TYPE Original Research PUBLISHED 01 June 2023 DOI 10.3389/fpsyg.2023.1163088 Dropout in youth and adult OPEN ACCESS education: a multilevel analysis of EDITED BY Carolyn Gentle-Genitty, Indiana University Bloomington, United States students and schools in Chile REVIEWED BY Sotiris Kotsiantis, Tabata Contreras-Villalobos1,2 , Verónica López1,3 , University of Patras, Greece Enrique Baleriola4* and Luis González3 Cleveland Hayes, Indiana University–Purdue University 1 Escuela de Psicología, Facultad de Filosofía y Educación, Pontificia Universidad Católica de Valparaíso, Indianapolis, United States Valparaíso, Chile, 2 Departamento de Psicología Básica, Evolutiva y de la Educación, Universitat David Kondrat, Autònoma de Barcelona, Barcelona, Spain, 3 Centro de Investigación para la Educación Inclusiva, Indiana University–Purdue University Viña del Mar, Chile, 4 Estudis de Psicologia i Ciències de l’Educació, Universitat Oberta de Catalunya, Indianapolis, United States Barcelona, Spain *CORRESPONDENCE Enrique Baleriola ebaleriola@uoc.edu Introduction: School dropout is an increasing worldwide phenomenon, marked RECEIVED 10February 2023 by inequality and educational exclusion. In Chile, many students who have ACCEPTED 16 May 2023 dropped out of regular schools attempt to reenter youth and adult education PUBLISHED 01 June 2023 (YAE). However, some of them drop out again from YAE. CITATION Contreras-Villalobos T, López V, Baleriola E Objective: The aim of this study was to identify and jointly analyze school and and González L (2023) Dropout in youth and adult education: a multilevel analysis individual factors that influence dropout in YAE. of students and schools in Chile. Front. Psychol. 14:1163088. Methods: This secondary multilevel analysis of official datasets from Chile’s doi: 10.3389/fpsyg.2023.1163088 Ministry of Education focused on students enrolled in YAE (N = 10,130). COPYRIGHT © 2023 Contreras-Villalobos, López, Baleriola Results: According to the findings, YAE dropout can be explained by the individual and González. This is an open-access article risk factors of age (19–24 years), low academic achievement, and school-level distributed under the terms of the Creative Commons Attribution License (CC BY). The factors such as number of teachers (raw and student-to-teacher ratio), economic use, distribution or reproduction in other resources, and school management quality. forums is permitted, provided the original author(s) and the copyright owner(s) are Discussion: We discuss the need to develop school-level protective factors that credited and that the original publication in this build connections, foster student engagement, and ultimately, promote students’ journal is cited, in accordance with accepted academic practice. No use, distribution or permanence and progress in YAE. reproduction is permitted which does not comply with these terms. KEYWORDS school dropout, school abandonment, youth and adult education, multilevel analysis, Chile 1. Introduction Recent years have seen international organizations call for education policies that improve the quality, equity, inclusiveness, and fairness of teaching and school processes (UNESCO, 2015). However, many gaps remain to overcome, such as access to education and school completion, regardless of students’ educational pathways (OECD, 2012; de Cristo, 2019). In this regard, more than 12 million children, young people, and adults are excluded from education worldwide (UNESCO, 2020). This issue exacerbates the education crisis, translating into exponential growth in school dropout rates (EduCo, 2021). School dropout, or leaving school, is broadly defined as people who discontinue their studies (European Commission, 2003). It is used as an indicator to measure the relationship between academic progress, age, and education level. Students who are unable to keep up may leave the education system, contributing to the school dropout rate (Camacho, 2016). School dropout is not only an indicator; it also has negative and far-reaching effects on people’s lives in terms of finances, employment, mental health, wellbeing, cohesion, and Frontiers in Psychology 01 frontiersin.org Contreras-Villalobos et al. 10.3389/fpsyg.2023.1163088 social engagement (Bae, 2020; Andrew and Blake, 2021). Thus, in YAE must be addressed, not only to provide an opportunity school dropout is regarded as the outcome of a process marked by to complete compulsory education, but also to promote more school failure, demotivation, and disengagement with school and comprehensive personal and social development, including non- the education system (Jabbari and Johnson, 2021). Its precursors cognitive or relational aspects (Nada et al., 2020; Paniagua, 2022). are absenteeism, low performance, failure, grade repetition, and In Chile, YAE falls under the oversight of the Ministry of educational backwardness (Jurado and Tejada, 2019). Hence, Education. It was originally created to give adults an officially school dropout is a major cause of inequality and social exclusion recognized way to pursue their education. However, due to and one of the most entrenched phenomena in education systems, cultural, social, and economic shifts, it is currently used more affecting millions of vulnerable young and adult people (Lee-St by young people, 75% of whom are between 15 and 24 years of et al., 2018; Tarabini, 2020; Contreras-Villalobos et al., 2022). age (MINEDUC, 2017), a phenomenon that is seen worldwide Many of those who drop out of school graduate later than (Novella, 2020). usual (OECD, 2018). They often complete their studies in another Such institutions are internationally regarded as high- educational setting, namely, youth and adult education (YAE) complexity schools, given the high-risk social environments their programs. The institutions offering these programs are essential students face (Murillo and Duk, 2020). However, the students because they strive to get people—mostly those younger than are not the only ones coping with adverse circumstances: The 25 years of age—back into school so that they can complete their institutions themselves often encounter complex conditions compulsory education. However, despite their good intentions, that lead to an exclusionary organizational environment, scant YAE schools confront another challenge with respect to school resources, and frequent reliance on self-management (Opazo, abandonment. Chile is a paradigmatic case because although 1.7% 2017). In Chile, YAE institutions’ capacity for action is limited due of students drop out of regular education, this figure rises to 24% to the short supply of basic funding and lack of an education policy in YAE (MINEDUC, 2020). In other words, almost 34,000 people that might otherwise provide additional grants. These schools do drop out again each year after having reentered the school system not currently have enough fixed resources to plan and execute via YAE without receiving further intervention or support to finish effective responses to the needs of their students (MINEDUC, their compulsory education. 2016; Rodríguez, 2019). The low-coverage, unsustainable, swing- This means that between 5.2 and 8.9% of the population budget project funding model and attendance-based grant system between the ages of 6 and 21 years old who once entered the currently in place further challenge public education (Bellei, 2015). education system have dropped out and have not since returned Youth and adult education policies are currently designed (de Cristo, 2019). The inexactitude of these figures is a consequence around high-stakes testing, privatization, and demand-side grants of the disengagement and invisibility of YAE students, which are that determine the resources schools must have to operate, further a result of the existing segregation of the most vulnerable groups deepening inequality in this type of education (Author; Mikulec, (Verger, 2019). This scenario has become even more complex in 2021). This model affects school management, creating new the wake of the COVID-19 pandemic, given official figures estimate problems and challenges for both institutions and their students. that school dropout rates will increase by 200% (MINEDUC, 2020). Hence, strategic planning is needed to comprehensively monitor School dropout in the context of YAE is even more complex, and assess these institutions to understand and manage the work they do to prevent students from dropping out (Grinberg, 2006; with higher figures than in mainstream education. So, in this Quiroga, 2017). context, two questions arise: What variables are decisive in At the school level, research on school dropout has shown understanding school dropout in YAE? What actions can be that school management is key in tackling this problem (Mduma deployed to intervene in this phenomenon? To answer these et al., 2019; Contreras-Villalobos et al., 2022). Certain factors at questions, we carried out a multilevel exploration of the underlying the institutional level can help keep students in school, including factors to identify lines of action to address this situation policies and legal regulations, resources (Alvariño et al., 2000), and (Bronfenbrenner, 1992). Our research objective, therefore, was to school governance that determine how schools are organized and identify and collectively analyze individual and school factors that managed (Ball, 2000). affect school dropout in Chile’s YAE programs. Through this line In turn, school-level management determines educational of study, we generated recommendations on how to address this practices and nurtures environments that benefit learning and problem, which affects not only young people but also their social help keep students on their educational pathway, albeit not environments. always with the same level of success (Bolívar, 2019). Important factors include physical arrangements (Veiga-Neto and Noguera, 2010), the availability of learning means and resources, the size 1.1. Managing school dropout in YAE of the school, the number of people it can accommodate, the opportunities it provides for gathering and participation, and School dropout requires urgent attention at different levels. even the educational practices, including whether they enable Action is needed at the political level, given its direct consequences personalized work that emphasizes recognition and representation on young people and adults. These effects include reduced social (Belavi and Murillo, 2020). cohesion, less engagement and civic participation, and threats to However, in recent years, the scientific literature has identified their educational trajectories, entry into the labor market, and certain factors that are complex to establish or measure yet are key economic development. These threats exacerbate inequalities and in the school’s management of dropout. Among these factors is social vulnerability, affecting students’ pursuits in life (Causton- the bond between teachers and students as a protective element, Theoharis et al., 2011; UNESCO, 2017). In this way, school dropout whereby pedagogical work, communication, and individualized Frontiers in Psychology 02 frontiersin.org Contreras-Villalobos et al. 10.3389/fpsyg.2023.1163088 work generate transformations in educational trajectories (Farmer 2.2. Instruments and variables et al., 2018). This situation is highlighted in alternative schools, where new opportunities for learning are generated, providing We consulted official public datasets from the Ministry of support to students at risk of school dropout and shaping new Education, available in the open data section of its online study experiences in the school environment (Wilkerson et al., 2016). center. Using this platform, we requested four datasets for 2018: At the individual level, indicators of disengagement between (a) performance summary, (b) enrollment summary by school, (c) students and the institution should be addressed if dropout is to national performance evaluation system, and (d) school grants. be tackled by school management. Among the most frequently Table 1 identifies the datasets used to derive certain information mentioned in the literature are poor attendance, grade repetition, and how variables were operationalized for the two levels of our perceived school failure, financial struggles, family burdens, analysis: students (level 1) and schools (level 2; Woltman et al., behavioral difficulties, certain geographical contexts, and drug use 2012). Based on the information available and previous empirical (Opazo, 2017; Tarabini, 2017; Rodríguez, 2019; Valenzuela et al., studies, we hypothesized that school dropout would be explained 2019). at the individual level by gender, age, and grade point average However, the complex nature of school dropout demands (GPA). This implies that young men (≤19 years old) with low an integrative perspective that analyzes educational management academic performance would have the highest risk of dropping across all levels. It is impossible to consider the impact of any out a second time during YAE. At the school level, we expected of these dimensions or factors in isolation because how schools that depersonalized work, low school resources, and hierarchical are run is linked to macro-level social, cultural, and political and standardized school management would make the highest contexts and the micro-level circumstances of individual students contribution to explaining school dropout in YAE. and their immediate environment (Bronfenbrenner, 1992; Castro, 2008). Hence, a comprehensive, structural approach is needed to analyze school dropout in YAE because it allows moving away from reductionist, instrumental, or mechanistic interpretations to 2.3. Procedure provide insightful recommendations for transforming school and public policies. For this reason, this research explored factors After retrieving these datasets, we identified relevant variables related to school dropout in YAE at the school and individual levels from the codebook provided. As a starting point for the merging to determine their attributed variability. This allowed us to identify of databases, we used the performance summary dataset. This lines of action for the transformation and improvement of school was merged with the enrollment summary by the school dataset dropout. through an encrypted code assigned to each student. This dataset has information on individual characteristics and the school in which each student was enrolled in 2018. To clean the resultant 2. Materials and methods database, we dropped observations that had missing information on individual characteristics or where a student appeared more To accomplish the explanatory objective of this research than once (e.g., because the student had transferred schools). and address the hierarchical nature of the data, we performed The resultant sample featured 10,356 students. Then, we added a multilevel analysis, an extension of linear regression models students’ school information by merging the national performance that recognizes and accounts for the nested nature of data. evaluation system and school grants datasets with the previous This type of analysis allowed us to ascertain the proportion of merged dataset using school identification numbers. We also variance explained by differences between cases (students) in a recoded certain variables, grouping education levels and age by certain context (schools)—known as within-school variance—and school, to simplify our analyses. During this process, we cleaned compare it with the proportion of variance explained by differences our data of invalid cases with missing school information. The final among contexts (schools)—known as between-school variance sample consisted of 10,130 students. (Martínez-Garrido and Murillo, 2013). During this process, we cleansed our data of invalid cases where student information on grades and attendance was missing or where information appeared more than once (e.g., because the 2.1. Participants and materials student had transferred schools). In all, we discarded 62% of our original sample (n = 16,298). The analysis was conducted with a national sample of 10,130 Table 2 compares the means of the study variables between the students (51% men; age: M = 24.2 years, SD = 9.6) from original and final samples. The statistical significance of the mean 179 schools in 11 regions in Chile. These 179 schools had differences was estimated through Student’s t-test. There were an average of 56.6 students enrolled (SD = 72.5; Min = 3; statistically significant differences between samples in all variables Max = 592), where 88.3% were public municipal schools, 10.0% but GPA, the percentage of students in high school (technical– were subsidized private schools, and 1.7% were public education professional), percentage of students in municipal and partially local services. Although the Ministry of Education’s (MINEDUC, subsidized schools, percentage of students with a 60% grant, and 2020) records indicate more students enrolled in YAE nationwide the achievement score of the students’ schools. The final sample was (n = 139,581), information is not available or complete on many composed of students with a lower dropout rate, a lower proportion of them. Therefore, the sample only included students for which of men students, and lower average age. This loss of sample data complete information had been provided by schools to the central due to the process of merging datasets increases bias, and therefore government. findings should be regarded as only representative of this sample. Frontiers in Psychology 03 frontiersin.org Contreras-Villalobos et al. 10.3389/fpsyg.2023.1163088 TABLE 1 Operationalization of variables. Level Variable Operationalization Data source Type of measure Response options Individual School dropout Understood as a student’s status at the end of Performance summary, 2018 Nominal Passed; failed; dropped out (abandonment) the 2018 school year, i.e., whether they completed the year or withdrew and did not enroll at another school during the same year GPA Student’s overall grade point average at the end Performance summary, 2018 Continuous 1–7 of the school year Gender Identifies a person as man or woman Enrollment summary by Nominal 0 = male; 1 = female according to patterns, as indicated by the school, 2018 student at enrollment Age Age of the student in years as of June 30, 2018 Enrollment summary by Continuous 13–82 school, 2018 Education level Education level in which the student is Enrollment summary by Nominal 1 = elementary school; enrolled; could be elementary or high school in school, 2018 2 = high school science–humanities or technical–professional (science–humanities); 3 = high school (technical–professional) School Schedule Time of day that students attend school Enrollment summary by Nominal 1 = morning; 2 = afternoon; (structure) school, 2018 3 = morning and afternoon; 4 = evening and night Average no. of Average number of students per classroom Enrollment summary by Continuous 0–44 students school, 2018 No. of teachers Total number of teachers at the school during National performance Continuous 2–120 the selected year evaluation system, 2018–2019 Student–teacher Number of students divided by the number of National performance Continuous 4–27 ratio teachers evaluation system, 2018–2019 School Additional Amount of additional financial aid provided School grants, 2018 Ordinal 0 = no additional grant; (management) quality-based due to low performance 60 = additional 60% grant; grants 100 = additional 100% grant Effectiveness The educational outcome achieved by the National performance Continuous school 0–100 school in relation to its student sample evaluation system, 2018–2019 self-reporting through a standardized evaluation card Achievement The school’s educational achievement National performance Continuous school 0–100 differentials over time evaluation system, 2018–2019 self-reporting through a standardized evaluation card Innovation The school’s capacity to innovate in education National performance Continuous school 0–100 and engage the support of external partners for evaluation system, 2018–2019 self-reporting through a its pedagogical work standardized evaluation card Improvement The improvement of working conditions and National performance Continuous school 0–100 the proper running of the school evaluation system, 2018–2019 self-reporting through a standardized evaluation card Integration Integration and participation of teachers, National performance Continuous school 0–100 parents, and guardians in realizing the school’s evaluation system, 2018–2019 self-reporting through a educational goals standardized evaluation card Equality Equality in terms of the student sample being National performance Continuous school 0–100 able to access and remain in the school and evaluation system, 2018–2019 self-reporting through a integration of groups with learning disabilities standardized evaluation card 2.4. Analysis plan 0 = completed). Model 0 was the intercept-only model. Model 1 included the individual characteristics (level 1). Model 2 included We performed our analyses using Stata 15 statistical software. individual variables and school-level variables (level 2), accounting for schools’ structural attributes. Finally, Model 3 included all As part of our descriptive analysis, we carried out a frequency variables from the first two models and the school-level variables analysis to further characterize this educational sample at the characterizing school management. individual and school levels. We also performed a t-test to draw a mean comparison between students who did not complete the 2018 school year (dropped out) and those who did (passed or failed). 2.5. Ethical considerations Then, we conducted a multilevel logistic analysis, constructing four models based on the two levels. The dependent variable for this For this study, we requested information through the analysis was the dichotomous dropout variable (1 = dropped out, mechanism provided by the Chilean Government’s Transparency Frontiers in Psychology 04 frontiersin.org Contreras-Villalobos et al. 10.3389/fpsyg.2023.1163088 TABLE 2 Differences between initial sample and final sample. Original sample (n = 16,298) Final sample (n = 10,130) Difference Variables Mean (or %) SD Mean (or %) SD Dropout 31.2 0.46 6.4 0.24 24.8*** Gender (Man = 1) 58.7 0.49 50.8 0.49 7.9*** Age 25.7 10.7 24.3 9.6 1.4*** GPA 5.5 0.5 5.5 0.5 0.0 Education level Elementary school 18.0 0.38 0.0 0.02 18.0*** High school (humanities–science) 71.6 0.45 90.0 0.30 −18.4*** High school (technical–professional) 10.5 30.6 10.0 0.30 0.5 Schedule Mornings 11.8 0.32 13.0 0.33 −1.2*** Afternoons 9.8 0.30 8.0 0.28 1.8*** Mornings and afternoons 8.4 0.28 3.0 0.17 5.4*** Evenings and nights 70.0 0.46 76.0 0.43 −6.0*** Administrative dependence Municipal 89.1 0.31 89.0 0.31 0.1 Partially subsidized 10.4 0.31 10.0 0.30 0.4 Local public education service 0.5 0.07 1.0 0.08 −0.5* Rural school (Yes = 1) 9.8 0.29 6.0 0.24 3.8*** *p < 0.05, ***p < 0.001. The t-tests were carried out to estimate the difference between the original and final sample. Law 20.285 (2015), ensuring that it was only used for scientific Hence, to characterize our sample, students who dropped out purposes and safeguarding the integrity and anonymity of the of the education system were more likely to be women, young schools, their students, and their performance indicators. We later people, and people with low academic achievement. This occurred consulted data provided by the Ministry of Education through its at a significantly higher rate among students who attended online study center, an open data website with detailed information technical–professional high schools. on Chile’s education system with free access to various datasets. Regarding the schools’ structural attributes, a Student’s t-test In both the analyses and presentation of results, attention was showed that students who attended school in the evening and given to maintaining and safeguarding the privacy of these data. It at night dropped out more frequently (p < 0.05). Likewise, the should be noted that the researchers did not have access to data that results indicate that a significantly smaller proportion of students could have allowed them to identify individual students because the attending school in the morning dropped out a second time datasets are encrypted. Although it was possible to merge the data (p < 0.001). using student identifier codes, this is an encryption of personal data There were no significant differences for the students’ and is therefore not in the public domain. enrollment in neither type of administrative dependence nor school location. There was a difference, however, with respect to the number of teachers per school, whereby the average was higher among students who left school again. In other words, students 3. Results were more likely to drop out of schools that had more teachers (p < 0.001). Table 3 displays the descriptive results of the variables In terms of school management, it is worth noting that very we studied, showing means for the continuous variables and few YAE institutions received additional quality-based grants, at percentages for the dichotomous variables. Regarding the neither 60% nor 100%. A Student’s t-test revealed significant mean individual variables, we found several significant differences differences in this respect (p < 0.05), indicating that people who between students who left school and those who did not, based on dropped out tended to be enrolled in schools that did not receive their own attributes. In terms of gender, more women (p < 0.01) additional funding. Regarding performance data reported by the dropped out than men. Students who dropped out also had a lower National Performance Evaluation Service, which were based on mean age (p < 0.001) and grade point average (p < 0.001) than evidence-based self-reports submitted by school principals, the those who remained in school. Finally, looking at education level, results showed significant differences in school dropout based on those who had a technical–professional high school education effectiveness, achievement, and improvement (p < 0.05). Indeed, were more likely to drop out (p < 0.001). The mean comparison the higher these values were for schools, the less likely students tests showed that these differences were indeed significant. were to drop out. Frontiers in Psychology 05 frontiersin.org Contreras-Villalobos et al. 10.3389/fpsyg.2023.1163088 TABLE 3 Descriptive data and differences according to dropout rate and t-test. Not drop out (n = 9,482) Dropped out (n = 648) Total M SD M SD M SD Difference Individual Gender 0.49 0.50 0.44 0.50 0.51 0.49 0.06** Age 24.4 9.8 22.6 7.1 24.3 9.6 1.7*** GPA 5.5 0.5 5.2 0.4 5.5 0.5 0.3*** Education level Elementary school 0.00 0.01 0.00 0.04 0.00 0.02 0.00 High school (humanities–science) 0.91 0.29 0.85 0.36 0.90 0.30 0.06*** High school (technical–professional) 0.90 0.29 0.15 0.36 0.10 0.30 −0.06*** School (structure) Schedule Mornings 0.13 0.34 0.08 0.27 0.13 0.33 0.05*** Afternoons 0.08 0.28 0.09 0.28 0.08 0.28 −0.01 Mornings and afternoons 0.03 0.17 0.04 0.19 0.03 0.17 −0.01 Evenings and nights 0.76 0.43 0.79 0.41 0.76 0.43 −0.03* Administrative dependence Municipal 0.89 0.31 0.90 0.30 0.89 0.31 −0.01 Partially subsidized 0.10 0.30 0.09 0.29 0.10 0.30 0.01 Local public education service 0.01 0.08 0.00 0.07 0.01 0.08 0.01 Location Rural 0.06 0.24 0.05 0.22 0.06 0.24 0.01 Urban 0.94 0.24 0.95 0.22 0.94 0.24 −0.01 Average no. of students 23.5 6.7 23.9 7.3 23.5 6.7 −0.4 No. of teachers 46.7 26.7 50.6 30.6 46.9 27.0 −3.9*** Student–teacher ratio 12.3 4.0 12.2 4.4 12.3 4 0.4 School (management) Additional grants 100% grant 1.9 0.3 2.0 0.2 1.9 0.3 0.0* 60% grant 2.0 0.1 2.0 0.0 2.0 0.1 0.0* Performance Effectiveness 42.4 6.3 41.7 6.9 42.3 6.4 0.6* Achievement 62.3 3.9 61.9 3.7 62.3 3.9 0.4* Innovation 55.0 43.0 55.8 42.8 55.1 42.9 −0.8 Improvement 95.0 10.3 94.0 12.1 94.9 10.4 1.0* Integration 45.0 37.4 45.4 36.9 45.0 37.3 −0.4 Equality 87.1 9.7 87.7 9.8 87.1 9.8 −0.6 The right-most column shows the difference between the means of those who did not drop out and those who did; t-tests were carried out to show if differences between those who dropped out and those who did not were statistically significant. *p < 0.05, **p < 0.01, ***p < 0.001. Table 4 shows the results of the estimation of the multilevel appeared to affect dropout rates (OR = 0.31, p < 0.001); specifically, logistic regression. The dependent variable used in the estimation the lower the GPA, the higher the likelihood of dropping out. was the dropout status of students, controlling for individual and For Model 1, the null hypothesis was rejected for age and GPA school factors. According to Model 1, which included the individual because students between 22 and 30 years of age and those with variables, one age group was more likely to drop out of school again low grades were more likely to drop out of school again. The null compared to students between 13 and 21 years old: those between hypothesis was accepted for gender, however, because there were the ages of 22 and 30 (OR = 1.27, p ≤ 0.05). Grade point average also no significant differences in this respect in the sample analyzed. Frontiers in Psychology 06 frontiersin.org Contreras-Villalobos et al. 10.3389/fpsyg.2023.1163088 TABLE 4 Results of multilevel logistic regression of dropout based on individual and school-level variables (N = 10,130 students at 179 schools). Model 0 Model 1 Model 2 Model 3 OR SD OR SD OR SD OR SD Individual Gender 1.05 0.09 1.04 0.09 1.04 0.08 Age range (ref: 13–21 years) 22–30 years 1.27* 0.15 1.28* 0.14 1.29* 0.15 31–40 years 0.88 0.16 0.88 0.15 0.90 0.16 41 years or older 0.74 0.17 0.74 0.17 0.74 0.17 GPA 0.31*** 0.03 0.29*** 0.03 0.29*** 0.03 School Average no. of students 1.01 0.02 1.00 0.01 No. of teachers 1.00 0.00 1.00 0.00 Enrollment 0.99 0.00 0.99 0.00 Student–teacher ratio 20.6*** 19.2 Administrative dependence (ref: municipal) Partially subsidized 0.68 0.20 0.68 0.20 Local public education service 1.62 0.75 1.17 0.59 Education level (ref: elementary school) High school (humanities–science) 0.10* 0.12 0.10* 0.11 High school (technical–professional) 0.24 0.29 0.20 0.25 Rural school (1 = yes) 1.08 0.30 1.00 0.28 Schedule (ref: morning) Afternoon 1.74** 0.39 1.75** 0.39 Morning and afternoon 1.08 0.44 1.23 0.51 Evening and night 1.52* 0.26 1.49* 0.29 Additional quality-based grants No 100% grant 1.63** 0.32 1.31 0.38 No 60% grant 12.02*** 8.68 10.67*** 7.94 Performance Effectiveness 1.00 0.01 Achievement 0.96** 0.01 Innovation 1.01 0.00 Improvement 1.01 0.00 Integration 0.98* 0.00 Equality 1.01 0.00 Constant 0.06*** (0.00) 30.82*** 16.43 8.94 13.30 17.75 38.00 Variance partitioning coefficient School level 0.111 0.108 0.086 0.079 Log-likelihood −2,360.58 −2,269.01 −2,248.11 −2,243.22 *p < 0.05, **p < 0.01, ***p < 0.001. In this line, the data may not have had enough power to detect no significant differences in terms of student enrollment, this association. average number of students per classroom, or number of Model 2 included the individual variables and the variables teachers. However, a difference emerged for the student– relating to school structure. It should be noted that adding teacher ratio. Indeed, the higher the ratio, the more likely these variables did not change the significance of the individual students were to drop out of school again (OR = 19.96, variables noted in the previous model. Thus, Model 2 showed p < 0.001). Frontiers in Psychology 07 frontiersin.org Contreras-Villalobos et al. 10.3389/fpsyg.2023.1163088 Regarding students’ school schedule, those who attended in the the literature has reported that men drop out the most, our afternoon (OR = 1.78, p ≤ 0.01) and those who attended in the analysis suggested that women were more likely to drop out of evening and at night (OR = 1.52, p ≤ 0.05) were significantly more school again in the context of YAE. Although this variable was likely to drop out of school than those who attended in the morning. no longer significant in Model 2, our results are nevertheless The model also showed higher dropout rates in schools without consistent with previous research showing that young women from an additional 100% grant (OR = 1.63, p ≤ 0.01) and those without vulnerable communities are sometimes more likely to have their an additional 60% grant (OR = 12.02, p < 0.001). In other words, educational journeys cut short (Hubert et al., 2019). This can schools that lacked these additional resources were more conducive be largely explained by teenage pregnancy, but research has also to students dropping out. identified financial instability, various types of violence, and lack Finally, we added the variables relating to institutional of backing from institutions as additional factors (Vázquez-Nava management to produce Model 3. From this model, we ascertained et al., 2019). Support for these women should be sought from a whether school performance explained dropout, per the National gender perspective, with a focus on keeping them in school and, Performance Evaluation System criterion. It should be noted that in so doing, closing the significant gap in this respect (Banda et al., when the management variables were added, the odds’ ratio for 2019). whether schools had a 100% grant was no longer significant. As Our results also indicate that the highest dropout rates and for the management variables, achievement (OR = 0.96, p ≤ 0.01) poorest school performance in YAE occurred among young people. and integration (OR = 0.99, p ≤ 0.05) predicted dropout. In this These data suggest that significant efforts are required to meet the regard, students were less likely to drop out of schools with higher needs of these young people, who continue to be outsiders in school educational achievement differentials over time and those with and, thus, have dropped out a second time. To delve further into higher levels of teacher integration and participation. this matter, syllabus design must be analyzed. In addition, teaching In summary, school variables such as when students attended methodologies and assessment models should be examined to class, the grants that schools received, educational achievement, identify factors that need to be addressed to support these students and teacher integration were shown to affect the likelihood and promote learning and participation (Allison and Attisha, 2019; of dropout. The null hypothesis was rejected because no Låg and Sæle, 2019). significant differences were found for variables such as the average At the school level, the management, and accountability number of students, number of teachers, and other management- indicators of achievement and integration could have a significant related factors. impact on school dropout. As these indicators increased, the In order to calculate the variance partition coefficient (VPC), likelihood of YAE students dropping out of school again decreased. we used Snijders and Bosker (2012) approach known as the latent The achievement indicator reflects the schools’ educational variable method. The VPC allows quantifying the proportion of the achievements over time and, thus, can be taken as a proxy for total variance that is attributable to the different levels in the model academic performance. Students are more likely to complete their for a given set of variables (Goldstein et al., 2002). compulsory education in high-performing schools. Contrarily, The VPC for Model 0 (intercept-only model) showed that students have a tougher time continuing their education in schools 88.89% of the variance in students dropping out of YAE was that do not do so well in this respect, thus perpetuating school explained by individual variables, whereas 11.11% of the variance exclusion and negative experiences (Warne et al., 2020). was explained by school-related attributes. In Model 3, the VPC Our findings regarding school integration are also noteworthy. for level 2 (school-level) dropped to 7.9%, showing the relevance This indicator reflects teacher participation and opportunities for of individual characteristics. Therefore, we found support for parents, guardians, and students to lead and participate. As such, the hypothesis that variance in school dropout would be more it considers multiple stakeholders in the school community. This management indicator also measures educational commitment, greatly explained by individual factors than by differences between based on the type and number of activities organized (Lee YAE institutions. and Desjardins, 2019). Therefore, it is a fairly broad indicator, accounting for multiple practices in the school environment that may diminish the likelihood of students dropping out. Considering 4. Discussion our findings, we can identify strategies that address dropout from a relational perspective in the school environment (Itzhaki, 2019). Returning to school after having left regular education by These strategies should take a systemic, inclusive approach that entering a YAE school does not guarantee graduating and, thus, emphasizes the construction of democratic and participatory spaces finishing the compulsory school trajectory. In fact, dropping out (De la Cruz Flores and Ortega, 2019; Portela et al., 2019). of a YAE school constitutes, for these students, a second school In terms of school management, it is evident that more dropout experience that ultimately hinders their “last chance” resources are needed to support learning processes because this possibility of graduating from high school. In this vein, this study could reduce the number of students who discontinue their sought to analyze individual and school-level factors that, together, schooling. These resources should include physical improvements contribute to explaining why students who have already dropped to school facilities and more support for students in terms of out (at least once) of school and are willing to try again in a food, transportation, and educational materials, thus encouraging YAE school may drop out of YAE too, increasing the risk of not and helping them remain in the education system. However, an completing their formal education. enormous risk for YAE school dropout is presented by the effects Our findings show that in terms of students’ individual of distance learning education that has been carried out in most attributes, there are gender differences in school dropout. Although YAE schools in Chile during the COVID-19 pandemic. Unlike Frontiers in Psychology 08 frontiersin.org Contreras-Villalobos et al. 10.3389/fpsyg.2023.1163088 regular K-12 schools, which have progressively advanced toward available resources, and how the school environment is managed. in-person learning with safety measures, YAE schools have been By addressing these issues, recognizing the needs of students, and physically closed for almost 2 years, with students receiving guides giving them the right support, we can change their lives. Not and materials that they must complete alone at home. This likely only will they receive their right to education, but they will also has had a large effect on the relational perspective (Itzhaki, 2019) gain new opportunities to shape their future and realize their life that young, lower-achieving students in YAE schools need to engage goals. and keep studying in these schools. Even so, many obstacles remain that push these young adult At the school level, the teacher–student ratio was found to and adult students to drop out of school again. To reverse be a determining factor in school dropout. The literature has this situation, we propose searching for other non-privatizing already noted that personalized work helps keep students in school, models, which have dominated Chile’s education system for over given the social support that they perceive from their teachers forty years as a result of neoliberal logic, affecting the most in such cases. Importantly, the climate and relationships that vulnerable students, such as those in YAE. These educational this approach creates translate into greater school commitment models should focus on gender differences, context-specific (Terrenghi et al., 2019) and transform the educational experience, curricular and assessment plans, and school management that promoting students’ continued engagement and keeping them promotes inclusivity, democratic school experiences, and student from dropping out. engagement. Considering these variables from a wellbeing and Following the latent variable method used by Snijders and social justice framework, the consequences and effects of school Bosker (2012), the variance partition coefficient (VPC) was dropout on young and adult people could be modified, configuring estimated in order to quantify the total variation attributable to new educational trajectories. differences between schools. The VPC in the estimated models Among the limitations of this study, although the sample had values in the range of [0.079–0.108]—in other words, between was national, the dataset analyzed represented a subsection of 7.9 and 11.1% of the total variation could be attributable to the total universe of students enrolled in YAE schools in Chile. level 2 differences, showing that a high proportion of the The final sample analyzed also differed in the proportion of unexplained variance is related to the differences between students’ men students, ages, and grades at the individual level, and the characteristics within schools. students’ schools differed in structural characteristics from the Because individual factors made the highest contribution to original sample. The datasets analyzed did not include information explaining school dropout in this national sample of students, there regarding student engagement, participation, or motivation toward is a risk of putting all the responsibility on students. However, learning. Another important limitation to address in follow-up as stated previously, the analysis and possible lines of action and studies is the lack of monitoring and updating of the YAE datasets intervention cannot be reduced to one level, given the complex since the COVID-19 pandemic. It is due to this that it has nature of school dropout. Research on school improvement has not been possible to carry out a cross-over study. Therefore, we shown that schools can and do make a difference when they suggest that future studies include a cross-sectional study over provide opportunities for inclusion, learning, and participation time and broaden the scope of analysis, data, and information for all students (Ainscow, 2015). The findings from this study of the experience of leaving school, such as the educational and suggest that school interventions and public policies based on socioemotional effect, for comprehending the factors involved in school improvement and educational inclusion may strengthen YAE school dropout. Finally, although this study contributes to the YAE school experience such that student characteristics may the existing scientific evidence by incorporating contextual, school- have lower predictive power, and in contrast, school characteristics, related attributes that contribute to explaining school dropout in particularly school management practices, may play a greater the YAE educational setting, future studies should complement role in preventing school dropout. To foster this transformation, these analyses by incorporating other algorithmic and machine researchers and educational communities should focus on both learning approaches. the strengthening and visibility of YAE, as well as on facilitating evidence-based decisions that consider the situated indicators that predict YAE student dropout (Rodríguez et al., 2023). Summing up, school dropout leads to exclusion and has direct Data availability statement repercussions on the educational attainment, life projects, and subjectivities of the students affected. Although attention is often The original contributions presented in this study are included placed on the mainstream education system, school dropout also in the article/supplementary material, further inquiries can be occurs in educational settings for students that have already left directed to the corresponding author. school at least once. Returning to the questions that guided this study—namely, what variables are decisive in understanding early school leaving in YAE and what actions can be deployed to intervene in this Ethics statement phenomenon—we can conclude that school dropout in YAE can only be adequately explained by considering individual, school, The studies involving human participants were reviewed and and social factors jointly (Prenkaj et al., 2020). Our results approved by the Pontificia Universidad Católica de Valparaíso. reveal individual risk factors such as age and low grades and Written informed consent for participation was not required for protective factors that help to build connections and foster student this study in accordance with the national legislation and the engagement, such as the teacher–student ratio at the school, institutional requirements. Frontiers in Psychology 09 frontiersin.org Contreras-Villalobos et al. 10.3389/fpsyg.2023.1163088 Author contributions Conflict of interest TC-V: conceptualization, formal analysis, methodology, The authors declare that the research was conducted writing—original draft, and writing—review and editing. VL: in the absence of any commercial or financial relationships conceptualization, writing—review and editing, supervision, and that could be construed as a potential conflict of project administration. EB: methodology, writing—original draft, interest. and writing—review and editing. LG: methodology and formal analysis. All authors contributed to the manuscript revision, read, and approved the submitted version. Publisher’s note All claims expressed in this article are solely those Funding of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, Funding was obtained from SCIA ANID CIE160009, the editors and the reviewers. Any product that may be FONDECYT 1191267, Beca de Doctorado Nacional Anid-Chile, evaluated in this article, or claim that may be made by and Spain Ministry of Science and Innovation. Grant project: its manufacturer, is not guaranteed or endorsed by the PID2021-127076OB-I00. publisher. References Ainscow, M. (2015). “The index for inclusion,” in Struggles for Equity in Education, https://www.observatoriodelainfancia.es/ficherosoia/documentos/7434_d_informe- ed. M. Ainscow (London: Routledge), 85–95. doi: 10.4324/9781315688213 educo-covid-19.pdf Allison, M. A., and Attisha, E. (2019). The link between school attendance and good European Commission (2003). The Role of the Universities in the Europe of health. Pediatrics 143:e20183648. doi: 10.1542/peds.2018-3648 Knowledge. Brussels: European Union. Alvariño, C., Arzola, S., Brunner, J. J., Recart, M. O., and Vizcarra, R. (2000). Gestión Farmer, T. W., Dawes, M., Hamm, J. V., Lee, D., Mehtaji, M., Hoffman, A. S., escolar: un estado del arte de la literatura. Paideia 29, 15–43. et al. (2018). Classroom social dynamics management: why the invisible hand of the teacher matters for special education. Remedial Spec Educ. 39, 177–192. doi: 10.1177/ Andrew, M., and Blake, M. K. (2021). The long arm of early exclusionary school 0741932517718359 discipline? a multi-model analysis. Youth Soc. 55:0044118X2110426. doi: 10.1177/ 0044118X211042643 Goldstein, H., Browne, W., and Rasbash, J. (2002). Partitioning variation in multilevel models. Understanding Statistics 1, 223–231. doi: 10.1207/ Bae, S. M. (2020). Long-term effect of adverse childhood experiences, school S15328031US0104_02 disengagement, and reasons for leaving school on delinquency in adolescents who dropout. Front. Psychol. 11:2096. doi: 10.3389/fpsyg.2020.02096 Grinberg, S. (2006). Educación y gubernamentalidad en las sociedades de gerenciamiento. Rev. Argentina Sociol. 4, 67–87. Ball, S. J. (2000). Sociology of Education: Major Themes. London: Psychology Press. Hubert, C., Villalobos, A., Abreu, A. B., Suárez-López, L., and Castro, F. D. (2019). Banda, E., Svanemyr, J., Sandøy, I. F., Goicolea, I., and Zulu, J. M. (2019). Factors associated with pregnancy and motherhood among Mexican women aged Acceptability of an economic support component to reduce early pregnancy and 15-24. Cad Saude Publica 35:e00142318. doi: 10.1590/0102-311x00142318 school dropout in Zambia: a qualitative case study. Glob. Health Action 12:1685808. doi: 10.1080/16549716.2019.1685808 Itzhaki, Y. (2019). The different role of mentor support along the high-school dropout process. Youth Soc. 51, 981–1008. doi: 10.1177/0044118X18803260 Belavi, G., and Murillo, F. J. (2020). Democracia y justicia social en las escuelas: dimensiones para pensar y mejorar la práctica educativa. REICE. Rev. Iberoam. Sobre Jabbari, J., and Johnson, O. (2021). The process of “pushing out”: accumulated Calid. Efic. Cambio Educ. 18, 5–28. doi: 10.15366/reice2020.18.3.001 disadvantage across school punishment and Math achievement trajectories. Youth Soc. 54, 911–934. doi: 10.1177/0044118X211007175 Bellei, C. (2015). El Gran Experimento: Mercado y Privatización de la Educación Chilena. Santiago: LOM. Jurado, P., and Tejada, J. (2019). Disrupción y fracaso escolar. Un estudio en el contexto de la educación secundaria obligatoria en cataluña. Estudios Sobre Educ. 36, Bolívar, A. (2019). Un currículum inclusivo en una escuela que asegure el éxito para 135–155. doi: 10.15581/004.36.135-155 todos. E-Curriculum 17, 827–851. doi: 10.23925/1809-3876.2019v17i3p827-851 Bronfenbrenner, U. (1992). Ecological Systems Theory. Philadelphia, PA: Jessica Låg, T., and Sæle, R. G. (2019). Does the flipped classroom improve student learning Kingsley. and satisfaction? a systematic review and meta-analysis. AERA Open 5, 1–17. doi: 10.1177/2332858419870489 Camacho, M. (2016). Fracaso Escolar y Abandono Educativo Temprano en Educación Secundaria Obligatoria: Un Estudio Integrado. Huelva: Universidad de Huelva. Lee, J., and Desjardins, R. (2019). Inequality in adult learning and education participation: the effects of social origins and social inequality. Int. J. Lifelong Educ. Castro, A. (2008). Gestión y política: dos modos de organizar y gobernar las escuelas. 38, 339–359. doi: 10.1080/02601370.2019.1618402 Rev. Escuela Ciencias Educ. 6, 31–46. Lee-St, T., Walsh, M., Raczek, A., Vuilleumier, C., Foley, C., Heberle, A., et al. (2018). Causton-Theoharis, J., Theoharis, G., Bull, T., Cosier, M., and Dempf-Aldrich, The long-term impact of systemic student support in elementary school: reducing high K. (2011). Schools of promise: a school district–university partnership centered school dropout. AERA Open 4, 1–16. doi: 10.1177/2332858418799085 on inclusive school reform. Remedial Spec. Educ. 32, 192–205. doi: 10.1177/ 0741932510366163 Martínez-Garrido, C., and Murillo, F. J. (2013). “El uso de los modelos multinivel en la investigación educativa: estadísticas avanzadas para conocer y cambiar la educación Contreras-Villalobos, T. A., Baleriola, E., and Opazo, H. S. (2022). Enactment legal en América Latina,” in Estadística en la Investigación: Competencia Transversal en la de la educación para jóvenes y adultos. Del enfoque laboral al enfoque educativo. Foro Formación Universitaria, ed. A. Salcedo (Caracas: Universidad Central de Venezuela), Educ. 20, 13–38. doi: 10.14516/fde.994 41–71. de Cristo, H. (2019). Del Dicho al Derecho: Modelo de Calidad de Escuelas de Mduma, N., Kalegele, K., and Machuve, D. (2019). A survey of machine learning Reingreso Para Chile. Santiago: Dirección Social Nacional. approaches and techniques for student dropout prediction. Data Sci. J. 18:14. doi: De la Cruz Flores, G., and Ortega, D. (2019). Why did I go back to school?” school 10.5334/dsj-2019-014 dropout and return experiences from high school students. Perf. Ed. 41, 8–26. Mikulec, B. (2021). The influence of international intergovernmental organizations EduCo (2021). COVID-19: Impacto de la Pandemia y sus Escuelas en on Slovenian adult education policies. Int. J. Lifelong Educ. 40, 37–52. doi: 10.1080/ la Educación. Diagnóstico de un año de Pandemia. Available online at: 02601370.2021.1871674 Frontiers in Psychology 10 frontiersin.org Contreras-Villalobos et al. 10.3389/fpsyg.2023.1163088 MINEDUC (2016). Medición de Deserción Escolar en Chile. Santiago: Ministerio de dropout in school systems: the case of Chile. Educ. Inform. Technol. doi: 10.1007/ Educación. s10639-022-11515-5 MINEDUC (2017). Educación de Personas Jóvenes y Adultas. Santiago: Ministerio Snijders, T. A. B., and Bosker, R. J. (2012). Multilevel Analysis: An Introduction to de Educación. Basic and Advanced Multilevel Modeling. Thousand Oaks, CA: Sage Publishing. MINEDUC (2020). Aprendo en Línea. Santiago: Ministerio de Educación. Tarabini, A. (2017). La Escuela no es Para ti: El rol de los Centros Educativos en el Abandono Escolar. Barcelona: Octaedro. Murillo, F. J., and Duk, C. (2020). El COVID-19 y las brechas educativas. Rev. Latinoamericana Educ. Inclusiva 14, 11–13. Tarabini, A. (2020). ¿Para qué sirve la escuela? reflexiones sociológicas en tiempos de pandemia global. Rase 13, 145–155. doi: 10.7203/RASE.13.2.17135 doi: 10.4067/S0718-73782020000100011 Terrenghi, I., Diana, B., Zurloni, V., Rivoltella, P. C., Elia, M., Castañer, M., et al. Nada, C., Santos, S., Macedo, E., and Araújo, C. (2020). Can mainstream and (2019). Episode of situated learning to enhance student engagement and promote alternative education learn from each other? an analysis of measures against school deep learning: preliminary results in a high school classroom. Front. Psychol. 10:1415. dropout and early school leaving in Portugal. Educ. Rev. 72, 365–385. doi: 10.1080/ doi: 10.3389/fpsyg.2019.01415 00131911.2018.1508127 UNESCO (2015). Recomendación Sobre el Aprendizaje y la Educación de Adultos. Novella, C. (2020). What educational pact do we need in Spain to fulfill the Paris: UNESCO. indicators of et 2020? Rev. Española Educ. Comparada 36, 74–97. doi: 10.5944/reec. 36.2020.26132 UNESCO (2017). El Poder del Aprendizaje y la Educación de Adultos en Todo el Mundo. Revisión de la Sexta Conferencia Internacional de Educación de Adultos. Paris: OECD (2012). Promoting Growth in all Regions. Paris: OECD. doi: 10.1787/ UNESCO. 9789264174634-en UNESCO (2020). Global Education Monitoring Report 2020: Inclusion and OECD (2018). Education at a Glance. Paris: OECD. doi: 10.1787/eag-2018-en Education: All Means All. Paris: UNESCO. Opazo, A. (2017). Trayectoria Escolar de los Estudiantes más Vulnerables del Sistema Valenzuela, J. P., Ruiz, C., and Contreras, M. (2019). Revisión de Antecedentes Sobre Educativo Chileno. Santiago: Centro de Estudios MINEDUC. Exclusión Educativa: Una Mirada a la Deserción y Repitencia Escolar. Santiago: CIAE. Paniagua, A. (2022). Programas de Segunda Oportunidad ¿Qué Funciona Vázquez-Nava, F., Vázquez-Rodríguez, E. M., Vázquez-Rodríguez, C. F., and Para Mejorar el Retorno Educativo y las Transiciones al Trabajo de las y Betancourt, N. V. O. (2019). High school dropout: association with family structure, los Jóvenes?. Available online at: https://ivalua.cat/sites/default/files/2022-04/qf20_ maternal employment, and health-risk habits among female Mexican adolescents. segundaoportunidad_220421.pdf (accessed December 15, 2022). J. Child Fam. Stud. 28, 3307–3314. doi: 10.1007/s10826-019-01505-6 Portela, A., Nieto, J., and Torres, A. (2019). La reincorporación formativa de jóvenes Veiga-Neto, A., and Noguera, C. (2010). Gubernamentalidad neoliberal: que abandonan tempranamente la educación. Rev. Española Pedagogía 77, 103–122. implicaciones para la educación. Rev. Educ. Pedagogía 22, 213–235. doi: 10.22550/REP77-1-2019-07 Prenkaj, B., Velardi, P., Stilo, G., Distante, D., and Faralli, S. (2020). A survey of Verger, A. (2019). A política educacional global: conceptos y marcos teóricos clave. machine learning approaches for student dropout prediction in online courses. ACM Práxis Educ. 14, 9–33. doi: 10.5212/PraxEduc.v.14n1.001 Comp. Surveys 53, 1–34. doi: 10.1145/3388792 Warne, M., Svensson, Å, Tirén, L., and Wall, E. (2020). On time: a qualitative Quiroga, A. R. (2017). Escuela y producción de subjetividad: El papel de la educación study of Swedish students’, parents’ and teachers’ views on school attendance, with a en las sociedades del gerenciamiento y el paradigma de la gestión escolar. IXTLI - Rev. focus on tardiness. Int. J. Environ. Res. Public Health 17:1430. doi: 10.3390/ijerph1704 Latinoamericana Filosofía Educ. 4, 221–235. 1430 Rodríguez, P. (2019). Caracterización Cuantitativa de Trayectorias Escolares en Wilkerson, K., Afacan, K., Yan, M., Justin, W., and Datar, S. (2016). Academic Jóvenes con Algún Grado de Exclusión Educacional. Available online at: http:// remediation–focused alternative schools: impact on student outcomes. Remedial Spec. ciae.uchile.cl/download.php?file=noticias/00_1552671019.pdf (accessed December 09, Educ. 37, 67–77. doi: 10.1177/0741932515620842 2022). Woltman, H., Feldstain, A., MacKay, J. C., and Rocchi, M. (2012). An introduction Rodrí,guez, P., Villanueva, A., Dombrovskaia, L., and Valenzuela, J. P. (2023). A to hierarchical linear modeling. Tutor Quant Methods Psychol. 8, 52–69. doi: 10.20982/ methodology to design, develop, and evaluate machine learning models for predicting tqmp.08.1.p052 Frontiers in Psychology 11 frontiersin.org