846619 BHDXXX10.1177/0198742919846619Behavioral DisordersAfacan and Wilkerson research-article2019 Original Research The Effectiveness of Behavior-Focused Alternative Middle Schools for Students With Disabilities Behavioral Disorders 1–12 © Hammill Institute on Disabilities 2019 Article reuse guidelines: sagepub.com/journals-permissions https://doi.org/10.1177/0198742919846619 DOI: 10.1177/0198742919846619 journals.sagepub.com/home/bhd Kemal Afacan, PhD1 and Kimber L. Wilkerson, PhD2 Abstract Behavior-focused alternative schools serve students who are identified by school personnel as exhibiting behavior difficulties, often coupled with low academic achievement. Students can be referred to behavior-focused alternative schools as an alternative to expulsion. In this study, we examined the demographic characteristics of students who attended behaviorfocused alternative middle schools, as well as the effectiveness of these schools on two outcomes: (a) standardized state reading assessment scores and (b) number of suspensions received. Using a retrospective cohort design study, we investigated whether students attending behavior-focused alternative middle schools experienced significantly different reading and suspension outcomes compared with a matched sample of students attending traditional middle schools. The majority of students in behavior-focused alternative middle schools were male, Black, and receiving special education services. Results showed that students attending behavior-focused alternative middle schools performed significantly lower on standardized assessments of reading in the eighth grade. No significant differences were noted for the number of suspensions experienced. Keywords behavior-focused, middle school, disability, alternative, remediation During the middle school years, students attain the academic and behavior skills necessary for a successful transition to high school. These skills typically include not only mastery of academic content areas such as reading and math but also mastery of the social skills and behavioral repertoires that allow them to flourish in high school (Mizelle & Irvin, 2000). During the past 20 years, researchers have documented two challenges to students’ successful transition from middle to high school: (a) middle school students experience a spike in behavior problems (Arcia, 2007; Gagnon, Gurel, & Barber, 2017; Skiba, Peterson, & Williams, 1997) and (b) middle school behavior and academic outcomes significantly predict later outcomes (McIntosh, Flannery, Sugai, Braun, & Cochrane, 2008; Tobin & Sugai, 1999). With regard to the spike in behavior problems, a host of researchers have reported that the frequency of behavior incidents reported by schools increases during the middle school years (Gagnon et al., 2017; Mendez & Knoff, 2003; Skiba et al., 1997). Furthermore, researchers have reported that patterns of school-reported behavior infractions vary across groups of students. For example, Skiba et al. (1997) reported that nearly half of the students attending middle school received disciplinary actions (e.g., suspension, office discipline referrals [ODRs]). However, analyses revealed that males, Black students, students receiving free or reduced price meals (FRM), and students with emotional and behavioral disorders (EBD) were more likely to receive disciplinary actions than females, White students, students who did not qualify for FRM, and students who received special education services under other disability categories. Similarly, Mendez and Knoff (2003) found that the suspension rate for a sample of middle school students was significantly higher than for elementary and high school students, with male and Black middle school students receiving a higher number of suspensions than their female, White, and Hispanic counterparts. Arcia (2007) reported similar findings: middle school students from a large urban school district experienced significantly more suspensions than elementary school students in the same district. Arcia (2007) also found that Black middle school students were more likely to be suspended than their Hispanic counterparts. In a recent study that provided a statewide analysis of punitive discipline practices in the state of Florida, Gagnon 1 Artvin Coruh University, Turkey University of Wisconsin–Madison, USA 2 Corresponding Author: Kemal Afacan, Department of Special Education, Faculty of Education, Artvin Coruh University, Artvin 08000, Turkey. Email: kemalafacan@artvin.edu.tr 2 et al. (2017) reported that males, Black students, and middle school students received more school suspensions than their female, non-Black, and elementary school counterparts. Both the academic and behavioral outcomes that students experience during their middle school years have been linked to later outcomes both in terms of academic achievement in high school and their experience with exclusionary discipline practices. For example, students with low standardized reading test scores in middle school had significantly lower rates of making timely progress toward degree in high school than students with higher reading scores (Balfanz, Herzog, & Mac Iver, 2007). McIntosh et al. (2008) found that as students’ ODRs increased in eighth grade, their grade point average (GPA) significantly decreased in ninth grade, and as their reading test scores increased in eighth grade, their GPA significantly increased in ninth grade. Tobin and Sugai (1999) also reported that when students had a high number of behavior incidents during their middle school years, they had significantly less chance to be on track for graduation from high school compared with students with none or very few formally recorded behavioral incidents during the middle school years. Researchers have also found that the number of behavior incidents in sixth grade positively and significantly predicts behavior incidents and academic outcomes in later middle school grades. For example, Lassen, Steele, and Sailor (2006) examined students’ sixth grade standardized reading and math scores, as well as the number of suspensions that they received, and found that the number of suspensions in sixth grade significantly predicted standardized reading and math scores in eighth grade. Students who received fewer suspensions performed significantly higher in both reading and math. Arcia (2007) found that students who experienced a high number of suspensions and had low standardized reading test scores in sixth grade tended to also have a high number of suspensions in seventh grade. These findings suggest the potentially predictive impact of middle school academic and behavior performance on later student outcomes. Behavior-Focused Alternative Middle Schools By definition, students who are viewed by school personnel as exhibiting chronic or persistent behavioral problems pose challenges for educators. The decades-long increase in the use of and research on systemic approaches to prevent behavior problems (e.g., notably, Positive Behavior Interventions and Supports [PBIS]) is a testament to the seriousness with which this problem has been approached (Sugai, Horner, & McIntosh, 2016). PBIS has been effective in decreasing exclusionary discipline actions overall; however, educators continue to log many formal exclusionary responses to individual behavior incidents, Behavioral Disorders 00(0) particularly for students who are Black (Vincent & Tobin, 2011). These responses may include ODRs, suspensions, expulsions, or referral to behavior-focused alternative schools. Behavior-focused alternative schools are designed to support students who are identified as exhibiting challenging behavior; the primary aim of these schools is to address behavior problems that have contributed to students’ lack of success in the traditional public school (Aron, 2006; Lehr, Lanners, & Lange, 2003). Enrollment in this school type is typically by referral, rather than by the student or their family’s choice (Raywid, 1994). Although the majority of these schools are designed to serve high-school-age students, it is also common for school districts to offer similar alternative school options to middle school students; specifically, 41%, 57%, and 63% of all public school districts reported that they offered alternative school options to middle school students in sixth, seventh, and eighth grades, respectively (Carver & Lewis, 2010). Characteristics and Outcomes of Students in Behavior-Focused Alternative Schools Past research examining student enrollment in alternative schools has documented a general concern that these schools are likely to perpetuate racial segregation within school districts. Carver and Lewis (2010) reported that students from minority backgrounds enroll in alternative middle schools at higher rates than their White counterparts. Specifically, students from minority backgrounds account for 55%, 69%, and 74% of students in sixth, seventh, and eighth grade alternative middle schools, respectively. In addition, more than 70% of students attending alternative middle schools come from homes considered to be low income (Carver & Lewis, 2010). Similarly, Foley and Pang (2006) found that male and Black students were overrepresented in alternative schools compared with female, Hispanic, Native American, and Asian students. Foley and Pang (2006) also reported that nearly 50% of students served in alternative schools were students receiving special education services for EBD. This is in stark contrast to the less than 1% of students who have this label in regular public schools (U.S. Department of Education, Office of Special Education and Rehabilitative Services, Office of Special Education Programs, 2016). Lehr and Lange (2003) also reported that alternative schools predominantly served students from minority backgrounds and students with disabilities. In a recent study examining enrollment patterns in one Midwestern urban school district, Perzigian, Afacan, Justin, and Wilkerson (2017) found that a higher percentage of students with disabilities and Black students enrolled in behavior and academic remediation-focused 3 Afacan and Wilkerson alternative schools. In contrast, White students were more likely to attend either traditional or innovative alternative schools. These past studies did not aim to serve as tests of racial disproportionality. Researchers defined disproportionality by comparing averages of unmatched or nonstatistically controlled students. Nonetheless, they provide evidence of differential enrollment across school types. In addition to examining student enrollment in alternative schools, it is also important to document these schools’ effectiveness on students’ academic and behavior outcomes. Past studies have reported that students’ educational experiences in alternative schools were not always positive. Chiang and Gill (2010) examined statewide-standardized reading and math test scores of students attending traditional and alternative schools in the city of Philadelphia. They reported lower standardized math and reading scores for students attending alternative schools compared with students attending traditional schools. Similarly, Drame (2010) compared the fourth and fifth grade standardized reading and math test scores of students with and without disabilities attending alternative schools in an urban school district. Drame (2010) reported that students with and without disabilities’ reading and math performance did not differ significantly and even decreased after attending alternative schools after one school year. Wilkerson, Afacan, Perzigian, Justin, and Lequia (2016) examined the impact of behavior-focused alternative high schools on students’ outcomes. They reported that students attending behaviorfocused alternative high schools had significantly lower secondary school attendance, earned significantly fewer credits, but experienced significantly fewer ODRs compared with a matched sample of students who remained in traditional schools. Existing evidence suggests differential outcomes for matched (e.g., Wilkerson, Afacan, Perzigian, et al., 2016) and unmatched (e.g., Chiang & Gill, 2010) groups of students who attend alternative versus traditional high schools. The current examination extends this research by examining outcomes for students who have similar characteristics (i.e., matched samples) but received their middle school instruction either in a behavior-focused alternative middle school or a traditional middle school. characteristics examined included the following: gender, race/ethnicity, qualification for FRM, qualification for special education services, fifth grade standardized reading test scores, and fifth grade suspensions received. Second, we compared the middle school academic and behavior outcomes of students attending behavior-focused alternative middle schools to a matched sample of students with similar fifth grade characteristics who remained in traditional schools. Middle school academic and behavior outcomes examined included eighth grade standardized reading test scores and number of suspensions received in the eighth grade. We chose eighth grade because it was the last year of middle school and the last opportunity to report data that was not influenced by the effect of high school. We used prior studies and available data to select covariates associated with students’ propensity to be placed in alternative schools. For example, we used gender, ethnicity/ race, and special education status as covariates because male students, students from ethnic minority backgrounds, and students with disabilities (e.g., EBD) are more likely to enroll in alternative schools (Carver & Lewis, 2010; Foley & Pang, 2006; Lehr & Lange, 2003 Perzigian et al., 2017). Also, history of poor behavior and academic outcomes including, but not limited to, excessive suspensions received and low reading achievement were used as covariates as they are two main criteria for students’ placement into alternative schools (Lehr, 2004; Lehr, Tan, & Ysseldyke, 2009; Raywid, 1994). Thus, covariates that reflect students’ background characteristics and past behavior and academic outcomes are included in this study. We answered the following two research questions: The Current Study and Research Questions Method The current study’s purpose was twofold. First, we examined demographic characteristics as well as fifth grade academic and behavior outcomes of students attending behavior-focused alternative middle schools and compared them with the same characteristics and outcomes of students who remained in traditional middle schools. We used fifth grade data to capture achievement and suspension rates before students head to middle school, where data suggests they will experience differential trajectories. Student Research Question 1: Do the baseline (fifth grade) characteristics of students attending behavior-focused alternative middle schools differ from those of students attending traditional middle schools? Research Question 2: Do eighth grade students attending behavior-focused alternative middle schools experience improved school outcomes (i.e., higher reading scores or lower number of suspensions) compared with students with similar fifth grade characteristics who remain in traditional middle schools? Participating School District Data for this study were provided by a local education agency (LEA) located in an urban Midwestern city in the United States. The LEA is situated in a city with a population of approximately 600,000, with 40% of the city’s total population identifying as Black, 38% identifying as White (nonHispanic), and 17% identifying as Hispanic/Latino. The LEA operates approximately 200 PK–12 schools and serves about 4 75,000 students annually, 56% of whom identify as Black, 24% as Hispanic or Latino, 14% as White (non-Hispanic), and 6% as Asian. Nearly 80% of students in the LEA receive FRM. Standardized scores on the state-mandated annual assessment in reading and math were below average compared with other similar districts in the same state. Data Source and Study Design Longitudinal data for the project were comprised of de-identified student-level data for all 21,162 students who were attending secondary schools during the 2012–2013 academic year. For the current study, we relied on a subset of that data that included only those students for whom there was complete demographic, academic, and behavior data from fifth grade through at least 1 year of high school. We only included students who were enrolled in schools in 2012–2013 that were coded as either traditional or behavior-focused. Students who were enrolled in other school types (e.g., innovative, academic-focused) were not included in this study’s sample. Creating this data subset resulted in a selected sample of 5,929 students, 5,848 (98.6%) of whom attended traditional middle schools and 81 (1.4%) who attended behavior-focused alternative middle schools. We utilized a retrospective cohort design study. We retrospectively gathered historical information about just one group of students (Ruspini, 2002). The cohort was not dynamic. It was closed against new entries or dropouts (Ruspini, 2002). Thus, attrition was not a confounding factor. We excluded students who took the state’s alternate reading assessment instead of the state’s regular assessment. This resulted in removal of two students from the study sample, both of whom attended behavior-focused alternative middle schools, thus reducing the total number of students in behavior-focused alternative schools in the study to 79. Privacy and Confidentiality The institutional review board at the University of MadisonWisconsin approved this study in 2013. The LEA provided us with permission to analyze and disseminate the results. The LEA removed any identifiable information from the database using a seven-digit identification number for each student to avoid revealing student names or other sensitive information before the data became accessible to researchers. Also, we secured the data in a password-protected file and did not access the data outside the institution. School Coding Procedures Training. The principal investigator of the project trained two research team members for school coding. The trainer and two coders examined the schools’ self-reported Behavioral Disorders 00(0) descriptions, curricula used, programs offered, and student population targeted as described on the school district’s website. Then, the trainer randomly selected 10 schools for the coders to review independently. The coding results were compared with the trainer’s coding results and discussed at a subsequent meeting. The training phase continued until the coders reached 100% agreement with the trainer. School coding. Once reliability was established, the two coders reviewed all schools in the sample based on a researcherdeveloped coding protocol. To determine which students attended behavior-focused alternative or traditional middle schools, the coders coded the schools using publicly available school information from the LEA’s website. The coding protocol included four areas that the coders assessed: (a) Is the school identified by the LEA as an alternative, charter, intensive, transformative, or something other than comprehensive PK–12 school? (b) Do the majority of students attend by choice or by referral/assignment? (c) Is the school focused on a specific skill area (e.g., arts, technology, language) or targeting a specific student population identified as “at-risk”? and (d) Is this school aimed at academic recovery or behavior modification? The answers provided to the four questions helped the coders to determine a final school type. Schools coded as traditional were identified as comprehensive, regular middle schools with the following characteristics: they served a majority of students who attended because the school served their geographic neighborhood; they did not target any particular student demographic or specific skill area; and they did not have academic or behavior remediation as a primary aim. Schools coded as behavior-focused alternatives were identified as nontraditional and also identified as providing behavior remediation for students referred or assigned due to behavioral difficulty as a primary aim. Schools coded as other types of alternative schools (e.g., innovative, academic-focused) were not included in the study. Coding reliability. The two coders reviewed all schools in the district with 91% reliability. The coders discussed any discrepancies in person until they reached 100% agreement on the final school code. A total of 108 middle schools were included in the data set. Of those, 55 were traditional middle schools and five were behavior-focused alternative middle schools. The remaining 48 middle schools were coded as other types of alternative schools. Outcome Measures Reading. We used students’ fifth grade reading scores from a statewide-standardized test for the first research question and eighth grade reading scores from the same test for the second research question. The state uses four categories to determine students’ proficiency levels in reading: advanced, Afacan and Wilkerson proficient, basic, and minimal. The following proficiency level ranges were used to place students into these categories respectively: 690 to 497, 496 to 444, 443 to 401, and 400 to 290 for fifth grade and 790 to 539, 538 to 480, 479 to 445, and 444 to 330 for eighth grade. Suspension. The participating LEA maintains a district-wide suspension policy and disciplinary handbooks. Individual school administrators must review charges that are made against a student and determine whether the student has violated the Code of School/Classroom Conduct. This determination is made in consultation with the student and appropriate witnesses. Each suspension is accompanied by communication to the student’s family that includes the specific reason for the suspension. We used the number of suspensions each student received in fifth grade and eighth grade as the behavioral outcomes in the current study. The data are number of cumulative days suspended across the school year. Data Analyses Research question 1. To answer our first research question, we used descriptive statistics to summarize characteristics (i.e., gender, race/ethnicity, special education status, and FRM status) and previous outcomes (i.e., fifth grade standardized reading test scores and fifth grade suspensions) of students attending behavior-focused alternative and traditional middle schools. Research question 2. To answer our second research question, we used a propensity score matching (PSM) technique. A propensity score indicates the conditional probability that a participant can be assigned to one condition rather than another (e.g., assigned to a treatment group rather than a control group). To predict the participant’s condition, a set of theoretically or statistically identified covariates is used in the analyses (Rosenbaum & Rubin, 1983). PSM is helpful when observational data are used because it provides nearly the same conditions as randomization for control and treatment groups. Two unequal groups are balanced and the effectiveness of the treatment condition over a control condition is tested (Rosenbaum & Rubin, 1983). Propensity score matching. We followed Guo and Fraser’s (2010) three-step procedure for conducting PSM. First, we conducted logistic regression to calculate propensity scores based on the eight student characteristics: gender (male), EBD, Black, Hispanic, other health impairment (OHI), special education status, fifth grade suspension, and fifth grade standardized reading test score. OHI was included as a potential covariate because of the large percentage of students with OHI in our sample. The school type was the binary treatment condition in the run of the logistic regression. School type was dummy coded as 0 indicating atten- 5 dance at a traditional school and 1 indicating attendance at a behavior-focused alternative middle school. This process resulted in a propensity score between 0 and 1 for each student in the data set. We checked if this logistic regression model was a good fit with the data using the Hosmer– Lemeshow goodness-of-fit test. A nonsignificant test result suggested that the logistic regression model with the eight 2 predictors fit the data well ( χ8 = 6.19, p = .63). Second, we used nearest neighbor pair matching or oneto-one matching method (without caliper) to match students attending behavior-focused alternative middle schools with students attending traditional middle schools. Guo and Fraser (2010) described the purpose of this stage as “balanc[ing] the two groups on the observed covariates” (p. 138). We conducted balancing analyses to check if the matching worked by comparing the two groups for significant differences. The two groups were compared using the aforementioned eight student characteristics and they were no longer statistically significantly different after the matching. This suggested that an unbiased estimate of difference between two groups could be made based on the selected covariates. An average treatment effect on treated (ATT) was calculated for each outcome measure. The t test statistics were calculated to determine the significance of difference between two group means. In addition, we calculated effect sizes and confidence intervals for each outcome variable of the study. Reading. As a third step, postmatching estimates were performed on the matched sample. For the eighth grade standardized reading test outcome, we used multiple regression. Eighth grade standardized reading test score was the dependent variable in the regression test. In the beginning model, we entered three predictors: (a) fifth grade standardized reading test score, (b) school type (0 = traditional, 1 = behavior-focused), and (c) special education status (0 = no disability label, 1 = yes, receiving special education services for a labeled disability). This model, which included three predictors, accounted for 43.6% of the variance in the dependent variable. Then, we entered seven additional predictors: OHI (0 = no, 1 = yes), Black (0 = no, 1 = yes), Hispanic (0 = no, 1 = yes), EBD (0 = no, 1 = yes), FRM (0 = not qualified, 1 = yes, qualified), gender (0 = female, 1 = male), and fifth grade suspensions. This model, including 10 predictors, explained 45% of variance in the dependent variable (R2 = .45). None of the seven additional predictors entered in the model significantly predicted the dependent variable; thus, they were excluded. The test statistic between the two models was not significantly different, F(7, 147) = .696, p = .676. This verified that exclusion of the seven variables was legitimate. We conducted and checked assumptions of multiple regression in SPSS 22. We checked multicollinearity by calculating variance inflation factor (VIF) for each predictor in 6 Behavioral Disorders 00(0) Table 1. Baseline Characteristics of the Sample (N = 5,929). Student baseline characteristics Male Ethnicity Black Hispanic White Asian Qualified for FRM Students with SPED status EBD OHI Fifth grade reading score Fifth grade suspensions Traditional schools n = 5,848 (98.6%) BFAS n = 81 (1.4%) 3,083 (52.7%) 54 (66.7%) 3,782 (65.1%) 1,213 (20.9%) 597 (10.3%) 214 (3.7%) 4,921 (84.1%) 1,611 (27.5%) 115 (2.0%) 655 (11.2%) 74 (91.4%) 4 (4.9%) 3 (3.7%) 0 (0%) 68 (84.0%) 43 (53.1%) 13 (16.0%) 19 (23.5%) M (SD) M (SD) 442.71 (50.75) .62 (1.50) 418.75 (54.94) 2.59 (3.80) Note. BFAS = behavior-focused alternative schools; FRM = free or reduced price meals; SPED = special education; EBD = emotional or behavioral disorder; OHI = other health impairment. the model. A mean 1.20 VIF score, with a range from 1.31 and 1.00, suggested no multicollinearity existed in the regression model. The normality assumption was checked with a histogram and was satisfied in this analysis. Suspensions. For the eighth grade suspensions outcome, we used zero-inflated Poisson (ZIP) regression analysis because the dependent variable was a count variable and included excessive zeros. It is recommended to use ZIP for data sets having count dependent variables with excessive zeros (Cheung, 2002). We applied Vuong hypothesis test to check if the ZIP regression was appropriate to use. A significant test result suggested that the ZIP model was appropriate for the suspension outcome (z = 3.42, p < .001). We report results from the ZIP model including a binary independent variable, school type, along with three other predictors: gender (male), EBD, and fifth grade suspension to predict the change in the log count of eighth grade suspensions. Measures of regression model fit were McFadden R2 = .06 and Bayesian information criterion (BIC) = −103.205. We completed the PSM and ZIP regression in Stata 14. Results Do the Fifth Grade Characteristics of Students Attending Behavior-Focused Alternative Middle Schools Differ From Students Attending Traditional Middle Schools? Table 1 shows results from the descriptive analyses. Percentages of male and female students differed by school type, with the percentage of males higher than the percentage of females in behavior-focused alternative middle schools (66.7% vs. 33.3%). In comparison, male and female students had nearly equal representation in traditional middle schools (52.7% vs. 47.3%). Student ethnicity also differed across behavior-focused alternative and traditional middle schools. The percentage of Black students was higher in behavior-focused alternative middle schools than in traditional schools (91.4% vs. 65.1%), whereas the percentage of Hispanic students was lower in behavior-focused alternative middle schools than in traditional middle schools (4.9% vs. 20.9%). Only three White students (3.7%) attended behavior-focused alternative middle schools, with no enrollment reported for Asian students. FRM status was nearly equal across behavior-focused alternative middle schools (84.0%) and traditional middle schools (84.1%). The percentage of students identified with disability was higher in behavior-focused alternative middle schools than in traditional middle schools (53.1% vs. 27.5%). Furthermore, the percentage of students identified with an EBD was higher in behavior-focused alternative middle schools than in traditional middle schools (16.0% vs. 2.0%). Similarly, the percentage of students identified with OHI was higher in behavior-focused alternative middle schools than in the traditional middle schools (23.5% vs. 11.2%). The mean fifth grade reading test score for students attending behavior-focused alternative middle schools was lower than for students attending traditional middle schools (418.75 vs. 442.71). The mean number of fifth grade suspensions experienced by students attending behavior-focused alternative middle schools was higher than for students attending traditional middle schools (2.59 vs. 0.62). 7 Afacan and Wilkerson Table 2. Balance of Variable Means Before and After Matching for Fifth Grade Student Characteristics. Before matching Variable Male EBD Black Hispanic OHI SPED status Suspension Reading After matching BFAS Traditional t BFAS Traditional t % R bias .66 .16 .91 .05 .23 .52 2.60 418 .52 .02 .65 .21 .11 .27 0.62 443 2.51* 8.89*** 4.98*** −3.52*** 3.48*** 6.21*** 11.14*** −4.44*** .66 .16 .91 .05 .23 .52 2.60 418 .64 .14 .91 .05 .29 .55 2.53 419.2 0.33 0.44 0.00 0.00 −0.90 −0.48 0.13 −0.03 83.0 83.0 100.0 100.0 48.2 87.1 96.2 99.0 Note. BFAS = behavior-focused alternative school; % R Bias = percentage of reduced bias; EBD = emotional or behavioral disorder; OHI = other health impairment; SPED = special education; Suspension = fifth grade suspensions; Reading = fifth grade standardized reading score. *p < .05. ***p < .001. Table 3. Mean Eighth Grade Reading and Suspensions Before and After Matching (n =158). Before matching Outcome Reading Suspension BFAS 421.49 3.14 After matching Traditional BFAS Traditional SE 486.96 1.26 421.49 3.14 455.35 3.07 9.33 0.55 t −3.63*** 0.11 d 95% CI −.58 .02 [–0.89, –0.26] [–0.29, 0.33] Note. CI = confidence interval; BFAS = behavior-focused alternative school; Reading = eighth grade standardized reading score; Suspension = number of suspension received during eighth grade. ***p < .001. Do Eighth Grade Students Attending BehaviorFocused Alternative Middle Schools Experience Improved School Outcomes Compared With Students With Similar Fifth Grade Characteristics Who Remain in Traditional Middle Schools? To answer our second research question, the first step was to balance the treatment (i.e., behavior-focused alternative) and the control (i.e., traditional) groups on their observed differences. To make an unbiased estimate of difference between these two groups, we performed nearest neighbor one-to-one matching using students’ propensity scores from the logistic regression. Table 2 shows the balancing analysis before and after the matching. All variables were significantly different across the two school types before matching; however, the differences in these variables were no longer significant after matching. Table 3 shows means for each school group before and after the matching for the mean eighth grade standardized reading test scores and mean eighth grade suspensions received. ATT results indicate if there are any significant differences between the means of the treatment and control groups after the matching. After matching, the mean eighth grade standardized reading test score was 421.49 for the behavior-focused alternative middle schools and 455.35 for the traditional schools. The behavior-focused alternative middle schools’ eighth grade standardized reading test score mean was 33.86 points lower than the traditional schools’. This difference was statistically significant (p < .001). An effect size of d = −0.58 was associated with attending behavior-focused alternative middle schools for the eighth grade reading outcome. Another outcome of interest was the number of suspensions received by students in the eighth grade. The mean number of eighth grade suspensions received was 3.14 days for the behavior-focused alternative middle schools while it was 3.07 days for the traditional schools. The traditional schools’ mean number of eighth grade suspensions received was 0.07 days more than the behavior-focused alternative middle schools’. This difference was not statistically significant (p = .45). An effect size of d = 0.02 was calculated for the eighth grade suspensions outcome. Postmatching Estimates Reading. Table 4 reports the results from a multiple regression test with eighth grade standardized reading test score as the dependent variable and fifth grade standardized reading test score, school type (i.e., 0 = traditional vs. 1 = behavior-focused), and special education status (0 = no vs. 8 Behavioral Disorders 00(0) Table 4. Prediction of Eighth Grade Standardized Reading Test Score From Fifth Grade Standardized Reading Test Score, Special Education Status, and School Type (n = 158). Predictor Fifth grade reading SPED status School type R2 F(3, 154) B SE β 95% CI 0.445*** −33.280*** −36.820*** .436 39.72*** 0.079 8.328 7.277 .390 −.277 −.306 [0.28, 0.60] [–49.73, –16.83] [–51.19, –22.44] Note. CI = confidence interval; SPED = special education. ***p < .001. 1 = yes) as the predictors. Together, these three predictors explained 43.6% of variance and significantly predicted eighth grade standardized reading test score, F(3, 154) = 39.723, p < .001. The prediction equation is based on the unstandardized coefficients, as follows: Y= 289.082 + 0.445 ( fifth grade reading ) _ 33.280 ( special education status ) _ 36.820 ( school type ) . The slope of fifth grade standardized reading test score is 0.445. This means that for every one unit increase in fifth grade standardized reading test score, predicted eighth grade standardized reading test score increases by 0.445, controlling for school type and special education status. The slope of school type is −36.820. This means that predicted eighth grade standardized reading test scores for students attending behavior-focused alternative middle schools are 36.820 points lower than students attending traditional middle schools, controlling for fifth grade standardized reading test score and special education status. The slope of special education status is −33.280. This means that predicted eighth grade standardized reading test scores for students with disabilities are 33.280 points lower than students without disabilities, controlling for fifth grade standardized reading test score and school type. Suspensions. Table 5 shows the logit and Poisson results of the ZIP regression analysis. With regard to the log odds of an excessive number of zeros, school type was significant (B = −1.38, z = −2.77, p = .006). In the Poisson model, school type significantly predicted eighth grade suspensions (B = 0.28, z = 2.40, p = .016). As students enroll in behavior-focused alternative schools, the log count of eighth grade suspension is expected to increase by 0.28, given other variables in the model held constant. This model also showed that the log count of eighth grade suspensions is expected to increase by 0.39 for male students (p = .002), 0.39 for students identified with EBD (p = .002), and 0.02 for every unit increase in fifth grade suspension (p = .046). Discussion The purpose of this study was to examine characteristics of students who enroll in behavior-focused alternative and traditional middle schools, as well as to compare academic and behavior outcomes of matched (statistically controlled) groups of students attending these two types of middle schools. In addition to analyzing demographic data for students across behavior-focused alternative and traditional middle schools, we also used a PSM technique to compare eighth grade reading and suspension outcomes of students attending behavior-focused alternative middle schools to those of a matched sample of students attending traditional middle schools. In the following sections, we discuss the findings from these analyses organized by the two main research questions. We also provide directions for future research. Student Characteristics and Baseline Outcomes Findings from demographic analyses suggested that the majority of students attending behavior-focused alternative middle schools were male, Black, and receiving special education services compared with students attending traditional middle schools. In our sample, more than 50% of students attending behavior-focused alternative schools were identified with disabilities. With regard to specific disability categories, our findings show that higher numbers of students with EBD and students with OHI attended behavior-focused alternative middle schools. Our study also revealed differences in gender and ethnicity: Male students and students who are Black constituted 66.7% and 91.4% of students attending behavior-focused alternative middle schools, respectively. These findings were consistent with past research illustrating that higher percentages of males, students from minority backgrounds, and students with EBD attended remediation-focused alternative schools (Carver & Lewis, 2010; Lehr & Lange, 2003; Perzigian et al., 2017; Wilkerson, Afacan, Yan, Justin, & Datar, 2016). As Lehr (2004) argued, alternative schools may function as a forced choice for specific groups of students. Denying 9 Afacan and Wilkerson Table 5. Prediction of Eighth Grade Suspensions Received From Fifth Grade Suspensions Received, Gender, EBD Status, and School Type (n = 158). Predictor Fifth grade suspension Gender (male) EBD School type (Inflate) school type B SE z 0.025 0.399 0.397 0.281 −1.38 .012 .130 .128 .117 .499 1.99* 3.05** 3.09** 2.40* −2.77** 95% CI [0.00, 0.05] [0.14, 0.65] [0.14, 0.64] [0.05, 0.51] [–2.36, –0.40] Note. EBD = emotional or behavioral disorder; CI = confidence interval. *p < .05. **p < .01. students’ access to traditional general educational experiences remains a concern for students with disabilities in general, but especially for those with EBD (Lehr et al., 2009). In relation to the socioeconomic status of students attending behavior-focused alternative and traditional middle schools, the representation of students who were qualified for FRM were nearly equal across the two school types: 84% of students from both school types were qualified for FRM. This finding may be an artifact of the fact that the sample school district serves a high percentage of families living in poverty. We additionally examined the students’ baseline academic and behavior data (i.e., fifth grade) prior to their entry into middle school. Findings from baseline outcome data analyses suggest that students who eventually attended behavior-focused alternative middle schools had lower fifth grade standardized reading test scores and a higher number of fifth grade suspensions than students who remained in traditional middle schools. These findings suggest a pattern that students who experience academic challenges and also receive suspensions connected to their behavior during the elementary school years are likely to be referred to behavior-focused alternative middle schools. Middle School Outcomes Across School Types Reading. The mean standardized reading score for students attending behavior-focused middle schools was 421.49 compared with 455.35 for a matched sample of students attending traditional middle schools, demonstrating that students’ eighth grade reading performance was significantly lower in behavior-focused alternative middle schools than in traditional middle schools. A reading score of 480 on the state’s standardized reading assessment indicates a proficient reading level as determined by the state’s criteria. This finding is consistent with past research suggesting that students attending alternative schools performed lower on standardized reading tests (Wilkerson, Gagnon, Melekoglu, & Cakiroglu, 2012). It is important to note, however, that mean scores for both groups were below the proficient level in reading. Knowing that students attending behavior-focused alternative middle schools had lower reading outcomes compared with their counterparts in the matched sample attending traditional schools, it is logical to consider the characteristics and qualifications of teachers in those settings who might influence reading outcomes. Wilkerson et al. (2012) reported that teachers working at alternative schools had more years of teaching experience than their counterparts in traditional schools. This greater experience level might be expected to be associated with greater effectiveness on the part of these teachers in delivering reading instruction. Yet, results from the current study showing lower reading performance for students in the alternative middle schools could be used to support an opposite view. In a recent study, Mason-Williams and Gagnon (2017) found that teachers in alternative schools less often held a master’s degree and a degree in English language/arts than teachers in regular schools. The researchers stated that “alternative schools were left to select lesser prepared teachers in terms of degree status from a pool of applicants” (Mason-Williams & Gagnon, 2017, p. 246). To better understand why students attending alternative schools are performing lower in reading than matched peers in traditional schools, future research should examine the actual reading instruction provided to students in alternative schools along with characteristics and qualifications of teachers who provide instruction in those schools. Importantly, results from our multiple regression analyses showed that student disability status significantly and negatively predicted their eighth grade reading performance. Given that reading skills are associated with improved student outcomes in many areas, such as later school achievement, active involvement in home and community, employment, and general quality of life (Conners, 2003), improving the reading skills of students with disabilities in alternative settings is imperative. Past research suggested that once students were provided with high quality reading instruction in alternative settings, they were able to demonstrate significant improvements in reading skills within a short period of time (Allen-DeBoer, Malmgren, & Glass, 2006; Malmgren & Leone, 2000). Therefore, reading 10 intervention studies targeting students who are at risk of school failure attending behavior-focused alternative middle schools can play an important role in helping these students improve their reading skills. Another related finding from our multiple regression test was that students’ fifth grade reading performance significantly and positively predicted their eighth grade reading performance. This suggests that focus on effective reading instruction should happen as early as possible and that students who are falling behind in reading in the elementary school years—including those who also are identified for behavioral concerns— should receive targeted intervention even earlier than fifth grade. If we hope to break the pattern of low early reading achievement leading to poor school outcomes (e.g., Hernandez, 2011), our attention must focus on more than just behavior. Suspensions. The mean number of eighth grade suspensions received was 3.14 for students attending behavior-focused alternative middle schools and 3.07 for students attending traditional middle schools; this difference was not statistically significant. A typical characteristic of behavior-focused alternative schools is to help students who are struggling in traditional schools with remediating their behavior difficulties. However, students attending behavior-focused alternative middle schools did not have a significantly lower mean number of eighth grade suspensions than matched students who stayed in the traditional middle schools. This finding was similar to past research examining outcomes for youth in behavior-focused alternative schools at the high school level compared with experiences for a similar (matched and statistically controlled) group of students who remained at traditional high schools (Wilkerson, Afacan, Perzigian, et al., 2016). Future inquiry into student behavior and discipline practices in alternative schools across grade levels could help shed light on whether suspension rates are influenced by differing behavioral expectations in alternative settings at the middle and high school levels and whether students’ behavior varies across the settings. Findings from the ZIP regression analysis supported existing literature suggesting gender (i.e., male), EBD status, and number of suspensions received during the early grades were significant predictors for later suspensions (Skiba et al., 1997; Tobin & Sugai, 1999; Wilkerson, Afacan, Perzigian, et al., 2016). Authors of a review examining the existing literature on behavior interventions implemented in alternative settings reported that the majority of intervention studies did not include effective practices and the quality of these studies were relatively low (Flower, McDaniel, & Jolivette, 2011). Based on the preliminary findings from the current study and existing literature, there may be a need for more evidence-based behavior interventions to be designed and implemented for students who are at risk of school failure and attending behavior-focused alternative middle schools. Alternatively, researchers could Behavioral Disorders 00(0) examine alternative schools’ discipline systems; inclusive teams of diverse stakeholders could also be formed to analyze and re-design schools’ existing discipline systems to better address potential behavior outcome disparities that students face in alternative schools (Bal, Afacan, & Cakir, 2018; Bal, Kozleski, Schrader, Rodriguez, & Pelton, 2014; Bal, Schrader, Afacan, & Mawene, 2016). Limitations and Future Research Our study has limitations. First, our analysis of demographic characteristics of students across school types was limited due to the fact that it was a simple comparison of counts and percentages. Future research should utilize analytical approaches that provide additional rigor in the examination of representation across specific educational settings. Second, although the district’s overall low reading proficiency might be a confounding issue for our study, even with suppressed scores there was still ample variability within reading achievement scores to be able to detect differences. Future research should determine if these patterns are consistent or perhaps even more pronounced in a district with overall higher achievement. In addition, our data were limited only to students’ reading outcomes in the state’s general assessment and middle school suspensions. Future research should examine other student outcome measures such as math and ODRs. In our analysis of eighth grade outcomes, although we intentionally matched students based on their past education history to eliminate systemic differences, we did not have access to data or information to explain why some students were assigned behavior-focused alternative schools while their matched peers stayed in traditional schools. We also do not know how many years each student spent in behavior-focused alternative schools; we only know that they were enrolled there in eighth grade. Future research should examine the impact of “dosage” or “length of time” spent in an alternative school as a predictor of later school outcomes. We acknowledge that matching students based on elementary school characteristics cannot control for all differences. Some students may be transferred to alternative schools on the basis of behavior or attributes that cannot be captured by quantitative data. For this reason, we also recommend qualitative observational and interview studies to examine the educational trajectories of students who end up in behavior-focused alternative schools. Finally, another limitation was that the analyses did not take into account clustering of students within schools; future research should utilize analytical approaches that take this clustering into account. Conclusion An Office for Civil Rights report (U.S. Department of Education, Office for Civil Rights, 2012) emphasized the Afacan and Wilkerson national problem of poor outcomes in schools that serve predominantly students from minority backgrounds and students with disabilities. Findings from this study confirmed that the majority of students attending behaviorfocused alternative middle schools were Black and receiving special education services. These high percentages of students who are Black and students with disabilities in behavior-focused alternative middle schools were coupled with poor student academic and behavior outcomes. The current research underscores that the system of assigning students to behavior-focused alternative middle schools does not necessarily portend more positive outcomes for students in secondary school. It is important for diverse stakeholders to take action to develop more effective educational environments for students who struggle early on in elementary school to succeed academically and also to meet teachers’ behavioral expectations. Effective middle school options are especially important for maximizing students’ chances of positive academic and behavior outcomes in high school. Declaration of Conflicting Interests The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. Funding The author(s) received no financial support for the research, authorship, and/or publication of this article. References Allen-DeBoer, R. A., Malmgren, K. W., & Glass, M. E. (2006). Reading instruction for youth with emotional and behavioral disorders in a juvenile correctional facility. Behavioral Disorders, 32, 18–28. Arcia, E. (2007). A comparison of elementary/K-8 and middle schools’ suspension rates. Urban Education, 42, 456–469. doi:10.1177/0042085907304879 Aron, L. Y. (2006). An overview of alternative education. Washington, DC: Urban Institute. Bal, A., Afacan, K., & Cakir, H. I. (2018). Culturally responsive school discipline: Implementing learning lab at a high school for systemic transformation. American Educational Research Journal, 55, 1007–1050. doi:10.3102/0002831218768796 Bal, A., Kozleski, E. B., Schrader, E. M., Rodriguez, E. M., & Pelton, S. (2014). Systemic transformation from the ground– up: Using learning lab to design culturally responsive schoolwide positive behavioral supports. Remedial and Special Education, 35, 327–339. doi:10.1177/0741932514536995 Bal, A., Schrader, E., Afacan, K., & Mawene, D. (2016). Using learning labs for culturally responsive positive behavioral interventions and supports. Interventions in School and Clinic, 52, 122–128. doi:10.1177/1053451216636057 Balfanz, R., Herzog, L., & Mac Iver, D. J. (2007). Preventing student disengagement and keeping students on the graduation path in urban middle-grades schools: Early identification and 11 effective interventions. Educational Psychologist, 42, 223– 235. doi:10.1080/00461520701621079 Carver, P. R., & Lewis, L. (2010). Alternative schools and programs for public school students at risk of educational failure: 2007–08 (NCES 2010–026). Washington, DC: U.S. Department of Education, National Center for Education Statistics, Government Printing Office. Cheung, Y. B. (2002). Zero-inflated models for regression analysis of count data: A study of growth and development. Statistics in Medicine, 21, 1461–1469. doi:10.1002/sim.1088 Chiang, H., & Gill, B. (2010). Student characteristics and outcomes in alternative and neighborhood high schools in Philadelphia. Cambridge, MA: Mathematica Policy Research. Conners, F. A. (2003). Reading skills and cognitive abilities of individuals with mental retardation. International Review of Research in Mental Retardation, 27, 191–229. Drame, E. R. (2010). Measuring academic growth in students with disabilities in charter schools. Education and Urban Society, 42, 379–393. doi:10.1177/0013124510362031 Flower, A., McDaniel, S. C., & Jolivette, K. (2011). A literature review of research quality and effective practices in alternative education settings. Education and Treatment of Children, 34, 489–510. Foley, R. M., & Pang, L. S. (2006). Alternative education programs: Program and student characteristics. The High School Journal, 89, 10–21. doi:10.1353/hsj.2006.0003 Gagnon, J. C., Gurel, S., & Barber, B. R. (2017). State-level analysis of school punitive discipline practices in Florida. Behavioral Disorders, 42, 65–80. Guo, S., & Fraser, M. W. (2010). Propensity score analysis: Statistical methods and applications. Thousand Oaks, CA: SAGE. Hernandez, D. J. (2011). Double jeopardy: How third-grade reading skills and poverty influence high school graduation. Baltimore, MD: The Annie E. Casey Foundation. Lassen, S. R., Steele, M. M., & Sailor, W. (2006). The relationship of school-wide positive behavior support to academic achievement in an urban middle school. Psychology in the Schools, 43, 701–712. doi:10.1002/pits.20177 Lehr, C. (2004). Alternative schools and students with disabilities: Identifying and understanding the issues. Information Brief: Addressing Trends and Developments in Secondary Education and Transition, 3(6). Retrieved from https://conservancy.umn.edu/handle/11299/172844 Lehr, C. A., & Lange, C. M. (2003). Alternative schools serving students with and without disabilities: What are the current issues and challenges? Preventing School Failure, 47, 59–65. doi:10.1080/10459880309604431 Lehr, C. A., Lanners, E. J., & Lange, C. M. (2003). Alternative schools: Policy and legislation across the United States (Research Report 1). Minneapolis: University of Minnesota, Institute on Community Integration. Lehr, C. A., Tan, C. S., & Ysseldyke, J. (2009). Alternative schools: A synthesis of state-level policy and research. Remedial and Special Education, 30, 19–32. doi:10.1177/0741932508315645 Malmgren, K. W., & Leone, P. E. (2000). Effects of a short-term auxiliary reading program on the reading skills of incarcerated youth. Education and Treatment of Children, 23, 239–247. 12 Mason-Williams, L., & Gagnon, J. C. (2017). An analysis of teacher sorting in secondary special education and alternative schools. The Journal of Special Education, 50, 239–250. doi:10.1177/0022466916656174 McIntosh, K., Flannery, K. B., Sugai, G., Braun, D., & Cochrane, K. (2008). Relationships between academics and problem behavior in the transition from middle school to high school. Journal of Positive Behavior Interventions, 10, 243–255. doi:10.1177/1098300708318961 Mendez, L. M. R., & Knoff, H. M. (2003). Who gets suspended from school and why: A demographic analysis of schools and disciplinary infractions in a large school district. Education and Treatment of Children, 26, 30–51. Mizelle, N. B., & Irvin, J. L. (2000). Transition from middle school to high school: What research says. Middle School Journal, 31, 57–61. doi:10.1080/00940771.2000.11494654 Perzigian, A. B., Afacan, K., Justin, W., & Wilkerson, K. L. (2017). Characteristics of students in traditional versus alternative high schools: A cross-sectional analysis of enrollment in one urban district. Education and Urban Society, 49, 676– 700. doi:10.1177/0013124516658520 Raywid, M. A. (1994). Alternative schools: The state of the art. Educational Leadership, 52, 26–31. Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensity score in observational studies for causal effects. Biometrika, 70, 41–55. doi:10.2307/2335942 Ruspini, E. (2002). Introduction to longitudinal research. New York, NY: Routledge. Skiba, R. J., Peterson, R. L., & Williams, T. (1997). Office referrals and suspension: Disciplinary intervention in middle schools. Education and Treatment of Children, 20, 295–315. Sugai, G., Horner, R., & McIntosh, K. (2016). Multi-tiered systems of behavior support: Implementing PBIS at scales of social significance. Retrieved from https://www.pbis.org/Common/Cms/ files/pbisresources/MTSS-B-Implement%202016%2005-16.pdf Behavioral Disorders 00(0) Tobin, T. J., & Sugai, G. M. (1999). Using sixth-grade school records to predict school violence, chronic discipline problems, and high school outcomes. Journal of Emotional and Behavioral Disorders, 7, 40–53. doi:10.1177/106342 669900700105 U.S. Department of Education, Office for Civil Rights. (2012). Helping to ensure equal access to education: Report to the president and secretary of education, under section 203(b)(1) of the department of education organization act, FY 2009-12. Washington, DC: Author. U.S. Department of Education, Office of Special Education and Rehabilitative Services, Office of Special Education Programs. (2016). 38th annual report to congress on the implementation of the individuals with disabilities education act. Washington, DC. Retrieved from https://www2.ed.gov/about/reports/annual/ osep/2016/parts-b-c/38th-arc-for-idea.pdf Vincent, C. G., & Tobin, T. J. (2011). The relationship between implementation of school-wide positive behavior support (SWPBS) and disciplinary exclusion of students from various ethnic backgrounds with and without disabilities. Journal of Emotional and Behavioral Disorders, 19, 217–232. doi:10.1177/1063426610377329 Wilkerson, K. L., Afacan, K., Perzigian, A. B., Justin, W., & Lequia, J. (2016). Behavior focused alternative schools: Impact on student outcomes. Behavioral Disorders, 41, 81– 94. doi:10.17988/0198-7429-41.2.81 Wilkerson, K. L., Afacan, K., Yan, M., Justin, W., & Datar, S. (2016). Academic remediation focused alternative schools: Impact on student outcomes. Remedial and Special Education, 37, 67–77. doi:10.1177/0741932515620842 Wilkerson, K. L., Gagnon, J. C., Melekoglu, M. A., & Cakiroglu, O. (2012). Reading instruction in secondary day treatment and residential schools for youth with emotional or behavioral disorders. Remedial and Special Education, 33, 78–88. doi:10.1177/0741932510364546