Educational attainment in Swedish (L1) and English (L2) for students with reading difficulties: a longitudinal case study from primary to the end of secondary school

ABSTRACT

This longitudinal case study examines reading difficulties identified in Grade 2 and Grade 6 national assessment test (NAT) scores in relation to Grade 9 NAT in Swedish (L1) and English (L2). A norm-referenced screening was used in Grade 2 to assess word reading and reading comprehension. In line with the simple view of reading, four subgroups were identified: (1) Typical reading development, (2) Word reading difficulties, (3) Reading comprehension difficulties, and (4) Mixed difficulties. Results indicate that reading difficulties in primary school have long-standing implications for educational attainment in Swedish and English. In logistic regression models students with reading difficulties in Grade 2 and low Grade 6 NAT-scores had a higher probability to achieve low attainment levels in Grade 9. This implies that a norm-referenced screening in Grade 2 can be useful for identifying students at risk of low attainment in language subjects at the end of compulsory school.

KEYWORDS:

• Reading difficulties
• educational attainment
• foreign language learning
• early identification
• national assessment test

Students with reading difficulties in primary school risk leaving compulsory school without prerequisites for further study at non-compulsory upper secondary school or with lower grades that reduce the range of study options (Cain & Oakhill, 2006; Fouganthine, 2012; Holopainen et al., 2017; Ricketts et al., 2014). This paper focuses on reading skills that have been shown to contribute to educational outcomes (Conti-Ramsden et al., 2009; Dockrell et al., 2011; Nordström et al., 2016). The context for this case study is Sweden, where in 2018, 15% of secondary school students failed to leave compulsory school with the grades required to continue to national upper-secondary school programmes (SNAE, 2018b). These students follow personalized routes to reach the requirements for a national programme, and approximately only 20% succeed (SNAE, 2019). Such failure increases the risk of unemployment with associated personal and societal implications (Smart et al., 2017; SNAE, 2014). Moreover, not achieving high grades affects students’ abilities to meet the requirements for some upper secondary programmes.

In the current case study, we investigate how well reading difficulties in Swedish primary school Grade 2 (G2), and national assessment test scores (NAT) in Grade 6 (G6) predict educational attainment as evidenced by NAT grades in Grade 9 (G9), the final year of compulsory school. To our knowledge, only one large-scale longitudinal study has examined reading in relation to educational attainment in Sweden (Fouganthine, 2012; Nordström et al., 2016). Given that a pass in the language subjects Swedish and English in G9 are prerequisites for upper-secondary school, we specifically focus on these linguistically demanding subjects. If the risk of low-grade achievement for these subjects in G9 can be predicted by a short classroom screening in G2, this could assist in designing support for the students. The Swedish Education Act (2010:800) requires that teachers give the support that is needed to provide further knowledge development to all students. Previous studies (Levlin & Nakeva von Mentzer, 2020; Vellutino et al., 2008; Wolff, 2011) indicate early reading interventions promote reading development for at-risk students.

Many primary schools in Sweden use classroom norm-referenced screening tools, often in conjunction with mandatory, criteria-based NATs in Grade 3 (G3) and G6, to identify students who may be at risk of not achieving the minimum knowledge requirements of compulsory school. Identified students are typically provided with additional learning support for a short period of time, and less typically with permanent support throughout compulsory schooling (SNAE, 2018c). It is unclear how short classroom norm-referenced screening tools, and the criteria-based NATs complement each other. Herkner (2011) and Herkner et al. (2014) have questioned whether the criteria-based NATs in G3 are sensitive enough to identify students at risk of reading difficulties, and even questioned the Grade 3 NATs validity (Herkner et al., 2014).

In this case study, we explore to what extent norm-referenced tests in G2, that aim to detect students with reading difficulties, and G6 NAT scores, that aim to detect pupils at risk of not achieving the minimum educational goals, may predict educational attainment in Swedish (L1) and English (L2) in the G9 NATs. Moreover, we use the simple view of reading’s (SVR) (Hoover & Gough, 1990) definitions of reading difficulties to further explore how the components of word recognition and reading comprehension may predict achievement in G9 NATs.

Simple view of reading

Reading and listening comprehension are two modes for acquiring knowledge in education and are viewed as prerequisites for critical reflection and inference skills. According to the simple view of reading, reading comprehension is defined as the outcome of decoding x oral language comprehension (Hoover & Gough, 1990; Tunmer & Greaney, 2010).

Reading difficulties according to SVR

According to SVR (Hoover & Gough, 1990) reading difficulties are due to variations in word recognition and oral language comprehension. Longitudinal studies have shown word recognition to contribute primarily to reading comprehension in the early stages of learning to read, while oral language comprehension increases in importance over the years of schooling (Fraser & Conti-Ramsden, 2008; Lervåg et al., 2018).

According to the SVR-model individuals with reading difficulties can be divided into: (1) specific reading comprehension difficulties, that is poor reading and listening comprehension despite age-adequate word recognition; (2) specific word reading difficulties, that is poor word recognition in combination with age-adequate oral language comprehension, and (3) mixed reading difficulties, that is poor reading and listening comprehension together with poor word recognition (Hoover & Gough, 1990; Tunmer & Greaney, 2010).

Individuals identified with specific reading comprehension difficulties have often already been identified as having poor oral language comprehension before school-start (Catts et al., 2016; Elwér et al., 2013; Nation et al., 2010). This indicates that poor oral langauge comprehension may be a precursor of reading comprehension difficulties.

Specific word reading difficulties are usually associated with phonological processing difficulties (see: Melby-Lervåg et al., 2012), yet with age-typical oral language comprehension. However, they can also occur in combination with poor comprehension as in mixed reading difficulties (Hoover & Gough, 1990; Tunmer & Greaney, 2010). Such individuals share the underlying cognitive and language profiles common in specific reading comprehension difficulties in addition to their poor word recognition (Tunmer & Greaney, 2010). In early reading development, mixed difficulties could also be a result of severe poor word recognition.

Reading difficulties in relation to educational attainment in language subjects

Several studies have considered the long-term impact of reading difficulties in relation to attainment in language subjects. For example, Holopainen et al. (2017) investigated educational performance in Finnish in G4–G9 (10–15 years of age) and found that students with specific word reading difficulties achieved on a par with students with typical reading development, whereas students with mixed reading difficulties performed significantly less well.

In contrast to Holopainen et al. (2017), Cain and Oakhill (2006) and Ricketts et al. (2014) focused on students with poor comprehension. Looking at educational performance in English, these studies suggest that the significant difference in attainment found in students aged 8-11, in comparison to age-matched controls, vanishes by age 16.

In the Swedish context, one longitudinal reading and educational outcome study has been conducted previously. This study followed students from schools G2–G9 (8—15 years of age). Nordström et al. (2016) found that word reading in G2 significantly explained educational attainment in G9 Swedish. However, no differentiation was made between typical and atypical reading development. Yet, an earlier publication based on a subset of the same data (Fouganthine, 2012) differentiated between various forms of reading difficulties; When considering G9 students, those with oral language comprehension difficulties performed significantly less well in the NATs in Swedish than those with word reading difficulties. However, the two groups had the same low attainment level in their end-of-compulsory school Swedish grade.

Few studies have explored word reading difficulties in relation to attainment in school foreign language subjects (SFLS). Several studies suggest that underlying difficulties in phonological processing in a first language (L1) are also evident in the second language (L2) (e.g., Bonifacci et al., 2017; Chung & Ho, 2010; Van Der Leij & Morfidi, 2006). However, some studies indicate that word recognition difficulties in L1 do not automatically transfer to L2, at least not for adult learners with dyslexia (Downey et al., 2000). Miller-Guron and Lundberg (2000) reported that a sub-group of dyslexic readers performed better in word reading in L2 English than in L1 Swedish.

To our knowledge, no previous study has investigated educational attainment in SFLS for students with poor reading comprehension in their L1. A few studies have explored the learning of SFLS by students with specific learning difficulties (DSM-5; American Psychiatric Association, 2013). Kormos (2017) systematic review of these studies identified the challenges as learning new vocabulary, grammar, word reading, reading comprehension, and writing. Considering that reading comprehension is a core component assessed in SFLS, it is likely that attainment in SFLS will be a challenge for students with L1 reading comprehension difficulties. It is, however, important when reading this article to be aware that English is ubiquitous in Sweden—films and tv programmes are subtitled rather than dubbed, and students use English in their out-of-school literacies (Holm et al., 2021).

Nordström et al. (2016) pointed out that for a long time, intelligence was the factor most frequently used for predicting educational outcomes, and that more recently other factors have been found to affect general school achievement. These factors include gender, motivation, personality, self-discipline, persistence, and self-esteem. However, our literature review suggests reading difficulties can lead to lower attainment levels in both English and Swedish and that students with specific word reading difficulties may be less vulnerable to lower attainment than students with reading comprehension difficulties.

Moreover, the reading ability profile children possess as early as second grade might predict performance in different ways as children progress in English and Swedish during compulsory school. It is also unclear which, if any, of these reader subgroups is most at risk of not achieving the grades necessary in Swedish and English to continue to upper secondary school, or at risk of having a limited range of upper secondary school study options. This case study is also one of the first exploring how students with reading difficulties perform after the Swedish 2011 curricula revision (SNAE, 2018a) that places a greater focus on critical reflection and inference skills.

Present study

The aim of this longitudinal case study is to examine how well reading difficulties in G2 and the NAT scores in G6 in English and Swedish predict NAT grades in G9 in English and Swedish, respectively. Reading difficulties are defined using the simple view of reading (Hoover & Gough, 1990). How well the four SVR subgroups, alone and in combination with G6 NAT scores, can predict the comprehensive (overall) NAT grades as well as the subcomponents (comprehension, oral communication, and writing) in Swedish and English in G9 is explored. A pass grade (E) in Swedish and English is required for upper-secondary school, however, higher attainment levels may be necessary to get access to popular programmes. Following grades are used in Sweden; F (fail), E (pass), D, , C, B, and A (highest attainment level). G6 marks the end of primary school with many students changing schools for their secondary education. G9 marks the end of compulsory school.

The following questions were explored in G9 for both Swedish (L1) and English (L2):

1. Are reading difficulties in G2 predictors of educational attainment in G9?

2. Are educational attainment scores in G6 predictors of educational attainment in G9?

3. Are reading difficulties in G2 in combination with educational attainment scores in G6 predictors of educational attainment in G9?

Reading difficulty was operationalized as typical vs non-typical reading (all subgroups with reading difficulties collapsed into one group) and as the SVR four subgroups in the regression models.

Method

Participants

A total of 214 G2 students from 14 public schools in two municipalities in a semi-rural area in the northern part of Sweden participated in the annual screening of reading comprehension and word reading (Jacobson, 2001; Järpsten, 1999). These students formed our recruitment pool. Only Swedish as L1 students were included in the study. We used both prospective and retrospective recruitment approaches. Participants who enrolled prospectively in G2 were 45 students with reading difficulties identified through the annual screening. Their caregivers gave written consent. Participants who enrolled retrospectively in secondary school were recruited without consideration of their reading ability; 64 students gave written consent. Hence, 109 provided access to their G2 screening results and their G6 and G9 English and Swedish NAT grades; 87 students completed all the required tests and met inclusion criteria.

Criteria for reading group inclusion and classification

Reader subgroups were selected based on their G2 screening test performance. Raw scores were converted to z scores according to tests’ standard reference data (Jacobson, 2001; Järpsten, 1999).

Following SVR, the 87 students were divided into four subgroups based on their word reading and reading comprehension z-scores (Cain & Oakhill, 2006; Nation et al., 2010). Three cut-offs were implemented: z ≤ −0.6, −0.7, and −0.8. The −0.6 and −0.7 cut-offs created identical subgroup membership. Comparing − 0.7 and −0.8 cut-offs to determine group membership yielded slightly different groups in a few instances, however, the results from the models were nearly identical. Therefore, we included a cut-off of z ≤ −0.7 in the final identification of subgroups. The following subgroups were identified:

1. Typical reading development (TRD, 42 students, 59% girls), word reading, and reading comprehension ≥ −0.5.

2. Word reading difficulties (WRD, 20 students, 40% girls), word reading ≤ - 0.7, and reading comprehension ≥ −0.5.

3. Reading comprehension difficulties (RCD, 13 students, 62% girls), reading comprehension ≤ −0.7 and word reading ≥ −0.5.

4. Mixed difficulties (Mixed, 12 students, 33% girls) in reading comprehension and word reading.

The group (n = 87) mean values fell within the expected age-norms in G2 (Jacobson, 2001; Järpsten, 1999). Detailed information is presented in Table 1.

Table 1. Descriptives for word reading and reading comprehension in Grade 2.

Measures	TRD (n = 42)		WRD (n = 20)		RCD (n = 13)		Mixed (n = 12)		Total (n = 87)
	M (SD)	Min/Max	M (SD)	Min/Max	M (SD)	Min/Max	M (SD)	Min/Max	M (SD)
Word reading
z-score	.51 (.87)	-.43/3.1	-.98 (.29)	−1.7/-.71	.16 (.74)	-.43/2.3	−1.1 (.33)	−1.9/-.71	-.11 (1.0)
Points	18.5 (6.1)		8.1 (2.0)		16.2 (5.2)		7.0 (2.3)		14.2 (7.0)
Reading comprehension
z-score	.62 (.66)	-.27/2.2	.37 (.58)	-.27/1.4	−1.1 (.32)	−1.9/-.83	−1.6 (.49)	−2.2/-.83	-.00 (1.0)
Points	12.2 (2.4)		11.3 (2.1)		6.0 (1.2)		4.2 (1.8)		9.9 (3.7)

Note. TRD = Typical reading development; WRD = Word reading difficulties; RCD = Reading comprehension difficulties; Mixed = RCD + WRD.

Materials

Screening materials—Grade 2

Reading comprehension. Diagnostic reading and writing (Järpsten, 1999). This test consists of short paragraphs that are read silently before four-alternatives multiple-choice questions are answered. Students have 30 min and can be awarded a maximum of 18 points. Internal consistency measured by Cronbach’s alpha was .62 for Grade 2 according to the manual (Järpsten, 1999).

Word reading. Word-chains (Jacobson, 2001). This test consists of 80 chains of three semantically unrelated words without blank spaces between them. Students mark each word boundary by drawing a line. The score is the number of correctly marked word boundaries in two minutes. Test–retest correlations were .89 for children in Grade 2 according to the manual (Jacobson, 2001)

National assessment test in Swedish and English—Grade 6 and 9

In the G6 and G9 NATS, the criteria-based grading scale is A–F, where A is highest pass, E is the lowest pass, and F is failed. The comprehensive grade for each subject is based on the performance in sections A, B, and C.

Swedish G6. Section A examines oral skills through an individual presentation, and a group discussion task (90 min). Section B examines reading comprehension through two fiction and two non-fiction texts (150 min) that require inferential skills, and the ability to identify facts. Section C examines writing. Students write one story and one argumentative text (130 min). In the writing assignments content, text organization, language use, spelling, and punctuation are assessed.

English G6. Section A examines students’ oral communication in a group discussion task (15 min). Section B examines reading comprehension and listening comprehension (70 min). Section C examines writing (45 min).

Swedish G9. Section A examines students’ ability to present a subject orally, lead and participate in a group discussion with 4–5 students (110 min). Section B examines reading comprehension (200 min). Section C examines writing – students can choose between argumentative, descriptive, and narrative assignments (200 min).

English G9. Section A examines students’ oral communication in group discussion (15-25 min). Section B examines reading (90 min) and listening comprehension (50 min). Section C examines writing (80 min).

Ethical considerations

The Regional Ethics Board, Umeå University, approved the research and data access in G2 and G6 (dnr 09-220), and evaluated the research questions and data access for G9 (dnr 2015/334-331Ö).

Procedure

Grade 2 screening

Schoolteachers conducted the screening between October and February.

National assessment tests

The students sat G6 papers between January and March, and G9 papers between November and May.

Data analysis

Logistic regression models were built in Stata to investigate educational attainment in Swedish G9 as predicted by reader subgroup scores from G2 (Table 4, models 1 and 2) and from educational attainment in G6 (Table 4, models 3 and 4). Tables 4–11 report estimates arising from logit specifications on educational attainment variables. Column 1 lists the predictors included in the model: reader subgroups, and G6 NAT scores. These were transformed into numerical values following the SNAE guidelines: A = 20; B = 17,5; C = 15; D = 12,5; E = 10; F = 0.

Model 1 investigates the G2 distinction of typically developing children compared to children with reading difficulties (without distinction by type of difficulties). Model 2 also models G2 scores, but it includes the reader subgroups as potential predictors. Both models exclude G6 NAT scores. Model 3 only includes G6 scores as potential predictors. Models 1, 2, and 3 control for gender and date of birth (dob). Model 4 includes both the G2 predictor (typical/non-typical) and the G6 NAT score predictor to model how both pieces of information contribute to the model. G9 educational outcomes were modelled as the dependent variable (DV) where a poor grade was considered as the outcome of getting a D/E/F.

All tables with estimates follow the structure described above and include the same model specifications. Alpha values were set at .10, .05, and .001. We are modelling complex behaviours with real outcomes for children who are struggling in school. Lenient alpha values are reported, as are commonly applied in contexts with similar models in various branches of economics (Filippini et al., 2017), in economics more generally (Hansen, 2021, p. 148), and in educational economic contexts (García et al., 2021). To remain cautious, these trends (p < .10) are only reported in the tables. We report these results as a longitudinal case study that describes our sample with precision; generalizations must be made with caution.

Results

In Table 2, the distribution of grades for all participants (n = 87) is described in comparison to the national statistics from the same year. There is generally a low proportion of students failing, apart from the subcomponents reading comprehension and writing in Swedish. The participants in this case study follow the same pattern as in the national statistics in the distribution of grades. However, a lower proportion achieves higher grade levels (A/B/C) compared to national attainment levels. This is particularly the case for the subcomponents reading comprehension and writing in both languages.

Table 2. Distribution of grades in the National assessment tests (NAT) in Swedish and English compared to national attainment levels.

NAT in Grade 9	All participants n = 87			National Statistics
		%			%
	F	D E	C B A	F	D E	C B A
Swedish
Oral communication	0	40	60	3	31	66
Comprehension	13	41	46	10	32	58
Writing	19	53	28	16	39	45
Comprehensive grade	2	61	37	6	41	53
English
Oral communication	0	28	72	2	27	71
Comprehension	5	37	58	5	13	72
Writing	2	48	50	4	31	65
Comprehensive grade	1	36	63	3	25	72

In Table 3, the distribution for each SVR subgroup is described. The proportion of students reaching higher attainment levels (A/B/C) in Swedish in G9 is considerably lower in all subgroups with reading difficulties compared to the group with TRD. In English, the pattern is slightly different with the two groups with reading comprehension difficulties showing a considerably lower proportion reaching higher attainment levels in G9, while the group with WRD performed almost on par with the TRD group.

Table 3. Distribution of grades in the National assessment tests (NAT) in Swedish and English across reading profiles identified in Grade 2.

NAT in Grade 9	TRD n = 42			WRD n = 20			RCD n = 13			Mixed n = 12
		n (%)			n (%)			n (%)			n (%)
	F	D E	C B A	F	D E	C B A	F	D E	C B A	F	D E	C B A
Swedish
Oral communication	0 (0)	14 (33)	28 (67)	0 (0)	7 (35)	13 (65)	0 (0)	8 (62)	5 (38)	0 (0)	6 (50)	6 (50)
Comprehension	4 (9)	13 (31)	25 (60)	1 (5)	11 (55)	8 (40)	4 (31)	6 (46)	3 (23)	2 (17)	6 (50)	4 (33)
Writing	6 (14)	18 (43)	18 (43)	6 (30)	11 (55)	3 (15)	2 (15)	8 (62)	3 (23)	3 (25)	9 (75)	0 (0)
Comprehensive grade	0 (0)	21 (50)	21 (50)	0 (0)	14 (70)	6 (30)	2 (15)	8 (62)	3 (23)	0 (0)	10 (83)	2 (17)
English
Oral communication	0 (0)	7 (17)	35 (83)	0 (0)	4 (20)	16 (80)	0 (0)	9 (69)	4 (31)	0 (0)	4 (33)	8 (67)
Comprehension	1 (2)	9 (21)	32 (76)	0 (0)	7 (35)	13 (65)	1 (8)	10 (77)	2 (15)	2 (17)	6 (50)	4 (33)
Writing	0 (0)	14 (33)	28 (67)	1 (5)	8 (40)	11 (55)	1 (8)	12 (92)	0 (0)	0 (0)	8 (67)	4 (33)
Comprehensive grade	0 (0)	8 (19)	34 (81)	0 (0)	6 (30)	14 (70)	1 (8)	10 (77)	2 (15)	0 (0)	7 (58)	5 (42)

Note. TRD = Typical reading development; WRD = Word reading difficulties; RCD = Reading comprehension difficulties; Mixed = RCD + WRD.

Predictors of educational attainment in Swedish (L1) in Grade 9: comprehensive scores

As early as G2, students with reading difficulties are more likely to have lower grades in the D/E/F range (p < .05) in the comprehensive score (Table 4). Model 1 is slightly preferred over Model 2 for G2 results. G6 NAT scores were statistically significant for models 3 and 4 (p < .001). Model 3 is preferred with more significant variables (including gender). Females tended to be less likely than male students to have poor grades across the models where gender was included (p < .05).

Table 4. Swedish comprehensive scores.

Predictor	Model 1			Model 2			Model 3			Model 4
	Grade 2:			Grade 2:			Gr 6 NAT:			Gr 2 TRD vs Non TRD +
	TRD vs Non TRD			Reader subgroups			Including gender/dob			Gr 6 NAT
	B	SE	OR	B	SE	OR	B	SE	OR	B	SE	OR
TRD	−1.07**	0.50	0.35							−0.20	0.58	0.82
WRD				0.62	0.63	1.85
RCD				1.56*	0.80	4.76
Mixed				1.33	0.88	3.79
Gr 6 NAT							−0.48***	0.12	0.62	−0.49***	0.13	0.61
Gender	−1.75**	0.54	0.17	−1.84**	0.57	0.16	−1.19**	0.60	0.30
Dob/2nd half	0.91*	0.53	2.47	0.97*	0.55	2.64	0.96	0.59	2.62
Constant	1.71***	0.50		0.67	0.49		7.16***	1.70		7.27***	1.66
N	87			87			87			87
r2_p	0.17			0.18			0.33			0.28
Ll	−47.75			−47.11			−38.40			−41.40
Aic	103.50			106.20			84.80			88.80
Bic	113..40			121.00			94.66			96.20

Predictors of educational attainment in Swedish (L1) Grade 9: subcomponent scores

Comprehension (reading)

The same models were modified to predict the probability of getting a D/E/F on each subcomponent. Again, as early as G2 (Table 5, model 1), students with typical reading were less likely to have poor grades for the comprehension score (p < .05). Furthermore, students with RCD (model 2) were 5 times more likely to have a poor grade on this subcomponent (OR = 5.28, p < .05) compared to students with TRD. The G6 NATs were a significant predictor of performance in models 3 and 4 (p < .05).

Table 5. Swedish comprehension scores.

Predictor	Model 1			Model 2			Model 3			Model 4
	Grade 2:			Grade 2:			Gr 6 NAT:			Gr 2 TRD vs Non TRD +
	TRD vs Non TRD			Reader subgroups			Including gender/dob			Gr 6 NAT
	B	SE	OR	B	SE	OR	B	SE	OR	B	SE	OR
TRD	−1.02**	0.45	0.36							−0.30**	0.53	0.74
WRD				0.68	0.57	1.98
RCD				1.67**	0.74	5.28
Mixed				0.94	0.70	2.56
Gr 6 NAT							−0.25**	0.08	0.78	−0.22**	0.08	0.81
Gender	−0.55	0.46	0.58	−0.65	0.48	0.52	−0.64	0.49	0.53
Dob/2nd half	0.24	0.47	1.27	0.29	0.47	1.34	0.32	0.49	1.38
Constant	0.85**	0.41		−0.14	0.44		3.87***	1.17		3,41**	1.11
N	87			87			87			87
r2_p	0.06			0.08			0.14			0.13
Ll	−56.24			−55.48			−51.63			−52.44
Aic	120.50			123.00			111.30			110.90
Bic	130.30			137.80			121.10			118.30

Oral communication

Model 2 accounting for reader subgroups is preferred over model 1 comparing typical vs non-typical reading (Table 6). Students with RCD were nearly 5 times (OR = 4.81, p < .05) more likely than students with TRD to fall into the lower grade ranges. Regarding G6 NAT scores, model 3 is preferred over model 4 as it includes gender which is a significant predictor in all of the models where it was included (p < .001).

Table 6. Swedish oral communication scores.

Predictor	Model 1			Model 2			Model 3			Model 4
	Grade 2:			Grade 2:			Gr 6 NAT:			Gr 2 TRD vs Non TRD +
	TRD vs Non TRD			Reader subgroups			Including gender/dob			Gr 6 NAT
	B	SE	OR	B	SE	OR	B	SE	OR	B	SE	OR
TRD	−0.37	0.49	0.69							−0.03	0.52	0.97
WRD				−0.38	0.65	0.69
RCD				1.57**	0.76	4.81
Mixed				0.24	0.75	1.28
Gr 6 NAT							−0.13**	0.07	0.88	−0.12*	0.06	0.88
Gender	−1.86***	0.51	0.16	−2.20***	0.57	0.11	−1.92***	0.52	0.15
Dob/2nd half	0.14	0.51	1.15	0.24	0.53	1.27	0.12	0.52	1.13
Constant	0.59	0.41		0.34	0.48		2.30**	1.00		1.33	0.83
N	87			87			87			87
r2_p	0.15			0.19			0.18			0.05
Ll	−50.11			−47.44			−47.94			−55.76
Aic	108.20			106.90			103.90			117.50
Bic	118.10			121.70			113.70			124.90

Writing

Students with TRD in G2 (Table 7, model 1) had a lower probability of receiving a low score on the writing exam (p < .05). The model with the more general distinction between typical vs non-typical reading (model 1) was preferred over the four subgroups (model 2). The mixed subgroup was automatically omitted from model 2 due to collinearity.¹ The lower the G6 NAT scores (models 3 and 4) the higher the probability to perform poorly in writing (p < .05). Gender was significant in models 1–3 (p < .05).

Table 7. Swedish writing scores.

Predictor	Model 1			Model 2			Model 3			Model 4
	Grade 2:			Grade 2:			Gr 6 NAT:			Gr 2 TRD vs Non TRD +
	TRD vs Non TRD			Reader subgroups			Including gender/dob			Gr 6 NAT
	B	SE	OR	B	SE	OR	B	SE	OR	B	SE	OR
TRD	−1.51**	0.57	0.22							−0.96	0.61	0.38
WRD				1.28*	0.73	3.61
RCD				1.07	0.77	2.90
Mixed				0	(.)	1.00
Gr 6 NAT							−0.25**	0.09	0.78	−0.18**	0.09	0.84
Gender	−1.66**	0.60	0.19	−1.60**	0.61	0.20	−1.77**	0.61	0.17
Dob/2nd half	0.22	0.55	1.25	0.16	0.56	1.17	0.28	0.57	1.33
Constant	2.76***	0.64		1.23**	0.56		5.68***	1.48		4.19**	1.34
N	87			75			87			87
r2_p	0.18			0.15			0.21			0.14
Ll	−41.91			−40.18			−40.48			−44.12
Aic	91.83			90.36			88.97			94.24
Bic	101.70			102.00			98.83			101.60

Predictors of educational attainment in English (L2) in Grade 9: comprehensive scores

Model 2 with reader subgroups was judged to be a better statistical fit than model 1 (Table 8). Students with RCD were about 23 times more likely to have poor comprehensive scores (OR = 23.27, p < .001) and students with Mixed difficulties were nearly 6 times as likely to have poor scores (OR = 5.87, p < .05) than their peers with TRD. G6 NATs were statistically significant for models 3 and 4 (p < .001).

Table 8. English comprehensive scores.

Predictor	Model 1			Model 2			Model 3			Model 4
	Grade 2:			Grade 2:			Gr 6 NAT:			Gr 2 TRD vs Non TRD +
	TRD vs Non TRD			Reader subgroups			Including gender/dob			Gr 6 NAT
	B	SE	OR	B	SE	OR	B	SE	OR	B	SE	OR
TRD	−1.61**	0.50	0.20							0.30	0.72	1.35
WRD				0.59	0.64	1.81
RCD				3.15***	0.86	23.27
Mixed				1.77**	0.72	5.87
Gr 6 NAT							−0.67***	0.14	0.51	−0.70***	0.16	0.50
Gender	0.18	0.49	1.19	−0.05	0.54	0.96	−0.09	0.65	0.91
Dob/2nd half	−0.21	0.49	0.81	−0.11	0.53	0.89	−0.09	0.66	0.92
Constant	0.14	0.40		−1.37**	0.52		8.61***	1.98		8.78***	2.04
N	87			87			87			87
r2_p	0.10			0.19			0.44			0.44
Ll	−51.41			−46.37			−31.89			−31.82
Aic	110.80			104.70			71.77			69.65
Bic	120.70			119.50			81.64			77.05

Predictors of educational attainment in English (L2) Grade 9: subcomponent scores

Comprehension (reading and listening)

Model 2 was judged to be a better statistical fit than model 1 (Table 9). The model predicted that students with RCD were almost 18 times more likely to have poor comprehension (OR = 17.93, p < .001), and students with mixed difficulties were over 7 times as likely to have poor scores (OR = 7.45, p < .05) than their peers with TRD. G6 NATs were statistically significant for models 3 and 4 (p < .001).

Table 9. English comprehension scores.

Predictor	Model 1			Model 2			Model 3			Model 4
	Grade 2:			Grade 2:			Gr 6 NAT:			Gr 2 TRD vs Non TRD +
	TRD vs Non TRD			Reader subgroups			Including gender/dob			Gr 6 NAT
	B	SE	OR	B	SE	OR	B	SE	OR	B	SE	OR
TRD	−1.59**	0.49	0.20							0.49	0.72	1.64
WRD				0.65	0.61	1.92
RCD				2.89***	0.86	17.93
Mixed				2.01**	0.74	7.45
Gr 6 NAT							−0.70***	0.15	0.50	−0.75***	0.17	0.48
Gender	0.61	0.50	1.84	0.51	0.53	1.66	0.66	0.65	1.94
Dob/2nd half	−0.24	0.48	0.79	−0.17	0.51	0.84	−0.01	0.67	0.99
Constant	0.15	0.40		−1.41**	0.52		8.94***	2.03		9.70***	2.16
N	87			87			87			87
r2_p	0.10			0.17			0.46			0.46
Ll	−52.89			−48.74			−31.68			−31.98
Aic	113.80			109.50			71.36			69.96
Bic	123.60			124.30			81.22			77.36

Oral communication

Model 2 was judged to be a better statistical fit than model 1 (Table 10). Students with RCD were almost 12 times more likely to have poor exam scores (OR = 12.25, p < .001) than their peers with TRD. G6 NATs were statistically significant for models 3 and 4 (p < .001).

Table 10. English oral communication scores.

Predictor	Model 1			Model 2			Model 3			Model 4
	Grade 2:			Grade 2:			Gr 6 NAT:			Gr 2 TRD vs Non TRD +
	TRD vs Non TRD			Reader subgroups			Including gender/dob			Gr 6 NAT
	B	SE	OR	B	SE	OR	B	SE	OR	B	SE	OR
TRD	−1.06**	0.52	0.35							0.87	0.76	2.39
WRD				0.08	0.71	1.08
RCD				2.51***	0.75	12.25
Mixed				0.73	0.76	2.08
Gr 6 NAT							−0.51***	0.13	0.60	−0.58***	0.15	0.56
Gender	−0.38	0.51	0.69	−0.76	0.58	0.47	−0.64	0.65	0.53
Dob/2nd half	−0.26	0.52	0.77	−0.16	0.56	0.85	−0.37	0.66	0.69
Constant	−0.23	0.41		−1.14**	0.53		6.20***	1.75		6.28***	1.79
N	87			87			87			87
r2_p	0.06			0.15			0.36			0.35
Ll	−48.26			−43.52			−33.05			−33.22
Aic	104.50			99.04			74.10			72.43
Bic	114.40			113.80			83.97			79.83

Writing

Model 1 was judged to be a better statistical fit than model 2 (Table 11). Students classified with TRD were almost 74 percent less likely to have poor writing scores in G9 (OR = .26, p < .05) than their peers classified with non-TRD. The group with RCD was automatically omitted from the model due to collinearity and this could be one limitation of the model. G6 NATs were statistically significant for models 3 and 4 (p < .001). Gender was significant in model 2 (p < .05) however model 2 did not have good measures of statistical fitness.

Table 11. English writing scores.

Predictor	Model 1			Model 2			Model 3			Model 4
	Grade 2:			Grade 2:			Gr 6 NAT:			Gr 2 TRD vs Non TRD +
	TRD vs Non TRD			Reader subgroups			Including gender/dob			Gr 6 NAT
	B	SE	OR	B	SE	OR	B	SE	OR	B	SE	OR
TRD	−1.34**	0.46	0.26							0.04	0.60	1.04
WRD				0.32	0.58	1.38
RCD				0	(.)	1.00
Mixed				1.21*	0.72	3.34
Gr 6 NAT							−0.57***	0.13	0.57	−0.50***	0.12	0.60
Gender	−0.64	0.47	0.53	−1.07**	0.54	0.35	−1.15*	0.60	0.32
Dob/2nd half	0.34	0.48	1.41	0.60	0.53	1.82	0.80	0.60	2.23
Constant	0.85**	0.42		−0.38	0.46		8.41***	1.88		7.23***	1.66
N	87			87			87			87
r2_p	0.10			0.09			0.37			0.33
Ll	−54.36			−45.88			−38.25			−40.73
Aic	116.70			101.80			84.50			87.46
Bic	126.60			113.30			94.37			94.86

Discussion

This longitudinal case study examined how well reading difficulties in G2 and the G6 NAT scores predict G9 NAT grades in English and Swedish. The students reading abilities in Swedish were assessed in G2 using a standardized norm-referenced screening tool for quick group assessment and placed into one of the four SVR reader subgroups. These subgroups also allowed us to create a more general distinction between students with typical reading versus non-typical reading. Using G9 NAT scores in English and Swedish, the predictability of the G2 reader subgroups and the G6 NAT scores in Swedish and English was tested using four logit-regression models.

Typical versus non-typical reading and educational attainment in L1 and L2 in Grade 9

Students with reading difficulties (regardless of reader subgroup) in G2 are represented to a greater degree than students with typical reading among those awarded lower G9 NAT grades in both Swedish and English. Except for the Swedish oral communication subcomponent of the G9 NATs, the distinction typical reading/non-typical reading was found to significantly predict G9 NAT performance (model 1) in the subcomponents comprehension and writing in both languages, and oral communication in English. This is in line with previous studies that have identified reading difficulties as a risk factor for lower educational attainment generally (e.g., Cain & Oakhill, 2006; Fouganthine, 2012; Holopainen et al., 2017; Ricketts et al., 2014) and for writing difficulties (Carretti et al., 2016; Sumner et al., 2013; Wengelin et al., 2014).

Given that writing is a complex task (Hayes, 2012) with high demands on language and cognition, it is expected that writing will be challenging for students with reading difficulties. This was confirmed in the current case study in model 1, indicating that reading difficulties (regardless of type) were associated with a higher probability of receiving a lower score in writing in both L1 and L2. Previous studies have posited a range of underlying reasons why writing is particularly challenging for students with reading difficulties. For example, students with poor word reading usually have underlying difficulties in phonological processing that affect decoding and spelling (Sumner et al., 2013; Wengelin et al., 2014), making transcription taxing. This has been found to impact general text quality, lexical diversity, and grammar due to un-automatized transcription skills leaving less capacity in working memory for higher-level processes in writing (Berninger et al., 2008; Sumner et al., 2016). For students with reading comprehension difficulties, text organization, the complexity of written language and general text quality have all been found to be hampered due to underlying limitations in oral language (Carretti et al., 2016; Cragg & Nation, 2006).

SVR reader subgroups and educational attainment in L1 and L2 in Grade 9

Overall, the models with a more general typical reading vs non-typical reading distinction for the Swedish (L1) NAT grades indicated a better statistical fitness over the models including the SVR reader subgroups. However, teasing apart the non-typical readers into the three SVR subgroups highlighted that the students with poor reading comprehension in G2 (RCD and Mixed subgroups) are generally more likely to perform poorly than both the typical group of readers (TRD) and the students with word reading difficulty (WRD). For Swedish, the logit-regression model 2 found that belonging to the RCD group in G2 significantly (p < .05) predicted lower grades on the reading comprehension and oral subcomponents of the G9 NAT indicating a risk of low performance in L1 reading comprehension for the duration of compulsory school.

It is not unexpected that reading comprehension difficulties in G2 persist over time (Catts et al., 2012; Elwér et al., 2013), and especially so when the demands on inferencing skills and reflection increase in secondary school. Although not unexpected, this finding shows that a simple screening of reading comprehension in G2 can be useful in schools. Furthermore, the oral communication subcomponent was challenging for the students with RCD, indicating that underlying oral language difficulties may influence both receptive as well as expressive processes.

For English, unlike Swedish, the models including the SVR subgroups overall indicated better statistical fitness over a more general typical reading vs non-typical reading distinction. It was found that especially the students with RCD and Mixed difficulties in G2 were generally more likely to perform less well than both students with TRD and WRD. These findings suggest that students with reading comprehension difficulties may need support in English in addition to any classroom or special educational needs support that they already receive in Swedish.

Only a few studies have considered RCD in relation to SFLS learning in contexts where the language is not the societal language. Kormos (2017) undertook a literature review of specific learning difficulties (SLD) and SFLS learning. Kormos reported that vocabulary, word reading, reading comprehension and writing challenged students with SLD when learning a non-societal SFLS. The students in our case study identified with RCD demonstrated similar challenges. Even though English in Sweden is much nearer a societal language than in many European countries, this situation does not appear to help Swedish students with RCD. Swedish research has found that students learn English vocabulary from social media (Sundqvist, 2009; Sylvén & Sundqvist, 2012), and learn English listening and writing skills from digital out-of-school literacies (Holm et al., 2021). None of these studies considered students with reading difficulties. Hence, these findings may relate only to students with typical reading.

The impact of these out-of-school literacy practices (Holm et al., 2021; Sundqvist, 2009; Sylvén & Sundqvist, 2012) does however appear to have lifted the performance of the students identified with WRD in G2. In G9 these students performed almost as well as students with TRD on the English comprehensive NAT. This result differs from other studies (e.g., Bonifacci et al., 2017; Chung & Ho, 2010; Van Der Leij & Morfidi, 2006). Differences in research design potentially explain this difference. Bonifacci et al.’s (2017) and Chung and Ho’s (2010) participants were younger and Van Der Leij and Morifidis (2006) participants were beginners.

Grade 6 NAT-scores and educational attainment in L1 and L2 in Grade 9

Models 3 and 4 found that G6 NAT scores are highly predictive of G9 NAT grades and its subcomponents. This is not unexpected given that these tests are based on the same national curriculum and constructed by the same university-based units. However, it also suggests that lower grades in G6 NATs should be considered when designing pedagogical support in secondary school. Different interventions could be factored into future regression models for English and Swedish that focus on the secondary school context, and the predictivity of G6 NAT scores for G9 NAT grades.

Confounding factors: date of birth and gender and educational attainment in L1 and L2 in Grade 9

Three models included two potential confounding factors, date of birth that was operationalized as being born in the first or the second half of the year, and gender. Date of birth was not found to be a strong predictor for any of the Swedish and English NATs G9 models. Gender was found to be a significant predictor for the G9 Swedish NAT subcomponents writing and oral communication. Girls were 15%−30% more likely to achieve higher grades than boys. However, boys and girls had similar receptive language skills in Swedish.

For the models analysing English G9 NATs, gender was generally not a good predictor. This result aligns well with earlier research showing boys perform better on the English NATs than all other subjects (see e.g., Holm et al., 2021). English has an out-of-school importance for gaming and social media that is of interest to boys and builds community. This could be part of the reason for the low predictive power of gender in English. However, some gender differences in the subcomponent writing could be interpreted as suggesting explicit language instruction is necessary to develop reasoning skills in boys’ English writing. Introducing out-of-school literacy activities into the logic regression models could be a potential way to investigate informal learning on NAT results.

Study limitations

This case study has three main limitations that mean the findings need to be interpreted with a degree of caution. First, the small sample size reduces the statistical power specifically when we separate the students into the four subgroups. However, we have interpreted instances of collinearity with caution which were likely overfitted due to the small sample size. Future research is needed to determine if these case study findings hold more generally.

Arguably the RCD group, for which we are able to rule out the word reading difficulties, is the most useful group in the context of this case study for predicting educational attainment in G9. However, more work is needed with larger sample sizes to explore the other reader sub-groups in greater detail.

Second, we were unable to influence the form of the annual G2 school-based screening, and this included no measure of listening comprehension. A measure of listening comprehension would have strengthened the validity of the reader sub-groups and the possibilities to understand the comprehension component in the SVR-model.

Third, we have not collected information about any additional support the participants may have received during their compulsory schooling and cannot, therefore, include support as a predictive factor in the regression models.

Implications and conclusions

In line with many previous studies, this case study indicates that reading difficulties in primary school have long-standing implications for attainment in language subjects and therefore also highlights the importance of implementing evidence-based early reading instruction in primary school (Levlin & Nakeva von Mentzer, 2020; Vellutino et al., 2008; Wolff, 2011).

Low attainment in L1 and L2 dominates the grades of students with reading difficulties, regardless of the reader subgroup. However, the students with reading comprehension difficulties are slightly more challenged compared to students with word reading difficulties only. This difference is especially evident in English. This suggests there is a need for more support focusing on learning English as an SFLS for students with reading difficulties. At the moment instruction in English as SFLS is not a part of the national requirements in Sweden for the special needs trainer programme. We suggest policy-makers consider if it could be appropriate to introduce such a national requirement given the importance of English in compulsory school and beyond. Indeed, entering upper secondary school with low attainment levels in the compulsory language subjects may affect the development of the high-level inferencing and reasoning skills, and literacy skills taught and required to pass upper secondary school, and later on gain a place at university or employment.

Overall, these results, providing they hold for a more general population, suggest adopting a two-stage procedure where a norm-referenced quick screening of reading ability in G2 may be used in a first step to identify children who may struggle in L1 and L2 language learning during primary school, and the second step in G6 when the NAT-scores are an even stronger predictive tool to identify students in need of support to reach higher attainment levels in the compulsory language subjects.

Acknowledgements

The authors thank all participating students, teachers, and principals for their valuable contributions and for making this project possible. The authors would also like to thank Adan L. Martinez Cruz for valuable support with the statistics.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Collinearity implies that both groups performed similarly. However, this similarity in performance may not be a general finding. We recommend caution in this respect as the size of our sample is small. Indeed, as one of our reviewers has kindly suggested, this collinearity implies that we may want to consider including only two subgroups instead of three. An alternative is to group all three in two categories. However, as evidence of collinearity is found in only 2 models out of 32, we have kept the three-category distinction for purposes of comparability across models.