Economic life for most American households is quite dynamic. Such income instability is an understudied aspect of households’ economic contexts that may have distinct consequences for children. We examine the empirical relationship between household income instability, as measured by intrayear income change, and adolescent school behavior outcomes using a nationally representative sample of households with adolescents from the Survey of Income and Program Participation 2004 panel. We find an unfavorable relationship between income instability and adolescent school behaviors after controlling for income level and a large set of child and family characteristics. Income instability is associated with a lower likelihood of adolescents being highly engaged in school across the income spectrum and predicts adolescent expulsions and suspensions, particularly among low-income, older, and racial minority adolescents.
Economic life for most Americans is quite dynamic: household earnings and income as well as eligibility for social assistance programs have been shown to fluctuate across and within years (Bane and Ellwood 1996; Dahl et al. 2011; Duncan 1988; Gottschalk and Moffitt 2009; Hacker 2008). Such instability may have distinct consequences for families and children that have not been fully incorporated into existing theories or empirical investigations (Hill et al. 2012). What is understood about the effects of poverty, and of the household economic context more broadly, is largely derived from analyses that treat income as a static condition (averaged annually; Cancian and Danziger 2009; Duncan et al. 1998; Gennetian et al. 2010) or as a snapshot of dramatic income change (e.g., resulting from an economic crisis or job loss; Conger and Conger 2002; Elder 1974). To begin to fill this gap, we use a nationally representative data set to examine the descriptive empirical relationship between income instability, as measured through intrayear income dynamics, and adolescent school behavior outcomes across heterogeneous income groups.
Adolescents may be affected by unstable family income irrespective of income level. Adolescence is a period of rapid development that includes dramatic physical and physiological changes, cognitive advances, and formation of identities separate from peers and family (Allen et al. 1990; Steinberg and Morris 2001). Pernicious disruptions to the home environment may be especially consequential by altering adolescents’ developmental trajectories. An 11- to 15-year-old American child, on average, will spend 11 % of his or her life in poverty (Wagmiller and Adelman 2009). Such experiences in poverty contribute to lower academic achievement, poorer social-emotional and physical health, and increased risk of engaging in risky behaviors (Holzer et al. 2007). We hope to extend the poverty research knowledge base by descriptively examining whether income instability has distinct consequences for adolescent development.
We focus on school behavior, partly because of the availability of national representative data on adolescents with frequent measurement of household income but also, and importantly, because both positive and negative school behaviors are contributors to school achievement and educational completion. School behaviors are also sensitive to the contemporaneous family processes that may occur as a result of economic instability (Fredricks et al. 2004). By requiring parents to continuously manage changing household finances, unstable income may create parental stress, interrupt household routines (e.g., after-school activities), and result in less-effective parenting practices (e.g., less monitoring, greater harshness; Hill et al. 2012). Such changes in parental behaviors may reduce children’s ability to focus on the learning tasks of schooling (Steinberg et al. 1992), and through this, may result in lowered academic achievement. Indeed, it is our expectation that children’s behavior may be the most proximal targets of family stress and parenting behavior and, in turn, affect children’s academic performance (rather than through changes in cognitive ability, for example).
This study makes several important contributions to the study of income and child development. First, we examine the relationship between intrayear household income dynamics and adolescent schooling outcomes. Because so little research has been conducted conceptually or empirically on this topic in the context of adolescent development, our work is descriptive. It is a foundational step that extends the existing literature on the effects of poverty and family economic conditions on adolescent development (Ruhm 2008) as well as studies that have considered single negative income shocks (Ananat et al. 2011; Conger et al. 1994; Elder et al. 1985; Kalil and Ziol-Guest 2005, 2008; McLoyd et al. 1994; Randolph et al. 2004; Weitzman 1985). Second, we consider a broad range of measures that capture various features of income instability that are hypothesized to have meaningful and distinct influences on adolescent development, including a measure of total income instability most commonly used by economists who track trends over time, as well as measures that capture frequency, magnitude, and overall direction of income change. Third, we use a nationally representative data set of households with adolescents, allowing us to examine these questions in a sample that spans the range of income levels (and financial assets) represented in the United States. We are also able to examine empirical associations between income instability and school behavior by a variety of adolescent characteristics (gender, age, and race).
Our analysis sample is drawn from the 2004 panel of the Survey of Income and Program Participation (SIPP), spanning a relatively stable economic period from February 2004 to September 2006 to minimize confounds of broader economic instability triggered by the recent Great Recession. We find that income instability is nearly double among households in the lowest compared with the highest income quintile over this period. Controlling for a range of demographic characteristics and income level, we find that income instability is adversely associated with adolescents’ school behavior (both engagement and suspensions/expulsions). However, these relations do not hold across all subgroups: greater income instability is associated with lower levels of engagement across the income spectrum, but with a greater likelihood of expulsions and suspensions only for those adolescents at the lowest levels of household income. Our findings suggest that income instability may indeed have distinct negative associations with certain aspects of adolescents’ school behavior beyond income level.
Background and Conceptual Motivation
According to some estimates, annual income volatility has increased by nearly one-third over the past 50 years (Bania and Leete 2007; Dynan 2010; Dynan et al. 2007; Gottschalk and Moffitt 2009; Keys 2008; Newman 2008). Income instability, net of low income level, has also been shown to predict food insecurity, an indicator of material well-being (Bania and Leete 2007; Mills and Amick 2010; Newman 2008; Ribar et al. 2008). Empirical examinations of the influence of income instability on children’s well-being are relatively limited. One exception is a study by Yeung et al. (2002) using the PSID-CDS, in which income instability was measured as the proportion of years in which a family experienced at least a 30 % decrease in income level from the prior year. Income instability (and not income level) was positively associated with maternal depressive affect, a key maternal predictor of outcomes for children; income instability was negatively associated with children’s cognition and behavior.
Although theoretical work directly addressing income instability is limited, well-developed theories on the effects of income on children’s development provide a starting point and emphasize two theoretical frameworks: the family investment model and the family stress model. The family investment model emphasizes the role of parental time and monetary investments in children’s human capital development (Becker and Thomes 1986) and the harmful consequences of material deprivation on children’s development (Brooks-Gunn and Duncan 1997; Gershoff et al. 2007). A lack of material investment by low-income parents has been shown to contribute to lower school performance and diminished cognitive development in children (Gershoff et al. 2007; Mistry et al. 2008; Yeung et al. 2002).
The family stress model highlights the ways in which economic strain can increase parent stress, affect family interactions, and impede parenting quality (McLoyd 1990; McLoyd et al. 1994). Greater household economic hardship is associated with increased marital conflict, conflicts with parents and children over money, and increased emotional and behavior problems in adolescents (Conger et al. 1994, 2000). Much of the formative evidence supporting the importance of the family stress pathway comes from research examining the impact on children’s well-being of large, single, negative income shocks, such as job loss (Ananat et al. 2011, 2013; Conger et al. 1994; Elder et al. 1985; Kalil and Ziol-Guest 2005, 2008; McLoyd et al. 1994; Randolph et al. 2004; Weitzman 1985).
These two theories provide a foundation for understanding the effects of income instability (for example, by pointing to the importance of erratic investments or heightened stress) but do not fully consider ways in which frequent income instability, per se, may have distinct influences from income level and/or from one-time changes in income. Theories from developmental psychology and sociology suggest household chaos and family instability as two facets of environmental turbulence that could affect child development (Ackerman et al. 1999; Wachs and Evans 2010). Family instability more generally is associated with higher stress and less regular sleep for both parents and children, in addition to lower school engagement and higher social-emotional problems for young adults (e.g., Adam and Chase-Lansdale 2002; Brown and Low 2008; Marcynysyn et al. 2008). Chaotic home environments have overstimulating physical (e.g., noise and overcrowding) and psychosocial (e.g., divorce and parental job loss) characteristics. Chaotic environments can disrupt children’s development by reducing the quantity and quality of interactions parents have with their children and the regularity of household routines (Corpaci and Wachs 2002; Wachs and Evans 2010).
Although adolescence is a stage marked by increasing independence from parents, responsive parenting is essential to monitor and support adolescents’ academic persistence and achievement (Brown et al. 1993; Crouter et al. 1990). By draining parents’ time and attention (Gennetian and Shafir 2015; for an application to the more general context of poverty, see Shah et al. 2012), income instability may interfere with parents’ available psychological resources to consistently practice responsive parenting. Parents who are coping with a financial crisis at hand may unintentionally neglect opportunities to enforce school participation and homework or to engage with children on educational tasks. Parents with depleted energy may give in to decisions that might increase adolescents’ engagement in risky behaviors (e.g., giving into an adolescent’s demands to stay out late). Repeated challenges to family finances may have additional effects on school-related investments, especially those that require regularity of payments (such as extracurricular activities). Finally, family stress may spill over to an adolescent’s own stress about family economic circumstances that could distract attention away from school.
Dimensions of Income Dynamics
Different dimensions of income instability—including the magnitude, frequency, and direction of income change—could translate into qualitatively different experiences for families. The magnitude of income fluctuations determines the amount of relative adjustment required. Patchwork strategies to compensate for income loss may be more feasible and less noticeable with small changes, whereas a relatively large change could require substantial adjustment (e.g., a residential move to a new community). Also, a higher frequency of income changes may be at least as detrimental as a single change because of its potential contribution to enduring chaos and stress in the home environment (see Hill et al. 2012). Finally, interpreting how magnitude and frequency of change relate to well-being will also likely depend on the direction of income change. A sizable positive income shock (e.g., a $1,000 cash windfall from the Earned Income Tax Credit) may or may not have the similar impact as a sizable negative income shock (e.g., a $1,000 decrease in earnings due to reduced work hours), although both would be similarly considered in measures of magnitude (a $1,000 change) and frequency (a single shock). Other dimensions of family income dynamics may moderate the influence of income variability. For instance, an upward trajectory, even if occurring through a series of income shocks, may mitigate the negative effects of the instability. Also potentially important, but difficult to measure, is the predictability or intentionality of income changes. Regular, predictable changes in family income, such as those experienced by seasonal workers, might not disrupt family processes because they can be anticipated and plans can be made to accommodate or smooth consumption. Income instability that occurs because of intentional reallocations of parental time might be less disruptive if parents are making choices to substitute time at home for income.
By adolescence, increased stress and chaos in the home, the disruption of routines, and decreased monitoring (all potential mediators of income instability discussed earlier) are less likely to affect the level of cognitive or academic skills and knowledge than they are to affect the ability to focus attention on the learning tasks of school and behavior at school (with expulsions and suspensions representing one end of that behavioral distribution; Fredricks et al. 2004). Such behavioral outcomes have been hypothesized to drive associations between early childhood interventions and adult outcomes (Heckman 2006).
Several features of the local economic environment and individual and household characteristics could play a role in exacerbating or attenuating the influence of income instability on the lives of families and adolescents. Here we focus on three key developmentally relevant moderators—adolescent age, gender, and race—and three key policy-relevant moderators—household income level, household liquid assets, and state economic and policy context.
Age, Gender, and Race
The effects of income instability may depend on the timing of income change relative to a child’s age or developmental stage. Although the entire period of adolescence may be particularly vulnerable to income instability, these effects may be exacerbated during key transition periods within adolescence, including periods of physical growth (early adolescence) and school transitions (e.g., entering high school; Graber et al. 1996). Transitions to middle school are associated with a loss in self-esteem, school engagement, and achievement (Alspaugh 1998; Seidman et al. 1994), and the potential harmful effects of this already sensitive period may be exacerbated by fluctuations in income. On the other hand, older adolescents may be relied on to take on adult-like tasks and responsibilities in the face of increased household chaos; and although such increased responsibilities could enhance confidence and self-esteem, less time might be available to school-related tasks (Gennetian et al. 2004).
In prior research, boys, less-adaptable children, and children with more behavior problems have been found to be more vulnerable to family structure instability and residential mobility (Ackerman et al. 1999; Clampet-Lundquist et al. 2011). On the other hand, girls have been found to benefit the most from residential moves to lower-poverty neighborhoods, particularly through feelings of safety and improved mental health (Ludwig et al. 2013). Finally, families of African American and minority youth often live in areas of concentrated neighborhood poverty (Kasarda 1993). These households may have limited information about the range of financial credit and loan options, as well as limited access to low-cost financial alternatives to buffer negative income shocks; they may therefore be more vulnerable to the risks of income instability (Barr 2012).
Income Level and Assets
The poorest households are the least likely to have compensating economic or psychological resources to buffer against the ramifications of negative income shocks. Low-income households have a higher risk of precarious employment (Kalleberg 2009), relationship instability and thus contributions of additional earners (Teachman et al. 2000), and limited savings or credit (Barr 2012). When income fluctuates above and below a particular threshold, low-income households may lose their eligibility for public benefits and employment-based safety net programs (Romich 2006). Indeed, the unfavorable relationship between income volatility and participation in a variety of food assistance programs is well documented (Joliffe and Ziliak 2008; Moffitt and Ribar 2008; Newman 2008).
The extent of the consequences of income instability on family and children’s lives may also depend on the presence of savings or access to other types of liquid assets. For the wealthy, cutting back on nonessentials and leisure, or dipping into savings, may be available strategies to cushion and minimize the effects of negative income shocks. The poor have fewer low-cost ways to buffer against income instability. Two-thirds of low-income families do not have enough liquid financial resources to cover three months of consumption at the federal poverty level (McKernan and Ratcliffe 2009). Many of these same households lack access to liquid financial resources either in the form of savings or low-cost credit and do not have insurance to buffer against the financial consequences of unexpected events (Barr 2012).
State Economic and Policy Contexts
The generosity of safety net benefits and the availability of jobs are two aspects of the local context that might moderate the relationship between income instability and adolescent outcomes. We hypothesize that more-generous benefit regimes (which may provide a cushion for households in times of income drops) and greater availability of jobs (which may reduce the incidence or length of earnings loss) might reduce the potential harm from income instability. We use maximum Temporary Assistance for Needy Families (TANF) benefits for a family of three as a measure of the generosity of safety net benefits; we use state-level unemployment rates as a measure of job availability.
Data and Sample
Data for this study come from the SIPP 2004 Panel, a nationally representative survey of households conducted by the U.S. Census Bureau. The SIPP is uniquely positioned to answer the questions posed in this study because it collects monthly income information, select parenting measures, and meaningful adolescent schooling outcomes. The 2004 panel followed households for 12 waves collected every four months, collecting data on select schooling outcomes for adolescents in the third and eighth waves. We use the first eight waves, collected from February 2004 to September 2006.
The unit of observation in this sample is adolescents residing in households with at least one child aged 12–17 by the eighth wave of data collection and missing no more than two waves of income data (N = 6,933 adolescents). Approximately 69 % of households have one adolescent (N = 4,783), 26 % have two adolescents (N = 1,803), and 5 % have more than two adolescents (N = 347). We restrict our sample to households in which the reference person is the parent, grandparent, brother/sister, other relatives, or foster guardian of the adolescent.
Household Average Monthly Income
Average monthly income is a composite variable computed by the U.S. Census Bureau, which adds the reported pretax income of everyone in the household. This measure includes earned income, cash transfer payments (i.e., means-tested income including cash values of food stamps), lump-sum and one-time payments, regular salary or other income from self-owned business, property income, and any interest and dividend income (Westat 2007). It is collected at each wave, once every four months, and reported on for the previous four months. We calculate the natural log of this value so that a one-unit change in income level is approximately 1 percentage point. Imputed values of income are available in the SIPP data. The imputed values are generated using hot-deck imputation techniques from a randomly selected case that was observationally similar on a number of variables (Westat 2007). For cases in which the entire household survey was missing data for a wave, the SIPP did not impute any value and indicated this as a missing value. In our main analyses, we use all reported and imputed values, but we also test the sensitivity of those results to excluding observations with substantial imputed income. Finally, studies have shown that the income data collected in the SIPP are subject to reporter seam bias, such that income is reported with more error when recalled back to previous months but much less error when reporting for the current month (Hill 1987; Moore 2007). Therefore, we use only income reported for the current month in which the data were collected at each wave. Given this restriction, monthly income is reported three times per year for each household.
We use total household income, rather than family income, to represent the household economic context. Unrelated household members may contribute to expenses, such as rent and food costs, and changes in an unrelated household member’s income may alter the resources and stress levels of both parents and children. In addition, household income as a unit of analysis maintains the national representativeness of the data set, in which households were randomly selected to participate in the study.
Household Intrayear Income Instability
We construct five measures capturing the magnitude, frequency, and direction of intrayear income changes across eight waves of data. Given the limited prior empirical work as guidance, we are intentionally judicious in the number of income instability measures examined to minimize chance findings.
We capture the magnitude of income instability with two measures. The first measure is the more common coefficient of variation (CV), the ratio of the household’s standard deviation of monthly income to the household’s mean monthly income. The advantages of the CV are that it is invariant to scale or absolute level changes in income and lends itself to subgroup analysis (Newman 2008). The second measure is the absolute value of the average percentage change of income, or 100 × [(Yt – Yt – 1)| / Yt – 1]), which captures the relative size of wave-to-wave changes in income (see Dynan et al. 2007). We use the natural logarithm of the raw value to adjust for positive skewness for both measures.
We measure the frequency of monthly income changes using two additional measures. The first counts total number of income shocks (positive and negative) as changes in income across consecutive waves above a certain magnitude of 33 % (see Elder 1974 for the orignal developmentally motivated criterion for this threshold). The second is a measure of the number of consecutive waves without an income shock of 33 % or greater. This measure captures the longest period within the eight waves that a household experienced relative income stability.
Finally, we explore one measure of the direction of income change—large negative shock—depicted as a binary variable indicating whether a household experienced a negative change in income equal to 50 % or more of the previous wave’s income. This is a higher threshold than we use to measure the frequency of shocks; we set this threshold higher in order to capture large changes relative to base income. All income instability measures are computed using the value of monthly income reported in the most recent month of each of the eight waves leading up to the school outcome measure.
Adolescent School Behavior
School engagement is reported by the household reference person in response to three statements that generally tap behavioral engagement in school (rather than emotional or cognitive engagement; Fredricks et al. 2004): (1) In general, (child’s name) likes to go to school; (2) (child’s name) is interested in school work; (3) (child’s name) works hard in school. For each item, the respondent indicates whether each statement is not true, sometimes true, or often true on a response scale ranging from 1 to 3, with higher scores indicating more engagement. Consistent with prior work in the field on this measure and to increase the reliability of the measure of school engagement (Le Menestrel et al. 1999), we combine items to create a composite scale. We then dichotomize the summary scale to indicate whether the adolescent is engaged in school at the highest level—that is, had a score of 3 on all items (52.1 % of the sample was highly engaged). Expulsion/suspension is measured with a dichotomous variable, answered by the household reference person, about whether the child has ever been suspended, excluded, or expelled from school (true for 10.1 % of the sample).
A set of control variables are included in all models, which capture parent and household characteristics at Wave 1. These include parental race (white, black, Asian, or other), parental gender (1 = female); parental age, education (less than high school, high school/GED, college, or beyond college), employment status (full-time, part-time, mixed part-time/full-time, or unemployed), and marital status (married, never married, widowed/separated/divorced); indicators for whether three or more children live in the household and for whether the youngest child in the family is under 6 years old; and the relationship of the reference person to the adolescent (approximately 95 % of reference persons are a parent of the adolescent). Finally, we control for adolescent gender and age at Wave 8.
Table 1 presents weighted descriptive characteristics of the sample overall and by income quintile. Approximately 79 % of the sample is white (includes Hispanic whites), average monthly household income is $5,911, and two-thirds of household heads have a high school diploma or less. The bottom panel of Table 1 also presents weighted descriptive statistics of the income instability measures.1 The intrayear CV for the first income quintile is .55, compared with CVs of .23–.33 in the higher quintiles (p < .05, except the comparison between quintiles 3 and 5). Whereas all households experience nonnegligible income instability according to these measures, the magnitudes of these changes are largest among households in the lowest-income quintiles. Over the roughly 2.5-year period of data (from 2004 to 2006), households in the highest income quintile experienced an average of 1.7 income shocks, compared with 3.2 for households in the lowest income quintile. The lowest-income households are nearly 20 percentage points more likely to have experienced a negative shock than households in the second through fifth income quintiles.
The correlations of each of the income instability dimensions are presented in Table 2. Correlations among instability dimensions are high, ranging from .55 to .92 in magnitude. The two measures of frequency—total shocks and length of stability—are the highest correlated dimensions of income instability (r = –.92), followed by total shocks and average percentage change (r = .86). Measures of income instability are not as highly correlated with income level (r = –.18 to –.17). We also find higher correlations between income level and instability in the lowest two income quintiles (results not shown).
In light of the richness of the data, we also constructed a variety of alternative measures including a measure of interyear income instability, comparing Waves 2–4 with Waves 5–8, and a measure of intrayear income instability over a shorter time frame of 12 months using data from Waves 6–8. Both of these alternative constructions more closely mimic the measurement options available with other data sets (e.g., measuring annualized income). The correlation matrix of the income instability measures under these reconfigurations are presented in Table 6 in the appendix. In almost all cases, the between-year (or interyear) instability measures show higher correlations among the five income instability measures than the equivalent correlations among the within-year (or intrayear) measures, suggesting that the intrayear measures are potentially better suited for unpacking distinctions in the dimensions of income instability. We also find that measures of intrayear instability constructed over a one-year period are relatively highly correlated with measures of intrayear instability constructed over Waves 1–8 (i.e., a period of 32 months). And, more generally, varying the measurement period for intrayear income instability has less impact on the correlations of intrayear income instability than does redefining income instability as year-to-year change. Throughout this article, we note our results with these alternative measures, but our main results use intrayear income instability across the full eight waves of data.
We begin with an estimate of income instability with no controls (Model 1). We then reestimate the model adding income level (Model 2) and then adding covariates (Model 3). Additionally, we run Model 3 by income quintile groups (measured in Wave 1) to minimize any confounds resulting from large fluctuations in income level over time (that might move households from one quintile to another in either direction) with variations in income within level. We then examine subgroup differences through interactions between adolescent characteristics (gender, race, and age) and the measures of income instability for adolescents in the bottom quintile.
Even with a potentially large set of family- and child-level control variables, these analyses are observational, and any ensuing statistical associations will be vulnerable to omitted variable, measurement, and simultaneity bias. Our estimates may be upwardly biased if, for example, particularly ambitious or risk-seeking individuals move frequently between jobs generating many fluctuations in income even if average income level remains high. Our estimates may be biased toward 0 if parents are juggling multiple types of public assistance, with altering periods of recertification, and resulting payments are lumpy and unpredictable, but their housing and health care are stable. The relative influences of these biases are also likely to vary by income quintile.
Notably, these types of biases are no different from those that troubled early research attempting to identify the effects of income. Over time, research on the effects of income level and poverty has made substantial progress in deriving causal estimates by exploiting variation in income produced by policy changes (as a type of natural experiment) or by experimental social programs (Dahl and Lochner 2012; Duncan et al. 2011; Milligan and Stabile 2011). We considered alternative analytic strategies, such as fixed-effect models using data from multiple time points in the SIPP, but had insufficient variability in the outcomes across time to reliably estimate these models that would have controlled for unobserved within-child heterogeneity. Similar to the trajectory of research on income level and child development, our admittedly descriptive analysis is an important first step to furthering research on income instability during childhood.
Relationships Between Income Instability and Adolescent Schooling Behavior
Table 3 presents the relationship between income instability and adolescent school engagement and expulsions/suspensions from the regression models. As shown in Model 1, there is a statistically significant unconditional relationship between each dimension of income instability and both schooling outcomes in the expected direction, with higher instability being negatively associated with adolescent school engagement and positively associated with expulsion/suspension rates.
As shown in Models 2 and 3, the direction of these relationships remains the same, but the size of the odds ratios is attenuated (approach 0) when income level and parent, household, and adolescent covariates are added to the model. For school engagement, the magnitude of the relationship with average percentage change remains negative and statistically significant (OR = 0.72, p < .05), as does the relationship with total number of shocks (OR = 0.97, p < .05). Consistent with instability having adverse effects, the number of consecutive waves with no income shock is associated with a small increase in the odds of high school engagement (OR = 1.0, p < .01). These effects are small to moderate in size: for instance, the odds of high school engagement are reduced by 3 % for each additional income shock and by 28 % for each percentage point increase in average percentage change in income, net of all control variables.
The relationships between income instability and expulsions/suspensions are quite large in the bivariate regressions (Model 1) but less robust to the inclusion of income level and control variables (Models 2 and 3). In Model 3, only the variable consecutive waves with no income shock remains statistically associated with expulsions/suspensions (OR = 0.95, p < .05).
Table 4 presents results of the fully controlled model (Model 3) by income quintile. For school engagement, the odds ratios are negative (less than 1) and statistically significant for adolescents at the top and bottom of the income distribution, but not for those in the middle. The results suggest a negative relationship between income instability and school engagement for both the lowest and highest income children. (Note that the differences in the size of the odds ratios are not statistically significant by income quintile.) Unlike school engagement, the relationships between the various dimensions of income instability and adolescent expulsions/suspensions are only observed among adolescents in the lowest income quintile (quintile 1). The odds ratio estimates statistically differ between quintiles 1 and 2 (for average percentage change and total shocks).
The models from Table 4 include measures of average income level (odds ratios not shown). Notably, there are no detectable differences in the association between average monthly income and school engagement by income quintile. For adolescents in the lowest income quintile, income instability is statistically related to expulsions/suspensions but average income level is not. For adolescents in the second-lowest quintile, the opposite is true: average income level is statistically related to expulsions/suspensions, but income instability is not. We also ran models combining adolescents in quintiles 4 and 5 (N = 183, or 6.8 % of the analytic sample), finding similar results on the income instability measures with slightly more precise standard errors, but also a statistically significant relationship between income level and reductions in expulsions/suspensions.
Our initial hypotheses propose that the frequency of income change might matter for adolescent schooling behavior irrespective of the direction of income changes. As a supplementary analysis, we estimated Model 3 for the full sample and by income quintile groups, separating the number of negative income shocks (a 33 % or greater income decrease from the previous wave’s income) from the number of positive income shocks (a 33 % or greater increase from the previous wave’s income; not shown). Of the nearly 70 % of the adolescent sample that experienced at least one positive shock, one-half experienced at least one negative shock, two-thirds (N = 4,122) experienced at least one negative and one positive shock, and 24 % (N = 1,689) experienced neither negative nor positive shock. In supplemental analyses, we did not find patterns that suggest statistically differing relationships of positive shocks from those of negative shocks.
To place our main results in the context of prior research focused on income level as the key independent variable of interest, we also offer results (shown in Table 7 of the appendix) of models using average monthly income to predict the adolescent outcomes (Models 1a and 1b), with and without control variables that include income instability (Models 2 and 3, which replicate Models 3 and 4 in Table 3). Average monthly income has no statistically significant relationship with school engagement when covariates are included in the model but does have a relatively stable relationship when each of the measures of income instability is entered separately into the model (with no covariates). On the other hand, average monthly income has a positive statistically significant relationship with adolescent expulsions and suspensions; this relationship is of similar size and statistical significance when each of the measures of income instability is entered separately. The odds ratios are reduced with the addition of covariates but remain statistically significant (p < .05). Thus, for the full sample, covariates seem to explain much of the relationship between income level and school engagement but not between income instability and school engagement. Neither covariates nor measures of income instability appear to qualitatively affect the relationship between income level and adolescent expulsions and suspensions.
Finally, we tested the sensitivity of our results to excluding observations with substantial imputation of income data (e.g., more than 25 %). The results are qualitatively similar, with similarly sized coefficients but larger standard errors because of decreased sample size.
We next assess whether the observed relationships within each income quintile are moderated by three key adolescent characteristics—gender (male vs. female), race (minorities vs. whites), and age (12- to 14-year-olds versus 15- to 17-year-olds)—by adding an interaction term in each of separate regressions by income quintile. No statistically significant interaction terms appear for age, gender, or racial group characteristics among adolescents in households in the second, third, fourth, and fifth quintile (results not shown).
Table 5 shows results among adolescents in the lowest income quintile. The relationship between income instability and schooling outcomes does not differ by gender. Although minority status does not differentially predict the relationship between income instability and school engagement, minority status moderates the relationship between expulsions/suspensions and average percentage change (OR = 8.4, p < .05), the total number of income shocks (OR = 1.3, p < .05), income stability (OR = 0.80, p < .05), and any large income shock (OR = 2.4, p < .01). These results consistently show no statistically detectable relationship between income instability and the odds of expulsion/suspension for white students but large and statistically significant adverse effects of income instability on the odds of expulsion/suspension for nonwhite students.
Adolescent age moderates the relationship between school engagement and three measures of income change: average percentage change (OR = 0.33, p < .05), consecutive waves with no income shock (OR = 1.16, p < .05), and any large negative shock (OR = 0.59, p < .05). In this case, income instability lowers the odds of high school engagement for older (15–17 years) but not for younger (12–14 years) adolescents. Similarly, in models predicting having ever been expelled, the age by instability interaction with average percentage change is statistically significant (OR = 4.4; p < .05), suggesting that an increase in average percentage change of income increases older but not younger adolescents’ odds of expulsion/suspension.
We tested two other potential moderators: household liquid assets and state economic context (results not shown). We hypothesized that the relationship between income instability and adolescent school behavior may depend on the availability of compensating resources to smooth consumption or buffer the consequences of unexpected dips in income. The SIPP collected information on household liquid assets in Waves 3 and 8. We constructed a measure of household liquid assets measured in Wave 3 using the natural log of this composite variable, which includes the value of investments in IRAs, Keogh accounts, 401(k) accounts, and other savings accounts. The level of liquid assets substantially varies across income quintiles: nearly 70 % of households with adolescents in quintile 1 have no reported liquid assets, compared with only 10 % of households in quintile 5. The mean value of assets among households who do have them is just over $1,300 for households in quintile 1, compared with nearly $21,000 for households in quintile 5. We reestimate our models controlling for a continuous version of the liquid assets variable, the interaction of a continuous measure of liquid assets with income instability for the full sample, and by income quintile (results not shown). The results suggest that the relationship between instability and adolescent school behaviors does not vary by asset level.
In addition, we hypothesized that state economic context might buffer the effects of household income instability on adolescent school outcomes. To test this hypothesis, we estimated models including main effects and interactions terms for state unemployment rates and state welfare benefits levels. The results show that state-level unemployment rates, but not levels of safety net benefits, have a buffering influence for families at the bottom of the income distribution. The interactions between unemployment and the instability measure are consistently positive and statistically significant for the bottom quintile, suggesting that an increase in income variability is less detrimental, or possibly even beneficial, to school engagement in the context of higher unemployment rates. This is a counterintuitive finding, but it is possible that parents are more available and increase their monitoring of their adolescent’s schooling during economic downturns. Adolescents may also see the negative ramifications of the poor economy through their parents’ underemployment or unemployment, which may act as a motivation to increase their engagement in school.
Discussion and Conclusion
The economics literature has documented an increase in earnings and income volatility since the 1970s, although the timing and magnitude of that increase is still debated (Celik et al. 2012; Dynan et al. 2012; Hacker 2008; Shin and Solon 2011). A complementary study that we have conducted with multiple panels of SIPP data further suggests that the gap in income instability has increased fivefold since the 1980s, with the lowest-income households experiencing more-frequent income change than the highest-income households (Morris et al. 2014). To date, no study has documented the consequences of such increased income instability for children’s well-being, particularly during children’s most vulnerable developmental transitions.
We take advantage of data from the SIPP that is one of the few (or, only) data sets that provide information on household monthly income, measured multiple times within a year, and have selected measurement of children’s developmental outcomes. We find that frequent income change, as measured within year, is nearly double among households with adolescents in the lowest-income quintile compared with those in the highest income quintile. When we control for a range of demographic characteristics, income instability predicts a lower likelihood of school engagement and a higher likelihood of a school expulsion or suspension. The associations between income instability measures and school engagement hold for adolescents overall—at the low and high ends of the income distribution—and are not moderated by gender or race of the child. We observe larger (more unfavorable) associations, however, between income instability and school engagement for older than for younger adolescents. In contrast, we find that income instability is related to an increased likelihood of expulsions/suspensions only for those adolescents in the lowest income quintile, but this relationship does not vary systematically by age.
The different patterns of findings on the school behavior outcomes may be due to the fact that school engagement is a positive and affective construct, related to attitudes about school, while expulsion/suspensions are negative outcomes and are consequences of misbehavior. Both are important but different aspects of schooling behavior that may be associated with academic achievement. Low-income minority children are at the greatest relative risk of being expelled and suspended (KewalRamani et al. 2007), and the resulting loss of in-classroom time from such infractions may reduce adolescent’s ability to achieve in school (Gregory et al. 2010). The sensitivity of older adolescents’ school engagement to income instability may be due to a variety of factors, including older adolescents taking on additional hours of work to contribute to household finances or take on more duties at home (e.g., caring for younger children, chores) in light of parents taking on additional employment, both of which can interfere with schooling. Some of these hypotheses are testable and should be examined in future work.
Several important limitations of this study are worth noting. First, although the measures of school behavior are used regularly in research on educational outcomes, they suffer from the standard limitations, and potential biases, of being parent-reported. Parents may not have full information about their child’s school engagement (especially) or even of all cases of expulsion/suspension, and parents’ own state of mind may bias their reports. Measurement error of this type would be most likely to bias our estimates toward 0, masking even larger unfavorable association between income instability and adolescent school behaviors.
Second, although several variables were included as controls in the model, the resulting observed relationships between income instability and adolescent schooling—in particular, expulsions/suspension among the lowest-income adolescents—could be confounded by a variety of unobserved characteristics of adolescents, their families, or households of residence, and their local environments. In theory, longitudinal data offer opportunities to better control for observed and unobserved sources of heterogeneity. We are limited in opportunities to successfully conduct these types of analyses because the SIPP’s expulsions/suspensions measure is a lifetime measure (not subject to change across the panel). In addition, although the school engagement scale is meaningful in a cross-sectional sense (i.e., there is enough variation in the scale to capture differences in the distribution of school engagement at any one point in time), it is limited to three very specific items and thus is not a sensitive enough indicator to capture intra-academic-year differences in school behavior. We also note that it is methodologically more challenging to identify the causal effect of frequent income change than, for example, to identify the causal impact of a one-time income shock, as might be feasible with methodologies such as fixed effects with panel data.
Third, our study is limited in examining the co-occurring and longitudinal nature of positive and/or negative income change, the timing of this change as it interacts with children’s developmental trajectories, and the interrelationships with precipitating factors. Fluctuating positive changes in income through childhood could be preferable to chronically low income for supporting longer-term developmental outcomes. On the other hand, fluctuations in family resources at any point of childhood could be a source of chaos in the home environment, and may disrupt the regularity or quality of parent–child interactions and daily activities. Adolescents may be particularly vulnerable because of the co-occurrence of these events with their own dramatic changes in physical and emotional development and heightened awareness of their external environments to implications of these changes. Whereas the SIPP might be well suited to investigating some aspects of these questions—for example, on selected precipitating factors related to employment or household structure—it is not well suited to investigate others, such as children’s lifetime experience with income change.
Fourth, it is beyond the scope of this study to fully examine the effects of instability by its predictability or intentionality, by the specific source of income, or by local policy or economic context. It is conceivable that regular, predictable changes in family income, such as those experienced by seasonal workers, might not disrupt family processes because they can be anticipated, allowing families to plan to accommodate or smooth consumption. The role of income from means-tested programs is also unclear: safety net assistance cushions households during periods of earnings fluctuations, but eligibility rules and processes for the recertification of benefits may amplify fluctuating earnings (see, for example, recent work on the role of the social safety net in the context of the recent 2008 recession by Bitler et al. 2013; Moffitt 2012). The measures of income and income instability in our analysis encompass income from all sources and income fluctuations of both predictable and unpredictable forms. Future extensions of this work can incorporate policy variables to examine the potential buffering role of the safety net, as well as characteristics of the local labor market to examine heterogeneous responses or repercussions to earnings instability.
Research on income instability is particularly timely: the recent U.S. economic crisis has further exposed millions of families to volatile economic situations. Sensitivity to economic fluctuations is particularly acute for those families living on the margins of poverty. Understanding the implications of income instability on family life and children lays the groundwork for more effective federal and state policy making across numerous facets of policy design. Most social policies targeting low-income families are not designed to flexibly accommodate frequent income change. For example, steady employment and earnings are an (implicit) eligibility requirement for several important safety net programs. A job loss not only translates into lost earnings but also to lost benefits from employment-based safety net programs (Aber et al. 2012).
Important directions for future research include opportunities to untangle the causal effects of income level from income change. To do this, studies need frequent measurement of income, detailed and well-validated measures of family life and children’s developmental outcomes, and strategies to leverage exogenous difference that will separately identify income from income instability. One promising example may be experimental studies that were designed to enhance income but, because of the design of the payment structure, may also have increased income instability. Another example is naturally occurring changes in income induced by the economic or policy context: that is, a natural experiment that exploits income instability imposed by the timing of public assistance income or the temporal nature of the institution of particular policies (Gennetian et al. 2013). Given today’s economic reality, the implications of income instability on family life as well as the role that income instability plays in the context of public programs should be actively integrated as critical components of future research on the impacts of income and poverty on family and child well-being.
We also created a measure of interyear income instability comparable to what was constructed for Yeung et al. (2002) using the PSID-CDS, and we similarly found that 20 % of households experienced at least one income change of 30 % or more from year to year.