An Exploratory Analysis of the Student Connections Survey in Rhode Island

The purpose of this study was to complete a data-driven exploratory analysis of integrated data from the Connections Project collected across several school sites during the 2016-2017 academic school year. Using data from 1,309 middle school and high school students in Rhode Island, the study examined the relationship between student connectedness with adults and peers and student outcome variables commonly assessed in schools across the U.S., namely tardy arrivals, attendance, disciplinary referrals, and failed courses. Results indicated that students with higher levels of perceived connectedness to adults and peers in their school building had more positive school outcomes. Specifically, students with higher levels of connectedness had fewer instances of disciplinary referrals and fewer failed courses when compared to peers with lower levels of perceived connectedness. Further, students who named their advisory teacher as an adult connection had fewer instances of tardy arrivals, absences, and failed courses. However, studentperceived connectedness was not a significant predictor of drop-out risk. Implications for practice and research with the Connections Project are discussed.


LIST OF TABLES
described feelings of connectedness and social belonging as a fundamental human need. In examining feelings of belonging in schools, social belonging has been referred to using various terms including school engagement, school bonding, school attachment, and school connectedness (Libbey, 2004;Shochet, Dadds, Ham, & Montague, 2006). Across the plethora of definitions for the construct of school connectedness (Blum, 2005; Center for Disease Control, 2009a; Gillen-O'Neal & Fuligni, 2013;Goodenow, 1993;Sulkowski, Demaray, & Lazarus, 2012), there are three key elements: connectedness to adults in the school, connectedness to peers in the school, and connectedness to the school itself (Lohmeier & Lee, 2011). For the purposes of this study, the CDC (2009a) definition of school connectedness, which states that it is "the belief by students that adults and peers in the school care about their learning as well as about them as individuals," will be used.
Feelings of school connectedness are not unique to one developmental period, and are salient across all students, from preschool to post-doctoral settings (Lohmeier & Lee, 2011). Most research on school connectedness has focused on the transitions to and from middle school, as this time is seen as critical to the remainder of students' academic careers (Tillery, Varjas, Roach, Kuperminc, & Meyers, 2013;Appendix A). Indeed, it is common for feelings of school connectedness to decline in middle school years (Gillen-O'Neel & Fuligni, 2013;Monahan, Oesterle, & Hawkins, 2010;O'Brennan & Furlong, 2010). Research on the stability of school connectedness over time has yielded inconsistent results. Gillen-O'Neal and Fuligni (2013) report that feelings of connectedness tend to increase again when students reach secondary school. Other researchers, such as Monahan, Oesterle, and Hawkins (2010), report that by high school, as many as 40% to 60% of all youth report feeling disconnected from school across urban, suburban, and rural settings. Additional research is needed to examine school level differences in school connectedness. Presently, results have been inconclusive, though they do show clear differences based on grade level (O'Brennan & Furlong, 2010).

Adult Connections
Student connectedness to teachers and adults has long been heralded as an important factor in the demonstration of positive student outcomes. For example, Metz (1983) reported that one of the most frequently mentioned reasons students gave for leaving school prior to graduation was poor relationships with teachers (as cited by Davis & Dupper, 2004). In addition to these consequences, teacher connectedness has also been linked as a protective factor for initiation of health risk behavior, including smoking, escalation of smoking, suicidal attempts, and age of first intercourse (McNeely & Falci, 2004). It is important to note that all adults (i.e., lunch personnel, janitorial staff, coaches, etc.) in a school building are important components of school connectedness, not just teachers and administrative staff (Blum, 2005).

Perception of Support.
Perception of teacher support may have more powerful effects on student outcomes than the actual level of support teachers provide. Murray, Murray, and Waas (2008) investigated self-reported child and teacher perceptions of teacher-child relationships among kindergarten students of color in a large urban district.

Using the My Family and Friends -Teacher (MFF-T) and My Family and Friends -
Child (MFF-C) measures, teachers and students reported on the child's perceptions of the child-teacher connection. Additional information was gathered regarding the child's school adjustment through teacher reports and self-reports from the child. Results showed minimal concordance between teacher and child reports of perceptions of teacher support.
The children who reported greater perceived support from teachers also reported greater school liking on the school adjustment scale than children with lower levels of perceived support. The authors discuss the need to utilize methodology that provides a more direct test of child versus teacher perceptions. At present, no data are available on student perceptions of teacher support beyond elementary school. The current study examined student perceptions of teacher support during middle school and secondary school.
Advisory. Increasingly, secondary schools in the U.S. are employing an advisory system. An advisory program is a school scheduling configuration in which an adult meets with a group of students regularly during school hours to provide mentorship, to create personalization within the school, and to form a peer community of learners (Shulkind & Foote, 2009; Appendix A). To provide empirical evidence on the effectiveness of advisory programs, Shulkind and Foote (2009) , 1997), which asks participants to nominate the person or persons that play an attachment-related role (e.g., a safe haven to relieve stress in difficult situations). Data were also gathered on students' closeness with their advisor, security with their advisor, school engagement, perceptions of support from peers, and academic achievement. Overall, 40.7% of students nominated their advisors as a secondary attachment figure in their attachment hierarchy; their mother and best friend were the most frequently cited otherwise. Students who nominated their advisor also reported more engagement in school. In order to reinforce the role of advisory in facilitating adult connections in the school environment, these results must be replicated across various student populations.

Peer Connections
Buchanan and Bowen (2008)  teacher-child support, teacher-child conflict, peer acceptance, peer popularity, and physical and relational aggression in three data waves from grade four to grade six.
Results showed that peer relationships mattered above and beyond the effect of teacherchild relationships. Behavioral engagement was positively associated with teacher-child support and peer acceptance, while it was negatively associated with teacher-child conflict and peer popularity.
A secondary goal of De Laet et al. was to examine the normative development of behavioral engagement, teacher support, and teacher conflict. The results showed a general trend of decline in behavioral engagement, decline in teacher-child support, and an increase in teacher-child conflict over time. From grade four to grade six, children with fewer declines in teacher-child support also had fewer declines in behavioral engagement. Furthermore, children who were endorsed as being more physically aggressive had less initial teacher-child support and peer acceptance, more initial teacherchild conflict and peer popularity, and a greater decrease in engagement over time. The present study will examine behavioral engagement at the school level (i.e., number of tardy arrivals, number of absences, number of failed courses, and number of disciplinary referrals).

Correlates of School Connectedness
The current study addresses the relationship between student-perceived connectedness and known correlates of connectedness cited in the literature, including student disability status, socioeconomic status, tardy arrivals, attendance, disciplinary referrals, number of failed courses, and student dropout risk.

Mental Health. The National Longitudinal Study of Adolescent Health (Add
Health) collected data on more than 36,000 7 th grade to 12 th grade students nationwide to investigate adolescents' health and risk behavior trajectories over time. A large body of research has emerged from this data, including an examination of the relationship between student connectedness and mental health outcomes (Loukas, Ripperger-Suhler, & Horton, 2009;McNeely & Falci, 2004;Wormington et al, 2016). Contained within the Add Health survey is a five-item measure of school belonging. Items include: "I feel close to people at this school"; "I am happy to be at this school"; "I feel like I am a part of this school"; "The teachers at this school treat students fairly"; and "I feel safe at this school." Additional measures, including the California Healthy Kids Survey, have utilized these same items (O'Brennan & Furlong, 2010).
Using Add Health data, connectedness has been found to be the strongest protective factor for decreases in substance use, early sexual initiation, violence, and risk of unintentional injury across girls and boys (CDC, 2009a). Further, connectedness is negatively related to the development of conduct problems, engagement in substance use, antisocial and violent behavior, depression, anxiety, emotional distress, and suicidality (Lohmeier & Lee, 2011;Sulkowski, Demaray, & Lazarus, 2012). In fact, the CDC has promoted "building and strengthening connectedness or social bonds within and among persons, families, and communities" as a prevention strategy for suicidal behavior (CDC, 2008, p. 1).
Vulnerable Populations. School connectedness may be especially important to foster in students from vulnerable at-risk populations, such as LGBTQ students, students with disabilities (e.g., identified status, Appendix A), students with physical or mental health problems, and students who live in poverty (CDC, 2009a;Sulkowski et al, 2012;Tillery et al, 2013 Similarly, Doren, Murray, and Gau (2014) examined the predictors of school dropout for high school students with learning disabilities (LD) using a nationallyrepresentative sample of 13-17 year old students. Twenty-six predictors across four domains (e.g., sociodemographic, individual, family, and school-based factors) were examined. The final multivariate model indicated that grades, risk behaviors, parent expectations, and the quality of students' relationships (i.e., getting along with teachers and other students) remained salient predictors to school dropout among students with LD. Perceived quality of students' relationships were measured using the sum of two items, "gets along with teachers" and "gets along with other students," on a four-point scale (1 = not at all well; 2 = not very well; 3 = pretty well; and 4 = very well). Given the increased dropout risk among students with disabilities and the importance of positive relationships with teachers and peers, student connectedness should be considered in models of dropout risk and monitoring student outcomes. One aim of the present study was to examine differences in connectedness based on SES (using free and reduced lunch status as a proxy) and differences in connectedness based on disability status in the school environment.

Student Outcomes.
Besides its association to mental health, the relationship between school connectedness and student outcomes has been widely studied. In her literature review of student relationships to schools, Libbey (2004) found that across all studies, connectedness was highly related with positive student outcomes, both academically and behaviorally. School connectedness is positively correlated with classroom test scores, grades earned, academic motivation, academic self-efficacy, and student engagement (CDC, 2009b;Klem & Connell, 2004;Niehaus et al, 2012).
Considerably less research has been done on the relationship between school connectedness and behavioral outcomes, such as disciplinary referrals or school suspensions (i.e., De Laet et al, 2015). Further, the formation of interpersonal relationships in the school building is an important factor in school retention, dropout prevention, and graduation rates (Davis & Dupper, 2004;Doll, 2010;Sulkowski et al, 2012).
The dropout prevention literature indicates that differences exist between high school dropouts and graduates as early as kindergarten in areas such as academics, problem behavior, and family factors (Hickman, Bartholomew, Mathwig, & Heinrich, 2008). These differences can be stark among students from vulnerable populations, particularly students with disabilities and low-income students (Balfanz & Byrnes, 2012).
Long term negative outcomes associated with school dropout include lower average income, higher rates of unemployment, increased likelihood of being incarcerated, and death at a younger age (Schoenberger, 2012).
While there has been increased concern regarding school dropout and its deleterious effects, research has only begun to study early indicators of school dropout longitudinally (Schoenberger, 2012). McKee and Caldarella (2016) argue that risk factors can be considered in two categories: social (e.g., race, ethnicity, gender, and socioeconomic status) and academic (e.g., prior academic performance, course grades, and test performance). In recent years, several states and districts have developed early warning systems (EWS) to identify at-risk students in middle and high school with the intention of designing and implementing interventions to keep them on track to graduate (Frazelle & Nagel, 2013). EWSs use student-level data as indicators of student progress toward graduation. An effective EWS should utilize indicators and thresholds that have been verified in the local context in which the system is being used. Given the statistical knowledge needed to create localized systems, districts are encouraged to use attendance, behavior incidents, and course performance (the "ABCs") as their base set of indicators when building an EWS (Frazelle & Nagel, 2013). In line with the Response to Intervention framework, tiered systems of intervention are suggested in order to address the complexity of student needs.
As mandated by the Rhode Island Secondary School Regulations, local education agencies are required to monitor and analyze student indicators beginning in grade six and continuing to grade 12 (Rhode Island Department of Education, 2017). In 2012, the Rhode Island Department of Education (RIDE) developed the state's initial early warning system as a tool to identify and intervene with students at-risk of not graduating high school on time or dropping out based on seven years of historical student data from districts across Rhode Island (RIDE, 2013). Using student demographic and performance data as independent variables, the development team completed regression modeling to determine the most salient predictors of on-time graduation for each grade. On-time graduation was represented as a binary dependent variable with students who graduated within four years of entering high school considered on-time graduates and students who took longer than four years were considered non-on-time graduates (RIDE, 2012).
Results from the regression models were cross-validated to determine accuracy rates for the grade-based model of on-time graduation. Of the 17 possible indicators, results indicated that the following six indicators were the most robust predictors: 1) attendance, 2) years overage (i.e., the number of years a student is older than the standard age for a given grade), 3) number of suspensions, 4) New England Common Assessment (NECAP) reading scores, 5) NECAP math scores, and 6) aggregate on-track percentage. The aggregate on-track indicator is an equation that provides a percent likelihood that a student will graduate on-time given the student's current year performance and demographic data, and varies by grade level. It should be noted that although student gender was highly predictive of on-time graduation, this variable was removed from the list of indicators as it is not an "actionable" variable as nothing can be done to change it. Response to Intervention (RTI) framework, all students complete a universal screening measure designed to ascertain the names of adults and peers in the building with whom they feel they have a good personal connection (Appendix A). In conjunction with the student screening measure, teachers and staff also complete a survey wherein they name students in the building whom they feel they have a good personal connection with.
Localized data obtained from the screening measure has been used to target students who may be in need of social-emotional intervention. Presently, there are two middle schools and three high schools involved in the Connections Project. Four of the five schools are located in suburban and rural school districts in the Northeast, while the fifth school is in a suburban district in the upper Midwest.
Individual schools or school districts that participate in the Connections Project are provided assistance and support in implementation from the Connections Project Team. Two primary support people conduct four remote, web-based meetings per academic school year to prepare schools for screening administration, discuss data organization and entry, review data and identify individuals and groups for follow up, and to plan for the following school year. Additionally, a team of graduate students from the University of Rhode Island provides on-site assistance as needed and data support.
The team from URI analyzes the de-identified data to provide descriptive statistics as well as correlational analyses to the each individual school's Problem-Solving Team in a consolidated report. It is this project that served as the basis of this thesis project.

Purpose of the Present Study
The purpose of the study was to complete a data-driven exploratory analysis of integrated data from the Connections Project collected over the 2016-2017 academic school year. The research will contribute to the development of the Connections Screening as a valid universal screening measure to be used to examine middle school and secondary students' connectedness to important others in their school community.
The following hypotheses were tested:

Participants
The present study of secondary data included 1,309 students and corresponding data from 140 school personnel in their respective school buildings in the state of Rhode Island. Table 1 provides the full complement of data collected about the students, including year of graduation, disability status, and socioeconomic status (free/reduced lunch: FRL). Neither students nor teachers were asked to respond to demographic or personal background questions. No data were collected about gender, race, or ethnicity of students or teachers. Year building (including teachers, staff, and support personnel) to provide data regarding student-adult relationships by identifying the names of up to six students with whom they feel they have a good personal connection. Adults are told that these students may be those who seek advice and guidance for personal or academic matters. Instructions to teachers note that the students they name may not necessarily be current students in their classrooms. The measure is scored by identifying the number of perceived student connections for a total score of six possible connections. Adult-perceived connections are tallied for each student and added to the student data as "number of faculty/staff connections," which can range from zero to seven or more. At present, no studies have examined the psychometric properties of the Adult Connections Survey.

Procedure
The present study uses secondary data from the Connections Screening Data and Evaluation Project (Pristawa & Marraccini, 2013), an on-going project designed to assist school personnel in identifying potentially at-risk students in the social-emotional area of

Preliminary Analyses
Data were analyzed using IBM SPSS 24.0. Prior to conducting analyses to address the study hypotheses, descriptive statistics were examined to determine if the data met the assumptions of normality, linearity, and homogeneity of variance.
Preliminary analyses revealed that the data did not meet the assumptions of normality, linearity, and heteroscedasticity. Therefore, student outcome data variables (e.g., tardy arrivals, attendance, disciplinary referrals, and failed courses) which contained several zero values, were transformed using the square root method in order to normalize the distribution, similar to McKee and Calderella (2016). After performing square-root transformations, tardy arrivals, absences, and failed courses were in the acceptable range for skewness and kurtosis (|1.0| and <2.0, respectively; Harlow, 2014). However, skewness and kurtosis for disciplinary referrals remained elevated (e.g., 3.62 and 14.76).
In order to assess whether any statistically significant group differences existed between school sites, a multivariate analysis of variance ( ; data were not available to inform differences in tardy arrivals and failed courses. Nevertheless, no significant differences existed between middle school students (School A) and secondary school (School B) students' perceived adult connectedness or peer connectedness.
Additionally, a logistic regression was used to examine group differences in categorical variables (e.g., connection to advisor, student connectedness, disability status, and SES) across school sites. As a set, connection to advisor, student connectedness, disability status, and SES showed a significant relationship with school site identification among the sample of 1,309 students across two schools, χ 2 (8)=25.16, p = .001. The average pseudo R 2 value was 0.02, indicating a small effect size (ES) according to Cohen's guidelines for multivariate ES (Harlow, 2014). For disability status, SES, and student connectedness, the first category was used as the reference category, all of which indicated little to no risk based on the literature (e.g., no identified disability, no qualification for free or reduced lunch, and high levels of connectedness, respectively).
Inversely, the last category for connection to advisor (i.e., student-and adult-perceived connection) was used as the reference category. Two of the four predictors, connection to advisor and student connectedness, significantly predict school site. Odds ratios greater than 1.0 suggest higher odds of being in the high school group, and results less than 1.0 suggest lower odds of being in the high school group. Using the odds ratios and their respective confidence intervals, results suggest that high school students had four times more odds than middle school students of having an adult-perceived connection to their  (Harlow, 2014).
Results indicated a significant multivariate effect for the combined independent variables after controlling for student SES and disability status, F(12, 3906) = 6.46, p<.001, Pillai's trace = 0.58, η 2 = .019, indicating a small effect size between student levels of support and student outcome variables when controlling for student disability status and SES. Follow-up ANCOVAs were completed to analyze micro-level results.
Significant univariate effects were found for disciplinary referrals, F(1) = 14.76, p<.001, R 2 = .033, and failed courses, F(1) = 16.14, p<.001, R 2 = .036, indicating that disciplinary referrals and failed courses explained 3.3% and 3.6%, respectively, of the variance with student-perceived levels of support after disability status and SES were taken into consideration. Both of these are considered to have small effect sizes (Harlow, 2014). As there were more than two groups in the independent variable, post hoc tests using the Bonferroni approach were completed. Post hoc tests revealed that lower levels of support (i.e., High Risk: No Adult, No Peer) had significantly higher rates of disciplinary referrals and failed courses when compared to peers with greater levels of support (Table   3).
To further test the first hypothesis, a logistic regression was used to extend the study results from Buchanan and Bowen (2008) to school-based student outcome variables. Student background variables (i.e., disability status and SES) were entered in stage one, followed by number of adult connections, number of peer connections, and the adult connection by peer connection interaction in subsequent stages. Given that attendance percentage was the only Rhode Island Early Warning System variable available in the data set, each student's attendance data was coded to reflect the level of drop-out risk (i.e., low risk, some risk, at-risk, and high risk) based on the benchmark for their respective grade, which served as the dependent variable. As the majority of students fell in the low drop-out risk category (n = 1,000), drop-out risk was collapsed into two categories, low risk and at-risk (i.e., some risk, at risk, and high risk), as opposed to four categories. For the purpose of this analysis, the low risk group served as the reference category. Two-tailed Pearson correlations did not reveal any evidence of collinearity among the variables in this analysis. Results indicated that the set of variables, disability status, SES, adult connectedness, peer connectedness, and the adult connectedness by peer connectedness interaction term, significantly related to student drop-out risk, χ 2 (5) = 14.22, p = .01. The average pseudo R 2 value was 0.01 indicating that differences between groups did not reach substantive significance (i.e., .02) according to Cohen's guidelines for multivariate ES (Harlow, 2014;Sullivan & Feinn, 2012). From an examination of the odds ratios and their respective confidence intervals (Table 4), students in this sample who qualified for free or reduced lunch (FRL) had 1.57 times more odds than students who did not qualify for FRL to be considered atrisk for school drop-out (OR = 1.57, p = 0.001, 95% CI [1.19, 2.07]). Adult connectedness, peer connectedness, and disability status did not predict school drop-out above and beyond student SES.

Summary of Logistic Hierarchical Regression Analysis for Variables Predicting Student
Drop-out Risk

Hypothesis 2
It was hypothesized that students who felt connected to their advisor, regardless of reciprocity, would have more positive student outcomes. To address Hypothesis 2, a multivariate analysis of variance (MANOVA) was conducted using student connection to advisor as the independent variable (e.g., no perceived connection, student-perceived connection, no student-perceived connection, adult-perceived connection, no adultperceived connection) and student outcome data as the dependent variables. Results from the MANOVA indicated a significant multivariate effect for the relationship between student outcome variables on student-and advisor-endorsed connection to advisor, F(12, 3912) = 3.18, p < .001, Pillai's trace = .029, partial η 2 = .010, indicating a nonmeaningful multivariate effect size. Micro-level results revealed significant univariate effects for tardy arrivals (F(3) = 6.32, p < .001, R 2 = .014), absences (F(3) = 5.67, p = .001, R 2 = .013), and failed courses (F(3) = 4.31, p = .005, R 2 = .010; however, there was no significant effect for number of disciplinary referrals on connection to advisor. Post hoc Tukey HSD tests were conducted on all possible pair-wise comparisons (See Table   5). Regarding tardy arrivals and absences, significant differences (p < .05) were present between students with no endorsed connection to their advisor and student-perceived connection to the advisor, indicating students with no endorsed connection had higher rates of both tardy arrivals and absences. Additionally, when examining failed courses, post hoc tests showed significant differences (p < .05) between students with no endorsed connection to their advisor and those who had a self-perceived and advisor-perceived connection to their advisor. Students with no perceived connection had higher numbers of failed courses in their first quarter of school. Note: *The mean difference is significant at the .05 level.
Due to the vastly uneven group sizes represented in the student connection to advisor variable in the first MANOVA (no perceived connection = 797; adult-perceived connection = 27; student-perceived connection = 413; student-and adult-perceived connection = 72), an additional MANOVA was completed wherein the independent variable was collapsed into two groups: student-perceived connection to advisor (n = 824) and no student-perceived connection to advisor (n = 485). Similarly, results indicated a significant multivariate effect for the relationship between student outcome variables on student-and advisor-endorsed connection to advisor, F(4,1304) = 5.25, p < .001, Pillai's trace = .016, partial η 2 = .016, indicating a small effect size. Significant univariate effects were found for all four student outcome variables. However, there were no meaningful Cohen's d effect sizes; effect sizes ranged from 0.004 to 0.011 (Table 6).

Discussion
The purpose of this study was to complete a data-driven exploratory analysis of integrated data from the Connections Project data collected over the 2016-2017 academic school year. The findings indicate that less student-perceived connectedness to adults and peers in the school building were inversely related to positive school outcome data.
Specifically, students with lower levels of connectedness had a greater number of disciplinary referrals and failed courses when compared to peers with greater levels of connectedness. Additionally, students who named their advisory teacher as an adult connection had fewer instances of tardy arrivals, school absences, and failed courses.
Unfortunately, student-perceived connectedness was not a significant predictor of student drop-out risk.
When controlling for the effects of disability status and socioeconomic status, students who reported lower levels of support had significantly higher rates of disciplinary referrals and failed courses when compared to peers with greater levels of support; however, level of support was not significantly related to tardy arrivals or number of absences in the first quarter. This finding provided partial support for Hypothesis 1, as it was expected that greater levels of connectedness would be related to lower rates of all four student outcome variables. This finding may be related to the fact that the Student Connections Survey and Adult Connections Survey are administered at the end of the first quarter after approximately 45 total school days. The mean number of days absent and number of tardy arrivals are 2.28 and 1.16, respectively. Results may have been different if the measure was administered at a later date given typical increases in absences and tardy arrivals through the progression of the academic year. The relationship between levels of support and attendance and tardy arrivals may have also been influenced by the square root transformations completed on those variables. In school psychology applied practice, these results can be used to examine differences between students who would be identified as low, moderate, or high risk according to the Student Connections Survey, perhaps indicating that these students should be targeted for additional interventions under multi-tiered systems of support.
Student perceptions of adult and peer connectedness did not significantly predict school drop-out risk, contrary to the expectation that levels of connectedness would be inversely related to poor student outcomes (e.g., higher levels of drop-out risk).
Therefore, these results did not extend the findings from Buchanan and Bowen (2008) to student outcome variables. Socioeconomic status was the only salient factor in the model, which included disability status, SES, adult connectedness, and peer connectedness. One possible reason for this finding is that the outcome variable only consisted of attendance data from the Rhode Island Early Warning System, as opposed to the full algorithmic model used by the Rhode Island Department of Education. The full model includes years overage, number of suspensions, NECAP reading and math scores, and the aggregate ontrack percentage. Use of the full model would have allowed for the creation of a more robust measure of drop-out risk. Further, the use of attendance to measure dropout risk may have also been problematic given the well-known connection between student income level and school attendance (National Center for Children in Poverty, 2008).
However, SES may have had stronger effects in this particular population given the amount of socioeconomic diversity present in the district. District-level data indicates that the median household income in the participatory district is $67,693, whereas the per capita income is $32,073, suggesting a considerable discrepancy between the two (U.S. Census Bureau, 2016). According to the U.S. Census Bureau, "median household income" refers to the income of the householder and all individuals in the house over age 15, whereas "per capita income" is derived by dividing the aggregate income of a particular group by the total population in that group (U.S. Census Bureau, 2016). In areas where there is not such a large discrepancy in SES, this factor may not be as influential.
The importance of relationships to advisors continues to be well-supported in the literature for undergraduate and graduate students (Craft, Augustine-Shaw, Fairbanks, & Adams-Wright, 2016;Khalil & Williamson, 2014;Zhang, 2016); however, there is still a dearth of information regarding the effects of advisor-student relationships in secondary school. In the present sample, 37.1% of students named their advisor as a connection.
Student-perceived connection to advisor was related to lower rates of tardy arrivals and absences. This finding adds to the body of literature that suggests that student-perceived support, rather than adult perception of given support, has a greater impact on student outcome data (Murray, Murray, & Waas, 2008). Regarding failed courses, students with no perceived connection had higher numbers of failed courses in their first quarter of school when compared to those with both a student-perceived and adult-perceived connection to advisor. In this instance, reciprocity of the endorsed relationship between students and their advisors mattered. In the present study, connection to advisor did not have any significant relationship to number of disciplinary referrals in quarter one. These findings are in line with previous research by Van Ryzin (2010), who found that 40.7% of students who participated in the study nominated their advisor as an attachment figure.
Similarly, students who nominated their advisor as an attachment figure were more engaged in school.

Limitations
Several limitations are notable in this study. First, the psychometric characteristics of the Student Connections Screening and the Adult Connections Screening have not yet been established. Only one study has explored the concurrent validity of the Student Connections Survey in relation to the Strengths and Difficulties Questionnaire (SDQ), a 25-item questionnaire developed to screen for behavioral and emotional difficulties and social skills with school-aged youth (Ruise, 2017). The study hypothesized that there would be no significant difference between students identified as "connected" (i.e., identifying more than one school connection) using the SCS and students identified as "normal" on the SDQ. Findings indicate that there is a negative relationship between students' self-reported peer connectedness and the Peer Relationships Problems subscale of the SDQ, suggesting that as peer connections increase, peer problems decrease. Thus, it is possible that these tools could be measuring similar constructs. Further, results indicated that the SCS classified students as at-risk more frequently than the SDQ, overidentifying up to 15% of students. Ruise (2017) also sought to evaluate the social validity of the Student Connections Screening. Teachers who participated in the study perceived the administration of the SCS to be useful and appropriate for the school setting, suggesting that the screening tool is practical for use by schools.
Second, the measure of connectedness is based solely on self-report at one sampling point during the school year. However, under the Response to Intervention framework, universal screeners are typically administered multiple times per school year (i.e., Fall, Winter, Spring) to accurately track all students (National Center on Response to Intervention, 2012). Previous research has indicated that student perception of connectedness outweighs other indicators of connectedness, thereby negating the need for additional support beyond self-report (Murray, Murray & Waas, 2008). Further, no follow-up data from participating schools exists on students identified as needing Tier 2 or Tier 3 intervention, particularly with those that endorsed having no connections. The Connections Project provides a follow-up social-emotional screening assessment for students who endorsed having few connections (i.e., no connections, no adult connections, etc.; Appendix E). Also, based on anecdotal comments, in some cases, students do not understand the directions on the Student Connections Survey or they indicate that they have adult connections outside of the school environment (i.e., coaches, scout leaders). Moreover, the Connections Project does not prescribe a uniform way of completing additional intervention beyond the initial follow-up. Rather, the Project suggests the use of local resources existing in each participating school, such as previously implemented interventions (e.g., Check & Connect: Christenson, Stout, & Pohl, 2012) to follow up with students lacking connections in the school building.
Longitudinal data from multiple points in the same academic year would be beneficial to determining if connectedness status changed as the result of school interventions or additional time to create connections with adults and peers.
Third, this study created drop-out risk categories based on the Rhode Island Early Warning System; therefore, the results may not be generalizable to samples outside the state. However, it should be noted that several individual districts and states (i.e., Sioux Falls School District, Houston Independent School District, Delaware Department of Education) have implemented similar systems to track drop-out risk (Frazelle & Nagel, 2015).
Furthermore, the present study created very liberal categories for connectedness (e.g., No Adult Support, No Peer Support, etc.). This limitation is two-fold. First, the Connections Project suggests follow-up screening for students who endorse no connections and those who endorse no connections to adults with some peer connections.
Given these criteria, a student who endorsed one adult connection and one peer connection was placed in the same risk category (i.e., Low Risk) as a student who endorsed three adult connections and three peer connections. This coding system increases the likelihood of Type II error in that students who are placed in lower risk categories based on one adult connection may actually be more appropriately placed in higher risk categories. Second, it would be useful to use existing Connections Project data to complete discriminant function analyses to determine if student background variables and student outcome variables could predict levels of connectedness. This process could aid in creating more rigid categories for connectedness based on associated student level variables.
Finally, the large number of zeros in the student outcome variables (e.g., tardy arrivals, absences, disciplinary referrals, and failed courses) in the present data set resulted in a highly non-normal distribution. Although the data was normalized using square-root transformations, future studies examining data from the Connections Project may want to consider the use of zero-inflated regression procedures as these statistical methods have the capacity to handle an overabundance of zero count data (Yang, Harlow, Puggioni, & Redding, 2017).

Implications
A secondary goal of this study was to contribute to the development of the Connections Screening as a valid universal screening measure to be used to examine middle school and secondary students' connectedness to important others in their school community. Although the present study cannot offer evidence for psychometric validity or reliability, results indicate that the Connections Screening can be used in conjunction with Early Warning Systems employed by schools to provide additional quantitative and qualitative data to explain student progress and behavior. For example, the originator of the survey has used its results to target school climate issues, such as safety and social relationships for students who are new to the district. Further, future research should focus on localized measures of student connectedness that serve specific school environments.
It must be noted that the Connections Project is an on-going project that implements improvements based on feedback from the preceding year. Results from this study can be used to inform future iterations of the survey in practice as well as research.
The Connections Project may want to consider including additional demographic data, such as gender, racial or ethnic group in future administrations. If used for research purposes, investigators may also want to gather additional measures of socioeconomic status, such as parental income level or parental education level. Researchers could also consider using the Student Connections Survey and the Adult Connections Survey with student subpopulations such as students with identified disabilities or students who identify as LGBTQ. Further, it may be beneficial for schools implementing the Connections Project to collate data that aligns with their state's early warning system to track students who are at-risk of dropping out. Finally, the Connections Project might also consider including a third aspect of student connectedness, connectedness to the school itself, as delineated by Lohmeier and Lee (2011). To achieve this end, the Student Connections Survey and Adult Connections Survey could be administered alongside psychometrically sound measures of school climate.
Universal screening. Brief assessment of all students conducted at the beginning of the school year designed to identify students who may be at risk for poor learning outcomes (National Center on Response to Intervention, 2012) Year of graduation. The year in which a student is scheduled to graduate high school based on their current class standing and credits earned; the class cohort to which a student belongs. MMM DD, 2017 Hello School Department Superintendent, My name is Erin Churchill. I am currently a third-year doctoral student in the APA-accredited, NASPapproved School Psychology Program at the University of Rhode Island. For the last two years, I have worked closely with the Connections Project as a data analyst. One of the schools in your district is involved in the Connections Project. During that time, I have become interested in looking at student connections and social-emotional learning in my own body of research.

Appendix D Cover Letter
My proposed thesis project seeks to gain a better understanding of the data provided by the Student Connections Survey and the Adult Connections Survey. I intend to use the combined de-identified data from each of the six participating schools to examine the relationship between Connections Survey data and school outcome data (e.g., tardy arrivals, absences, disciplinary referrals, and failed courses). Additionally, I intend to examine the relationship between student-advisor connection and those same school outcome variables. I feel that my study will contribute to the current body of literature on the importance of student connections to school dropout prevention and student retention.
Collectively, my major professor, Dr. Margaret Rogers, Kim Pristawa, and I have created a letter of authorization to be signed by each of the participating schools' superintendents. The text that is italicized in red is intended to be personalized for each school. Additionally, the IRB requires that the letter be placed on department letterhead. Please note that all data shared with me will be coded numerically and will not contain any identifying information.
If you feel comfortable with this request, please place the attached letter on district letterhead, sign, and return to me by April 19, 2017. If you would prefer to discuss this request further, feel free to email me at edchurchill@my.uri.edu or call at (928)  The University of Rhode Island is an equal opportunity employer committed to community, equity, and diversity and to the principles of affirmative action.