ARE PUPILS IN SPECIAL EDUCATION TOO “SPECIAL” FOR REGULAR EDUCATION? YSBRAND J. PIJL and SIP J. PIJL
Abstract – In the Netherlands special needs pupils are often referred to separate schools for the Educable Mentally Retarded (EMR) or the Learning Disabled (LD). There is an ongoing debate on how to reduce the growing numbers of special education placements. One of the main issues in this debate concerns the size of the difference in cognitive abilities between pupils in regular education and those eligible for LD or EMR education. In this study meta-analysis techniques were used to synthesize the findings from 31 studies on differences between pupils in regular primary education and those in special education in the Netherlands. Studies were grouped into three categories according to the type of measurements used: achievement, general intelligence and neuropsychological tests. It was found that pupils in regular education and those in special education differ in achievement and general intelligence. Pupils in schools for the educable mentally retarded in particular perform at a much lower level than is common in regular Dutch primary education. Zusammenfassung – In den Niederlanden werden Schüler mit besonderen Bedürfnissen oft an separate Schulen für erzieherisch mental Zurückgebliebene (EMR) oder Lernbehinderte (LD) verwiesen. Es gibt eine Diskussion über die Reduzierung der wachsenden Anzahl an Sonderschuleinrichtungen. Ein Hauptthema des Artikels behandelt die Unterschiede in den kognitiven Fähigkeiten zwischen Schülern in regulären Bildungseinrichtungen und Schülern in Sonderschuleinrichtungen. In dieser Studie werden meta-analytische Techniken angewandt, um Ergebnisse von 31 Studien zusammenzufassen über Unterschiede zwischen Schülern in regulärer Primarausbildung und Schülern in Sonderschulen in den Niederlanden. Die Studien wurden nach Bewertungsarten in drei Kategorien unterteilt: Leistung, allgemeine Intelligenz und neuropsychologische Tests. Man fand heraus, dass Schüler im regulären Schulwesen und Schüler im Sonderschulwesen in Leistung und allgemeiner Intelligenz Unterschiede aufweisen. Besonders Schüler in Schulen für geistig Zurückgebliebene arbeiten auf viel geringerem Leistungsniveau als gemeinhin in holländischer Primarausbildung üblich. Résumé – Aux Pays-Bas, les élèves présentant des besoins particuliers sont souvent adressés aux écoles distinctes pour retardés mentaux en éducation (EMR) ou pour apprenants handicapés (LD). Un débat est actuellement en cours sur le moyen de réduire le nombre croissant des placements en éducation spécialisée. L’un des axes de ce débat porte sur l’importance des différences des capacités cognitives entre élèves de l’éducation régulière et ceux admissibles à l’éducation LD ou EMR. La présente recherche a utilisé des techniques de méta-analyse pour procéder à la synthèse des résultats de 31 études menées aux Pays-Bas sur les disparités entre élèves de l’enseignement régulier et ceux de l’éducation spécialisée. Les études ont été réparties en trois catégories selon le type d’évaluation utilisé: performances, intelligence globale et tests neuropsychologiques. L’analyse a constaté que les élèves de l’éducation régulière et ceux de l’éducation spécialisée présentent des résultats divergents International Review of Education – Internationale Zeitschrift für Erziehungswissenschaft – Revue Internationale de l’Education 44(1): 5–20, 1998. 1998 Kluwer Academic Publishers. Printed in the Netherlands.
6 quant aux performances et à l’intelligence globale. En particulier les élèves des écoles pour retardés mentaux en éducation produisent des résultats beaucoup plus faibles que le niveau courant dans l’enseignement primaire régulier des Pays-Bas. Resumen – En los Países Bajos, los alumnos con necesidades especiales frecuentemente se remiten a escuelas separadas para retardados mentales educables o alumnos con dificultad de aprendizaje. Se está desarrollando un debate de cómo reducir el creciente número de plazas de educación especial. En este debate, uno de los problemas principales se refiere al grado de diferencia que existe en cuanto a capacidades cognoscitivas entre los alumnos que asisten a una enseñanza regular y aquellos destinados a la enseñanza especial. En este estudio se han usado técnicas de meta-análisis a efectos de sintetizar los resultados de 31 estudios sobre las diferencias que existen en los Países Bajos entre alumnos en enseñanza primaria regular y aquellos de enseñanza especial. Los estudios se agruparon en tres categorías según el tipo de parámetro usado, con tests en cuanto a nivel de conocimientos, inteligencia general y neuropsicológicos. Se ha comprobado que los alumnos en enseñanza regular y aquellos en enseñanza especial difieren en cuanto a nivel de conocimientos e inteligencia general. En particular, los alumnos que asisten a escuelas de retardados mentales educables se desempeñan a un nivel muy inferior al que es usual en la enseñanza primaria regular holandesa.
The education system in the Netherlands consists of regular schools and special schools. Compared with many other European countries, the Dutch special education system is extensive, differentiated, and segregated. Since
7 the sixties, Dutch special education has developed into a wide-ranging system for students who cannot keep up in regular schools. In 1972, the total population enrolled in the 14 special education school types was 2.2 per cent of all pupils between 4 and 11 years of age (Dutch primary school age). In 1995, this percentage had increased to 4.3 per cent (Pijl and Pijl 1995; Pijl 1997). This growth was primarily caused by the expansion of two of the system’s major school types: schools for children with learning disabilities (LD) – in the Netherlands referred to as “LOM” schools – and schools for educable mentally retarded (EMR) children – in the Netherlands referred to as “MLK” schools. The “LD” and “EMR” labels cover a wide range of pupils and pupil needs. These labels are not defined very clearly, either by law, by teachers, or by experts in psychological and educational assessment who are involved in special education placements. Hence, circular characterisations of these groups (“pupils are learning disabled or mildly mentally retarded because they are eligible for placement in schools for children with LD or EMR”) are rather common. For several decades, this highly differentiated, extensive and separated system of special education was considered a sign of great concern for pupils with special learning needs. However, more recently this point of view has become the subject of much debate. Various reviewers have pointed out that special education placement does not diminish the problems and academic difficulties of the pupils referred. Placement in a separate special school merely functions as a “safety valve” for regular schools (Tomlinson 1982). This safety valve can be considered an improper way of relieving regular education of pupils who are difficult to handle or “time-consuming” (Pijl 1989). Parents, it is said, want their special needs child to attend a regular school because they like to send their child to the same school as their other children, to a neighbourhood school, and to educate their child with other non-special needs children. They want their child to receive as normal a schooling as possible. A growing group of policy-makers, educators and parents became convinced that segregation in education had gone too far. The high costs and sideeffects of a segregated system, such as labelling or a shattered school career, were said to be unjustified. The advocates of integration believe that integration is possible by referring to examples in other systems (e.g. Denmark, Sweden, England and the United States) (Meijer, Pijl and Hegarty 1994). In 1990, the central government issued a document “Together to school again” which intended to make a fresh start in integrating special needs pupils (Dutch Ministry of Education and Sciences 1990). As a result, regular and special schools started working together, special needs coordinators were appointed in every regular school, teacher training programmes were launched, new legislation was passed, and new regulations for funding regular and special schools were proposed. All these measures are intended to act as a push towards integration. Although there is consensus about the necessity of halting the growth
8 of special education attendance, uncertainty still exists as to how to stop this trend and to what extent it should be reversed into a less segregated structure. It is this last step in particular – maintaining pupils now labelled LD or EMR within regular primary schools – that is the subject of much discussion. Substantial numbers of both regular and special education teachers as well as parents of pupils now in special education are sceptical about integration. They do not reject the push for more integration in principle, but believe that pupils with learning difficulties and/or mental retardation are better off in segregated special schools with their highly differentiated, individually focused teaching and counselling. In accordance with this view of the specialness of special education they consider LD or EMR pupils to have profound and special problems which make a regular school placement inappropriate. They regard these pupils as “different” from those in regular education: after all, why would they have been referred in the first place? Those who argue for integration regard cognitive differences as less relevant. For them, the issue is more a civil rights argument: segregation should be avoided and teachers will just have to learn to accommodate pupils with special needs. They also claim to have great difficulty distinguishing pupils labelled LD or EMR from slow learners in regular primary schools (Meijer 1988). In their view, in spite of referral, there is a considerable overlap between the two groups. Surprisingly, both advocates and opponents make use of the same research findings. Both refer to studies in which LD and/or EMR pupils are compared with those in regular education on a number of variables, e.g. achievement, IQ, concentration and behaviour. Both use these data to argue either for implementing inclusive education or for taking great care and restraint in changing the status quo. Clearly, integration all these research findings could greatly contribute to ending this debate.
The research question The confusion about the outcomes of some 20 years of research on the differences between groups in regular and special education is understandable. Differing variables, different sets of instruments and all kinds of samples have been used, the results of the studies differ from each other, and the outcomes have been presented and interpreted in almost every possible way. A meta-analysis of the existing studies on the characteristics of the pupils might yield a conclusive answer as to the similarities and differences between the three groups. The answer does not mean that integrating some or all of these groups is, or is not, desirable or possible. It just states the size of the potential differences between the groups and thereby ends the confusion on the topic.
9 Method Data collection In order to locate appropriate studies, computer literature searches were conducted, relevant journals scanned and the annual review of standardized tests and test research (Evers, van Vliert-Mulder and ter Laak 1992) studied. In this way 81 documents were acquired for further analysis. The final selection of documents and the studies these documents represented were based on the following criteria: – each study must contain a comparison of the cognitive abilities of either regular and LD pupils, or regular and EMR pupils, and/or LD and EMR pupils; – the study findings had to be quantifiable using the statistic g (= the standardized mean difference between two samples); – the samples of regular, LD or EMR pupils compared within each study should have the same age distribution; – within each pool of redundant or overlapping documents and studies, the best documented study was chosen. Several documents reported on more than one study, mostly because different age groups were distinguished. A total of 33 studies which met the criteria were described in 13 articles, research reports and dissertations. Five studies compared regular with LD pupils, three compared regular with EMR pupils and nine compared LD with EMR pupils. The remaining 16 studies contained comparisons of all three groups (see Table 1). Bias in finding studies, known as publication bias or the file drawer problem, was unlikely for two reasons. First, none of the studies found and selected was aimed specifically at testing mean differences between regular (denoted by “R” in Table 1), LD and EMR pupils. Even if such studies did exist, it would be virtually impossible to get data causing the null hypothesis to be accepted. All the studies found were aimed at estimating differences rather than testing them; establishing base-rates at estimating differences rather than testing them; establishing base-rates for experimental curricula; standardization of test scores for special populations or some other goal which resulted in the comparison of regular, LD and EMR groups as a by-product. Thus, the chance of a relevant study being published is just remotely linked to the outcome of significance tests (if any). Secondly, the number of Dutch researchers in the special education field is limited. Large sample studies would be known throughout the field, though an occasional failure to detect a relevant but unpublished study can never be ruled out.
1990 1985 1986 1988 1988 1990
1990
1989
1989
1989 1990 1988 1988
Bon, W.H.J. van Bos, K.P. van den
Bos, K.P. van den
Brugman, G.
Koppel, J.M.H. van de
Schonewille, B.
Wolters, M.A.
Vinjé, M.J.
Resing, W.C.M.
Doef, M.P. van der Dekker, R. Snijders, J.Th. Vergeer, M.M.
and recognition) and recognition) and recognition)
and recognition) and recognition) and recognition)
WISC-R full scale IQ WISC-R full scale IQ SON-R non-verbal IQ test Reading and writing tasks
Word recognition Reading (comprehension Reading (comprehension Reading (comprehension Word recognition Reading (comprehension Reading (comprehension Reading (comprehension RAKIT full scale IQ
Metacognition task
Reading, math achievement
CPM-Raven, tangrams, concentration, visual analogies, CEFT non-verbal IQ WISC-R full scale IQ
Memory
Reading, spelling, arithmetic, CPM-Raven Memory
Year of Dependent variables Publication Description
Study authorship
05 01 01 02 02 05 05 01 01 02 02 02 02 02 01 01 01 01 02 02 02 01 02 02 02 01 01 01 01 01 01 01 15
Number
Table 1. Studies of differences in cognitive abilities between regular, LD- and EMR-pupils.
Achievement Non-verbal abilities Non-verbal abilities Non-verbal abilities Non-verbal abilities Non-verbal abilities Non-verbal abilities General IQ General IQ Achievement Achievement Achievement Achievement Achievement Non-verbal abilities Non-verbal abilities Non-verbal abilities Achievement Achievement Achievement Achievement Achievement Achievement Achievement Achievement General IQ General IQ General IQ General IQ General IQ General IQ Non-verbal abilities Achievement
Domain
07.5–9.0 07.5–9.1 10.9–12.5 07.7–9.1 10.5–11.9 08.5–9.5 10.7–11.7 09.0–10.0 10.0–11.0 07.0–8.0 08.0–9.0 09.0–10.0 10.0–11.0 11.0–12.0 09.4–9.4 10.7–10.7 11.5–11.5 08.7–9.7 09.7–10.7 10.7–11.7 11.7–12.7 08.7–9.7 09.7–10.7 10.7–11.7 11.7–12.7 07.1–8.1 08.1–9.1 09.1–10.1 10.1–11.1 07.0–12.0 07.8–12.8 06.5–11.5 10.6–13.6
Ages of pupils
2,594 2,661 2,761 2,632 2,422 2,374 2,528 2,458 0,200 0,200 0,200 0,200 0,190 0,190 0,874
-.040 -.086 -.086 -.030 -.030 -.020 -.020 -.190 -.190
Regular
130 015 483
014 044 028 022 033 007 010 010 202 268 349 322 164 237 265 274 045 050 051 050
040 086 086 030 030 020 020
LD
Sample sizes
046 055 018 033 032 042 037 007 010 010 146 181 172 189 127 176 182 141 048 051 053 053 022 069 011 251
086 086
EMR
0.86 0.60
1.69 1.35 1.28 1.42 1.75 1.51 1.27 1.29 1.31 1.24 1.25 1.07
1.07 0.35 0.35 0.23 0.60 1.03 1.27
R vs LD
2.46 2.56 2.33 2.22 2.30 2.29 2.28 1.90 2.64 2.30 2.52 2.51 1.66 2.04 1.66
2.30 2.00
0.64 0.64
R vs EMR
1.43 1.06 0.54
0.01 0.48 0.40 1.06 1.80 0.28 0.52 0.39 0.77 0.69 0.68 0.73 0.53 0.37 0.63 0.55 1.42 1.32 1.64 2.03
0.30 0.30
LD vs EMR
Standardized mean difference g
10
11 Coding study findings and characteristics Study findings may prove to be quite heterogeneous. Therefore, it is important in a meta-analysis to anticipate study characteristics which could account for variance in the findings. In this study the following study characteristics were coded (see Table 1): – – – –
the year in which the data were collected; sample sizes; the age of the sampled students; the measured cognitive abilities (general IQ, verbal IQ, performance IQ, general achievement, reading comprehension, word recognition, reading rate, spelling, mathematical achievement, concentration, memory and visual perception); – the instruments used to measure each cognitive ability; – for each instrument: its reliability; – for each instrument: the use of raw versus standardized scores. This list is for the most part fairly common to all meta-analyses. The special interest in the ages in the samples stems from the fact that age correlates with both cognitive abilities and the probability of being labelled as eligible for special education. More variance in ages causes more variance in cognitive abilities, thereby obscuring mean differences between R, LD and EMR groups unless standardized test scores have been used. Study findings were coded separately for each measure used within a single study. Coding was simple and straightforward, since conceptual problems in coding treatments often encountered in meta-analysis do not apply in this case, since each group is defined by attending either a regular primary school, an LD or an EMR school. Constructs measured and instruments used were also easily coded as most researchers used well-known instruments which are common in educational assessment. Findings were quantified by using the standardized mean difference between two samples to estimate the effect size in the study: g = (M1 – M2)/Sp where Sp is the pooled sample standard deviation. Since most studies report means and standard deviations, calculating g was straightforward. For one study g had to be derived from F-tests, using: Sp = (MSw)1/2 = (MSb/F)1/2 (Glass, McGaw and Smith 1981). While some studies only report on one measure of cognitive ability, such as full-scale WISC-R IQ, others make use of sub-tests or multiple measures of the same or very similar constructs. In order to avoid overestimating the reliability of the outcomes of the meta-analysis by using interdependent data stemming from multiple measurements of the same sample, “studies” must be the unit of analysis. Therefore g was calculated for a linear combination of (sub)tests reported in each of the studies rather than for each separate (sub)test. When means and standard deviations of a linear combination were not provided in the study’s report, but results were given of separate sub-tests or multiple measures of a construct, the g’s within a study were averaged. The
12 only way in which this solution to the problem of multiple measures may cause difficulties is when the findings vary with respect to the instruments used in the studies. Analysis As Hedges has pointed out (Hedges 1981; Hedges and Olkin 1985), the statistic g is a biased estimator of the mean standardized difference in a theoretical population of studies (δ). To estimate δ from a single study Hedges proposes
(
)
3 d = 1 – –––––––– g, 4N – 9
with variance
d2 n1 + n2 s2(d) = –––––––– + –––––––––– . n 1n 2 2(n1 + n2)
When a series of k independent studies share a common effect size δ, δ can be estimated by the weighted combination of all k d’s: d+ (Hedges and Olkin 1985, ch. 6, expression 6). The weights are inversely proportional to the variances in each study in order to give large sample studies more weight than smaller ones. Hedges even provides us with the simple statistic Q (Hedges and Olkin 1985, chapter 5, expression 25) to test the hypothesis of a common effect size. Q has a chi-square distribution with k – 1 degrees of freedom. When the hypothesis of a common effect size is falsified, Q can be split into tests of a common effect size in each of p groups of studies and a test of differences between the group-common effect sizes, very much like the partitioning of the total sum of squares in the analysis of variance.
Results First of all the attempt was made to estimate Hedges’ weighted average effect size (d+) across all studies for each comparison of regular (R), learning disabled (LD) and educable mentally retarded (EMR) pupils. As can be seen from Table 2, large differences between R and LD groups (d+ = 1.3 standard deviations) and especially between R and EMR groups (d+ = 2.2) were found. As in all meta-analyses one should beware of interpreting mean effect sizes without having a closer look at the assumption that all studies share a common effect size. The Q-tests of homogeneity proved to be highly significant (p < 0.00001) for all three comparisons of the R, LD and EMR groups. So we must reject the hypothesis that differences between studies in study findings are due to sampling variance. From graphic representations of the data it became clear that the broad concept of “cognitive abilities” had to be abandoned. Studies were divided into three groups according to the class of cognitive abilities measured:
13 Table 2. Average effect size by comparison of R, LD and EMR pupils.
R vs. LD R vs. EMR LD vs. EMR
Mean effect size (d+)
Standard deviation of d+
Number of studies
Total number of pupils
Q-Test of homogeneity
p
1.3 2.2 0.7
0.022 0.027 0.029
21 19 25
25,340 24,930 05,456
193.4 263.5 143.5
0.00 0.00 0.00
– Achievement (reading, mathematics); – IQ (general/full-scale IQ tests); – Non-verbal tests (concentration, memory, visual perception). The aforementioned Q-tests for each comparison were split into tests of the homogeneity of effect sizes within each group of studies and between groups of studies. The division of studies into achievement, IQ and non-verbal subgroups was successful in reducing the within-group variance of effect sizes and raising between-group variance (Table 3). As expected, all three tests of between-group differences in effect size are highly significant. However, the pooled within-grouped tests of homogeneity Table 3. Q-tests of homogeneity of effect size by class of cognitive abilities and by comparison of R, LD and EMR pupils. N R vs LD Between groups Within groups Within group:
R vs EMR Between groups Within groups Within group:
LD vs EMR Between groups Within groups Within group:
(pooled) Achievement IQ Non-verbal
09 05 07
(pooled) Achievement IQ Non-verbal
08 08 03
(pooled) Achivement IQ Non-verbal
14 05 06
Q
df
p
129.0 064.4
02 18
0.00 0.00
047.2 007.0 010.2
08 04 06
0.00 0.14 0.12
202.9 060.6
02 16
0.00 0.00
033.3 017.4 009.9
07 07 02
0.00 0.02 0.01
101.2 042.3
02 22
0.00 0.01
035.5 005.9 002.9
13 04 05
0.00 0.21 0.71
14 are also significant for each comparison of R, LD and EMR groups, due to the significant variance of effect sizes within the group of achievement studies. This result signifies systematic differences between the findings in the studies which measure achievement. The fact that variance in effect size with respect to achievement proved significant, does not mean that the remaining variance between achievement studies is relevant in a practical perspective for a number of reasons. First, the variance in effect size may be significant, but is not very large when inspecting ranges and unweighted standard deviations of effect sizes in the achievement group (see Table 4). Secondly, the Q-statistic is extremely sensitive to within-group variance when the findings of large sample studies are integrated. As can be deduced from the counts in Table 4 of the number of studies and of the total number of students in these studies, the achievement group contains several studies with sample sizes well over 2000 students. These large samples also explain the low standard deviation of the mean achievement study effect size. A third reason to consider the achievement group variance to be irrelevant is that the large sample studies in this group stem from the same research program and have many study characteristics in common. No remaining study characteristics were found covarying with effect size. After grouping studies into the aforementioned non-overlapping groups of studies using either achievement, IQ for non-verbal measures, we shall now return to the main purpose of this meta-analysis: the quantitative summary of educational research on differences between regular, LD and EMR pupils in primary education. The largest differences between regular and LD pupils and between regular and EMR pupils (see Table 4) are revealed by the group of Table 4. Average effect size by class of cognitive abilities and comparison of R, LD and EMR pupils. Mean effect size (d+)
Standard deviation of d+
Standard deviation of d’s
Range effect sizes d
Number of studies
Total number of pupils
R vs LD Achievement IQ Non-verbal
1.4 1.1 0.5
0.02 0.07 0.08
0.22 0.18 0.38
1.1–1.8 0.9–1.3 0.2–1.3
09 05 07
22,591 01,316 01,433
R vs EMR Achievement IQ Non-verbal
2.3 2.2 0.8
0.03 0.07 0.10
0.19 0.33 0.60
1.9–2.6 1.7–2.6 0.6–1.7
08 08 03
21,744 01,957 01,229
LD vs EMR Achievement IQ Non-verbal
0.6 1.5 0.3
0.03 0.09 0.10
0.40 0.28 0.29
0.0–0.18 14 1.3–2.0 05 0.2–0.5 06
4,432 0,600 0,424
15 achievement studies (1.4 and 2.3 standard deviations respectively). Assuming normal distributions, these effect sizes show that about 92 per cent of LD pupils and about 99 per cent of EMR pupils achieve less than the pupils in regular education with average performance. In other words: the average LD and EMR pupil would have a percentile ranking of 8 and 1 respectively on a theoretical standardized achievement scale. The IQ group reveals slightly smaller differences between regular and special education pupils but differentiates the LD and EMR groups much more clearly (d+ = 1.5) than the achievement studies do (d+ = 0.6). Studies using non-verbal instruments in the neuropsychological/modality skills domain do a relatively poor job in distinguishing regular, LD and EMR pupils from each other.
Discussion This meta-analysis has shown that regular pupils differ mainly in achievement and to a lesser extent also in IQ tests from pupils in special schools. IQ tests differentiate mainly between LD and EMR pupils. This interaction between the class of cognitive ability and the pattern of differences between R, LD and EMR pupils reflects very much the way in which pupils experiencing difficulties in regular education are referred and placed in special schools in the Netherlands. The initiative for a referral is often taken by a primary school child’s teacher, in close cooperation with parents and headteacher. If they decide to refer, the child is extensively assessed by a multidisciplinary team from a special education school or a group of special education schools working together. Almost always, placement in a special school is the result (Meijer 1988; Pijl 1989). IQ tests are the focal point of the assessment and result in the final placement decision. Thus, the first phase in decision-making (the decision to refer) is solely based on achievement, the second phase (the placement decision by a LD or EMR school or a group of special schools) depends very much on the child’s performance on IQ tests. IQ correlates strongly with achievement measures, which explains why regular education pupils differ from special education pupils in IQ as well. This does not mean that IQ is causally related with the distinction between regular education and special education pupils. IQ tests can only have a small effect on the chance that a pupil will stay in the same or another regular primary school, because such an outcome of the assessment procedure seldom occurs. The large effect of IQ tests on the choice between a LD or EMR school is both widely accepted and empirically corroborated (Pijl 1989). Of course, other explanations of the pattern of differences between R, LD and EMR pupils are possible. A popular hypothesis by the advocates of integration is that the slow pace of instruction in special education and the special education pupils’ reference to other low-achieving pupils cause pupils in special education to achieve much less than they would have done in regular schools. This hypothesis is not as plausible as it may seem. The slower pace
16 in special education is meant to maximize the effect of instruction by adapting it to the skills of its pupils. For teachers in regular education one effect of the relatively high pace in regular education in particular, the debilitating failure anxiety, is very often an important reason for referral (Pijl 1997). Negative frog-pond effects are more likely to occur when the difference between low achievers and other pupils are large (Guldemond 1994), as is often the case in regular education for special needs pupils before referral and placement. Still, as long as experimental data about the effects of special education are lacking, comparing pupils in special education with pupils in regular education means that pupil characteristics will be confounded with school characteristics. In the meantime, we would like to go along with critics of special education who argue from empirical though partly circumstantial, evidence that special education does not raise the cognitive abilities of its pupils any more than regular education would have done. We would not go as far as assuming that special education lowers cognitive abilities. This study supports this view since the mean age of the samples, which of course correlates strongly with the length of stay in special education, does not covary with effect size. From an integration perspective, the decision to refer is the more important one since much depends on the willingness of the regular school teacher to adapt to pupils with problems. In the view of the role of achievement and IQ in referral and placement decisions mentioned a few paragraphs earlier, the group of achievement studies provides us with the most relevant data. The results of our study show that in the Netherlands a very strict differentiation is made between regular and LD pupils, and between regular and EMR pupils concerning achievement (effect sizes of 1.4 and 2.3). Comparable data from other countries seem to be scarce. In the USA Kavale and Nye (1985) found in their meta-analysis a mean effect size of 0.68 for differences in the achievement domain between regular and LD students. The greatest differentiation Kavale and Nye found was in the linguistic domain (0.88), which contrasts with an effect size of 1.4 in the Netherlands. It may well be that these differences in effect size are partly explained by differences in LD attendance figures (in the Netherlands 1.4 and in the USA 4.5 per cent of the resident population aged 6–17 (Pijl and Pijl 1995; U.S. Dept. of Education 1990), but the differences in effect sizes most likely reflect differences in the criteria for learning disabilities. A practice of non-segregation would mean that pupils now labelled LD would be educated in regular primary schools. Roughly 75 per cent of the LD pupils would perform at the same level as other low performers now in regular classrooms, while about 25 per cent would perform at levels now quite unusual in regular education. The majority of EMR pupils perform at exceptionally low levels compared with regular education pupils. It should be noted that this meta-analysis does not suggest that regular school and special school pupils only differ in cognitive abilities. This is almost certainly not the case. It is highly likely that at least a number of
17 referrals to schools for learning disabilities are based on behaviourial problems and personality disorders of the pupils involved. There is as yet insufficient knowledge on the nature and magnitude of those problems. The conclusions neither imply that the majority of pupils now referred to schools for the learning disabled can be placed in regular school settings while those now referred to schools for the educable mentally retarded cannot. Decisions on integration and on how to integrate are much more complex than that. Unlike in other systems, such as Norway, England and the United States, the integration debate in the Netherlands has not been initiated by parents (Pijl, Meijer and Hegarty 1996). In general, parents value the current special education system and do not complain about the disadvantages of the segregated system. It is not surprising then that the values discussion plays a minor role in the integration debate. Parents and teachers endorse the strive towards integration in principle, but feel it can only be implemented if an adequate education for each pupil can be guaranteed. Adequate education for special needs pupils in a regular education setting is considered very difficult because of the presumed large differences in performance between the pupils. It would be too much to handle for a regular teacher in a regular classroom. This meta-analysis clarifies the persistent discussion on the magnitude of some of the differences between regular and special education pupils. Advocates and opponents of integration are both partly put in the right. The advocates of integration in the Netherlands rightly state that the majority of students now in separate schools for the learning disabled perform at a level comparable to low-performing pupils in regular education. The opponents are right when they point to the large gap in performances between lowperforming pupils in regular education and pupils now in schools for educable mentally retarded pupils. However, the debate on integration encompasses more than pupil differences. Both groups view integration as a technical operation which will succeed only if pupil performances do not diverge too much. They tend to take the states quo in regular and special education as fixed and end up in debates about pupil groups “equal” enough to be integrated in the current regular system. It has been argued, though, that integration should not be confined to placement questions or to providing access to pre-set norms of learning and behaviour; it is about fitting schools to meet the needs of all their pupils (Hegarty 1991). In this wider connotation, integration (or inclusion) stands for an education system that includes a large diversity of pupils and provides differentiated education to serve this diversity. Inclusion has to do with changing schools, organizations, curricula, teacher training, legislation, etc. (Pijl, Meijer and Hegarty 1996). The real challenge is beyond puzzling about differences, it is about developing inclusive schools and ensuring that the percentage of pupils in segregated settings goes down. In that perspective the aim
18 of this meta-analysis has been modest: to end the debate on the “real” pupil differences, thereby giving way to discussions on how to further inclusion.
References Bon, W. H. J. van, Derx, M. H. G. and Klerkx, J. J. M. P. 1990. De morfologische competentie van kinderen met leerproblemen. Pedagogische Studiën 67: 59–69. Bos, K. P. van den and Kingma, J. 1985. Het aanvankelijk leren en het leren tot criterium bij jonge en oudere MLK-, LOM- en GLO-school-kinderen in een memoriseertaak. Tijdschrift voor Orthopedagogiek 24: 239–253. Bos, K. P. van den, Kingma, J. and Bakker, A. C. 1986. De invloed van leeftijd bij het memoriseren van normaal lerende en leergestoorde kinderen. Tijdschrift voor Orthopedagogiek 25: 172–181. Brugman, G. and Koster, B. 1988. Het oplossen van visueel-ruimtelijke taken door kinderen op LOM-scholen en kinderen op basisscholen. Tijdschrift voor Onderwijswetenschappen 18: 189–207. Dekker, R., Mackaay-Cramer, E. M. and Bruin, E. E. J. de 1990. De Nederlandstalige WISC-R bij kinderen in het speciaal onderwijs. Kind en Adolescent 11: 170–179. Doef, M. P. van der, Kwint, J. M. and Koppel, J. M. H. van de 1989. Wat leren moeilijk lerende kinderen van de WISC-R. Kind en Adolescent 10: 136–141. Evers, A., Vliet-Mulder, J. C. van and Laak, J. ter 1992. Documentatie van tests en testresearch in Nederland. Assen/Maastricht: van Gorcum. Glass, G. V., McGaw, B. and Smith, M. L. 1981. Meta-analysis in Social Research. Beverly Hills: Sage. Guldemond, H. 1994. Van de kikker en de vijver. Leuven: Garant. Hedges, L. 1981. Distribution Theory for Glass’s Estimator or Effect Size and Related Estimators. Journal of Educational Statistics 6: 107–128. Hedges, L. and Olkin, I. 1985. Statistical Methods for Meta-analysis. New York: Academic Press. Hegarty, S. 1991. Towards an Agenda for Research in Special Education. European Journal of Special Needs Education 6, 87–99. Kavale, K. A. and Nye, C. 1985. Parameters of Learning Disabilities in Achievement, Linguistic, Neuropsychological, and Social/Behavioral Domains. The Journal of Special Education 19(4): 443–458. Koppel, J. M. H. van de, Nuyten, J. P. M. and Kaanders, M. L. J. H. J. 1988. De nieuwe WISC-R en het moeilijk lerende kind. Enkele feiten. Kind en adolescent 9, 177–186. Meijer, C. J. W. 1988. Verwijzing gewogen. Groningen: RION. Meijer, C. J. W., Pijl, S. J. and Hegarty, S. 1994. New Perspectives in Special Education. London: Routledge.
19 Ministry of Education and Sciences. 1990. Weer samen naar school. Zoetermeer: Ministerie van Onderwijs & Wetenschappen. Pijl, S. J., Meijer, C. J. W. and Hegarty, S. 1996. Inclusive Education: A Global Aganda. London: Routledge. Pijl, Y. J. 1989. Toelatingsonderzoek in het LOM- en MLK-onderwijs. Dissertation. Groningen: RION. Pijl, Y. J. and Pijl, S. J. 1995. Omwikkelingen in de deelname aan het (voortgezet) speciaal onderwijs. Pedagogische Studiën 72(2): 102–113. Pijl, Y. J. 1997. Stabiliteit deelname aan LOM en MLK onafhankelijk van WSNSbeleid. Tijdschrift voor Orthopedagogiek 36: 131–139. Resing, W. C. M. and Bleichrodt, N. 1989. Intelligentiemeting bij kinderen uit het speciaal onderwijs: de betekenis van de RAKIT voor LOM- en MLK-leerlingen. Nederlands Tijdschrift voor Psychologie 44: 382–389. Schonewille, B., Roth, D. and Ley, A. van der. 1990. Handelingsalternatieven van leerkrachten in het speciaal onderwijs. Amsterdam: SCO. Snijders, J. Th., Tellegen, P. J. and Laros, J. A. 1988. Handleiding SON-R, 5.5–17. Groningen: Wolters-Noordhof. Tomlinson, S. 1982. A Society of Special Education Routledge and Kegan Paul. U.S. Department of Education 1990. Annual report to Congress on the implementation of the Education for All Handicapped Children Act. Washington DC: Author. Vergeer, M. M. and Oostdam, R. J. 1988. Voorstudie PPON in het SO. Deel III: Prestatie van LOM- en MLK-leerlingen op functionerende taaltaken. Vinjé, M. P. 1989. Ontwikkelingen in de leesvaardigheid van leerlingen in het basisonderwijs. Resltaten van het landelijk onderzoek leesvaardigheid in 1986 en 1988. Arnhem: CITO. Wolters, M. A. 1990. Leerproblemen en metacognitieve ontwikkeling. In: M. K. van der Heijden and J. Rispens (red.), Zorgverbreding en afgrenzing (131–141). Amsterdam/Lisse: Swets & Zeitlinger.
The authors Sip Jan Pijl is senior researcher at GION, the Groningen Institute for Educational Research, University of Groningen, the Netherlands. He is involved in studies on the integration of students with special needs into regular education and has conducted international comparative research on integration. He works part-time for the European Agency for Development in Special Needs Education. Contact address: Dr Sip J. Pijl, GION, University of Groningen, Westerhaven 15, PO Box 1286, 9701 BG Groningen, The Netherlands. Ysbrand Johan Pijl is senior researcher at GION, the Groningen Institute for Educational Research, University of Gronningen, the Netherlands. He has
20 published on a variety of topics in educational research. These publications concern for a large part questions about assessment, selection and decisionmaking in both special and regular education. Articles on artifacts in educational research show his interest in statistical methods and research design. Contact address: Dr Ysbrand J. Pijl, GION, University of Groningen, Westerhaven 15, PO Box 1286, 9701 BG Groningen, The Netherlands.