In second language acquisition and applied linguistics research, few topics have attracted the attention of researchers as much as anxiety. Horwitz et al. (Reference Horwitz, Horwitz and Cope1986) first differentiated foreign language anxiety (FLA) from other learning anxieties and conceptualized it as a unique and situation-specific construct referring to the tension and apprehension experienced during the foreign language learning process. Since then, studies that aim to explore FLA in their contexts have mushroomed, with initial work demonstrating a negative effect of FLA on language learners’ performance or achievement (Aida, Reference Aida1994; Koch & Terrell, Reference Koch, Terrell, Horwitz and Young1991; Young, Reference Young1990). Most studies sought to determine whether FLA hinders the language learning process and define its role in influencing student success. This almost forty-year research journey has shown that anxiety (a) is an important psychological construct in language classrooms; (b) has a negative role in second/foreign language learning (Teimouri et al., Reference Teimouri, Goetze and Plonsky2019); and (c) its effect should not be ignored regardless of learners’ proficiency levels (Zhang, Reference Zhang2019).
Though FLA, as an umbrella term referring to nervousness in foreign language classrooms, was initially associated with speaking skills, the late 1990s and early 2000s witnessed an emergence of research interest in receptive skills, resulting in two relatively independent concepts from general FLA. Foreign language reading anxiety (FLRA) is defined as “a specific anxiety type distinguishable from the more general types of foreign language anxiety that have been linked to oral performance” (Saito et al., Reference Saito, Horwitz and Garza1999, p. 215). In contrast, foreign language listening anxiety (FLLA) is associated with fear, nervousness, worry, and lack of confidence in foreign language listening (Kim, Reference Kim2000). Development of scales by Saito et al. (Reference Saito, Horwitz and Garza1999) for reading and by Kim (Reference Kim2000) and Elkhafaifi (Reference Elkhafaifi2005) for listening has particularly made the investigation of those phenomena more practical for researchers. With the help of these instruments, scholars interested in anxiety research presented a descriptive picture of anxiety types in line with demographic variables and their relationship with general FLA. For instance, Wu (Reference Wu2011) found that FLRA and FLA are two distinct phenomena, and students’ performance in foreign language reading comprehension does not change according to their FLA. On the other hand, Liu and Yuan (Reference Liu and Yuan2021) tracked the changes in FLA and FLLA over a semester among Chinese EFL learners. The findings revealed high levels of both anxiety types among students throughout the semester and a high correlation between them. It was also concluded that these anxiety types predicted the learners’ self-rated listening and speaking English proficiency.
At the beginning of the 2010s, scholars also began to embrace a different research perspective focusing on exploring the relationship between FLA and skill-based anxieties through advanced statistical models. For instance, Pae (Reference Pae2013) tested the intrarelationship between skill-based anxieties and FLA and found that each anxiety type was a distinct and distinguishable construct. All skill-specific anxieties contributed independently to FLA. Recently, in a similar vein, Ran et al. (Reference Ran, Wang and Zhu2022) compared FLA in relation to four language skills and found that Chinese EFL learners’ FLA was mainly driven by listening anxiety. In another recent study, Öztürk et al. (Reference Öztürk, Şahin, Ölmezer Öztürk and Elmas2022) demonstrated a high and significant correlation between all skill-based anxieties. They also concluded that the short forms of four skill-based anxiety scales can be used “to measure a single latent variable called foreign language learning anxiety and this single variable explained the 88% of foreign language classroom anxiety by the FLCAS, a standard instrument in foreign language anxiety measurement” (p. 356). Taking the research questions, methods, and statistical procedures followed in those and similar studies into consideration, it would not be wrong to say that the descriptive perspective in anxiety research is saturated.
The Current Study
FLA has already received much attention as a variable in SLA research, embracing the aforementioned research trend with advanced statistical models and procedures, especially those using mediation or moderation analysis. For instance, aiming to explore the relationship between FLA, emotional intelligence (EI), and English proficiency, Shao et al. (Reference Shao, Yu and Ji2013) found that FLA partially mediates the relationship between EI and proficiency. Similarly, Fallah (Reference Fallah2017) examined the relationship between mindfulness, coping self-efficacy, and FLA among Iranian EFL learners and found that coping self-efficacy significantly mediates the relationship between FLA and mindfulness. In a recent study, Feng and Hong (Reference Feng and Hong2022) investigated the relationship between behavioral engagement, FLA, enjoyment, and self-reported achievement in the Chinese EFL context. The findings demonstrated a significant mediating effect of engagement between enjoyment and FLA.
In sum, FLA has found its place in the current research trend focusing on its mediating/moderating effect on the relationship among other psychological variables such as enjoyment, mindfulness, and self-efficacy. However, to the best knowledge of the researcher, no studies so far have focused on a potential mediating effect of foreign language classroom anxiety (FLCA) on skill-based anxieties. Such a relationship between FLCA and skill-based anxieties might be a new territory in anxiety research to be explored. As the first step in this exploration of anxiety research, and based on previously reported significant correlations between FLA and anxiety in receptive skills (Çapan & Karaca, Reference Çapan and Karaca2013; Chow et al., Reference Chow, Chiu and Wong2018; Liu & Yuan, Reference Liu and Yuan2021; Merç, Reference Merç, Arabski and Wojtaszek2009), this study aims to examine the relationship between foreign language classroom anxiety (FLCA), foreign language reading anxiety (FLRA), and foreign language listening anxiety (FLLA) and to test the mediating effect of FLCA on the relationship between FLRA and FLLA. The following research questions are addressed throughout the study:
1. What is the level of FLCA, FLRA, and FLLA in Turkish EFL learners?
2. To what extent are FLCA, FLRA, and FLLA related?
3. To what extent does FLCA mediate the relationship between FLRA and FLLA?
Methodology
Research Context and Participants
In Turkey, English preparatory schools present one-year intensive English language teaching programs for freshman students at the tertiary level. After initial placement via proficiency tests, students are placed in different groups based on their proficiency level and receive highly intensive instruction and assessment throughout the academic year. The main purpose of these programs is to help them gain English language skills that might help in their subsequent programs of study. The data for this study was collected from such English-intensive programs in four public universities at the beginning of the 2022 fall semester through an online survey. The participants included 341 students—145 of whom were male and 196 were female. Their proficiency levels, determined by institutional exams using the CEFR scale, ranged from A1 to B2.
Data Collection Tools
Three scales were used to collect the data in the current study: (a) the Foreign Language Classroom Anxiety Scale (FLCAS); (b) the Foreign Language Reading Anxiety Scale (FLRAS); and (c) the Foreign Language Listening Anxiety Scale (FLLAS). The FLCAS, probably the most widely used scale in anxiety research, was developed by Horwitz et al. (Reference Horwitz, Horwitz and Cope1986) as a self-report instrument with thirty-three items. The second instrument, FLRAS, was developed as a 5-graded Likert scale by Saito et al. (Reference Saito, Horwitz and Garza1999) to measure reading anxiety in foreign language classrooms, and the third instrument, the thirty-three-item FLLAS, was developed by Kim (Reference Kim2000).
These scales were translated into Turkish by several prior studies to investigate anxiety in Turkish EFL contexts. During the data collection process, these translated versions were used: FLCAS by Aydın (Reference Aydın1999), FLRAS by Kuru Gönen (Reference Gönen and İ2005), and FLLAS by Kılıç (2007). Since the factorial structures of FLCAS, FLRAS, and FLLAS were not fully revealed in the Turkish adaptation studies, and validation of the scales was important for further mediation analysis, exploratory factor analysis (EFA) was used to assess the construct validity of the instruments. First, whether the data set met the assumptions of EFA was examined. In this regard, missing data, multicollinearity, multivariate normality, multivariate extreme values, Bartlett's test of sphericity, and Kaiser-Meyer-Olkin (KMO) values were calculated for each instrument. The following presents the detailed EFA analysis for each scale used in the current study.
Foreign Language Classroom Anxiety Scale (FLCAS)
The data set in the scale was 5-graded, thus, EFA was performed using the polychoric correlation matrix. First, the KMO value and Bartlett's test of sphericity were examined. The KMO value was 0.92, suitable for factorization (Kaiser, Reference Kaiser1970), and Bartlett's test of sphericity was statistically significant (χ2 (351) = 3640, p < 0.05). The minimum average partial (MAP) test developed by Velicer (Reference Velicer1976) and parallel analysis methods proposed by Horn (Reference Horn1965) were used to decide the number of factors in EFA, and two dimensions were suggested. Examination of the explained variance rates showed that the eigenvalue of the first factor was 11.34, explaining 34.37% of the total variance, and the eigenvalue of the second factor was 2.26, with 6.87% of the variance explained by this factor. The two factors together explained 41.25% of the total variance. Considering the parallel analysis, MAP analysis, and explained variance rates, it was seen that the two-dimensional structure would be simpler and more understandable.
Since the number of factors was determined as two, the Simplimax (Kiers, Reference Kiers1994) method was used as the factor rotation method because it overcomes more complex situations (Lorenzo-Seva, Reference Lorenzo-Seva2000). As a result of the EFA performed with these features, it was observed that an item (V7) was cross-loaded, and some items (V10, V11, V21) had factor loads below 0.30 (Costello & Osborne, Reference Costello and Osborne2005). Thus, these items were excluded from the analysis one by one, and the EFA was repeated. In this analysis, since the factor loads of some more items (V22, V25) were below 0.30, they were also excluded from the analysis one by one, and the EFA was repeated. At the end of this process, expert opinions from three scholars in the field of SLA research were received regarding the items that should be removed, and they agreed that it would not harm the scale's content validity. For this reason, six items in total were excluded from the scale, and EFA was repeated with twenty-seven items. In the two-dimensional structure obtained from EFA, the eigenvalue of the first factor was found to be 10.42 and explained 38.60% of the total variance, and the eigenvalue of the second factor was 2.12 and the variance rate explained was 7.87%. The two factors together explain 46.48% of the total variance. The interfactor correlation is 0.25. The structure obtained as a result of EFA is presented in Table 1.
Table 1. EFA Results of FLCAS
The EFA analysis of FLCAS showed that it had sufficient construct validity within the scope of this study. The Cronbach's alpha and McDonald's omega coefficients were also calculated for reliability analysis. Cronbach's alpha was 0.59 for factor 1 and 0.93 for factor 2. McDonald's omega coefficients were 0.64 and 0.94, respectively. Finally, for the whole scale, Cronbach's alpha coefficient was 0.91, McDonald's omega coefficient was 0.93, and the stratified alpha coefficient was 0.92.
Foreign Language Listening Anxiety Scale (FLLAS)
Just as in FLCAS, similar statistical procedures were followed to validate the scale, and EFA was performed using the polychoric correlation matrix. First, the KMO value was found as 0.93, meaning suitability for factorization, and Bartlett's test of sphericity was statistically significant (χ2 (496) = 3609.3, p < 0.05). Based on the parallel analysis, MAP analysis, and explained variance rates, a two-dimensional factorial structure was obtained. Thus, the Promin method was used as the factor rotation method. As a result of the EFA performed with these features, it was observed that one item (V17) was cross-loaded, so it was removed from the analysis, and the EFA was repeated. In the two-dimensional structure as a result of EFA, the eigenvalue of the first factor was 13.62 and explained 42.58% of the total variance. The eigenvalue of the second factor was 2.24, explaining 7.02% of the variance. Together, the two factors explain 49.60% of the total variance, as presented in Table 2. The interfactor correlation is 0.77.
Table 2. EFA Results of FLLAS
The EFA results showed that FLLAS had sufficient construct validity. Cronbach's alpha and McDonald's omega coefficients were also calculated for reliability analysis. Cronbach's alpha was 0.90 for factor 1 and 0.91 for factor 2. Moreover, McDonald's omega coefficients were 0.92 and 0.93. Cronbach's alpha coefficient for the whole scale was 0.94, McDonald's omega coefficient was 0.93, and the stratified alpha coefficient was 0.95, all referring to a high level of reliability.
Foreign Language Reading Anxiety Scale (FLRAS)
After the same initial procedures performed in FLCAS and FLLAS, KMO was found as 0.91 and Bartlett's test of sphericity was significant (χ2 (136) = 3438.1, p < 0.05), and the data set was suitable for factorization. The minimum average partial (MAP) test and parallel analysis methods were used to determine the number of factors, and both proposed a two-factor structure. Since the number of dimensions was determined as two, the Promin method was used as the factor rotation method, as performed in the other two scales. The EFA performed with these features showed that the factor load of the eighteenth item was below 0.30 (Costello & Osborne, Reference Costello and Osborne2005), so it was excluded from the analysis, and the EFA was repeated. In the two-dimensional EFA structure, the first factor's eigenvalue was 7.81, explaining 45.94% of the total variance. The eigenvalue of the second factor was 1.95 and the variance rate explained is 11.50%. The two factors explain 57.44% of the total variance, as seen in Table 3. The interfactor correlation is 0.56.
Table 3. EFA Results of FLRAS
FLRAS had sufficient construct validity for this study as well. As for reliability, Cronbach's alpha was 0.83 for factor 1 and 0.90 for factor 2. McDonald's omega coefficients were 0.87 and 0.92. For the whole scale, Cronbach's alpha coefficient was 0.90, McDonald's omega coefficient was 0.92, and the stratified alpha coefficient was 0.92.
Data Analysis
To find out the level of FLCA, FLRA, and FLLA among Turkish EFL learners, descriptive statistics were calculated, and the relationship between these constructs was analyzed through Pearson correlations. In addition, the Factor program (Lorenzo-Seva & Ferrando, Reference Lorenzo-Seva and Ferrando2022) was used to calculate the EFA, parallel analysis, MAP analysis, and McDonald's omega coefficient, which were used in collecting evidence for the construct validity of the scales. For EFA, the codes written by Kılıç (Reference Kılıç2020) were used to check the assumptions. EFA was performed using the polychoric correlation matrix and ULS as the factoring method. Cronbach's alpha coefficient was calculated in the psych package (Revelle, Reference Revelle2022) in the R software, and the stratified alpha coefficient in the sirt package (Robitzsch, Reference Robitzsch2021).
The mediation relationship between the constructs was examined by PROCESS (v4.2) macro written by Hayes (Reference Hayes2022). Bootstrap confidence intervals were created in the PROCESS, and the analysis was conducted according to these confidence intervals. Five thousand samples were made for Bootstrap. Examining whether the estimations were statistically significant using Bootstrap confidence intervals was conducted to determine whether these intervals contained zero. If the bootstrap confidence interval did not contain zero, it was concluded that the estimation was statistically significant (MacKinnon, Reference MacKinnon2008). As for the assumptions of the mediation analysis, the dataset was examined in terms of linearity, lack of correlation of residuals, multicollinearity, homoscedasticity, and outliers. The scatter-dot graph showed that the relationships between the variables were linear. Since the Durbin-Watson statistic was 1.83, the assumption of uncorrelated residuals was met. Since the TV value was 0.54, the VIF value was 1.86, and the CI value ranged between 1 and 12.45, there was no multicollinearity problem (Kline, Reference Kline2011; Tabachnick & Fidell, Reference Tabachnick and Fidell2019). The Regression Studentized Residual and Regression Standardized Predicted Value scatter plot indicated that the homoscedasticity assumption was held.
Findings
Anxiety Levels and Correlations
The first research question aimed to determine the FLCA, FLRA, and FLLA levels among the participants and the relationship between these constructs. The descriptive statistics for the FLCAS, FLLAS, and FLRAS variables and the Pearson correlations between these anxiety types are presented in Table 4.
Table 4. Descriptive Statistics and Correlations Between FLCA, FLLA, and FLRA
** p < 0.01
Based on the skewness and kurtosis values indicated in the table, the data in all scales were normally distributed (Tabachnick & Fidell, Reference Tabachnick and Fidell2019). The mean scores derived from the descriptive analysis demonstrate that participants experienced moderate levels of anxiety in all types (FLCA, M = 2.69; FLLA, M = 2.91; FLRA, M = 2.69). Moreover, the correlation coefficients in the table reveal that FLCA, FLLA, and FLRA were significantly correlated, and the level of this correlation is high according to L2 field-specific benchmarks for effect sizes (small, r = .25; medium, r = .40; large, r = .60) (Plonsky & Oswald, Reference Plonsky and Oswald2014). The highest correlation (0.76) was found between FLRA and FLLA.
Results of the Mediation Analysis
The analysis was performed with the mean scores obtained from the scales to explore the mediating effect of FLCA on the relationship between FLRA and FLLA. Within this study's scope, the scales’ structures were examined. All were two-dimensional and the correlations between the dimensions were 0.56, 0.77, and 0.25 for FLRA, FLLA, and FLCA, respectively. As the correlation between the dimensions increases, the scores obtained from the two dimensions become additive (Cohen et al., Reference Cohen, Swerdlik and Sturman2013). For this reason, considering the correlations between the factors in the scales, mediation analyses were carried out with the mean scores. The standardized coefficients obtained from the model as a result of the mediation analysis are shown in Figure 1. The nonstandardized coefficients obtained from the analysis are presented in Table 5.
Figure 1. Standardized Coefficients obtained from the model.
Table 5. Unstandardized Coefficients and 95% Bootstrap Confidence Intervals (BCI)
** p < 0.01
Table 5 shows that the FLRA predicted the mediating variable FLCA at a statistically significant level (β = 0.62, p < 0.01). This variable explains 47% of the variability in the FLCA. Similarly, the outcome variable FLLA is statistically predicted by FLRA (β = 0.47, p < 0.01) and FLCA (β = 0.37, p < 0.01) at a significant level, and these two variables explain 64% of the variability in FLLA. Direct and indirect effects and standardized coefficients are presented in Table 6.
Table 6. Direct and Indirect Effect of FLRA on FLLA and its 95% Confidence Interval
**p < 0.01
When the mediator variable in the effect of FLRA on FLLA was added to the model, the overall effect of FLRA on FLLA (0.70) decreased (0.47). However, it was still statistically significant (β = 0.47, BLCI = 0.39, BUCI = 0.55). Therefore, the FLCA partially mediated the effect of FLRA on FLLA. Examining the indirect effects shows that the indirect effect of FLRA on FLLA over FLCA is significant (β = 0.23, BLCI = 0.16, BUCI = 0.30). Directly or indirectly, FLRA significantly predicted a moderate increase in FLLA (β = 0.75, p < .001). All together, FLRA and FLCA explained 64% of the variance in the FLLA. In addition, it was found that FLCA explained 33% (0.25/0.75) of the total effect of FLRA on FLLA.
Discussion and Conclusion
The main focus of this study was to examine the extent to which FLCA mediates the relationship between foreign language reading (FLRA) and listening (FLLA) anxieties. In addition, participants’ anxiety levels in all types and the correlation between them were investigated. The findings show that the participants experienced moderate levels of FLCA, FLRA, and FLLA in their language learning process, in line with many other studies demonstrating moderate or high levels of anxiety among EFL learners, not only in the Turkish context (Akar, Reference Akar2021; Aydın, et al., Reference Aydın, Harputlu, Çelik, Uştuk and Güzel2018; Özer & Altay, Reference Özer and Altay2021) but also in other EFL contexts (Gawi, Reference Gawi2020; Su, Reference Su2022; Toubot et al., Reference Toubot, Seng and Azizah2018). Furthermore, FLCA, FLRA, and FLLA were found to be positively and highly correlated, with the strongest correlation between FLLA and FLRA. This means that if one of those anxiety types increases or decreases, it is quite probable that the other will also increase or decrease. These findings are parallel to those of Merç (Reference Merç, Arabski and Wojtaszek2009), who also found a significantly positive relationship between FLRA and FLLA. Similarly, in a recent study in the Chinese EFL context, Liu and Yuan (Reference Liu and Yuan2021) demonstrated a positive high correlation between FLCA and FLLA. For this reason, the descriptive findings of the current study provide more evidence for the moderate level of anxiety among EFL learners and positive correlations between foreign language anxiety and reading and listening anxieties in the foreign language learning process.
More importantly, in the mediation model tested in the current study, reading anxiety significantly predicted classroom anxiety, and listening anxiety (FLLA) was predicted by both classroom anxiety (FLCA) and reading anxiety (FLRA). The mediation analysis revealed that FLCA partially mediates the relationship between FLRA and FLLA, which means the effect of FLRA on FLLA can be partly explained by foreign language classroom anxiety. In other words, an increase or decrease in the level of FLCA, as the “umbrella” anxiety, might influence the level of other skill-based anxieties and the strength of the relationship among them. This partial mediating effect and explanatory power of FLCA, even on the relationship between two anxiety types, can lead to a rediscussion of studying and conceptualizing skill-based anxieties in foreign language classrooms. Particularly after the validation and development of tools with the capacity to measure each skill-based anxiety during the early 2000s, there has been a growing interest in investigating skill-based anxieties as separate and independent constructs. Although this skill-based perspective has made unique contributions to our understanding of anxious learners’ feelings, they distract attention from anxiety as a holistic psychological construct in foreign language classrooms. For this reason, together with some relatively recent studies demonstrating a single construct covering skill-based anxieties (Kutuk et al., Reference Kutuk, Putwain, Kaye and Garrett2020; Öztürk et al., Reference Öztürk, Şahin, Ölmezer Öztürk and Elmas2022; Pae, Reference Pae2013), the finding of the current study demonstrating the mediating power of an umbrella-type of anxiety (FLCA, in this case) might provide a rationale for researching anxiety with “a ring that binds them all” perspective. This argument suggests new research opportunities in the field. As discussed above, FLCA has already received much attention as a moderator/mediator variable in many studies examining the relationship among psychological variables. However, such a perspective is currently missing in anxiety research. This study investigated the mediating effect of FLCA on the relationship between reading and listening anxieties. Further studies might focus on the moderating role of FLCA on receptive skills. Additionally, examining both its mediating and moderating roles on anxiety in productive skills might be another research path for scholars in the field. Finally, accepting that the literature no longer needs to present a sole description of how anxious learners feel in language classrooms, structural and factorial investigations of anxieties in various contexts will potentially contribute more to whether foreign language anxiety should be handled as a single psychological construct and will lead us to consider whether or not to embrace a back-to-the-past tendency to research and conceptualize anxiety as in the 1980s.
