Characteristics of Weak and Strong Readers in a Foreign Language

Reading in a Foreign Language: A Reading Problem or a Language Problem?

Alderson (1984) raised the question as to whether reading in a FL was a reading problem or a language problem. In a speculative article, he reported many FL teachers as holding the position that their students had problems reading in an FL because they were poor readers in their L1. Alderson concluded tentatively that “although both L1 reading ability and FL language knowledge affect FL reading comprehension, FL reading appears to be both a language and a reading problem, but with firmer evidence that it is a language problem, for low levels of FL competence, than a reading problem” (cited in Bossers, 1991, p. 48). Alderson speculated that there is likely to be a language threshold beyond which FL readers have to progress before their L1 reading abilities can transfer to the FL—which Clarke (1979) called the ‘short-circuit hypothesis.’

Since then, however, a number of empirical studies have investigated this question, many of which have administered tests of L1 and second or foreign language (SFL) reading comprehension, as well as tests measuring L1 and SFL vocabulary and grammar. Hacquebord (1989) tested 50 L1 Turkish students in basic vocational or basic general secondary education in the Netherlands. Their average age was 13.9 years, and most had immigrated to the Netherlands before the age of 4. Their knowledge of Dutch as a second language (L2; language spoken in the learner's environment) was four times stronger as a predictor of their L2 reading than their L1 reading, at both times of testing (2 years apart). It was claimed that this may have been due to loss of the L1 over time (but no evidence was offered).

Bossers (1991) tested 50 adult native speakers of Turkish, all of whom had passed the highest level of general secondary education in Turkey. They were recent immigrants to the Netherlands and were learning Dutch as an L2 in the Netherlands. He found a correlation between L1 Turkish and L2 Dutch reading comprehension of .59, but the correlation between L2 comprehension and L2 vocabulary and structures was much stronger at .83. Comparing the low L2 reading group (n = 35) with the high L2 reading group (n = 15), L2 proficiency was the only significant predictor for the low group, whereas L1 reading was the only significant predictor for the high group. Bossers (1991) concluded that “knowledge of the target language plays a dominant role initially and L1 reading becomes a prominent factor at a more advanced level. This is exactly what a threshold or language ceiling hypothesis would predict” (p. 56).

Bernhardt and Kamil (1995) found moderate correlations between L1 English and L2 Spanish reading tests and claimed that L1 literacy is a strong predictor of L2 reading at around 20% of variance, but L2 linguistic knowledge consistently accounted for around 30% of the variance. Taillefer (1996) studied L1 French and FL English, and found that L1 reading comprehension and FL knowledge accounted for 35% of the variance on the FL reading test, but FL knowledge accounted for three times more of the variance than did the L1 reading test. Yamashita (2002) tested Japanese students on their L1 Japanese reading, FL English reading and FL English knowledge. Together, L1 reading and FL knowledge accounted for 40% of the variance in FL reading, with FL knowledge being five times stronger than L1 reading.

Despite the differences in L1s tested, status of the SFL, the nature of the tests used (grade point averages, standardised reading and knowledge tests, cloze tests, recall measures, etc.), and the varying sizes and ages of the samples tested, it would appear that the findings consistently indicate greater importance for SFL knowledge in SFL reading than L1 reading ability, at least until some threshold (undetermined) of SFL proficiency has been passed.

L1 and FL Learning

In fact, the largest influence on SFL reading research has been the L1 research on how children learn to read, and in particular among at-risk readers, rather than dyslexics per se (for an overview of reading development, see Nunes & Bryant, 2009; Pang, 2008; readers can be ‘at risk’ for various reasons such as disability, poor instruction or lack of literacy support at home). Researchers have sought to identify the sorts of learning problems that are experienced by at-risk readers. Vellutino and Scanlon (1986) compared good and poor L1 readers to find that the poor readers had difficulties with basic L1 language skills, particularly phonology and syntax. A wealth of research since then has confirmed that phonological coding (sometimes known as word decoding) and difficulties with phonological awareness (the conscious ability to recognise, categorise, produce, and manipulate phonological information of the language at the syllable, onset and rime, and phoneme levels) lie at the heart of the problems that children experience when learning to read in their L1, with most early research being done in English (Bradley & Bryant, 1983; Liberman & Shankweiler, 1985; Stanovich, 1986a).

Sparks, Ganschow, and Pohlman (1989) proposed that at-risk FL learners also experience a linguistic coding problem when learning the FL. They called their explanation for this problem the Linguistic Coding Deficit Hypothesis (LCDH, later renamed as Linguistic Coding Differences Hypothesis); Sparks and his associates (Sparks & Ganschow, 1993a, 1993b, 1995; Sparks et al., 1998, 2008) have since published widely on the research justifying this hypothesis.

Numerous studies (reported in Sparks & Ganschow, 1993a) showed that no differences were found in L1 reading comprehension and vocabulary or IQ between at-risk and successful students, but significant differences were found in L1 phonology and syntax, as well as on the Modern Language Aptitude test (MLAT; Carroll & Sapon, 1959). Sparks and Ganschow also reported that methods that directly teach phonology in L1 also result in significant gains on the MLAT. Case studies by the same authors reinforced these findings and the authors argued that cognitive problems relating to basic language skills in L1 are the main cause of problems in learning an FL. They recommended identifying at-risk FL learners by using a battery of L1 measures, including phonology (pseudo-word recognition, spelling, phoneme segmentation), syntax (written grammar), semantics (vocabulary, reading comprehension, oral language), and FL aptitude. Sparks and Ganschow (1993b) argued that poor phonological processing and lack of phonological awareness may give rise to Matthew effects (Stanovich, 1986a, 1986b, 1991), which leads to motivational and attitudinal problems, and not vice versa.

Motivation for FL Learning

MacIntyre and Gardner (1991, 1994) argued that language anxiety is an important cause of individual FL achievement and of differences in FL learning (see also Skehan, 1991). MacIntyre (1995) asserted that Sparks et al.’s LCDH understates the importance of affective variables; he argued that, among other factors, language anxiety can play a causal role in creating individual differences in language learning.

Sparks and Ganschow (1995) reacted to MacIntyre's criticism by restating their belief that “problems with FL learning are not likely to be primarily the result of low motivation, poor attitude or high anxiety” (p. 235). They pointed out that anxiety can be related to L1 skills as well as to FL learning, and thus they argued that anxiety cannot be considered on its own, but alongside learners’ overall language skills, be they first or foreign language.

However, since the study to be reported here is specifically interested in the diagnosis of strengths and weaknesses in FL reading performance, it is important to explore whether motivation can indeed predict FL reading performance, despite Sparks and Ganschow's 1995 assertions.

More recently, studies of the relationship between motivation and FL proficiency have proliferated (Dörnyei, 2005; Dörnyei & Ushioda, 2011; Iwaniec, 2014; Kormos & Csizér, 2008), and that research has developed and used a wide range of motivational constructs, including Instrumentality, Intrinsic Interest, Motivational Intensity, Parental Encouragement, Anxiety, Self-Regulation, and SFL Self-Concept. Research (e.g., Netten, Droop, & Verhoeven, 2011) has indeed shown that motivation is connected with reading performance in both L1 and L2. Previous research has also shown that motivation is connected with individual differences in L2 learning (e.g., Skehan, 1991). Aspects of motivation, including anxiety, are potentially interesting factors for diagnosis because they may not only explain strong versus weak performance in reading in an SFL, but the educational system can try to do something about them. For example, teachers can try to change learners’ negative attitudes and persuade them to see learning in a more positive light and thus become more motivated learners.

Based on recent findings by Kormos and Csizér (2008), it is suggested that learners’ age also plays a role in the importance of different motivational constructs. In the study to be reported in this article, we therefore included measures of such motivational constructs as well as the age of learners among the potentially important factors in the difference between weak and strong FL readers.

Methodological Issues

Most of the evidence for the link between L1 language skills and SFL learning is based upon correlational and regression studies of the relationship between various L1 measures, language aptitude (MLAT, Carroll & Sapon, 1959), and SFL achievement, as well as on in-depth case studies. In two studies reported in Sparks et al. (1998), however, the authors examined the differences in L1 skills, FL aptitude, and FL achievement among high school students classified as high-, average-, and low-proficiency FL learners. Because this is a methodology that we partially follow in the current study we now provide details of the Sparks et al. studies.

Both studies reported in Sparks et al. (1998) examined differences in L1 language skills, FL aptitude and FL second-year final course grades. The first study involved 60 females divided into three groups according to their FL proficiency level. High proficiency learners had a mean score at least 1 standard deviation (SD) or more above the mean (n = 13); the average proficiency group (n = 35) scored between .99 SDs above the mean to .99 SDs below the mean. The low proficiency group (n = 12) had a mean score of −1 SD or more below the mean. A multivariate analysis of variance (MANOVA) showed overall differences among the three proficiency groups on all L1 language measures and the FL aptitude test. There were significant (p <.05) differences between High and Low groups on all L1 and aptitude measures except for phoneme deletion.

Study 2 was similar to Study 1. Thirty-six 10th-graders enrolled in second-year Spanish, French and German courses took several measures of their L1 language skills, some of which were similar to those in Study 1, plus the MLAT, during their first-year FL course. Analyses of variance (ANOVAs) and post hoc tests showed similar significant differences between the groups as in Study 1. Significant differences between the High and Low groups were found on all the L1 measures (where phoneme deletion was replaced by an informal “Pig Latin” test of phonemic awareness). Sparks et al. (1998) concluded that “students with low average to average (L1) language skills as reflected in a variety of tests measuring phonology/orthography, semantics, group achievement, and foreign language aptitude may still experience moderate to severe difficulties with foreign language learning in school” (p. 206). There may well be implications for the prediction and diagnosis of potential weaknesses in FL learning: the authors suggested that “foreign language word decoding is a good predictor of overall (oral and written) proficiency in a foreign language” (p. 207).

Although Sparks and associates used standardised measures of L1 cognition and reading and of IQ, they only used FL measures of the same cognitive skills in one study (Sparks et al., 2008), and their measures of FL proficiency were not standardised: They were either course grades, or informal, researcher-made measures somehow linked to the descriptors on the American Council on the Teaching of Foreign Languages (ACTFL). In addition, reliability coefficients were not reported in most studies, other than for the researcher-made ACTFL-based proficiency tests.

Meta-Analyses

Despite the limitations mentioned before, the studies reported present interesting findings about the importance of L1 and SFL skills in SFL reading, whose tentative conclusions would be strengthened by rigorous meta-analyses.

Melby–Lervåg and Lervåg (2014) conducted a meta-analysis of 82 studies comparing L1 and L2 learners (‘L2’ was defined as the additional language used in the context of English) that looked at reading comprehension and its underlying components, which they list as language comprehension, decoding, and phonological awareness. Moderator variables included age, socioeconomic status, home language, instructional language, differences between L1 and L2, writing system in L1, consistency of L1 orthography, test type, and nonverbal IQ as a moderator to rule out the influence of this factor on group differences. The authors concluded that “unless specific decoding problems are detected, interventions that aim to ameliorate reading comprehension problems among second language learners should focus on language comprehension skills” (p. 409). Thus, this meta-analysis would appear to support the conclusions of research into Alderson's (1984) Threshold Hypothesis, namely that reading in an SFL is more of a language problem than a reading problem, rather than Sparks et al.’s (1998) LCDH. Another conclusion was that there are at least two kinds of struggling readers, those who have decoding problems and those who have comprehension problems, and therefore the prediction and treatment of such different weaknesses in SFL reading might be somewhat different.

Wide-ranging though this meta-analysis is, it still has limitations. A rather low-level or basic set of cognitive skills (primarily decoding) was the focus, rather than higher-order cognitive variables which were not measured, including working memory and automaticity (fluency). Although L1 and L2 vocabulary were measured, other linguistic variables like spelling, syntax, semantics, and writing were not. Nor were reading skills included, like inferencing; gist and main idea comprehension; scanning and skimming (so-called expeditious reading); integrating/synthesising; and interpreting, reflecting, and evaluating.

Jeon and Yamashita (2014) also reported a meta-analysis of L2 reading comprehension and its correlates. The authors argued that a components approach to reading comprehension allows one to observe how different processes interact with each other, as predicted by Stanovich's (1980) Interactive–Compensatory Model of reading, as well as to assess how much the individual components contribute to the overall skill, be that of reading in L1 or L2. The meta-analysis asked “whether the individual ability difference observed in L2 reading comprehension is better explained by the variance observed in L2 language knowledge (e.g., vocabulary, grammar) or by variance stemming from more language-general, cognitive, or metacognitive processes (e.g., working memory, metacognitive knowledge)” (Jeon & Yamashita, 2014, p. 162). This meta-analysis is thus relevant both to the claims of Sparks and associates as to the role of L1 knowledge in FL learning, and to the research inspired by Alderson's (1984) question whether FL reading poses a reading problem or a language problem.

Ten passage-level SFL reading correlates were included in Jeon and Yamashita's (2014) study, namely SFL decoding, vocabulary knowledge, grammar knowledge, phonological awareness, orthographic knowledge, morphological knowledge, listening comprehension together with L1 reading comprehension, working memory, and metacognition.

The authors concluded that grammar and vocabulary are equally important correlates of SFL reading comprehension. The correlations for SFL language variables were larger than the mean correlations for language-general variables, and the authors argued therefore that SFL reading poses a language problem more than a reading problem. Moreover, SFL reading correlated strongly with SFL listening, providing support for the Simple View of Reading (Hoover & Gough, 1990).

Jeon and Yamashita (2014, p. 196) called for more research on SFL orthographic and morphological knowledge, and cited Jeon (2011) as showing FL morphological knowledge and awareness to contribute substantially to FL reading comprehension, and Kato (2009) as showing that L2 orthographic knowledge was an important contributor to L2 comprehension among higher proficiency readers.

However, Jeon and Yamashita (2014) acknowledged similar limitations of their study to those previously described with respect to the Melby–Lervåg & Lervåg (2014) study, namely the limited variety of processing efficiency variables, largely confined to decoding in this study. They also acknowledged the importance of the distinction between L2 and FL contexts.

Summary Discussion

It is clear that weak SFL readers are worse readers than strong SFL readers, but weak at what and why? Are they weak because of L1 reading problems or because their SFL is weak? If they have a reading problem, what is that problem, exactly? If they are weaker in their SFL, why, and in which areas do they have weaknesses? In short, what factors underlie their SFL reading weaknesses? Do weak readers have a different profile across a range of factors that distinguishes them from stronger readers? If so, are these factors cognitive, linguistic, affective, motivational, or are they bound up with background features that affect their ability to read in the SFL?

The two major hypotheses that we have now explored—the Threshold Hypothesis and the Linguistic Coding Differences Hypothesis—seem to be in contrast with each other, and related research seems to have reached different conclusions. A number of variables have been explored within the broad areas of cognition, language, and affect but within somewhat different frameworks and with somewhat different outcomes. Could it be that insufficient attention has been paid to the range of possible measures that operationalize the researchers’ constructs? Might it be that the measures of L1 and/or SFL reading are inadequate in some way, or that L1 and/or SFL language knowledge, skills, and ability have been inadequately construed?

In short, many questions remain to be answered, and if we wish to have a better understanding of what underlies weaknesses in SFL reading in order to develop diagnostic procedures, relevant feedback, and more appropriate interventions in order to help weak readers become stronger readers, then we need to investigate further and more deeply, in order to arrive at clearer answers to at least some of these questions. That is what the exploratory project ‘Diagnosing reading and writing in a second or foreign language’ (DIALUKI), to be described in the next section, set out to do.

Research Questions

Previous studies (e.g., Sparks et al., 1998) and meta-analyses (Jeon & Yamashita, 2014; Melby–Lervåg & Lervåg, 2014) suggested that to improve our understanding of weak FL reading performance we need to investigate FL readers with a wider repertoire of measures. The DIALUKI project sought to do precisely that, and therefore the research questions addressed here were the following:

Given the exploratory nature of the study, it did not set out to investigate specific hypotheses. However, our general expectation based on previous research was that learners’ L1 and FL skills, and possibly also their cognitive performance and motivation, should distinguish weak and strong FL readers, but which of these areas would turn out to be the clearest separators was difficult to predict. As regards the three age groups, we expected to find some differences between them simply because of their different age and amount of prior English studies but the exact nature of these differences could not be predicted, as we were not aware of any previous research comparing age groups in the way we did.

Participants

The participants were three groups of native Finnish speakers learning EFL in primary and secondary schools in Finland: 10-year-old children in Grade 4 of primary school, who had been learning English for about 18 months; 14-year-old adolescents in Grade 8 of lower secondary school; 17-year-old students in the second year of academic upper secondary school (gymnasium). These groups were selected since they represent the widest possible range of ages and proficiency levels in English available for systematic study in the Finnish educational system: The 4th-graders had just enough prior English exposure to make it meaningful to investigate literacy skills in English, and the second year in the gymnasium is the last stage at which students preparing for the high-stakes final examination (typically in their third year) could be expected to participate in such school-external activities as the study reported here. The English test results indicate that we did indeed manage to cover a wide range of proficiencies. The 8th-graders and gymnasium students took the DIALANG English reading tests (see Appendix A); therefore, we have estimates of their Common European Framework of Reference (CEFR; Council of Europe, 2001) levels in reading. The most typical level for the 8th-graders was A2 and for the gymnasium students B1 or above, even up to C1/C2. The 4th-graders’ English reading scores cannot be reliably related to the CEFR but presumably most of them were at A1 or below. The CEFR-based ratings of participants’ writing ability in English yielded very similar results. The study reported here is a cross-sectional study across these three groups. The total number of participants involved was 203 4th-graders (girls 52%, boys 48%); 204 8th-graders (girls 56%, boys 44%); 205 gymnasium students (girls 53%, boys 47%).

Instruments

The independent variables were the categorizations of FL readers into weak and strong readers based on their performance on standardized English reading tests (see the Analyses section). The 4th-graders were given the Pearson Test of English (PTE) Young Learners Test at levels A1/A2 on the CEFR. The 8th-graders were given the PTE General (Reading) at CEFR levels A2/B1 as well as the DIALANG Test of Reading (Intermediate) at levels B1/B2. The gymnasium students were given the PTE General Reading Test and the DIALANG Reading Test (Advanced) at a higher level (B1–C1 on the CEFR) than the tests taken by the 8th-graders (see Appendix B for the descriptive statistics and reliabilities).

A number of cognitive (L1 and FL), linguistic (L1 and FL), and motivational measures as dependent variables were also administered within 2 months. The 17 cognitive measures used covered phonological processing, working memory, and lexical retrieval; both L1 and FL versions of the measures were administered. A total of five L1 and four FL measures were also taken by the learners (e.g., vocabulary, writing, and segmentation tasks). In addition, they responded to an extensive questionnaire covering seven dimensions of motivation on using and learning English (e.g., intrinsic and instrumental motivation, anxiety, self-regulation). For details of the instruments see Appendixes A to C.

Analyses

To compare learners’ reading proficiency in English with the various cognitive, linguistic, and motivational variables, learners’ English reading scores (in fact, Rasch measures since they provide more valid indicators of ability than raw scores; see, e.g., Engelhard, 2013) were first converted to z-scores and then the learners were divided into three groups: strong, medium, and weak readers. We used one standard deviation above/below the mean as the cut-off points (the same procedure was used by, e.g., Sparks et al., 1998). In the current study, however, only the weak and strong FL readers are investigated. The number of weak and strong readers varies to some extent between different analyses, for example, due to absence of some students on the day particular measures were administered. In most analyses, there were 15‒22 weak FL readers and 20‒30 strong FL readers (see Appendixes D‒G for details).

Before comparing weak and strong FL readers, we condensed the 17 different cognitive measures into a more manageable number of constructs. The other measures had either already been turned into factors (motivation) or were large enough units in their original form (e.g., reading, writing, vocabulary). Since the 8th-graders and gymnasium students took exactly the same cognitive measures, a joint factor analysis could be performed for them. However, the 4th-grade data had to be analysed separately because some of the measures were shortened or were not administered at all in order to lessen the burden to these young learners.

Factor analyses of the cognitive measures yielded five factors for the 4th-graders and four factors for the two older age groups. The decision on the number of factors was based on examining the eigenvalues, scree plots, and the interpretability of the factors. The factor content was remarkably similar, but not identical, across the groups. In the two older groups, the biggest factor featured rapid naming and reading of individual words, in both the L1 and the FL, but in the 4th-grade group the key components of this factor were the rapid word recognition tasks, again in both languages, even though the rapid word list reading tasks also loaded heavily on this factor (see Appendix H). Most of these loadings were at the .7 or .8 level. In some cases the three types of measures which aimed at tapping automatization and speed of access to vocabulary (i.e., rapidly presented words, word list reading, and rapid alternating stimulus naming) also loaded on other factors than the fluent reading factor(s). In grade 4, the Finnish word list reading task also had a lot in common (over .5 loading) with the Finnish phonological skills factor. In the two older groups both the word list and alternating stimulus tasks, in both L1 and FL, also loaded on the phonological factor, and in particular the English versions of the word list and alternating stimulus tasks also loaded on the working memory factor. Thus, there is a clear overlap between these and the other factors, which might suggest that there is a higher-order cognitive factor in the background. However, since our aim was to obtain more detailed, potentially diagnostic information about the distinguishing characteristics of weak and strong FL readers, a multi-factorial approach was used.

In both weak and strong reader groups it was possible to identify a working memory factor in which the core comprised the two backwards digit span tasks. As might be expected, working memory plays a significant role in many other cognitive tasks and, consequently, several other cognitive measures loaded at least moderately (typically between .4 and .5) on that factor.

Two phonological factors were also identified in both groups. One was based on specific Finnish phonological tasks, whereas the other differed: For the older students, the factor related mainly to English phonological tasks and for the 4th-graders it was based on phonological tasks in both languages.

Overall, the first ‘fluent reading’ factor explained 26% of variance in the L1 and FL cognitive measures in both the combined 8th-grade and gymnasium group and in the 4th-grade group, whereas the second factor (i.e., mainly English phonology in the two older groups or another ‘fluent reading’ factor in the 4th grade) explained only 6‒8% of the variance.

To examine the differences between the strong and weak FL readers statistically, a number of MANOVAs were run on groups of independent variables (e.g., L1 linguistic measures, FL linguistic measures, cognitive factors, and motivation) to investigate any overall differences between the weak and strong FL readers in each group of variables of interest. These were followed by univariate tests (in MANOVA) to examine the differences in more detail.

Trends and Differences Across Age Groups

It should be borne in mind that the three age groups represent three broadly different proficiency groups as the 4th-grade, 8th-grade, and gymnasium students had studied English for 1½, 5½, and 8½ years, respectively. As mentioned earlier, the 4th-graders were typically at the CEFR level A1 or below, the 8th-graders at A2, and the gymnasium learners at B1 or above as far as their English reading level was concerned.

Two broad similarities were evident across the three age groups. First, the content of the motivational and L1 and FL cognitive factors was rather similar in all groups. Second, the relative importance of FL, L1, cognitive, and motivational factors in distinguishing weak and strong EFL readers was the same: FL skills were always clearly the best separators.

Despite these two similarities, some differences across groups may be important diagnostically as they imply that certain indicators may be important at certain ages or stages of learning. It is worth remembering that the 4th- and 8th-graders represent whole age groups whereas the gymnasium is typically attended by the more academically oriented students.

Even though command of the FL skills other than reading was the most important separator of weak and strong FL readers in all age groups, their importance increased with learners’ age and proficiency (see Appendix E). FL vocabulary knowledge is a good example: Its effect size increased from d = 2.33 in grade 4 to d = 3.86 and d = 5.01 in the two older groups. The same trend could be observed for FL writing, and, less clearly, for dictation. Segmentation accuracy was also a clear separator (with d = 1.88‒2.72) for all ages. These findings do not suggest a threshold such that L1 reading becomes more important with age or increasing FL proficiency and thus might appear to contradict Bossers (1991). However, he investigated L2 rather than FL learners.

Interestingly, segmentation speed did not separate weak readers from strong ones in the youngest group but did so for the 8th grade and gymnasium (with d = .71 and .93, respectively). This suggests that the speed of performing a task that involves both vocabulary and syntactic knowledge is potentially a useful indicator of reading (comprehension) problems for learners who have studied the language for some considerable time.

Among first language skills, L1 reading was the only consistently good separator of weak and strong FL readers (with d ≥ .98). L1 writing separated weak from strong FL readers in the two younger groups only and L1 vocabulary knowledge only in the two older groups. The latter finding is particularly interesting, as the vocabulary test was very easy (see Appendix C): Learners who failed to achieve the maximum score on L1 vocabulary were likely to be among the weakest FL readers. Another diagnostically interesting finding was that both L1 segmentation and L1 spelling error detection often distinguished weak and strong FL readers (Appendix D). Spelling error detection was administered only in the two older groups but both the speed and accuracy in detecting the errors separated the two reading groups. Segmentation accuracy did the same in both the youngest and oldest age groups but was only a marginal separator in the middle, 8th-grade group. Noteworthy was the large effect size (d = 1.55) of L1 segmentation speed in the gymnasium group; it appears that problems in rapid segmentation of either L1 or FL text may be an indicator of FL reading comprehension problems even among learners with considerable experience of study of the FL.

Cognitive factors tended to separate weak and strong readers less clearly in the youngest (4th-grade) group (with only one d > 1.00). In contrast, three cognitive factors distinguished the weak and strong readers very clearly (mostly d > 1.00) in the two older groups. First, fluent word level reading (based on both L1 and FL tasks) separated FL readers quite well among the two younger groups but less well in the gymnasium. Second, in contrast, the working memory factor (based on both L1 and FL tasks) only distinguished weak from strong readers in the two older groups. Third, a factor based on phonological tasks presented in English also separated weak and strong readers clearly in the two older groups, which reinforces the trend noticed earlier that FL skills in general are increasingly important as separators of weak versus strong FL readers as learners’ age and experience with the language increases.

As for motivation, although self-concept distinguished weak from strong readers in all three age groups, as learners’ age increased, more and more motivational variables separated weak readers from strong ones. Thus we might tentatively conclude that with increasing age, motivation to learn or read the FL decreases among the weakest readers, or that the gap between the strong and weak readers seems to become wider with respect to motivation. Alternatively, a lack of improvement in one's reading ability may bring about decreasing motivation. However, it was somewhat encouraging to find that some aspects of motivation that can potentially be affected by teachers and parents, such as Intrinsic motivation and Motivational intensity, did separate weak and strong FL readers, particularly among the gymnasium learners.

Moreover, as one reviewer of this study pointed out, while FL Self-concept was found to be a significant separator in all age groups, Anxiety was significant only for the gymnasium group, which is the oldest and the most proficient. This finding is plausibly related to the fact that our gymnasium learners in the second year of upper secondary school were closer than the younger learners to taking their high-stakes school-leaving examinations.

Theoretical Implications

Previous studies have led to several different hypotheses about the nature of problems with reading an SFL. One hypothesis is that FL reading is more of a reading problem (including weakness in L1 reading) than an FL language problem. Conversely, problems in FL reading may be simply a problem of FL knowledge. Earlier we considered two other hypotheses—the Threshold Hypothesis (Alderson, 1984) and the Linguistic Coding Differences Hypothesis (Sparks et al., 1989). According to our findings, FL knowledge was the best separator of weak and strong FL readers. Much bigger effect sizes were discovered for the FL skills, sometimes exceeding d = 5.00 (see Appendix E), than for the L1 and cognitive skills. These findings, then, support the Threshold Hypothesis more than the LCDH and are in line with the conclusion of the meta-analysis by Melby–Lervåg and Lervåg (2014) that training L2 skills is the best treatment for most learners with L2 reading problems (even if their operationalisation of L2 skills is oral vocabulary and oral comprehension more generally).

However, our study demonstrated that certain aspects of cognition (usually based both on L1 and FL tasks) and L1 skills also separated weak and strong FL readers, albeit often only with moderate effect sizes. As reported earlier, some studies reported in, for example, Sparks & Ganschow (1993a) suggest that differences in L1 cognitive and linguistic skills can explain differences in FL skills to some extent. Research on dyslexia (e.g., Helland & Morken, 2015) has demonstrated how important basic cognitive skills are in L1 reading.

Melby–Lervåg and Lervåg (2014) suggested that their recommendation of focusing on improving weak L2 readers’ proficiency in L2 may not apply to learners with specific decoding problems. It may be, therefore, that the two theories (the Threshold Hypothesis and the LCDH) apply to somewhat different kinds of learners. It might be that the Threshold Hypothesis is most appropriate in explaining FL reading among the general population of learners, most of whom are without specific reading problems, as was the case in our study, and that the LCDH is most appropriate for at-risk learners. A full investigation of this would probably require the inclusion of a fairly large number of at-risk L1 readers and a comprehensive measurement of their cognitive, L1 and FL skills.

However, there are several caveats to this speculation. One is that previous research has looked at L2 learners acquiring the language in contexts where they are surrounded by the L2, which very likely facilitates learning. Our study, however, has investigated the issue in an FL setting where learners are not surrounded by the FL, even if the language in question (English) has some presence in many youngsters’ lives in a country such as Finland.

Second, previous studies have not focused on age differences, whereas we have. Furthermore, our learners were arguably still developing their cognitive skills in their L1 and in their FL. Moreover, although we have discovered that weak FL readers are clearly weaker in a range of FL skills than stronger readers, in other words, that FL reading problems are more of an FL proficiency problem than a problem of L1 reading abilities, we do not know whether even the stronger readers have reached a threshold of FL proficiency that might allow their L1 reading abilities (or strategies) to transfer to their FL reading.

Importantly, we have reported on a cross-sectional study, not a longitudinal one, although we have also conducted a longitudinal study, which many earlier SFL reading researchers have not. We may eventually find evidence of individual development over time that reveals an FL threshold beyond which L1 reading abilities can transfer to FL reading. But in this study, we cannot claim that a threshold has been reached that is associated with increasing age.

Implications for Further Research

Specific implications from our results for further research on diagnosing SFL reading are that the instruments used in research, the characteristics of the learners, and the context in which learning is taking place should be elaborated more carefully than has usually been the case. The nature and quality of the data-gathering instruments need serious attention.

First, the FL reading construct should be defined and operationalised carefully, since diagnosing reading for synthesising information, for example, may be different from diagnosing skimming and scanning, inferencing, or other component skills. Specifically we have not referred to or used a number of theories of comprehension developed in L1 reading research, like that of Stanovich's (1980) Interactive–Compensatory Model, or Kintsch's (2004) ideas on discourse processing and his Construction–Integration Model of text comprehension, or the role of background knowledge and long-term working memory in comprehension.

Second, the selection and operationalisation of the predictors is important. Previous research, including ours, has covered a range of potential predictors but there are still gaps such as orthographic and morphological knowledge of the SFL (see Aro & Wimmer, 2003; Jeon & Yamashita, 2014). How particular predictors are operationalised also matters. For example, vocabulary size is already a well-known predictor of SFL reading ability, but it remains to be seen whether improving weak readers’ depth or diversity of vocabulary and vocabulary learning strategies, for instance, will increase their reading ability.

In addition, the language through which the instruments are presented is an important consideration as it is very difficult to find tasks that do not involve any degree of language knowledge.

A broad enough range of both dependent and independent variables should be used in diagnostic research, unless there are good reasons, based on theory and previous studies, to limit their range. Even our study with its wide coverage of skills was limited, for practical reasons, in what was measured compared with what could ideally have been measured.

A related study looking at the relation between learners’ background characteristics and SFL reading (Huhta et al., 2016) has suggested that very few background variables are able to predict performance in SFL reading, but it may be that some such variables are useful diagnostically. Particularly interesting is the amount of SFL reading or other use of the SFL in individual learners’ free time. L1 reading research has found that primary school students’ reading frequency was the best predictor not only of performance on L1 reading tests but also the student's growth as a reader from the second to the fifth grade (Anderson, Wilson, & Fielding, 1998). Since free-time use of an SFL is clearly variable, it may be a useful factor to manipulate when attempting to improve learners’ SFL reading skills. More generally, Huhta et al. (2016) point out the need to consider the characteristics of the learner and the context more carefully, as the diagnosis—and possible treatment—of learners of different ages, proficiency levels, and contexts may differ. In particular, we need research in different L2 and FL contexts and across different L1 and SFL pairs than were studied in this research. Finnish has a transparent orthography and English an opaque one (cf. Jeon & Yamashita's [2014] previously mentioned point). Different L1/SFL pairs might well give different results.

Finally, the current study was cross-sectional, and while we argue that it has shed new light on the role of different factors that are characteristic of weak FL readers, in contrast to strong readers, longitudinal and intervention studies are also needed to confirm and complement the present results on how development in FL reading relates to cognitive, linguistic, and motivational characteristics and which of such characteristics are important simply for predicting or identifying problems in development and which are also useful for treatment purposes.