cognitive stimulation scale	0.11	0.87	47616.90	47757.83
3	Parent support factor excluding warmth toward child scale	0.06	0.96	47435.57	47576.50
4	Parent support factor excluding school attendance item	0.06	0.96	47433.53	47574.46
5	Parent support factor excluding expected educational attainment item	0.09	0.91	47521.53	47662.46
6	One-factor model of parent distress and parent support combined	0.08	0.91	47520.37	47656.44
		Difference from Full Model (Model 1)
Model		χ2	df	AIC	BIC
2	Parent support factor excluding cognitive stimulation scale	64.76***	1	62.76	57.90
3	Parent support factor excluding warmth toward child scale	62.72***	1	60.72	55.86
4	Parent support factor excluding school attendance item	150.71***	1	148.72	143.86
5	Parent support factor excluding expected educational attainment item	151.55***	2	147.56	137.83
6	One-factor model of parent distress and parent support combined	246.09***	1	244.09	239.23

^***p < .001

Table 4.

SEM Structural Model of Relations of Parent Variables to Income, Parent Variables, and Control Variables

	Material Hardship 2002			Parent Distress 2002			Parent Support 2002
	b	β	SE	b	β	SE	b	β	SE
Average income 1998 – 2001	−0.22*	−0.11	0.10	0.00	−0.01	0.02	0 11***	0.13	0.03
Material hardship 2002				0.06***	0.28	0.01	0.03**	0.07	0.01
Parent distress 2002							−0.77***	−0.38	0.05
Control Variables
Single biological parent in home	0.39*	0.10	0.16	0.07*	0.08	0.03	−0.04	−0.02	0.05
Number of siblings in home	0.10	0.06	0.06	0.00	0.01	0.01	0.02	0.02	0.02
Parent education: < HS	0.52*	0.10	0.23	0.13**	0.11	0.05	−0.05	−0.02	0.07
Parent education: college	−0.63***	−0.13	0.16	−0.04	−0.04	0.03	0.39***	0.19	0.05
Parent education: advanced	−0.96***	−0.11	0.21	0.03	0.01	0.05	0.75***	0.20	0.08
Child’s ethnicity: Black	0.31	0.08	0.17	0.02	0.02	0.03	−0.19***	−0.11	0.05
Child’s ethnicity: Hispanic	−0.62*	−0.09	0.28	0.05	0.03	0.06	−0.17*	−0.05	0.09
Child’s gender: female	−0.18	−0.04	0.12	−0.02	−0.03	0.02	0.18***	0.10	0.04
Child’s age	−0.10***	−0.11	0.03	0.01	0.05	0.01	−0.04***	−0.10	0.01
Parent cognitive achievement	0.06***	0.18	0.02	−0.01***	−0.16	0.00	0.02***	0.15	0.01
Constant	2.83**		1.11	0.18		0.23	−1.64***		0.33

^***p < .001

^**p < .01

^*p < .05

Table 5.

Relations of Adolescent Behaviors to Independent and Control Variables

	Cognitive Achievement 2007			Externalizing Behavior 2007			Internalizing Behavior 2007
	b	β	SE	b	β	SE	b	β	SE
Average income 1998 – 2001	0.19	0.02	0.37	−0.22	−0.06	0.18	−0.15	−0.05	0.12
Material hardship 2002	0.08	0.01	0.17	0.19**	0.09	0.07	0.14*	0.09	0.06
Parent distress 2002	−1.28	−0.04	0.93	3.09***	0.33	0.37	2.65***	0.36	0.29
Parent support 2002	8.50***	0.59	0.55	−0.76***	−0.16	0.20	−0.32*	−0.09	0.16
Control Variables
Single parent in home	0.42	0.02	0.71	0.59*	0.07	0.30	−0.08	−0.01	0.23
Number of siblings in home	−0.07	−0.01	0.25	−0.11	−0.03	0.11	−0.12	−0.05	0.08
Parent education: <HS	−0.68	−0.02	0.85	−0.38	−0.04	0.39	−0.29	−0.03	0.30
Parent education: college	−0.24	−0.01	0.76	0.31	0.03	0.33	−0.10	−0.01	0.26
Parent education: advanced	1.38	0.03	1.22	1.36*	0.08	0.62	0.58	0.04	0.48
Child’s ethnicity: Black	−3.96***	−0.15	0.74	−1.11***	−0.13	0.30	−1.34***	−0.21	0.25
Child’s ethnicity: Hispanic	−0.93	−0.02	1.26	−1.47***	−0.10	0.51	−0.58	−0.05	0.39
Child’s gender: female	0.09	0.00	0.54	0.07	0.01	0.24	0.11	0.02	0.19
Child’s agea				−0.05	−0.03	0.06	0.01	0.01	0.05
Parent cognitive achievement	0.23***	0.10	0.07	0.02	0.03	0.03	0.00	0.00	0.02
Constant	−7.45		4.69	7.50***		2.07	5.03		1.47

^aCognitive achievement scores are age-adjusted so age is not included in the analysis.

^***p < .001

^**p < .01

^*p < .05.

Table 6.

Decomposition of Adolescent Behavior Total Associations into Direct and Indirect Associations

	Direct Association			Indirect Association			Total Association
	b	β	SE	b	β	SE	b	β	SE
Cognitive Achievement
Average income 1998 – 2001	0.19	0.02	0.37	0.98**	0.08	0.32	1 17**	0.10	0.45
Material hardship 2002	0.08	0.01	0.17	−0.21	−0.03	0.12	−0.14	−0.02	0.20
Parent distress 2002	−1.28	−0.04	0.93	−6.53***	−0.22	0.63	−7.81***	−0.27	0.98
Parent support 2002	8.50***	0.59	0.55				8.50***	0.59	0.55
Externalizing
Average income 1998 – 2001	−0.22	−0.06	0.18	−0.18*	−0.05	0.09	−0.41*	−0.10	0.20
Material hardship 2002	0.19**	0.09	0.07	0.20***	0.10	0.03	0.38***	0.19	0.07
Parent distress 2002	3 09***	0.33	0.37	0.59***	0.06	0.16	3.68***	0.39	0.34
Parent support 2002	−0.76***	−0.16	0.20				−0.76***	−0.16	0.20
Internalizing
Average income 1998 – 2001	−0.15	−0.05	0.12	−0.11	−0.04	0.07	−0.27	−0.09	0.14
Material hardship 2002	0.14*	0.09	0.06	0.16***	0.10	0.03	0.30***	0.19	0.06
Parent distress 2002	2.65***	0.36	0.29	0.24*	0.03	0.13	2 89***	0.39	0.26
Parent support 2002	−0.32*	−0.09	0.16				−0.32*	−0.09	0.16

^***p < .001

^**p < .01

^*p < .05

Table 7.

Comparison of Alternate Structural Models to the Full Model, Listing Fit Statistics when the Variable is Included or Excluded and Difference from Full Model

		Fit Statistics
Model		RMSEA	CFI	AIC	BIC
1	Base model with income predicting child outcomes	0.19	0.41	43810.53	44403.40
2	Income and parent distress predicting child outcomes	0.19	0.56	43427.30	44088.21
3	Income and parent support predicting child outcomes	0.14	0.77	42836.96	43497.87
4	Income, parent distress, and parent support predicting child outcomes	0.11	0.93	42431.09	43164.89
5	Full model with all variables (income, parent distress, parent support and material hardship) predicting child outcomes	0.08	0.99	42257.20	43068.75
		Difference from Full Model (Model 5)
Model		χ2	df	AIC	BIC
1	Base model with income predicting child outcomes	1643.33***	45	1553.33	1334.65
2	Income and parent distress predicting child outcomes	1232.10***	31	1170.10	1019.46
3	Income and parent support predicting child outcomes	641.76***	31	579.76	429.12
4	Income, parent distress, and parent support predicting child outcomes	205.89***	16	173.89	96.14

^***p < .001

3.1. Results of Measurement Model

I used SEM to first calculate a measurement model, which produced three latent variables (Table 2); I then tested the measurement model for goodness of fit (Table 3). Scales for these measures were used instead of the items of which they are composed because the scales had been validated in previous research and separating them into individual items components would have prevented direct comparison of results obtained here to earlier published findings. The two parent distress scales were based on Likert scaling, and rated by the parents; the other scales represented either test scores (i.e., the three child cognitive achievement measures) or count variables of numbers of behaviors exhibited (i.e., child behavior problems, parent cognitive stimulation score, parent warmth and support score, parent involvement score, and parent expected educational attainment score). The first indicator for each latent variable was set to 1.0 to set the metric for that factor.

The four parent behaviors showed significant loadings (p < .001) on a single latent variable which I termed “parent support”: cognitive stimulation (β = .65), parent school involvement (β = .32), warmth and support (β = .33), and expected educational attainment (β = .52). These relatively strong associations with the latent variable suggested an overall construct of parent support that combined components of parent investment theory and cultural theory, consistent with Lareau’s (2011) theory of concerted cultivation and with findings from Cheadle and Amato (2011) that parent investment and cognitive stimulation loaded together in an SEM analysis of younger children. This finding differs from that of Gershoff et al. (2007), who identified two different factors for these variables. However, they used a different program for confirmatory factor analysis (AMOS), different measures, younger children, and entered a few variables at a time due to small sample size and restricted power. The relatively larger sample size in this study permitted analysis of all variables at once with the SEM program from Stata, which produced a single latent variable that was robust, with strong model fit.

The second latent variable was composed of scales measuring parents’ psychological distress and aggravation in parenting. The loadings for the two scales were roughly equal (β = .68 and .66, respectively; p < .001). I named the latent variable “parent distress.”

Finally, the three Woodcock-Johnson Revised Test of Achievement scores loaded (β = .76 to .87, p < .001) on a single latent variable termed “cognitive achievement.” In contrast, I kept the scales for the other dependent variable, behavior problems, separate because a model that loaded them both on one latent variable did not converge, suggesting that they differ empirically as well as conceptually. The scales, externalizing and internalizing, were therefore treated as separate observed variables. Previous research (Peterson & Zill, 1986) empirically supports the idea that the two subscales are distinct, and recent studies (e.g., Gershoff et al., 2007) have separated them, so the model tested here permits direct comparison of results to those obtained in earlier studies.

Results from Table 3 describe fit statistics for the measurement model in comparison to alternative models of parent support and parent distress. Model 1 represents the two-factor measurement model used in the later analyses, with two latent variables, parent support and parent distress. Models 2–5 show the models for parent support when each one of the four component indicators (cognitive stimulation, warmth, parent school event attendance, and educational expectation, respectively) is excluded. Model 6 merges the two latent parent variables into one to test for their independence. Model 1—the two-factor model—shows stronger fit statistics than all of the other models (RMSEA = .03; CFI = .99). Further, comparison of chi square, AIC, and BIC results indicate that all of the other models differ significantly from Model 1. These findings suggest that model fit is best with two latent variables, referring to parent behaviors (i.e., parent support) and parent psychological reactions (i.e., parent distress).

3.2. Results of Structural Models

The relations between average family income at child ages 6–9 years and parent behaviors at child age 10 years are shown in Table 4. The associations of average family income at child ages 6–9 to adolescent behaviors at age 15 are listed in Table 5. Figure 2 provides a graphic representation of the linkages among variables.

Figure 2.

Structural Model

The first set of hypotheses about associations with average family income at ages 6–9 was mostly supported (Table 4). Family income was inversely associated with material hardship (β = −.11, p < .01) as predicted by H1a, but not significantly associated with parent distress as predicted by H1b, net of material hardship. Hypotheses 1c and 1d referred to two parent variables also measured when the mean child age was 10 years (investment and parent support) that were merged into a single variable of parent support based on results of the measurement model described above. As expected, average family income at ages 6–9 was positively related to parent support at age 10 (β = .13, p < .001). The relation of family income to each parent variable was assessed while controlling the effects of demographic variables.

The mediating variables were also associated with several control variables, net of income. Relative to parents with a high school education, parents with less education had significantly greater material hardship and distress, and parents with degrees had significantly lower material hardship and higher parent support. Relative to White children, Hispanic children experienced less material hardship and parent support, and Black children experienced less parent support, net of income and other controls. Child age was inversely associated with material hardship and parent support. Having only one biological parent in the house was associated with material hardship and parent distress. Parent cognitive achievement was directly associated with material hardship and parent support, and inversely associated with parent distress.

As shown in Table 5, net of controls and mediating variables, average family income at child ages 6–9 was not significantly related to adolescent cognitive achievement and behavior problems measured at age 15. Thus, the direct path from family income to adolescent behaviors, depicted in Figure 1, was not significant for cognitive achievement, externalizing behaviors, or internalizing behaviors.

As predicted (H2), material hardship was associated with externalizing (β = .19, p < .01) and internalizing (β = .09, p < .05); it was not associated with cognitive achievement, contrary to prediction. Parent distress (H3) was also associated with externalizing (β = .33, p < .001) and internalizing (β = .36, p < .001), as expected; again, contrary to prediction, it was not associated with cognitive achievement. Hypotheses 4 and 5 were tested simultaneously using the merged variable of parent support, composed of investment and other supportive behaviors, based on results from the measurement model that were explained above. Parent support was positively associated with cognitive achievement (β = .59, p < .001) and negatively associated with externalizing (β = −.16, p < .001) and internalizing (β = −.11, p < .09), supporting Hypotheses 4 and 5.

Adolescent externalizing was positively related to having a parent with an advanced degree and inversely related to child’s ethnicity being Black or Hispanic rather than White, and internalizing was negatively related to child’s ethnicity being Black relative to White. Adolescent cognitive achievement was positively related to parent’s cognitive achievement and negatively related to child’s ethnicity being Black. These associations with demographic variables were not predicted and cannot be readily interpreted.

The results in Tables 4 and 5 show the strength and significance of the individual path associations, but they do not provide a full test of the different mechanisms predicted by the three theoretical explanations. I used the results to decompose the total associations of income and mediating variables with later variables into direct effects and total indirect effects using the Sobel test for significance with bootstrapping (Preacher & Hayes, 2004, 2008). Family income at child ages 6–9 did have significant total associations with cognitive achievement (β = .10, p < .001) and externalizing (β = −.10, p < .05), and significant indirect associations with cognitive achievement (β = .08, p < .01) and externalizing (β = −.05, p < .05) through the mediating variables. Thus, it appeared that average family income of young children was associated with adolescent outcomes, but only indirectly via material hardship, parent distress, and parent support. Additionally, material hardship had significant indirect associations (through parent distress and parent support) and total associations with externalizing and internalizing, and parent distress had significant indirect associations (through parent support) and total associations with all three child behaviors. Parent support had the strongest association with cognitive achievement (β =.59, p < .001), whereas parent distress had the strongest total associations with externalizing and internalizing (β = .39 on each, p < .001)—both much stronger than income’s total associations with each child behavior.

Table 7 compares fit statistics and results from likelihood-ratio chi-squared tests for the structural model and several simpler alternative models that effectively exclude important mediating variables by constraining their effects to zero. In this series of nested models, Model 1 includes only paths from family income and the control variables to adolescent behaviors, Model 2 adds paths from parent distress to adolescent behaviors to Model 1, Model 3 adds paths from parent support to adolescent behaviors to Model 1, Model 4 includes both parent distress and parent support, and Model 5 (the full model) includes the material hardship variable as well. Fit statistics suggest that the full model (Model 5) has a better fit (RMSEA = .08, CFI = .99) than the simpler models. For example, Model 1, showing the effects of average family income alone, is much weaker than Model 5, which includes all variables. Further, likelihood-ratio chi-squared tests of the fit of the alternative models relative to the full model are all significant, suggesting that the alternative models lose significant amounts of information by excluding paths from any of the mediating variables to adolescent behaviors. Thus, all three mediating variables (parent distress, parent support, and material hardship) play important roles in predicting adolescent behaviors.

4. Discussion

As noted above, the overall goal of this study was to use data collected in multiple waves to test an integrated model of the association among average family income of children aged 6–9, parent behaviors toward children (age 10), and adolescent behavior problems and cognitive achievement (age 15). My hypotheses are based heavily on three independent but not mutually exclusive theories about possible mediators of the relations of income to adolescent behaviors. This study confirms previous studies’ findings but expands on them in two important ways.

First, previous studies included fewer variables so that comparison among competing theories was not feasible. Modeling of a measurement model provided support for an integrated model of child development that draws on material hardship theory, parent distress theory, parent investment theory, and cultural theories of parent concerted cultivation of children. Second, data used in earlier studies were usually cross-sectional, not longitudinal, and so developmental associations could not be assessed. In this study it was possible to test a structural model of the association of average family income when children were young (6–9 years old) to parent behaviors when they were 10 years old, and to cognitive achievement and behavior problems at age 15. Both models are discussed below.

4.1. Parent Behaviors, Income and Adolescent Behaviors: Measurement Models

As discussed at the outset, there are three primary theoretical frameworks that describe mechanisms by which family income may be associated with parent behaviors, which in turn are associated with child behaviors. Parent distress theorists argue that low income is related to material hardship and parent distress, which then reduces parent support, which in turn is associated with child behavior problems; parents’ psychological distress is the mechanism that links income to child behaviors. Parent investment theorists contend that low income reduces parent investment in educational and other child supports, which is related to child behavior problems and low child cognitive achievement; parents’ investment in child supports is the mechanism that links income to child behaviors. Cultural theorists expand on investment theory by suggesting that investment and other supportive parent behaviors are associated with child cognitive achievement and low behavior problems; parents’ supportive behaviors, including investment are related to their cultural milieu, which is associated with socioeconomic status, but not driven by income. Theorists in each tradition thus posit one mechanism that explains why income influences a particular parent behavior that then is related to child behavior. SEM analysis was used to examine measurement models of the parent behaviors to assess whether the three predicted mechanisms were independent or interrelated. Results produced two latent variables, parent distress and parent support; comparison to alternate models showed that the two-factor solution was strongest. There are three important implications for theory and policy.

First, these findings suggest that parent distress is a separate construct from parent support; they may be correlated for some parents but they are not inherently associated with one another for all parents. This supports existing theories of parent distress and concerted cultivation, in that parent distress does not directly determine the warmth and support they show their children, but is associated. It is theoretically plausible that distressed parents would never have the psychological capacity to engage in concerted cultivation of children, in which case the two constructs might have loaded together on the same construct—but results did not support this argument. Rather, parent distress was associated with less support for some, but not all, parents. Lareau’s (2011) qualitative research identified parents who engaged in concerted cultivation of their children (i.e., investment, cognitive stimulation efforts, warmth and high expectations) but also parents who allowed their children to roam more freely and find cognitive stimulation outside the family’s orbit, more recently termed “free-range” parents (Calarco, 2018). Lareau’s work showed that parents in the two groups were distinct due to cultural or ideological differences in how they viewed norms for child-rearing, not in psychological distress levels; the reason parents engaged in concerted cultivation was their cultural beliefs, not their psychological states. The policy implication is that actions to reduce parent distress may change parent behaviors in some ways (e.g., less corporal punishment) but not change behaviors that reflect parents’ cultural views about child-rearing. The results also show that parent distress theory and cultural theory each advance our understanding of parent actions toward their children, in different ways. Specifically, income is inversely correlated with parent distress and positively correlated with parent support; it is therefore reasonable to expect that increased income would reduce parent distress and increase parent support, although the analyses presented here represent only a first step toward testing that expectation. Future research should take the second step of assessing whether changes in income are associated with reduced parent distress and increased parent support.

Second, the results suggest that parent investment and parent support are not independent constructs. That is, parents who sought enrichment activities for their children and were heavily involved in their schooling also tended to be emotionally supportive, consistent with Laureau’s primary research on concerted cultivation. Researchers in the parent investment literature (Dumais et al., 2012; Lee & Bowen, 2006; Sui-Chu & Willms, 1996), however, did not predict that investment and support were interrelated. It is intuitively reasonable that the two constructs would be related but the empirical support provides greater confidence for this idea.

Thus, findings from the measurement model provide some clarity as to why increases in parent income may not always reduce parent distress or increase their efforts to provide cognitive stimulation to their children. Results from structural models analyses provide further details about why this might be the case. We examine them next.

4.2. Parent Behaviors, Income and Adolescent Behaviors: Structural Models

Prior research (e.g., Mayer, 1997) suggested that family income in early childhood was not strongly related to behaviors for young adults, which may mean that the effect of income on adolescent behaviors is mediated by parent behaviors. Results here support the idea that income’s association with adolescent behaviors is largely mediated, but still significant. I used three methodological tools not used in previous work that strengthen confidence in these results. First, I used bootstrapped standard errors for the mediation model, which produces more valid and reliable results (Preacher & Hayes, 2004). Second, data were longitudinal, with income measured prior to parent behaviors and both measured prior to adolescent behaviors; this reduced bias that could be caused by measurement in the same time period. Third, structural equation modeling (SEM) allowed separation of the “total effects” of income with adolescent behaviors (i.e., its association without taking into account the mediating variables) into a direct association of income net of the parent behaviors and an indirect association through the mediating parent behaviors.

Three parent behaviors measured when the child was 10 years of age were significantly related to average family income during the time period when the child was 6–9 years of age. Income was inversely related to material hardship and parent distress and positively related to parent support, consistent with earlier research (e.g., Cheadle & Amato, 2011; Gershoff et al., 2007; and several studies from Conger and colleagues). These findings support the first link in the chain of mediated relations predicted among variables.

The relations of each individual parent behavior to adolescent behaviors were assessed while controlling effects of the other parent behaviors. Adolescent externalizing and internalizing behaviors (age 15) showed significant relations to all parent behaviors: they were most strongly related to parent distress, followed by parent support (inverse) and material hardship. Adolescent cognitive achievement (age 15) was significantly related to parent support but not to parent distress or material hardship. These findings support the second expected link in the chain of mediated relations among variables. The mediators are thus related to both family income and adolescent cognitive achievement and behavior problems. Examining total and indirect associations of income with adolescent behaviors confirms these linkages: though income was not significantly associated with child behaviors net of parent behaviors, it did have significant total and indirect associations with cognitive achievement and externalizing behavior. This suggests that most of income’s association with child behaviors was through its association with material hardship and parent behaviors. However, even the direct associations of the parent behaviors with child behaviors were much stronger than the total associations of income with child behaviors.

The final question concerning the proposed mediation is whether all three parent behaviors contribute to the prediction of adolescent behaviors. Tests for goodness of fit showed that the strongest structural model included all three mediators; exclusion of paths to and from any one of these constructs significantly decreased model fit, suggesting that the alternative models lose significant amounts of information by excluding paths from any of the mediating variables to adolescent behaviors. In particular, the relations of income alone to the dependent variables are much weaker than those seen when the parent behaviors are also included. Thus, all three mediating variables (parent distress, parent support, and material hardship) play important roles in predicting adolescent behaviors.

These results have implications for both theory and policy. First, findings for material hardship support the earlier research of Gershoff, Raver and colleagues, cited earlier. Material hardship accounts for additional variance in adolescent behaviors, beyond that accounted for by family income alone. For example, even families with moderate income might suffer eviction if they are confronted with unexpected medical bills that deplete their financial reserves. Thus, material hardship can reflect more than average income, and may have a more substantial impact on some child behaviors (i.e., externalizing and internalizing).

Second, the structural model produced here shows that parent distress and parent support (already shown by the measurement model to be independent from one another) each account for variance in adolescent cognitive achievement and behavior problems. These findings suggest that parent distress (e.g., Conger and others) and parent support (e.g., Cheadle & Amato, 2011; Lareau, 2011) are complementary paths by which parents’ psychological and behavior responses to their children are related to the children’s development. It should be noted that parent distress is more strongly related to adolescent behavior problems but parent support is more strongly related to adolescent cognitive achievement. This finding is wholly consistent with predictions of the two theories but provides important confirmation that the two mediator variables have somewhat different relations to adolescent behaviors. It should be cautioned that these results do not test whether increased family income for young children will be associated with reduced behavior problems or increased cognitive achievement. Future research should investigate whether changes in family income are associated with changes in parent distress and support and change in adolescent behavior problems and cognitive achievement. The current study extends existing theory and suggests that increased income could have these beneficial effects but empirical assessment is needed to test this idea. Previous research (Votruba-Drzal, 2003, 2006) has used fixed effects analysis to assess this for children in younger age groups but it has not been assessed for adolescents. Further, the associations between the parents’ responses when the children were 10 and the children’s responses at age 15 were quite strong—despite the fact that five years had elapsed. From a policy perspective, it suggests that adolescent behaviors could be improved by increases in family income, but even greater improvements could be achieved if parents received direct support that would reduce the material hardship events that they experience (e.g., housing or food support), if their psychological distress could be reduced directly (e.g., through counseling), or if their efforts to support their children could be enhanced (e.g., through “parenting classes” that teach skills for providing more cognitive stimulation and using less harsh discipline).

4.3. Limitations

The sample used here was relatively small, but generally representative of the population cohort of American children aged 15 years old in 2007. Attrition between waves may have reduced generalizability but power was sufficient for the analyses. Second, it was not possible to check temporal priority, and therefore directionality and causality, of the relations among the parent behaviors. It was, however, possible to assess sequence of these variables in relation to family income (average of data from 1998–2001) and adolescent behaviors (measured in 2007). Additional research is needed to assess whether these findings would be replicated with more recent samples of children, but one benefit of the current sample is that the economy was relatively stable during this time period and volatility would not have been a source of bias in results, as it might be with data collected in subsequent periods (e.g., 2008–2012 or 2020–2021).

4.4. Conclusions

Results suggest that family income of young children is related to adolescent cognitive achievement and behavior problems—but that this relation is mediated by several parent behaviors, including material hardship, parent distress, and parent support in the pre-adolescent years, at about age 10. These results support an integrated model of the multiple developmental mechanisms by which family income becomes associated with adolescent cognitive achievement and behavior problems over time. Future research should examine the question of whether changes in family income are related to changes in adolescent cognitive achievement and behavior problems in order to provide a more comprehensive picture of the developmental process.

Data Availability Statement:

This study used data from the PSID and CDS, which are produced and distributed by the Survey Research Center, Institute for Social Research, University of Michigan, Ann Arbor, MI. The collection of these data was partly supported by the National Institutes of Health under grant number R01 HD069609 and R01 AG040213, and the National Science Foundation under award numbers SES 1157698 and 1623684.