Content uploaded by Neil Humphrey
Author content
All content in this area was uploaded by Neil Humphrey on Jun 11, 2015
Content may be subject to copyright.
This article was downloaded by: [The University of Manchester Library]
On: 31 August 2012, At: 03:21
Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
International Journal of Inclusive
Education
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/tied20
The role of school and individual
differences in the academic attainment
of learners with special educational
needs and disabilities: a multi-level
analysis
Neil Humphrey
a
, Michael Wigelsworth
a
, Alexandra Barlow
a
&
Garry Squires
a
a
Educational Support and Inclusion, School of Education,
University of Manchester, Oxford Road, Manchester, M13 9PL, UK
Version of record first published: 31 Aug 2012
To cite this article: Neil Humphrey, Michael Wigelsworth, Alexandra Barlow & Garry Squires
(2012): The role of school and individual differences in the academic attainment of learners with
special educational needs and disabilities: a multi-level analysis, International Journal of Inclusive
Education, DOI:10.1080/13603116.2012.718373
To link to this article: http://dx.doi.org/10.1080/13603116.2012.718373
PLEASE SCROLL DOWN FOR ARTICLE
Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-
conditions
This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation
that the contents will be complete or accurate or up to date. The accuracy of any
instructions, formulae, and drug doses should be independently verified with primary
sources. The publisher shall not be liable for any loss, actions, claims, proceedings,
demand, or costs or damages whatsoever or howsoever caused arising directly or
indirectly in connection with or arising out of the use of this material.
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
The role of school and individual differences in the academic
attainment of learners with special educational needs and
disabilities: a multi-level analysis
Neil Humphrey
∗
, Michael Wigelsworth, Alexandra Barlow and Garry Squires
Educational Support and Inclusion, School of Education, University of Manchester, Oxford
Road, Manchester M13 9PL, UK
(Received 14 March 2012; final version received 1 August 2012)
Students with special educational needs and disabilities (SEND) are at a greatly
increased risk of poor academic outcomes. Understanding the factors that
influence their attainment is a crucial first step towards developing more
effective provision. In the current study we present a multi-level, natural
variation analysis which highlights important determinants at school and
individual levels in two core aca demic subjects (English and Maths) using a
nationally representative sample of over 15,000 students with SEND attending
more than 400 schools across England. We found that at the school level,
inclusivity, attainment, free school meal (FSM) eligibility, behaviour (in primary
schools) and linguistic diversity (secondary schools), and at the student level,
age, sex, FSM eligibility, SEND provision, SEND primary need, attendance,
behaviour and positive relationships each contributed to the distribution of
academic attainment. Implications of these findings are discussed and study
limitations are noted.
Keywords: academic attainment; special educational needs and disabilities;
hierarchical linear modelling; inclusion
Introduction
The study reported in this article examines the role of school and individual differences
in predicting the academic attainment of learners with special educational needs and
disabilities (SEND) attending mainstream schools in England. Our objectives were
(i) to determine the amount of variation in the academic attainment of students with
SEND that exists between schools (school effects) and within schools (individual
effects) and (ii) to determine which characteristics at school and individual levels
explain significant variation in this attainment. The research constitutes a significant
and original contribution to knowledge as the first large-scale multi-level empirical
investigation of the factors associated with academic attainment in a sub-group of
the school population who are widely considered to be the most vulnerable to poor
outcomes. Developing our understanding of which factors are important and the role
ISSN 1360-3116 print/ISSN 1464-5173 online
# 2012 Taylor & Francis
http://dx.doi.org/10.1080/1360311 6.2012.718373
http://www.tandfonline.com
∗
Corresponding author. Email: neil.humphrey@manchester.ac.uk
International Journal of Inclusive Education
2012, 1–23, iFirst Article
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
that they play in determining the academic attainment of students with SEND is a
crucial first step towards developing more effective provision.
Prevalence estimates for SEND vary by country – for example, 21% in England
(Department for Education 2010a; Office for Standards in Education 2010), 13.2%
in the USA (Institute of Education Sciences 2011) and 7.6% in Australia (Australian
Institute of Health and Welfare 2004). This variation is accounted for by differential
definitions of what constitutes SEND and the social, political and legal influences
that affect schooling systems across the world (Robson 2005). In developed nations,
the overwhelming majority (for example, 91.4% in England – Department for
Education 2010a; 95% in the USA – Institute of Education Sciences 2011) attend
mainstream (‘regular’) schools. The academic attainment of such children and young
people is a matter of concern. Analysis of national statistics in England, for example,
has shown that an ‘attainment gap’ exists between students with and without SEND
in core curriculum subjects such as Maths, English and Science. This gap persists
throughout the different phases of schooling, such that by the end of compulsory edu-
cation, just 16.5% of students with SEND achieve the expected level of academic
attainment (at least 5 A
∗
–C General Certificate of Secondary Education (GCSE)
grades including English and Maths), compared to 61.3% of those with no SEND
(Department for Education 2010a). Similar findings have been reported elsewhere,
including the USA (Zhang, Katsiyannis, and Kortering 2007). This has obvious impli-
cations for the future life chances of these learners, including greatly reduced opportu-
nities for further study (e.g. further and higher education), training and employment
(and, for those who gain employment, reduced earning power) (Robinson and Oppen-
heim 1998). Put simply, the low academic attainment of students with SEND puts them
at an extremely high risk of experiencing social exclusion as adults (Sparkes 1999).
However, whilst we know a great deal about the size and persistence of the achievement
gap for students with SEND, we know very little about the factors that influence the
magnitude of this disparity. Furthermore, the fact that the SEND achievement gap
exists should not be seen as an excuse for lowered expectations or inaction; rather, it
should serve as a reminder of the need for a co-ordinated response to ensure that vul-
nerable learners are given effective support.
Influences on academic attainment of students with SEND – a (bio)ecosystemic
perspective
The academic progress of children and young people varies as a function of influences
at different levels, including the individual, family and school (Rutter and Maughan
2002; Rutter et al. 1979; Sellstro¨m and Bremberg 2006). Uri Bronfenbrenner’s influen-
tial (bio)ecosystemic model of human development is a useful theoretical framework in
this regard (see Bronfenbrenner 2005, for a comprehensive review of his theory). Bron-
fenbrenner proposed a series of nested systems – the microsystem, mesosystem, exo-
system, macrosystem and chronosystem
1
– with which the individual interacts through
the lifespan, shaping their development. The most proximal of these systems – the
microsystem – represents the immediate social context of the individual, including
school, family and peers. Bronfenbrenner proposed that the individual’s endogenous
features (e.g. gender) were an important part of this microsystem (hence the reference
to ‘bio’ ecosystemic theory). The mesosystem describes the links between different
elements of the microsystem – for example, the relationship between a child and
his/her peers. Finally, the exosystem and macrosystem represent broader, more distal
2 N. Humphrey et al.
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
influences, including (respectively) educational, political and economic systems and
overarching cultural beliefs and values (Bronfenbrenner and Morris 1998).
The (bio)ecosystemic theory of human development is a useful ‘organising idea’ in
the context of the current research as it helps us to understand the variety of influences
to which students with SEND are subjected. At the level of the macrosystem, core
beliefs and values about individual differences and, in particular, disabilities, have
changed over time to reflect a more inclusive philosophy (Atkinson et al. 2005),
although some authors argue that basic ideologies regarding difference and disability
still need to be challenged (Johnson 2010). A key element of the exosystem has
been the policy directives and subsequent changes in educational systems, structures
and practices pertaining to educational placement and provision for learners with
SEND. In England, the Excellence for All Children Green Paper (Department for Edu-
cation and Skills 1997) continued the international shift towards more inclusive edu-
cation that followed the Salamanca and Dakar agreements (United Nations Scientific
Educational and Cultural Organisation 1994, 2000). Accompanying politically driven
developments in thinking and practice around identification and assessment of
SEND have also occurred (Department for Education and Skills 2001), in addition to
the evolution of provision in schools (for example, the Achievement for All (AfA) intia-
tive – see Humphrey and Squires 2011b).
The current study focuses on understanding which aspects of the micro- (e.g. school
size) and meso- (e.g. relationships with teachers and peers) systems influence the aca-
demic outcomes of students with SEND. School and individual differences are of
primary concern. This is the first study to focus solely on students with SEND; although
there is a body of research that has examined school and student effects on academic
attainment (e.g. Cervini 2009; Konishi et al. 2010; Sellstro¨m and Bremberg 2006),
this has focused almost exclusively on the school population as a whole. Where strati-
fication has occurred, the focus has been on particular ethnic groups considered to be at
risk for poor outcomes (e.g. Strayhorn 2010). Nonetheless, the extant literature in this
area provides a useful starting point from which to consider the factors that may play a
role in determining the distribution of attainment for learners with SEND.
School effects on academic attainment
The research in this area has primarily made use of hierarchical linear modelling
(HLM), a statistical technique that takes into account the clustered and hierarchical
nature of school-based data sets (that is, students reside within schools, which reside
within Local Authorities (LAs); scores of students attending the same school will be
correlated) and enables an estimation of the proportion of variance in a response vari-
able (e.g. academic attainment) that is attributable to the each level (student, school,
LA). The technique also allows researchers to examine which explanatory variables
at each level explain significant variation in the response variable (Paterson and Gold-
stein 2007; Raudenbush and Bryk 2002a; Raudenbush and Willms 1995).
An overarching theme in the literature on school effects is that school differences
are important in determining students’ attainment (Rutter and Maughan 2002; Sell-
stro¨m and Bremberg 2006). Studies have consistently demonstrated large intra-
cluster correlations (ICC) of around 15–30%, although this figure can vary based
upon the academic subject and phase of education (Cervini 2009; Opdenakker and
Damme 2001; Sellstro¨m and Bremberg 2006). In accounting for these school effects,
researchers have generally sought a range of contextual variables, such as school
International Journal of Inclusive Edu cation 3
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
size, and ethnic composition. In this vein, several studies have identified the roles
played by school factors such as average socioeconomic status (e.g. Battistich et al.
1995; Blau et al. 2001), urbanicity (e.g. Webster and Fisher 2000; Young 1998), size
(Opdenakker and Damme 2007), climate and/or sense of community (e.g. Battistich
et al. 1995; Lee and Bryk 1989) and levels of bullying (Konishi et al. 2010). Interest-
ingly, a couple of studies have examined the impact of the proportion of students with
SEND in mainstream schools on the academic attainment of learners without SEND
(Demeris, Childs, and Jordan 2007; Farrell et al. 2007); both found weak
2
relationships
that were unlikely to be practically significant or meaningful, indicating that the
increased inclusion of learners with SEND in mainstream schools does not adversely
affect the attainment of their non-disabled peers.
Student effects on academic attainment
Individual differences have repeatedly been shown to be the most important predictor of
academic outcomes among students (Veenstra and Kuyper 2004), explaining around
70–85% of the variance in scores (although as above, this figure can vary based upon
the academic subject and phase of education). Differences that appear to be key are stu-
dents’ sex (girls outperform boys, and vice versa, depending upon the subject examined,
e.g. Al-Nhar 1999; McNiece and Jolliffe 1998; Veenstra and Kuyper 2004), ethnicity,
language and/or minority status (learners from minority groups and/or those who
speak the instructional language is not their primary language tend to perform worse,
e.g. McNiece and Jolliffe 1998; Veenstra and Kuyper 2004), parental education (children
of parents with higher levels of education typically perform better, e.g. Al-Nhar 1999;
Opdenakker and Damme 2001) and socio-economic status (children from poorer back-
grounds usually perform worse academically, e.g. Cervini 2009; Farrell et al. 2007;
Young 1998). Psychosocial variables, such as students’ attitudes towards academic sub-
jects and/or schooling in general (with positive attitudes being linked to greater attain-
ment – Webster and Fisher 2000), their locus of control (learners with a greater
internal sense of agency perfom better – Strayhorn 2010) and self-concept (with positive
self-perceptions being associated with increased attainment – Veenstra and Kuyper
2004), also appear to play on their academic attainment. Finally, the relationship
between students and their teachers has been shown to make a contribution (with stronger
relationships leading to better attainmnent – Konishi et al. 2010).
Although there are no multi-level studies that have examined individual differences
in attainment among students with SEND, there are a number of traditional, ‘single
level’ studies that give an indication of the factors that might be important. Type or cat-
egory of need (e.g. social, emotional and behavioural difficulties (BESD) vs. difficulties
in cognition and learning) appears to be important (Anderson, Kutash, and Duchnowski
2001; Department for Education 2010a; Zhang, Katsiyannis, and Kortering 2007), as
does educational placement (e.g. mainstream vs. special) (see Freeman and Alkin
2000, for a review) and level of provision for need (Department for Education
2011). Similar to the general population of students, sex, ethnicity, language and/or
minority status, and socio-economic status have also been found to co-vary with the
attainment of those with SEND (Department for Education 2011). Finally, in terms
of psycho-social variables, levels of social support (Rothman and Cosden 1995) and
the extent of behavioural difficulties (Lane et al. 2008) are related to attainment,
although these effects have only been demonstrated for students with particular
needs rather than the full spectrum of SEND.
4 N. Humphrey et al.
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
The current study
The study reported in this article is the first large-scale multi-level empirical investi-
gation of the factors associated with academic attainment among students with
SEND. Such learners are amongst the most vulnerable to poor outcomes in the
school population as a whole; developing understanding of which factors are important
and the role that they play in determining their academic attainment is a crucial first step
towards developing more effective provision. The knowledge base reported above was
useful in helping to determine the types of school (e.g. urbanicity) and student (e.g. sex)
level factors that might be important in explaining significant proportions of the var-
iance, and also in shaping our analytical framework (e.g. separate analyses by academic
subject and phase of education).
The data set used in our analyses was drawn from the baseline of a national evalu-
ation of a major SEND initiative in England, AfA (Humphrey and Squires 2010, 2011a,
2011b). This initiative was piloted in 454 schools (including mainstream primary and
secondary schools, special schools and pupil referral units) in 10 LAs across the
country. Baseline data collected by the authors in relation to this project were combined
with data held in the national pupil database (NPD) (for individual-level data), from
EDUBASE performance tables and/or provided by LAs (for school-level data). Data
were matched at individual and school levels using unique identification numbers
that are used in pupil and school censuses in England. With access to a wide range
of explanatory variables, we were able to perform a comprehensive series of analyses
to address the predictions outlined in Table 1.
However, it is important to acknowledge some of the inherent challenges and limit-
ations of analyses that make use of the types of data reported herein. Florian et al.
(2004) note several issues that warrant brief discussion. First, in spite of a national
code of practice (Department for Education and Skills 2001), the identification and
assessment of SEND is extremely variable, with various drivers (e.g. funding) influen-
cing decisions regarding need and provision (Humphrey and Squires 2010); further-
more, the concept of SEND is by nature contextual and comparative rather than
being fixed and absolute. Such issues belie the ‘nominal’ approach underpinning quan-
titative data sets (e.g. ‘SEND’ vs. ‘no SEND’). The classification of needs is similarly
problematic – the 12 categories of need recorded in England
3
may oversimplify the
complexity of children’s difficulties (Florian et al. 2004). In relation to academic attain-
ment, the system/scale used to quantify students’ progress (a common points score (PS)
scale that incorporates P-levels, National Curriculum (NC) levels, and latterly, GCSE
grades or equivalent – see (Humphrey and Squires 2011b) also carries limitations; it
may not, for example, accurately capture small-step difference in progress experienced
by students with more severe difficulties (Humphrey and Squires 2011b).
Method
We used a cross-sectional, multi-level, natura l variation design. As noted above, data were
derived from baseline measures taken as part of the national evaluation of the AfA initiative.
Sample
Sampling was purposive and multi-stage in nature. The 10 participating LAs were
selected by the Department for Education to broadly represent the diversity inherent
in LAs across the country. Around 40 –50 schools were then chosen in each LA by
International Journal of Inclusive Edu cation 5
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
Table 1. Predictions for school- and student-level variables in the current study.
Level Measure Description/scoring Justification and prediction
School Urbanicity Urban (0) vs. rural (1) Rural schools are by definition more ‘isolated’ (geographically, economically and socially) than
urban schools, and previous research has demonstrated the negative effect this can have on
attainment (Webster and Fisher 2000; Young 1998). We predict that this effect will be
accentuated among students with SEND, who may be more vulnerable to the effects of this
isolation effect
Size Number of full-time equivalent (FTE) students on roll
(e.g. 1000 ¼ a school with 1000 students)
Opdenakker and Damme (2007) found a positive association between size and attainment, citing
the role of the former in mediating practice characteristics such as teacher co-operation;
however, we predict the reverse pattern for students with SEND, the largest proportion of
whom (those at SA in terms of provision) depend upon their needs being met within normal
school resources, and are thus more liable to ‘slip through the net’ in larger schools
EAL Proportion of students at school speaking EAL
(expressed as %, from 0 to 100)
There has been little research focusing on school-level linguistic factors. However, like SEND
proportions (see below), higher EAL proportions reflect greater diversity in the student
population. It is theoretically plausible that in schools with such diversity, ‘difference’ becomes
the norm and therefore factors that may previously have presented barriers to effective learning
in vulnerable students are more easily accommodated – hence, a positive association with
attainment is predicted
FSM Proportion of students at school eligible for FSM
(expressed as %, from 0 to 100)
Research consistently points to low school-level socio-economic status being associated with
poorer attainment (e.g. Battistich et al. 1995; Blau et al. 2001) – a pattern expected to be
repeated in the current study
SA Proportion of students in school following SEND
provision procedures at the SA stage
(expressed as %, from 0 to 100)
These variables are a proxy for school inclusivity. Previous studies examining associations with
attainment among students without SEND have been inconclusive (e.g. Farrell et al. 2007).
However, it is theoretically plausible that increased inclusivity will lead to increased attainment
of students with SEND, since the former presumably reflects increased capacity within the
school to support students’ needs
SA+/SSEN As above but for SA+ and SSEN combined
Academic
attainment
Proportion of students achieving at least NC Level 4 in
English and Maths (primary) or at least 5 A
∗
–C GCSE
grades including English and Maths (secondary)
(expressed as %, from 0 to 100)
Used as a proxy for overall school effectiveness, school-level attainment is expected to be
positively associated with student-level attainment because higher attaining schools are more
likely to reflect a culture of aspiration and achievement that is considered important for
vulnerable learners, and also more effective pedagogic strategies
Attendance Average proportion of student absence from school
(expressed as %, from 0 to 100)
School-level attendance is expected to be negatively associated with attainment for two reasons –
first, through reduced opportunities for learning when students are absent, and second, as a
possible reflection of a culture of disaffection and disengagement among the student body
Behaviour Aggregated student behaviour scores (0– 3) Taken as proxies for the psychosocial climate of the school, these factors are likely to be associated
with attainment (see, for example, Battistich et al. 1995; Konishi et al. 2010) (aggregated
behaviour and bullying negatively, positive relationships positively)
Positive
relationships
Aggregated student positive relationships scores (0– 3)
Bullying Aggregated student bullying scores (0– 3)
6 N. Humphrey et al.
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
Student Sex Male (0) vs. female (1) Studies have pointed towards a sex by subject interaction effect (Al-Nhar 1999; McNiece and
Jolliffe 1998; Veenstra and Kuyper 2004). Hence, it is predicted that females will perform
significantly better in English than males, with the reverse pattern evident for Maths
Year group Year 1 vs. Year 5 (primary) We take it as axiomatic that attainment increase with age, and hence older students with SEND
will demonstrate significantly higher attainment than their younger counterpartsYear 7 vs. Year 10 (secondary)
Ethnicity White British, Asian, Black, Mixed,
Chinese, Other, Unclassified
Patterns of attainment by ethnic background are expected to follow established trends – e.g.
relatively lower attainment among black students, and relatively higher attainment among
Asian and Chinese students (Department for Education 2012)
Language English, Other, Unclassified EAL may compound the difficulties students with SEND face in accessing the curriculum
(evidence by Department for Education 2011) – hence, it is predicted that this will result in
significantly lower attainment
FSM Not eligible (0) vs. eligible (1) The deeply entrenched relationship between poverty and poorer educational outcomes is expected
to be borne out once more among students with SEND
SEND provision SA, SA+, SSEN, Unknown Students at the latter stages of SEND provision (e.g. SA+, SSEN) typically experience more
complex and/or profound difficulties (hence the need for more intensive provision to support
their needs), and this can impact upon their attainment (Department for Education 2010a,
2011). It is therefore predicted that students at SA+ and with SSEN will be associated with
significantly lower levels of attainment
SEND primary
need
SpLD, MLD, SLD, PMLD, BESD, SLCN,
ASD, VI, HI, MSI, PD, Other, Unclassified
Difficulties associated with cognition and learning, particularly those that are more severe and/or
complex (e.g. SLD, PMLD) will have more profound effects on attainment than those
associated with other aspects of development (e.g. communication and interaction – ASD)
(evidenced by Anderson et al. 2001; Department for Education 2011)
Behaviour WOST behaviour score (0–3) Behaviour problems present a significant barrier to learning and participation (see, for example,
Lane et al. 2008); hence, a negative association with attainment is anticipated
Positive
relationships
WOST positive relationships score (0–3) Relationships with teachers (Konishi et al. 2010) and peers (Rothman and Cosden 1995; Stewart
2008) are crucial facilitators of school adjustment and learning, and hence a positive association
with attainment is predicted
Bullying WOST bullying score (0–3) Like behaviour problems (above) being bullied presents a significant barrier to learning and
participation; hence, a negative association with attainment is expected
Attendance Proportion of days attending school school
(expressed as %, from 0 to 100)
Students with higher attendance rates have increased opportunities for learning and they are more
likely to be engaged with the education process – hence, a positive association with attainment
is predicted
Academic
attainment
PS English and Maths (both 1–65) N/A – response variable
International Journal of Inclusive Edu cation 7
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
senior staff on the basis of them representing the diversity of schools inherent within the
area. Finally, within each school, students with SEND in Years 1 and 5 (primary) and 7
and 10 (secondary) were selected to participate (see below).
Schools
Four hundred and twenty schools drawn from 10 LAs across England were represented in
the analysis,
4
made up of 346 primary and 73 secondary. Given the non-random nature of
the sampling/recruitment procedures for the AfA project, it was important to establish
that these schools broadly represented the broader population of schools in England in
terms of key characteristics. In this vein, Table 2 presents a comparison of our school
sample and all state-funded mainstream schools in England by size, attendance, attain-
ment, free school meal (FSM) eligibility, proportion of students speaking English as
an additional language (EAL) and proportion of students with SEND (SEND).
As can be seen, the school samples were very similar to the population from which
they were drawn in terms of their extant characteristics, with only two differences emer-
ging with an effect size greater than small (using Cohen’s (1992) thresholds); specifi-
cally, these were the somewhat larger overall absence rates and proportion of students
with SEND in our secondary school sample.
Students
Fifteen thousand six hundred and forty students with SEND in Years 1 and 5 in primary
schools (aged 5/6 and 9/10, respectively) and 7 and 10 in secondary schools (aged 11/
Table 2. School sample characteristics and national averages.
School characteristic
Sample mean
(SD)
National
average
Magnitude of
difference (expressed
as Cohen’s d)
Size
a
– number of FTE
students on roll
Primary 276.89 (152.37) 233.4 0.28 small
Secondary 1106.11 (391.78) 977 0.33 small
Attendance
b
– overall
absence (% half days)
Primary 6.01 (4.13) 5.21 0.19 small
Secondary 7.69 (1.35) 6.88 0.6 medium
Attainment
c
– proportion
of students achieving
expected level of
academic attainment
Primary 71.10 (15.56) 73 20.12
Secondary 48.03 (14.95) 53.4 20.36 small
FSM
a
– proportion of
students eligible for
FSMs
Primary 23.39 (15.51) 18.5 0.32 small
Secondary 18.75 (11.00) 15.4 0.3 small
EAL
a
– proportion of
students’ EAL
Primary 18.70 (26.45) 16.0 0.1
Secondary 14.37 (20.20) 11.6 0.14
SEND
a
– proportion of
students with SEND
Primary 22.50 (5.93) 19.9 0.44 small
Secondary 25.08 (6.45) 21.7 0.52 medium
a
Department for Education (2010b).
b
Department for Education (2010c).
c
Department for Education (2010d, 2010e).
8 N. Humphrey et al.
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
12 and 14/15, respectively) drawn from the above schools at the beginning of the 2009/
10 school year were represented in the analysis. As above, the non-random nature of the
sample made it important to establish that these students broadly represented the
broader population of learners with SEND in England in terms of their key character-
istics. In this vein, Table 3 presents a comparison of our student sample and all students
with SEND in England by sex, FSM, EAL, SEND provision and SEND primary need.
As can be seen, characteristics of the student sample are very similar to those of all
students with SEND in England, with some minor exceptions. For example, the
proportion of students eligible for FSM was somewhat larger in the current sample.
Similarly, the proportion of students with MLD was also somewhat larger than in
the national average, while the proportion of those with behavioural, social and
emotional difficulties (BESD) or SLCN was somewhat lower.
5
None of these differ-
ences constitute major deviations from the overall trends in the national data (for
instance, the ranking of proportions for both SEND provision and primary need are
both identical to the national picture), and are most likely the result of random variation.
Sample size computation was based around statistical power considerations for
multi-level modelling (MLM). Our initial calculations indicated that 54 schools
with an average of 37 students per school would be required to detect a small effect
(f
2
¼ 0.03) with 22 explanatory variables. These calculations were completed in two
stages. First, standard power and sample size calculations for a single-level regression
to detect a small effect (f
2
¼ 0.03) using 22 explanatory variables with both Power (0.8)
and alpha (0.05) set to standard levels indicated a minimum of 740 participants would
be needed. This was then corrected for the design effect (e.g. the multi-level nature of
the data) using the formula, N/1 + (estimated average cluster size 2 1) (1 – estimated
ICC) (Twisk 2006), applied in this case as 740/1 + (17 2 1) (1 2 0.20), therefore 54
Table 3. Student sample characteristics and national averages.
Student characteristic Sample (N)
National
average (%)
% difference
(effect size)
Sex – proportion of male students 64% (10,013) 63.5 0.5 (0.001)
FSM – proportion eligible for FSMs 32.5% (5070) 28.0 4.50 (0.16)
EAL – proportion 15.7% (2442) 14.8 0.90 (0.06)
SEND provision –
proportion of students at
each stage of provision
SA 56.3% (8806) 60.0 3.70 (0.06)
SA+ 32.1% (5020) 32.0 0.10 (0.03)
SSEN 7.5% (1179) 8.0 0.50 (0.06)
Unclassified 4.1% (635) – –
SEND primary need –
proportion in each
category of need
SpLD 15.8% (2476) 13.35 2.55 (0.18)
MLD 38.6% (6042) 24.7 13.90 (0.56)
SLD 1.6% (246) 1.2 0.40 (0.33)
PMLD 0.3% (53) 0.3 0.00 (0.00)
BESD 18.1% (2836) 24.4 6.30 (0.26)
SLCN 11.2% (1754) 17.2 6.00 (0.35)
ASD 3.8% (599) 6.6 3.8 (0.42)
VI 0.7% (115) 1.3 0.60 (0.46)
HI 1.3% (200) 2.3 1.00 (0.25)
MSI 0.1% (18) 0.2 0.10 (0.50)
PD 2.3% (357) 3.7 1.40 (0.38)
Other 2.4% (375) 5.1 2.50 (0.53)
Unclassified 3.6% (569) – –
International Journal of Inclusive Edu cation 9
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
schools. Our sample size greatly exceeded this minimum threshold, even when strati-
fied by phase of education (primary/secondary) in the ensuing analyses.
Measures
Table 1 outlines the data collected at the various levels of the study, and provides
associated descriptions, justifications and predictions. A psychometric instrument
was used to measure students’ behaviour, bullying and positive relationships, and is
discussed in more detail below.
Wider Outcomes Survey for Teachers
The Wider Outcomes Survey for Teachers (WOST) (Humphrey and Squires 2010,
2011b) provides indices of students’ behaviour, bullying and positive relationships.
It consists of series of statements (e.g. ‘[student name] is called names or teased by
other children’) to which the respondent indicates a level of agreement on a four-
point rating scale (e.g. never, rarely, sometimes, often). The WOST consists of 20
items, with 7 each for bullying and positive relationships and 6 covering behaviour.
Psychometric analyses conducted during the development and refinement of the
WOST indicate that it meets several of the key criteria set out by Terwee et al.
(2007). Specifically, it has good content validity (exemplified by the clear measurement
aims, target population, concepts, item selection and reduction, and item interpretability
reported by its developers, ibid), strong internal consistency (established through
acceptable fit indicators in confirmatory factor analysis and Cronbach’s alpha co-effi-
cients of .0.9 for each domain), excellent construct validity (demonstrated by analyses
showing that scores are consistent with a range of theoretically derived hypotheses con-
cerning the concepts under scrutiny) acceptable floor (.15% evident in the behaviour
and bullying domains) and no ceiling effects (less than 15% for all domains) and good
interpretability (aided by normative scores) (Humphrey and Squires 2011b). Item
responses are averaged for each domain of the WOST such that overall scores
always range from 0 to 3, with higher scores indicative of higher levels of the measure-
ment domain in question.
Procedure
‘Key Teachers’ – defined as the individual members of staff who knew and understood
the student in question well, had regular contact with them and had an influence on their
provision arrangements
6
(Humphrey and Squires 2011b) completed the WOST online
in early 2010. Assessment of students’ academic attainment in English and Maths was
provided by the relevant subject teachers. These academic data were collated first at
school level, where it was checked and verified by the Special Educational Needs
Co-ordinator or equivalent, before being securely transferred to the relevant LA, and
then on to the National Strategies (representing the Department for Education), who
performed data cleaning duties before handing the final data set over to the research
team via secure channels. The research team then converged this data set with the
WOST data, and school- and student-level socio-demographic data (see Table 3)
drawn from EDUBASE and the NPD. Data matching was made possible via the use
of unique school (‘LAESTAB’ code) and student (‘UPN’ code) identification numbers.
10 N. Humphrey et al.
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
Results
In view of the clustered and hierarchical nature of the data set, MLM (also known as
HLM) using MLWin version 2.16 was the primary analytical tool for the current
study. Data were available on the LAs in which each school resided, and so four
three-level ‘intercept-as-outcomes’ models (Raudenbush and Bryk 2002a, 2002b)
(students nested within schools, nested within LAs) were produced that covered the
two different phases of education (primary and secondary) and academic subjects
(English and Maths). In each case, an initial unconditional (or ‘empty’) model for
the response variable of academic attainment was produced in which no explanatory
variables were included. This enabled approximations of the proportion of unexplained
variance in academic attainment attributable to each level. The unconditional models
demonstrated that differences between LAs did not account for any variance. School
differences were more important, with an average ICC of 19.7 (i.e. just under 20%)
across the four models. Finally, differences between students always accounted for
the largest share of the variance in attainment (average of 80.3%). These findings
were generally consistent with ICCs reported in other studies (e.g. Cervini 2009;
Opdenakker and Damme 2001; Sellstro¨m and Bremberg 2006).
The second stage of the analysis involved adding the various explanatory variables
(see Table 3) to produce conditional (or ‘full’) models. Data fit was assessed by com-
puting and then comparing the 2
∗
log-likelihood values of the empty and full models.
Chi-square analysis of these values confirmed that the inclusion of the explanatory vari-
ables improved model fit in all four models (all p , 0.001). However, comparison of
the level co-efficients across the empty and full models in each analysis indicated a
clear divergence between primary and secondary schools in the amount of unexplained
variance accounted for in the full models. For the primary analyses, the inclusion of the
school and student-level explanatory variables accounted for large proportions of the
unexplained variance identified in the unconditional models (82.94% for English,
87.41% for Maths at the school level; 65.03% for English, 68.60% for Maths at the
student level), but this was not the case in the secondary analyses (12.42% for
English, 13.07% for Maths at the school level; 25.67% for English, 27.52% for
Maths at the student level).
The empty and full models can be seen in Tables 4–7. In the interests of brevity and
clarity, only those explanatory variables that reached statistical significance (p , 0.05)
are displayed. In terms of school-level variables, a finding common to all four models
was the hypothesised ‘inclusivity effect’, where increased proportions of students at
school action plus (SA+) and SSEN (but not those at SA) were significantly associated
with greater student-level attainment. School-level academic attainment emerged as a
significant predictor in all models, but with opposing trends in the primary and second-
ary school analyses, such that it was associated with increases in student-level attain-
ment in the former, but decreases in the latter. In all but one analysis (primary Maths),
the hypothesised inverse relationship between school-level FSM and attainment was
borne out. The proposed influence of school-level aggregated behaviour scores on
the response variable was evident in the primary analyses, but not in the secondary ana-
lyses. Finally, in secondary schools but not primary schools, the proportion of students
speaking EAL was positively associated with attainment, confirming the proposed
‘diversity effect’.
At the student level, where most of the variance in attainment resided, the proposed
age trends, gender by subject interactions (significantly higher attainment in English for
International Journal of Inclusive Education 11
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
Table 4. School- and individual-level variables associated with statistically significant changes in students’ English scores (primary schools).
Empty model:
b
0ijk
¼ 12.540 (0.145) Full model:
b
0ijk
¼ 3.298 (1.175)
Co-efficient
b
SE p Co-efficient
b
SE p
LA level 0.000 0.000 – LA level 0.328 0.170 0.043
0.000% 3.5%
School level 5.890 0.548 ,0.001 School level 0.931 0.123 ,0.001
20.4% 9.8%
% eligible for FSM 20.021 0.008 0.005
% overall school absence 0.035 0.019 0.033
% of students at SA+ or SSEN 0.048 0.016 0.001
Aggregated behaviour 21.543 0.405 ,0.001
Aggregated bullying 0.662 0.370 0.037
% students achieving Level 4 in Eng/Maths 0.016 0.006 0.004
Individual level 22.976 0.368 ,0.001 Individual level 8.219 0.159 ,0.001
79.6% 86.7%
Year group (compared to ‘Year 1’) 8.816 (if ‘Year 5’) 0.098 ,0.001
Gender (compared to ‘male’) 0.234 (if ‘female’) 0.084 0.003
FSM eligibility 20.368 (if ‘yes’) 0.089 ,0.001
Ethnic group (compared to ‘white British’) 0.466 (if ‘Asia’) 0.222 0.018
Language group (compared to ‘English’) 20.868 (if ‘other’) 0.186 ,0.001
SEND provision (compared to ‘SA’) 21.535 (if ‘SA+’) 0.096 ,0.001
23.813 (if ‘SSEN’) 0.183 ,0.001
12 N. Humphrey et al.
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
SEND primary need (compared to ‘MLD’) 22.258 (if ‘SLD’) 0.259 ,0.001
21.582 (if ‘PMLD’) 0.664 0.009
2.440 (if ‘BESD’) 0.133 ,0.001
0.797 (if ‘SLCN’) 0.136 ,0.001
2.564 (if ‘ASD’) 0.219 , 0.001
2.975 (if ‘VI’) 0.523 ,0.001
1.974 (if ‘HI’) 0.371 ,0.001
4.177 (if ‘MSI’) 1.311 ,0.001
2.888 (if ‘PD’) 0.316 ,0.001
1.179 (if ‘other’) 0.294 ,0.001
1.047 (if ‘unclassified’) 0.242 ,0.001
% attendance (09/10) 0.040 0.006 ,0.001
Behaviour 20.179 0.083 0.015
Positive relationships 0.513 0.095 ,0.001
22
∗
log likelihood ¼ 49169.06 22
∗
log likelihood ¼ 28214.93
X
2
(46, 5640) ¼ 20954.13, p , 0.001
International Journal of Inclusive Education 13
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
Table 5. School- and individual-level variables associated with statistically significant changes in students’ Maths scores (primary schools).
Empty model:
b
0ijk
¼ 13.158 (0.145) Full model:
b
0ijk
¼ 3.889 (1.053)
Co-efficient
b
SE p Co-efficient
b
SE p
LA level 0.000 0.000 – LA level 0.158 0.089 0.054
0.000% 1.8%
School level 5.782 0.546 ,0.001 School level 0.634 0.093 , 0.001
18.8% 7.3%
% of students at SA+ or SSEN 0.027 0.014 0.028
Aggregated behaviour 21.290 0.356 ,0.001
Aggregated bullying 0.545 0.326 0.48
% students achieving Level 4 in Eng/Maths 0.014 0.005 0.003
Individual level 24.958 0.398 , 0.001 Individual level 7.891 0.152 ,0.001
81.2% 90.9%
Year group (compared to ‘Year 1’) 9.491 (if ‘Year 5’) 0.095 ,0.001
Gender (compared to ‘male’) 20.813 (if ‘female’) 0.082 ,0.001
Language group (compared to ‘English’) 20.338 (if ‘other’) 0.181 0.031
FSM eligibility 20.258 (if ‘yes’) 0.087 0.001
SEND provision (compared to ‘SA’) 21.229 (if ‘SA+’) 0.093 ,0.001
23.397 (if ‘SSEN’) 0.178 ,0.001
SEND primary need (compared to ‘MLD’) 0.386 (if ‘SpLD’) 0.133 0.002
21.915 (if ‘SLD’) 0.253 ,0.001
21.342 (if ‘PMLD’) 0.620 0.015
2.042 (if ‘BESD’) 0.130 ,0.001
0.891 (if ‘SLCN’) 0.132 ,0.001
1.961 (if ‘ASD’) 0.213 ,0.001
2.197 (if ‘VI’) 0.496 ,0.001
1.619 (if ‘HI’) 0.364 ,0.001
3.289 (if ‘MSI’) 1.282 0.005
2.204 (if ‘PD’) 0.311 , 0.001
1.004 (if ‘other’) 0.287 , 0.001
% attendance (09/10) 0.033 0.006 , 0.001
Positive relationships 0.511 0.093 , 0.001
Bullying 20.263 0.087 0.001
22
∗
log likelihood ¼ 50252.211 22
∗
log likelihood ¼ 28165.491
X
2
(46, 5689) ¼ 22086.72, p , 0.001
14 N. Humphrey et al.
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
Table 6. School- and individual-level variables associated with statistically significant changes in students’ English scores (secondary schools).
Empty model:
b
0ijk
¼ 25.086 (0.453) Full model:
b
0ijk
¼ 21.204 (8.419)
Co-efficient
b
SE p Co-efficient
b
SE p
LA level 0.000 0.000 – LA level 0.000 0.000 –
0.0% 0.0%
School level 14.058 2.460 ,0.001 School level 7.177 1.374 ,0.001
20.4% 15.1%
School size 0.002 0.001 0.024
% FSM 20.143 0.084 0.047
% EAL 0.096 0.035 0.004
% of students at SA+ or SSEN 0.173 0.087 0.026
Individual level 54.939 0.931 ,0.001 Individual level 40.365 0.794 ,0.001
79.6% 84.9%
Year group (compared to ‘Year 7’) 6.435 (if ‘Year 10’) 0.189 ,0.001
Gender (compared to ‘male’) 1.391 (if ‘female’) 0.201 ,0.001
Ethnicity (compared to ‘White British’) 0.816 (if ‘mixed’) 0.487 0.047
Language (compared to ‘English’) 21.461 (if ‘other’) 0.445 ,0.001
FSM eligibility 20.876 (if ‘yes’) 0.208 ,0.001
SEND provision (compared to ‘SA’) 21.855 (if ‘SA+’) 0.215 , 0.001
24.428 (if ‘SSEN’) 0.338 ,0.001
SEND primary need (compared to ‘MLD’) 2.672 (if ‘BESD’) 0.267 ,0.001
3.591 (if ‘ASD’) 0.514 ,0.001
3.354 (if ‘VI’) 1.085 0.001
4.878 (if ‘HI’) 0.757 ,0.001
4.049 (if ‘PD’) 0.564 ,0.001
1.220 (if ‘unclassified’) 0.561 0.015
% attendance (09/10) 0.087 0.010 ,0.001
Behaviour 20.683 0.177 ,0.001
Positive relationships 1.238 0.209 ,0.001
22
∗
log likelihood ¼ 48392.846 22
∗
log likelihood ¼ 34415.537
X
2
(48, 5240) ¼ 13977.309, p , 0.001
International Journal of Inclusive Education 15
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
Table 7. School- and individual-level variables associated with statistically significant changes in students’ Maths scores (secondary schools).
Empty model:
b
0ijk
¼ 24.9836 (0.440) Full model:
b
0ijk
¼ 19.947 (8.188)
Co-efficient
b
SE p Co-efficient
b
SE p
LA level 0.000 0.000 – LA level 0.000 0.000 –
0.0% 0.0%
School level 13.176 2.334 ,0.001 School level 6.550 1.296 ,0.001
19.2% 12.3%
% FSM 20.171 0.081 0.020
% EAL 0.100 0.034 0.002
% of students at SA+ or SSEN 0.169 0.084 0.025
Aggregate bullying 23.795 1.874 0.24
Individual level 55.583 0.945 ,0.001 Individual level 46.830 0.924 ,0.001
80.8% 87.7%
Year group (compared to ‘Year 7’) 3.872 (if ‘Year 10’) 0.205 ,0.001
Gender (compared to ‘male’) 21.235 (if ‘female’) 0.218 ,0.001
Ethnicity (compared to ‘White British’) 21.544 (if ‘Asia’) 0.646 0.008
21.067 (if ‘Black’) 0.572 0.031
3.765 (if ‘Chinese’) 2.039 0.032
22.058 (if ‘other’) 1.036 0.023
23.098 (if ‘unclassified’) 1.044 0.001
16 N. Humphrey et al.
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
Language (compared to ‘English’) 4.234 (if ‘unclassified’) 1.522 0.003
FSM eligibility 20.652 (if ‘yes’) 0.224 0.002
SEND provision (compared to ‘SA’) 21.589 (if ‘SA+’) 0.233 , 0.001
24.659 (if ‘SSEN’ 0.366 ,0.001
SEND primary need (compared to ‘MLD’) 1.210 (if ‘SpLD’) 0.286 ,0.001
3.196 (if ‘BESD’) 0.288 ,0.001
1.062 (if ‘SLCN’) 0.471 0.012
4.515 (if ‘ASD’) 0.561 ,0.001
4.054 (if ‘VI’) 1.168 ,0.001
5.417 (if ‘HI’) 0.813 ,0.001
4.353 (if ‘PD’) 0.612 ,0.001
1.828 (if ‘other’) 0.626 0.002
1.904 (if ‘unclassified’) 0.607 ,0.001
% attendance (09/10) 0.067 0.011 ,0.001
Behaviour 20.675 0.192 ,0.001
Positive relationships 1.535 0.226 ,0.001
22
∗
log likelihood ¼ 48132.417 22
∗
log likelihood ¼ 34917.681
X
2
(48, 5201) ¼ 13214.736, p , 0.001
International Journal of Inclusive Education 17
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
females and Maths for males), and socio-economic status effects were borne out across
all models. In the case of the latter, the ameliorating effects of poverty appeared to
accentuate with age, such that the co-efficient roughly doubled in magnitude from
primary to secondary school. The predicted facilitating influence of positive relation-
ships was evident in all four models and, like the aforementioned socio-economic
status effect, grew in magnitude between primary and secondary schools. Similarly,
the hypothesised association between stage of SEND provision was found in every
analysis, such that students at SA+ and with SSEN were associated with significantly
worse academic outcomes than those at SA. The four analyses also confirmed a priori
expectations regarding the role of primary SEND need, with students experiencing
more complex and/or severe problems in cognition and learning (e.g. SLD, PMLD)
being associated with lower attainment, and those with difficulties in other aspects of
development (e.g. communication and interaction – such as ASD) associated with
higher attainment, relative to the reference group of students with MLD. Attendance
emerged as a significant factor, with students with lower absence rates attaining
better in every model. Finally, in all but one analysis (primary Maths), behaviour pro-
blems were inversely associated with the response variable, as originally predicted.
Discussion
The objectives of the current study were to (i) determine the amount of variation in the
academic attainment of students with SEND that exists between (school effects) and
within (individual effects) schools and (ii) determine which characteristics at school
and individual levels explain significant variation in this attainment. In the case of
the former, our MLM analyses demonstrated an average school effect of just under
20%, with the remaining variance attributable to differences between individual stu-
dents. In the case of the latter, we found that school-level inclusivity, attainment,
FSM eligibility, behaviour (in primary schools) and linguistic diversity (secondary
schools) and student-level age, sex, FSM eligibility, SEND provision, SEND
primary need, attendance, behaviour and positive relationships each contributed to
the distribution of academic attainment.
In relation to the first objective, our research confirmed that the notion of school
differences being important in determining students’ attainment, well established in
the general population of students (e.g. Rutter and Maughan 2002; Sellstro¨m and Brem-
berg 2006), can also be applied to learners with SEND. However, once a school effect
has been established, it is of course important to determine what school differences con-
tribute to this effect. It is of particular interest – given the discourse, debate and
research centring on the effects of school inclusivity on the attainment of students
without SEND (e.g. Demeris, Childs, and Jordan 2007; Farrell et al. 2007) – that the
current study demonstrated that for students with SEND themselves, being part of a
more inclusive school (using the school-level proportion of students with SEND as a
proxy, as in the aforementioned studies) is certainly conducive to their academic
achievement. On a similar note, although only demonstrated in the secondary school
analyses, school-level EAL proportions also appear to be associated with the attainment
of students with SEND. Taken together, an explanation for these findings is that in
schools with higher levels of diversity (e.g. EAL, SEND) – where ‘difference’ is the
norm, and to be expected – staff think about and work with this diversity in qualitat-
ively different ways to those in more homogenous environments (for example, that
differences between students can be used as an educational tool rather than being
18 N. Humphrey et al.
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
thought of as a problem to overcome) (Humphrey et al. 2006). As such, higher levels of
diversity can drive achievement. This notion is given some support by Blau et al.’s
(2001) finding that students situated in more ‘cosmopolitan’ educational environments
made greater gains in their social studies grades.
The facilitative role of positive relationships (evident in all four models) in the aca-
demic attainment of students with SEND is also worthy of discussion. Determining the
exact mechanisms by which positive relationships may contribute to attainment is
beyond the remit of this study and the extant literature offers several explanations
(for example, positive relationships may improve the effectiveness of teaching and
learning (Wentzel 1993), and/or may have an effect on motivation and academic
engagement (Martin and Dowson 2009)). On an applied note though, these findings
reinforce the need for us to consider the factors in mainstream schools that are condu-
cive to the development of positive relationships for students with SEND. The work of
Murray and Pianta (2007) is helpful here – they propose that organisational structures
and resources (for example, an overall ethos that places a high value on relationships),
classroom structures and practice (such as clear rules and consequences) teacher beliefs,
behaviours and actions (for example, high expectations for student achievement and be-
haviour) and individual skills for developing relationships (such as explicit instruction
in self-awareness and self-management skills) can all contribute to the development and
maintenance of positive relationships among students with SEND.
That behaviour problems emerged as a significant barrier to attainment in all but one
analysis is also worthy of note. This study serves as a reminder that even when such
problems are not the primary difficulty experienced by a given student with SEND,
relatively low level behavioural difficulties can still interfere with the learning
process. Where this occurs, it may primarily be a consequence of other extant difficul-
ties that are not being effectively accommodated in the classroom (for example, a child
who is struggling to read may experience frustration that is expressed outwardly in
maladaptive ways), which reinforces the need for more effective early identification,
assessment and the development of appropriate provision. In this sense, primary diffi-
culties in one area need not necessarily give rise to secondary difficulties in another.
Notwithstanding the aforementioned challenges and limitations of using data sets
such as the one reported here (see ‘The current study’), it is important to note
several other limitations of this research. First and foremost, the data and analyses
are cross-sectional in nature. Although this is not unusual in multi-level research (Sell-
stro¨m and Bremberg 2006), it does mean that greater caution has to be applied in
drawing conclusions about the nature of the associations found between our different
explanatory variables and the response variable. Future research should explore
whether these relationships hold true in a temporal/longitudinal design. Second, in
the interests of brevity and parsimony, the analyses presented above did not consider
interactions between variables within and between levels. It would be useful, for
example, to examine whether the school-level inclusivity effect varies for students
with different primary SEND types or at different stages of SEND provision.
Finally, it is also important to note that, given that this article essentially reports on a
secondary analysis (that is, all of the data described herein were gathered for a different
purpose, as noted earlier), there were natural parameters on what variables could be
included, meaning that several key school- and student-level factors (e.g. student
locus of control and/or self-concept – Strayhorn 2010; Veenstra and Kuyper 2004)
could not be included.
7
A natural extension of this point is to consider the fact that
– for the secondary school analyses – there remained large proportions of unexplained
International Journal of Inclusive Education 19
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
variance in our full models, indicating that there are certainly other variables to consider
at both school and individual level in future research.
Conclusion
The current study reported the first large-scale multi-level empirical investigation of the
factors associated with academic attainment among students with SEND. Determinants
at school and individual levels in two core academic subjects (English and Maths) were
assessed using a nationally representative sample. We found that school-level inclusiv-
ity, attainment, FSM eligibility, behaviour (in primary schools) and linguistic diversity
(secondary schools) and student-level age, sex, FSM eligibility, SEND provision,
SEND primary need, attendance, behaviour and positive relationships each contributed
to the distribution of academic attainment. The identification of these socio-demo-
graphic and psycho-social ‘signposts’ of the factors influencing how well these students
with SEND perform in school represents a significant initial step towards the develop-
ment of more effective provision. A crucial natural corollary of this study is research
that builds upon it to determine what school processes and practices make a difference
to these particularly vulnerable learners over time. Data from the aforementioned AfA
study (Humphrey and Squires 2010, 2011a, 2011b) have provided some broad indi-
cations here – including the need for strong leadership for SEND, improved assess-
ment, tracking and intervention, work to strengthen parental engagement and
confidence, and the development of provision to address wider outcomes (such as be-
haviour problems and positive relationships). At the more specific level, meta-analyses
such as that produced by Kavale (2007) have been helpful in identifying the types of
instructional regimes, services and activities that are likely to be effective in meeting
the needs of those with SEND, including increased instruction in core areas (e.g.
reading comprehension, math) developing skills for learning (e.g. memory training),
and psychosocial interventions (e.g. attribution training). However, it is vital that the
implementation of such programming is sensitive to the variety of individual differ-
ences and contextual factors noted above that influence attainment.
Notes
1. The chronosystem is proposed as an overarching system the reflects the fact that an indi-
vidual’s ecology changes over time, such that certain influences may become more or less
important in his/her development.
2. Although weak, the relationship between SEND proportion and achievement in the latter
study was statistically significant.
3. These are specific learning disability (SpLD), moderate learning difficulty (MLD), severe
learning difficulty (SLD), profound and multiple learning difficulties (PMLD), BESD,
speech, language and communication needs (SLCN), hearing impairment (HI), visual
impairment (VI), multi-sensory impairment (MSI), physical difficulty (PD), autism spec-
trum disorder (ASD), and ‘other’. Children identified as having SEND receive provision at
three stages: school action (SA – where needs are met within adaptations to normal school
practice), school action plus (SA+ – where input from external professionals (such as
educational psychologists) is sought, and statement of SEND (SSEN – where a statutory
assessment of need has been conducted and a legal document is drawn up outlining the
support required to meet the student’s needs) (Department for Education 2010a).
4. This figure is lower than the overall number of school participating in the Achievement for
All initiative as (a) data were missing for a small number of schools, and (b) since our
focus was on the attainment of pupils with SEND in mainstream schools, data from
special schools and pupil referral units were excluded from the analysis.
20 N. Humphrey et al.
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
5. It should be noted that the national averages are based on students in receipt of provision at
the SA+ and statement of SEND levels of provision only, which may explain this
discrepancy.
6. In primary schools the Key Teachers were typically the individual students’ class teachers;
in secondary schools, they were typically form tutors.
7. However, we feel that this limitation is outweighed by the benefits of the otherwise fairly
comprehensive list of explanatory variables that we were able to include (see Table 3).
Notes on contributors
Neil Humphrey is Professor of Education at the University of Manchester. He specialises in
social and emotional wellbeing and inclusive education.
Michael Wigelsworth is Lecturer in Psychology of Education at the University of Manchester.
His areas of expertise are statistical analysis (particularly multi-level modelling) and measure-
ment/assessment.
Alexandra Barlow is a Senior Research Associate at the University of Manchester. She has
project managed a num ber of large scale evaluations, including the national Achievement for
All pilot.
Garry Squires is a Lecturer in Educational Psychology at the University of Manchester. His
research interests focus on special educational needs and cognitive behavioural therapy.
References
Al-Nhar, T. 1999. Determinants of Grade 8 achievement in Jordan: A multilevel analysis
approach. Educational Psychology 19, no. 1: 37–52.
Anderson, J.A., K. Kutash, and A.J. Duchnowski. 2001. A comparison of the academic progress
of students with EBD and students with LD. Journal of Emotional and Behavioral
Disorders 9, no. 2: 106–15.
Atkinson, D., M. Nind, S. Rolph, and J. Welshman. 2005. Witnesses to change: Families, learn-
ing difficulties and histo ry. Kidderminster: British Institute of Learning Disabilities.
Australian Institute of Health and Welfare. 2004. Children with disabilities in Australia.
Canberra: Commonwealth of Australia.
Battistich, V., D. Solomon, D. Kim, M. Watson, and E. Scha ps. 1995. Schools as communities,
poverty levels of student populations, and students’ attitudes, motives, and performance: A
multilevel analysis. American Educational Research Journal 32, no. 3: 627–58.
Blau, J.R., V.L. Lamb, E. Stearns, and L. Pellerin. 2001. Cosmopolitan environments and ado-
lescents’ gains in social studies. Sociology of Education 74, no. 2: 121 – 38.
Bronfenbrenner, U. 2005. Making human beings human: Bioecological perspectives on human
development. London: SAGE.
Bronfenbrenner, Urie, and P.A. Morris. 1998. The ecology of developmental processes. In
Handbook of child psychology: Volume 1: Theoretical models of human development, ed.
W. Damon and R.M. Lerner, 5th ed., 993 – 1028. Hoboken, NJ: John Wiley.
Cervini, R.A. 2009. Class, school, municipal, and state effects on mathematics achievement in
Argentina: A multilevel analysis. School Effectiveness and School Improvement 20, no. 3:
319–40.
Cohen, J. 1992. A power primer. Psychological Bulletin 112, no. 1: 155 –9.
Demeris, H., R.A. Childs, and A. Jordan. 2007. The influence of students with special needs
included in grade-3 classrooms on the large-scale achievement scores of students without
special needs. Canadian Journal of Education 30, no. 3: 609 – 27.
Department for Education. 2010a. Children with special educational needs 2010: An analysis.
Nottingham: DFE Publications.
Department for Education. 2010b. Schools, pupils and their characteristics. Nottingham: DFE
Publications.
Department for Education. 2010c. Pupil absence in schools in England, autumn term 2009 and
spring term 2010. Nottingham: DFE Publications.
Department for Education. 2010d. National curriculum assessments at Key Stage 2 in England,
2009/10 (revised). Nottingham: DFE Publications.
International Journal of Inclusive Education 21
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
Department for Education. 2010e. GCSE and equivalent results in England, 2009/10 (revised).
Nottingham: DFE Publications.
Department for Education. 2011. Special educational needs information act: An analysis.
Nottingham: DFE Publications.
Department for Education. 2012. GCSE and equivalent attainment by pupil characteristics in
England (2010/11). Nottingham: DFE Publications.
Department for Education and Skills. 1997. Excellence for all children: Meeting special edu-
cational needs. Nottingham: DfES Publications.
Department for Education and Skills. 2001. Special educational needs: Code of practice.
Nottingham: DFE Publications.
Farrell, P., A. Dyson, F. Polat, G. Hutcheson, and F. Gallannaugh. 2007. The relationship
between inclusion and academic achievement in English mainstream schools. School
Effectiveness and School Improvement 18, no. 3: 335–52.
Florian, L., M. Rouse, K. Black-Hawkins, and S. Jull. 2004. What can national data sets tell us
about inclusion and pupil achievement? British Journal of Special Education 31, no. 3:
115–21.
Freeman, S.F.N., and M.C. Alkin. 2000. Academic and social attainments of children with
mental retardation in general education and special education settings. Remedial and
Special Education 21, no. 1: 3– 26.
Humphrey, N., P. Bartolo, P. Ale, C. Calleja, T. Hofass, V. Janikova, A. Mol Lous, V. Vilkiene,
and G.-M. Wetso. 2006. Understanding and responding to diversity in the classroom: An
international study. European Journal of Teacher Education 29, no. 3: 305–18.
Humphrey, Neil, and G. Squires. 2010. Achievement for All national evaluation: Interim report.
Nottingham: DFE Publications.
Humphrey, N., and G. Squires. 2011a. Achievement for All national evaluation: Second interim
report. Nottingham: DFE Publications.
Humphrey, Neil, and G. Squires. 2011b. Achievement for All national evaluation: Final report.
Nottingham: DFE Publications.
Institute of Education Sciences. 2011. Digest of education statistics 2010. Washington, DC: IES.
Johnson, K. 2010. People with intellectual disabilities: Towards a good life? Bristol: Policy
Press.
Kavale, K.A. 2007. Quantitative research synthesis: Meta-analysis of research on meeting
special educational needs. In The SAGE handbook of special education, ed. L. Florian,
207–21. London: SAGE.
Konishi, C., S. Hymel, B.D. Zumbo, and Z. Li. 2010. Do school bullying and student-teacher
relationships matter for academic achi evement? A multilevel analysis. Canadian Journal
of School Psychology 25, no. 1: 19–39.
Lane, K.L., S.M. Barton-Arwood, J.R. Nelson, and J. Wehby. 2008. Academic performance of
students with emotional and behavioral disorde rs served in a self-contained setting. Journal
of Behavioral Education 17, no. 1: 43–62.
Lee, V.E., and A.S. Bryk. 1989. A multilevel model of the social distribution of high school
achievement. Sociology of Education 62, no. 3: 172–92.
Martin, A.J., and M. Dowson. 2009. Interpersonal relationships, motivation, engagement, and
achievement: Yields for theory, current issues, and educational practice. Review of
Educational Research 79, no. 1: 327 –65.
McNiece, R., and F. Jolliffe. 1998. An investigation into regional differences in educational
performance in the National Child Development Study. Educational Research 40, no. 1:
17–30.
Murray, C., and R.C. Pianta. 2007. The importance of teacher-student relationships for
adolescents with high incidence disabilities. Theory into Practice 46, no. 2: 105–12.
Office for Standards in Education. 2010. The special educational needs and disability review: A
statement is not enough. London: OFSTED.
Opdenakker, M.C., and J. Damme. 2001. Relationships between school composition and
characteristics of school process and their effect on mat hematics achievement. British
Educational Research Journal 27, no. 4: 407–32.
Opdenakker, M.C., and J.V. Damme. 2007. Do school context, student composition and school
leadership affect school practice and outcomes in secondary education? British Educational
Research Journal 33, no. 2: 179 – 206.
22 N. Humphrey et al.
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
Paterson, L., and H. Goldstein. 2007. New statistical methods for analysing social structures: An
introduction to multileve l models. British Educational Research Journal 17, no. 4: 387 – 93.
Raudenbush, S.W., and A.S. Bryk. 2002a. Hierarchical linear models. Thousand Oaks, CA:
SAGE.
Raudenbush, Stephen W., and A.S. Bryk. 2002b. Hierarchical linear models: Applications and
data analysis methods. 2nd ed. Thousand Oaks, CA: SAGE.
Raudenbush, Stephen W., and J.D. Willms. 1995. The estimation of school effects. Journal of
Educational and Behavioral Statistics 20, no. 4: 307 –35.
Robinson, P., and C. Oppenheim. 1998. Social exclusion indicators. London: IPPR.
Robson, C. 2005. Students with disabilities, learning difficulties and disadvantages: Statistics
and indicators. France: OECD.
Rothman, H.R., and M. Cosden. 1995. The relationship between self-perception of a learning
disability and achievement, self-concept and social support. Le arning Disability
Quarterly 18, no. 3: 203–12.
Rutter, M., and B. Maughan. 2002. School effectiveness findings 1979 –2002. Journal of School
Psychology 40, no. 6: 451–75.
Rutter, M., B. Maughan, P. Mortimore, J. Ouston, and A. Smith. 1979. Fifteen thousand hours:
Secondary schools and their effects on chil dren. Cambridge, MA: Harvard University Press.
Sellstro¨m, E., and S. Bremberg. 2006. Is there a ‘school effect’ on pupil outcomes? A review of
multilevel studies. Journal of Epidemiology and Community Health 60, no. 2: 149–55.
Sparkes, J. 1999. Schools, education and social exclusion. London: Centre for Analysis of
Social Exclusion.
Stewart, E.B. 2008. School structural characteristics, student effort, peer associations, and par-
ental involvement: The influence of school- and individual-level factors on academic
achievement. Education and Urban Society 40, no. 2: 179–204.
Strayhorn, T.L. 2010. The role of schools, families, and psychological variables on Math
achievement of black high school students. The High School Journal 93, no. 4: 177 –94.
Terwee, C.B., S.D.M. Bot, M.R. de Boer, D.A.W.M. van der Windt, D.L. Knol, J. Dekker, L.M.
Bouter, and H.C. de Vet. 2007. Quality criteria were proposed for measurement properties of
health status questionnaires. Journal of Clinical Epidemiology 60, no. 1: 34– 42.
Twisk, J. 2006. Applied multi-level analysis. Cambridge: Cambridge University Press.
United Nations Scientific Educational and Cultural Organisation. 1994. The Salamanca state-
ment and framework for action on special needs education. Salamanca, Spain: UNESCO.
United Nations Scientific Educational and Cultural Organisation. 2000. The Dakar Framework
for Action. Education for all: Meeting our collective commitments. Paris: UNESCO.
Veenstra, R., and H. Kuyper. 2004. Effective students and families: The importance of individ-
ual characteristics for achievement in high school. Educational Research and Evaluation 10,
no. 1: 41–70.
Webster, B.J., and D.L. Fisher. 2000. Accounting for variation in science and mathematics
achievement: A multilevel anal ysis of Australian Data Third International Mathematics
and Science Study (TIMSS). School Effectiveness and School Improvement 11, no. 3:
339–60.
Wentzel, K.R. 1993. Does being good make the grade? Social behavior and academic compe-
tence in middle school. Journal of Educational Psychology 85, no. 2: 357 –64.
Young, D.J. 1998. Rural and urban differences in student achievement in Science and Mathematics:
A multilevel analysis. School Effectiveness and School Improvement 9, no. 4: 386–418.
Zhang, D., A. Katsiyannis, and L.J. Kortering. 2007. Performance on exit exams by students
with disabilities: A four-year analysis. Career Development for Exceptional Individuals
30, no. 1: 48 – 57.
International Journal of Inclusive Education 23
Downloaded by [The University of Manchester Library] at 03:21 31 August 2012
