Professional Documents
Culture Documents
Applied Linguistics-2013-Pae-232-52
Applied Linguistics-2013-Pae-232-52
TAE-IL PAE
Department of English Language Education, Yeungnam University, South Korea
E-mail: paet@ynu.ac.kr
INTRODUCTION
This article is concerned with the issue of the relationships between individual
skill-based foreign language (FL) anxieties. So far a lot of studies have been
conducted to examine FL anxiety, and existing research findings indicate that
anxiety has detrimental influence on the processes of FL learning since FL
learning is inherently a learning activity wherein learners are forced to com-
municate with others using ‘an imperfectly mastered second language’ (Saito
et al. 1999: 202). It has been suggested that anxiety interferes with FL learning
(Horwitz et al. 1986; MacIntyre and Gardner 1989; Price 1991; Saito and
Samimy 1996), and anxious FL learners tend to show physiological as well
as psycholinguistic symptoms.
Recently, research in FL anxiety has extended to the examination of more
language-skill-specific anxieties. For example, Cheng et al. (1999) explored the
relationships between general L2 classroom anxiety and more skill-specific L2
writing anxiety for a sample of 433 Taiwanese university learners of English as a
FL (EFL). Based on both factor-analytic and correlation procedures, they
demonstrated that writing anxiety as measured by the second language version
T.-I. PAE 233
of Daly and Miller’s Writing Apprehension Scale (Daly and Miller 1975) was a
related but independent construct clearly distinguishable from more general FL
classroom anxiety, which was operationally defined by the Foreign Language
Classroom Anxiety Scale (FLCAS): while FLCAS was found to be a general
measure of FL classroom anxiety with a strong focus on speaking anxiety, writ-
ing anxiety was found to be a skill-based anxiety associated with writing skills.
The two constructs, however, shared low self-confidence as a common compo-
nent. Taking the research one step further, Cheng (2002) analyzed factors af-
fecting FL writing anxiety using 165 Taiwanese EFL students. The results
showed that the self-perception of L2 writing competence was the most power-
and investigated the psychometric properties of FLLAS using 312 Korean EFL
university students. To examine the construct validity of FLLAS, Kim corre-
lated FLLAS with different scales of anxiety (e.g. Trait Anxiety Inventory,
FLRAS, and FLCAS). The results of a series of correlation analyses identified
a Pearson correlation coefficient of 0.53 between FLLAS and FLRAS, 0.70 be-
tween FLLAS and FLCAS, and 0.33 between FLLAS and the Trait Anxiety
Inventory. The low correlation (r = 0.33) between FLLAS and the Trait
Anxiety Inventory (Spielberger 1983) reveals that a situation-specific language
anxiety construct, such as FL listening anxiety, is quite separable from general
personality-type anxiety. The higher correlation between FLLAS and FLCAS
METHOD
Participants
A total of 285 Korean university EFL students participated in the present study.
Participants were sampled from diverse academic areas, such as English,
Business, Political Science, Engineering, Hard Sciences, and Arts. The female
participants comprised 58 per cent, and the age range was from 18 to 39 years
(M = 22.1, SD = 3.13). More than half (53 per cent) of the participants were
first-year students. Upon close investigation of the students’ data, however,
missing values were found for 56 cases. A list-wise deletion of the missing
values resulted in the final sample of 229 students. In South Korea, English
is a compulsory school subject from grades 3 to 12. At the time of the data
collection, the university, from which the participants were recruited, required
all of its students to take General English Courses for at least two semesters or
more to be eligible for graduation. This suggests that all the study participants
had been learning English for at least 11 years.
Procedure
Initially, the author contacted individual instructors in charge of the General
English Courses (e.g. English Reading, English Conversation, and English
Writing), which emphasized a balanced development of English competence
236 SKILL-BASED L2 ANXIETIES REVISITED
in the four skill areas, and explained the study’s purpose. With the permission
from the individual instructors, the author visited individual classrooms and
invited students to voluntarily participate. All the students were instructed
about the study’s purpose, and all of them chose to participate in the present
study. They received detailed information about how to answer the question-
naire used. The data collection lasted for about 50 minutes. All the participat-
ing students were presented with a pre-paid lunch ticket for use in the
university cafeteria for their cooperation.
Instrumentation
students must have been familiar with English letters and symbols. Item ana-
lysis of the remaining 18 FLRAS items indicated that 5 of the 18 had very poor,
and even negative, item-total correlations. Therefore, these items were also
removed. Internal consistency evidence as measured by Cronbach’s a was 0.85
with the final 13 items.
Competing models
To analyze the exact relationships among the four skill-based anxieties, a set of
competing models were developed, and the adequacy of each model was stat-
istically evaluated.
Model 4 (one-factor)
This was the most restrictive model and assumed that a single general anxiety
factor can account for all the inter-correlations among the four skill-based
anxieties.
Hypotheses
Since Research Question 1 requires a series of model-testing procedures, three
hypotheses were generated to investigate the statistical relationship among the
four skill-based anxieties. Hypothesis 1 posited that the three-factor models
(i.e. Models 2A–2D), which were more constrained than the four-factor model
(i.e. Baseline Model), would not be significantly worse than the Baseline
Model (Model 1), indicating that the combination of two skill-specific anxieties
into one general anxiety factor (e.g. Oral, Written, Productive, or Receptive
Anxiety Factor) would be supported by the data. This hypothesis suggests that
the two skill-specific anxieties subsumed under a more general anxiety factor
(e.g. Oral Anxiety Factor) are not statistically different from each other.
240 SKILL-BASED L2 ANXIETIES REVISITED
Analyses
The relationships among the four skill-based anxieties (i.e. Research Question
1) were analyzed under the Confirmatory Factor Analysis (CFA) framework.
In this regard, a series of CFA models (i.e. Models 1–4) were developed in order
to test the formulated hypotheses. To facilitate the hypothesis testing, the
model data fit of each CFA model, as measured by chi-square, was compared
with that of the baseline model using a series of chi-square difference tests
(Anderson and Gerbing 1988; Hsiao and Oxford 2002; Pae 2008).
A chi-square difference test is based on the logic that the difference in
chi-square values between two nested models is distributed as chi-square
with degrees of freedom equal to the difference in degrees of freedom for
the two nested models (Anderson and Gerbing 1988; Reise et al. 1993).
Hence, a significant increase in chi-square difference value between two
nested models suggests that the fit of the nested model is significantly worse
than the baseline model. For example, if Model 2A, which combined speaking
and listening anxieties into one general anxiety factor (i.e. Oral Anxiety) but
kept reading and writing anxieties independent, showed a significant
chi-square increase from the baseline model at the given degrees of freedom
difference, this would demonstrate that the combination of listening and
speaking anxieties significantly worsens the fit of Model 2A compared with
the baseline model, thereby verifying that listening anxiety and speaking anx-
iety are statistically independent.
Because chi-square values tend to be inflated in a large sample size, three
supplementary fit statistics [Comparative Fit Index (CFI), Non-Normed Fit
Index (NNFI), and Normed Fit Index (NFI)], which are widely used in the
context of nested model comparison, were also reported (Schumacker and
Lomax 2004). Incremental fit indices, such as CFI, NNFI, and NFI evaluate
the fit of the proposed model in reference to the baseline model. Values of
these three fit indices range from 0 to 1, and a value of 0.9 or above generally
suggests an acceptable model data fit (Bentler and Bonett 1980). All the par-
ameters in the CFA models were calibrated through maximum likelihood
T.-I. PAE 241
estimation procedure via LISREL 8.5 program. Input data were provided to the
LISREL program in the form of covariance matrix.
Since the CFA models used in the present study involved a great number of
observed variables (83 skill-based anxiety items), a data reduction procedure
was performed based on exploratory factor analysis (EFA). To determine an
optimum number of factors for each anxiety scale, an Eigenvalue of one or
above served as the criterion. The number of factors for each anxiety scale was
further verified through CFA procedure by examining the significance of the
chi-square difference value between the factor model indicated by
the Eigenvalue criterion and a more parsimonious factor model. Once the
RESULTS
Correlation analysis
Table 1 presents the inter-correlation matrix among the five scale scores. Since
most of previous studies about L2 anxiety were conducted based on correlation
analyses, this matrix was expected to provide comparable data in reference to
the previous findings reported in the literature. The inter-correlation matrix
indicated that all of the five aggregate scores were significantly correlated with
each other.
1. FLCAS 1.00
2. LAS 0.77* 1.00
3. SAS 0.71* 0.63* 1.00
4. RAS 0.66* 0.65* 0.48* 1.00
5. WAS 0.61* 0.55* 0.62* 0.57* 1.00
*p < 0.01.
presents the results of the dimensionality check for each of the four skill-based
anxiety scales.
In the LAS, the initial EFA procedure identified four factors showing an
Eigenvalue of one or above. The fit of the four-factor model was compared
with that of a three-factor model to see if a more parsimonious model (i.e.
three-factor model) would represent the sample covariance of the 32 LAS
items better than the four-factor model does. Comparison of the model-data
fit between the two nested models (i.e. four-factor versus three-factor so-
lution) was made via chi-square difference tests, as explained in the ana-
lyses section. The three-factor model produced a chi-square value of
1419.93 with 461 degrees of freedom, whereas the chi-square value of
the four-factor model was 1420.02 with 458 degrees of freedom. The dif-
ference in the chi-square value between the two nested models was not
significant [2 (3) = 0.09, p > 0.05], indicating that a four-factor model was
not significantly better than a three-factor model. Therefore, a three-factor
solution was targeted for the LAS. The three factors accounted for 50.6 per
cent of the total variance. In interpreting a factor, if an item showed a
primary loading of 0.50 or above and secondary loadings (i.e. cross loadings)
of 0.25 or below, then this item was considered to be a member of a factor,
as consistent with previous research (e.g. Cheng et al. 1999). With regard to
the first factor of the LAS, 10 items were found to meet the factor criteria
and a majority of these items seem to have in common a sense of being
nervous and worried in English listening. Therefore, the first factor was
named Nervousness and Worry in Listening English (Items 2, 5, 7, 16,
17, 18, 21, 23, 30, and 33). The second factor received strong loadings
from six items and this factor appears to reflect the difficulties associated
with English listening. This factor was labeled Difficulties in Listening
English (Items 11, 12, 13, 19, 22, and 32). Three items were included in
the third factor, and this factor was assigned the label of Self-Confidence in
Listening English (Items 14, 25, and 31).
T.-I. PAE 243
The same analytical procedure was applied to the verification of factor struc-
ture for the remaining skill-based anxieties. In the SAS, EFA with a total of 12
items indicated a two-factor structure. Again, the fit of the two-factor model
was compared with that of a one-factor model through the CFA. The
one-factor model resulted in a chi-square value of 469.33 with 54 degrees of
freedom, compared with 310.62 with 53 degrees of freedom in the two-factor
model. The chi-square difference between the two models was significant [2
(1) = 158.71, p < 0.05], suggesting that a two-factor solution explains the
sample covariance of the 12 SAS items significantly better than a one-factor
solution does. Hence, a two-factor solution was selected for the SAS. The two
and most of these items tapped lack of confidence in writing English. This
factor was named Low Self-Confidence in Writing English (Items 16, 21, 22,
and 24). Finally, the third factor was defined by four items, and these items
seem to measure the extent to which learners avoid writing English. This factor
was labeled Reluctance toward Writing English (Items 1, 4, 5, and 8).
correspondence between the indicators and their related factors in the Baseline
Model.
Table 3 provides information about the degree to which constructs with
dissimilar theoretical orientations differ from each other (i.e. discriminant val-
idity). In order to evaluate the evidence of discriminant validity among the
four anxiety factors, AVE values of the four anxiety factors were compared
with the squared correlations among the anxiety factors, following the recom-
mendation from Fornell and Larcker (1981). As demonstrated by Table 3, none
of the squared correlations exceeded the AVE values. This suggests that the
Baseline Model with four factors has acceptable discriminant validity.
246 SKILL-BASED L2 ANXIETIES REVISITED
Note: N = 229; dependent variable = FLCAS; R2 = 0.71; AVE = Average Variance Extracted;
CCR = Composite Construct Reliability.
Note: The squared values of the correlations between factors are presented in the upper
off diagonal. AVEs are presented in boldface type along the diagonal.
to Model 2A (i.e. 290.81 101.31 = 189.5, p < 0.05) verified that listening anx-
iety was statistically different from speaking anxiety, when reading and writing
anxieties were kept separate. This finding suggests that the existence of oral
anxiety as an independent construct was not supported by the data. It was also
found that reading anxiety was statistically different from writing anxiety, as
indicated by the significant difference in chi-square value between the
Baseline Model and Model 2B [2 (3) = 135.25, p < 0.05], signaling that com-
bining reading and writing anxieties into a more general construct of written
anxiety did not receive empirical support from the data. A similar pattern was
found for the relationship between speaking and writing anxieties, as evi-
denced by the significant increase of the chi-square value from the Baseline
Model to Model 2C [2 (3) = 251.8, p < 0.05]. Lastly, the significant increase
in the chi-square value from the Baseline Model to Model 2D [2 (3) = 88.91,
p < 0.05] suggests that reading anxiety was statistically different from listening
anxiety when speaking anxiety was specified to be independent of writing
anxiety. The results of a series of chi-square difference tests between the
Baseline Model and Models 2A through 2D led to the rejection of
Hypothesis 1, which assumed that the three-factor models (i.e. Models 2A–
D) would not be significantly worse than the Baseline model. These results
showed a sharp contrast to the significant correlations among the four
skill-based anxieties reported in Table 1.
Results of chi-square difference tests between the Baseline Model and
Models 3A and 3B also supported rejection of Hypothesis 2. To be specific,
the chi-square difference between the Baseline Model and Model 3A, which
specified a two-factor solution (i.e. Oral versus Written Anxiety), was
253.42 (i.e. 354.73 101.31 = 253.42) with 5 degrees of freedom difference
248 SKILL-BASED L2 ANXIETIES REVISITED
(i.e. 42 37 = 5), and this increase in chi-square value was significant both at
0.01 and 0.05 a level [2 (5) = 253.42, p < 0.01], providing evidence that a
simultaneous specification of the four skill-based anxieties into the oral versus
written dichotomy was not empirically supported. In a similar vein, Model 3B,
which posited a two-factor solution (i.e. Productive versus Receptive Anxiety),
produced a chi-square value of 376.88 with 42 degrees of freedom. A com-
parison of Model 3B with the Baseline Model revealed a significant increase in
chi-square value from the Baseline to Model 3B [i.e. 2 (5) = 275.57,
p < 0.05], hence statistically disproving the re-conceptualization of the four
skill-based anxieties into two related constructs of productive versus receptive
Among the four skill-based anxieties, listening anxiety made the biggest con-
tribution (i.e. = 0.383) to FLCAS, followed by speaking anxiety
(i.e. = 0.302) and reading anxiety (i.e. = 0.220). In the case of listening
anxiety, one unit increase in listening anxiety was associated with a 0.383
unit increase in the FLCAS score when the effects of other anxieties were
controlled for. The predictive power of writing anxiety, however, was mar-
ginal, as indicated by the smaller value of regression coefficient (i.e. = 0.095,
p = 0.041).
indicating that oral versus written pairing is better than the receptive versus
productive division in explaining skill-based L2 anxieties.
The second research question was concerned with the relationship between
the four skill-based anxieties and general FL classroom anxiety, as operationa-
lized by FLCAS. Most of the previous studies examining this issue have focused
on the bivariate correlation between FLCAS and a selected skill-based anxiety,
thus ignoring any interdependence among the four skill-based anxieties. The
study results demonstrated that each of the four skill-based anxieties made an
independent contribution to the prediction of FLCAS, even when the effects of
the other skill-based anxieties were controlled for. This finding forms a mean-
SUPPLEMENTARY DATA
Supplementary material is available at Applied Linguistics online.
T.-I. PAE 251
ACKNOWLEDGEMENTS
I would like to thank Jane Zuengler, Co-Editor of Applied Linguistics, and the four anonymous
reviewers for their insightful and informative feedback on the earlier versions of this article.
REFERENCES
Aida, Y. 1994. ‘Examining of Horwitz, Horwitz, Horwitz, E., M. Horwitz, and J. Cope. 1986.
and Cope’s construct of FL anxiety: The case of ‘Foreign language classroom anxiety,’ Modern
students of Japanese,’ Modern Language Journal Language Journal 70: 125–32.
78: 155–68. Hsiao, T. and R. Oxford. 2002. ‘Comparing the-
(eds): Language Anxiety: From Theory and Schumacker, R. and R. Lomax. 2004. A
Research to Classroom Implications. Prentice Hall, Beginner’s Guide to Structural Equation Modeling.
pp. 101–8. Lawrence Erlbaum.
Reise, S., K. Widaman, and R. Pugh. 1993. Spielberger, C. D. 1983. Manual for the State-trait
‘Confirmatory factor analysis and item re- Anxiety Inventory. Consulting Psychologists Press.
sponse theory: Two approaches for exploring Wang, J. H. and J. Guthrie. 2004. ‘Modeling
measurement invariance,’ Psychological Bulletin the effects of intrinsic motivation, extrinsic
114: 552–66. motivation, amount of reading, and past read-
Saito, Y. and K. Samimy. 1996. ‘Foreign lan- ing achievement on text comprehension be-
guage anxiety and language performance: A tween U.S. and Chinese students,’ Reading
study of learning anxiety in beginning, inter- Research Quarterly 39: 162–86.
mediate, and advanced-level college students Woodrow, L. 2006. ‘Anxiety and speaking