Student Perceptions of School Climate as Predictors of Office Discipline Referrals

Abstract

Research indicates that school climate influences students’ academic, social, and behavioral outcomes. Therefore, improving school climate provides a promising avenue for preventing academic, social, and behavioral difficulties. Research has examined school-level measurement of school climate, but few studies have examined student-level responses to school climate and student perceptions of school climate and their academic, social, and behavioral performance in school. In this study, we examined latent classes of students, based on their perception of school climate, and identified specific items within each class that predicted student social and behavioral performance as measured by office discipline referrals (ODR). Finally, we explored the academic, social, and behavioral profiles and demographic profiles within each class and discussed implications for practice and research.

Keywords

achievement gap early childhood policy poverty

Successful schools create environments that promote academic achievement, social competence, and prosocial behaviors by providing safe, orderly, and positive learning environments for all students (Bradshaw, Koth, Thornton, & Leaf, 2009; Herman et al., 2008). School climate, or the quality and character of school life (Cohen, McCabe, Michelli, & Pickeral, 2009), influences students’ social and behavioral outcomes (D. Gottfredson, 2000; G. D. Gottfredson, Gottfredson, Payne, & Gottfredson, 2005; McIntosh, Chard, Boland, & Horner, 2006; Walker & Shinn, 2002). Therefore, improving school climate provides a promising avenue for preventing social and behavioral difficulties. Further, successful student behavior support has been linked to school environments and school climates that are effective, safe, preventive, and positive (Bradshaw, Koth, Bevans, Ialongo, & Leaf, 2008; Horner, Sugai, & Anderson, 2010), supporting the relationship between school climate and behavior. However, what facets of school climate are predictive of positive social and behavioral outcomes is unclear. We designed this study to identify (a) classes of students based on risk status for school-based behavioral difficulties and (b) specific facets of school climate that are predictive of decreased risk.

School Climate

School climate is a complex multidimensional construct encompassing the atmosphere, culture, values, resources, and social networks of a school (Thapa, Cohen, Guffey, & Higgins-D’Alessandra, 2013; Wang & Degol, 2015) and has been defined as the shared beliefs, values, and attitudes that shape interactions between students, teachers, and administrators and set the parameters of acceptable behavior and norms for the school (Cohen et al., 2009; Fan, Williams, & Corkin, 2011; Koth, Bradshaw, & Leaf, 2008). School climate is based on patterns of student and teacher experiences of school life and reveals norms, goals, values, interpersonal relationships, teaching and learning practices, and organizational structures of schools that support feeling socially, emotionally, and physically safe in school (Cohen et al., 2009).

Positive school climate has been linked to a number of academic, social, and behavioral outcomes, including academic achievement (Brand, Felner, Shim, Seitsinger, & Dumas, 2003; Wang, Selman, Dishion, & Stormshak, 2014); student academic, social, and personal attitudes and motives in school (Battistich, Solomon, Kim, Watson, & Schaps, 1995); increased attendance (Brand et al., 2003; Welsh, 2000); and decreased student delinquency (G. D. Gottfredson et al., 2005; Welsh, 2000), use of illegal substances (Brand et al., 2003), bullying (Nansel et al., 2001), victimization (Gage, Prykanowski, & Larson, 2014; G. D. Gottfredson et al., 2005), depression and self-esteem (Brand et al., 2003; Way, Reddy, & Rhodes, 2007), and general behavior problems (Battistich & Horn, 1997; Kuperminc, Leadbeater, & Blatt, 2001; Welsh, 2000). Therefore, measurement and use of school climate data may be a promising approach to addressing both school-wide and individual problem behaviors.

School Climate and Behavior

Irvin, Tobin, Sprague, Sugai, and Vincent (2004) contend that school climate varies as a function of several factors, including (a) student behavior and attitudes, (b) school and classroom characteristics, and (c) educator and student values and related perceptions regarding considerations of school safety and school effectiveness. The interrelationship between school climate and behavior, particularly behavior problems in schools, has received attention in the research literature (G. D. Gottfredson et al., 2005). For example, Wang (2009) examined the relationship between middle school students’ perceptions of school climate and students’ deviant behaviors and depressive symptoms. Results suggested that adolescents who perceived their schools to have positive school climate were less likely to engage in deviant behaviors and report depressive symptoms. However, the study utilized structural equation modeling and examined path coefficients between latent constructs, thereby not identifying specific aspects of school climate influencing student behaviors. Wang et al. (2010) conducted a more nuanced study of school climate, examining students’ perception of school climate in sixth grade and the probability of engaging in problem behaviors in seventh and eighth grade as reported by student. They found that positive student perceptions of school climate, particularly positive teacher-student interactions, were related to decreased frequency of reported problem behaviors. Two limitations should be noted: (a) The study did not utilize a school climate measure but instead a measure of social nomination, and (b) the measure of student problem behavior was self-report, which may have been biased.

From a social cognitive perspective (Bandura, 2001), students tend to react to experiences as they subjectively perceive them, not necessarily to the objective nature of the experience (Koth et al., 2008). Consequently, students’ perceptions of the school environment may have an impact on their behavior at school. Further, researchers agree that student personal experiences of the school climate mediate actual school climate effects on their behavior (Kuperminc et al., 2001; Loukas & Robinson, 2004; Roeser, Eccles, & Sameroff, 1998). Individual perceptions of school climate appear to contribute to student outcomes, including problem behavior (Loukas & Murphy, 2007). Therefore, school climate is an important target for school improvement initiatives that aim to reduce discipline problems (Haynes, Emmons, & Ben-Avie, 1997). Unfortunately, little is known about which specific facets of school climate (e.g., positive teacher interactions) are predictive of positive behavioral outcomes.

Purpose

School climate may serve as a protective factor for student-level problem behavior; however, there is a need to determine if certain facets of school climate predict decreases in problem behaviors. Further, if a predictive relationship is established between specific facets of school climate, those facets could be targets for intervention and have positive distal impacts on the frequency of office discipline referrals (ODRs). Therefore, in this study, we identified latent classes of students based on their frequency of ODRs and their perceptions of school climate in order to identify the predictive relationship between specific facets of school climate and ODRs within each identified latent class. In addition, we examined the academic, social, and behavioral profiles and demographic profiles of each class to create an overall profile of each class of student based on the relationship between ODR and student perceptions of school climate.

Method

Sample

Data were collected from a large school district in New England, comprised of eight elementary schools, two middle schools, and two high schools, with a total enrollment of ~8,200 students. The district serves a diverse student population, with 62.2% receiving free or reduced lunch; 11.5% not fluent in English; 30.4% of district’s students coming from homes where English is not the primary language, including 43 different languages; and 13.4% receiving special education services. Total minority population was 61.1%, including 44.7% Hispanic, 13.5% African American, 2.5% Asian, and 0.3% Native American. The district reported a 91.2% graduation rate and a 3.0% dropout rate for Grades 9 through 12. Of those graduating, approximately 71.6% pursue a higher education degree.

A sample of 3,797 students (~46% of the district) completed the school climate survey. Teachers of students in 3rd through 12th grades were invited to participate in the study. Sixty-eight percent of teachers consented, brought all students in their classroom to a school computer lab, and asked them to complete the online survey. Students were not excluded from participation, and if students had questions or needed item clarifications, teachers would assist by reading the item out loud and prompting students to respond. Forty-five percent of respondents were in elementary school, 22% in middle school, and 32% in high school. Exactly 50% of the sample was female, 40% were White, 42% were Hispanic, and 15% were African American. Sixty-three percent of the sample received free or reduced lunch, 7% were classified as English language learners by the district, and 6% received special education services, primarily for learning disabilities (55% of students with disabilities in the sample).

Measures

Meriden School Climate Survey-Student Version

The Meriden School Climate Survey-Student Version (MSCS-SV; Gage, Larson, & Chafouleas, 2016) is a 47-item survey of students’ perception of adult support at school; school safety; respect for differences; adult support at home, including academic support; student aggression toward others; and peer support (see Table 4 for specific item definitions). Preliminary psychometric evaluation identified a seven-factor structure with acceptable reliability at the full scale and subscale levels (e.g., full scale alpha = .91). The MSCS-SV is delivered online to students in Grades 3 to 12 during grade-level testing sessions. Data for this study were collected in November and December of 2011 during regular school hours as part of the initial validation study; therefore, not all students in the district completed the survey. Although subscale factor scores were available, the goal of this study was to identify predictive relationships between specific facets of school climate (e.g., having a friend that a student can trust), not broader subscales. For example, the adult support subscale provides a single score for all teacher-related items, but in this study, we wanted to identify the specific items on the survey that predicted class membership. Therefore, modeling was conducted at the item level.

Office Discipline Referrals

All schools were using the School-Wide Information System (SWIS; May et al., 2005) to collect ODR data. SWIS provides a database framework for systematically defining, collecting, and presenting individual and school-wide patterns of ODRs. School personnel entered ODR data into SWIS, and each record included the type of problem behavior and administrative action resulting from the problem behavior (e.g., in-school suspension). The total number of ODRs reported for the sample of students in this study was 3,778 ODRs, recorded from August 2011 through March 2012. The majority of ODRs reported in the study sample were for student disrespect (n = 909), followed by aggression toward a peer (n = 482), disruptions (n = 458), and inappropriate language (n = 299). For this study, the number of ODRs per student was summed and recorded as the ODR value for each student. A total of 1,190 students received at least one ODR (range of 0 to 35 ODRs). The average per student rate of ODR was 0.47 ODRs. The average per student ODR rate for the district during the 2011–2012 school year was 0.45 ODRs (4,821 ODRs committed by 8,227 students).

Suspension

In addition to ODRs, administrative decisions were collected in the SWIS system. Two administrative decisions were retained for analysis: in-school and out-of-school suspension. Not all ODRs resulted in a suspension; however, students with repeated ODR occurrences and students exhibiting significant behavior problems (e.g., fighting, bringing weapons to school) received more severe administrative decisions, including suspensions. To calculate in-school and out-of-school suspensions, the total number of days suspended were summed for each student. A total of 327 students received in-school suspension, and total of 110 students received out-of-school suspension. Across all students, the mean number of days of in-school suspension was 0.27 days (SD = 1.38), with a maximum of 20 days. Across all students, the mean number of days of out-of-school suspension was 0.17 days (SD = 1.26), with a maximum of 22 days. District-wide and school-wide suspension data were not available.

Academic Achievement

During the time of data collection, all students in Connecticut were assessed in reading, mathematics, and writing on the Connecticut Mastery Test (CMT) for Grades 3 through 8 and the Connecticut Academic Performance Test (CAPT) for Grade 10, each providing a level of academic performance. No academic achievement level was available for Grades 9, 11, and 12. The reading test included two reading tests, the Degrees of Reading Power and a Reading Comprehension test. The mathematics test assessed student mastery of grade-specific mathematics skills and concepts, and the writing tests included two tests, the Direct Assessment of Writing and the Revising and Editing tests. Both the CMT and the CAPT provide a grade-specific level of performance for each student in each academic content area (reading, mathematics, and writing). The levels include Level 1 below basic, Level 2 basic, Level 3 proficient, Level 4 goal, and Level 5 advanced. Each corresponding level number was retained as the academic achievement score for each student (e.g., a value of 1 for below basic). Across all students with available data, the mean mathematics performance was 3.33 (SD = 1.21, n = 2,868), mean reading performance was 2.92 (SD = 1.31, n = 3,383), and mean writing performance was 3.10 (SD = 1.20, n = 2,951). To confirm sample equivalence with district-level average performance, we compared the percentage of students at or above goal across the district and in the study sample. The percentage of students in the study sample at or above goal was 50.5% in math and 50.8% in reading. The percentage of students in all elementary and middle schools at or above goal in elementary and middle school was 49.1% in math and 53.8% in reading. The percentage of study sample students in high school at or above goal was 23.1% in math and 17.4% in reading. The percentage of al high school students at or above goal was 27.4% in math and 20.7% in reading.

Grade

The grade of each student was included in the data set, and each grade (e.g., third) was retained as a numeric value (e.g., 3).

Data Analysis

The purpose of this study was to identify whether specific items or clusters of items from the school climate survey predicted student ODRs. Underlying the assumption of item clustering is the presence of latent classes of students based on their response patterns on the school climate survey and their respective number of ODRs. That is, each observation (student) is a member of one, and only one, of T latent, or unobserved, classes (Magidson & Vermunt, 2004) contingent on patterns of ODRs. Statistically, this modeling approach is called latent class regression modeling (also referred to as finite mixture regression). The goal of latent class regression is the identification of classes of students based on patterns of responses on the school climate survey with respect to student ODRs.

The analysis was conducted using LatentGOLD 4.5 (Vermunt & Magidson, 2005), a Windows-based statistical software program designed specifically for latent class regression modeling, utilizing maximum likelihood and posterior mode estimation and Bayes constants to eliminate boundary solutions. Because each student’s cumulative number of ODRs served as the dependent variable (i.e., a count variable), a Poisson distribution was used, resulting in a log-linear Poisson latent class regression model. To address the heterogeneity across grades (3rd–12th), the grade variable was included as a covariate across all models. Six models were sequentially calculated, using all 47 items from the MSCS-SV as predictors, and compared for best model fit. Each model successively increased the number of possible classes, from one class to six classes following procedures outlined by Vermunt and Magidson (2005). The determination of the best fitting model and final number of latent classes was made based on the minimization of the Bayesian Information Criteria (BIC), the Akaike Information Criterion (AIC), the minimization of classification error, and the proportion of total variance explained (R²) (Lanza, Collins, Lemmon, & Schafer, 2007). The modeling was conducted with a complete case analysis approach to missing data. Across all items, an average of 1.5% of data were missing (range, 0% to 3.1%). Assuming missing at random, missing data were modeled in LatentGOLD 4.5, which utilizes the expectation maximization (EM) and maximum likelihood estimates in the parameter calculations (Harel, Pellowski, & Kalichman, 2012).

Based on the final model, each student’s classification was retained and used as the fixed factor in multivariate analysis of covariance (MANCOVA) model with reading achievement, math achievement, writing achievement, ODRs, in-school suspension, and out-of-school suspension as dependent variables and controlling for grade. The MANCOVA model included Helmert contrasts and univariate test with Tukey honest significant difference (HSD) post hoc comparisons performed after obtaining a significant multivariate effect. The MANCOVA was developed to assess whether or not latent classes of students’ social-behavioral and academic performance in school were significantly different by identified latent class. The MANCOVA was conducted in SPSS 19.0. To examine class differences across demographic variables, including gender, ethnicity, lunch status, English learner status, and special education status, we calculated descriptive statistics by class and χ² tests to determine whether or not differences were statistically significant.

Results

Latent Class Model Identification

In this study, we examined latent classes of students based on their perceptions of school climate and ODRs to identify clusters of responses predicting students in need of school-based social and behavioral interventions. Model fit statistics and descriptive features of each of six latent class Poisson regression models are presented in Tables 1 and 2. The goal with the BIC and AIC is to identify the model with the lowest value (Magidson & Vermunt, 2004). The two-class regression model had the smallest BIC, whereas the four-class model had the lowest AIC. The two-class model only explained 84% of the variance, and the classification error was impacted by the 50/50 chance of classifying students accurately. The four-class regression model accounted for 95% of the variance, but the classification error was high at almost 50%. Based on all available information, the three-class regression model was retained as the final model, having the lowest BIC and AIC with correspondingly low classification error and high variance explained (94%). The choice of the three-class regression model was further confirmed by examining the class size and mean ODRs. The two-class regression model included 92% of the sample in a single group. The four-class regression model appeared to segment the sample into groups with face validity (i.e., three approximately equal groups and one outlier group with high mean ODRs). However, as indicated by the results in Table 1, the error rate was almost 50% across the four groups. Therefore, based on the model fit and descriptive statistics, the three-class regression model was retained.

Table 1

Latent Class Poisson Regression Model Fit Statistics, Classification Errors, and Variance Explained

Model	Number of Classes	LL	BIC	AIC	Class Err	R ²
1	One-class regression	−3,445.48	7,282.51	6,986.98	0.00	0.16
2	Two-class regression	−2,257.41	5,314.22	4,710.83	0.02	0.84
3	Three-class regression	−2,086.69	5,380.64	4,469.39	0.17	0.94
4	Four-class regression	−1,978.50	5,572.10	4,353.01	0.49	0.95
5	Five-class regression	−1,936.11	5,895.19	4,368.24	0.48	0.97
6	Six-class regression	−1,897.08	6,224.98	4,390.17	0.51	0.97

Note. LL = log likelihood; BIC = Bayesian Information Criterion; AIC = Akaike Information Criterion; Class Err = classification error.

Table 2

Class Size and Class-Specific Mean of ODRs

Model		Class 1	Class 2	Class 3	Class 4	Class 5	Class 6
One-class model
	Size (%)	100
	M ODR	0.45
Two-class model
	Size (%)	92	8
	M ODR	0.14	4.02
Three-class model
	Size (%)	74	22	4
	M ODR	0.11	2.92	9.66
Four-class model
	Size (%)	41	31	23	5
	M ODR	0.59	2.65	6.48	7.58
Five-class model
	Size (%)	33	28	23	13	3
	M ODR	0.29	1.65	1.70	4.90	9.92
Six-class model
	Size (%)	23	22	20	19	8	7
	M ODR	0.50	1.92	1.27	2.11	6.88	10.17

Note. Size is the proportion of students within each class. M ODR is the mean number of ODRs within each class. ODRs = office discipline referrals

The results displayed in Table 2 summarize the percentage of students in each class and their respective mean ODRs. The first class of students included ~74% of the sample, and their mean ODR per student was almost 0. The second class of students included ~22% of the sample, and their mean ODR was almost 3. The last class included ~4% of the sample, and their mean ODR per student was almost 10. Based on the frequency of ODRs per class, we subjectively named each class based on their need for intervention support. The first class was labeled Primary, the second class was labeled Secondary, and the third class was labeled Tertiary, similar to the school-wide positive behavior support (SWPBS; Sugai & Horner, 2002) tiered frameworks. In Table 3, we summarize the proportion of students within each of the three classes by grade level, included as a covariate to control for grade-level differences. The results suggest that the percentage of students in each class within each grade was equivalent, providing further support for the accuracy of the three-class regression model.

Table 3

Proportion Students by Latent Class by Grade Levels

Grade	Primary	Secondary	Tertiary
3–4	.15	.13	.14
5–6	.30	.29	.26
7–8	.11	.13	.16
9–10	.22	.21	.24
11–12	.22	.22	.20

Three-Class Poisson Regression Parameter Estimates

The parameter estimates from the three-class Poisson regression model are presented in Table 4. The Wald statistic, which is a chi-square test, indicates which survey items (predictors) were significant predictors within the model. Analogous to low factor loadings in factor analysis, a nonsignificant Wald statistic indicates that the survey item did not contribute significantly to class identification. For example, the survey item “At home, I have a parent or other adult who always wants me to do my school work” did not significantly contribute to the model. Therefore, only items with significant Wald statistics were interpreted. The Wald(=), also a chi-square test, indicates which item parameter estimates were significantly different from each other. For example, the survey item “At home, I have a parent or other adult who talks with me about my problems” is a significant predictor within the model, but the parameter estimates, contingent on item response patterns, are not different from each other, meaning the item does not provide discrimination between classes, which is the focus of this study. Therefore, only items with significant Wald and Wald(=) statistics were interpreted (Magidson, & Vermunt, 2004).

Table 4

Three-Class Poisson Regression Parameter Estimates and Raw Mean Score for Predictors by Class

			Primary (n = 3,075)		Secondary (n = 340)		Tertiary (n = 72)
Predictors	Wald^a	Wald(=)^b	β^c	M ^d	β	M	β	M
A person’s skin color can cause problems at my school (r)	54.67**	20.42**	0.35*	4.18	0.49**	4.21	−0.06	4.10
At home, I have a parent or other adult who always wants me to do my best	29.25**	27.05**	−0.22*	4.57	0.30*	4.45	−0.73**	4.50
At home, I have a parent or other adult who cares about my school work	6.27	6.21*	0.01	4.53	−0.06	4.39	0.52*	4.51
At home, I have a parent or other adult who expects me to follow school rules	15.52**	15.11**	0.04	4.52	−0.21*	4.40	0.79**	4.49
At home, I have a parent or other adult who listens to me when I have something to say	1.34	0.49	0.02	4.27	−0.08	4.14	0.01	4.35
At home, I have a parent or other adult who talks with me about my problems	9.05*	2.50	−0.07	4.02	0.28*	3.90	0.11	3.96
At home, I have a quiet place to do my homework	9.84*	0.09	0.13	3.95	0.10*	3.92	0.12*	4.11
At home, I have time to do my homework	7.71*	7.70*	−0.06	4.26	0.14*	4.06	−0.21*	4.25
At home, if I need help with homework, a parent or adult will help me	11.78*	3.54	0.22	4.07	0.05	3.92	0.27**	4.24
At my school, I have a friend who I can really trust	17.15**	12.12**	−0.01	4.28	−0.23**	4.09	0.22*	4.38
At my school, there is a teacher or other adult who always wants me to do my best	9.89*	9.01*	0.41*	4.41	−0.16	4.21	0.22	4.51
At my school, there is a teacher or other adult who listens to me when I have something to say	18.18**	4.92	−0.01	4.04	−0.19*	3.81	−0.38**	3.99
At my school, there is a teacher or other adult who tells me when I do a good job	55.65**	14.20**	−0.01	4.03	0.35**	3.96	0.69**	4.11
At my school, there is a teacher or other adult whom I can trust	28.70**	1.78	−0.29**	4.03	−0.18**	3.89	−0.33**	3.96
At school, the color of my skin can get me in trouble (r)	54.24**	41.06**	−0.54**	4.52	0.16*	4.41	−0.21*	4.61
During the past few months, I have hit or pushed other students at school in anger (r)	53.22**	1.26	−0.19*	4.61	−0.27**	4.31	−0.34**	4.31
During the past few months, I have hit, pushed, or spread mean rumors at the bus stop or on the bus (r)	109.16**	56.62**	−0.24*	4.75	−0.55**	4.59	0.28**	4.51
During the past few months, I have spread mean rumors or lies about other students (r)	17.30**	14.86**	0.43**	4.74	−0.04	4.61	0.36**	4.65
I am happy to be at this school	32.05**	14.56**	−0.53**	3.81	−0.04	3.54	−0.32*	3.92
I do my homework on time	23.41**	2.66	−0.05	4.19	−0.26**	3.85	−0.24**	3.86
I feel sad in school (r)	26.64**	12.21**	−0.13	3.81	−0.27**	3.69	0.07	4.14
I feel safe at school	21.78**	14.38**	0.41**	3.86	−0.08	3.68	0.32**	3.83
I feel safe on my way to and from school	24.26**	14.94**	0.03	4.14	0.24**	4.07	−0.14*	4.13
I get hit or threatened by other students (r)	35.25**	3.47	−0.18	4.51	−0.37**	4.38	−0.54**	4.60
I have a friend about my own age that really cares about me	19.79**	13.98**	0.19	4.27	−0.08	4.14	−0.45**	4.33
I have a friend about my own age who talks with me about my problems	11.53*	10.84**	−0.13	3.82	−0.11*	3.76	0.20*	3.86
I know the school rules	25.17**	18.90**	0.51**	4.37	−0.37**	4.10	−0.18	4.22
I try to do my best at school	48.95**	11.26**	−0.50**	4.47	−0.61**	4.15	−0.09	4.46
I try to understand how other students feel	35.21**	30.92**	−0.07	4.00	−0.06	3.84	0.66**	4.01
I worry about many things (r)	3.21	0.40	0.00	2.73	0.07	2.76	0.07	2.61
In class, I try to understand other students who disagree with me	6.82	4.83	−0.13	3.64	0.06	3.41	−0.19*	3.54
In the future, I feel I will be successful in life	2.91	2.53	0.02	4.40	−0.02	4.24	0.26*	4.43
My school handles student behavior problems fairly	16.21**	16.12**	−0.24**	3.88	0.25**	3.71	0.08	3.88
My teachers want me to work hard and do well	25.43**	23.68**	−0.05	4.38	0.45**	4.31	−0.18*	4.22
Other students at school have spread mean rumors or lies about me (r)	54.18**	16.20**	−0.42**	4.23	−0.34**	4.05	0.09	4.28
Other students in my school hurt my feelings (r)	69.97**	0.25	0.36**	3.96	0.42**	3.98	0.39**	4.04
Other students in this school are polite and listen to what I say	0.73	0.70	0.10	3.26	−0.03	3.14	0.02	3.32
Students being mean to other students (harassment) is a problem in my school (r)	6.47	5.23	0.15	3.22	−0.12*	3.01	0.05	3.39
Students in my school respect differences in other students	10.61*	3.90	0.03	3.48	0.05	3.41	0.25**	3.50
The adults in my school treat all students fairly	55.19**	17.00**	−0.05	3.93	−0.44**	3.80	0.14	3.92
The adults in my school treat students with respect	19.15**	17.47**	0.23*	4.17	0.19**	3.97	−0.43**	4.19
The teachers in my school make learning fun	11.81*	2.48	0.02	3.45	−0.20**	3.22	−0.11	3.42
There are teachers at my school who care about me	28.78**	6.55*	−0.40**	4.13	−0.20**	3.91	−0.06	4.08
There are teachers in my school that help me to really want to learn	21.45**	0.19	0.30*	4.12	0.23**	4.01	0.28*	4.15
There is physical fighting between students at my school (r)	19.04**	2.38	−0.24*	3.33	−0.25**	3.09	−0.09	3.54
This school wants all students to do their very best	8.66*	1.34	0.24*	4.27	0.15*	4.17	0.04	4.32
When I have a problem, I find someone to talk with	24.06**	16.41**	−0.30**	3.77	0.10	3.64	−0.25**	3.83

Note. (r) = that the survey item scaling was reverse coded.

The Wald statistic is a chi-square test and indicates whether the set of parameters are significant.

The Wald(=) statistic is also a chi-square test, but tests whether the coefficients are equivalent across classes.

The β is the estimated Poisson regression coefficients for each class controlling for student grade.

The mean is the raw mean score for each survey item (predictor) for each estimated class.

p < .05, **p < .01, ***p < .001.

Unlike linear regression, the coefficients are not the unit change in the dependent variable for each unit change in the predictor holding all other predictors constant. Instead, the Poisson regression coefficient is interpreted as follows: For a one-unit change in the predictor variable, the difference in the logs of expected counts is expected to change by the respective regression coefficient given the other predictor variables in the model are held constant. For ordinary least squares (OLS), if a given case were one unit higher on a covariate, all things being equal, the mean of the conditional distribution should be β₁ units higher. Here, if a case were one unit higher, the expected conditional mean should be e^β times higher. To determine the strength of the relationship relative to the scaling, the class-specific coefficient (β) becomes the exponent of e, and the result can be subtracted from 1 and multiplied by 100 to identify the relative percentage increase in ODR counts per one scale unit increase on of the survey item holding all other survey items constant. This calculation was conducted in Microsoft Excel for each coefficient using the following function: =1-(exp(β)).

School Climate Items Predicting Decreases in ODRs

We identified significant indicators of decreasing within class ODRs for each class of students (Primary, Secondary, and Tertiary) (Table 4). Because a number of survey items are significant predictors across different classes, we described the survey items with the largest within-class percentage of ODR change. For the Primary class of students, four survey items were related to a ~50% within-class decrease of ODRs: (a) I know the school rules; (b) During the past few months, I have spread rumors or lies about other students (r); (c) At my school, there is a teacher or other adult who always wants me to do my best; and (d) I feel safe at school.

For the Secondary class, three survey items were related to ~50% within-class decrease of ODRs: (a) A person’s skin color can cause problems at my school (r), (b) My teachers want me to work hard and do well, and (c) At my school, there is a teacher or other adult who tells me I do a good job. A fourth item was associated with a 35% within-class decrease of ODRs: At home, I have a parent or other adult who always wants me to do my best.

The Tertiary class had four items that significantly decreased within class ODRs: (a) At home, I have a parent or adult who expects me to follow school rules (~122% decrease); (b) At my school, there is a teacher or other adult who tells me I do a good job (~101% decrease); (c) I try to understand how other students feel (~94% decrease); and (d) At home, I have a parent or other adult who cares about my school work (~70% decrease). Two other items were associated with a ~40% decrease in ODRs: (e) During the past few months, I have spread mean rumors or lies about other students and (d) I feel safe at school.

Academic, Social, and Behavioral Profiles of Latent Classes

To examine whether significant differences were evident across the three latent classes of students (Primary, Secondary, and Tertiary), controlling for grade level, a MANCOVA was modeled with the latent class as the fixed factor and student academic achievement in reading, mathematic, and writing; ODRs; and in-school and out-of-school suspension as dependent variables. Because examination of Box test indicated that the model violated the assumption of homogeneity (F = 237.52, p < .000), Pillai’s Trace was used to assess the multivariate difference (Leech, Barrett, & Morgan, 2008). Overall, a significant difference was found across the three classes and the dependent variables (Pillai’s Trace = .398, F = 104.38, p = .000). Examination of the univariate effects in Table 5 indicate significant academic, social, and behavioral profiles across the three classes of students, with ODRs and in-school suspension as the dependent variables with the largest mean differences.

Table 5

MANCOVA and ANOVA Between-Subjects Effects

Variable and Sources	SS	MS	F	P	η²
Mathematics
Between	153.13	76.57	56.20	.000	0.04
Error	3,437.34	1.36
Reading
Between	193.51	96.76	55.66	.000	0.04
Error	4,385.91	1.74
Writing
Between	117.60	58.80	43.61	.000	0.03
Error	318.19	0.45
ODRs
Between	2,544.18	1,272.09	727.98	.000	0.37
Error	4,408.74	1.75
In-school suspension
Between	805.55	402.78	319.03	.000	0.20
Error	3,185.26	1.26
Out-of-school suspension
Between	203.97	101.98	95.66	.000	0.07
Error	2,689.92	1.07

Note. MS = mean square; ODRs = office discipline referrals; SS = sum of squares.

We conducted post hoc comparisons for all dependent variables because all dependent variables were significant. Across all comparisons, the majority of mean differences across classes within dependent variables were also statistically significant at the p < .05 level. However, nonsignificant differences were found between the Secondary and Tertiary classes on all three academic achievement levels, indicating that the mean academic achievement was equivalent for both classes of students. Based on the estimated margin means and their respective confidence intervals, controlling for grade, the academic, social, and behavioral profiles of the three classes of students are provided in Figure 1 with 95% confidence intervals overlaid. Across all variables, clear differences are evident between those students in the Primary class and the Tertiary class. However, the error, as indicated by the confidence bands, in the Primary class is very small, whereas the error in the Tertiary class is wider, indicating greater heterogeneity of academic, social, and behavioral profiles for students in the Tertiary class.

Figure 1.

Academic, social and behavioral profiles of students based on a 3-class Poisson regression of school climate and office disciplinary referrals.

Demographic Profiles of Latent Classes

Last, we examined the demographic profiles for each class (see Table 6). A much larger percentage of males were found in the Tertiary class, indicating that males commit significantly more ODRs than females. Differences by ethnicity indicate that more African American students were in the Secondary class than either the Primary or Tertiary, while fewer White students were in the Tertiary class than in the Primary class. Students with more ODRs are more likely to receive free or reduced lunch and more likely to receive special education services. No differences were noted across EL status. Differences within each demographic characteristic by class membership were statistically significant for all characteristics, except for EL status.

Table 6

Demographic Characteristics by Class

Demographic Characteristic		Primary (%)	Secondary (%)	Tertiary (%)	χ² (p Value)
Gender					.000
	Male	47.9	59.6	75.0
Ethnicity					.000
	White	41.8	21.1	31.3
	Hispanic	41.4	51.3	52.1
	African American	13.8	26.3	16.7
Lunch status					.000
	Free and reduced	60.6	80.7	85.4
EL Status					.742
	Yes	7.1	8.3	8.3
Special education status					.000
	Yes	5.8	11.8	11.1

Note. EL = English learner.

Discussion

We designed this study to identify latent classes of students based on their ODRs and perception of school climate and the academic, social, behavioral, and demographic profiles of identified latent classes of students. We identified a three-class model based on a priori model fit criteria. Unexpectedly, the percentage of students within each class was congruent with a multi-tiered model framework for prevention and intervention, with a progressively smaller number of students exhibiting higher frequency of ODRs and subsequently needing more intensive support. Multi-tiered models of school-wide support (e.g., SWPBS) are based on three types of students: (a) typical students not at risk for academic or behavioral problems (~80% of the student population), (b) students at risk for developing academic or behavioral problems (~15% of the student population), and (c) students with persistent and chronic maladaptive academic and behaviors (~5% of the student population; Nelson, Benner, Reid, Epstein, & Currin, 2002). These descriptions appear appropriate for the identified classes in this study and were used subjectively for the naming convention for each class. The mean number of ODRs and percentage of students for each class aligned with Nelson and colleagues’ (2002) definition, with 74% of the sample recording an average of 0.1 ODRs (typical, not at risk), 22% recording an average of 3 ODRs (students at risk), and 4% with an average of 10 ODRs (persistent and chronic behaviors).

Although a number of survey items were statistically significant predictors of class membership, those items that significantly predicted the largest decreases in within-class ODRs were highlighted. Focusing on the Tertiary class, the most important factors contributing to decreases in ODRS include having school-involved parents, a caring adult at school that reinforces appropriate behaviors, and feeling safe at school. This finding supports the need for schools to create positive learning environments where students feel safe and suggests that schools need to connect with parents so that parents (a) support school rules and (b) care about their children’s performance in schools.

Based on the classifications from the three-class regression model, we explored the academic, social, behavioral, and demographic profiles of students within each latent class. Students in the Tertiary class were performing between the basic and proficient academic levels, experiencing the most office discipline referrals, and as a result, the most number of days suspended. Across all academic, social, and behavioral profiles, the Tertiary class of students performed statistically significantly worse than the Primary class. Clearly, these students are in need of academic, social, and behavioral interventions and also were more likely to be male and receive free or reduced lunch than students in the Primary class. The academic and demographic profiles of the Secondary class were similar to the Tertiary class.

Limitations and Recommendations for Future Research

This study represents a step in a line of continued research into the relationship between student perceptions of school climate and academic, social, and behavior outcomes. With that in mind, a number of limitations necessitate highlighting. First, this study included students from a single district in New England, thereby reducing generalizability of findings. Replication is necessary with students in schools and districts from broader geographic regions to confirm that the (a) three-class model predicts within-class decreases of ODRs, particularly for students in the Tertiary class, and (b) academic, social, and behavioral profiles identified in this study are consistent. Further, the study did not include student-level characteristics due to restrictions by the district- or school-level characteristics. Although these missing covariates in and of themselves do not invalidate findings, future research should examine differential patterns of classification and academic, social, and behavioral profiles by ethnicity, gender, and disability status.

The analysis in this study controlled for grade-level differences within the analysis, but more fine-tuned, age/grade-specific analysis should be conducted to identify whether different patterns emerge for different age/grade-level students. Academic, social, and behavioral interventions may be different depending on educational context, including grade (e.g., reading interventions for first graders compared with interventions for fifth graders) and school type (social skills programs in elementary schools compared to social skills programs in high schools). Additionally, this study reports on student perceptions in the fall of the school year and the number of ODRs by the early spring. Future research should include longitudinal measures of school climate, ODRs, and measures of academic, social, and behavioral performance. In addition, interventions should be tested that address the needs of students in the Secondary and Tertiary classes, with effects measured across time and outcome. Lastly, the small sample size of the Tertiary class may have underpowered the analysis and limited the findings. However, the smaller sample resulted in larger standard errors, and the results are not significantly impacted. Future research with larger samples will attenuate this concern.

Implications for Practice

Although preliminary and exploratory in nature, the results of this study have implications that can directly impact practice. Based on the results of the latent class regression analysis, clusters of school climate survey items significantly predicted within-class decreases of ODRs. Specifically, schools where teachers provide consistent and regular positive reinforcement to all students, particularly to students exhibiting chronic behavior problems, are more likely to decrease ODRs and increase school climate. The results further confirm the value of positively reinforcing students for appropriate behavior (Horner et al., 2010).

Schools should also work to increase connections with parents, particularly for students with frequent behavior problems. Based on the results, parents need to support their student’s performance at school. Although intuitive, parents need to become formal partners with schools helping these students, which means they need to be involved before behavior problems occur. Schools can do this by highlighting students’ positive behaviors and calling home when students do something well, not only when they do something wrong.

In addition to increasing school culture, schools need to provide a variety of interventions for students with increased ODRs. Students in the Secondary and Tertiary classes appear to be performing significantly worse than their Primary class peers across all outcomes. These results support integrated academic and behavioral interventions for students exhibiting chronic behavior problems as well as for students at risk for chronic behavior problems.

Conclusions

The results of this preliminary study confirm that (a) generally, three broad classes of students can be identified in schools based on their perceptions of school climate and ODRs, (b) students with the most problematic behaviors need teachers to reinforce their appropriate behavior and parents to reinforce the value and importance of school, and (c) students at risk and exhibiting increased problem behaviors as measured by ODRs perform statistically significantly worse across all core academic content areas. Taken together, we suggest that schools should (a) create positive learning environments by reinforcing appropriate behavior and working to connect with parents to reinforce the value of school and (b) provide social-behavioral and academic supports to students exhibiting increased problem behaviors. By increasing positive school climate, schools can decrease ODRs and increase positive academic, social, and behavioral outcomes for all students.

Footnotes

Nicholas A. Gage is an assistant professor of special education in the School of Special Education, School Psychology, & Early Childhood Studies in the College of Education at the University of Florida, 1403 Norman Hall, PO Box 117050, Gainesville, FL 32611; e-mail: gagenicholas@coe.ufl.edu . Specific research interests include identification of policies and practices at the national, state, local and classroom levels to support the academic, social, and behavioral needs of students with or at risk for emotional and/or behavioral disorders through rigorous and diverse research methods.

Alvin Larson is the research and evaluation specialist for Meriden Public Schools in Meriden, Connecticut. He provides research support for all research questions posed by the school board to evaluate and improve educational progress for students in Meriden. In addition, he provides technical assistance for survey and test development for the school district and other departments in the City of Meriden.

George Sugai is professor and Carole J. Neag Endowed Chair in the Neag School of Education at the University of Connecticut. His professional interests include effective classroom and behavior management practices, school-wide discipline, social skills instruction, and positive behavior supports.

Sandr M. Chafouleas is the associate dean for research and a professor in the Department of Educational Psychology within the Neag School of Education. Her research interests include behavioral assessment and school-based prevention of behavioral and emotional problems.

References

Bandura

(2001). Social cognitive theory: An agentic perspective. Annual Review of Psychology, 52, 1–26. doi:10.1146/annurev.psych.52.1.

Battistich

Horn

(1997). The relationship between students’ sense of their school as a community and their involvement in problem behaviors. American Journal of Public Health, 87, 1997–2001. doi:10.2105/AJPH.87.12.1997

Battistich

Solomon

Kim

Watson

Schaps

(1995). Schools as communities, poverty levels of student populations, and students’ attitudes, motives, and performance: A multilevel analysis. American Educational Research Journal, 32, 627–658. doi:10.2307/1163326

Bradshaw

C. P.

Koth

C. W.

Bevans

K. B.

Ialongo

Leaf

P. J.

(2008). The impact of school-wide Positive Behavioral Interventions and Supports (PBIS) on the organizational health of elementary schools. School Psychology Quarterly, 23, 462–473.

Bradshaw

C. P.

Koth

C. W.

Thornton

L. A.

Leaf

P. J.

(2009). Altering school climate through school-wide positive behavioral interventions and supports: Findings from a group-randomized effectiveness trial. Prevention Science, 10, 100–115. doi:10.1007/s11121-008-0114-9

Brand

Felner

Shim

Seitsinger

Dumas

(2003). Middle school improvement and reform: Development and validation of a school-level assessment of climate, cultural pluralism, and school safety. Journal of Educational Psychology, 95, 570–588. doi:10.1037/0022-0663.95.3.570

Cohen

McCabe

E. M.

Michelli

N. M.

Pickeral

(2009). School climate: Research, policy, practice, and teacher education. Teachers College Record, 111, 180–213.

Fan

Williams

C. M.

Corkin

D. M.

(2011). A multilevel analysis of student perceptions of school climate: The effect of social and academic risk factors. Psychology in the Schools, 48, 632–647. doi:10.1002/pits.20579

Gage

N. A.

Larson

Chafouleas

S. M.

(2016). The Meriden School Climate Survey-Student Version: Reliability and validity. Assessment for Effective Intervention, 41, 67–78. doi:10.1177/1534508415596960

10.

Gage

N. A.

Prykanowski

D. A.

Larson

(2014). School climate and bullying victimization: A latent class growth model analysis. School Psychology Quarterly, 29(3), 256–271.

11.

Gottfredson

(2000). Schools and delinquency. New York, NY: Cambridge University Press.

12.

Gottfredson

G. D.

Gottfredson

D. C.

Payne

A. A.

Gottfredson

N. C.

(2005). School climate predictor of school disorder: Results from a national study of delinquency in schools. Journal of Crime and Delinquency, 42, 412–444. doi:10.1177/0022427804271931

13.

Harel

Pellowski

Kalichman

(2012). Are we missing the importance of missing values in HIV prevention randomized clinical trials? Review and recommendations. AIDS and Behavior, 16(6), 1382–1393. doi:10.1007/s10461-011-0125-6

14.

Haynes

N. M.

Emmons

Ben-Avie

(1997). School climate as a factor in student adjustment and achievement. Journal of Educational and Psychological Consultation, 8, 321–329. doi:10.1207/s1532768xjepc0803_4

15.

Herman

Dawson

Dee

Greene

Maynard

Redding

Darwin

(2008). Turning around chronically low-performing schools: A practice guide (NCEE No. 2008-4020). Washington, DC: National Center for Education Evaluation and Regional Assistance, Institute of Education Sciences, U.S. Department of Education. Retrieved from http://ies.ed.gov/ncee/wwc/publications/practiceguides

16.

Horner

R. H.

Sugai

Anderson

C. M.

(2010). Examining the evidence base for school-wide positive behavior support. Focus on Exceptionality, 42, 1–14.

17.

Irvin

L. K.

Tobin

T. J.

Sprague

J. R.

Sugai

Vincent

C. G.

(2004). Validity of office discipline referral measures as indices of school-wide behavioral status and effects of school-wide behavioral interventions. Journal of Positive Behavior Interventions, 6, 131–147. doi:10.1177/10983007060080010301

18.

Koth

C. W.

Bradshaw

C. P.

Leaf

P. J.

(2008). A multilevel study of predictors of student perceptions of school climate: The effect of classroom-level factors. Journal of Educational Psychology, 100, 96–104. doi:10.1037/0022-0663.100.1.96

19.

Kuperminc

G. P.

Leadbeater

B. J.

Blatt

S. J.

(2001). School social climate and individual differences in vulnerability to psychopathology among middle school students. Journal of School Psychology, 39, 141–159. doi:10.1016/S0022-4405(01)00059-0

20.

Lanza

S. T.

Collins

L. M.

Lemmon

D. R.

Schafer

J. L.

(2007). PROC LCAS: A SAS procedure for latent class analysis. Structural Equation Modeling, 14, 671–694.

21.

Leech

N. L.

Barrett

K. C.

Morgan

G. A.

(2008). SPSS for intermediate users: Use and interpretation (3rd ed.). New York, NY: Psychology Press.

22.

Loukas

Murphy

J. L.

(2007). Middle school student perceptions of school climate: Examining protective functions on subsequent adjustment problems. Journal of School Psychology, 45, 293–309. doi:10.1016/j.jsp.2006.10.001

23.

Loukas

Robinson

(2004). Examining the moderating role of perceived school climate in early adolescent adjustment. Journal of Research on Adolescence, 14, 209–233. doi:10.1111/j.1532-7795.2004.01402004.x

24.

Magidson

Vermunt

J. K.

(2004). Latent class models. In Kaplan

(Ed.), The Sage handbook of quantitative methodology for the social sciences (pp. 175–198). Thousand Oaks, CA: Sage Publications.

25.

May

Ard

Todd

A. W.

Horner

R. H.

Glasgow

Sugai

Sprague

J. R.

(2005). School-wide information system. Eugene, OR: Educational and Community Supports, University of Oregon, Eugene.

26.

McIntosh

Chard

D. J.

Boland

J. B.

Horner

R. H.

(2006). Demonstration of combined efforts in school-wide academic and behavioral systems and incidence of reading and behavior challenges in early elementary grades. Journal of Positive Behavioral Interventions, 8, 146–154. doi:10.1177/10983007060080030301

27.

Nansel

T. R.

Overpeck

Pilla

R. S.

Ruan

W. J.

Simons-Morton

Scheidt

(2001). Bullying behaviors among US youth: Prevalence and associations with psychosocial adjustment. Journal of the American Medical Association, 285, 2094–2100. doi:10.1001/jama.285.16.2094

28.

Nelson

J. R.

Benner

G. J.

Reid

R. C.

Epstein

M. H.

Currin

(2002). The convergent validity of office discipline referrals with the CBCL-TRF. Journal of Emotional and Behavioral Disorders, 10, 181–188.

29.

Roeser

R. W.

Eccles

J. S.

Sameroff

A. J.

(2000). School as a context of early adolescents’ academic and social-emotional development: A summary of research findings. Elementary School Journal, 100, 443–471. doi:10.1086/499650

30.

Sugai

Horner

R. H.

(2002). The evolution of discipline practices: School-wide positive behavior supports. Child & Family Behavior Therapy, 24, 23–50. doi.org/10.1300%2FJ019v24n01_03

31.

Thapa

Cohen

Guffey

Higgins-D’Alessandro

(2013). A review of school climate research. Review of Educational Research, 83, 357–385. doi:10.3102/0034654313483907

32.

Vermunt

J. K.

Magidson

(2005). Latent GOLD 4.0 user’s guide. Belmont, MA: Statistical Innovations Inc.

33.

Walker

H. M.

Shinn

M. R.

(2002). Structuring school- based interventions to achieve integrated primary, secondary, and tertiary prevention goals for safe and effective schools. In Shinn

M. R.

Walker

H. M.

Stoner

(Eds.), Interventions for academic and behavior problems II: Preventive and remedial approaches (pp. 1–25). Bethesda, MD: National Association of School Psychologists.

34.

Wang

(2009). School climate support for behavioral and psychological adjustment: Testing the mediating effect of social competence. School Psychology Quarterly, 24, 240–251. doi:10.1037/a0017999

35.

Wang

Degol

J. L.

(2015). School climate: A review of the construct, measurement, and impact on student outcomes. Educational Psychology Review. doi:10.1007/s10648-015-9319-1

36.

Wang

Selman

R. L.

Dishion

T. J.

Stormshak

E. A.

(2010). A tobit regression analysis of the covariation between middle school students’ perceived school climate and behavioral problems. Journal of Research on Adolescence, 20, 274–286. doi:10.1111/j.1532-7795.2010.00648.x

37.

Way

Reddy

Rhodes

(2007). Students’ perceptions of school climate during the middle school years: Associations with trajectories of psychological and behavioral adjustment. American Journal of Community Psychology, 40, 194–213. doi:10.1007/s10464-007-9143-y

38.

Welsh

W. N.

(2000). The effects of school climate on school disorder. Annals of the American Academy of Political and Social Science, 567, 88–107. doi:10.1177/0002716200567001007