Absent Peers in Elementary Years: The Negative Classroom Effects of Unexcused Absences on Standardized Testing Outcomes
by Michael A. Gottfried - 2011
Background/Context: This article addresses the classroom contextual effects of absences on student achievement. Previous research on peer effects has predominantly focused on peer socioeconomic status or classroom academic ability and its effects on classmates. However, the field has been limited by not discerning the individual-level academic effects of being in classrooms with absent peers.
Purpose/Objective/Research Question/Focus of the Study: The purpose of this study is to determine the peer effects of absent students in urban elementary school classrooms.
Population/Participants/Subjects: The data set is longitudinal and comprises entire populations of five elementary school cohorts within the School District of Philadelphia, for a total of 33,420 student observations. Individual student records were linked to teacher and classroom data and to census block neighborhood information.
Research Design: To examine the educational effects of absent peers, this study employed an empirical specification of the education production function. The dependent variables were Stanford Achievement Test Ninth Edition (SAT9) reading and math scores.
Findings: Models differentiated between unexcused and total absence measures and indicated that the peer effect of absences was driven by negative effects associated with classroom rates of unexcused absences rather with rates of total absences. These findings were obtained after controlling for student, neighborhood, teacher, and classroom characteristics.
Conclusions/Recommendations: Not only are absences detrimental to the absentee, but they also have a pervasive effect on the achievement of other students in the classroom.
Since the dissemination of the Coleman Report (Coleman et al., 1966), economists, sociologists, education researchers, and policy makers have been assessing the relationship between classroom peer effects and educational outcomes. The basic premise is that stronger peers can produce better academic outcomes for their classmates. In the analyses of classroom composition, the research has focused predominantly on peer effects relating to socioeconomic status (SES) (Caldas & Bankson, 1997; Link & Mulligan, 1991; Willms, 1986) and classroom academic ability (Henderson, Mieszkowski, & Sauvageau, 1978, Summers & Wolfe, 1977; Zimmer & Toma, 2000). Nonetheless, even conditional on these characteristics, there remain other channels through which students can affect other students.
One uncharted line of research on the educational effects of peers is the relationship between classroom absences and student-level performance. The importance of this peer effect arises because research has thus far shown that being absent from school is detrimental to learning and academic achievement, and an increase in absences will exacerbate educational and sociological risk factors in concurrent and future years (Dryfoos, 1990; Finn, 1993; Lehr, Hansen, Sinclair, & Christenson, 2003; Stouthamer-Loeber & Loeber, 1988). As for individual test performance, absent students receive fewer hours of instruction and consequently are likely to perform more poorly on exams (Chen & Stevenson, 1995; Connell, Spencer, & Aber, 1994; Finn, 1993). In addition, when present in the classroom, highly absent students may feel a greater sense of alienation from their classmates, teachers, and schools and may thus disrupt classroom instruction through their negative interactions and social disengagement (Ekstrom, Goertz, Pollack, & Rock, 1986; Finn, 1989; Johnson, 2005; Newmann, 1981).
Hence, the peer effect of individual student absences can arise from both academic and behavioral sources. Academically, if absent students receive fewer hours of instruction and their in-school learning time decreases, then on their return, they often require remedial instruction (Monk & Ibrahim, 1984). If teachers respond to educational needs of absent students by allocating regular class time, then nonabsent students may be adversely affected because classroom instruction is slowed. Increasingly large numbers of absences in the classroom would suggest larger portions of instruction dedicated to remediation, thereby further slowing educational advancement for other students in the room. Similarly, if absences generate negative behavioral outcomes, such as school disengagement or alienation, and if these behaviors in turn produce further problems in school (Finn, 1989), then absent students can generate noninstructional (i.e., behavioral) disruptions on return to the classroom (Reid, 1983), which can also slow the learning process for nonabsent peers. That is, teachers must devote their instructional resources elsewhere rather than to teaching.
Lazear (2001) theorized about this phenomenon by proposing that education in the classroom can be classified as a public good, in which congestion effects exist on the teachers instructional time. As such, negative effects on learning are generated when one students actions impede the learning of other classmates. According to this notion, an absent student who disrupts regular instruction on return to school utilizes teaching time in ways that nonabsent students may not find useful. In essence, absent students produce both an individual effect by decreasing their own learning from having missed school, and a peer effect by slowing instruction and reducing the educational outcomes for others in the class.
Given the individual and classroom risk factors associated with being absent, this article empirically considers the classroom peer effects of students with unexcused absences. Although some studies have evaluated the relationship between student-level absences on individual-level achievement, and others have assessed school-level effects of absences on educational outcomes, the literature has been limited on discerning the effects of classroom-level absences on student-level achievement. That is, the extant body of research has not fully considered the classroom effect of absent peers on student-level test performance.
This study contributes to the literature on attendance by evaluating the classroom contextual effects of absences on individual student achievement. Because this study has been afforded multilevel data of elementary school students in the School District of Philadelphia from 1995 to 2001, it is possible to link students to classrooms, teachers, and schools, as well as other covariates, such as demographic information and neighborhoods. Therefore, having comprehensive multilevel and longitudinal data of elementary school students in a large urban district allows for the designation of student absences, identification of classroom peers, and subsequent construction of individual and classroom peer metrics of absences for every students schooling experience.
Furthermore, the field has predominantly focused on measures of total absences. However, absences can be defined in two waysexcused and unexcused. As such, students may be exposed to peers with high levels of excused absences, though absences by these students may not signal academic, family, or social disengagement or other problems. On the other hand, a classroom may contain students with high levels of unexcused absences, which may arise from delinquency or school disengagement and not for the same reasons as excused absences (Hess, Lynon, Corsino, & Wells, 1989; Rumberger, 1995).
The difficulty in relying on the current empirical literature is that most studies have not differentiated between unexcused absences and total absences. As a consequence, the findings from these studies may potentially contain confounding issues resulting from not parsing out the effects of absence behavior deemed unexcused (by the teacher and school) from other factors of absences (i.e., a general metric of absences does not differentiate between a high-performing student with the flu, and a student with behavior or disengagement issues). To avoid these problems, this article draws distinctions between total absences and unexcused absences in constructing measures of student- and classroom-level covariates as a way to discern individual and peer effects of missing school in general from those resulting from unexcused absence behavior.
Finally, the research on absences has predominantly focused on the educational outcomes of high school students. The effects of absences on elementary school performance are lacking in the research on attendance, and empirically, studying peer effects of absent students in elementary schools allows for two advantages. The first is that unlike high school students in large urban school districts like Philadelphia, elementary school students are generally contained within the same classroom throughout the school day and academic year. As a result, studying elementary classrooms allows for a more clear-cut identification of peers with whom other students interact. This effect is too confounded in empirical studies on high school students, who move classrooms throughout the school day with the start of each new period. Thus, because classroom peers potentially alternate five to six times per day, the peer effects for high school students (and some middle school students) are difficult to identify and are potentially diluted. There is a second advantage that is particularly germane to this sample: By identifying significant factors in the schooling experiences of urban elementary youth, it is possible to develop policy and support interventions for at-risk students early in school, before chronically absent students enter into secondary education, where their risk of dropout or postgraduation misbehaviors becomes exacerbated (Alexander, Entwisle, & Horsey, 1997; Barrington & Hendricks, 1989; Lehr et al., 2003).
EMPIRICAL LITERATURE ON ABSENCES
Given that the purpose of this study is to evaluate the relationship between absent peers and academic attainment, the relevant literature focuses empirically on the relationship between absences as a predictor and educational performance as an outcome. In early research, Summers and Wolfe (1977) implemented a student-level economic model of achievement to derive a relationship between unexcused absences and sixth-grade standardized test performance in the School District of Philadelphia during the 19701971 school year. Their cross-sectional results suggested a negative effect of unexcused absences on student achievement, which was heightened for low-SES students. However, their work did not evaluate the classroom peer effects of students with unexcused absences on individual test performance.
Among other hypotheses in their assessment of student absences, Monk and Ibrahim (1984) examined the peer effects of highly absent students. In particular, the authors evaluated the relationship between absences and achievement by utilizing a data set of 227 ninth graders in one middle school in upstate New York. The results suggested negative individual- and classroom-level effects on standardized ninth-grade test performance resulting from an increase in absences both for students and their peers. Although Monk and Ibrahim relied on a small sample of a single grade in one school within a homogenous school district, this early study laid the foundation for further empirical research in two capacities. First, in evaluating achievement, the authors did not distinguish between effects of total absences and unexcused absences. Therefore, additional research could parse out the effects of unexcused absences from overall absences of classroom peers on student testing performance. Second, rather than assessing the effects of absences on multiple grades, their analysis only relied on ninth-grade students. Thus, there is a need to evaluate the effects of absences at other school levels, such as early elementary years.
Several longitudinal studies have examined how attendance patterns in early years of schooling can predict future dropout rates (Alexander et al., 1997; Barrington & Hendricks, 1989; Hess et al., 1989). Rumberger (1995) identified student- and institution-level factors that significantly relate to dropout in middle school. The results suggested that the classification of students as having moderate or high absence patterns significantly predicted the probability of dropout. Odds ratios, which can be used as measures of effect sizes when both independent and dependent variables are binary, suggested odds of 2.03 (p < .01) for dropout related to moderate absenteeism and odds of 5.10 (p < .01) for students with high rates of absenteeism.
Other empirical studies have used metrics of current attendance or absences to evaluate contemporaneous educational outcomes. Neild and Balfanz (2006) utilized a cross-section of students in the 19992000 academic year and employed logistic regressions to predict the risk factors of nonpromotion from 9th to 10th grade. Among other results, they reported that for each additional percentage point increase in eighth-grade attendance, the odds of repeating ninth grade decreased by 5%. The odds ratio for this result was reported as .96 (p < .001). Although that paper focused on high school achievement, the work nonetheless provided insight into how student-level attendance and absence information is directly related to student-level performance.
Balfanz and Byrnes (2006) evaluated comprehensive school reform models aimed at closing the math achievement gap in urban middle schools. Among the span of covariates predicting math improvement was attendance, with results indicating a 20% difference in the probability of higher math performance for students with 60% attendance rates versus those who attended every day. Like Neild and Balfanz (2006), attendance was measured for each student, though no distinction was made between total absences and unexcused absences, nor was there an evaluation of classroom peer effects.
Other studies have evaluated the school-level contextual effects of elementary school attendance on standardized test performance, thereby providing initial insight into educational effects of absent peers. Assessing Louisiana public elementary schools, Caldas (1993) studied the effects of school attendance rates, among other covariates, on a composite index of test scores. The results indicted that a one standard deviation increase in average daily school attendance was associated with a .10 (p < .001) standard deviation increase in student-level test performance. This indicated that from within the context of the school environment, average daily student attendance was a positive and significant factor in predicting same-year academic performance for inner-city students. Similarly, Roby (2004) concluded that, based on the analysis of educational outcomes in Ohio, there were statistically significant correlation coefficients (specifically, Pearsons r) between measures of school-level attendance and student achievement in fourth, sixth, ninth, and twelfth grades. Specifically, the correlation coefficients were .57, .54, .78, and .55 for each grade, respectively. These two studies have employed measures of attendance at the aggregated school level rather than for individual students or classrooms. Doing so has facilitated the opportunity for further in-depth student- and classroom-level analyses into the peer effects of the attendanceachievement relationship.
Within the framework of this body of empirical literaturein examining how missing or attending school can impact a range of academic outcomesthis study investigates individual- and peer-level effects of absences on annual student standardized testing performance, holding constant other predictors of academic achievement. To truly capture this relationship requires the use of empirical methods based on a large-scale individual-level data set in which students, teachers, and classrooms can be identified and in which absence measures can be parsed out for each student and peer group.
To examine the educational effects of absent peers in urban elementary school classrooms, this study employs the standard education production function developed by education economists and sociologists. To produce an academic outcome, it is possible to utilize an education production function to model the relationship between schooling inputs and output measures of achievement. Building on the conceptual foundations provided by Coleman and his authors (1966) and the empirical research developed by Summers and Wolfe (1977), Henderson et al. (1978), and Hanushek (1979), the particular model selected in this article evaluates the outcome of standardized test scores in both reading and math as a function of various educational inputs relating to students, classrooms, families, schools, and neighborhoods.
Specifically, the analysis of the education production function begins with a contemporaneous model of achievement. That is, at any time period t, an educational output, such as test scores, is a function of the cumulative, concurrent influences of individual student characteristics, including ability and demographic characteristics, classroom environments, family factors, and school resources. Similarly, achievement in all other time periods, such at t-1, is a function of the cumulative, concurrent inputs from the same period, t-1. Conceptually, the model to be estimated for a student in a given school year t is:
Ait = f(ABit, Iit, Fit, Nit, Cit, Tit, Sit) (1)
where A is achievement at year t for student I; AB is classroom peer effects of absent students in year t; I is a series of individual student characteristics, including transcript information on individual absences; F is a vector of family background influences; N includes variables relating to the neighborhood census block where the student resides; C is a set of classroom inputs; T is teacher characteristics; and S is school characteristics.
Conceptually, the contemporaneous model can be estimated for each school year in a students educational history. As a result, it is theoretically possible to build a model of current achievement that captures both current and previous inputs to education. This model is known as the historical education production function. Incorporating the inputs of past years, however, requires an enormous, if not impossible, amount of data. A specification that mitigates some of the data requirements and that is commonly used in the education production function literature considers the change in achievement between two years. That is, by taking the difference between year ts historical model of education production (which contains all inputs to education from the first year of schooling through year t) and year t-1s historical model of education production (which contains all inputs to education from the first year of schooling through year t-1), all that remains in the model is current achievement, a 1-year lagged measure of prior achievement, and current inputs to education. This is known as the value-added specification and is expressed as follows:
Ait = f(ABit, Iit, Fit, Nit, Cit, Tit, Sit, Ai(t-1)). (2)
In this formulation, the inputs are contained strictly to measures from year t-1 through year t. As in the contemporaneous model of education production, educational output is represented by a students standardized test performance in reading or math in the current time period t. Also, the value-added specification contains the same vector of inputs for contemporaneous measures of peer effects, individual student characteristics, classroom environments, family factors, and school resources. The distinction of this value-added model, then, arises from the fact that in this specification, the effects of all prior inputs are assumed to be captured in the lagged achievement measure from the period t-1. As a consequence, biases created by omitted variables in the past only result in biases in the estimated coefficient on lagged achievement. Thus, current achievement is not confounded with omitted characteristics that persisted in prior periods of schooling (Hanushek, Kain, Markman, & Rivkin, 2003).
From this conceptual model of educational attainment, the literature assumes a linear empirical specification of the education production function:
where A is standardized test performance for student i in classroom j in school k in year t as the dependent variable on the left-hand side of the equation and in year t-1 as the lagged measure of achievement on the right-hand side of the equation. In this linear model, the 1-year lagged achievement score can be assumed to proxy for individual fixed effects. For example, if student ability is intransient over time, then this model accurately estimates the relationship between inputs and achievement.
Empirically, the sets of independent variables, described by the education production function, would be estimated as follows. As a vector of key independent variables, AB is the effect of absent peers (measured by average total absences and unexcused absences) that student i experiences in classroom j in school k in year t. At the student level, other sets of independent variables include: I, a vector of a students characteristics, including individual-level absence information, in year t; F, a function of family inputs for student i in time period t; and N, student neighborhood census block characteristics for student i in time period t. At the classroom level, the model assigns the following inputs: C is classroom-specific characteristics for classroom j in time t, and T is teacher effects in classroom j in school k in time period t. Finally, the error term γ includes all unobserved determinants of achievement. It is in the error term where school effects are identified.
A multilevel approach is taken in these data in which the error structure is decomposed as follows:
To better understand how absences affect student achievement in the early years of education, this study is facilitated by a unique and comprehensive data set of student, neighborhood, teacher, and classroom observations. Student and teacher data were obtained from the School District of Philadelphia via the districts office of student records and through the districts personnel office. Neighborhood data were obtained from the 2000 census flat files at the census block level. Information was collected on a students home address, including street number and name and zip code. The merging of neighborhood data with the student-level database was achieved by geo-coding each address to its longitude and latitude and by assigning each student to a census block group.
The data set in its entirety comprises complete populations of elementary school student cohorts within the School District of Philadelphia over the academic period from 1994 through 2001; the analysis in this study employs observations from 1995 through 2001, as explained next. Because student characteristics have not significantly changed since the years of observations in this data set (School District of Philadelphia, 2009), absence issues addressed in this article continue to be important in the educational experiences of youth in the district. Further, because the data set is unique in that it is longitudinal, nonselective, and comprehensive of entire cohorts of students within the district, the results derived from employing these data are representative of those needs facing at-risk urban children. Hence, the approach taken in this article remains applicable, and the results are generalizable.
Inclusive of both reading and math standardized tests, the analytical sample consists of a total of 33,4001 observations within 175 public neighborhood schools with elementary grades. The data were available for five cohorts of students. The first three cohorts began kindergarten, first grade, and second grade in the 19941995 academic year. The fourth cohort began kindergarten in the 19951996 school year, and the fifth cohort began kindergarten in the 19961997 school year.
In the analysis to follow, observations from 1995 through 2001 were employed; the analytical sample is restricted to third- and fourth-grade observations because students were included in the analyses if data existed on their current and lagged standardized achievement tests in reading or math, or both. Because students in the data set have standardized testing information for second, third, and fourth grades, only third- and fourth-grade observations could be used in the value-added (lagged model) specifications described in the previous section. Furthermore, to be included in the sample, data must also exist for other input measures to the education production function, including gender, race, academic indictors (some include lagged information), classroom and teacher characteristics, assignment information (room, grade, and school identifiers), and neighborhood information.
Table 1 provides details on the dependent and independent variables employed in this study. The dependent variables are the normal curve equivalent scores (NCE) for the Stanford Achievement Test (SAT9). The NCEs are the generally preferred measurement for methodological reasonsthey have statistical properties that allow for evaluating achievement over time (Balfanz & Byrnes, 2006). Normal curve equivalents range in value from 1 to 99.
For every student in a given academic year, the data set contains demographic information concerning personal characteristics, such as gender, race, and whether the student had been enrolled in kindergarten within the School District of Philadelphia. Additional student identifiers include indicators for special education status, English language learning status, free lunch status,2 and whether the student has a behavior problem, determined by his or her behavior grade from the previous academic year.3
Data at the student level of analysis also contain a vector of neighborhood information. The empirical model contains four attributes that describe the census block on which the student resides: the percentage of a students census block that is White, the percentage of a students block at or below poverty, the household vacancy rate for the block, and the blocks average household income. Whereas the first three attributes are calculated as percentages, the last is evaluated as a natural logarithm. Note that in the absence of other direct measures of family data, free lunch status and neighborhood information often serve in empirical models as proxies for family background (e.g., Hanushek et al., 2003) because they are based on direct observation of family and neighborhood characteristics (e.g., household and census block incomes).
Table 1 also presents teacher descriptive information. Data on teachers are sourced both from student records and from the districts personnel office. A student record provides the name of the teacher assigned to a students classroom in a given academic year. In addition, a detailed teacher data set was obtained from the districts personnel office. From these, four sets of variables were incorporated into the data set. First, for each teacher, basic characteristics include race and gender. Second, a measure of teacher experience is based on appointment date variables, including district appointment date, teaching seniority date, and present position appointment date. Third, a binary variable indicates whether a teacher had a masters degree, based on the record that provides detail on which graduate school the teacher attended. Finally, a binary variable indicates if a teacher had received Pennsylvania state certification, based on completion of either Level I or Level II certificates.
Students can be grouped unambiguously into classrooms because school and classroom assignment information is included in the student database. In contrast, the teacher data set does not include school or classroom assignment. Teachers are matched to their classrooms by identifying their names, as they appear on personnel records, to the teachers names as they appear in the student data set.4 The name that appears on the report card is not always the full name of the teacher, and thus the matching algorithm is conservative in requiring teacher first and last names to be present to have a successful student-classroom-teacher match.
Once students have been matched to teachers and classrooms, it is possible to construct classroom-level peer effects. The key peer variables in this study relate to within-classroom absences. There are two types of peer absence variables in the data relevant to this study: number of total absences and number of unexcused absences. The analyses to follow utilize both measures of total absences and unexcused absences to separate the effects of absences from the signals of those absences being unexcused.
Constructing measures of peer absences begins with identifying student-level indicators of absences. Having student records on total absences and unexcused absences first and foremost controls for the individual student effect of missing school while additionally parsing out information regarding missing school for unexcused reasons. The data in Table 1 highlight that students are more likely to have unexcused absences, given that they account for approximately 75% of total absences, on average. Only aggregate numbers of absences are provided for each student and academic year; it is not possible from the data to determine at what point in the year students were absent. Moreover, aside from the distinction on the students records between excused and unexcused absences, details are not provided regarding the reasons for a specific case. However, the attendance and truancy office in the School District of Philadelphias headquarters offices provided definitions for unexcused absences that are used as guidelines in all elementary schools.
The School District of Philadelphia defines an unexcused absence as one lacking a note for short-term illnesses that last no longer than 3 days. An unexcused absence for a long-term medical illness, such as the flu or hospitalization, results from lack of a note signed by a doctor. Additionally, if a note is not presented for a short-term emergency involving immediate family, such as a death, the absence is deemed unexcused. Even if the student does bring a note for missing school, an absence can still be classified as unexcused. Unexcused absences with notes include family problems that do not involve the child, such as a parents illness or unemployment. In addition, missing school for nonschool activities, such as recreational and extracurricular activities, is deemed an unexcused absence. For example, a students sporting game is not an excused absence. Finally, missing school for suspension is always considered an unexcused absence.
Individual-level absence data can be aggregated up to classroom-level covariates, thereby creating measures of the peer effects of absent students. These are constructed as rates of absences: the average number of total absences per student in a classroom and the average number of unexcused absences per student in a classroom. To avoid confounding empirical issues, the classroom-level absence variables for student i do not include student is own measure of absences. In other words, the effect of absent peers does not contain a students own absence information, but instead refers to the absence rates of the students classmates. As such, every student with absences will have a slightly different value for classroom peer effects, depending on his or her unique individual record. The exception is for all students in a classroom who do not have any absences or who have the same number of absences: They will have the same value for classroom-level absence rates, respectively.
To control for the possibility that high rates of absences may occur in special or tracked rooms, both class size and mean classroom lagged test scores (reading or math depending on the regression outcome) are included in the analyses. The average class size is approximately 28 students. Mean lagged test scores are based on the one-year lagged testing outcomes for students in the classroom. It is constructed analogously to the peer measure of absences: Student is lagged test outcome is not included in the aggregate class score. Thus, each student will have a slightly different average class score in the data. A squared term of the class mean is also included to capture nonlinear, in addition to linear, effects of classroom test performance.5 Table 2 presents the correlations and their significances between classroom peer absence indicators and outcomes, as well as other inputs in the model. The results suggest almost no correlation between class size and class absence rates or mean class testing ability and class absence rates. Thus, classrooms with higher absences do not appear to be related to particular types of classrooms, large or small, and high performing or otherwise.6
The results from Table 2 also suggest that Black students are more likely to be in classrooms with higher rates of unexcused absences and White students less likely, though this result does not hold when looking at classroom averages of total absences rather than averages of unexcused absences. Neighborhood information shows that students whose neighborhood blocks have higher income levels, higher percentages of Whites, or lower vacancy rates are in classrooms with lower rates of unexcused absences. Otherwise, no strong relationship stands out between absences and the predictors or outcomes in the model. However, in sum, this table does indicate that more significant relationships arise in terms of classrooms with high rates of unexcused absences as compared with those with high rates of total absences. Thus, there is a suggested peer effect: Students with high instances of unexcused absences are more likely to have peers with higher rates of unexcused absences.
Table 3 presents regression results for SAT9 reading and math outcomes. In the first two columns of each test subject outcome, the regressions include predictors of total absence measures at both student and classroom levels, among other covariates. The third and fourth columns of each subject test outcome have taken the first two regressions and have added to them measures of unexcused absences at student and classroom levels. Doing so allows this study to determine if, after controlling for measures of total absences, unexcused absence behavior remains a significant predictor of test performance.
In addition, within each pair of regressions and within each testing subject area (i.e., the models that have only measures of total absences versus those models that also incorporate unexcused absence measures), there are two specified empirical models: baseline and full. Baseline models are regressions estimated with ordinary least squares and include grade and year indicators and standard errors adjusted for clustering within classrooms. Baseline models do not include school fixed effects. Full models then incorporate school fixed effects to the baseline models.
Assessing the coefficients pertaining to the effects of absent peers (which are located in the top four rows of Table 3) indicates several findings of interest. First, within each testing subject area and within either total or unexcused absences regressions, the differences in sizes of the coefficients between the baseline and full models (i.e., with and without school fixed effects) are not large. What this implies, then, is that the inclusion of school fixed effects (which, recall, controls the average school effect) does little to alter this articles premise that higher classroom rates of peer absences have negative, significant effects on reading and math standardized testing achievement.
Second, even after controlling for student, neighborhood, teacher, and classroom characteristics, absence predictors remain significantly related to reading and math performance. In fact, the results on the effects of absent peers are always statistically significant and equally conceptually compelling throughout Table 3. Turning first to just the implications of total days absent in the first two columns of each test subject area, the coefficients on individual-level total absences and peer classroom rates of total absences demonstrate that, holding all else constant, there are two preliminary significant effects on students own reading and math achievement: a negative student effect of missing school and a negative classroom effect of having peers who have, on average, more total days absent. In other words, holding all else equal, students with more total absences perform worse in reading and math than students with lower instances of total absences. Moreover, after controlling for a students own record of absences (including other predictors), students in classrooms where peers have higher rates of total absences tend to have lower test scores than students in classrooms where peers have lower rates of total absences.
Appendix A presents effect sizes as standardized beta coefficients for absence measures of Table 3. When employing nonexperimental methods, the use of standardized beta coefficients is often implemented as the effect size in both the peer effects literature (e.g., Ammermueller & Pischke, 2006; Hoxby, 2000; McEwan, 2003) and in the research on absences (e.g., Caldas, 1993). Specifically, Ammermueller and Pischke reported that the majority of the peer effects literature presented standardized beta coefficients within a range of .05σ to .10σ. The effect sizes throughout this table for both reading and math regressions are consistent with effect sizes presented in the empirical literature on peer effects (Ammermueller & Pischke). For those regressions that only include measures of total absences at student and peer levels, the baseline and full models in Appendix A suggest that a one standard deviation increase in the average class rate of total absences leads to a -0.03σ to -0.04σ decline in test performance. For students with high records of total absences, there is the possibility of a compounded negative effect: the negative effect of their levels of total absences on test performance plus the potential negative spillover effect of being in a classroom in which their peers also have high rates of total absences. In other words, the negative peer effect of total absences heightens the already negative individual effect on standardized test performance.
The addition of peer measures for unexcused absences (as well as an additional unexcused absence control measure at the student level) provides a slightly different, and potentially more accurate, story of classroom peer effects. Incorporating unexcused absence measures into the previous models, which strictly had predictors relating to total absences, these new regressions are found in the second set of models within each subject testbroken out by baseline and full models. The coefficients on student- and classroom-level measures of unexcused absences demonstrate that as unexcused absences increase both for students and their peers, test scores decline. This implies two discernable effects in the model: a negative student effect and a negative classroom peer effect, much like in the previous results pertaining strictly to measures of total absences.
First, both baseline and full models of unexcused absences indicate that even after controlling for an individual record of total absences, students with a higher number of unexcused absences have lower test performance than do students with lower numbers of unexcused absences. Combined with the negative coefficients on total days a student is absent, the results here demonstrate that being absent for unexcused reasons exacerbates the negative effect of missing school.
The results also suggest that there is a negative spillover effect of students with unexcused absences onto others in the classroom. This peer effect of unexcused absence behavior is demonstrated statistically by the negative coefficients on peer averages of unexcused absences in reading and math test performance. That is, for classrooms in which peers have higher unexcused absence rates, controlling for the classroom rate of total absences, students perform worse than those in classrooms with lower rates of unexcused absences in both reading and math (-0.04σ to -0.06σ). This implies that even after controlling for total absences and covariates relating to students, neighborhoods, teachers, and classrooms, there nonetheless remains a negative peer effect from those classmates who are missing school for unexcused reasons. That is, there are negative effects associated with students who have high rates of unexcused absences, and this negative student effect on test performance is increased when these students have classmates with high instances of unexcused absences.7
Interestingly, the results indicate that it is the peer effect of unexcused absences, rather than the peer effect of total absences, that drives the negative effects of absenteeism. That is, when the models incorporate measures of unexcused absences in the second set of regressions for both reading and math, there is no longer statistical significance on the class average of total number of days absent. Rather, once having controlled for the class average rate of total absences, the statistically significant covariates on the measures of peer rates of unexcused absences imply negative effects of students in classrooms where peers have higher rates of unexcused absences compared with students whose peers have lower rates of unexcused absences. In other words, the classroom effect of absent peers arises from being exposed to peers with unexcused absence behavior. This indicates that perhaps it is not simply sufficient to describe the relationship between peer absences and student test performance in aggregated, or general, absence terms. That is, a higher classroom rate of total absences does not necessarily imply disruptive behaviorit is entirely possible that a classroom with higher levels of total absences may in reality have high-performing students who are sick and missing school for excused reasons.8
Finally, given that the state of literacy and math performance continues to be prevalent in the achievement effects spotlight, this study has differentiated between reading and math achievement outcomes in order to draw distinctions between the two subjects. As noted, the results show that declines in reading and math test scores are exacerbated for students with high levels of unexcused absences and for classrooms with high averages of unexcused absences. However, the results indicate that the negative effects of unexcused absences are more severe with math achievement, as evidenced by coefficients and effect sizes in math that are double those of reading. This result is germane within this studys urban school sample because it is particularly those minority and high-poverty students who fall behind in math achievement beginning as early as fourth grade (Balfanz & Byrnes, 2006).
Examining the control variables in Table 3 provides the following results. Males are associated with having a lower reading test score compared with females, and Black and Latino students tend to have lower reading and math test scores compared with White students (Caldas & Bankston, 1997; Ogbu, 1989; Summers & Wolfe, 1977). On the other hand, Asian students tend to have higher math testing outcomes than do White students. Similarly, the coefficients on being a special education student, English language learner, or free lunch recipient or having a behavior problem are negative and statistically significant in both reading and math achievement. Finally, the table does not indicate significant coefficients relating to student-level neighborhood attributes. The results of student-level neighborhood characteristics may suggest that it is the students characteristics and school environment that affect achievement levels in the education production function model.
As for teacher characteristics, the overall lack of significance in the reading achievement models is consistent with many education production studies, including Hanushek (1986). In the reading achievement model, teacher gender, race, and education do not significantly relate to test performance. In the math model, however, various teacher ethnicities do have statistically significant relationships to achievement. For both reading and math, consistent with Coates (2003), teacher experience is not a significant predictor of student achievement.
Nonabsence classroom characteristics generally do not provide significant findings. This is true for all models in Table 3. Ferguson (1991) and Coates (2003) asserted that class size enters nonlinearly into the education production function. The results here indicate that class size, as measured by the number of students in a single classroom, and class size squared may not be important factors in student achievement (Henderson et al., 1978). Finally, being in a classroom with a higher test class average is in general not significant on individual performance (except for two models in reading), once accounting for other covariates in the model.
TESTS OF ROBUSTNESS
This section tests two different structures of fixed effects to confirm the previous analyses. The models in the preceding section evaluate the classroom peer effect of absences by controlling for inputs related to student, neighborhood, teacher, and classroom characteristics and by implementing school fixed effects to control for systematic characteristics of a particular institution. This section briefly examines two alternative multilevel structures that account for either the systematic year-to-year or grade-to-grade changes in school factors. As with the preceding models, the specifications here control for school effectiveness, thereby allowing the principal source of variance to be identified within schools and across classrooms.
The models of educational production are similar to those from Table 3. For brevity, Table 4 presents the main findings for reading and math outcomes.9 Also analogous to Table 3, the first two regressions for each testing outcome present results for models containing only absence measures pertaining to total absences at the student and classroom levels. The second set of regressions then incorporates individual and peer measures of unexcused absences. In addition, within each set of models pertaining to either total absences or unexcused absences, the regressions are differentiated by having either school-by-year or school-by-grade fixed effects. All regressions have robust standard errors adjusted for classroom clustering.
The results of these models correspond to those in the previous section. The similarities between the coefficients in Tables 3 and 4 demonstrate the robustness of the empirical educational production function specified in this article. In particular, the coefficients indicate that classroom rates of unexcused absences represent statistically significant peer effects, as before. The effect sizes, again constructed as standardized beta coefficients, are presented in Appendix B. The results indicate that a one standard deviation increase in the classroom rate of unexcused absences is associated with a 0.03σ to 0.04σ decrease in reading or math test performance. As with the main analyses, these effect sizes of the peer variables are consistent with contemporary empirical literature on peer effects: Again, Ammermueller and Pischke (2006) have suggested that the majority of studies on peer effects reported standardized beta coefficients that hover around .05σ to .10σ.
In addition to the school, grade, and year fixed effects tests of robustness conducted in this section, another test of validity can examine the models. Specifically, this final test includes student-level fixed effects into each model presented in this article. If the results remain consistent, then this would yield evidence that measures relating to class average absences do not solely proxy for SES and other family characteristics that may not be adequately controlled for by the variables in the model. Hence, in a set of alternative specifications, models that incorporated student fixed effects (by coding dummy variables for each student identification number) were tested in the main empirical model and those presented in this subsequent section. The coefficients of student-level and peer absencesthe parameters of key interestwere quantitatively similar to the coefficients in Tables 3 and 4. As such, the statistical outcome remains valid, as do their implications.10
From all specifications in this research, the results indicate that the peer effects of having unexcused absences are significant, above and beyond the inclusion of student, neighborhood, teacher, and classroom variables or multilevel specification. The overarching conclusion on peer unexcused absence rates shines through: Students with unexcused absences are detriments not only to their own educational outcomes but also to others in the classroom. This negative spillover effect permeates throughout all models.
This study provides insight into the academic consequences of absent peers. Although much of the literature has focused on the effects of student-level absences or schoolwide attendance measures, this article has demonstrated that it is crucial to develop a relationship between classroom peers and student academic attainment. Furthermore, this study has parsed out a distinction in type of absence, based on official school records, in order to empirically differentiate between the effects of missing school because of unexcused reasons, and other absence behavior. Doing so for elementary school classrooms has not only its methodological advantages (i.e., children are contained in a single classroom through the day and year) but also policy implications regarding the predictors of early schooling risk.
Overall, the results suggest two negative effects of absences that remain significant even after controlling for student, neighborhood, teacher, and classroom characteristics and after modeling for school, year, and grade fixed effects. First, the results suggest an individual student effect of absences: Controlling for all other inputs in the model, students with more absences perform worse on standardized reading and math tests than do students with lower instances of absences. Second, the results indicate a negative peer effect of absences: Holding constant an individuals own record of absences, students whose classmates are absent more have lower test scores than students whose peers have lower rates of absences. In a sense, the negative peer effect of absences heightens the already negative individual effect on standardized test performance.
Furthermore, when the models in this study differentiate between unexcused and total absence measures, the results indicate that the peer effect on absences is actually driven by negative effects associated with classroom rates of unexcused absences rather with rates of total absences. Statistically, once controlling for the rates of peer total absences, the covariates on the peer effect of unexcused absences are significant, whereas those on total absences are not. Therefore, this study demonstrates that it is not sufficient to assert that classroom peer effects based on aggregate measures of total absences negatively relate to student test performance. Rather, the negative classroom peers effects of absences arise from those students with unexcused absences in particular.
Evidence suggests that students with high instances of unexcused absences are academically disengaged (Bealing, 1990; Harte, 1995; Reid, 1983; Southworth, 1992). Moreover, having a high proportion of unexcused absences in elementary school may be indicative of negative family environments (Sheldon, 2007) in which parents are absent from, unaware of, or uninvolved in their childrens schooling (Catsambis & Beveridge, 2001; Fan & Chen, 2001; Jeynes, 2003; McNeal, 1999; Muller, 1993). Although this particular study cannot pinpoint specific cases for each absence per student in a given year, the results of this article nonetheless suggest that behavioral indicators of academic disengagement and social problems that may arise in the form of unexcused absences affect not only absentees but also their classmates.
Absence behavior, then, not only has negative effects on individual student test performance, but also has potential to diminish academic outcomes for other students via a strain on classroom resources. If the classroom instructional environment can be classified as a public good, as Lazear (2001) has suggested, then any deviance from regular instruction caused by one student can be interpreted as a negative externality onto his or her peers. This negative outcome materializes as a decline in academic performance, which has been highlighted in this article. Thus, under the premise that absent students who are disengaged from school are also disengaged when in school (Bealing, 1990; Harte, 1995; Reid, 1983; Southworth, 1992), teachers must divert their attention from instruction and toward remediating and potentially disciplining these disengaged students when they return from cases of unexcused absenteeism. As such, peer effects from students with unexcused absences arise because the classroom environment has diminished in quality because teachers must devote more resources toward these absent students on return to the classroom.
If unexcused absences do in fact serve as signals of academic and social risk, and if these peers do have negative effects on their classmates and classrooms, then the policy implications of this article arise from the fact that disentangling empirical absence measures for students and their classroom peers into more refined indicators can lead to more realistic predictors of educational outcomes. By differentiating patterns of unexcused absences from total absences not only at the student level or as a schoolwide average, as seen previously in the research, but also as a classroom-level peer effect, as shown in this article, then researchers and practitioners can more efficiently identify at-risk students and at-risk peer groups early in schooling. Practically, if schools could home in on those students or classrooms with particularly high rates of unexcused absences, doing so might mitigate patterns of unexcused absences, thereby improving the use of classroom resources and subsequent test performance. Further, by providing ameliorating interventions to enhance school engagement for students with high rates of unexcused absences, schools can further decrease potential future individual academic and sociological negative effects generated from patterns of unexcused absence behaviors (Alexander et al., 1997; Broadhurst, Patron, & May-Chahal, 2005; Kane, 2006; Lehr et al., 2003; Slavin, 1999).
The results of this article are corroborated by intervention studies that have shown that early levels of attendance can be important predictors of school success. The empirical work in this article thus suggests that those school interventions that try to mitigate absences and stimulate engagement, such as Check & Connect (Lehr et al., 2003), have not gone unwarranted: They found a statistically significant decline of 23% in school disengagement (measured by declines in absences) over 2 years of the study. This article also supports previous efforts of attendance interventions regarding school outreach to parents and families, and the results from this study underscore a continued need for parental involvement in the school and community to improve academic outcomes for at-risk students and their peers (Epstein, 2001; Sheldon, 2003, 2007). Multiple intervention studies have supported the benefits of programs that improve the relationship between parents and schools: Epstein and Sheldon (2002), for instance, found significant partial correlation values of approximately .75 between increased familyschool relationships and student attendance. However, in determining how to allocate resources to such programs, practitioners and policy analysts can sharpen their efforts and analyses by identifying not only students with high rates of unexcused absences but also their classroom peers.
Conducting research and developing policies at these refined levels of analyses may also have spillover benefits for schools and communities. For instance, if schools can effectively mitigate students leaving school for unexcused reasons and increase academic engagement in their classrooms, not only would there be an immediate effect on student educational and sociological outcomes but also an aggregate effect on school performance. Because appropriations to individual schools are often based on a previous years aggregate performance, the improvement in the learning environment via a decrease in unexcused absence behavior may impact school funding. Moreover, some schools may receive financial benefits from lower levels of absences. Therefore, schools have an additional economic incentive to decrease absences because funding may be directly dependent on absences rates, as Epstein and Sheldon (2002) have suggested. Thus, in conjunction with mitigating student rates of unexcused absences and decreasing overall absences, schools can not only increase educational outcomes of their student body but also enjoy economic benefits.
Given that this study is based on a sample of primarily high-poverty minority students in Philadelphia, there are also urban policy implications. For instance, given the evidence that absences in early years of schooling are associated with a higher dropout rate, unemployment, and probability of involvement in illicit activities (Alexander et al., 1997; Broadhurst et al., 2005; Kane, 2006), policies that prevent these students from early rates of absences through stimulated engagement can directly benefit the student and improve the community. For instance, if programs can mitigate unexcused absence behavior and improve academic engagement in early schooling, thereby increasing the performance for students and their peers and decreasing the risk of future dropout, then there is the potential for increased future economic opportunities for such students and their classmates who can enjoy higher academic attainment. Moreover, the urban community also experiences a benefit: Lessened unexcused absences and improved academic outcomes might diminish potential neighborhood economic problems often associated with increased dropout and unemployment. Although these community benefits are not immediately actualized for elementary school students in this article, the outcomes from preventing students from missing school for unexcused reasons in early years may improve the future economic state of the city.
The data and analyses used in this study were multilevel and highly comprehensive of students backgrounds and classroom environments, though further research in this area would yield additional understanding. For instance, this study has provided specific data on unexcused absences, as defined by the district. Whereas the data set used provides information as to whether an absence was unexcused, it does not provide specific reasons for each absence. Having this detailed data would allow for future research to evaluate particular types of unexcused absences, thereby enabling researchers to analyze even more specific categories of absence behavior. Future distinctions may prove useful in examining how specific types of unexcused absences play a role in achievement.
For instance, because previous literature has suggested that absences may be indicators of academic disengagement or negative family environments (Kearney & Silverman, 1995; Sheldon, 2007), further research might examine longitudinal metrics of academic motivation and parental involvement and engagement, in addition to the multilevel covariates in this article regarding unexcused absences. Combining student psychological and sociological factors in a large-scale administrative database would thus parse out factors that predict unexcused absences and then enable a link to be established between student psychological and sociological backgrounds and peer effects in the classroom.
Finally, the scope of this research pertains to elementary school student outcomes. Because of the structure of the elementary school learning environment, where students remain contained in the same classrooms, among the same peers, throughout the school day, the peer group is more defined than in middle or high schools, where students and their peers shuffle on an hourly basis. Thus, with data that can identify peers classroom by classroom, further research may incorporate the analysis of absent peers of middle and high school classrooms. In addition, with a longitudinal data set that incorporates elementary, middle, and high school observations, it may be possible to evaluate the exacerbated effects of absent peers not only in elementary school but also on future years of learning.
The research here was supported in whole by the Institute of Education Sciences, U.S. Department of Education, through Grant No. R305C050041 to the University of Pennsylvania. The opinions expressed are those of the author and do not represent the views of the Institute of the U.S. Department of Education. The author wishes to thank Laura Desimone, Adele E. Gottfried, Allen Gottfried, and Ruth Neild for advice and commentary. The author also would like to thank Robert Inman and the School District of Philadelphia for providing an invaluable source of data. Finally, the author acknowledges the valuable insight of the journals editorial and review team.
1. Data used in the proceeding regressions present smaller sample sizes depending on the standardized test subject area. However, those subsamples employed for each subject test are no different than the third- and fourth-grade population. Results showing no statistical significance are available on request.
2. The measure of free lunch is a strict measure of high poverty because it only includes those students who are on fully funded programs. Reduced lunch is not included in the data, which would increase the percentage of low-income families to much higher numbers but would dilute the current measure from high poverty to poverty.
3. Though student observations were dropped from the analysis because of missing test score data, the correlation between absenteeism and test scores is extremely small. The correlation between missing test score data and number of days absent is .05. The correlation between missing test score data and number of absent days unexcused is .04. Hence, no systematic relationship appears to exist between students with missing test scores and absence rates.
4. The name of each students teacher in each year appears as part of the students record; that information is extracted from the students report card along with the classroom number.
5. Much of the literature assumes that the higher mean in the class is associated with a higher level of educational attainment but that this effect may diminish (the squared term will be negative; Henderson et al., 1978; Zimmer & Toma, 2000).
6. To address, in more detail, the allocation of students to classrooms, two ancillary models were tested in which lagged measures of student absences were regressed on all other covariates in the main specification: one model for all students in the entire analytical sample, and one strictly for students in the same school in years t-1 and t. In all models, there is a lack of statistical significance on class size and mean classroom ability and their squared terms (in both reading and math), thereby suggesting that a nonsystematic relationship exists between student absences in year t-1 and classroom assignment in year t. A similar interpretation exists when lagged measures of unexcused absences are regressed on the model.
7. In two additional specifications, the peer effect was first replaced with rates of unexcused absences per student with any unexcused absences (rather than the rate of unexcused absence per class student as run in the article). The logic behind this new variable is to assess the rate of unexcused absences among students with unexcused absences within the classroom. The results are comparable in magnitude and significance to those presented in the article. A second model incorporated the standard deviation of the original equation above, though the coefficient on this measure of standard deviation is insignificant. The results of both of these additional specifications indicate that is it the overall classroom rate of unexcused absences, rather than the distribution of absences among students, that influences the negative peer effect of unexcused absence behavior.
8. For instance, in a separate model, the unexcused student and peer effects were replaced with class averages of excused absences. Controlling for total absences, the peer effect on a higher room average of excused absences, indicates a positive, significant result. This indicates that classrooms with a higher average of students who miss school for excused reasons tend to also have higher performing students compared with classrooms with lower rates of excused-absence peers. Results are available on request.
9. Full results of all student, teacher, classroom, and neighborhood input variables are available on request.
10. The results of the student-level fixed effects analyses are available on request.
Alexander, K. L., Entwisle, D. R., & Horsey, C. S. (1997). From first grade forward: Early foundations of high school dropout. Sociology of Education, 70, 87107.
Ammermueller, A., & Pischke, J. S. (2006). Peer effects in European primary schools: Evidence from PIRLS (Discussion Paper No. 06-027). Mannheim, Germany: Centre for European Economic Research.
Balfanz, R., & Byrnes, V. (2006). Closing the mathematics achievement gap in high-poverty middle schools: Enablers and constraints. Journal of Education for
Students Placed at Risk, 11, 143159.
Barrington, B. L., & Hendricks, B. (1989). Differentiating characteristics of high school graduates, dropouts, and nongraduates. Journal of Educational Research, 82, 309319.
Bealing, V. (1990). Pupil perception of absenteeism in the secondary school. Maladjustment and Therapeutic Education, 8, 1934.
Broadhurst, K., Patron, K., & May-Chahal, C. (2005). Children missing from school systems: Exploring divergent patterns of disengagement in the narrative accounts of parents, careers, children, and young people. British Journal of Sociology of Education, 26, 105119.
Caldas, S. J. (1993). Reexamination of input and process factor effects in public school achievement. Journal of Educational Research, 86, 206214.
Caldas, S. J., & Bankston, C., III. (1997). Effect of school population socioeconomic status on individual academic achievement. Journal of Educational Research, 90, 269277.
Catsambis, S., & Beveridge, A. A. (2001). Does neighborhood matter? Family, neighborhood, and social influences on eighth grade mathematics achievement. Sociological Focus, 34, 435457.
Chen, C., & Stevenson, H. W. (1995). Motivation and mathematics achievement: A comparative study of Asian-American, Caucasian, and East Asian high school students. Child Development, 66, 12151234.
Coates, D. (2003). Education production functions using instructional time as an input. Education Economics, 11, 273 292.
Coleman, J. S., Campbell, E. Q., Hobson, C. J., McPartland, J., Mood, A. M., Weinfeld, F. D., et al. (1966). Equality of educational opportunity. Washington, DC: U.S. Government Printing Office.
Connell, J. P., Spencer, M. B., & Aber, J. L. (1994). Educational risk and resilience in African-American youth: Context, self, action, and outcomes in school. Child Development, 65, 493506.
Dryfoos, J. (1990). Adolescents at risk: Prevalence and prevention. New York: Oxford University Press.
Ekstrom, R. B., Goertz, M. E., Pollack, J. M., & Rock, D. A. (1986). Who drops out of high school and why? Findings from a national study. Teachers College Record, 87, 356373.
Epstein, J. L. (2001). School and family partnerships: Preparing educators and improving schools. Boulder, CO: Westview Press.
Epstein. J. L., & Sheldon, S. B. (2002). Present and accounted for: Improving student attendance through family and community involvement. Journal of Educational Research, 95, 308318.
Fan, X., & Chen, M. (2001). Parental involvement and students academic achievement: A meta-analysis. Educational Psychology Review, 13, 122.
Ferguson, R. F. (1991). Paying for public education: New evidence on how and why money matters. Harvard Journal of legislation, 28, 465497.
Finn, J. D. (1989). Withdrawing from school. Review of Educational Research, 59, 117142.
Finn, J. D. (1993). School engagement and students at risk. Washington, DC: National Center for Education Statistics.
Hanushek, E. A. (1979). Conceptual and empirical issues in the estimation of education production functions. Journal of Human Resources, 14, 35188.
Hanushek, E. A. (1986). The economics of schooling: Production and efficiency in public schools. Journal of Economic Literature, 24, 11411178.
Hanushek, E. A., Kain, J. F., Markman, J. M., & Rivkin, S. G. (2003). Does peer ability affect student achievement? Journal of Applied Econometrics, 18, 527544.
Harte, A. J. (1995). Improving school attendance: Responsibility and challenge. Toronto, Ontario, Canada: Canadian Education Association.
Henderson, V., Mieszkowski, P., & Sauvageau, Y. (1978). Peer group effects and educational production functions. Journal of Public Economics, 10, 97106.
Hess, G. A., Lyons, A., Corsino, L., & Wells, E. (1989). Against the odds: The early identification of dropouts. Chicago: Chicago Panel on Public School Policy and Finance.
Hoxby, C. (2000). The effects of class size on student achievement: New evidence from population variation. Quarterly Journal of Economics, 115, 12391285.
Jeynes, W. H. (2003). A meta-analysis: The effects of parental involvement on minority childrens academic achievement. Education and Urban Society, 35, 202218.
Kane, J. (2006). School exclusions and masculine, working-class identities. Gender and Education, 18, 673685.
Kearney, C. A., & Silverman, W. K. (1995). Family environment of youngsters with school refusal behavior: A synopsis with implications for assessment and
treatment. American Journal of Family Therapy, 23, 5972.
Johnson, G. M. (2005). Student alienation, academic achievement, and WebCT Use. Educational Technology and Society, 8, 179189.
Lazear, E. P. (2001). Educational production. Quarterly Journal of Economics, 116, 777803.
Lehr, C. A., Hansen, A., Sinclair, M. F., & Christenson, S. L. (2003). Moving beyond dropout prevention to school completion: An integrative review of data based interventions. School Psychology Review, 32, 342364.
Link, C. R., & Mulligan, J. G. (1991). Classmates effects on Black student achievement in public school classrooms. Economics of Education Review, 10, 297310.
McEwan, P. (2003). Peer effects on student achievement: Evidence from Chile. Economics of Education Review, 60, 131141.
McNeal, R. B. (1999). Parental involvement as social capital: Differential effectiveness on science achievement, truancy, and dropping out. Social Forces, 78, 117144.
Monk, D. H., & Ibrahim, M. A. (1984). Patterns of absence and pupil achievement. American Education Research Journal, 21, 295310.
Muller, C. (1993). Parental involvement and academic achievement: An analysis of family resources available to the child. In B. Schneider & J. S. Coleman (Eds.), Parents, their children, and schools (pp. 77114). Boulder, CO: Westview Press.
Neild, R. C., & Balfanz, R. (2006). An extreme degree of difficulty: The educational demographics of urban neighborhood schools. Journal of Education for Students Placed at Risk, 11, 123141.
Newmann, F. (1981). Reducing student alienation in high schools: Implications of theory. Harvard Educational Review, 51, 546564.
Ogbu, J. U. (1989). The individual in collective adaptation: A framework for focusing on academic underperformance and dropping out among involuntary minorities. In L. Weis, E. Farrar, & H. G. Petrie (Eds.), Dropouts from school: Issues, dilemmas, and solutions (pp. 181204). Albany: State University of New York Press.
Reid, K. (1983). Institutional factors and persistent school absenteeism. Educational Management Administration and Leadership, 11, 1727.
Roby, D. E. (2004). Research on school attendance and student achievement: A study of Ohio schools. Educational Research Quarterly, 28, 314.
Rumberger, R. W. (1995). Dropping out of middle school: A multilevel analysis of students and schools. American Education Research Journal, 32, 583625.
School District of Philadelphia. (2009). Retrieved from http://www.phila.k12.pa.us/
Sheldon, S. B. (2003). Linking school-family-community partnerships in urban elementary schools to student achievement on state tests. Urban Review, 35, 149165.
Sheldon. S. B. (2007). Improving student attendance with school, family, and community partnerships. Journal of Education Research, 100, 267275.
Slavin, R. E. (1999). Educating young students at risk of school failure: Research, practice, and policy. In R. Stevens (Ed.), Teaching in American schools (pp. 103119). Upper Saddle River, NJ: Prentice Hall.
Southworth, P. (1992). Psychological and social characteristics associated with persistent absence among secondary aged school children with special reference to different categories of persistent absence. Personality and Individual Differences, 13, 367376.
Stouthamer-Loeber, M., & Loeber, R. (1988). The use of prediction data in understanding delinquency. Behavioral Sciences and the Law, 6, 333354.
Summers, B., & Wolfe, A. (1977). Do schools make a difference? American Economic Review, 67, 639652.
Willms, J. D. (1986). Social class segregation and its relationship to pupils examination results in Scotland. American Sociological Review, 51, 224241.
Zimmer, R. W., & Toma, E. F. (2000). Peer effects in private and public schools across countries. Journal of Policy Analysis and Management, 19, 7592.
Effect Sizes (Standardized Beta Coefficients) for Absence Variables From Table 3
Effect Sizes (Standardized Beta Coefficients) for Absence Variables in Robustness Tests
Effect Sizes (Standardized Beta Coefficients) for Absence Variables From Table 3