Volume 17, Issue 2 p. 132-142
Original Article
Free Access

Comparison of the effects of mainstream and special school on National Curriculum outcomes in children with autism spectrum disorder: an archive-based analysis

Emma M. Waddington

Emma M. Waddington

Swansea University

Search for more papers by this author
Phil Reed

Corresponding Author

Phil Reed

Swansea University

Address for correspondence

Phil Reed,

Department of Psychology,

Swansea University,

Singleton Park,

Swansea SA2 8PP,

UK.

Email: [email protected].

Search for more papers by this author
First published: 30 September 2016
Citations: 29

Abstract

The literature dealing with the inclusion of children with autism spectrum disorder (ASD) in mainstream schools has increased over recent years, propelled by the argument that it will improve the quality of life, educational performance and social development of ‘included’ children. This area of research is currently an important one for the development of policy and practice. The literature on inclusion dealing with the inclusion of children with ASD is limited, so the implementation of inclusion has preceded research. The current study investigated whether children in mainstream placements show enhanced performance, relative to those in specialist provisions. The study used a combination of primary and secondary data analysis to explore the impact of inclusion on children with ASD in four authorities in the south east of England. The results suggest that mainstream children have no greater academic success than children in specialist provision. The study suggests that a number of specific provisions are involved in promoting success, such as Speech and Language Therapy, and the impact of Learning Support Assistants, and these are also reviewed and discussed.

Introduction

The inclusion of children with autism spectrum disorder (ASD) into mainstream schools has been argued to improve their quality of life, educational performance and social development (Connor, 2000; Knight, Petrie, Zuurmond, et al., 2009; Kurth and Mastergeorge, 2010; Strain, 1983). Mainstreaming is also thought to increase the social awareness of the other children exposed to the included children (Egel and Gradel, 1988; McGregor and Vogelsberg, 1998). In addition to these putative benefits, inclusion has been argued to relieve some of the financial strain on many external supporting agencies, such as educational, psychological and health services (Jarbrink and Knapp, 2001). Although the definitions of inclusion vary (e.g., children included for play times and meals versus children included all day), the fundamental concept is that children identified with special educational needs should be educated in the same setting as their mainstream peers (Kurth and Mastergeorge, 2010; Mesibov and Shea, 1996; Norwich, 2005; Reed and Osborne, 2014).

However, it has been argued that the promotion and implementation of this ideal has preceded research into the success of such inclusive practices, and that this is especially true concerning children with ASD (Humphrey and Parkinson, 2006; Reed and Osborne, 2014). A small number of studies have observed the effects of inclusion for children with ASD, but these studies have reported mixed results (Kurth and Mastergeorge, 2010; Reed and Osborne, 2014), a pattern which emerges in both the areas of social (Knight et al., 2009; Kurth and Mastergeorge, 2010) and academic (Ruijs and Peetsma, 2009; Smith and Matson, 2010) performance.

In terms of the social benefits of inclusion, Strain (1983) (Boutot and Bryant, 2005; Buysse and Bailey, 1993) found that young children with ASD in mainstream settings exhibited more pro-social behaviours than those in special schools, and that these social skills were generalised best in integrated rather than segregated settings. However, several other studies have shown no such pattern of gains associated with mainstream education for pupils with ASD (Durbach and Pence, 1991; Harris, Handleman, Kristoff, et al., 1990; Reed, Osborne and Waddington, 2011). Additionally, Panerai, Zingale, Trubia, et al. (2009) reported greater gains in a variety of domains for pupils in special school placements compared to those in a mainstream schools (although this effect was overcome when the teaching practices of the special school were imported into the mainstream school). In fact, there is a great deal of evidence to suggest that when children with ASD lack social competence, they can experience a number of negative academic and socio-behavioural outcomes in mainstream settings (Humphrey and Symes, 2010; McIntyre, Blacher and Baker, 2006). Myles, Simpson, Ormsbee, et al. (1993) examined the social interactions of preschool children with ASD when their non-disabled age-matched peers were either present or absent, and their results indicated that teachers interacted less with the students with ASD if their non-disabled peers were present. The children with ASD initiated very few interactions with anyone in either condition. The authors concluded that physical integration was not enough to create social interactions between children with ASD and their peers.

In terms of academic progress, there is very little evidence relating to the impact of inclusion on pupils with ASD (Reed and Osborne, 2014), although there are two reviews of the impact of inclusion on children with intellectual and/or behavioural difficulties. Ruijs and Peetsma (2009) suggest that mainstream placements offer some small advantages to children with mild intellectual disabilities, but acknowledge that there are a number of studies that report no difference between these placements. In contrast, Smith and Matson (2010) suggest that greater academic gains are made by children who displayed behaviour problems in special school.

At the very least, such a pattern of data warrants the conclusion that the ideal of inclusion is not founded on a strong evidence base (Humphrey and Parkinson, 2006; Reed and Osborne, 2014). The importance of identifying the success of this model is then paramount to the ongoing practice of inclusion in schools across the country. In fact, the importance of basing policy decisions on evidence-based practice is recognised, and is beginning to shape the delivery of educational services (Department of Health, 1998a,b). The fundamental argument is that there needs to be a link between professional practice and research (Fox, 2003).

Of course, evidence highlighting best practice could come from a number of sources. Obviously, studies involving the comparison of well-matched groups undergoing different interventions are necessary (Panerai, Zingale, Trubia, et al., 2009; Reed, Osborne, and Waddington, 2011), but there are many practical constraints on the conduct of such studies (e.g., these studies take time and money that might be used for the employment of teachers). However, alternatives to such experimental and quasi-experimental designs do exist. Although primary data analysis uses data collected by the researchers themselves, or through trained observers, often in settings constructed as a part of the research programme, secondary data analysis uses data that have previously been collected by other investigators, often in ‘naturally occurring situations’, and for reasons that differ from those of the research for which they are employed in the secondary analysis. This form of research is being used increasingly as an important source of evidence, especially in the initial stages of an investigation, where it can be used to highlight which of many possible factors could be important for further investigation. In addition to being less expensive than using primary research designs, secondary data can lead to increased sample sizes; number of observations; and ecological validity (all measures coming from actual cases, rather than designed studies, thus, increasing the ecological validity of the findings). Thus, under some conditions, secondary data analysis can be more representative (or more ecologically/environmentally valid), and offer more generalisation potential, than findings obtained from purposefully constructed research programmes.

Secondary data analysis has a long history of use in education both to cut costs, and to make use of the vast amount of data collected on students. For example, secondary data analysis was used in the USA to study the trends in achievements as a function of age at admission using data collected by the National Assessment of Educational Progress in the United States (Langer, Kalk and Searles, 1984). A further example of secondary data analysis relevant to special needs education comes from a proposed method to demonstrate accountability of decisions for students with disabilities in the USA. This study re-analysed extant data on educational performance of children with special educational needs in order to see how children with disabilities were performing both academically and non-academically as compared to their non-disabled peers (Ysseldyke, Thurlow, Langenfeld, et al., 1998). For this study, all of the publicly available reports produced by state departments of education, containing student outcome data such as achievement test performance, were collected. The summary of the performance data revealed lower performance for students with disabilities compared to other students and lower rates of participation on tests compared to students without disabilities (e.g., 50–80%).

Given the need to establish evidence for the policy of inclusion for children with ASD, and given the availability of secondary data in this area, the current study uses a similar methodology to Ysseldyke, Thurlow, Langenfeld, et al. (1998) to analyse educational provisions for children with ASD in the UK. Local authorities responsible for the education of children in a particular area hold archive data on all children with ASD in their local authority. This archive data could contain possible predictive and outcome measures of the success of the inclusion of the child, which could provide an invaluable source of information concerning the success of inclusion and may help identify the common factors leading to success. Consequently, such an analysis may help to improve the current provision of the participating local authorities. Additionally, the collection of this data will allow us to identify gaps where data collection needs to be improved in the participating local authorities. In particular, the current analyses focused on the impact of a wide range of factors [e.g., type of ASD diagnosis, autism severity, socio-economic status (SES), Learning Support Assistant (LSA) time, and types of intervention given to the children, such as Portage, Speech and Language Therapy (SLT), social skills training; these interventions were chosen purely on the basis of the data which were available] on both the school placement, and the national curriculum results, of the children (see Table 3, for a description).

Method

Sample

One hundred and eight children (18 girls and 90 boys) with a diagnosis of ASD, from four local authorities in the South East of England, formed the sample for this study. The criteria for inclusion of a participant in the study were that they had a diagnosis of an ASD, made according to the DSM-IV-TR, by a Paediatrician independent from the current study prior to the start of the study, and they could not have left school more than 5 years before. Local authorities were contacted, and those who agreed to take part provided a list of parents.

The parents were then sent a letter outlining the aim of the study, and asking them if they would consent to their child's data being accessed from the local authority archives. The letter stressed that no personally identifiable data (names) would be extracted from the files. A consent form was included with the letter, which could be returned to the study authors using a prepaid envelop if the parent consented. If they consented, then the data from that child were recorded from the archive without recording their name. A total of 213 parents were contacted, and 108 consent forms were returned, giving a response rate of 51%.

The distribution of the diagnosis of participants was gathered, and revealed that 72% of the participants had a diagnosis of Autism, 16% had a diagnosis of Asperger syndrome (AS), 7% had a diagnosis of attention deficit hyperactive disorder in addition to an ASD diagnosis, and 5% had an additional diagnosis of Tourette's syndrome, dyspraxia or depression. The age of the participants ranged between 5 and 17 years old, with a mean age of 13 years.

Location

The characteristics of the four local authorities in the South East of England that took part in the study are displayed in Table 1. These measures were obtained from the Census for each local authority. Local authorities A, B and C had the same index of unemployment as one another, while local authority D had a lower index than the others. All had indices slightly lower than the mean in the UK. A total of 46 mainstream schools, four units and 17 special schools were sampled for the study. The breakdown of the types of schools sampled per local authority (mainstream, special, etc.) is displayed in Table 2.

Table 1. Characteristics of the participating local authorities in terms of population, ethnicity and socio-economic status (unemployment)
Local authority Population Ethnic make-up Index of unemployment (percentage of available workforce not employed)
A 211 600

59% White British

41% non-White

3%
B 185 131

88% White British

12% non-White

3%
C 372 000

94% White British

6% non-White

3%
D 150 229 94% White British, 6% non-White 1.5%
UK 58 789 194 80% White British, 20% non-White 5%
Table 2. Breakdown of types of school and number sampled per local authority
Local authority Mainstream Unit Special
A 13 0 0
B 13 2 11
C 6 2 3
D 14 0 3

Measures

Archive measures

Measures were taken from the archives concerning child outcomes, measured by national curriculum results and by school placement. Additionally, the interventions that the child had undergone, such as access to SLT, Social Skills Training and Portage, were recorded through archive analysis. The measures found in the archives for each LAs. varied. There were 15 measures collected for local authority A, 14 measures collected for local authority B, 10 collected for local authority C and 16 measures for local authority D. In addition, the measures collected were not consistent from child to child within the LA. This was most evident in terms of the Educational Psychologists assessments for each child. Despite such inconsistencies, outcome and predictive measures were obtained for each child in all four LAs. Table 3 summarises the predictive measures and their potential outcome measures taken from the archives of the four LAs.

Table 3. Identified measures from the archive data broken down into predictor variables and potential outcome measures
Predictors Outcome
  • Diagnosis
  • Portage
  • Hours of Learning Support Assistant
  • Speech and Language Therapy
  • Social Skills training
  • Socio-economic status
  • Autism severity
  • School placement
  • Diagnosis
  • Portage
  • Years of statement
  • Hours of Learning Support Assistant
  • Speech and Language Therapy
  • Social Skills training
  • Socio-economic status
  • Autism severity
  • National curriculum results

Questionnaires

In addition to the archive data collected, two questionnaires were sent to parents covering three areas: diagnosis, developmental and medical history.

Autism severity

The Autism Behaviour Checklist (ABC; Krug, Arick and Almond, 1980) was employed to assess the severity of the autism of each child. The ABC is a 57-item checklist, grouped into five areas; sensory, relating, body and object use, language, and social and self-help skills. A total score of 67 or more is taken by Krug, Arick, and Almond (1980) to suggest probable autism, and scores between 55 and 67 suggest possible autism. The intra-rater reliability of the test is 0.94, and its validity is regarded as satisfactory (Volkmar, Cicchetti, Dykens, et al., 1988). However, it is important to note that the ABC measure may not give a similar picture of the child's autism as other instruments (Shaffer, Lucas and Richters, 1999). These issues tend to reflect the greater focus of the ABC, compared to other measures, on language skills. However, these issues were not regarded as a problem for the present study, because it is only used as an index of the autism symptomatology, and not as a diagnostic instrument. Additionally, the ABC was still considered useful in the present context as: (1) no special training in administration or scoring is required, and, in the current study, it was to be completed by parents, who tend, on average, to produce higher scores than teachers (Volkmar, Cicchetti, Dykens, et al., 1988); and (2) it was to be used as a research tool gauging the relative effects of autism symptomatology across the participants, rather than to make absolute judgements regarding the impact of symptoms.

Child's history

The ‘Parent's Questionnaire on Your Child's History’ was used to collect information on the child's medical and educational history. The questionnaire consists of questions regarding initial diagnosis, medical problems (allergies), vaccinations and early intervention. In addition, there were also questions about the current provision for the child (SLT or placement). This tool has previously been used in compiling background information concerning treatment integrity in studies of the outcome effectiveness of early intervention studies for ASD (Reed, Osborne and Corness, 2007).

Procedure

The archive data sample was identified in conjunction with the LA. Consent forms were sent out to parents. Once consent was obtained, the archive data for the children within each of the participating local authorities were accessed. The descriptive data on the children were collected, as well as possible predictors and outcome measures of success (see above). The data collection process was iterative, with repeated visits to each of the authorities’ archives, in turn, impacting on the decisions taken about which measures to employ. The initial assessment identified potential measures. The measures were then refined as the data that were common to all archives across the local authorities was identified. Schools were contacted, if necessary, to obtain national curriculum results. Each provision was identified as that named in the child's statement, and was the place where each child spent the majority of the day. Mainstream provision was defined as regular school placement (i.e., not special school). Special schools were schools with specialised provisions, while units were specialised classrooms attached to a mainstream school. In addition to this data collection, the family of the child was also contacted, the purpose of the project explained, and the questionnaires were sent to the families.

Analysis

For the purpose of analysis, there were two measures of outcome: school placement and national curriculum result. Each outcome had a set of predictors (displayed in Table 3). Each outcome measure was analysed in terms of the possible predictors in order to identify any possible relationships and interactions. When data were missing, it was replaced by mean substitution. Mean substitution was deemed a more appropriate method than listwise deletion, or regression replacement, as listwise deletion would lead to heavy data loss, and the use of regression was not applicable as there were no multiple measures available to assess related factors. Moreover, mean substitution is a very conservative and transparent method of dealing with missing data, although it does lead to a loss in variability in the data (Tabachnick and Fidell, 2007). In no cases was there more than 15% missing data, and no single measure had more than 10% missing data.

Results

Table 4 presents the mean, maximum and minimum values for age, school year, years of statement and hours of LSA a week (given specifically to the child, and not merely the presence of an LSA in the classroom), for the 108 children in the sample. There was a wide range of variation in terms of LSA help. The number of hours of LSA per week ranged from 1 to 35 hours per week, with an average of 18 hours a week per child. In addition, the proportion of children receiving SLT, Portage training and Social Skills Training are displayed in Table 4. Due to insufficient data, only access to, rather than amount of these interventions was recorded. SES was measured as the percentage of free school meals at the child's school. The schools involved came from areas that presented a large variance in SES (as measured in percentage of free school meals) ranging from 3% to 48% of children in the school having free school meals. The average autism severity for the entire sample was 55.7, with a range of 0–154, on the ABC, suggesting possible autism, and that the sample had moderate levels of autism severity.

Table 4. Descriptive statistics of selected variables for the total sample of students
Variable N Minimum Maximum Mean SD
Age (years) 108 5 18 12.9 3.2
School years 108 0 13 7.3 3.0
Years of statement 108 0 15 6.1 3.6
Hours of LSA 108 1 35 18.6 7.1
Visits of SLT (Yes/No) 67 0 1 N/A N/A
Portage (Yes/No) 108 0 1 N/A N/A
Social skills training (Yes/No) 108 0 1 N/A N/A
Free school meals (percentage) 108 3 48 18.3 7.5
Autistic severity 108 0 154 55.7 22.8
Parental coping 108 61 115 91.7 6.7

School placement

Table 5 displays the proportion of children with ASD placed in each of the provisions across the four local authorities. Across local authorities A and D, children were overwhelmingly more likely to be placed in mainstream schools. In local authority B, children were more likely to be placed in special school, while, in local authority C, children were equally placed in special school or in mainstream. Mainstream units had the lowest number of children across all local authorities. There were two children who were home educated in the sample of 108 children.

Table 5. Provision across the four local authorities
Local authority Mainstream Special Unit Home
A 94% (16) 0% (0) 0% (0) 6% (1)
B 36% (16) 48% (21) 14% (6) 2% (1)
C 45% (10) 45% (10) 5% (2) 0% (0)
D 70% (19) 30% (6) 0% (0) 0% (0)

Table 6 displays the diagnosis and the severity of autism for children in the different forms of school placement. The proportion of children with diagnoses of ASD and AS placed in each type of school placements was broadly similar to one another, and a chi-square analysis did not reveal any statistically significant differences between diagnosis and placement, so children with ASD, AS or ASD co-morbid were not more likely to be placed in either mainstream or special school (x2 = 1.41, NS).

Table 6. Autistic severity and school placement
School placement
Mainstream Special Unit Homea
Diagnosis
ASD 59% (46) 35% (27) 6% (5) 0% (0)
AS 61% (11) 28% (5) 11% (2) 0% (0)
ASD/co-morbid 33% (4) 42% (5) 8% (1) 17% (2)
Mean ASD severity (SD)
Total ABC (31–155) 50.9 (2.5) 64.0 (4.6) 54.0 (1.8) 55.7 (0.0)
Sensory subscale (0–27) 7.9 (0.5) 9.4 (0.9) 8.1 (0.3) 8.4 (0.0)
Relating subscale (4–38) 15.1 (0.7) 19.3 (1.2) 16.6 (0.1) 16.7 (0.0)
Body and object use subscale (0–38) 8.9 (0.6) 11.2 (1.2) 9.6 (0.2) 9.8 (0.0)
Language subscale (0–31) 8.5 (0.6) 10.3 (1.1) 8.0 (1.1) 9.1 (0.0)
Social and self-help skills subscale (6–25) 10.4 (0.5) 14.0 (0.8) 11.6 (0.1) 11.7 (0.0)

Note

  • a There were only two participants therefore it is not possible to compute the standard deviation.

Those children placed in mainstream had an average score of 50.9 on the ABC, which was lower than the mean score for children placed in special school (64.0), but only marginally lower than that for the special units attached to mainstream (54.0), and those educated at home (55.7). The children in special schools had statistically significantly more severe autism symptoms as measured by the total ABC score than those in mainstream settings. This difference was assessed by a non-parametric Mann–Whitney test, which revealed a statistically significant difference between the scores (z = −2.21, P < 0.05). The special school group also had more severe problems with the relating (Mann–Whitney, z = −2.82, P < 0.05) and social skills (Mann–Whitney, z = −3.45, P < 0.001) subscale of the ABC. However, there were no statistically significant differences between the mainstream children and those attending units or home educated. There were also no differences between the children in special schools and those attending units and home educated.

Table 7 shows the characteristics of the provision that the children in each placement had received. For the purpose of analysis, the children educated at home were removed due to insufficient numbers. There was no difference between placements in terms of whether the child had access to SLT, P > 0.05. Children in all placements had LSAs, and there were no statistically significant differences between the placements and the amount of learning support hours received, all ps > 0.05. Having Portage as an early intervention did not have a statistically significant impact on subsequent school placement, P > 0.05 (although it is important to note that the number of children who had Portage was small and conclusions need to be taken cautiously). The results also suggest that children across both mainstream and special were getting the same access to Social Skills Training, P > 0.05. Finally, there were no statistically significant differences between the provisions in free school meals, P > 0.05.

Table 7. Descriptive data on predictor variables
School placement (number in brackets)
Mainstream Special Unit Home
SLT
Yes 78% (32) 88% (14) 88% (7) No data
No 22% (9) 12% (2) 12% (1) No data
LSA
Mean hours (1–35) 18 19 19 19
Percentage receiving 100% 100% 100% 100%
Portage
Yes 8% (6) 8% (2) 33% (4) 0%
No 92% (65) 91% (21) 66% (8) 100% (2)
Social skills training
Yes 27% (19) 35% (8) 42% (5) 0% (0)
No 73% (52) 65% (15) 58% (7) 100% (2)
SES (3–48%) 19% 17% 20% 15%

Academic success

In order to determine whether the children included in mainstream schools were more or less successful academically than those not fully included (i.e., those in special units and special schools), the mean scores for their performance on national curriculum Tests were assessed. No significant correlations were found between the overall ABC scores and national curriculum outcomes. These correlations suggest little direct relationship between autism severity and outcome.

Figure 1 displays the national curriculum results for children in mainstream and special provisions (special schools, units and home tuition). In order for the data on national curriculum results to be comparable across students, all the levels were recoded so that: P-level 1 = 1, P-level 2 = 2, P-level 3 = 3, and so on up to P-level 8 = 8, the Level 1 = 9, Level 2 = 10, and so on. The results suggest the mean performance level across both mainstream and special schools is low (around P8). Despite the mean age of the current sample being 12.9 years, a performance at P8 level is below that which would normally be expected from this age group – that is Level 4/5 (or Key Stage 3).

Details are in the caption following the image
National curriculum results for children in mainstream and specialist provision (P-level 1 = 1, P-level 2 = 2, P-level 3 = 3, and so on, up to P-level 8 = 8, then Level 1 = 9, Level 2 = 10, and so on)

Due to the violation of the assumption of normality (tested by the Kolmogorov–Smirnov statistic), non-parametric tests were used to statistically analyse these data. These tests revealed that the children in specialist provision did statistically significantly better in English than those in mainstream provision (Mann–Whitney, z = 2.26, P < 0.05). The means for the rest of the national curriculum outcomes were similar to one another, and Mann–Whitney tests failed to note any statistically significant differences between the provisions, all zs < 1. As a number of tests were conducted, so caution is needed in interpreting a significance level of P < 0.05.

Relationship between school factors and academic success

To further determine if any aspect of the provisions that the children had previously received were associated with academic success, a series of correlations and partial correlations were performed between the school factors, autism severity and academic outcomes. All correlations were calculated using a non-parametric correlational procedure (either a Kendall correlation or a Kendall partial correlation test). These results have been broken down for mainstream placements, and special placements (special schools and units), and for the sample as a whole, and all are reported in Table 8.

Table 8. Correlation matrix of predictor and outcome measures in the sample
Predictors\Outcome Provision ABC NC English NC Reading NC Writing NC Science NC Maths
SES Mainstream

K = 0.12

NS

K = 0.04

NS

K = 0.10

NS

K = 0.14

NS

K = 0.15

NS

K = 0.10

NS

Special

K = 0.11

NS

k = −0.02

NS

k = 0.02

NS

K = 0.14

NS

K = 0.15

NS

K = 0.10

NS

Combined

K = 0.10

NS

K = 0.10

NS

K = 0.06

NS

K = 0.10

NS

K = 0.11

NS

K = 0.10

NS

LSA hours per week Mainstream

K = −0.17

NS

K = −0.30

P < 0.01

K = −0.27

P < 0.01

K = −0.29

P < 0.01

K = −0.32

P < 0.01

K = −0.28

P < 0.01

Special

K = −0.22

NS

K = 0.023

NS

K = 0.08

NS

k = 0.08

NS

k = 0.11

NS

k = 0.06

NS

Combined

K = −0.10

NS

R = −0.16

P < 0.05

R = −0.12

NS

R = −0.15

NS

R = −0.16

P < 0.05

K = −0.15

NS

Notes

  • Combined = Mainstream and Special (special school and units) combined together; SLT, Speech and Language Therapy; SES, socio-economic measure (number of free school meals); NC, National Curriculum; ABC, Autism Behaviour Checklist.

There were no correlations between SES and autism severity, SES and academic outcomes, or between hours of LSA support and autism severity, suggesting that those children who have more hours of LSA support are not more severe than those children who have less hours of LSA. There were several significant negative correlations between LSA support hours and outcome for the sample as a whole and for pupils in mainstream provisions. In contrast, for children in special schools, hours of LSA support were not significantly correlated with outcomes.

Figure 2 displays the mean academic outcomes for children who did, and who did not, have access to Portage, Social Skills Training and SLT. A Mann–Whitney test revealed no significant differences between academic outcomes depending on whether a child had had access to Portage, P > 0.10. Kendall's correlations between Portage and academic outcomes also revealed no significant correlations between access to Portage and outcomes for pupils in mainstream schools, special schools or combined across the whole sample. There was no significant correlation between autism severity and Portage, P > 0.10, and Kendall's partial correlations between Portage and academic outcomes, with autism severity controlled, revealed that there were actually negative correlations between access to Portage and outcomes for the mainstream group: English (T = −0.21, P < 0.05), Reading (T = −0.21, P < 0.05), Writing (T = −0.23, P < 0.05), Science (T = −0.18, P < 0.05) and Math (T = −0.26, P < 0.01). Again, these conclusions need to be taken very cautiously, due to the small number of children who had access to Portage.

Details are in the caption following the image
Relationship between intervention (present = yes; absent = no) and academic success measured in terms of p values (P-level 1 = 1, P-level 2 = 2, P-level 3 = 3, and so on, up to P-level 8 = 8, then Level 1 = 9, Level 2 = 10, and so on)

A Mann–Whitney analysis displayed significant differences between the outcomes of those children in mainstream accessing Social Skills Training and those who did not have such access. Children who accessed Social Skills Training had statistically significantly lower grades in English (z = 2.50, P < 0.05), Reading (z = 2.80, P < 0.01), Writing (z = 2.42, P < 0.05), Science (z = 2.40, P < 0.05) and Maths (z = 2.90, P < 0.01). In addition, a Kendall's correlation revealed statistically significant negative correlations between access to Social Skills Training and poorer outcomes for children in mainstream schools: English (T = −0.37, P < 0.001), Reading (T = −0.38, P < 0.001), Writing (T = −0.34, P < 0.01), Science (T = −0.33, P < 0.01) and Math (T = −0.35, P < 0.001). However, there was no statistically significant correlation between Social Skills Training and autism severity in the mainstream group, P > 0.10. This negative relationship between Social Skills Training and outcomes was not present in children in special schools in both correlations and partial correlations, all ps > 0.10. In addition, there was no correlation between severity and access to social skills for those children in special school, all ps > 0.10. However, the negative correlation between Social Skills Training and outcome was present when the two groups were combined: English (T = −0.21, P < 0.01), Reading (T = −0.24, P < 0.01), Writing (T = −0.21, P < 0.01), Science (T = −0.21, P < 0.01) and Math (T = −0.24, P < 0.01). As with the sub-group analyses, there was no correlation between autism severity and Social Skills Training in the combined group, P > 0.10. A partial correlation between Social Skills Training and outcomes, revealed that, even when autism severity was partialled out, access to Social Skills Training remained negatively correlated with outcomes in: English (T = −0.37, P < 0.001), Reading (T = −0.38, P < 0.001), Writing (T = −0.34, P < 0.001), Science (T = −0.33, P < 0.001) and Math (T = −0.35, P < 0.001).

Finally, a Mann–Whitney test revealed that those children who had access to SLT were performing statistically significantly better at English (z = 2.84, P < 0.01), Reading (z = 2.80, P < 0.01), Writing (z = 2.73, P < 0.01), Science (z = 2.51, P < 0.05) and Maths (z = 2.71, P < 0.01). The positive impact of SLT on outcomes was confirmed by a series of Kendall correlations. In both mainstream and special schools, there were no significant correlations between SLT and academic outcomes. However, when the groups were combined, statistically significant correlations emerged. Children in the combined group who had previously accessed speech and language therapy did better in English (T = 0.32, P < 0.01), Reading (T = 0.30, P < 0.01), Writing (T = 0.30, P < 0.01), Science (T = 0.28, P < 0.05) and Math (T = 0.30, P < 0.01). A partial correlation between access to SLT and outcomes, with autism severity partialled out, revealed that there were statistically significant correlations between access to SLT and outcomes in: Reading (T = 0.18, P < 0.05) and Writing (T = 0.18, P < 0.05), for those children in mainstream. For those children in special school, a partial correlation revealed statistically significant correlations between SLT and outcomes in English (T = 0.33, P < 0.001), Reading (T = 0.33, P < 0.001), Writing (T = 0.32, P < 0.001), Science (T = 0.32, P < 0.001) and Math (T = 0.32, P < 0.001).

Discussion

The recent debates over governmental policies regarding inclusion make investigating the success of inclusion an important area for research and practice. The current study was concerned with identifying, whether an archive-based analysis could identify whether children with ASD in mainstream do better than those in specialist provision, and whether there were any factors involved in mediating the outcome. The results suggest that children in mainstream are not more academically successful than those in specialist placements, but, instead, a range of alternative factors are associated with success.

The archive data suggest a pattern of practice that is not entirely in accordance with the ‘green paper’, in that children with ASD were just as likely to be placed in special school as in a mainstream school. In this respect, inclusion in mainstream appears to be at about the same level as 16 years ago, when Barnard, Prior and Potter (2000) noted that about 50% of such pupils were included in mainstream classes. The current report finds that mainstreaming practice varied across local authorities. However, there were significant differences in the severity of ASD across the school placements. Those children in special school generally had more severe ASD, and had poorer social relating, and social skills, than those children placed in mainstream schools. This suggests that children are being placed in the different provisions as a function of their ASD severity. There were no differences in the SES of the children and their placement. In terms of provision received by the children in either type of placement, there were no differences in the access to interventions between the different school placements in terms of Social Skills Training, SLT and LSA support.

The academic performance of children on national curriculum levels in mainstream and specialist provision was analysed in order to identify whether included children were more or less successful than those in special units or special schools. Children in special school performed better in English than those in mainstream; however, there were no further differences in the academic performance across the provisions, suggesting that inclusion in itself does not have a significant impact on academic success. The current study did not find that autism severity had an impact on national curriculum outcomes. The reason why no correlations between autism severity and outcomes were identified may be because children in the current study were performing at low levels overall on the national curriculum, performing significantly below the average level.

The impact of a variety of different factors, and different provisions (rather than school placement) on national curriculum results, also were analysed. It is worth noting that children with more hours of LSA support did not have more severe autism than those who had fewer hours of LSA. Of course, LSA support might not be allocated solely on the basis of severity of ASD (e.g., ability might be an additional consideration in allocation of LSA support). The rationale for providing such support needs to be further explored.

Hours of access to LSAs were negatively correlated with academic outcomes for those children placed in mainstream schools. Such findings have been found previously, and have formed the basis of a number of criticisms regarding the use of LSA support. For example, Ainscow (2000) (Osborne and Reed, 2011) suggests that having an LSA can create a barrier between students and their classmates, and can stall pupil's progress by consistently decreasing the challenges of the work in the classroom. Ainscow (2000) also raises a concern that having an LSA means that the teacher is less involved with the student. This in turn may mean that the child with SEN is benefiting less from their teacher's expertise than other pupils in the class. In addition, the differentiation process may indirectly affect the impact of the LSA on performance. Tasks are often differentiated in mainstream classrooms to accommodate the range of needs and abilities of the pupils. The problem with differentiation is that it can also lower the expectations on the child (Ainscow, 2000), which may in turn lead to lower outcomes. In order to identify whether teaching targets have an impact on outcomes, children's targets would need to be identified and assessed in conjunction with their abilities, in order to identify whether children are underperforming. It should also be noted that factors like the ability of the child may also play a role in these findings of negative relationships between LSA support and outcomes. Support from an LSA may be allocated on the basis of enhanced needs, meaning that the child with LSA support may start from a lower level of achievement to begin with, making the final outcome likely to be lower. Hence, the negative relationship between LSA support and outcomes may be a product of greater allocation of LSAs to those with poorer ability, rather than the LSA intervention producing a worse outcome. In addition, there are a number of LSA factors that have been identified as promoting their impact on the included child with ASD (Symes and Humphrey, 2011).

Those children who attended Social Skills Training in mainstream schools did worse across the national curriculum subjects than those who did not attend Social Skills groups, even when ASD severity was controlled. However, this association was not present for those children who were in special schools. The results did not suggest a difference in ASD severity between those children in mainstream school who were attending Social Skills Training and those who were not attending such training. Of course, children who attended Social Skills Training may have difficulties in communication and language other than those measured by the ABC; therefore, it follows that these children would perform worse than those that were not in need of Social Skills Training.

Access to SLT had significant positive impacts on academic success across all of the subjects (even with ASD severity controlled). Communication interventions can lead to decreased challenging behaviours, when individuals with autism are taught specific language skills to serve the same communicative function as the challenging behaviour (Carr and Durand, 1985; Durrand and Carr, 1987, 1992). The decrease of inappropriate behaviours in children with ASD may affect their academic achievement, as it does with children with challenging behaviour (Luiselli, Putnam, Handler, et al., 2005). In addition, speech and language therapy may improve social competence by targeting reciprocal interactions and peer initiations (McGee, Almeida, Sulzer-Azaroff, et al., 1992), and social behaviour (Goldstein, Kaczmarek, Pennington, et al., 1992). This may lead to improved academic outcomes as research suggests that children lacking social competence go on to develop a number of negative academic outcomes (Kupersmidt and DeRosier, 2004). In order to identify how SLT works best, future investigations will need to identify specific nature of treatment and the effects of intensity on outcomes.

There are limitations concerning the present study that do need to be mentioned in order to allow these findings to be viewed with appropriate caution. Firstly, the findings are not based on an experimental or a quasi-experimental approach, which means that any interpretation given about the causal structure of these data should be made with caution. Any of the findings reported here could imply any one of a number of causal structures between the variables. However, the current relationships do suggest a number of places to start in order to explore the structure of potentially important relationships; such as further exploration of the impact of LSA support, and early interventions, on outcomes (Osborne and Reed, 2011). Secondly, in any such analysis, there should be caution taken regarding the validity of the measures used, the present measures (e.g., the ABC score for autism severity, national curriculum results for academic achievement) have reasonable reliability for research purposes, but are rather more suited to exploring relative effects of the measures, rather than the impact of the absolute level on these measures. Thirdly, in any such archive-based analysis, there are missing data, which will impact on the analyses that are performed. In the present case, the levels of missing data were relatively small (under 10% for any measure), and the treatment was conservative (tending to reduce variance, and so reduce correlational values).

However, the main limitation to the study was inconsistencies in the archive material. Additionally, as with all secondary data analysis, one cannot be sure of the quality of the data. Nevertheless, it was one of the purposes of this study to use extant data to establish an evidence-based practice which could be used in the future for accountability. Additionally, the use of secondary data analysis in this case has led to more representative data, and generalisation potential, than findings obtained from primary research programmes, due to the number of children and local authorities involved. In order for evidence-based practice to be incorporated into LEAs, archives need to include up-to-date information on the children as well as national curriculum results, and educational psychologist reports and assessments. It would also be important to have consistent educational measures for the children within and across local authorities to help assess progress and accountability of placement.