Abstract
Implementing research-based curricula and instruction in inclusion-oriented schools is helped or hindered by having coherent models of service delivery accounting for the full range of student diversity. The current investigation offers data from 174 participants in 32 schools, analyzed using descriptive statistics, correlation, and hierarchical linear modeling (HLM). The findings offer replication of special education service delivery data from an earlier study, new descriptive data, and HLM analyses that identify special educator school density (the number of special educator full-time equivalents to total school population) and individual special educators’ Individualized Education Program (IEP) caseload size as variables predictive of special educators’ ratings of the conduciveness of their working conditions to providing effective special education for students on IEPs.
How many special educators are needed to adequately staff an inclusion-oriented school? This seemingly foundational question has received relatively little attention in the professional literature. When addressed, it has been most commonly framed in terms of special educator caseload parameters, namely, the number of students with Individualized Education Programs (IEPs) and their characteristics (e.g., types of disabilities, support intensity levels; Russ, Chiang, Rylance, & Bongers, 2001). Caseload parameters and corresponding corollaries (e.g., amount of paperwork, extent of service coordination roles) affect the availability of time special educators have to collaborate with classroom teachers and directly work with students.
Accurately characterizing special educator caseloads so they can be compared within and across schools is complicated by additional factors such as (a) employment status (e.g., full-time, part-time, itinerant), (b) percentage of time assigned to students on IEPs versus those not special education eligible (e.g., 504, Title I, at-risk), (c) range of grades and general education curricular content supported, (d) number of general education teachers supported, (e) number of paraprofessionals supervised, and (f) other duties (e.g., case management). Not only have caseload stresses been identified as key contributors to special educator attrition, shortages, and the research-to-practice gap (Gersten, Keating, Yovanoff, & Harniss, 2001; Kozleski, Mainzer, & Deshler, 2000; McLeskey & Billingsly, 2008; Russ et al., 2001), the variations within special educator caseloads create ambiguity, hampering efforts to make reasoned and accurate comparisons.
Suter and Giangreco (2009) suggested that relying on a school’s average special educator caseload size might be an inadequate measure to determine whether a school has sufficient availability of special educators to meet student needs. Among schools with comparable special educator average caseload sizes, those that identified a lower percentage of students as eligible for special education tended to be less well resourced in terms of availability of special educators than schools that identified more students as in need of special education, even though approximately one third of all students in these schools had some kind of special needs designation resulting in an individualized support plan (e.g., IEP, 504, at-risk). Ironically, some schools that helped students by implementing general education supports, without identifying them as disabled, inadvertently undercut their services because the number of special educators was based exclusively on the number of students on IEPs, rather than considering the total enrollment of the school, regardless of labels (Suter & Giangreco, 2009). While acknowledging the importance of caseload size and configuration, this study explores other personnel-related factors in inclusion-oriented schools; one such key factor is special educator school density, defined as the number of special educators in full-time equivalents (FTE) per total school enrollment.
The descriptor “inclusion-oriented” refers to schools striving to provide appropriate educational opportunities and supports to students with the full range of disability characteristics and levels in chronologically age-appropriate, general education classes those students would attend if they were not labeled disabled. Schools demonstrate their commitment to providing inclusion-oriented special education by (a) serving virtually all of the students eligible to attend their school (e.g., not routinely sending students with more intensive needs out to regional special programs) and (b) exceeding the national average (59%; U.S. Department of Education, 2011) by placing 70% or more of their students with disabilities in general education classes (80% of the time or more). Inclusion-oriented schools continually strive to improve their practices to increase access to supported general class placements for students who currently receive less than 80% of their instruction in general education classrooms. This study focuses on inclusion-oriented schools because they represent qualitatively different contexts than schools that include substantially fewer students with disabilities in general education classes, are satisfied with the status quo, or remain committed to special class or special school models.
Although our national focus on research-based curriculum and instruction remains vital, without effective service delivery configurations, most notably thoughtful personnel utilization and deployment, even the most advantageous innovations in curriculum, instruction, and social/behavioral interventions likely will not fully realize their intended impact. A potent example is available in general education service delivery. Ouchi (2009) reported that significant gains in student achievement could be realized by lowering total student load (TSL), “The number of students that a teacher has to get to know each term and the number of papers that teacher has to grade” (p. 10). Lowering TSL to a recommended level, around 80 in middle and high schools, is pursued by giving principals decentralized budgetary control and accountability to reallocate available resources (e.g., hiring more teachers by reducing noninstructional staff).
When a district has too few classroom teachers, student loads per teacher rise to the point where teachers can no longer know their students well enough to establish a bond of trust with them. Without this trust, a teacher can neither establish an orderly classroom nor push a student to do his or her best, and the teacher’s job becomes frustrating and constantly stressful. (Ouchi, 2009, pp. 9–10)
Similarly, special educator school density has potential implications for special education service delivery. The United States has an average of one special educator FTE for every 120 students in total enrollment (SD = 35.96) with substantial differences across states, ranging from ratios of 1:80 or less in six states (ME, NH, NJ, NY, OK, RI) to 1:190 or more students in seven others (CA, ID, MS, TX, UT, WA, WY; Giangreco, Hurley, & Suter, 2009). Such wide variation raises questions about whether students with disabilities are receiving equitable access to the promises of the Individuals With Disabilities Education Act (IDEA). Special educator school density is a potentially important and minimally understood variable which schools have the power to change to improve working conditions for school personnel and learning conditions for students.
Special educator school density is easily calculable and allows a school to consider the availability of special educator resources regardless of the percentage of students labeled disabled or the variations in caseload configuration. Suter and Giangreco (2009) hypothesized that schools with a special educator school density of below 1:80 were the healthiest in terms of meeting students’ educational needs and were better able to absorb routine fluctuations (e.g., enrollment of a new student with a disability). They characterized schools in the 1:80 to 1:100 range as “precarious” because these schools reported “being less able to absorb routine fluctuations” (Suter & Giangreco, 2009, p. 10) and were more likely to respond to fluctuation pressures by hiring more paraprofessionals. Schools with special educator school density levels at 1:100 or higher seemed to be the least healthy, less able to absorb routine fluctuations affecting service delivery, and quicker to add paraprofessionals in response to perceived stressors.
Available research suggests that offering effective service delivery is not just about having many people but those with the right skills and making sure they have adequate time to work with each other and directly with students. Although some schools have substantially increased their number of paraprofessionals, rather than special educators or teachers, reviews of available research literature have questioned the wisdom of that path as an avenue to meet the needs of students with disabilities placed in general education classrooms (Giangreco, Edelman, Broer, & Doyle, 2001; Giangreco, Suter, & Doyle, 2010). Recently, a large-scale, longitudinal, multimethod study conducted in more than 6,000 schools in England and Wales found predominantly negative relationships between the use of teacher assistants and student achievement in English, math, and science (Webster et al., 2010). An observational study by the same research team in a subset of these schools identified differences in instructional interactions students have with qualified teachers versus teacher assistants (Rubie-Davies, Blatchford, Webster, Koutsoubou, & Bassett, 2010). Researchers reported quality interactions and engagement between teachers and students characterized by (a) informing students about the focus of the lesson, (b) taking time to explain concepts, (c) using prompts and questions to encourage thinking and check for understanding, and (d) making links to prior knowledge. Conversely, teacher assistants tended to focus their concern on task completion and were observed more frequently (a) providing answers, (b) completing student work, and (c) providing confusing or inaccurate information.
The purpose of the present investigation was twofold. First, the study replicates data collection from the earlier study by Suter and Giangreco (2009) to ascertain whether those initial findings are similar in a larger sample of schools. Second, the study uses more sophisticated statistical analyses (i.e., hierarchical linear modeling [HLM]) to explore relationships between special educator school density and other school- and special educator–level variables, most notably a new special educator self-rating regarding whether their work responsibilities (e.g., caseload size and configuration) were conducive to providing effective special education for students served on IEPs. This self-rating regarding work responsibilities relates to earlier research in both general and special education that has demonstrated a ripple effect between educators’ self-efficacy ratings, their job satisfaction, and outcomes such as their decision to stay in the field and student achievement (Caprara, Barbaranelli, Steca, & Malone, 2006; Gersten et al., 2001; Viel-Ruma, Houchins, Jolivette, & Benson, 2010). These data fill a gap in the existing literature and are important because they (a) provide additional data on a variety of service delivery variables related to inclusion-oriented schools where minimal data exist, (b) provide a mechanism for schools to examine their special education personnel resources, and (c) either bolster or disconfirm the conceptualization of special education service delivery health, based on special educator school density, proposed by Suter and Giangreco (2009).
Method
Design
This study used a quantitative design, including descriptive statistics, correlation, and HLM, a form of multilevel modeling (Raudenbush & Bryk, 2002) to examine the influences of special educator variables and school demographic variables on special educators’ ratings of perceived conduciveness of their work responsibilities to meeting the needs of students receiving special education. Data were collected between September 2007 and October 2010 from a convenient, yet purposeful sample of respondents in inclusion-oriented schools.
Settings
Data were collected in 32 schools spanning all grades (K-12), including 16 elementary schools, 5 elementary/middle schools (K-8), 8 middle schools, and 3 high schools. They were located in rural (n = 16), small city (n = 7), suburban (n = 5), and town (n = 4) settings based on national codes (Sable & Plotts, 2010). Total student enrollment ranged from 86 to 1,099 (M = 368.22, SD = 224.51) with a mean class size of 18.28 (SD = 2.50). On average, schools’ students included 8% (SD = 9.59) from racial/ethnic minorities, 5% (SD = 8.46) English Language Learners (ELL), and 32% (SD = 25.19) who received free or reduced-price lunch.
On average, each participating school employed nearly 6 special educators (M = 5.69, SD = 3.51), to cover an average of 4.33 (SD = 2.59) special educator positions in full-time equivalency (FTE). Special educator FTE included a small portion of speech-language pathologists’ time, on average 0.17 FTE (SD = 0.25) per school, where they functioned as special educators rather than as related services providers (Giangreco, Prelock, & Turnbull, 2010). Total paraprofessional utilization in these schools averaged nearly 20 people (M = 19.69; SD = 12.21) to cover an average of 18.60 FTE (SD = 11.15) positions. On average, approximately 79% (M = 14.64, SD = 9.04) of paraprofessional FTE was assigned to special education.
Students with IEPs in these schools averaged 13.76% (SD = 4.31), with nearly 91% (SD = 8.11) of these students included in general education classes as their primary placement (at least 80% of the time). Students with disabilities served using 504 plans averaged 3.49% (SD = 4.38), and 12.76% (SD = 11.41) of students without disabilities considered educationally at-risk were served on Educational Support Team (EST) plans. On average, these schools had 30.39% (SD = 11.94) of students on one of these three types of support plans (i.e., IEP, 504, EST).
Participants
The study totaled 174 participants, including 145 special educators (with 5 speech-language pathologists who functioned in the role of special educators), 23 principals and assistant principals, and 6 special education administrators. Data were only collected from special educators employed 0.4 FTE or more at the participating schools. Approximately 84% (n = 123) of the special educators self-identified as female, 13% (n = 19) as male, and 3 people did not respond to this question. They had an average of 13.10 years (SD = 8.99) of experience working as special educators and 79% (n = 115) held graduate degrees.
Procedures and Instrumentation
School administrator data collection
Participating school administrators completed the 20-item School Demographics Questionnaire for their respective schools, responding to basic school characteristics such as enrollment information and various personnel numbers for special educators and paraprofessionals.
Special educator data collection
On a school-by-school basis, an institutional review board–approved lay summary of the study and consent procedure were shared with special educators to invite their participation. Next, investigators scheduled a 1-hr meeting with all special educators at a school to obtain informed consent and administer questionnaires that were collected before leaving the meeting; these on-site data collection procedures resulted in 94% response rate.
A total of 145 special educators completed the 23-item Special Educator Questionnaire, responding to items about their work roles, caseload parameters, time use, and the paraprofessionals they supervised. The 19-item Student Characteristics Questionnaire was completed by a subset of 78 special educators reporting on 196 students on their caseloads receiving full-day, one-to-one, paraprofessional support. This questionnaire included items on demographics, instructional time, and reasons for having a one-to-one paraprofessional.
Post data collection reporting and debriefing
Within 1 month of completing data collection at each school, investigators provide a written data summary and conducted a 2-hr debriefing meeting with school personnel to review the findings.
Data Analyses
Descriptive quantitative and correlation analyses were conducted to aide in the replication aspect of the study. Differences between the current and earlier study (Suter & Giangreco, 2009) were identified using independent samples t tests (two-tailed) assuming unequal variances (recommended for unequal samples; see Zimmerman, 2004) and calculating standardized mean differences effect sizes (Cohen’s d) for all continuous variables and chi-square analyses for all categorical variables. Findings that were significantly different from the earlier study at p < .05 and had medium (d = 0.50) or large effect sizes (d = 0.80; Cohen, 1988) are reported.
We used HLM to explore the hypothesis that special educator school density could be a key predictor of special educators’ perceptions about whether their work responsibilities were conducive to providing effective special education. We considered alternative explanations by including other special educator (Level 1) and school (Level 2) variables in the analyses (see Table 1) and accounting for the fact that special educators were nested within schools. For example, special educators in the same school will likely have different special educator ratings, but all share the same special educator school density. Therefore, we did not use a statistical technique like ordinary least squares (OLS) regression that would ignore the nested nature of the data. Using HLM to examine these multilevel relationships allowed us to model the effects of special educators nested within schools and to include special educator–level and school-level variables in our analyses (Raudenbush & Bryk, 2002). The selection of the variables presented in Table 1 were selected because they (a) address long-standing concerns pertaining to special educator caseloads such as size, amount of paperwork, time available to provide instruction to students, and number of paraprofessionals to supervise (Gersten et al., 2001; Giangreco & Broer, 2005; Kozleski et al., 2000; Russ et al., 2001) and (b) explicitly extend a line of research about service delivery concerns in inclusion-oriented schools such as special educator school density and the special educators to paraprofessionals ratio (Giangreco, Broer, & Suter, 2011; Giangreco, Smith, & Pinckney, 2006; Suter & Giangreco, 2009). All HLM analyses were conducted in Mixed-Models SPSS Version 19 using restricted maximum likelihood (REML) estimation, which produces less biased estimates of random effects than maximum likelihood (ML) estimation for small samples (Raudenbush & Bryk, 2002).
Level 1, Level 2, and Outcome Variables for HLM Analyses.
Note. HLM = hierarchical linear modeling; IEP = Individualized Education Program; FTE = full-time equivalents; EST = Educational Support Team; FRPL = free or reduced-price lunch.
The HLM analyses involved a series of steps to test our hypothesis. First, we examined whether HLM was warranted (see Model 1 Findings). We calculated the intraclass correlation coefficient (ICC) of special educator ratings to see whether they systematically varied among schools. An ICC of zero indicates no apparent school influence, with special educators in one school responding much like special educators in any other. As ICC values rise, they represent increasing similarity among the ratings given by special educators in the same school (relative to the total variance), thus increasing the rationale for using HLM. Second, we tested our primary research question: Are special educator ratings significantly influenced by special educator school density (Model 2)? Third, we tested the effect of special educator school density on special educator ratings, controlling for several school and special educator variables that we thought might influence the ratings (see Table 1). This allowed us to see the unique effect of special educator school density on the ratings when controlling for the effects of other variables (Models 3–8).
Level 1 variables were direct special educator responses or simple calculations from their responses, and school variables (Level 2) were calculated from responses provided by school administrators. All continuous Level 1 and 2 variables were transformed using grand mean centering, whereby each data point was subtracted by its variable’s overall mean (Enders & Tofighi, 2007). The new mean for each variable becomes 0, allowing for easier interpretation of the parameter estimates yielded by HLM. We did not center the two categorical variables (i.e., Highest Degree Earned, Setting Category) to permit easier interpretation of their parameter estimates. Finally, for HLM analyses in which we wanted to directly compare the predictor parameter estimates across variables (Model 3 to 7), we standardized all continuous and categorical variables. Variables remained unstandardized for analyses in which such direct comparisons were not needed (Models 1, 2, and 8).
Results
The findings are presented in two sections: (a) descriptive data from the three questionnaires, both replication and new data and (b) a series of HLM models examining the relationship of special educator school density and special educator ratings of effectiveness.
Descriptive Questionnaire Data
Data in the current study largely reinforce findings from the earlier study (Suter & Giangreco, 2009); only two interrelated sample setting characteristic differences were found to be statistically significant. The average student enrollment for schools in the current study (M = 368.22, SD = 224.51) was significantly lower than the earlier study, t(27.33) = −2.50, p = .02, by approximately 220 students, representing a large effect (d = −0.81), and on average these schools employed approximately two fewer special educator FTE, t(31.82) = −2.37, p = .02, d = −0.73. There were no significant differences among the remaining school demographic variables.
Special education service delivery
Special educators reported IEP caseloads averaging 9.34 (SD = 4.92). Adjusting for the fact that not all special educators work full-time in that role, the average full-equivalent caseload (i.e., the number of students with IEPs a special educator would have if 1.0 FTE of his or her time was directed toward students in IEPs) increases to 10.38 (SD = 5.18); this mean was significantly lower than special educators in the earlier study by 4.66 students, t(116.56) = −4.02, p < .001, d = −0.62. Average caseloads for special educators are higher when you consider the actual combined number of students supported on IEPs, 504 plans, and EST plans (M = 12.66, SD = 6.91); this mean also was significantly lower than special educators in the earlier study by 4.58 students, t(132.20) = −3.42, p < .001, d = −0.51. This suggests that the lower number of students and difference in the combined caseload is attributable primarily to the lower number of students on IEPs, rather than those on 504 or EST plans. Considering all the students special educators report working with, on and off their caseloads, the average number increases to 16.50 (SD = 8.71), only one student less than the earlier study.
Responses to how much time special educators spent in different activities were nearly identical to those in the earlier study, with all individual means within two percentage points of those previously reported. Instruction (38.93%, SD = 18.19) and paperwork (15.42%, SD = 9.43) consumed the majority of time. The remainder was distributed across (a) planning activities (11.55%, SD = 7.01), (b) collaboration (10.82%, SD = 6.1), (c) behavior support (7.78%, SD = 9.09), (d) supervision of paraprofessionals (7.46%, SD = 5.42), (e) working with families (6.59%, SD = 4.71), and (f) other activities (1.45%, SD = 4.07).
Approximately halfway through data collection we added a new item; special educators (n = 55) estimated the percentage of their instruction that occurred outside of the general education classroom for their students on IEPs. Special educators reported, on average, that they provided 75% (SD = 29.03) of their instruction to their students on IEPs outside of general education classrooms, with 40% of them (n = 22) reporting that they provide 100% of their instruction outside general education classrooms. Special educators reported supervising an average of 3.32 paraprofessionals (SD = 2.86). Coupled with the relatively small percentage of time special educators spent supervising paraprofessionals (M = 7.46%, SD = 5.42), on average each paraprofessional only received approximately 2% of a special educator’s time, closely replicating the earlier study findings.
The mean special educator school density of one special educator FTE for every 91.40 (SD = 28.81) students of total school enrollment was within three percentage points of the earlier study and included a similarly wide range. The most densely resourced school had one special educator FTE for every 38 students of total school enrollment, whereas the most sparsely resourced school had one special educator FTE for every 141 students. Comparing special educator school density with percentage of students on IEPs yielded a strong and significant correlation (r = −.50, p = .004), whereas no significant relationships were found between special educator school density and percentage of students with 504 (r = −.03, p = .88) or EST plans (r = −.22, p = .22).
Special education paraprofessional service delivery
Paraprofessional utilization closely replicated findings from the earlier study. On average, schools employed one special education paraprofessional FTE for every 3.36 students on IEPs (SD = 1.27) and 3.81 special education paraprofessional FTE for every special educator FTE (SD = 1.56). More than half of the special education paraprofessional FTE (M = 51.73, SD = 24.15) was devoted to providing one-to-one supports to students with disabilities and nearly 63% (SD = 35.61) of the one-to-one paraprofessionals supported students whose placements were in general education settings at least 80% of the time. Special educators reported a similar distribution of how paraprofessionals whom they supervised spent their time, with nearly three quarters of their time devoted to instruction (50.64%, SD = 24.62) and behavior support (23.21%, SD = 21.58). The remainder of special education paraprofessional time was distributed across (a) self-directed activities (10.42%, SD = 16.51), (b) personal care (5.96%, SD = 14.14), (c) supervision of students (5.34%, SD = 6.25), (d) clerical (2.96%, SD = 3.98), and (e) other activities (1.47%, SD = 5.81).
In the current study, we were able to calculate a more accurate comparison of time use provided by special education personnel by accounting for an average of 10.07 hr (SD = 8.30) per week special educators reported working beyond their contracted time. The current data indicated that 76% (SD = 9.32) of instruction provided by special education personnel was provided by special education paraprofessionals and 24% (SD = 9.32) by special educators.
Students with disabilities receiving one-to-one supports in general education
Special educators reported on 196 students who were receiving full-day one-to-one paraprofessional support. Of these students, 86% were Caucasian, 8% African American, and 3% or less were identified as Asian, American Indian, or Other. Students ranged in age from 5 to 21 (M = 10.52, SD = 3.92); approximately 68% were in elementary school and 68% were male. Of the students receiving one-to-one paraprofessionals supports, 46% were placed in general education settings 80% of the time or more and approximately 31% participated in alternate assessment. Students had primary disabilities in all IDEA disability categories except for hearing impairment, with the distribution similar to the earlier study. Again, the most common category was autism (24.5%, n = 48). Although the order and percentage varied slightly, the next five categories remained similar: health impaired (14.8%, n = 29), emotional disturbance (13.8%, n = 27), intellectual disability (13.3%, n = 26), multiple disabilities (11.2%, n = 22), and developmental delay (10.7%, n = 21), with all other categories being below 4%.
As primary disability categories do not adequately describe disability-related characteristics, special educators rated their students’ level of severity in six domains: (a) behavioral, (b) intellectual, (c) physical, (d) health, (e) visual, and (f) hearing. Approximately 74% (n = 145) of students were characterized as having moderate or severe behavior problems and 72% (n = 142) moderate or severe intellectual disabilities, rated the two most common characteristic categories in both the current and earlier study. Students characterized as moderate or severe in the remaining four categories ranged from 29% (physical) to 6% (visual).
Special educators reported that they were the most common advocates for one-to-one paraprofessional supports (n = 188, 96%), with teachers (n = 174, 89%), parents (n = 172, 88%), and administrators closely trailing (n = 158, 81%). Students with disabilities were reported to less frequently advocate for this support (n = 42, 21%). Cited reasons for providing one-to-one supports remained in the same ranking order as the earlier study with the most commonly cited being instructional support (n = 166, 85%) and behavior support (n = 151, 77%), followed by (a) student safety (n = 130, 66%), (b) communication support (n = 101, 52%), (c) personal care support (n = 85, 43%), (d) safety of others (n = 67, 34%), and (e) other reasons (n = 15, 8%).
Special educators reported that, on average, students with one-to-one paraprofessional supports received more of their instruction from paraprofessionals (M = 40.48%, SD = 28.81) than teachers (M = 38.05%, SD = 31.29) and special educators (M = 20.45%, SD = 16.03). Compelling aspects of these data are reflected in the large standard deviations. The percentage of instruction students with one-to-one paraprofessional supports received from teachers ranged from 0% to 100%; more than 16% (n = 32) received none of their instruction from teachers and 43% (n = 80) received 20% or less. Only 33% (n = 65) of students being supported by a one-to-one paraprofessional received more than half of their instruction from teachers. The distribution of percentage of instruction received by these students was similarly wide for paraprofessionals, ranging from 0% to 94%; 43% of these students received half or more of their instruction from paraprofessionals, with nearly 20% (n = 50) receiving 70% or more. Conversely, just over 10% (n = 20) of these students did not receive any instruction from the one-to-one paraprofessionals assigned to them. Special educators followed suit with a similarly wide range, 0% to 100%, though with a smaller standard deviation. A total of 66% (n = 130) of these students received between 10% and 30% of their instruction from special educators.
HLM of Special Educator Ratings
Data considerations
Before conducting the HLM analyses, we screened the data for possible violations of statistical assumptions. Two participants were removed from the HLM analyses because they did not provide special educator ratings, yielding a revised n of 143 special educators. In addition, we found that the special educator rating outcome variable was negatively skewed. According to Raudenbush and Bryk (2002), deviations from normality will not bias estimated effects of the school-level variables but might introduce bias in the standard errors associated with those effects. Across all schools, there were no outliers in the special educator rating. However, within one school, which had a mean rating of 7.31 (SD = 2.12), two teachers gave ratings of 3 representing moderate outliers (greater than two standard deviations below the mean). Because these ratings were plausible and only slightly beyond the expected range within the school, they were retained. Overall, Level 1 residuals were slightly negatively skewed (Shapiro–Wilk W = .96, p < .05), indicating that some variables may have been omitted from the model. Given our small sample size, particularly at Level 2, it was not feasible to include all relevant predictors in the model. Means, standard deviations, and correlations among continuous study variables are presented in Table 2 (Level 1) and Table 3 (Level 2). Parameter estimates for the sequence of models tested to explore the study hypothesis are summarized in Table 4.
Correlations, Ms, and SDs for Level 1 (Special Educator) Continuous Variables (n = 143).
Note. IEP = Individualized Education Program; EST = Educational Support Team.
p < .05. **p < .01.
Correlations, Ms, and SDs for Level 2 (School) Continuous Variables (n = 32).
Note. IEP = Individualized Education Program; FRPL = free or reduced-price lunch.
p < .05. **p < .01.
Unstandardized Fixed Effects, Random Effects, and Model Statistics for Models Predicting Special Educator Work Ratings.
Note. Standard errors are in parentheses. Models 3 to 7 are not shown because they were exploratory steps to choose Level 2 variables for Model 8. Proportion Variance Explained calculated following Raudenbush and Bryk (2002). IEP = Individualized Education Program; EST = Educational Support Team.
p < .05. **p < .01. ***p < .001.
Model 1
The purpose of Model 1 was to estimate the ICC to determine the extent of clustering of special educator ratings within schools. Model 1 represents an unconditional model, with an intercept that is allowed to vary randomly across schools and without any special educator (Level 1) or school (Level 2) predictors included in the model. Dividing the between school variance (1.52) by the total variance (6.09) yields an ICC of .25, indicating that a quarter of the variability in special educator ratings is attributed to clustering within schools.
Model 2
In Model 2, we examined our main hypothesis: Special educator school density is a key predictor of special educators’ ratings of their effectiveness to meet their students needs given their work responsibilities. To examine this, we added special educator school density as a Level 2 predictor to the unconditional model. Special educator school density was found to have a significant effect, t(32.98) = −2.94, p = .006. The special educator school density parameter estimate of –.03 indicates that a one-unit increase in special educator school density (i.e., fewer special educator FTE available to support students) lowers the predicted special educator rating by 0.03. In other words, an increase in special educator school density of 33 is expected to lower the special educator rating by a full point.
The school-level variance dropped from 1.52 in Model 1 to 1.02 in Model 2, indicating that the inclusion of special educator school density explained some of the school influence on special educator ratings. In OLS regression it is common to calculate a R2 statistic to show the percentage of variance explained by a model. Although it is not possible to calculate R2 in HLM, Raudenbush and Bryk (2002) recommended calculating a proportion variance explained (PVE) to provide an estimate of the amount of variance accounted for by a set of independent variables. To compute the PVE, the difference between the variance components from a full model (e.g., Model 2) and the unconditional model (Model 1) is divided by the variance component from the unconditional model. This can be done separately for Level 1 (special educator) and Level 2 (school) variance components. Level 2 PVE for Model 2 is .33, indicating that 33% of the school-level variability in special educator ratings is explained by special educator school density.
Models 3 to 7
This series of models included variables that we anticipated might influence the relationship between school density and special educator ratings and thus would be important to consider along with special educator school density. In particular, we wanted to control for the effects of the eight special educator variables (Level 1) and five school variables (Level 2), in addition to special educator school density (see Table 1). Given our small sample size at Level 2 (n = 32), it is not recommended to include all six Level 2 variables in our model (Snijders & Bosker, 1999). Therefore, we chose to run five separate models, each including all special educator (Level 1) variables, special educator school density, and one additional school-level variable. The purpose was to identify the Level 2 variables that were the strongest predictors of special educator ratings, then include those in the final model to best isolate the unique effect of special educator school density. All variables in Models 3 to 7 were standardized to permit comparisons among the predictor parameter estimates.
With the inclusion of eight Level 1 variables, special educator school density, and another Level 2 variable, Models 3 to 7 yielded three primary findings relevant for revising the model predicting special educator ratings. First, the effect of special educator school density on special educator ratings remained largely unchanged across these five models with a standardized effect ranging from –.24 to –.22, with p values ranging from .04 to .10. Second, the Level 1 variable, special educator IEP caseload, was found to have standardized effects (–.29 to –.27) similar to special educator school density and that were consistently significant at p < .01. Third, the only Level 2 variable with a standardized effect much above 0 was Setting Category (i.e., rural or nonrural); this effect (–.15) was not significant, t(27.68) = −1.30, p = .20. All other Level 2 variables had nonsignificant standardized effects ranging from –.03 to .02.
Model 8
Our final model included all Level 1 special educator variables, special educator school density, and setting category predicting special educator ratings. The overall intercept of 6.89 represents the predicted rating with mean values on all continuous special educator variables in the model, for a teacher whose highest degree earned is a bachelor’s degree, working at a school with an average special educator school density, and in a nonrural setting. The fixed effects for both special educator school density, t(31.55) = −2.16, p = .04, and IEP caseload, t(128.54) = −2.88, p = .005, were significant. No other variables were found to have significant effects predicting special educator ratings. The PVE for this model was .12 at Level 1 and .18 at Level 2, indicating that the model explained 12% and 18% of the variance at each level, respectively.
The significant parameter estimate of –.12 for IEP caseload indicates that while controlling for all other Level 1 and Level 2 variables, for each additional student a special educator has on his or her caseload, his or her rating is predicted to drop by .12 (or by a full point for approximately every 8 students added to the caseload). Similarly, the special educator school density effect represents a predicted drop in special educator ratings by .02 for each one-unit increase in special educator school density while holding all other predictors, including individual caseload, constant at their means. Because special educator school density changes due to fluctuations in special educator FTE, total enrollment, or both, practical examples may help interpret this effect. For example, a school with a total enrollment of 368 and 4.33 in special educator FTE (the study sample means) would have a special educator school density of 84.99 and a predicted special educator rating of 6.89. The current model would predict an increase to approximately 7.25 if either total enrollment dropped by 75 students or the school hired another full-time special educator.
Discussion
In considering the findings, the reader is encouraged to remain cognizant of the study’s limitations. All data were collected from a convenience sample in Vermont. In addition, our sample of 145 special educators in 32 schools was relatively small with an effective sample size of 77 (Snijders & Bosker, 1999). These two factors necessarily limit the generalizability of our findings to other settings. Similarly, because only three high schools participated in this study, the findings are driven largely by data from elementary and middle schools. Although the smaller number of schools limited the number of parameters we could include in the models, all of the models converged. Furthermore, data provided by special educators were based on self-reporting without other corroborative measures. An inherent limitation of any questionnaire is the potential for idiosyncratic interpretation of items. Finally, the HLM outcome variable (i.e., special educator work rating) was based on single-item, rather than multi-item scale. Despite these limitations, the findings offer potentially valuable information regarding special education service delivery with implications for schools seeking to extend inclusive opportunities to more students with disabilities in general education classes.
Overall, the replication aspect of this study strongly confirmed the findings reported in the earlier study (Suter & Giangreco, 2009). These data have numerous implications for practice—here we offer a subset of those we think are most notable in developing coherent special education service delivery in inclusion-oriented schools. The average number of students that special educators supported, with and without IEP designations, rivaled average class size in these schools. With the proliferation of Response to Instruction (RtI), special educators are increasingly providing instructional supports to at-risk students not currently identified as requiring special education (Council for Exceptional Children, 2008); therefore, the ways we think about caseload issues and availability of special educators in a school needs to consider the total school population and needs, rather than exclusively counting students on IEPs. This is particularly important as, like the earlier study, schools that identified fewer students as eligible for special education were less well resourced with special educators than schools that identified higher percentages of students as eligible for special education. These are some of the reasons why a measure such as special educator school density is important to consider in concert with individual special educator caseload parameters.
An imbalanced ratio of special educators to special education paraprofessionals, in combination with personnel time use, contributes to the persistent problem of insufficient time available, about 2%, for supervision of paraprofessionals. Limiting the number of paraprofessionals supervised by any special educator to two and simultaneously developing shared models of supervision with general education teachers may begin to address this problem. This will also require special educator caseload parameters that are suited to two or fewer paraprofessionals to assist them. The percentage of out-of-class instruction (75%) reported by special educators raises a concern about its conceptual compatibility with offering inclusion-oriented instructional supports. As the field develops improved models of service delivery, it will be essential to reconceptualize the roles of special educators in ways that shift away from a narrow emphasis on skill improvement through pull-out instruction (Rea, McLaughlin, & Walther-Thomas, 2002) toward models that facilitate capacity building of general education teachers by embedding and generalizing specialized instructional approaches within general education classes to support students with disabilities and other educational needs. Such a shift will not preclude individual or small group instruction when needed, but rather is meant to encourage a more productive balance with other options (e.g., coteaching) that offer more potential for capacity building. This will require ongoing collaboration between special and general educators, the presence of special educators in general education classes as well as limiting the number of teachers and grade levels any single special educator is assigned to support. As schools pursue multiple school improvement initiatives, it becomes increasingly important that they be integrated to ensure minimally that they are not working at cross-purposes (e.g., intensity of instruction increases whereas access to the regular classroom decreases) and preferably to yield synergistic effects (Guskey, 1990).
As the percentage of out-of-class instruction reported by special educators pertained exclusively to their students on IEPs and did not include students on 504 plans or those considered “at-risk” with whom some of them worked, it is unknown to what extent, if at all, additional pull-out instruction by special educators occurred under the auspices of RtI. As the deployment and potential roles of special educators in RtI are explored, it is vital to keep in mind that Tiers 2 and 3 of RtI need not be synonymous with location of service—these tiers are instructional services, not places. Therefore, when considering possible interventions and locations of service, we should be cognizant of not requiring students to be pulled out of the regular classroom to receive necessary instruction or supports.
The replication data continue to present serious equity concerns for students with disabilities that may constitute violations of the free appropriate public education (FAPE) provisions of the IDEA. It is troubling that approximately three quarters of all instruction provided by special education personnel in this sample was provided by paraprofessionals. This concern extends even further for students receiving one-to-one paraprofessional supports, where 43% received half or more of their instruction from paraprofessionals, rather than highly qualified teachers or special educators. We need to continually ask ourselves whether our models of service provision represent equitable practices and opportunities that would be considered appropriate and desirable for students without disabilities. In addition, the substantial percentage of special education paraprofessional supports provided one-to-one (51%) can wreak havoc with a school’s ability to develop an equitable model of service delivery by restricting the ability to flexibly shift resources. Combined, the replication findings continue to suggest that the current proportion of special educators to paraprofessionals and their respective roles is contributing a ripple effect of concerns that require proactive approaches to developing schoolwide service delivery models (Giangreco, Doyle, & Suter, 2011).
The HLM findings establish special educator school density as a school-level variable that can predict special educator ratings of their work responsibilities as conducive to providing effective special education to students on IEPs. As suggested in the introduction, although a school’s average special educator caseload may not be an adequate predictor of special educator ratings, the HLM findings suggest that individual special educators’ IEP caseloads are predictive of their ratings. Although these two variables account for only a modest amount of the variance contributing to special educator ratings, this study has provided initial data to eliminate other variables and establish these two variables as those to be included in future research attempting to identify what contributes to special educator ratings of their effectiveness.
Special educator school density holds promise as a potential measure to make international service delivery comparisons among countries that utilize some version of a special educator role, yet have differing social constructions and definitions of disability leading to different percentages of students identified as disabled. For example, Italy identifies only about 2% of students as disabled, with 98% placed in general classes, and maximum caseloads of four students with disabilities for insegnante di sostegno (support teachers). Although Italian teachers report having generally positive attitudes toward inclusion of students with disabilities, they are concerned about whether resources for inclusive efforts are adequate (Cornoldi, Terreni, Scruggs, & Mastropieri, 1998). This may seem surprising to American special educators serving substantially higher caseloads. Special educator school density may hold a clue to explaining Italian educators’ concerns. Current legislation in Italy “envisages the ratio of one support teacher for every 138 enrolled students (disabled and non disabled), but the school principal can appoint more support teachers to face school needs” (D’Alessio, 2008, p. 59); based on our initial findings, this ratio would predict low special educator ratings regarding the conduciveness of their work responsibilities to providing effective special education.
In addition, future research would benefit from the development of a more robust measure to ascertain the perceived conduciveness of special educator work responsibilities to effectively meeting the needs of students on IEPs by replacing the single-item rating presented in this study with a validated, multi-item measure. Correspondingly, future research should explore the relationships between various service delivery elements and more direct measures of student access to general education classrooms and curricula, social/behavioral outcomes, and academic/functional achievement. Future research on academic and social interventions could be bolstered by describing a wider array of relevant service delivery parameters within which such interventions have been successfully implemented, extending beyond typically reported parameters (e.g., class size, school enrollment) to less frequently, but contextual important parameters (e.g., special educator caseload size and configuration, special educator school density). Finally, as RtI proliferates in an effort to improve academic achievement, it will be important to simultaneously monitor its impact on the percentage of students identified as requiring special education, the extent of classroom versus pull-out instruction, and to remain cognizant of potential unintended effects. Although we cannot yet unequivocally state how many special educators are needed to staff an inclusion-oriented school, these data contribute to understanding some key parameters schools should consider as they develop coherent models of inclusive service delivery.
Footnotes
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) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was partially funded by the schools that participated in the data collection via their participation in Project EVOLVE Plus at the Center on Disability and Community Inclusion at the University of Vermont.