THE GRADE-LEVEL AND CUMULATIVE OUTCOMES OF ABSENTEEISM DURING ELEMENTARY SCHOOL

Abstract

Nationally representative data from the Early Childhood Longitudinal Study Kindergarten Class of 2011 (n = 14,370) were used to examine the grade-level and cumulative outcomes of school absenteeism between kindergarten and fifth grade for students’ school performance in the United States. Students who were more frequently absent in any year of elementary school demonstrated lower academic, executive function, and socio-emotional outcomes. Although there was little variation in the magnitude of associations across grade levels, there was evidence of cumulative associations. Specifically, students who were consistently absent throughout elementary school tended to have lower outcomes across developmental domains in the long-term. The negative links between absenteeism and outcomes were larger for Black than White students, but few other subgroup differences emerged.

It is clear that students who are more frequently absent from school are at greater risk for school difficulty (e.g., Ansari & Pianta, 2019; Gershenson et al., 2017; Gottfried, 2011; Morrissey et al., 2013; Smerillo et al., 2018). Such evidence has motivated policymakers and practitioners to use student absenteeism as one of the key indicators of school performance (Gottfried & Hutt, 2019; Jordan & Miller, 2018). Nonetheless, the extant literature that has studied issues related to absenteeism and student outcomes has almost exclusively focused on end-of-year performance (i.e., in the same year that the absences took place) or on how absences link to students’ academic achievement. In doing so, these studies have overlooked non-testing outcomes as well as the possible cumulative outcomes of school absences and variation across grade-levels and for subgroups of children. Understanding the cumulative outcomes of absenteeism and the extent to which it varies across time and subgroups has important implications for policy and practice as it can inform key targets and periods for intervention.

To fill in these gaps in scientific knowledge, we rely on data from the Early Childhood Longitudinal Study, Kindergarten Class of 2011 (ECLS-K: 2011), which is a nationally representative sample of approximately 14,370 students who were followed from kindergarten through fifth grade, and is the only known longitudinal dataset to include attendance measures and numerous child outcomes. With that in mind, we use these data to examine the cumulative and grade-specific outcomes of absenteeism for students’ academic achievement, executive function, and socio-emotional development across the elementary school years, which represents a critical juncture of study because (a) absenteeism is at its peak in the first few years of children’s educational careers and (b) these are habits that become established in the earliest years and remain stable from there forth Ansari & Pianta, 2019). As part of this effort, we also consider whether the outcomes of absenteeism are larger for different groups of children cited as key groups in attendance research (Gottfried & Hutt, 2019), including children from racial/ethnic minority groups, children learning English as a second language, low-income children, and for boys as compared with girls. In doing so, this investigation is poised to provide one of the most comprehensive snapshots of elementary school absenteeism in the United States.

The Consequences of Absenteeism

Even though much of the extant literature has frequently studied absenteeism among middle school and high school students, with 12–19% considered chronically absent as a result of missing at least 10% of school (U.S. Department of Education, 2016), this is not a behavior that starts in the older grades. Indeed, recent national estimates reveal that there is an absenteeism crisis in the United States and it begins at the very start of children’s educational careers (Balfanz & Byrnes, 2012). Moreover, across the elementary school years, 5–15% of students are chronically absent (Bauer et al., 2018). As such, understanding and ultimately reducing student absenteeism has been of great research and policy interest.

With regards to the academic consequences of missing school, a number of studies have linked school absences with students’ math and literacy achievement in the primary and secondary school years (Ansari & Pianta, 2019, Gottfried, 2011; Gershenson et al., 2017; Morrissey et al., 2013). Poor school attendance between grades six through twelve has also been linked with increased odds of grade retention and high school dropout (Neild & Balfanz, 2006). In fact, once students enter high school, regular attendance is among the strongest predictors of graduation (Allensworth & Easton, 2007). In Chicago, ninth graders who missed more than one month of school per semester had less than a one in ten chance of graduating (Allensworth & Easton, 2007). Based on previous quasi-experimental studies for students between third and tenth grade (Gottfried, 2010; Goodman, 2014) as well as a descriptive study in which Gottfried & Gee (2017) explored the individual characteristics of kindergartners as well as the ramifications of their immediate home environments for school absenteeism, it is clear that absenteeism is not simply a proxy for other factors (e.g., family dysfunction), but rather, is a key independent factor that has implications for students’ maladjustment in areas of academic learning. Ultimately, students do less well academically when absent from school due to unequal opportunities to learn.

A small number of studies have also considered the link between absenteeism and non-testing outcomes, namely socio-emotional development and executive function outcomes. Early research in this area had suggested that there were social-behavioral concerns associated with school absences (Ekstrom et al., 1986; Finn, 1989; Newmann, 1981). More recent studies of children in the first years of school suggested that that early school absences predicted worsened socio-emotional (Gottfried, 2014) and executive function (i.e., working memory, inhibitory control, and the ability to focus attention; Fuhs et al., 2018) skills. Newer research also finds that middle and high school-aged students who miss more days of school also tend to demonstrate elevated levels of delinquent and aggressive behaviors and anxiety and depression, and lower levels of cooperation and self-control (Ansari & Pianta, 2019; Eaton et al., 2008; Hallfors et al, 2002).

The scarcity of work on these non-academic domains, especially in the early years of school, likely reflects the fact that schools are often perceived of as academic institutions; however, schools and classrooms also play important roles in shaping students’ social-behavior and executive function skills in addition to achievement (Rimm-Kaufman et al., 2002; Ryan & Patrick, 2001), which represent skills that shape children’s long-term educational success and well-being (Jones et al., 2015; Watts et al., 2018). Thus, the earliest years of education can (and do) have a significant impact on children’s academic growth and also on development more broadly (Duncan et al., 2007; Jones et al., 2015; McCoy et al., 2017), and consequently, the limited studies in this non-testing sphere are certainly merited and require further attention.

Indeed, theories of change regarding absenteeism argue that not only do absent students receive fewer hours of instruction, hence impacting achievement, but they also have fewer opportunities to interact with teachers and classmates, hence impacting social development (Finn, 1989). Thus, absenteeism is linked with less optimal outcomes across key developmental domains, especially in the early years, because missing school increases feelings of disengagement and provides students with fewer opportunities to develop and practice skills in the context of high quality and structured environments (Fuhs et al., 2013; Goble & Pianta, 2016; Pianta et al., 2020; Weiland et al., 2013).

Despite the importance of regular school attendance, there has been little systematic evaluation as to whether school absenteeism matters simultaneously for a broad range of student outcomes. And even though the educational and developmental literatures discussed above have provided important insight regarding the negative ramifications of student absenteeism, they have focused on either a single year of absenteeism and/or one year of student outcomes. In focusing only on one year of absenteeism or outcomes, this has greatly limited our understanding of the ways in which absenteeism operates as part of long-term developmental and educational cycles that begin early on in students’ educational careers and that could have long-term consequences. Rather surprisingly, very few studies have explored whether (a) the outcomes of school absences vary across the elementary school years or (b) school absences may be associated with multiple years of student outcomes. The above is an important gap in scientific knowledge for both theoretical and practical reasons.

Theoretically, the bioecological model of human development (Bronfenbrenner & Morris, 2006) has long emphasized the importance of developmental timing and sensitive periods in understanding how environments shape students’ learning. Reflecting this theoretical model, there is ample evidence to suggest that students learn more during the earliest years of school (Blair & Razza, 2007; Duncan et al., 2007; Heckman, 2008). As such, absenteeism in the earliest years of elementary school may pose greater problems for students than later absenteeism because students are missing more learning opportunities for every day missed. Practically, however, many parents continue to believe that absenteeism in the early years is not as important as absenteeism in the older grades (Ehrlich et al., 2013; Malcolm et al., 2003), though new research has brought to light the negative consequences of believing such a myth (Ansari & Pianta, 2019). Moreover, theories of cumulative risk (Sameroff et al., 1993) would contend that absenteeism across the elementary school years compounds and contributes to accumulated risk of maladjustment and functioning over time. Thus, it is critical to consider both the timing specific and cumulative outcomes of absenteeism.

In the only known test of these competing possibilities, Ansari and Pianta (2019) used non-nationally representative data and found that school absences in eighth grade mattered more for students’ school success in the following year than school absences in elementary school. At the same time, however, school absences in the elementary school years were found to have sizable cascading associations that resulted in less optimal school performance in ninth grade, in large part because absenteeism in the early years was linked with absenteeism in the later years (Ansari & Pianta, 2019). Accordingly, there is some indication that absenteeism may have both timing specific and cumulative consequences. Unfortunately, Ansari and Pianta (2019) only considered student outcomes in ninth grade and did not examine outcomes across grades to systematically evaluate the cumulative and grade-level outcomes of absenteeism.

The present study

Taken together, the present study sought to address four research questions. Specifically, we considered: (1) To what extent is absenteeism in kindergarten through fifth grade associated with students’ academic achievement, executive function, and socio-emotional development at the end of those school years? (2) Do the links between absenteeism and student outcomes vary across grade levels? (3) Are there cumulative patterns of absenteeism between kindergarten and fifth grade? And (4) Do the aforementioned associations vary as a function children’s socio-economic status, race/ethnicity, home language, and sex? Although we expected that students who were more frequently absent would do less well across all outcome measures, the remainder of our research questions were largely exploratory; thus, we did not make directional hypotheses.

It is important to note that our work is descriptive in nature, as we are using non-experimental data to identify patterns in a nationally representative sample of children in the U.S. population. Even though causal research methods are often the “gold standard” for educational and developmental research, some aspects of human behavior (such as absenteeism) cannot be readily randomized (see also Gershoff et al., 2018). Thus, although we do not attribute causality to our study, answers to our research questions can provide insight on key patterns and sources of risk for multiple stakeholders of research, including both school leaders and educational policy makers (Loeb et al., 2017). As described by Loeb and colleagues (2017), rigorous descriptive and correlational methods, such as those used in the present investigation, can help build our cumulative knowledge on child development and help us better understand associations of interest, such as the link between absenteeism and child outcomes. Additionally, the use of robustness tests and sensitivity analyses, as done in the present study, has the potential to strengthen inference in the context of correlational research (Dearing & Zachrisson, 2019).

Finally, it is important to note that in total, our study is novel to the developmental and educational sciences – taking into account multiple grade levels, multiple outcomes, as well as exploring the timing specific and cumulative outcomes of absenteeism. That said, some slices of our analyses (i.e., linking grade level absenteeism to some of our same-grade outcomes) have been previously explored in the literature. Though not a direct goal of the current study, some of our analyses serve as a means of replicating prior research on absenteeism, such as these same-grade analyses. Recent research has claimed that too few replications and robustness checks are being conducted in the educational and developmental sciences (Duncan et al., 2014). Accordingly, our study contributes in this domain as well.

Method

The present investigation utilized data from the Early Childhood Longitudinal Study – Kindergarten Class of 2010–11 (ECLS-K: 2011; Tourangeau et al. 2013), which is a nationally representative sample of kindergarten students across the United States. To date, data have been collected from multiple informants through the spring of 2016, which corresponds to the end of fifth grade. Data have been collected in the fall and spring of kindergarten, and then annually in the spring between first through fifth grade. To ensure a nationally representative sample, the ECLS-K: 2011 used a three-stage stratified sampling design. The first sampling unit was geographic region, which corresponded to approximately 3,150 counties in the United States from which a subset of sampling units were sampled. The second stage involved the selection of public and private schools that had kindergarten programs. In order to reach the target number of schools, approximately 1,320 schools were initially sampled. Then, the third and final stage of sampling involved the selection of kindergartners from the list of all enrolled students in the selected schools, stratified by race/ethnicity. A little over 770 of the approximately 1,260 schools deemed eligible consented to participate. An additional 200 substitute schools were added into the study, yielding a total of 970 schools. Within these schools, approximately 20,250 children were eligible for the base-year data collection, but only 18,170 kindergartners enrolled in the study. Of the baseline sample of 18,170 students, there was approximately 32% attrition by the end of fifth grade (first grade n ≈ 15,650; second grade n ≈ 14,450; third grade n ≈ 13,580; fourth grade n ≈ 12,920; and fifth grade n ≈ 12,350).

For the purposes of the present investigation, we used data from the surveys administered to parents and teachers and direct assessments of children between kindergarten and fifth grade. We limited our sample to students who had a valid kindergarten weight, which was required to ensure our analyses were nationally representative. After addressing missing data (average = 20%, range = 1–35%), our analytic sample consisted of approximately 14,370 students. The weighted sample consisted of 51% males and with 51% identifying as White, 14% as Black, 25% as Hispanic, 4% as Asian, and 6% as other. Approximately 15% of students spoke a language other than English at home and 50% were considered to be low-income (as measured by receipt of the Special Supplemental Nutrition Program for Women, Infants, and Children). Finally, 89% of study participants attended a public school (for other weighted sample descriptives, see Table 1).

Table 1.

Weighted descriptive statistics for focal covariates.

	Mean	Standard Deviation
Child characteristics
Male	.51
White	.51
Black	.14
Hispanic	.25
Asian	.04
Other	.06
English language learner	.15
Poor health rating	.14
Educational experiences
Full-day kindergarten	.83
Public kindergarten	.89
Hours per week in center-based prekindergarten	15.56	(13.99)
Hours per week in before/after school care during kindergarten	5.93	(9.92)
Out-of-home care prior to prekindergarten	.75
Household characteristics
Two-partner household	.78
Number of siblings	1.49	(1.12)
Low-income	.50
Mother has less than a college degree	.69
Mother has at least a college degree	.31
Father has less than a college degree a	.68
Father has at least a college degree a	.32
Mother not employed	.37
Mother employed part time	.21
Mother employed full time	.42
Father not employed a	.10
Father employed part time a	.06
Father employed full time a	.84
Number of books at home	86.87	(136.37)
Home activities (scale)	2.92	(0.47)
School-Going Practices and Routines
Bus to school	0.32
Distance	4.99	(3.86)
Time to school	12.29	(7.69)
Regular breakfasts at home	5.44	(1.66)
Regular dinners at home	5.50	(1.82)

Notes.

^aEstimates for father education and employment are provided for the group of families with a father in the home.

Measures

Table 2 presents descriptive statistics for all focal predictors and outcomes used in this study.

Table 2.

Weighted descriptive statistics for focal predictor and outcome variables.

	Days absent a	Math	Language and literacy	Cognitive flexibility	Working memory	Social skills	Internalizing behavior	Externalizing behavior
Fall of kindergarten	n/a	35.28 (11.42)	53.90 (11.22)	14.21 (3.30)	432.77 (30.07)	2.99 (0.59)	1.46 (0.49)	1.60 (0.62)
Spring of kindergarten	5.92 (4.77)	49.39 (13.23)	68.55 (14.22)	15.16 (2.78)	449.55 (30.37)	3.13 (0.60)	1.51 (0.50)	1.64 (0.64)
Spring of first grade	5.02 (4.27)	72.04 (15.49)	94.22 (17.75)	16.06 (2.32)	469.30 (25.58)	3.14 (0.60)	1.55 (0.51)	1.73 (0.62)
Spring of second grade	5.13 (4.36)	89.40 (17.97)	111.75 (17.07)	6.70 (1.34)	480.55 (23.08)	3.14 (0.61)	1.59 (0.53)	1.71 (0.62)
Spring of third grade	4.59 (4.03)	103.36 (17.97)	120.44 (15.55)	7.20 (1.098)	489.67 (21.90)	3.15 (0.61)	1.60 (0.53)	1.68 (0.61)
Spring of fourth grade	4.58 (3.88)	111.90 (17.92)	128.88 (14.86)	7.63 (0.96)	497.01 (21.47)	3.16 (0.60)	1.59 (0.55)	1.64 (0.59)
Spring of fifth grade	4.71 (4.01)	119.28 (17.69)	135.96 (15.62)	7.98 (0.95)	503.11 (22.02)	3.18 (0.60)	1.57 (0.52)	1.62 (0.59)

Note. Estimates correspond to means and those in brackets correspond to standard deviations.

^aNote, even though absenteeism in kindergarten is listed in the “spring of kindergarten” row, this estimate corresponds to absenteeism across the kindergarten year.

Days absent.

During the spring of each school year, students’ primary teachers were asked to “indicate the total number of absences for this child for the current school year.” Responses options were based on 6-point Likert scale (0 = no absences, 1= 1–4 absences, 2 = 5–7 absences, 3 = 8–10 absences, 4 = 11–19 absences, and 5 = 20 or more absences). To increase interpretability, we recoded the scale values to equal the midpoint of the response options (e.g., 1–4 absences was recoded as 2.5 absences). Students who were never absent (scale value of 0) and those who were absent for 20 or more days of the school year (scale value of 5) were coded as being absent for 0 and 20 days, respectively. It is important to note that during fourth and fifth grade, both students’ English language arts teacher and their science or math teacher responded to questions of absenteeism. Because the correlations of absenteeism across subject areas were high, we created a composite of fourth, and then, fifth grade absences (rs = .74-.85).

Academic achievement.

Students’ math and language and literacy skills were assessed at the beginning and end of the kindergarten year in addition to the spring of each subsequent grade using measures developed by the National Center for Education Statistics (α = .86-.95). The language and literacy test measured students’ print familiarity, letter recognition, and vocabulary knowledge, whereas the math assessment measured students’ problem solving, geometry, and measurement skills. For the purposes of the present study, we used the IRT-based overall scale scores for each domain, which is an estimate of the number of items a student would have answered correctly if they had been administered all of the math (146 unique questions) and language and literacy (155 unique questions) questions.

Executive function.

Students’ executive function skills were also measured with two tools that were administered in the fall and spring of kindergarten and then again in the spring of first through fifth grade. The Dimensional Change Card Sort (α = .90-.94, DCCS: Zelazo, 2006) was used to measure students’ cognitive flexibility (i.e., the ability to flexibly use rules to govern their behavior). As part of this assessment, students were asked to sort cards into different trays based on rules that changed periodically. For our purposes, we used the combined raw scores from the DCCS, which was based on the number of correct matches across the Color, Shape, and Border Games. Note that for kindergarten and first grade, the DCCS was administered with physical cards and trays. Because these assessments would have been too easy for older students, in second through fifth grade, students were administered the DCCS via a tablet. Differences in assessment reflect the logical progression of students’ development.

The second dimension of executive functioning assessed was working memory, which captures students’ ability to hold onto information to complete a task, was measured with the Numbers Reversed subtest of the Woodcock-Johnson (α = .87; Mather & Woodcock, 2001). During this assessment, students were asked to repeat a series of numbers that were dictated to the child. If study participants responded incorrectly, then the task ended. If children responded correctly to the task, then the number span increased by one digit at a time.

Social-emotional skills.

Teachers reported on different dimensions of students’ social-emotional skills using the Social Skills Rating System (SSRS; Gresham & Elliott, 1990). The SSRS is based a 4-point scale (0 = never to 3 = very often) and includes four subscales: interpersonal skills (5 items; α = .86-.87), self-control (4 items; α =.80-.82), internalizing behavior problems (4 items; α = .76-.79), and externalizing behavior problems (5 items; α = .86–89). In addition to the four subscales from the SSRS, the ECLS-K: 2011 also measured students’ approaches to learning with 6 items developed by NCES (α = .91-.92). Following an approach similar to Claessens and colleagues (2009), we collapsed these indicators into three dimensions: Internalizing behavior problems, externalizing behavior problems, and social skills (a combination of approaches to learning and socioemotional skills). And like the assessments of achievement and executive function, these teacher surveys were administered in the fall and spring of kindergarten and again in the spring of each subsequent grade.

Analytic Strategy.

All analyses were estimated within a path analytic framework in Mplus (Muthen & Muthen, 1998–2013). These models included robust standard errors to safeguard against violations of normality and missing data were accounted for with full information maximum likelihood estimation, which utilizes all available data points from each individual in estimating model parameters. All models were weighted to be nationally representative and error term were clustered at the school level. Note that, variations on the clustering of the error did not change our findings reported below. And given the large number of outcomes under study, all of our main “effect” analyses adjust for multiple comparisons using the Benjamini adjustment (Benjamini & Hochberg, 1995).

To minimize the possibility of spurious associations, all models controlled for a large number of covariates that were informed by prior studies that have examined the different reasons why students miss school (e.g., Gottfried & Gee, 2017; Gershenson et al., 2017). These indicators capture children’s characteristics (i.e., gender, race/ethnicity, English language learner status, and an indicator for whether a parent rated the child as having poor health), children’s educational experiences (i.e., enrollment in full-day kindergarten, school type, and the number of hours that the child spent in center-based pre-kindergarten and before/after school care during the kindergarten year), household characteristics (i.e., household structure, number of siblings, poverty status, parent education, parent employment, number of books in the home, home learning activities [9 items, α = .72]), and school-going practices and routines (i.e., whether the child took a school bus to school, how far the child lived from school, in miles, number of breakfasts and dinners that the family regularly had together at home). In addition, all models adjusted for lagged dependent variables (discussed in more detail below), which is recognized as one of the strongest adjustments for omitted variable bias (NICHD & Duncan, 2005).

With the above analytic framework in mind, our analyses proceeded in several steps. Our first set of analyses examined the associations between absenteeism and students’ academic, executive function, and social skills between kindergarten and fifth grade. Separate models were estimated for the different study outcomes (e.g., one model for math, one model for literacy etc.). Given the focus on timing specific associations, these models controlled for lagged dependent variables from the wave immediately prior. For example, when looking at the links between absenteeism in first grade and students’ first grade math achievement, models controlled for the spring of kindergarten math achievement. In contrast, when examining the links between second grade absenteeism and students’ math achievement, we controlled for students’ math scores in first grade. To determine whether the links between absenteeism vary across grade levels, we used an omnibus chi-square difference test to determine whether constraints across grade-levels or outcomes resulted in a significant decrease in model fit, which would be indicative of global moderation. If the omnibus test was not significant, we concluded that there was no variation across grade levels or outcome. If the omnibus test was significant, we then compared each coefficient one at a time (Paternoster et al., 1998) to determine which coefficients were significantly different across grade levels or outcome domain.

In addition to models that considered the grade-specific links between absenteeism and student outcomes, we estimated additional models that examined the cumulative outcomes of absenteeism. For these models, we created aggregate indicators of absenteeism. For example, when examining the cumulative outcomes of absenteeism between kindergarten and second grade, we created an index of absenteeism between these periods. These models adjusted for either: (a) lagged dependent variables from kindergarten entry or (b) lagged dependent variables from the wave immediately prior. In doing so, these cumulative analyses assess the extent to which absenteeism has cumulative associations with student learning and development over time in addition to year to year learning. Finally, to determine whether the outcomes of absenteeism varied for different subgroups of children, interaction terms were added between student absenteeism and the socioeconomic status, race/ethnicity, home language, and sex indicators.

Results

Before addressing our focal research questions regarding the grade-specific and cumulative outcomes of absenteeism, we examined the descriptive patterns of absenteeism between kindergarten and fifth grade. Results from these analyses revealed that absenteeism peaked in kindergarten (M = 5.29, SD = 4.77) and steadily dropped over time to 5.02–5.13 days (SD = 4.27–4.36) in first and second grade and to 4.58–4.71 days (SD = 3.88–4.03) in third through fifth grade (ps < .001). Even with the dip in absenteeism over time, there was modest cross-time correlations, ranging from .25-.43 (ps < .001). Collectively across the first six years of education, students missed approximately 32.29 days of school (SD = 21.48). But as can be seen by the standard deviation of the absenteeism indicator, there was considerable variability, with students who were infrequently absent (one standard deviation below the mean) missing only 10.81 days of school between kindergarten and fifth grade and frequent absentees (one standard deviation above the mean) missing 53.77 days of school. The top 10% of absentees missed closer to 83.48 days of school. And, finally, between kindergarten and fifth grade, rates of chronic absenteeism ranged from approximately 8–13%, with kindergarten representing the peak.

Grade-level links between absenteeism and students’ outcomes

Having established the descriptive snapshot of student absenteeism across the elementary school years, we proceeded to examine the within grade-level associations between absenteeism and students’ academic achievement, executive function, and socio-emotional skills. Results from these analyses revealed that 34 of the 42 associations estimated were statistically significant (33 of the 34 significant associations remained statistically significant with a Benjamani adjustment; see Table 3) and all path models demonstrated good fit (CFIs > .99 and RMSEAs < .02; Kline, 2010). In general, students who were more frequently absent between kindergarten and fifth grade demonstrated fewer gains in math (ds = .02-.03), language and literacy (ds = .02-.04), working memory (ds =.02-.03), and social skills (ds = .03-.08) in addition to greater levels of externalizing (ds = .02-.05) and internalizing behavioral problems (ds = .07-.11). In contrast, the links between absenteeism and students’ cognitive flexibility were somewhat less consistent. However, students who were more frequently absent in second and third grade were found to demonstrate less optimal cognitive flexibility (ds = .03-.04).

Table 3.

Results from path models examining the timing specific associations between absenteeism and student outcomes.

	Student outcome
Predictor	Math	Language and literacy	Cognitive flexibility	Working memory	Social skills	Internalizing behavior	Externalizing behavior
Days absent K → Fall of K-Spring of K	−.024 *** (.006)	−.021 ** (.007)	.000 (.010)	−.019 ** (.007)	−.028 *** (.008)	.073 *** (.010)	−.002 (.007)
Days absent G1 → Spring of K-G1 outcomes	−.026 *** (.006)	−.040 *** (.007)	−.005 (.010)	−.032 *** (.009)	−.055 *** (.010)	.092 *** (.012)	.020* (.009)
Days absent G2 → G1-G2 outcomes	−.021 *** (.006)	.001 (.006)	−.035 *** (.010)	−.018 * (.009)	−.054 *** (.010)	.086 *** (.012)	.027** (.010)
Days absent G3 → G2-G3 outcomes	−.023 *** (.005)	−.005 (.006)	−.028 * (.012)	−.024 * (.010)	−.060 *** (.010)	.079 *** (.011)	.032*** (.009)
Days absent G4 → G3-G4 outcomes	−.017 *** (.006)	−.020 ** (.007)	.002 (.011)	−.016 (.009)	−.077*** (.010)	.089 *** (.012)	.049 *** (.009)
Days absent G5 → G4-G5 outcomes	−.027 *** (.006)	−.024 *** (.006)	−.015 (.011)	−.021 * (.009)	−.064 *** (.011)	.109 *** (.013)	.032 *** (.010)

Note. All estimates are weighted and account for the complex sampling design. Coefficient in bold were statistically significant with a Benjamini false discovery adjustment. Models include a full set of controls. All continuous predictors and outcomes have been standardized to have a mean of 0 and standard deviation of 1 and, thus, coefficients can be interpreted as effect sizes. Estimates in brackets correspond to standard errors.

^***p < .001

^**p < .01.

^*p < .05.

Although the aforementioned associations may appear small at first past, it is important to benchmark these associations against other indicators. For example, the links between maternal education (i.e., a college degree vs less than a college degree) and student’s math achievement across time in the ECLS-K: 2011 yielded an effect size of approximately .075. Thus, the links between absenteeism (average d = .023) and students’ math achievement was approximately 30% of the maternal education gap in children’s math skills (.023/.075). When taken as a whole, these results suggest that students who were more frequently absent in any year of elementary school consistently performed less well across a variety of key educational benchmarks within the same year of school.

To determine whether absenteeism was more or less strongly associated with student outcomes during the early versus later years of elementary school, we estimated an omnibus test that constrained all coefficients, and as needed, we followed these models with a series of post-hoc coefficient comparisons. Results from our omnibus test revealed that there was no evidence of variation in the outcomes of absenteeism across grade-levels for students’ math (Scaled χ2 = 2.08, df = 5, p = .84), cognitive flexibility (Scaled χ2 = 11.03, df = 5, p = .051), working memory (Scaled χ2 = 2.71, df = 5, p = .74), and internalizing behavior (Scaled χ2 = 5.88, df = 5, p = .32). In contrast, there was some evidence of variation for students’ language and literacy skills (Scaled χ2 = 26.15, df = 5, p < .001), social skills (Scaled χ2 = 18.34, df = 5, p < .001), and externalizing behavior (Scaled χ2 = 23.45, df = 5, p < .001). When comparing coefficients within these three dimensions, we found that 17 of the 45 pairwise comparisons were statistically significant. In particular, absenteeism was found to be most strongly linked with students’ language and literacy skills in first grade and the smallest associations emerged in second grade, whereas for students’ social skills and externalizing behavior, the associations of absenteeism to these outcomes were smallest in kindergarten. In the main, however, there was far more consistency in the magnitude of associations between absenteeism and outcomes than variability across grade levels.

The cumulative links between absenteeism and students’ learning

All analyses presented thus far considered whether absenteeism in a given school year was linked with students’ performance in the same year. Our next set of analyses considered to what extent absenteeism across grade-levels predict students’ educational success and well-being. When looking at the results from these models, we find that 31 of the 35 associations estimated were statistically significant (see Table 4) and these models also demonstrated good model fit (CFIs > .90 and RMSEAs < .08). More specifically, students who were more frequently absent over time demonstrated less optimal outcomes across their educational careers. For example, students who were more frequently absent from school between kindergarten and first grade performed less well on five of the seven outcomes of interest at the end of first grade, with effect sizes of d = .05-.12. In contrast, students who were more frequently absent between kindergarten and fifth grade performed less well on all outcome measures at the end of fifth grade, with effect sizes ranging from d = .04-.16. And when examining whether the cumulative associations reported above are reflected in the outcome indicators from year to year, we find that cumulative absenteeism is consistently associated with year to year changes in outcomes across all domains but externalizing behavior. In total, 26 of the 35 associations estimated were statistically significant (see Table 5).

Table 4.

Results from path models examining the cumulative associations between absenteeism and student outcomes.

	Student outcome
Predictor	Math	Language and literacy	Cognitive flexibility	Working memory	Social skills	Internalizing behavior	Externalizing behavior
Days absent K-G1 → Fall of K-G1 outcomes	−.048 *** (.008)	−.067 *** (.009)	−.001 (.011)	−.048 *** (.011)	−.055 *** (.011)	.115 *** (.012)	.000 (.011)
Days absent K-G2 → Fall of K-G2 outcomes	−.067 *** (.009)	−.056 *** (.010)	−.056 *** (.012)	−.051 *** (.012)	−.065 *** (.012)	.105 *** (.013)	.017 (.011)
Days absent K-G3 → Fall of K-G3 outcomes	−.082*** (.010)	−.043 *** (.011)	−.063 *** (.015)	−.061 *** (.013)	−.064 *** (.013)	.103 *** (.014)	.014 (.012)
Days absent K-G4 → Fall of K-G4 outcomes	−.088 *** (.012)	−.048 *** (.012)	−.032 * (.015)	−.054 *** (.014)	−.100 *** (.013)	.128 *** (.016)	.047 *** (.014)
Days absent K-G5 → Fall of K-G5 outcomes	−.107 *** (.012)	−.067 *** (.014)	−.071 *** (.016)	−.058*** (.013)	−.101 *** (.015)	.156 *** (.016)	.039 ** (.014)

Note. All estimates are weighted and account for the complex sampling design. Coefficient in bold were statistically significant with a Benjamini false discovery adjustment. Models include a full set of controls. All continuous predictors and outcomes have been standardized to have a mean of 0 and standard deviation of 1 and, thus, coefficients can be interpreted as effect sizes. Estimates in brackets correspond to standard errors.

^***p < .001

^**p < .01.

^*p < .05.

Table 5.

Results from path models examining the cumulative associations between absenteeism and year to year change in student outcomes.

	Student outcome
Predictor	Math	Language and literacy	Cognitive flexibility	Working memory	Social skills	Internalizing behavior	Externalizing behavior
Days absent K-G1 → Spring of K-G1 outcomes	−.030 *** (.007)	−.053 *** (.008)	−.004 (.011)	−.041 *** (.010)	−.045 *** (.011)	.102 *** (.012)	.004 (.010)
Days absent K-G2 → G1-G2 outcomes	−.032 *** (.006)	−.004 (.006)	−.059 *** (.012)	−.029 ** (.011)	−.046 *** (.011)	.078 *** (.013)	.014 (.011)
Days absent K-G3 → G2-G3 outcomes	−.025*** (.007)	−.001 (.007)	−.035 ** (.013)	−.036 *** (.011)	−.045 *** (.011)	.073 *** (.013)	.006 (.010)
Days absent K-G4 → G3-G4 outcomes	−.023 *** (.007)	−.023 ** (.008)	−.004 (.013)	−.022 (.011)	−.076 *** (.012)	.092 *** (.015)	.039 ** (.013)
Days absent K-G5 → G4-G5 outcomes	−.037 *** (.007)	−.035 *** (.008)	−.054 *** (.013)	−.028 ** (.010)	−.059 *** (.013)	.118 *** (.015)	.017 (.012)

Note. All estimates are weighted and account for the complex sampling design. Coefficient in bold were statistically significant with a Benjamini false discovery adjustment. Models include a full set of controls. All continuous predictors and outcomes have been standardized to have a mean of 0 and standard deviation of 1 and, thus, coefficients can be interpreted as effect sizes. Estimates in brackets correspond to standard errors.

^***p < .001

^**p < .01.

^*p < .05.

When taken together, these results suggest that (a) for a number of outcomes, the cumulative outcomes of absenteeism grew larger over time and this was also reflected in year to year learning and (b) the cumulative associations were almost always larger than the grade-level links between absenteeism and student outcomes. Accordingly, these results imply that the associations of absenteeism to our outcomes accumulate year after year and, therefore, simply focusing on end-of-year outcomes would mask the longer-term consequences of missing school.

With that said, the above analyses and effect sizes do not address the practical implications of absenteeism as measured cumulatively between kindergarten and fifth grade. To facilitate this interpretation, we assess the value of each school day missed as a function of daily student learning. We were able to do so for three of our outcome measures that were based on a growth framework (i.e., math, language and literacy, working memory). For these three dimensions we found that across the five-and-a-half year window between the fall assessment window in kindergarten and the spring assessment in fifth grade, children gained approximately 84 points in math, 82 points in language and literacy, and 70 points in working memory, which correspond to daily gains of approximately .035 to .041 points. When benchmarking our findings against the outcomes of absenteeism, we find that each day of school missed was the equivalent of approximately 2.50 days lost of math learning, 1.50 days of lost language and literacy learning, and 2.00 days of lost working memory learning.

Variability in the outcomes of absenteeism

Our analyses thus far have considered the mean level associations between absenteeism and student learning, but not whether the outcomes of absenteeism vary for key subgroups of children. Accordingly, our next set of analyses considered whether absenteeism was more or less strongly associated with outcomes for different groups of students. Of the 784 interactions estimated, roughly 10% were statistically significant, which is only a little more than chance. More specifically, among the race/ethnicity interactions, 11% of the Asian, 11% of the Hispanic, 7% of the other race, and 14% of the Black interactions were statistically significant (relative to White children). In terms of the remainder of the interactions, only 4% of the low-income, 8% of the home language, and 14% of the child sex interactions were statistically significant.

With that said, one of the more consistent patterns that did emerge was that the associations between absenteeism and children’s academic achievement and executive function skills were larger for Black children than White children. In these domains, roughly 25% of the interactions estimated were statistically significant as compared with 2% for the domains of socioemotional development. To illustrate this pattern of findings consider the interactions from the cumulative outcomes model, which revealed that the associations between absenteeism and children’s fifth grade math (interaction d = .10, p < .05), language and literacy (interaction d = .12, p < .01), working memory (interaction d = .11, p < .05), and cognitive flexibility (interaction d = .17, p < .01) were larger for Black children as compared with White children. Thus, the Black-White opportunity gap was further acerbated as a result of absenteeism (see Figure 1 for an illustration of the aforementioned interactions).

Figure 1.

An illustration of the conditional associations between absenteeism between kindergarten and fifth grade and students’ fifth grade academic achievement and executive function skills as a function race/ethnicity. Note. Low absences correspond to one standard deviation below the mean, whereas high absences correspond to one standard deviation above the mean.

The only other consistent pattern that emerged was that boys did less well in language and literacy than girls as a result of absenteeism (56% of the absenteeism * sex interactions estimated within this domain were statistically significant as compared with roughly 6% for the remainder of the outcome domains). For example, although there were no gender differences in children’s language and literacy skills in fifth grade for children who were infrequently absent across grades, boys who were more frequently absent did less well than girls who were also more frequently absent (interaction d = .08, p. < .01).

Supplemental analyses

Below, we discuss supplemental analyses that (a) consider student-reported outcomes from fifth grade that were not consistently available in the earlier years of school and (b) assess the robustness of the main “effects” reported above.

Absenteeism and student reported outcomes in fifth grade.

In addition to the kindergarten through fifth grade outcomes reported as part of our focal analyses, students also reported on their motivation and grit (11 items, α = .86; e.g., trying to improve oneself, even when it takes a long time to get there), school belonging (14 items, α = .90; e.g., closeness with teachers and classmates), experiences of bullying (4 items, α = .81; e.g., teasing, name calling) school-related stress (5 items, α = .71; e.g., worrying about tests and doing well in school), and social anxiety (3 items, α = .88; worrying about what others think) in fifth grade. Accordingly, to highlight the other potential outcomes of absenteeism, we estimate exploratory models that consider the links between absenteeism in fifth grade and these self-reported outcomes in addition to the cumulative outcomes of absenteeism between kindergarten and fifth grade, net of the covariates outlined above. Because there were no lagged dependent variables from kindergarten, these models look at differences in level of student report. Importantly, however, the findings reported below were not sensitive to the inclusion of children’s baseline skills in other domains (i.e., academic achievement, executive function, and social behavior).

Results from these analyses revealed that both absenteeism and cumulative absenteeism were associated with higher levels of self-reported victimization (fifth grade absenteeism d = .05, p < .001; cumulative absenteeism d = .04, p < .01), school-related stress (fifth grade absenteeism d = .06, p < .001; cumulative absenteeism d = .05, p < .001), and social anxiety (fifth grade absenteeism d = .04, p < .001; cumulative absenteeism d = .03, p < .05) along with lower levels of motivation and grit (fifth grade absenteeism d = .06, p < .001; cumulative absenteeism d = .07, p < .001) and school belonging (fifth grade absenteeism d = .08, p < .001; cumulative absenteeism d = .07, p < .001). But unlike our focal study outcomes, for these student self-reported benchmarks, the cumulative outcomes of absenteeism were not larger than the grade-specific associations. Moreover, there was no consistent evidence of heterogeneity in these student-reported outcomes of absenteeism as a function of their socio-economic status, race/ethnicity, home language, and sex.

What would it take to change an inference?

Despite our use of extensive controls to minimize the possibility of spurious associations, our data are not experimental. Thus, the estimated outcomes of absenteeism may be biased because of variables omitted from our analysis. To address these concerns, we take two approaches to quantify how much bias there must be to invalidate an inference, both within our sample and considering omitted variables.

The first approach is a sample replacement framework, which uses Rubin’s causal model to quantify bias within the context of non-random assignment by comparing the reported estimate with a threshold to represent how much bias there must be in our sample to switch the inference (for more details see: Frank et al., 2013). The more the estimates exceed the threshold, the more robust the inference with respect to that threshold. We chose statistical significance as a threshold for inference. Results from our replacement of cases analyses revealed that to invalidate the main “effects” reported above (at the aggregate level), 54% of the estimates would have to be due to bias; that is, to invalidate the inference 54% of cases would have to be replaced with cases for which there is an effect of zero. Note that this replacement must occur even after observed cases have been conditioned on our covariates. When looking within the specific domains of student outcomes, we see some variation in the robustness of reported findings (listed in order from least robust to most robust): Externalizing behavior (37%); working memory (38%), cognitive flexibility (42%), language and literacy (53%), math (58%), social skills (62%), and internalizing behavior (74%). Importantly, the above reported thresholds—even for our least robust associations—all exceeded the estimates documented in other observational studies (~32%; Frank et al., 2013).

The second technique is through impact threshold for confounding variables analyses (ITCV; Frank, 2000). ITCV measures the degree to which an unknown variable would have to be correlated with both the predictor and outcome variables to negate the observed associations (for more details see: Frank, 2000). Results from our ITCV analyses revealed that an omitted variable would have to correlate with both the predictor and outcome at roughly r = .15 (aggregated across predictors and outcomes) to wash out the significant associations of interest. When looking within the specific domains of student outcomes, we see some variation in the robustness of reported findings (listed in order from least robust to most robust): Externalizing behavior (average r =.10), working memory (average r =.10), cognitive flexibility (average r =.11), language and literacy (average r =.14), math (average r =.16), social skills (average r =.17), and internalizing behavior (average r =.22). Importantly, however, none of the covariates included in our analyses—which represent some of the most common factors associated with absenteeism—approached these thresholds. And, when looking at other potential confounds (e.g., teacher education, experience, receipt of professional development), conditional on our covariates, these other factors did not approach the aforementioned thresholds, suggesting that our findings are robust to these variables and likely other unmeasured variables. Thus, despite some variation across domains, these findings from the replacement of cases framework and the ITCV analyses suggest that the links between absenteeism and student outcomes were robust.

Discussion

Even though school absenteeism has been found to undermine students’ school success (Ansari & Pianta, 2019; Gottfried, 2009, 2011, Ehrlich et al., 2013; Chang & Romero, 2008; Morrissey et al., 2013; Ready, 2010), the extant literature has rarely considered the potential cumulative associations between school absences and student outcomes or variation across grade-levels and different subgroups of students. Accordingly, our knowledge regarding the potential outcomes of absenteeism has been generally limited to thinking about single-year outcomes on same-year academic test scores. To address these gaps in scientific knowledge, we leverage data from roughly 14,370 students who participated in the ECLS-K: 2011 to examine the grade-specific and cumulative outcomes of absenteeism for students’ school success across elementary school. In considering these possibilities, several key themes emerged.

To begin, in following children from kindergarten through fifth grade, results from the present investigation revealed that absenteeism mattered for students’ school success across multiple dimensions. Results suggest comparable patterns for different dimensions of school performance: Students who were more frequently absent consistently performed less well on end of year assessments of academic achievement, executive function, and socioemotional development. Exploratory analyses also revealed that students who were more frequently absent reported themselves as experiencing greater school-related stress, lower levels of motivation and grit, and lower levels of school belonging. This finding – that absenteeism was associated with numerous domains of student success and well-being– certainly has implications for research and policy. As mentioned in the introduction of this study, recent research in absenteeism has predominantly focused on testing and achievement outcomes. There are certainly reasons why a focus on the attendance-achievement relation has been the dominant one, including federal accountability policies like No Child Left Behind as well as the general ‘academization’ of early grades (Bassok et al., 2016). That said, our study highlights the importance of looking beyond achievement when it comes to examining the outcomes of missing school. A portrait of decline in critical outcomes, including executive function, socioemotional development, and students’ self-reported motivation, stress, and engagement has been mostly obscured. Yet, policies around reducing absenteeism and supporting school attendance cannot be properly designed until we have full knowledge of the extent to which student outcomes are affected.

The second key theme that emerged from our work was that, contrary to parental beliefs (e.g., Ehrlich et al., 2013) or societal beliefs (Gottfried & Hutt, 2019), the magnitude of the association between absenteeism in the earliest years of school was equally as large as absenteeism in the later years of school. The main exception to this general pattern of findings was for children’s social skills and externalizing behavior. In these instances, absenteeism in kindergarten mattered less than absenteeism in the older grades. In general, the fact that patterns are consistent across the elementary school span corresponds to the most recent policy dialogue around student absenteeism (Gottfried & Hutt, 2019) – that though students age, there remain significant issues regarding absenteeism across all years of K-12 education, regardless of grade level. Our work would support this claim. Yet, when policies are designed to address absenteeism, they often focus on older students. This approach certainly has value, yet our study here suggests that a too-narrowed focus on older grades obscures attendance issues at earlier grades. Hence, the first and second conclusions of the current study suggest that the prior portrait of elementary school absenteeism was only partial – missing was an analysis of both multiple grades as well as numerous outcomes.

The third theme that emerged from this investigation was that the cumulative outcomes of absenteeism were larger than the timing specific associations. These findings align with Ansari and Pianta (2019) who found that absenteeism between kindergarten and eighth grade had cumulative consequences for student’s school performance in ninth grade. Yet the work here is far more nuanced. Rather than looking strictly at outcomes at one point in time, the work here looked across all grades in elementary school. The fact that there were cumulative associations across all of elementary school from absenteeism suggests the importance of addressing absences in each year of school, as the “effects” of absenteeism grow with continued habit. The implication of such a finding is consistent with what has been discussed thus far – that a well-rounded portrait of the outcomes of absenteeism will help to guide policy that addresses this issue during all years of schooling, rather than focusing on specific years, time-points, or outcomes. Each year matters for learning in that year, but also for future years. And when quantifying these lost opportunities, we find that each day missed for students’ academic achievement and executive function corresponded to approximately 1.50–2.50 days of lost learning. When taken together, the trajectory of missing school is of concern, and addressing it requires a multi-grade, multi-faceted perspective.

Finally, the results of the present investigation revealed that the links between absenteeism and student outcomes were far more similar than different for various groups of children across the elementary school years. These findings are in line with the work of Entwisle and colleagues (2001) who argue that, given equal exposure, all students benefit roughly equally from school. The one major exception, however, was for Black versus White students. In this case, our results revealed that absenteeism acerbated the White-Black opportunity gap in key domains of academic achievement and executive function. When interpreting these findings it is important to acknowledge that prior studies have shown that Black students in the United States start kindergarten about one half of a standard deviation behind their White peers on academic benchmarks (Hanushek & Rivkin, 2006; Rouse et al., 2005) and that these disparities continue to grow by about a tenth of a standard deviation a year (Fryer & Levitt, 2004). The findings of the present study align with the extant literature that has highlighted racial disparities in student learning (Burchinal et al., 2011), but advance our understanding by suggesting that absenteeism may be one potential culprit for the growing disparities across time between Black and White students. Accordingly, developing absenteeism interventions for Black students may represent one key way to minimize the White-Black opportunity gap.

Despite these contributions to the extant literature and policy conversation, the results of the present investigation should be interpreted in light of some key limitations. First, even though our sample followed a nationally representative sample of students over the course of the elementary school years, and we adjusted for a range of factors shown to be associated with absenteeism and students’ school performance (including lagged dependent variables), the ECLS-K: 2011 data are correlational in nature. As such, the results reported herein should be interpreted with caution as they may be affected by unmeasured factors. And even though we estimated how robust are findings were to omitted variables and found that our findings would not be altered by factors such as teacher characteristics, there were some aspects of classrooms that were missing from the ECLS-K, such as classroom quality, that we could not consider. With that said, work by Ansari and Gottfried (2020), suggests that the links between classroom quality and student absenteeism is generally small (d = .05) and so are the links between classroom quality and student learning (d = .05; Keys et al., 2013); therefore, even if such data were available, they are unlikely to reach the thresholds necessary to negate our findings. But even though absenteeism itself cannot be ethically randomized, future studies may consider the use of interventions that reduce absenteeism and then assess whether the effects on students’ outcomes match what is predicted by correlational findings, such as those reported in the present investigation (e.g., Cook et al., 2017).

Second, the ECLS-K did not provide information as to why children were absent. This limitation is somewhat tempered by the fact recent empirical evidence which highlights that all student absenteeism matters (Gottfried, 2009; Gershenson et al., 2017). Nonetheless, future studies should continue to consider the differential outcomes of both excused and unexcused absences. Third, although assessments of students’ academic achievement and executive function were based on direct assessments, children’s socio-emotional skills were reported on by teachers and, consequently, our outcome and predictor had shared-method variance. To our knowledge, however, there are not many direct assessments of children’s socio-emotional skills available that can be used in large-scale and nationally representative studies.

Fourth, our study provided an important snapshot of absenteeism between kindergarten and fifth grade, but what is missing is an understanding of absenteeism in early childhood education and how absenteeism across these years (and subsequent years) is linked to longer-term benchmarks of learning and well-being as students move through middle school, high school, and beyond. And even though the present study illustrated the links between absenteeism across grade-levels and outcomes, we did not consider the processes that could be related to absences as consequences and/or antecedents. Such efforts are necessary as they can begin to shed light on the developmental processes that absences may set in motion and potential points of intervention and prevention.

Finally, one may wonder whether our findings might be expected to generalize to adolescents. At present, it is unknown because absenteeism at various points in schooling may represent different processes that affect outcomes in different ways. For example, when compared with elementary schools, middle schools are larger and more bureaucratic, and tend to emphasize social comparison and competition through academic tracking, curricular differentiation, and structures for rewarding and recognizing academic success (e.g. Oakes, 2005; Wang & Degol, 2016). Secondary school teachers also tend to be less emotionally supportive (Roeser et al., 2006) and academic work is often incommensurate with adolescents’ skills and experiences (Juvonen et al., 2004). Accordingly, future studies should carefully consider whether the findings reported herein extend to adolescence and beyond.

With these limitations and future directions in mind, the present study used nationally representative data to report on the outcomes of student absenteeism. When taken together, the current investigation moved beyond studying the end-of-year outcomes of absenteeism (e.g., Connolly & Olson, 2012; Ehrlich et al., 2013; Gershenson et al., 2017; Miller & Johnson, 2016; Morrissey et al., 2013; Ready, 2010) by highlighting the grade-specific and cumulative consequences of missing school for students across the elementary school years. In doing so, our findings underscore the importance of viewing absenteeism from a longitudinal perspective.