Classrooms with high rates of absenteeism and individual success: Exploring students’ achievement, executive function, and socio-behavioral outcomes

Abstract

Research and policy dialogue surrounding absenteeism has predominately focused on the school when it comes to reducing student absences, with little focus on the classroom. Further, there has also been minimal attention paid to effects of absenteeism beyond achievement outcomes. To address both, we focused on the classroom and asked whether classrooms with typically higher rates of absenteeism were linked to students’ individual achievement, executive function, and social skills. We used a nationally representative dataset of children who started in kindergarten in 2010–2011 (N = 18,170) – when absenteeism is at its highest point not seen again until adolescence. Using school and student fixed effects, our findings revealed that as the percent of absent classmates increases, individual student performance worsens consistently across achievement and executive function domains. Evidence for links between classmate absenteeism and student performance in socio-behavioral domains was less conclusive. Finally, the findings were unique to different student groups.

Introduction

According to national estimates (e.g., Chang, Bauer, & Byrnes, 2018), approximately 15% of students in the United States miss enough school to be labled as chronically absent (i.e., around 10% of the school year). Contrary to common belief, absenteeism is not strictly an issue for middle and high school students; in fact, children miss a disproportionate amount of school in the earliest grades (Balfanz & Byrnes, 2012; Gottfried, 2019). Within these early years of schooling, evidence suggests that children in kindergarten and first grade miss more school than at any other point in elementary school (Balfanz & Byrnes, 2012). In other words, students are absent for excessive days of school throughout the totality of the K-12 educational pipeline, not just in the teenage years.

These high rates of absenteeism are problematic because students are missing numerous opportunities to learn and to develop from both teachers and classmates (Gottfried & Kirksey, 2017). The evidence of missing these opportunities is clear: Students who miss more days of school demonstrate lower academic performance and feel more alienated (Easton & Engelhard, 1982; Gershenson, Jacknowitz, & Brannegan, 2017; Goodman, 2014; Gottfried, 2009, 2010, 2011, 2014; Moonie, Sterling, Figgs, & Castro, 2008; Roby, 2004; Ansari & Gottfried, 2018). Students who are absent also generally are reported to have more distress in their lives (Finning et al., 2019; Ansari & Gottfried, 2018).

A common misperception among parents, however, is that the opportunity costs from missing school are minimal for young children (Gottfried, 2019). In contrast, research has documented both short- and long-term educational consequences of young students engaging in such behavior with effect sizes as large as those found in studies of older students (Chang & Romero, 2008; Connolly & Olson, 2012; Ehrlich, Gwynne, & Allensworth, 2018). Beyond the aformentioned ramifications for individual students, student absenteeism also affects school budgets. For example, over a recent 3-year period, California’s schools did not receive $4.5 billion that they would have otherwise received from the state had more students shown up to school (California Attorney General & Press, 2015).

Given the sheer numbers of students who are absent – as well as the consequences – there has been an amplified policy interest on student absenteeism. As a key example, via the federal Every Student Succeeds Act, the majority of states have now chosen to hold its schools accountable for student absenteeism (Jordan and Miller, 2017) – a first time in U.S. schooling history that both federal and state governments have aligned to improve actions about this issue (Hutt, 2018). Yet, holding schools accountable for reducing student absenteeism relies on a key assumption that schools are indeed places where absenteeism can be reduced (Gottfried, 2019). To test this assumption, over recent years there has been a surge of studies examining which schoolwide programs might be linked to absenteeism reduction, such as school-to-home mailers (Robinson, Lee, Dearing, & Rogers, 2018; Rogers & Feller, 2017, 2018), school-to-home texting (Page & Smythe-Leistico, 2019), full-day kindergarten programs (Gottfried, 2017), mentoring (Childs & Grooms, 2018), and access to transportation (Gottfried & Plasman, 2018).

Though the number of absence-reduction studies is growing, the focus of these analyses remains very much at the school level. The question as to whether schoolwide context has an effect on individual-level absenteeism is important, as it provides points of intervention for school administrators. But largely missing from this conversation is a focus on children’s most proximal experiences, including the classroom setting. This omission is problematic because elementary school students spend the majority (if not all) of the school day in classroom. In this vein, to date, little work has documented how absenteeism in the context of the classroom might be linked to students’ learning and development. The 1 exception is (Gottfried, 2011) who found that absent classmates negatively affected individual-level state test scores for children in the latter elementary school years. But if we support that absenteeism is problematic because it inhibits opportunities to learn and develop (Gottfried, 2017), then students who miss school not only affect their own academic, executive function, and socio-behavioral development but also might lower the opportunities for others from benefitting across these domains as well. In this study, academic outcomes are defined as reading and math achievement; executive function outcomes are defined as working memory and cognitive flexibility; and socio-behavioral outcomes are defined as problem behaviors and social skills.

Framework

The conceptualization of why classrooms with high rates of absenteeism might exert an influence on academic, executive function, and socio-behavioral outcomes begins first by thinking about the individual student. The negative individual effects are then considered in the context of classroom absenteeism.

Academic effects:

At the individual level, students who miss school more frequently tend to have weaker academic outcomes (Gershenson, Jacknowitz, & Brannegan, 2017; Goodman, 2014; Gottfried, 2009, 2010). One key reason behind these associations is that when students miss school, they receive fewer hours of instruction and hence learning declines (Gottfried, 2011). Yet, the academic performance of classmates are known to affect the academic outcomes of other classmates (Hanushek, Kain, Markman, & Rivkin, 2003; Hoxby, 2000). One reason these associations might surface in the context of absenteeism is that when absent students return back to the classroom, teachers often must often reallocate regular instructional time to remediate these students (Chen & Stevenson, 1995; Connell, Spencer, & Aber, 1994; Finn, 1993; Monk & Ibrahim, 1984). But in doing so, academic progress slows for the entire class (Gottfried, 2011). With higher rates of classroom absenteeism, this implies higher rates of absent students returning to the classroom throughout the year. Hence, remediation could potentially increase, and larger portions of instruction are dedicated to ensuring that those who were absent are caught up with classroom material, and hence learning – and achievement – declines for all students (Gottfried, 2019).

Executive function:

Although rarely examined as an outcome of absenteeism, there is new evidence to suggest that when students miss more days of school, they also demonstrate weaker executive function skills, which include their working memory, inhibitory control, and the ability to focus attention (Fuhs, Nesbitt, & Jackson, 2018). It has been hypothesized that these associations emerge because absenteeism introduces unpredictability in students’ educational experiences and it provides students with fewer opportunities to develop executive function skills in the context of high-quality environments (Fuhs et al., 2018). Little is known about classroom effects on executive function with regards to absent students. We speculate that with higher rates of classroom absenteeism, there is greater unpredictability for all students, hence hurting individual executive function. Furthermore, to the extent that, on top of unpredictability arising from high rates of absenteeism, teachers: (1) have to reallocate regular instructional time upon the return of absent students to school (Chen & Stevenson, 1995; Connell et al., 1994; Finn, 1993; Monk & Ibrahim, 1984) and (2) provide lower quality and engaging classrooms because of instructional repetition, then this is likely to disrupt the executive function skills of not just individual students, but the classroom as a whole.

Socio-behavioral development:

Research has also shown that when kindergartners are absent, they demonstrate less optimal social skills and greater levels of problem behaviors (Gottfried, 2014). The above associations are likely attributed to the fact being away from school results in feelings of alienation from classmates, teachers, and schools (Ekstrom, Goertz, Pollack, & Rock, 1986; Newmann, 1981). When a greater proportion of the classroom is absent, lower social skills and increased problem behaviors emerge for a greater proportion of the classroom. Hence, upon returning back to school, absent students might have a larger number of negative interactions with classmates and teachers (Finn, 1989). Consequently, negative interactions in the classroom may spur negative behaviors between classmates (Ribeiro & Zachrisson, 2017), even those who were not absent. And as the percent of classroom absenteeism rises, so might these disruptions for all students, hence slowing classroom progress (Bonesronning, 2008; Lazear, 2001) including behavioral development (Gottfried, 2019).

Research Questions

We asked the following research questions:

1. Do young elementary school children in classrooms with higher absenteeism rates demonstrate lower academic, executive function, and socio-behavioral outcomes?

2. Do these patterns differ for children who are considered most resilient to absenteeism, namely those with fewer absences, less school mobility, and/or being taught by more experienced teachers?

Our first aim is to examine the link between classroom absenteeism and individual outcomes. Given prior literature on the negative effects of absenteeism and possible spillover effects of classroom absenteeism, we hypothesize that as absenteeism increases in the classroom, individual outcomes will be lower. As mentioned above, children in kindergarten and first grade miss more school than at any other time in elementary school – levels that do not reach such peaks again until adolescence. Moreover, little work has examined the influence of classrooms with high rates of absenteeism even though we expect the classroom to be a setting where children both learn and develop from their teachers and classmates. If absent students hurt their own outcomes across academic, executive function, and socio-behavioral domains as a result of missing school, a next step would be to determine how being in a classroom with high rates of absenteeism plays out for other children in that same setting. In doing so, the above would provide for a more robust portrait of absenteeism, thereby motivating policymakers to extend their perspectives beyond school-level action-points and into the dynamics of the classroom.

Our second research aim is to examine group differences. The different groups we explored were designated based on groups that might be most resilient to absenteeism, with determinations based on previous literature: Students with minimal absences (Gottfried, 2014), less mobility (Welsh, 2018), and having more experienced teachers with 5 or more years of experience (Gershenson, 2016; Ingersoll, 2012; Ladd & Sorensen, 2017). Given the importance of resiliency to absenteeism, we hypothesize that students from more resilient groups will not experience as low of outcomes in classrooms with higher rates of absenteeism. Finally, though not a research question in of itself, a third aim of this study is to test the robustness of our findings. We do so through a series of sensitivity tests that are presented in the results section.

Method

Participants

We analyzed the Early Childhood Longitudinal Study – Kindergarten Class of 2010–2011 (ECLS-K:2011), which was created by the National Center for Education Statistics (NCES) at the U.S. Department of Education. Both nationally representative and longitudinal, the ECLS-K:2011 followed a cohort of kindergartners beginning in the 2010–2011 school year and through elementary school. Direct assessments were administered to the sample of children on a range of academic, executive function, and socio-behavioral outcomes. Parents, teachers, and school administrators received surveys. As mentioned, the sampling design is nationally representative, thus reflecting kindergarten children in public and private schools across the country as well as from different racial/ethnic backgrounds and different socioeconomic levels. Approximately 87% of the sample in kindergarten attended public schools.

In this study, we focused on children when they were in kindergarten and first grade, given: (1) the prominence of absenteeism in these 2 years, (2) the consistency of child-level assessments (some of which change beginning in second grade), as well as (3) sample attrition (which increases over time). To arrive at a final sample size of 18,170 kindergarteners, we employed chained multiple imputation (Royston, 2004). Using sample weights, 20 datasets were imputed to resemble the original observed distribution of all variables. In first grade, this sample was approximately N = 14,900 – slightly lower due to attrition. In sum, we had approximately 33,070 kindergarten and first grade observations. Note that all sample sizes have been rounded to the nearest tens, as required to use the restricted version of the ECLS-K:2011 data.

Measures

Academic achievement:

Reading and math achievement scores were based on assessments developed by NCES that were administered to children (α’s = 0.92–0.95). The math assessment included up to 96 questions about number sense, properties, and operations, measurement, geometry and spatial sense, data analysis, and patterns, algebra, and functions. The reading assessment included up to 100 questions about print familiarity, letter recognition, and recognition of common words. Higher scores on the math and reading assessments are indicative of more optimal performance.

Executive function:

Two key dimensions of executive function assessed in the ECLS-K:2011 were working memory and cognitive flexibility (Tourangeau et al., 2015). Working memory, which assesses the ability to store and manage information during complex cognitive tasks, was measured with the Numbers Reversed (NR) subtest of the Woodcock-Johnson III (α = 0.87; Woodcock, McGrew, & Mather, 2001). A child’s score on the NR assessment is based on reciting numbers in the reverse order. Cognitive flexibility assesses a child’s ability to switch between thinking about different concepts and was measured with the Dimensional Change Card Sort (α’s = 0.90–0.94; DCCS; (Zelazo, 2006)). The DCCS score reflects a child’s performance over a series of tests associated with accuracy on card sorting tasks (i.e., by shape, color, and border games). Higher scores on the working memory and cognitive flexibility are indicative of better executive function skills.

Socio-behavioral outcomes:

Teachers were asked to rate children’s problem behavior and social skills through a series of questions (not available to the public). NCES created 2 problem behavior scales (α’s = 0.78–0.89) and 3 social skills scales (α’s = 0.81–0.91) based on these individual items. Called the Teacher Social Rating Scales, they all exist on a 4-point Likert-metric, with higher scores indicating that the child engaged in those behaviors more frequently. The problem behaviors scales include externalizing and internalizing scales, with higher scores indicative of less optimal behavior. The externalizing behaviors scale included 5 questions about the frequency with which a child argues, fights, gets angry, acts impulsively, and disturbs ongoing activities. The internalizing behaviors scale included 4 questions about the extent that the child exhibits anxiety, loneliness, low self-esteem, and sadness. As mentioned above, there are 3 social skills scales, with higher scores indicative of more optimal behavior. A self-control scale measure included 4 questions about how well the child controls his/her temper, respects others’ property, accepts his/her peers’ ideas, and handles peer pressure. An approaches to learning scale measure included 6–7 questions about how well the child keeps his/her belongings organized, shows eagerness to learn new things, adapts to change, persists in completing tasks, pays attention, and follows classroom rules. An interpersonal skills scale included 5 questions about how well the child gets along with others, forms and maintains friendships, helps other children, shows sensitivity to the feelings of others, and expresses feelings, ideas, and opinions in positive ways.

Classroom rate of absenteeism:

In both kindergarten and first grade, teachers reported the number of children who were absent on a typical day. Teachers reported the absolute number, though we divided this by class size in order to obtain the classroom rate of absenteeism. This measure is self-reported by the teacher, and no administrative records with classroom absent tallies exists in the dataset. To determine whether there was systematic overcounting or undercounting that would bias our analyses, we regressed reported percent of absent students on the set of teacher characteristics (described in the next section). Comparing teachers within the same school, the characteristics were not statistically significant, and the coefficients were practically zero, limiting any practical significance in any case. Accordingly, it does not appear that there is systematic over/undercounting.

Control variables:

Table 1 presents all control variables utilized in this study. Measures of the majority of variables were available in both kindergarten and first grade, though it is noted here twice when this was not the case. The first section of the table presents child characteristics. These include gender, race/ethnicity, English Language Learner status, disability status, whether the child attended some form of center-based preschool prior to kindergarten, whether the kindergarten program was full vs part day, the hours of before/afterschool care in kindergarten, and the number of absences in the school year. The second section presents household characteristics. These include whether the child lives in a household of 2 partners, the number of siblings, whether the family chose their residence location due to the location of kindergarten (no equivalent; the question was only asked in the first grade survey only), whether the school was neighborhood-assigned (no equivalent; the question was only asked in the first grade survey), parental education and employment, and poverty status.

Table 1.

Descriptive statistics at kindergarten wave (N = 18,170).

Child characteristics	Mean	SD
Male	0.51	0.50
Latinx	0.25	0.44
Black	0.17	0.37
Asian	0.10	0.31
Health (1 highest rating, 5 lowest rating)	1.57	0.78
English language learner	0.16	0.37
Disability	0.20	0.40
Center-based care prior to kindergarten	0.53	0.50
Full-day kindergarten	0.83	0.38
Hours of care in kindergarten year	0.87	0.35
Absences	5.96	4.85
Household characteristics
Two-partner household	0.75	0.41
Number of siblings	1.51	1.12
Home location chosen for kindergarten location	0.34	0.47
Kindergarten school was assigned (not choice)	0.67	0.47
Mother has college degree or more	0.31	0.47
Father has college degree or more	0.28	0.47
Mother works full time	0.42	0.49
Father works full time	0.78	0.37
At or below poverty threshold	0.51	0.50
Classroom characteristics
Percent absent	0.05	0.05
Class size	20.37	5.07
Percent girls	0.49	0.10
Percent Black	0.15	0.25
Percent Latinx	0.22	0.29
Percent Asian	0.05	0.12
Percent below grade level: Reading	0.18	0.13
Percent below grade level: Math	0.14	0.12
Percent with disability	0.09	0.11
Percent English Language Learners	0.08	0.05
Teacher Black	0.07	0.25
Teacher Latinx	0.10	0.30
Teacher Asian	0.02	0.15
Years teaching	14.52	9.77
Teacher holds graduate degree	0.45	0.50

The third section of the table presents classroom and teacher variables. Based on the teacher’s report in the spring of both kindergarten and first grade, it was possible to construct the percentages of their classroom students by gender, race/ethnicity, academic ability, disability, and English Language learner status. Finally, the table includes teacher race/ethnicity, years of experience, and education.

We explored several of these control variables separately in the analyses to follow. Namely, we explored those that pertained to student resilience when in a classroom with a higher rate of absenteeism. These included: individual absenteeism, not changing schools, and teacher experience.

Procedures

Baseline model:

To explore whether students have different achievement, executive function, and socio-behavioral outcomes based on classroom absenteeism, we began with the following baseline model:

$${\text{Y}}_{\text{ijst}}={\beta }_0+{\beta }_1{\text{ABS}}_{\text{ik}}+{\beta }_2{\text{I}}_{\text{it}}+{\beta }_3{\text{H}}_{\text{it}}+{\beta }_4{\text{C}}_{\text{jst}}+{\epsilon }_{\text{ijst}}$$ (1)

where Y is one of the outcomes described above for student i in classroom j in school s in year t. In this model, ABS represents the percent of absenteeism as reported by the teacher for a typical day, I represents individual child characteristics, H includes household characteristics, and C represents classroom and teacher characteristics described above. The error term is clustered for students i in the same classroom-year such that we are accounting for the fact that students are observed in classrooms in each year and thus share unobserved yet similar experiences. Note that all models include a lagged measure of the outcome.

School fixed effects:

The experiences of children differ across schools for both observed and unobserved reasons, and the model above does not take this into account. For instance, in schools’ responses to high absenteeism rates, some administrators might work to ensure that students are less frequently absent. Yet, these same administrators might be making other investments to improve students’ outcomes. Students who enter these schools, then, might have lower chances of being absent, of therefore having a higher classroom rate of absenteeism on a typical day and might also have better outcomes. Therefore, the coefficient on the typical percent absent might reflect these unobserved school-specific efforts and comparing schools to each other across the sample might not yield unbiased results. Or, alternatively, it might be the case that parents who are highly involved in their child’s schooling might ensure that their children attend more frequently; and parents may in aggregate be sending their children to schools where the value of school attendance is also made a priority. Children in these schools would likely be in classrooms with lower absenteeism, hence biasing the estimate of β₁. To address this, a second model is employed:

$${\text{Y}}_{\text{ijst}}={\beta }_0+{\beta }_1{\text{ABS}}_{\text{ik}}+{\beta }_2{\text{I}}_{\text{it}}+{\beta }_3{\text{H}}_{\text{it}}+{\beta }_4{\text{C}}_{\text{jst}}+{\delta }_{\text{s}}+{\epsilon }_{\text{ijst}}$$ (2)

where ẟ_s represents school fixed effects. Including indicators for school in the regression model implies that only classrooms within schools are compared to each other. In other words, by including school fixed effects, we are controlling for school-to-school differences, such as policies and practices as well as aggregate student and family characteristics (i.e., if certain types of families attend specific schools).

Student fixed effects:

A concern not addressed thus far is that neither our baseline nor school fixed effects models take into account unobserved individual-level heterogeneity or within-school (i.e., between classroom) sorting. For instance, a principal might sort together students he/she suspects might be high-attending that year in the same classroom, hence biasing downward previous estimates of β₁. And this underscores the potential for within-school sorting. As a test of this possibility, we regressed the percent absent on the individual child characteristics from Table 1 using a school fixed effects model. None of the observed student characteristics were statistically significant, though we suspect there may be other unobserved ways that principals might sort students with regards to expected absenteeism, like signs of engagement in the child or in the family. To address this possibility, a student fixed effects model is employed:

$${\text{Y}}_{\text{ijst}}={\beta }_0+{\beta }_1{\text{ABS}}_{\text{ik}}+{\beta }_2{\text{I}}_{\text{it}}+{\beta }_3{\text{H}}_{\text{it}}+{\beta }_4{\text{C}}_{\text{jst}}+{\delta }_{\text{i}}+{\epsilon }_{\text{ijst}}$$ (3)

In this model, ẟ_i represents indicators for student ID – that is, student fixed effects. This model relies on repeated observations of the same student, thereby exploiting within-student variation year-after-year in the typical percent absent that each student has in his/her classroom. Effectively, each student becomes his/her own comparison, and in doing so, only time varying covariates – like percent absent on a typical day – remain in the model. In other words, all observed and unobserved time invariant characteristics remain fixed, allowing the model to better isolate the effect of classroom absenteeism.

Results

Research question 1:Effects on achievement and socio-behavioral outcomes

Baseline models:

Table 2 presents the findings for students’ reading and math outcomes. The coefficient on typical percent absent is presented, with standard errors in parentheses. Following is the coefficient on a student’s own absences. In this table, all control variables are also presented. For the sake of clarity, however, they are not placed in the remainder of the tables, given the wide swath of outcomes and large set of control measures.

Table 2.

Effects of absenteeism on achievement (baseline model).

	Reading	Math
Absence measures
Percent absent	−26.13*** (4.37) −0.06	−26.64*** (3.71) −0.07
Individual absences	−0.47*** (0.03) −0.10	−0.42*** (0.03) −0.10
Child characteristics
Male	−2.01*** (0.21)	1.46*** (0.19)
Latinx	0.75* (0.35)	−0.54 (0.31)
Black	−0.41 (0.42)	−3.48*** (0.35)
Asian	1.14* (0.57)	−0.59 (0.49)
Health	−0.37* (0.17)	−0.33* (0.14)
English language learner	−4.39*** (0.39)	−3.35*** (0.34)
Disability	−7.39*** (0.31)	−7.03*** (0.30)
Center-based care prior to kindergarten	−0.03 (0.38)	−0.50 (0.31)
Full-day kindergarten	2.14*** (0.52)	1.97*** (0.46)
Hours of care in kindergarten year	0.85 (0.54)	0.02 (0.39)
Household characteristics
Two-partner household	1.21*** (0.29)	0.91*** (0.25)
Number of siblings	−0.52*** (0.10)	−0.02 (0.09)
Home location chosen for kindergarten location	0.36 (0.31)	0.34 (0.25)
Kindergarten school was assigned (not choice)	1.10** (0.38)	0.54 (0.32)
Mother has college degree or more	4.35*** (0.34)	3.82*** (0.27)
Father has college degree or more	3.01*** (0.37)	2.67*** (0.31)
Mother works full time	0.20 (0.26)	0.51* (0.24)
Father works full time	1.75*** (0.40)	1.74*** (0.38)
At or below poverty threshold	−2.11*** (0.31)	−2.09*** (0.27)
Classroom characteristics
Class size	0.07 (0.06)	0.08 (0.05)
Percent girls	2.29 (2.11)	0.06 (1.77)
Percent Black	−11.25*** (0.98)	−10.67*** (0.78)
Percent Latinx	−18.83*** (0.71)	−17.80*** (0.60)
Percent Asian	25.55*** (3.92)	22.04*** (3.36)
Percent below grade level: Reading	−4.82** (1.51)	−1.21 (1.31)
Percent below grade level: Math	3.33 (1.96)	0.98 (1.63)
Percent with disability	−4.55** (1.71)	−5.87*** (1.47)
Percent English Language Learners	−21.23*** (5.03)	−18.04*** (4.22)
Teacher Black	1.02 (0.66)	0.98 (0.55)
Teacher Latinx	0.80 (0.53)	1.31** (0.45)
Teacher Asian	−1.50 (1.34)	0.91 (1.21)
Years teaching	−0.02 (0.02)	−0.01 (0.02)
Teacher holds graduate degree	0.99* (0.40)	0.93** (0.34)
n	33,070	33,070

Standard errors adjusted for clustering in parentheses.

^*P < 0.05.

^**P < 0.01.

^***P < 0.001.

From these initial baseline models, the findings suggest when students are in classrooms with a higher typical percent absent, individual achievement tends to be lower. The above is true for both reading and math achievement scores. Importantly, they are statistically significant even after taking into account a student’s own individual absenteeism, which is displayed as a negative coefficient on testing outcomes, in addition to our child, household, and classroom covariates.

In the table, the coefficients on percent absent appear large, given the variable is measured as percent of the classroom ranging from 0–1; thus, a 1 unit change in the predictor corresponds to having no classroom absenteeism to 100% classroom absenteeism. To provide a more meaningful metric and ease interpretation of the focal associations of interest, we also provide the standardized coefficient in bold below the coefficient and standard error. What this metric reveals is that a 1 standard deviation increase in the percent absent on a typical day is associated with a 0.06 standard deviation decrease in reading scores and a 0.07 standard deviation decrease in math scores. It is important to note that: (1) these are individual-level associations to class-level measures, hence making it possible for there to be quite large aggregated class-level effects and (2) the associations between classroom-absenteeism and students’ academic outcomes were not too dissimilar from the links between student’s own absenteeism and their academic test scores, which was associated with a 0.10 standard deviation decrease in both reading and math scores.

Results for all outcomes are presented in the top panel of Table 3 based on a model that includes both students’ individual absenteeism and all covariates. What these results reveal is that, as the typical percent of absent classmates increases, children not only demonstrate lower reading and math test scores, but they also demonstrate lower executive function outcomes. As for the socio-behavioral outcomes, the baseline models suggest that as the typical percent absent increases, children display greater levels of internalizing and externalizing behavior problems as well as lower levels of self-control, with standardized betas of 0.02–0.03.

Table 3.

Effects of absenteeism on achievement, executive function, and socio-behavioral outcomes (all models).

	Achievement Outcomes		Executive function outcomes		Socio-behavioral outcomes
	Reading	Math	DCCS	W-ability	Externalizing	Internalizing	Interpersonal skills	Self-control	Approaches to learning
Panel A: Baseline
Percent Absent	−26.13*** (4.37) −0.06	−26.64*** (3.71) −0.07	−0.98** (0.35) −0.02	−20.19*** (4.18) −0.03	0.32*** (0.09) 0.03	0.15* (0.08) 0.02	−0.16 (0.09) −0.01	−0.31*** (0.09) −0.02	−0.01 (0.10) 0.00
Individual Absences	−0.47*** (0.03) −0.10	−0.42*** (0.03) −0.10	−0.02*** (0.00) −0.02	−0.39*** (0.04) −0.06	−0.00 (0.00) 0.00	0.01*** (0.00) 0.10	−0.00*** (0.00) −0.04	−0.00* (0.00) −0.02	−0.01*** (0.00) −0.10
n	33,070	33,070	33,070	33,070	33,070	33,070	33,070	33,070	33,070
Panel B: School Fixed Effects
Percent Absent	−22.86*** (4.34) −0.04	−25.09*** (3.87) −0.06	−1.05* (0.44) −0.02	−17.78*** (5.18) −0.03	0.37** (0.13) 0.03	0.19 (0.11) 0.02	−0.30 (0.18) −0.03	−0.39** (0.13) −0.03	−0.13 (0.14) −0.01
Individual Absences	−0.46*** (0.03) −0.10	−0.42*** (0.02) −0.10	−0.02*** (0.00) −0.02	−0.43*** (0.04) −0.06	−0.00 (0.00) 0.00	0.01*** (0.00) 0.08	−0.01*** (0.00) −0.04	−0.00** (0.00) −0.02	−0.01*** (0.00) −0.09
n	33,070	33,070	33,070	33,070	33,070	33,070	33,070	33,070	33,070
Panel C: Student Fixed Effects
Percent Absent	−38.28*** (7.39) −0.07	−39.47*** (6.11) −0.09	−1.77** (0.65) −0.02	−26.52*** (7.01) −0.04	0.18 (0.13) 0.01	0.12 (0.15) 0.01	−0.19 (0.16) −0.01	−0.28 (0.15) −0.01	−0.07 (0.15) 0.00
Individual Absences	−0.50*** (0.05) −0.12	−0.44*** (0.04) −0.11	−0.02*** (0.01) −0.04	−0.44*** (0.06) −0.07	−0.00 (0.00) 0.01	0.00*** (0.00) 0.03	−0.00* (0.00) −0.03	−0.00 (0.00) −0.02	−0.00*** (0.00) −0.04
n	33,070	33,070	33,070	33,070	33,070	33,070	33,070	33,070	33,070

Standard errors in parentheses.

Each column represents percent absent and days absent from a unique regression analogous to the final, full models in Table 2. Student fixed effects regressions only include time-varying control variables from Table 2.

^*P < 0.05.

^**P < 0.01.

^***P < 0.001.

As a point of contrast, we present individual student absences in the bottom portion of Table 3. The results reveal that students who have more absences not only have lower reading and math scores, but also demonstrate lower executive function skills, higher problem behaviors, and less optimal social skills, with standardized betas of 0.02–0.10. These findings correspond with the framework of our study – that the effect of an individual absence might map onto a larger issue in the classroom specific for that outcome (i.e., more problem behaviors at the individual level might translate into more aggregate classroom problem behaviors). As before, the effect sizes for the percent absent were, in general, similar in magnitude to that of individual students’ absences, with an absolute difference of approximately 0.04 standard deviation units. In sum, the baseline models indicate that not only does individual absenteeism matter for student outcomes, but in classrooms with a higher percent absent, students demonstrated lower academic and executive function skills and higher levels of problematic behaviors.

School and student fixed effects:

In Table 3, the school and student fixed effects models are also presented. Each coefficient pairing represents the typical percent of absent classmates and individual absences from a unique regression model, where the outcome is designated down the left column. Standard errors are in parentheses, and underneath are the standardized coefficients. For clarity, control measures for each regression are not provided, though they are available upon request.

The school fixed effects models generally correspond to the findings reported above. When comparing the typical percent absent between classrooms within the same school, children tend to demonstrate lower achievement and executive function scores, higher externalizing problems, and lower self-control. Unlike our earlier model, the coefficient for internalizing problems is no longer statistically significant. Individual absences also correspond to the baseline models that we reported previously. Importantly, however, the standardized betas are comparable across specifications, suggesting that the focal associations reported above are not simply attributed to school-level differences.

In the student fixed effects models, the typical percent absent predicts statistically-significant differences in children’s achievement and executive function outcomes. Thus, when children have a higher percent of absent classmates, they exhibit lower reading, math, and executive function skills. Importantly, the effect sizes across specifications vary by roughly 0.01 standard deviation units across outcomes, suggesting that our findings reported above are not simply attributed to unmeasured fixed characteristics of children or their families. With that said, no associations emerged between the percent absent for students’ socio-behavioral outcomes (problem behaviors and social skills) in our student fixed effects models.

Research question 2: Group differences by resilience

Table 4 presents the findings for different groups. Interactions are noted at the top column, and all coefficients are the typical percent absent derived from the student fixed effects models. In the first column, we explored whether students with few-to-no absences (i.e., 3 or fewer) might be more resilient to having a higher typical percent of absent classmates. Results from these analyses revealed that none of the interactions were statistically significant. Similar patterns emerged when we explored whether these associations varied for those students who did and did not switch schools between kindergarten and first grade. That is, the outcomes of classroom absenteeism do not vary for those children who did and did not change schools.

Table 4.

Exploring effects of absenteeism for key subgroups.

	Percent absent interacted with the following:
	Students with few absences	Students, no change in schools	Teachers with 5+ years of experience
Academic outcomes
Reading	−32.05 (20.20)	17.59 (18.00)	−30.98 (16.21)
Math	−32.72 (17.60)	10.39 (15.17)	−32.99 (15.37)
Executive function outcomes
DCCS	−1.60 (1.72)	2.35 (2.49)	−2.08 (1.12)
W−Ability	−26.33 (20.20)	−13.56 (21.82)	−23.56 (13.12)
Socio-Behavioral Outcomes Problem Behaviors
Externalizing	0.26 (0.38)	−0.11 (0.40)	0.13 (0.27)
Internalizing	−0.08 (0.38)	−0.08 (0.49)	0.08 (0.25)
Social Skills
Interpersonal skills	0.25 (0.46)	−0.22 (0.51)	−0.20 (0.30)
Self-control	0.16 (0.48)	−0.08 (0.54)	−0.32 (0.31)
Approaches to learning	0.12 (0.44)	−0.09 (0.52)	−0.15 (0.27)
n	33,070	33,070	33,070

Note:

^***P < 0.001,

^**P < 0.01,

^*P < 0.05.

Standard errors in parentheses.

Each grouping represents percent absent from a unique regression that includes all covariates from Table 2.

Student fixed effects regressions only include time-varying control variables from Table 2.

Finally, it has been shown that teachers with more experience have students with fewer absences (Gershenson, 2016; Ladd & Sorensen, 2017). Therefore, we explored whether students with more experienced teachers might have different outcomes. There is no evidence of this being the case. In sum, there is little evidence that there are large group differences across any of the samples in this table.

Sensitivity tests

We ran several tests to address several concerns. For instance, 1 issue is that individual students themselves might start school and influence classroom absenteeism, or that student development over the course of a year might affect classroom absenteeism. Ideally, 1 way to address this would be to use lagged individual absences to construct this year’s classroom absenteeism rates, though this is not possible in ECLS-K data. With that said, we test the potential influence of individual students on classroom absenteeism rates in 2 ways, presented in Table 5.

Table 5.

Sensitivity tests.

	Outcome: percent absent
	Kindergarten	First grade	Kindergarten	First grade
	(1)	(2)	(3)	(4)
Independent variables
Achievement Outcomes
Reading	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)
Math	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)
Executive Function Outcomes
DCCS	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)
W-Ability	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)	0.000 (0.000)
Socio-Behavioral Outcomes Problem Behaviors
Externalizing	0.002 (0.001)	0.000 (0.001)	0.000 (0.001)	0.001 (0.001)
Internalizing	0.000 (0.001)	0.000 (0.001)	0.000 (0.001)	0.000 (0.001)
Social Skills
Interpersonal skills	0.000 (0.002)	0.000 (0.001)	−0.001 (0.002)	−0.001 (0.001)
Self-control	0.000 (0.001)	0.000 (0.002)	0.000 (0.001)	0.000 (0.001)
Approaches to learning	0.002 (0.001)	0.000 (0.001)	0.001 (0.001)	−0.001 (0.001)
Included in the regressions?
Child characteristics	Yes	Yes	Yes	Yes
Household characteristics	Yes	Yes	Yes	Yes
Classroom characteristics	Yes	Yes	Yes	Yes
School fixed effects models	Yes	Yes	Yes	Yes
n	17,720	15,340	11,590	10,060

Note:

^***P < 0.001,

^**P < 0.01,

^*P < 0.05.

Standard errors in parentheses.

In the first 2 columns of this table, we regressed the typical percent absent on kindergarten fall entry measures of our outcomes (plus all other controls) – we did this separately for kindergarten and first grade. The assumption would be that if a fall entry measure (which was measured just at the start of the school year before the student interacted with the classroom) predicted classroom absenteeism, that might provide evidence that individual students might be affecting other students’ absences. The first 2 columns suggest null results, however, and approximate a value of 0 even to the third decimal point, limiting any practical significance. In other words, individual students do not tend to predict classroom-level absences. In the second 2 columns, we regressed typical percent absent by grade on gains of these measures (i.e., the difference between spring kindergarten and fall kindergarten) while also controlling for all other variables. Doing so tests whether a student’s development across the school year might be influencing classmate absenteeism. Again, the results are null and approximate a value of 0. Thus, there does not seem to be evidence that an individual student’s skills nor growth over the course of the year affects current (or future) classroom absence behavior.

In Table 6, we ran 4 additional tests of sensitivity. In the first panel, we reran the student fixed effects models from Table 3, but did not include individual absenteeism as a control variable. We estimated these models given the possibility that individual student absences serve as a mechanism for the links between high levels of classroom absenteeism and students’ individual outcomes. Re-running these models without individual student absenteeism reveals much the same pattern as before: Students in classrooms with higher absenteeism demonstrate less optimal outcomes. Importantly, the significant coefficients with and without individual student absenteeism included in the models were only 5%–15% different. Because the ECLS-K does not allow us to disentangle the time sequencing (i.e., classroom vs child absenteeism), we preserve our primary specification.

Table 6.

Additional sensitivity tests.

	Achievement Outcomes		Executive function outcomes		Socio-behavioral outcomes
	Reading	Math	DCCS	W-ability	Externalizing	Internalizing	Interpersonal skills	Self-control	Approaches to learning
Panel A: No Individual Absences
Percent Absent	−42.37*** (7.39)	−42.94*** (5.95)	−1.85** (0.59)	−31.21* (8.35)	0.21 (0.12)	0.13 (0.19)	−0.21 (0.15)	−0.31 (0.16)	−0.10 (0.14)
n	33,070	33,070	33,070	33,070	33,070	33,070	33,070	33,070	33,070
Panel B: No Teacher Characteristics
Percent Absent	−38.78*** (7.53)	−39.62*** (6.07)	−1.68** (0.59)	−28.29* (8.17)	0.23 (0.12)	0.10 (0.18)	−0.19 (0.15)	−0.31 (0.15)	−0.08 (0.14)
Individual Absences	−0.51*** (0.05)	−0.45*** (0.04)	−0.02*** (0.01)	−0.44*** (0.06)	−0.00 (0.00)	0.00** (0.00)	−0.00* (0.00)	−0.00 (0.00)	−0.00*** (0.00)
n	33,070	33,070	33,070	33,070	33,070	33,070	33,070	33,070	33,070
Panel C: Public School Students Only
Percent Absent	−33.97*** (7.39)	−37.78*** (6.17)	−1.78* (0.66)	−27.09** (7.69)	0.15 (0.13)	0.11 (0.19)	−0.18 (0.17)	−0.29 (0.16)	−0.12 (0.15)
Individual Absences	−0.47*** (0.05)	−0.43*** (0.04)	−0.02** (0.01)	−0.43*** (0.06)	0.00 (0.00)	0.01*** (0.00)	−0.00** (0.00)	−0.00 (0.00)	−0.01*** (0.00)
n	29,160	29,160	29,160	29,160	29,160	29,160	29,160	29,160	29,160
Panel D: Public School Students Only, Same School K-1
Percent Absent	−34.42*** (7.60)	−37.84*** (6.50)	−1.79* (0.69)	−25.07** (7.62)	0.14 (0.13)	0.12 (0.18)	−0.18 (0.16)	−0.27 (0.16)	−0.09 (0.17)
Individual Absences	−0.49*** (0.05)	−0.44*** (0.04)	−0.02* (0.01)	−0.42*** (0.06)	0.00 (0.00)	0.01*** (0.00)	−0.00** (0.00)	−0.00 (0.00)	−0.00*** (0.00)
n	25,460	25,460	25,460	25,460	25,460	25,460	25,460	25,460	25,460
Panel E: Including Classroom Behavior Measures
Percent Absent	−38.20*** (6.98)	−39.33*** (5.92)	−1.97*** (0.57)	−26.28*** (6.48)	0.15 (0.12)	0.10 (0.15)	−0.17 (0.15)	−0.21 (0.14)	−0.05 (0.14)
Individual Absences	−0.49*** (0.05)	−0.43*** (0.04)	−0.02** (0.01)	−0.44*** (0.07)	−0.00 (0.00)	0.00*** (0.00)	−0.00 (0.00)	−0.00 (0.00)	−0.00*** (0.00)
n	33,070	33,070	33,070	33,070	33,070	33,070	33,070	33,070	33,070

Standard errors in parentheses.

Each column represents percent absent and days absent from a unique regression analogous to the final, full models in Table 2.

Student fixed effects regressions: only include time-varying control variables from Table 2.

^*P < 0.05.

^**P < 0.01.

^***P < 0.001.

In the second panel of Table 6, we removed teacher characteristics from the student fixed effects models in Table 3 and reran the models to test for whether the findings were robust to the inclusion/exclusion of teacher characteristics, per Blazar (2015). As shown in Table 6, the results are similar to our original estimates in Table 3. Practically speaking, the exclusion of teacher characteristics (i.e., education, experience, and race/ethnicity) from our models only changed the focal significant associations by approximately 1%–5%. Considering that: (1) the significant associations of interest were roughly 4.5x larger than the standard error and (2) the teacher characteristics included in our models are often discussed as some of the most important predictors of student engagement and school success, what this suggests is that it is unlikely that there are other unobserved teacher characteristics that affect both our predictor and outcomes (above and beyond the covariates and fixed effects included in our models) to the degree necessary to invalidate our inference. Thus, even though the ECLS-K does not include a large number of teacher characteristics nor is it possible to observe teachers over multiple years, the lack of meaningful change in our parameter estimates lends confidence to our conclusions.

In the third and fourth panels of Table 6, we reran our original student fixed effects estimates but only for the sample of students who were in public schools. As described by both Author et al. (2018) and (Wang, 2013), attending private schools might proxy for other unobserved (time-varying) variables that are not subsumed in the fixed effects analysis. On the other hand, public schools provide a more uniform sample of students (Wang, 2013). Panel C presents our results for students who were in public school in both kindergarten and first grade. Panel D presents findings for students who attended the same public school in both kindergarten and first grade, as a way to remove concerns about school mobility between the 2 grades (which may cause other issues related to absenteeism). Using only public school data confirms the findings from previous analysis, and thus including private school students does not pose a threat to validity of the estimates.

As a final test of robustness, Panel E presents findings based including 2 measures of classroom behavior, given concerns that classroom absenteeism may proxy for classroom behavior. Teachers were asked to rate the frequency of classroom misbehavior on a 5-point scale (i.e., group misbehaves very frequently, group misbehaves frequently, group misbehaves occasionally, group behaves well, group behaves exceptionally well). We coded the former 2 together as a single binary measure of frequent misbehavior, the middle score as occasional misbehavior, and the latter as infrequent misbehavior. The correlation coefficient between percent absent and any of these 3 measures is quite low, with the highest of the 3 being correlated with percent absent at 0.07. And, as can be seen in Panel E of the table, the conclusions from the findings remain consistent with all tables and tests. In other words, including classroom behavior measures do not alter our inferences.

Discussion

The extant literature has clearly established that students who are more frequently absent are more likely to struggle in school, particularly at an early age (Author, 2009, 2010, 2014; Gershenson et al., 2017; Goodman, 2014; Moonie et al., 2008; Roby, 2004). Many factors have been examined with regards to what drives absenteeism. These include: educational disengagement and alienation from school (Bealing, 1990; Harte, 1994; Lehr, Sinclair, & Christenson, 2004); family influences, such as family structure, father’s occupation, mother’s work status, parental involvement, and income status (Catsambis & Beveridge, 2001; Fan & Chen, 2001; Jeynes, 2003; McNeal Jr., 1999; Muller, 1993); school context and interventions, such as teacher-pupil relations, program interventions, and having health personnel (; Bealing, 1990; Gehlbach et al., 2016; Gehlbach, Brinkworth, & Harris, 2012; Marvul, 2012); and neighborhood context (Author, 2014).

Although these studies have certainly been critical in informing both policy and practice, little effort has been made to consider how the classroom itself influences absenteeism. This is a key gap in knowledge 2 reasons. First, the classroom has previously been established as a setting where children can affect the outcomes of other children, academically and behaviorally (Author, 2014; Bonesronning, 2008; Hanushek et al., 2003; Hoxby, 20 0 0; Lazear, 2001; Ribeiro & Zachrisson, 2017). Second, understanding these classroom dynamics specifically in this study can provide a better sense of the spread of absenteeism. The present investigation addressed this issue, and in doing so, 3 important themes emerged.

First, the results of the present investigation revealed both individual and classroom effects of absenteeism for key domains of children’s academic achievement and executive functioning, and importantly, these findings were robust to several methodological specifications. When taken together, these findings are in line with Author (2011) who found that chronically absent classmates negatively affected standardized state test scores for a sample of older students. The results of the present investigation add to this discussion by shifting the focus to the earliest school grades and also revealing that younger students in classrooms with a higher percent of absent peers demonstrate not only less optimal math and reading test scores, but also less optimal executive function. This extension downward and to domains of executive functioning represent key advances given the malleability of students’ skills in the early years and the role these early skills play in students’ long-term school success (Heckman, 2006).

The above pattern of findings is noteworthy because, even if students are not frequently absent themselves, what our results would imply is that they are potentially at risk in key developmental domains as a function of the absenteeism patterns of their classmates. These spillover effects are critical in these early years of elementary school given some of the misconceptions among parents of young children regarding the harms (or lack thereof) of early absences (Author et al., 2019). Thus, our findings not only confirm that is there an opportunity cost for missing school for individual children, but also their classmates. And even though we cannot test why these spillover effects emerge, 1 possibility worth considering in the future has to do with teachers and how they help absent children catch-up upon their return to the classroom (Chen & Stevenson, 1995; Connell et al., 1994; Finn, 1993; Monk & Ibrahim, 1984). In the meantime, however, the results of the present study indicate that, when trying to address issues of absenteeism, it is critical to take a whole classroom perspective.

Second, even though in our baseline and school fixed effects models, there also appeared to be negative associations between classroom absenteeism and students’ socio-behavioral outcomes (social skills and problem behaviors), these associations were no longer present in our student fixed effects models. There is no single explanation for the loss of statistical significance in the student fixed effects models for the links between absenteeism and socio-behavioral outcomes. One possibility of course is that no associations exist between classroom absenteeism and individual students’ behavior, once parsing out unobserved individual heterogeneity or the sorting by principals of students into classrooms. That is, principals might place children with observed behavioral difficulties in the same room, expecting higher absences in that classroom, given prior literature on classroom effects of behavior (Ribeiro & Zachrisson, 2017). Baseline and school fixed effects models would not have addressed this possibility.

Another explanation, however, is that there is no association due to the design of the ECLS-K survey. Unlike achievement and executive function which are measured through direct and objective assessments, the problem behavior and social skills outcomes were reported by teachers, and this introduces a great deal of subjectivity. Moreover, the same teacher did not rate the child across grades. In light of the above, future studies with more consistent ratings of children’s socio-behavior measures are needed, if at all possible. With that said, if taken at face value, what these results imply is that the spill-over effects of absent classmates have more to do with children’s achievement and executive functioning and less to do with their socio-behavior and classroom engagement.

Third, even though prior studies have found that educators with more experience have students with fewer absences (Gershenson, 2016; Ladd & Sorensen, 2017) and that school mobility might place students at greater risk of absenteeism, we found no evidence of heterogeneity in the focal associations of interest as a function of these factors. We also found no evidence of heterogeneity in outcomes as a function of low individual student absenteeism. Instead, we found that these different groups of children were similarly affected by having absent classmates. As such, the findings reported herein regarding the negative classroom effects of absenteeism for students’ academic achievement and executive functioning are generalizable to different groups of students in a given classroom. If these findings are replicated in the future and other sources of heterogeneity yield similar patterns, then what these results suggest is that absent classmates harm all students equally. In moving forward, it will be key to understand why having a higher percent absent yields less optimal outcomes throughout the classroom setting.

Limitations

As with any study, the results reported herein need to be interpreted in light of a few key limitations. First, when it comes to classroom-level absences, teachers in the ECLS-K:2011 only reported a typical day of absences and did not report precisely which (or why) students were absent on a particular day. The above is of note because it is possible that different groups of the students were absent each day, such that over the course of the school year, any given student would have only missed a few days of school. We make note of this limitation because it would likely have ramifications for the ways in which children experience the classroom environment (and the way in which teachers respond to absenteeism). And, thus, the source of absenteeism is a key factor that future studies should consider. Second, while we proposed mechanisms in the introduction as to how absenteeism might affect the entire class, these mechanisms were not available for exploration in the current study. They merit testing in either secondary data that include how teachers respond to absent students or in smaller-scale, in depth qualitative research. Finally, the work presented here is non-experimental. Although it is not always feasible to randomly assign students to classmates, future research could exploit random variation in why some students might be absent from school, such as illness that only affects some students’ absenteeism (i.e., an exceptionally strong year for pollen for those students with asthma) or natural disasters that only affect some students but not all (i.e., fires that affect neighborhoods but not school closures).

With these limitations and future directions in mind, the results of the present study provide new insight into the ramification of classroom absenteeism for not only students’ academic learning, but also their executive functioning and socio-behavioral development. What these results make clear that even if students are not absent themselves, they are at risk for school difficulty across key developmental domains when in classrooms with more missing students.