Early View
Open Access

Addressing social disparities in special education placement in three welfare states: Student demographic correlates of the share of students identified with special educational needs at the school level using TALIS data

Monica Reichenberg

Corresponding Author

Monica Reichenberg

Department of Education and Special Education, University of Gothenburg, Gothenburg, Sweden


Monica Reichenberg, Department of Education and Special Education, University of Gothenburg, 405 30 Gothenburg, Sweden.

Email: [email protected]

Search for more papers by this author
Girma Berhanu

Girma Berhanu

Department of Education and Special Education, University of Gothenburg, Gothenburg, Sweden

Search for more papers by this author
First published: 22 May 2024


The number of students with special educational needs (SEN) is growing rapidly. This study compared the correlations between the share of students identified with SEN and student diversity (socioeconomic status and ethnicity) at the school level in three countries. We used the principal questionnaire from the 2018 Teaching and Learning International Survey (TALIS) to examine data from principals in three welfare states (the United Kingdom, France, and Sweden) and whether minority students in these three countries also receive special education. We conducted an ordinal regression analysis to examine the data. First, our results suggest that the share of immigrants in schools does not reliably predict the share of students placed in SEN. Second, the schools' share of refugees predicts the share of students placed in SEN, although the results vary by educational stage and country. Third, the schools' share of socioeconomically disadvantaged students predicts the share of students with SEN in all countries. We conclude that our study both agrees and disagrees with overrepresentation theory and equity theory. Finally, we suggest that welfare state theory may explain these differences.

Key Points

  • A recurrent issue in research on special educational needs (SEN) placement is whether over- or under-representation exists in different social categories, such as ethnic or socioeconomically disadvantaged students.
  • Few studies have taken a welfare perspective when investigating SEN placement. Thus, this study adopts a welfare perspective.
  • This study uses school-level data from the principal questionnaire (2018) of the Teaching and Learning International Survey (TALIS) to examine principals' survey responses in three welfare states (the United Kingdom, France, and Sweden) and whether minority students in these three welfare states receive special education.
  • We conclude that our study both agrees and disagrees with overrepresentation theory and equity theory. Finally, we suggest that welfare state theory may explain these differences.


The number of students identified for and placed in special education is increasing rapidly (Berhanu, 2010). However, what consequences can special educational needs (SEN) placement have and who are those identified for it?

Most researchers agree that SEN placement has two consequences, with one being positive (intended) and the other negative (unintended) (Giddens, 1984). The intended consequence is that providing special education can be considered a response tailored to the experiences of students with special needs (Berhanu & Dyson, 2012). Special education placement allows students to receive support in, for example, reading, vocabulary development and mathematics, to compensate for social background; hence, a positive consequence. Thus, one might consider access to SEN programmes as a right and entitlement to pedagogical expertise. However, SEN placement may also have unintended negative consequences, such as stigmatisation (or discrediting) and labelling (i.e. identified as deviant). Stigmatisation and labelling may in turn result in additional negative consequences, such as lowering students' self-esteem and teachers' expectations of them (Artiles, 2019; Artiles & Trent, 1994; Bywaters et al., 2003; Foubert et al., 2014), resulting in the so-called ‘self-fulfilling prophecies’ that hamper academic attainment (Merton, 1948). Therefore, we suggest that providing students with special needs services implies a delicate problem of matching students' pedagogical needs with appropriate instruction and assessment.

Research has repeatedly demonstrated that minority students are often misidentified and placed in special education when they do not actually have special needs. For example, in some countries, learners from particular social backgrounds, such as those who are from ethnic minorities groups or socioeconomically disadvantaged, run the risk of being erroneously labelled as having special needs and thus placed in special education (Artiles, 2022; Berhanu, 2010; Blanchett, 2006). The identification of SEN among immigrant students is often based on subjective high-incidence special needs, such as specific learning disabilities, speech and language disorders, cognitive impairments and emotional disabilities (Sullivan & Bal, 2013). Hosp and Reschly (2003) noted that cultural differences between teachers and students may result in students being unnecessarily identified as having SEN (Hosp & Reschly, 2003). Teachers who are unfamiliar with their students' racial or ethnic backgrounds may be at risk of misidentifying them as having SEN. Assessment methods are another cause for concern. Minority students are typically tested in their second rather than dominant language, and the notifications immigrant parents receive from schools regarding their child's potential SEN are not in their dominant language (Artiles & Trent, 1994; Monsrud, Rydland, Geva, Thurmann-Moe, & Lyster, 2022; Monsrud, Rydland, Geva, & Lyster, 2022 Subasi Singh, 2020).12

However, despite decades of research, researchers disagree on whether minority students are over- or under-represented in special education or not (Morgan et al., 2015, 2017). This disagreement typically depends on (a) how overrepresentation is measured (Artiles et al., 2005), (b) the level of analysis (e.g. country, region, city, school, or individual), (c) the method of analysis, and (d) the type of overrepresentation (e.g. socioeconomic or ethnic). This issue is discussed in more depth later in this article.

Our study first contributes to the literature by emphasising the school level.3 We consider overrepresentation aggregated at the school level (a ‘social fact’) as opposed to the individual level, ‘misplacement’ (i.e. the ‘social action’). Second, we address two types of overrepresentation: socioeconomic and ethnic (immigrant and refugee). Third, we compare overrepresentation in the three types of welfare states. Comparing these three welfare states allows us to analyse placement from a welfare perspective.

In summary, first, we suggest that socioeconomically disadvantaged students are overrepresented across welfare states. Second, we suggest that the overrepresentation of refugees (but not necessarily immigrants) depends on the type of welfare state. Thus, our study investigates the school level as opposed to the individual level. Therefore, we do not make any claims based on principal self-reports as to whether individual students differ in their probability of SEN (Piantadosi et al., 1988; Robinson, 2009).

Purpose and research questions

This study aims to compare the correlations between the share of students identified for SEN placement and student diversity (socioeconomic and ethnic) at the school level in three countries (France, Sweden and the United Kingdom [UK]) using data from the Teaching and Learning International Survey (TALIS).

Our research questions (RQs) are as follows:
  1. How does the share of refugee students correlate with the schools´ share of students identified for SEN placement in schools in the three countries?
  2. How does the share of immigrant students correlate with the share of students identified for SEN placement in schools in the three countries?
  3. How does the share of socioeconomically disadvantaged students correlate with the share of students identified for SEN placement in schools in the three countries?

Before proceeding, we would like to clarify how we may answer our RQs to reach our conclusions. To the best of our knowledge, TALIS provides the best comparative dataset for answering our RQs. However, even the best comparative data cannot provide a firm answer, as we can only examine the estimated share of students (based on principal self-reports) and not the actual share of students (‘head count’). Nevertheless, we contend our RQs are urgently needed for policy and theoretically intriguing for educational research. A lack of ideal data should not hinder us from addressing urgent and intriguing RQs; thus, we cautiously examine what the best data available show and invite readers to critically make their own judgements.

Research on disproportionality: Overrepresentation vs underrepresentation

As mentioned above, researchers disagree on the phenomenon of overrepresentation and have reported conflicting findings. Several reasons can help explain these discrepancies.

First, the results regarding over- or under-representation depend on the level researchers examine (i.e. country, region, city, school or individual). Artiles et al. (2005) noted that most American studies on over- and under-representation have been conducted at the country level; thus, the results agree that Black students are overrepresented in terms of intellectual disability and behavioural disturbances. Artiles et al. (2005) and Farkas et al. (2020) have argued that not dividing the analysis into subgroups is problematic; thus, Artiles et al. (2005) conducted a study at the district level in California and found that Hispanic students were overrepresented.

Second, the results depend on the specific minority group. Some groups are overrepresented, whereas others are underrepresented. For example, Cooc (2017) found that Asian-American and Pacific Islander students are underrepresented in special education (Cooc, 2017).

Third, the results depend on how overrepresentation is measured; focusing on between-group differences may hide within-group differences (Strand & Lindsay, 2009). For example, results between groups may hide age differences within minority groups (Strand & Lindsay, 2009).

Fourth, Morgan et al. (2017) found very little evidence that Black children's overrepresentation in special education can be explained by misidentification based on race or ethnicity, particularly after controlling for individual-level confounders. However, these findings have been questioned on theoretical and methodological grounds. Skiba et al. (2016) argued that Morgan et al. (2015) oversimplified disproportionality. Specifically, they disregarded that ethnicity is context-dependent. Thus, researchers must address ethnic disproportionality by interpreting the results within the local context of the study (Skiba et al., 2016). For example, ethnicity in Europe has a different historical legacy than ethnicity in the US. Therefore, in Europe, ethnicity concerns immigrants and refugees to a greater extent.

Level of analysis and measurement

The statistical analysis of disproportionality indicates that the magnitude and direction of regression coefficients depend on the data. Thus, regression analysis results depend on the level (nation, state, city and district), social background (disability category, race, language, and gender), geographic location (urban, suburban and rural), and measurement.

Research on overrepresentation has favoured school-level data over student-level data (Cooc & Kiru, 2018), with most studies using descriptive statistics. Surprisingly, researchers seldom conduct regression analyses and depend only on descriptive statistics. Five concerns have been raised regarding this descriptive research.

First, the research may suffer from spurious associations owing to omitted variable bias. Therefore, the reported direction and strength (magnitude) of the associations reported may be misleading.

Second, associations may shift when individual data are aggregated (Kievit et al., 2013). Similarly, inferring associations across levels may lead to errors. However, in this study, we focus on overrepresentation rather than on misidentification.

Third, descriptive emphasis has led to inattention to the mechanisms (i.e. process explanations) of overrepresentation (Cooc & Kiru, 2018). Thus, the choice of datasets, variables and statistical analyses influences RQs and conclusions.

Fourth, researchers have noted the importance of measuring disproportionality correctly and caution against oversimplifying the problem, particularly because some studies have focused on binary questions such as ‘is there overrepresentation’ or ‘is special education racist’ (Cavendish et al., 2020; Cooc & Kiru, 2018; Cruz & Rodl, 2018; Tefera et al., 2023).

Fifth, many minority students are incorrectly identified as having SEN. For example, studies on self-identified disability suggest that citizens in social-democratic welfare states are more likely to identify as disabled than citizens in other countries (Black et al., 2017). This greater probability of identifying oneself as disabled may be explained by access to entitlements once a citizen has been identified as having a disability. Social-democratic welfare states offer more generous entitlements to people with disabilities (Black et al., 2017; Esping-Andersen, 1990), leading to an increase in teachers, principals, and parents demanding diagnoses for students (overrepresentation) (Giota & Emanuelsson, 2016).


The theory of welfare states refers to various social policies found in most developed democratic countries. All welfare states are characterised by state-led support for people with disabilities. Thus, people with disabilities are offered appropriate opportunities in life and means to participate in work and education. The welfare state modifies the impact of the market, by providing some minimum guarantee (mitigating poverty), covering a range of social risks (security), and providing certain services (e.g. healthcare or child and elder care) at the best standards available. Welfare states create social norms regarding good education. Consequently, welfare educational arrangements are offered to make it possible for all students, regardless of their disabilities, to achieve curriculum goals. However, if students want to use these facilities, they must first be labelled as having a disability or problem before the school can take steps to help them function academically (Foubert et al., 2014).

Welfare states differ with respect to the level of ambition and the mix between these aspects. Esping-Andersen's (1990) typology distinguishes between three basic models of welfare states. The social democratic model has high levels of gross domestic product (GDP) spending on social services (e.g. schools, hospitals) and social insurance (e.g. unemployment and sickness insurance) to ensure access for citizens of all social classes. The social-democratic regime has the smallest cluster. Welfare provision is characterised by universal and comparatively generous benefits, a commitment to full employment and income protection, and a strongly interventionist state used to promote equality through a redistributive social security system. In the liberal model, the state provides minimal welfare, benefits are modest and often attract strict entitlement criteria, and recipients are usually means-tested and stigmatised. The conservative welfare state regime is distinguished by its ‘status differentiating’ welfare programmes in which benefits are often earnings-related, administered through one's employer, and geared towards maintaining existing social patterns. The roles of the family and church are also emphasised, and the redistributive impact is minimal. Group solidarity and moral integration are crucial in establishing and maintaining social inclusion and cohesion. The welfare state is a product of conflicts of interest, and more specifically, of class struggles and class alliances (Esping-Andersen, 1990). In this study, we use Esping-Andersen (1990) categorisation of welfare states.

In the educational context, France, the UK and Sweden represent three different types of welfare states, and as such, they have a history of close relations between education and welfare-building. Three questions need to be answered. Who governs the educational system? How differentiated is the educational system? (van de Werfhorst & Mijs, 2010). When do educational policy interventions begin (Esping-Andersen et al., 2012)?

Education is regarded as an integral part of the welfare state and different welfare state regimes are expected to frame and influence the general direction of educational politics in different ways. Esping-Andersen et al. (2012) noted the importance of childcare and early education policies that may not only raise average achievement levels but may also be of special benefit for less-advantaged children, particularly if high-quality programmes are provided.

The Swedish social welfare/educational system was shaped by a long period of social democracy. Under the Swedish school system, all citizens are entitled to equal access to free healthcare, special education for those in need of such support, career counselling, school transport, free hot meals during the school day, teaching materials and free education (Berhanu, 2010). Such entitlements offer citizens equal chances and the hope that they can succeed. This early educational policy has helped diminish the effects of different social, cultural and economic backgrounds on academic outcomes (Beach & Dyson, 2016; Berhanu, 2010).

The UK did not develop its welfare state education platform in the same direction as Sweden (Strand & Lindsay, 2009). Special education is required for those needing such support and free healthcare and careers guidance are provided. However, free school meals are means tested. During the 1990s, the UK began applying market principles to its educational system (Beach & Dyson, 2016). Students choose their schools; thus, schools compete for students. The importance of private schools has increased, owing to state funding. Poorly managed schools with low pupil achievement are shut down. These market reforms are said to have benefited disadvantaged neighbourhoods (Beach & Dyson, 2016).

Curiously, in the 1990s, Sweden began to imitate several market reforms in the school system: free school choice (vouchers) leading to school competition and state-funded private schools combined with the decentralisation of school governance (Giota & Emanuelsson, 2016).4

The French Central Government holds decisive steering authority at all levels of education. The state is expected to provide the same educational opportunities to all, regardless of socioeconomic standing, through centralised steering and control. Furthermore, various system features promote educational equality and strong performance, such as the wide availability of preschool, small average class sizes, and high public expenditure. Public French schools (écoles publiques) are free of charge and secular. Conversely, private schools are almost always fee-paying, although the scale varies greatly. Meals are not free in French schools. Not all teaching materials are free.

England, France and Sweden have extensive urban areas where a significant number of people experience poverty and a range of associated problems. These factors limit both educational attainment and wider opportunities for progression (Lupton et al., 2013). All three countries also have a tradition of receiving immigrants. The school systems of the three countries are described in Table 1 (PISA) (Eurydice) (OECD).

TABLE 1. School systems and three welfare states: France, England and Sweden.
Indicators France England Sweden
Welfare state Conservative Liberal Social-democratic
Curriculum governance and assessment Centralised Decentralised Decentralised
Curriculum differentiation Late Late Late
Educational policy interventions Late Late Early and throughout the life course

Curriculum differentiation is late in Sweden, France and the UK: 15 years of age in France and 16 years of age in the UK and Sweden (Table 1).

In summary, welfare states and educational systems share several similarities. However, the type of educational system constitutes the main alternative explanation for country-specific differences. Although educational systems and welfare states co-evolve, educational systems have their own unique histories that contribute to their peculiarities. We also acknowledge the peculiarities of the three educational systems studied. Therefore, instead of ruling out the contribution of educational systems, we contend that welfare states and educational systems share sufficient similarities to provide a coherent explanation for country-level differences. In particular, we underscore the importance of policies for early intervention (e.g. preschool) and intervention throughout the life course (e.g. adult education). We elaborate on this in the Discussion.

Welfare states and socioeconomic disadvantages

All countries have inequalities in education caused by socioeconomic disadvantages; however, the level of this inequality varies. One theory posits that social-democratic welfare states reduce inequality and improve opportunities for students from socioeconomically disadvantaged homes. For example, social-democratic welfare states provide policies such as free bus rides, free school meals, free textbooks, no tuition fees and mandatory school libraries to all citizens regardless of their socioeconomic position. Thus, the association between parents' socioeconomic disadvantages and children´s educational attainment tends to be weaker in social-democratic welfare states. Proponents of welfare state theory would suggest that the identification of SEN early in life may be a social right to mitigate socioeconomic disadvantages later. As schools with a large proportion of students of immigrant background often correlate with socioeconomic disadvantages (including economic poverty), one might expect that much of the variation in overrepresentation depends on socioeconomic disadvantages rather than ethnic overrepresentation itself. Thus, children are identified as having SEN because they live in socioeconomically disadvantaged families or neighbourhoods, and the share of the socioeconomically disadvantaged (refugees) purges the explanatory variance in the proportion of students with migrant background in SEN.

However, critics have contended that major educational changes such as school privatisation and decentralisation of schools have eroded social-democratic welfare states such as Sweden (Beach & Dyson, 2016). Thus, educational opportunities may vary greatly owing to municipal differences (i.e. local social policies). Critics of welfare state theory suggest that assignment to SEN programmes serves as a desperate resort for schools to cope with privatisation, decentralisation, refugee crises or austerity (Giota & Emanuelsson, 2016).

Hypotheses based on the theoretical arguments in the literature

Based on previous research and theories we propose five hypotheses.
  1. Overrepresentation hypothesis 1 (H1) posits that refugee or immigrant students are overrepresented in SEN programmes (Artiles, 2022).
  2. Underrepresentation hypothesis (H2) posits that refugee or immigrant students are underrepresented in SEN programmes (Cooc, 2017).
  3. The equity hypotheses (HO) can be meaningfully interpreted as support for no overrepresentation or underrepresentation of refugees or immigrant students in SEN programmes (Morgan et al., 2017).
  4. The socioeconomic disadvantage hypotheses (H3) suggests that SEN overrepresentation correlates with the share of socioeconomically disadvantaged students. Thus, differences in ethnicity (refugees or immigrants) disappear when adjusting for the share of socioeconomically disadvantaged students (Berhanu, 2010).
  5. The welfare hypothesis (H4) implies that the pattern depends on the welfare state and educational system. This hypothesis is based on Esping-Andersen (1990) and further developed by Berhanu (2010).


Data sources

We used the principal questionnaire from the 2018 TALIS to examine (1) principals in three welfare states, the UK, France and Sweden, and (2) whether minority students in these three welfare states receive special education (OECD, 2019) for International Standard Classification of Education (ISCED) Levels 1 and 2.

The reasons for choosing TALIS data are as follows: (a) TALIS uses a complex randomised sample, (b) TALIS allows us to compare countries, (c) TALIS aggregates responses of migration background and index of social and cultural status at the school level, and (d) TALIS reportedly offers the best available dataset at the school level. Although TALIS has self-reported variables to a higher degree, it is also more complete than PISA measures, as principals describe the entire school composition. Since TALIS measures reflect school leaders' perceptions, they are arguably more likely to be related to principals' practices (OECD, 2019).

Focal variables

The principals in TALIS 2018 were asked to estimate the broad percentage (none, 1%–10%, 11%–30%, 31%–60%, and more than 60%) of certain types of students in their schools who had the characteristics of students with special needs (outcome) (for more information, see the TALIS principal Questionnaire). Sharp-eyed readers may notice a linguistic discrepancy between special needs and SEN. Although TALIS does not use the exact corresponding term, we can nevertheless be confident that the principals interpret special needs as SEN. We suspect that any discrepancy should be attributed to the discrepancy between the actual share of students with SEN and not the principal-reported share of students with SEN. Using self-reports offers a more sensitive definition of what constitutes SEN, which in the best of cases can measure how schools define SEN.

The focal predictors include students from socioeconomically disadvantaged homes, students who are immigrants or have a migrant background, and students who are refugees. For simplicity, we standardised all the above predictors by transforming them into z-scores (subtracting the mean and dividing by the standard deviation).

Control variables

To simplify the interpretation, all control variables were either standardised or dummy coded. First, we included control variables for the principals´ characteristics: gender (dummy coded), age category (dummy coded), education (standardised), and years of experience (standardised).

Second, we include control variables for school characteristics: location (i.e. city, town or village, dummy coded), SEN shortage, shortage of teachers with competence in teaching students with special needs, number of special educators (standardised), student ratio indicating the student-to-teacher ratio (standardised), and school culture, for which principals were asked to estimate on a four-item response scale whether their school had a collaborative culture (standardised).

Data analysis strategies

We conducted an ordinal regression analysis. Unlike linear regression, ordinal regression treats the outcome as ordinal, that is, ‘bounded’ between 1 and 3 and unequal scale steps. Boundedness induces over-/under-prediction in linear regression (i.e. beyond the 1–3 scale). The unequal-scale steps violate the assumption of linearity in linear regression. Thus, ordinal regression provides more efficient estimates than multinomial regression.

We report the unweighted estimates. Our sensitivity checks suggest that adding weights does not change the conclusions. However, the weights resulted in convergence issues for some models, owing to the instability when adding several predictors.

For simplicity, we focus on three countries and avoid conducting a multilevel analysis of all countries surveyed for TALIS. All analyses were conducted in R using the following packages: MASS (Ripley et al., 2013; Wickham & Grolemund, 2016), stargazer (Hlavac, 2022), dplyr (Wickham et al., 2019; Wickham & Grolemund, 2016), and ggplot2 (Wickham, 2011).


In this section, we report the regression results for the three welfare states. Thus, we report the results of one regression per country. After reporting the regression results, we then report the predicted probabilities of our theoretically motivated predictors.

Table 2 shows the logit coefficients (i.e. logged odds). Logit coefficients cannot be compared across models because the residual variance is fixed (to approximately 1.6–1.8). In principle, we can ‘y-standardise’ the coefficient by dividing by 1.6. However, we would still be unable to compare the magnitudes of the coefficients because of the fixed variance. Therefore, we can compare the signs and significance levels of the coefficients.

TABLE 2. Ordinal regression. Dependent variable.
Dependent variable:
Special educational needs
z_socioec 0.874*** 0.787*** 1.212*** 1.099*** 0.964*** 0.478**
(0.266) (0.232) (0.275) (0.256) (0.262) (0.208)
z_immigrants 0.210 0.171 −0.297 −0.397* −0.046 −0.073
(0.267) (0.238) (0.265) (0.221) (0.273) (0.225)
z_refugees 0.559** −0.015 0.490* 0.378* 0.649*** 0.273
(0.222) (0.193) (0.261) (0.228) (0.211) (0.195)
z_lackspecteachers 0.038 −0.189 0.031 −0.150 0.334* 0.009
(0.176) (0.183) (0.203) (0.204) (0.182) (0.168)
urbanTown (3001 to 100,000 people) −0.633 −0.560 0.781 −1.133 0.052 0.019
(0.482) (0.538) (0.507) (0.782) (0.438) (0.473)
urbanCity (more than 100,000 people) −0.423 −0.544 0.822 −1.227 −0.125 0.085
(0.585) (0.614) (0.628) (0.839) (0.653) (0.616)
z_studentratio −0.255 −0.366* −0.057 −0.330 −0.102 −0.340*
(0.207) (0.206) (0.225) (0.226) (0.195) (0.205)
z_schoolculture 0.132 −0.252 0.094 0.467** 0.212 0.126
(0.181) (0.168) (0.190) (0.195) (0.172) (0.152)
z_yearsprincipaltot 0.530** −0.149 −0.391* 0.049 −0.161 0.107
(0.243) (0.224) (0.214) (0.268) (0.197) (0.189)
z_ISCED 0.189 −0.331* −0.123 0.239 0.140 0.145
(0.188) (0.175) (0.209) (0.201) (0.191) (0.154)
agegroup 40–49 0.315 0.357 0.667 −0.011 −0.353 −0.216
(0.668) (0.708) (0.835) (0.708) (0.518) (0.784)
agegroup 50–59 −0.268 0.677 1.082 −0.041 −0.414 −0.337
(0.704) (0.731) (0.892) (0.728) (0.544) (0.799)
agegroup 60 and above −0.829 1.538* 2.888** 1.167 −0.288 0.174
(0.877) (0.847) (1.251) (1.225) (0.817) (0.872)
Observations 150 154 152 137 169 177
  • Note: *p < 0.1; **p < 0.05; ***p < 0.01.
  • Abbreviations: ISCED, International Standard Classification of Education; SEN, Special educational needs; Z, standardised.

Notably, a lack of statistical significance does not mean that an association does not exist. Instead, we simply cannot reject the possibility that the pattern was due to chance alone (under the null hypothesis). The variability of the estimates is simply too high (i.e. uncertainty).

We observed a positive association between socioeconomic advantage and SEN at the school level in all three countries. All coefficients for socioeconomic disadvantage were statistically significant, suggesting that the pattern was unlikely due to chance alone. Thus, we rejected the null hypothesis (H0: equity). Instead, we favour the socioeconomic hypothesis (H3: ses). Again, we resist the temptation to compare magnitudes across countries and educational levels.

For the school's share of immigrants, only one coefficient is statistically significant (in the UK). Thus, we failed to reject the null hypothesis (H0: equity).

Considering the exception found for the UK, the coefficient which appears statistically significant has a considerably large standard error. We also fit several models that might make our models look ‘too good’ owing to multiple comparison issues. Thus, we caution against interpreting the coefficient as evidence against H0: equity.

Thus, regarding immigration, we found no support for H1: overrepresentation. However, we should not accept H0: equity; we only fail to reject it.

Notably, some coefficients of schools´ share of immigrants are negative. However, none of these coefficients are statistically significant owing to large standard errors.

Thus far, we fail to reject H0: equity, as we note that four of the six coefficients are positive and statistically significant for schools' share of refugees. However, the standard error is somewhat large in the UK relative to the estimates of the coefficients. Socioeconomic disadvantages may matter more than the share of refugees for SEN assignment in the UK. However, this does not rule out the possibility of H1: over representation in the UK, as significance does not portray importance. We have only tempered our expectations.

For Sweden and France, we observed positive statistically significant coefficients for ISCED1, but not for ISCED 2. Thus, the data in Sweden and France are incompatible with H0: equity for ISCED 1. Nevertheless, we failed to reject H0: equity for ISCED 2. Failing to reject the null hypothesis does not mean that it is true; we cannot, however, rule out the possibility that the association is due to chance alone (assuming the null hypothesis is true). Nevertheless, the coefficient is almost zero in Sweden for ISCED 2, indicating strong evidence for H0: equity.

Regarding predicted probabilities, in Figure 1, we plotted the predicted probabilities for socioeconomic disadvantage at z = [−1, 0, 1] for each outcome. For ISCED 1, the pattern appears to be strongest in Sweden, closely followed by France. For ISCED 2, the pattern appears to be strongest in the UK and roughly equal in Sweden and France.

Details are in the caption following the image
Predicted probabilities for the share of students in special needs by share of socioeconomic disadvantage (standardised). Data: TALIS.

Consider Sweden at ISCED 2. Comparing the average socioeconomic advantage to those with one SD above the mean corresponds to an average predicted difference of approximately five percentage points for the highest response option (more than 60% in SEN).

However, for ISCED 2, the pattern of socioeconomic disadvantage is reversed. For ISCED 2, the pattern appears strongest in the UK, followed by France (8%) and Sweden (2%). Considering the UK for ISCED 2, comparing the average socioeconomic advantage to those one SD above the mean corresponds to an average predicted difference of approximately 7 percentage points for the highest response option (more than 60% in SEN). France has roughly equal magnitudes as the UK, whereas, in Sweden, the average predicted probability is approximately 1 percentage point. Thus, a considerable difference can be observed.

Schools in Sweden and France may be more likely to refer socioeconomically disadvantaged students to SEN programmes early in their education,5 whereas schools in the UK may make such referrals at later educational stages. This implies a ‘wait-and-see’ strategy in the UK and an ‘early intervention’ strategy in Sweden.6 However, we do not know if ‘a change’ exists from ISCED 1 to ISCED 2. Thus, we cannot know for certain if we observe a ‘change’ or a mere ‘difference’, as the data is cross-sectional and not longitudinal. Nevertheless, we interpret the differences between countries as seemingly indicating differences between welfare states, thereby supporting H4: welfare.

Next, we plot the predicted probabilities for the share of refugees (Figure 2). Again, we observe the wait-and-see’ strategy in the UK, indicating larger predicted probabilities at later educational stages. However, in Sweden, we observe the early intervention strategy in the early educational stages. Schools in Sweden have the largest predicted probability of identifying students with SEN in refugee-dense schools. Notably, the predicted probabilities remain almost flat, suggesting some evidence supporting H0: equity in Sweden. Finally, France appears to be somewhere in between the UK and Sweden.

Details are in the caption following the image
Predicted probabilities for the share of students in special needs by share of refugee students (standardised). Data: TALIS.

In conclusion, we again find considerable support for H4: welfare, owing to considerable country differences. We also note no evidence supporting H2: underrepresentation.


SEN placement can have both intended and unintended consequences. Schools place students in SEN programmes with the intention of supporting them, thereby compensating for their social disadvantages (Berhanu & Dyson, 2012). However, SEN placement may have unintended consequences, such as stigmatising (or discrediting) and labelling students (Artiles & Trent, 1994).

Implications for research on social disproportionality in special education

Researchers disagree on whether overrepresentation or underrepresentation exists in SEN placement (Artiles et al., 2005; Farkas et al., 2020; Morgan et al., 2015, 2017). Our study found no support for underrepresentation (H2) (Cooc, 2017). This has two possible explanations: the sample in terms of the dispersion between refugees and immigrants (e.g. Syrians, Somalis), and the type of welfare state (social-democratic, Liberal or Conservative). This also depends on the country from which the refugees have fled. Whether all citizens have equal opportunities for education differs across countries. In some countries, girls are not permitted to attend school.

We found a high level of support for socioeconomic status (H3). One explanation for this may be that migration and socioeconomic status overlap (Berhanu & Dyson, 2012). First, overlap is caused by social and economic policies, as well as by children's unequal life chances due to their social background. Second, immigrants vary in their socioeconomic disadvantages; for example, Roma, Somalis, Algerians or Pakistanis, as opposed to Germans, Americans or Indians.

Thus, an examination of ethnic disproportionality in SEN must consider its overlap with ethnicity and poverty. Place of residence, segregation, and socioeconomic status overlap even in a social-democratic welfare state such as Sweden (Berhanu, 2010). We contend that the marginalisation and segregation of socially disadvantaged and ethnic minority groups have increased because of school deregulation. Municipality allocations have caused inequalities in results and resources to widen. For example, municipalities in Sweden may reallocate government funding from schools to healthcare to meet local needs (Berhanu, 2010).

This argument has been proposed in previous special pedagogical research (Strand & Lindsay, 2009). Even in social-democratic welfare states, poverty can be found in the vicinity of the suburbs. It is easy to forget that the social-democratic welfare state manages class mobility better than other states (Berhanu, 2010). However, this study demonstrates that concerns remain in social-democratic welfare states.

Implications for welfare state theories in special educational research

In support of welfare state theory, we found country-level differences in the identification of students for SEN. We found early interventions in French and Swedish schools, whereas UK schools implemented a wait-and-see strategy (H4). Non-liberal welfare states provide citizens with education early in life to equalise educational chances (Esping-Andersen et al., 2012).

However, we did not find support for welfare state theory with respect to socioeconomic disadvantages (H3). However, although these results do not appear to fit well with welfare state theory at first, we did find differences in the strength of the association in the UK compared to that in Sweden and France. These inconsistencies could perhaps be explained by the recent deregulation of social-democratic welfare states within the educational system. Thus, recent educational reforms have interfered with the inequality features of welfare states (Giota & Emanuelsson, 2016).

Implications for special educational needs policy

Care should be taken when posing policy advice based on a single study. Nevertheless, based on our results and previous research, our study may nevertheless offer some promising suggestions.

First, our study suggests that policymakers have much to gain and little to lose by learning from social-democratic types of policies. The social-democratic welfare state did better or at least no worse than the other welfare states. Specifically, we suggest that high quality early educational policy interventions that targets all children seem especially promising. Targeting all children both increases opportunities and reduces the risk of stigmatising or labelling students (e.g. discrediting or identifying students as deviant). Avoiding stigmatisation and labelling early in the life course may be crucial to avoid counterproductive polices with unintended consequences.

Second, we agree with earlier policy suggestions to strengthen the assessment and identification procedures for special education in each country.7 Currently, ethnic minority students are mostly tested in their second language and not their dominant language (Artiles & Trent, 1994; Monsrud, Rydland, Geva, & Lyster, 2022; Monsrud, Rydland, Geva, Thurmann-Moe, & Lyster, 2022; Subasi Singh, 2020). Future special educators need to be aware of the importance of testing minority students in their dominant language and monitor whether the tests used are translated appropriately (Monsrud, Rydland, Geva, Thurmann-Moe, & Lyster, 2022; Monsrud, Rydland, Geva, & Lyster, 2022). Exposure to these policies may result in attitudinal and behavioural changes in schools, such as revising SEN instructional and assessment practices (activities). These policy changes may reduce the probability of misidentifying students as having SEN.

Limitations and alternative explanations

This study has some limitations. The first limitation is the use of principals' self-reports to identify refugee students in classrooms and schools. Although the percentage of classrooms and schools with refugee students generally reflected the size of the refugee population in each country, principals may have over- or under-identified refugee students. Furthermore, since the data were reported by principals, it is difficult to identify what is socioeconomic status and what is class respectively cultural capital. The second limitation was grouping all refugee students under a single classification in the survey design. This overlooks their diverse contexts (see also Cooc & Kim, 2023). Third, principals were not asked about their special education credentials. A fourth limitation is that Germany, a country that hosts one of the refugee populations, did not participate in TALIS 2018. Finally, some might contend that the alternative explanation for our results would be ‘positive discrimination’. The positive discrimination explanation posits that socioeconomically disadvantaged schools simply gain more resources for SEN programmes. Although plausible, we defend our explanation as our model adjusts for SEN resources. Therefore, our model should adjust for such confounding variables explanation. Another alternative interpretation is that the quality of special education in the three countries differs; however, we have no data to support this supposition.


We aimed to compare the correlations between the share of students identified as having SEN and student diversity (socioeconomic and ethnic) at the school level in three countries (France, Sweden and the UK) using TALIS 2018 data.

This study makes two empirical contributions as well as a theoretical one. First, although studies have investigated TALIS data, no study has investigated the school level when studying over- and under-representation. Second, we address two types of overrepresentation: socioeconomic and ethnic (immigrants or refugees). Third, we compare different types of welfare states; thus, we were able to analyse SEN placement from a welfare perspective.

Our study proposes five hypotheses. The equity hypothesis (HO) holds for predicting migration at the school level. The overrepresentation hypothesis (H1) finds mixed support, depending on the type of welfare state and perhaps the level of education (ISCED1 vs. ISCED2). Thus, the underrepresentation hypothesis (H2) was rejected. We found support for the socioeconomic disadvantage hypothesis (H3). The overrepresentation hypothesis (H1) holds for the school in all three welfare states. Regarding The welfare hypothesis (H4), we found qualitative differences depending on the type of welfare state.


This study was conducted without any external financial support.


The authors have no conflicts of interest to declare relevant to the content of this article.


The datasets, TALIS, are publicly available for secondary analysis, Permission has already been obtained at the time of data collection.


The data is publicly available from the OCED at: TALIS 2018 Data—OECD.

  • 1 The example may be extended to include other mechanisms of misplacement such as the dominant cultural knowledge or language norms (Bourdieu, 1984).
  • 2 The overrepresentation of students with migrant backgrounds in special education can be tracked in many countries (Berhanu & Dyson, 2012; Strand & Lindsay, 2009; Subasi Singh, 2020).
  • 3 Studying overrepresentation as an aggregate measure aligns with sociological measurement through studying aggregate behaviours as social facts, which has a long tradition within the sociology of education.
  • 4 Swedish free schools are free of charge.
  • 5 In Sweden there is a Read, write, count—a guarantee for early support efforts.Teachers in Swedish public schools are given material to conduct qualitative assessments of reading progress in students in grade 1 (mandatory) and grades 2 and 3 (optional) (The Swedish National Agency for Education, 2019).Primary education in France has experienced a net drop in the number of students per class since 2017, in combination with an increase in teaching staff. In 2017–2018, the average number of students per teacher (19 in elementary schools) was higher than the EU average (Directorate of Evaluation, Forecasting and Performance Monitoring).
  • 6 Read, write, count—a guarantee for early support efforts.
  • 7 In the past, students with limited English knowledge, (e.g. Latino in the US) were placed in programmes för students with intellectual or learning disabilities. Today, they are placed in programmes for students with speech and language impairment (Artiles & Trent, 1994; Berhanu & Dyson, 2012; Sullivan & Bal, 2013; Sweller et al., 2012).