Family Environment Variables as Predictors of School Absenteeism Severity at Multiple Levels: Ensemble and Classification and Regression Tree Analysis

School attendance problems, including school absenteeism, are common to many students worldwide, and frameworks to better understand these heterogeneous students include multiple classes or tiers of intertwined risk factors as well as interventions. Recent studies have thus examined risk factors at varying levels of absenteeism severity to demarcate distinctions among these tiers. Prior studies in this regard have focused more on demographic and academic variables and less on family environment risk factors that are endemic to this population. The present study utilized ensemble and classification and regression tree analysis to identify potential family environment risk factors among youth (i.e., children and adolescents) at different levels of school absenteeism severity (i.e., 1 + %, 3 + %, 5 + %, 10 + %). Higher levels of absenteeism were also examined on an exploratory basis. Participants included 341 youth aged 5–17 years (M = 12.2; SD = 3.3) and their families from an outpatient therapy clinic (68.3%) and community (31.7%) setting, the latter from a family court and truancy diversion program cohort. Family environment risk factors tended to be more circumscribed and informative at higher levels of absenteeism, with greater diversity at lower levels. Higher levels of absenteeism appear more closely related to lower achievement orientation, active-recreational orientation, cohesion, and expressiveness, though several nuanced results were found as well. Absenteeism severity levels of 10–15% may be associated more with qualitative changes in family functioning. These data may support a Tier 2-Tier 3 distinction in this regard and may indicate the need for specific family-based intervention goals at higher levels of absenteeism severity.


INTRODUCTION
School attendance problems, including school absenteeism, are common to many students worldwide (UNESCO, 2012). School absenteeism has been linked to academic performance and achievement deficiencies, various mental health and social problems, and later school dropout (Bridgeland et al., 2006;Burton et al., 2014;Attwood and Croll, 2015). School problem-solving skills. Mixed families display characteristics of several of these patterns (Kearney and Silverman, 1995;Barber and Buehler, 1996).
In addition, researchers have begun to focus on the concept of multi-tiered systems of support (MTSS) and related models to conceptualize different layers of intervention for school attendance problems (Freeman et al., 2016;Kearney, 2016;Elliott and Place, 2019). MTSS aims to provide high-quality, individualized instruction, and intervention, informed by frequent progress monitoring, for all aspects of student education (McIntosh and Goodman, 2016). MTSS models are often arranged in three tiers that focus on prevention (Tier 1), early intervention for emerging, acute problems (Tier 2), and intensive intervention for chronic and severe problems (Tier 3; Eagle et al., 2015). MTSS models have been applied to academic, social, and behavioral problems and skills across various age ranges and school settings (August et al., 2018). Kearney and Graczyk (2014) were the first to apply MTSS principles to a model of school absenteeism directly. Each MTSS tier has a specific focus based on the severity of school absenteeism: (1) Tier 1 focuses on enhancing functioning and schoolwide attendance and preventing absenteeism for all students, (2) Tier 2 focuses on addressing students with emerging, acute, or mild to moderate school absenteeism, and (3) Tier 3 focuses on addressing students with chronic and severe school absenteeism (Kearney, 2016). Specific interventions are matched to each tier to help school personnel identify individualized responses. Recent research has demonstrated the value of applying MTSS models to school absenteeism. For example, schools that implement MTSS with higher fidelity have lower levels of school absenteeism than schools with less fidelity (Freeman et al., 2016). School districts may also include attendance measures in MTSS models (Coffey et al., 2018).
A key task for researchers utilizing MTSS models for school absenteeism has been to identify demarcations between the tiers. A distinction between Tiers 1 and 2 essentially means a distinction between nonproblematic and problematic behavior, such as between appropriate school attendance and school absenteeism in need of intervention (Pullen and Kennedy, 2019). However, no consistent, consensus definition for problematic school absenteeism exists across research disciplines or school districts (Gentle-Genitty et al., 2015;Spruyt et al., 2016). Greater consensus can be found with respect to distinguishing Tiers 2 and 3, or identifying at what point school absenteeism is chronic and severe (DePaoli et al., 2015). Researchers, school districts, and other agencies sometimes utilize a 10% absenteeism cutoff to identify chronic absenteeism, though this is somewhat arbitrary and not universal (Conry and Richards, 2018).
Specific data-based demarcations between these tiers remain sparse, despite the fact that such distinctions would help inform early warning systems and intervention assignments for student absenteeism (Chu et al., 2018). Kearney (2016, 2018) found that risk factors for levels of absenteeism at 10% or higher tended to be more restricted than risk factors at lower levels of absenteeism. These studies focused primarily on academic and demographic variables, however, without Frontiers in Psychology | www.frontiersin.org 3 October 2019 | Volume 10 | Article 2381 examining family factors that have been identified as a key correlate of school attendance problems (Dahl, 2016). The present study aimed to identify potential family environment risk factors among youth at different levels of school absenteeism severity (i.e., 1 + %, 3 + %, 5 + %, 10 + %). Participants included students referred for services due to substantial school absenteeism, which allowed for analysis of varying levels of severity. In accordance with recent calls to employ machine learning-based methods to examine risk factors for school absenteeism (Chung and Lee, 2019;Sansone, 2019), two sets of statistical approaches were utilized. Ensemble analysis, including chi-square adjusted interaction detection (CHAID), support vector machines, and neural network analyses, is a nonparametric method that combines multiple algorithmic models or classifiers to produce a single best model for a given data set (Berk, 2006). In addition, classification and regression tree analysis (CART) is a nonparametric method that identifies comprehensive subgroups based on interactions among multiple risk or predictor variables (Lemon et al., 2003). Nonparametric methods are increasingly used for academic variables denoted by categorical levels (e.g., Cordero et al., 2017;Lahti et al., 2019). Various levels of school absenteeism were examined, with a general expectation that risk factors at higher levels of absenteeism would be more restricted than risk factors at lower levels of absenteeism.

Measures
The Family Environment Scale: Form R (FES; Moos and Moos, 2009) is a 90-item true/false measure of current family relationships, personal growth, and family system maintenance.
The FES comprises 10 subscales based on standard scores (mean, 50): cohesion (family member support of one another; COH), expressiveness (encouraging expression of feelings; EXP), conflict (open anger and hostility; CON), independence (self-sufficient, assertive members; IND), achievement orientation (activities cast in a competitive framework; ACH), intellectual-cultural orientation (family interest in intellectual and cultural issues; ICO), active-recreational orientation (participation in recreational/social activities; ARO), moralreligious emphasis (emphasis on ethical and religious values; MRE), organization (clear structure in activities; ORG), and control (set rules and procedures to structure family life; CTL). Internal consistency (Cronbach's alpha) ranges between 0.61 and 0.78. Cronbach's alpha for the items in the present study was 0.72. Two-and four-month test-retest reliabilities range between 0.70 and 0.91 (Moos, 1990). FES item and subscale standard scores (M = 50.0) were utilized as the primary unit of analysis in the present study.
School staff or parents provided absenteeism severity data in the form of number of full school days missed. Percentage of full school days missed was calculated by dividing a student's total number of full school days missed by the number of days of school in that academic year, at the time of assessment, and then multiplying that number by 100.

Procedure and Data Analyses
Participants were recruited from a specialized outpatient therapy clinic or community setting. Participants in the community setting were referred to family court or a truancy diversion program by their school or parent(s)/guardian(s) based on prior school absences. Measures that included the FES were administered to youth and their parent(s)/guardian(s) independently and in the presence of a research assistant. Spanish versions of the measures were available. Study procedures, including parent consent and child assent, were approved by a university institutional review board.
Ensemble analysis was utilized to identify potential family environment risk factors among youth with school attendance problems across different levels of school absenteeism. Ensemble analysis is the combination of multiple algorithmic models or classifiers to produce one, best model that can be applied to the data (Berk, 2006). These models have been shown to outperform standard parametric methods, primarily due to the automation of identifying interactions and non-linearities and reducing overestimations of a model's predictive ability (Rosellini et al., 2018). Ensemble analysis can include many different statistical methods; the present study utilized chi-square adjusted interaction detection (CHAID) decision trees, support vector machines, and neural network analyses. Predictors were examined collectively and independently. A multiple imputation method was utilized; different plausible imputed data sets were examined, and combined results were obtained and reported here. Confusion matrices supported the use of CHAID decision trees as the best approach. In addition, CART analyses were utilized to more specifically examine clusters of FES items associated with enhanced risk for a particular level of absenteeism severity (i.e., 1 + %, 3 + %, 5 + %, 10 + %). Other absenteeism levels were examined on an exploratory basis (i.e., 15 + %, 20 + %, 30 + %, 40 + %). For brevity, significant results are reported.

DISCUSSION
The present study examined family environment variables as potential predictors of various absenteeism severity levels. The findings reveal that several family environment variables are indeed related to different severity levels in both broad and more nuanced ways. Broadly, as expected, family environment risk factors tended to be more circumscribed and informative at higher levels of absenteeism, with much greater diversity at lower levels. Higher levels of absenteeism (i.e., 15 + %) appear more closely related to lower achievement orientation, active-recreational orientation, cohesion, and expressiveness. Lower levels of absenteeism (i.e., 1, 3, and 5%) were generally associated with a wider array of family environment variables. Active-recreational standard scores were generally suppressed across absenteeism severity levels, a result that parallels Hansen et al. 's (1998) finding that less active families were associated with greater levels of school absenteeism among youth with anxiety-based conditions. These authors speculated that a low emphasis on social and physical activities and greater time spent at home may mean that some children may be more apt to spend school time at home. In addition, these children may be more predisposed to have difficulties with social skills and peer interactions that could also interfere with school attendance. Some have also found that school absenteeism is related to less participation in school sports (Hunt and Hopko, 2009), though others have not (Skedgell and Kearney, 2018). Lower activerecreational scores were evident as well in Kearney and Silverman's (1995) study that led those authors to conclude that some families of youth with absentee problems are isolated in nature.
A number of nuanced findings were also revealed in the present study, however, that deserve detailed description. With respect to achievement orientation, for example, elevated standard scores were associated with less absenteeism severity but lower standard scores were associated with greater absenteeism severity. Higher school performance is generally associated with higher competition (Harrison and Rouse, 2014), though effects can depend on gender and age (Little and Garber, 2004;Wang and Holcombe, 2010). At the family level, achievement orientation could translate into specific activities such as modeling academic advancement, reading frequently, encouraging a strong work ethic, and providing enrichment opportunities that distally affect school attendance (Dubow et al., 2009).
In addition, lower standard scores for expressiveness were evident at less severe (3, 5%) and more severe (20, 30%) levels of absenteeism, though elevated standard scores were predictive of 10 + % absenteeism. As noted earlier, Bernstein and Borchardt (1996) found that families of youth with school refusal displayed significant problems with respect to role performance and communication. Findings from the present study indicate that such difficulties may be less evident during periods when families are working together to solve an absentee problem and during periods when frustration over long-term absenteeism has led to greater disengagement and less opportunities for direct expression (Kearney and Silverman, 1995).
Family cohesion represented another nuanced finding. Cohesion was not predictive at 1 + % and 3 + % absenteeism but lower standard scores were more predictive of higher levels of absenteeism. This result parallels Bernstein et al. 's (1999) finding that adolescents with school attendance problems and their parents viewed their families as particularly rigid and disengaged on a cohesion dimension. In addition, several researchers have found, broadly speaking, that parent and family involvement and support are crucial variables with respect to school attendance, performance, and dropout (Sheldon, 2007;Topor et al., 2010;Parr and Bonitz, 2015). Cohesion in the form of help with homework, support for academic progress, and commitment to education may be a key in this regard (Wilder, 2014).
Family conflict was expected to be an important predictor of absenteeism severity in the present study. Elevated conflict standard scores were more predictive of 5 + % absenteeism severity, whereas lower conflict standard scores were more predictive of 10 + % absenteeism severity. Some have found family conflict to be elevated in this population in general, and advocate for the problem to be resolved clinically in this population (Kearney and Silverman, 1995;Kearney and Albano, 2018), though others have found family conflict to be unrelated to school attendance problems (McShane et al., 2001). As with expressiveness, some families may display increased conflict at a point of urgency when trying to resolve a school attendance problem but later become frustrated and disengaged from the process (Kearney, 2019).
Finally, control was a family environment variable that did not appear until higher levels of absenteeism severity. Lower levels of control were more predictive at higher levels of absenteeism severity, particularly at the 20 + % and 30 + % levels. A less structured home environment has been associated with school absenteeism in other studies (Hunt and Hopko, 2009). In addition, as mentioned earlier, Bernstein et al. (1990) found that inconsistency of family rules related to some youth with school attendance problems. Conversely, family rules are part of a parent involvement process often associated with academic success (Catsambis, 2001).
Analyses of individual FES items also revealed interesting findings. First, items were sometimes endorsed differently in different nodes, indicating a high level of variability in these groups. This applied particularly to lower levels of absenteeism. Second, fewer items were predictive of 10 + % absenteeism than at lower levels, mirroring the subscale finding that predictors tended to be more restricted at higher absenteeism severity levels. Overall, however, examining subscale scores appeared to be more useful than examining item scores.
The present study may thus have some applicability to MTSS models of school absenteeism and how tiers within these models may be demarcated. In particular, absenteeism severity levels of 10-15% appear to be associated with more defined sets of risk factors, which may indicate more qualitative changes in family functioning at these levels. More intense drops in achievement orientation, active-recreational orientation, cohesion, and expressiveness, in addition to less conflict, may indicate that families become substantially more disengaged at these levels. Such disengagement could come in the form of sharply reduced parent-school official contact, consequences for school absenteeism, academic assistance, attendance monitoring, and parent supervision (Kearney and Albano, 2018).
The results may also have implications for MTSS development in educational settings. Many local educational agencies, for example, are moving toward systemic, evidence-based systems of academic and behavioral supports to meet the unique needs of diverse students (McIntosh and Goodman, 2016). A better understanding of how these needs intersect with family-based challenges is essential in this respect. Parental involvement, for example, has been found to be a key element of success in MTSS programs, and such programs often benefit from a wider array of stakeholders that include parents (August et al., 2018). In addition, MTSS models are increasingly moving toward a "whole child" approach that more fully considers ecological levels outside of school, such as family factors (Sailor et al., 2018). Results of the present study and related studies may thus help inform such an approach.
Results of the present study also have implications for further research work in this area, particularly with respect to how these findings intersect with other family-based risk factors for school absenteeism. Gubbels et al. (2019), for example, conducted a meta-analytic review of such factors for school absenteeism and dropout and found several pertinent family domains. These included low parental school involvement, lack of nuclear family structure, and low parental control, among others. An understanding of how the family environment dynamics identified in the present study intersect with these broader domains, particularly with respect to specific levels of school absenteeism, would be quite instructive for subtyping and demarcation purposes. Such information may also help inform family-based treatment for this population. For example, Tobias (2019) found that family-based intervention for persistent school absenteeism was often hindered by an insecure home environment. The latter construct could be investigated in greater detail in future work to identify whether the dynamics noted in the present study would apply.
Limitations of the present study should be noted. First, the sample was a diverse one ranging from having no formal school absences to having many school absences. Second, more detailed analyses of absenteeism type or of demographic or developmental differences were not examined in accordance with sample constraints and diversity of settings. Third, the primary dependent measure was based on parent-report. Future researchers should endeavor to explore a more wide-ranging assessment of family functioning in this population.

CONCLUSION
Despite these limitations, findings from the present study may have some clinical implications. Educators, mental health professionals, and others who address these families, particularly at higher levels of absenteeism severity, will likely need to prioritize certain goals given the problematic family dynamics involved. With respect to school attendance, such goals may include repairing parent-school official communications, educating family members about creative educational options, and establishing contracts or agreements to improve problem-solving ability and increase incentives for attending school (Kearney, 2019). More broadly, such goals may include interventions to enhance family engagement and communication as well as contacts with outside sources of support (Kelly et al., 2018).

DATA AVAILABILITY STATEMENT
The datasets generated for this study are available on request to the corresponding author.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by UNLV IRB. Written informed consent to participate in this study was provided by the participants' legal guardian/ next of kin.

AUTHOR CONTRIBUTIONS
All authors revised and approved the submitted version. MF helped collect data, performed the initial analyses, and assisted in the writing of the manuscript. CK helped with data analysis, assisted in the writing of the manuscript, and supervised the study.

FUNDING
This work was supported by the University of Nevada, Las Vegas. A portion of the publication fees for this article were supported by the UNLV University Libraries Open Article Fund.