A measure of second language writing anxiety:
Scale development and preliminary validation
Y.-S. Cheng
Department of English, National Taiwan Normal University, 162 Hoping East Road,
Section 1, Taipei 10610, Taiwan
Abstract
Evidence has been accumulating that shows the promise of multidimensional conceptualizations
of anxiety in investigating the effects of anxiety on different aspects of human behavior and
intellectual performance. In view of the lack of an L2 writing anxiety scale explicitly developed
from a multidimensional perspective, this study aims to develop and evaluate a self-report L2 writing
anxiety measure that conforms to a three-dimensional conceptualization of anxiety. Sixty-five EFL
learners’ reports of L2 writing anxiety were drawn upon to generate an initial pool of scale items. A
pilot test was conducted on the initial pool of items to help establish a preliminary version of L2
writing anxiety scale for further refinement and evaluation in the formal study. A sample of 421 EFL
majors enrolled in seven different colleges in Taiwan participated in the formal study. Exploratory
factor analysis was employed to determine the final make-up of the Second Language Writing
Anxiety Inventory (SLWAI) that consists of three subscales: Somatic Anxiety, Cognitive Anxiety,
and Avoidance Behavior. In addition to reliability coefficients, the validity of the SLWAI total scale
and subscales was assessed by means of correlation and factor analysis. The results suggest that both
the total scale and the individual subscales of the SLWAI have good reliability and adequate validity.
# 2004 Elsevier Inc. All rights reserved.
Research background
For the past three decades, a great body of research has been devoted to examine the role
of anxiety in second language (L2)1 learning. However, as shown in Scovel’s (1978) review
of the literature then available, early studies on anxiety and L2 learning produced
Journal of Second Language Writing 13 (2004) 313–335
E-mail address: t22035@cc.ntnu.edu.tw.
1 In this paper, second language or L2 is used to refer to either a second language or a foreign language.
1060-3743/$ – see front matter # 2004 Elsevier Inc. All rights reserved.
doi:10.1016/j.jslw.2004.07.001
contradictory results regarding the relationship between anxiety and L2 achievement or
performance. Many researchers have attributed these discrepant findings in part to the use
of inadequate anxiety measures, such as scales of test anxiety and general trait anxiety,
which do not assess an individual’s responses to the specific stimulus of second language
learning (Horwitz, 1986; Horwitz, Horwitz, & Cope, 1986; MacIntyre, 1999).
Furthermore, these researchers proposed conceptualizing second/foreign language anxiety
as a unique form of anxiety specific to the L2 learning context. Since then, several
instruments have been developed and widely adopted to measure this anxiety, including
Gardner’s (1985) French Class Anxiety Scale and French Use Anxiety Scale, and Horwitz
et al.’s (1986) Foreign Language Classroom Anxiety Scale.
These L2-learning-specific measures of anxiety have contributed to a clearer picture
regarding the relationship between anxiety and L2 learning. In general, research adopting
these anxiety measures has indicated a consistent negative association of second language
anxiety with students’ L2 learning attitudes (Phillips, 1992), self-ratings of proficiency
(MacIntyre, Noels, & Cle´ment, 1997), language processing (Steinberg & Horwitz, 1986;
MacIntyre & Gardner, 1994a, 1994b), and with L2 academic achievement (Aida, 1994;
Horwitz, 1986; Saito & Samimy, 1996).
Nevertheless, some researchers, taking heed of the dominance of speaking-related items
in the above-mentioned second language anxiety measures, began to question the adequacy
of using them to measure anxiety aroused in performing language skills other than
speaking (Aida, 1994; Cheng, Horwitz, & Schallert, 1999; Phillips, 1992). Some language
anxiety researchers even took a step further, proposing to distinguish language-skill-
specific anxiety from general second language classroom anxiety that seems to be more
associated with oral aspects of L2 use (Cheng et al., 1999; Horwitz, 2001). Meanwhile,
researchers have increasingly focused their attention on identifying and examining anxiety
associated with specific language skills such as reading (Saito, Horwitz, & Garza, 1999;
Sellers, 2000), listening (Vogely, 1998; Kim, 2000), and writing (Cheng et al., 1999; Leki,
1999). This trend of investigation suggests a pressing need to develop proper and
standardized measurement instruments for researchers who are interested in the
quantitative assessment and investigation of various kinds of skill-specific second
language anxiety. For ‘‘adequate measures are a necessary condition for valid research’’
(DeVellis, 1991, p. 11).
There have been some studies on second language writing anxiety. In them, the Daly–
Miller Writing Apprehension Test (WAT; Daly & Miller, 1975) was the most commonly
used measurement instrument of second language writing anxiety (e.g., Cheng et al., 1999;
Hadaway, 1987; Lee, 2001; Masny & Foxall, 1992; Wu, 1992). Although the Daly–Miller
WAT as a whole has been shown to be an instrument of satisfactory internal consistency
reliability as well as concurrent and predictive validity, there seems to be plenty of room for
further improvement if the WAT is to be used in future studies of second language writing.
First of all, the WAT was originally developed with reference to first language learners,
particularly English native speakers. It might not tap the most essential aspects of second
language writing anxiety. Moreover, several researchers have raised questions about the
construct validity of the WAT. In his investigation of L1 writing anxiety, McKain (1991, pp.
22–25) did a content analysis of the WAT and classified the 26 items on the WAT into five
categories: (1) nine items on positive feelings that are more or less incompatible with
Y.-S. Cheng / Journal of Second Language Writing 13 (2004) 313–335314
anxiety (e.g., ‘‘I enjoy writing’’ and ‘‘I like to write my ideas down’’); (2) nine items on
self-efficacy or outcome expectancies (e.g., ‘‘I feel confident in my ability to clearly
express my ideas in writing’’ and ‘‘I expect to do poorly in composition class even before I
enter them’’); (3) five items on the presence (n = 4) or absence (n = 1) of anxiety (especially
evaluation anxiety) (e.g., ‘‘I have no fear of my writing being evaluated’’ and ‘‘I am afraid
of writing essays when I know they will be evaluated’’); (4) one item on the value of writing
(i.e., ‘‘Expressing ideas through writing seems to be a waste of time’’); and (5) two items
dealing with miscellaneous events or behaviors that might be caused by anxiety (i.e., ‘‘I
avoid writing’’ and ‘‘My mind seems to go blank when I start to work on a composition’’).
Because only 14 of the 26 WAT items deal with feelings (among them, only four items
concern the presence of anxiety), McKain called into question the WAT as a pure measure
of writing anxiety.
On the grounds that 9 of the 26 WAT items have to do with individuals’ self-efficacy
beliefs and outcome expectancies, McKain (1991) further argued that the WAT could be
considered as ‘‘a measure of writing self-esteem just as much as a measure of writing
apprehension’’ (p. 25). McKain’s argument has received some empirical support from
Cheng et al.’s (1999) principal components analysis of an L2 version of the WAT.
Originally, the WAT was reported to represent a unidimensional structure and capture a
single construct by Daly and Miller (1975). All of the items on the WAT are thus expected
to load on one factor or one component in a factor analysis of the measure. However,
different from the expectation, data in Cheng et al. (1999) supported a three-component
solution. That is, the WAT items loaded on three components. Among the three
components, the Low Self-confidence component accounted for the largest part of the total
variance of the L2 version of the WAT, followed by the components of Aversiveness of
Writing and Evaluation Apprehension. Similarly, one of the three components that
consistently emerged from Shaver’s (1990) three principal components analyses of the
WAT on L1 learners was Writing Self-concept, although it did not always explain the
greatest amount of variance in the three analyses. The other two components were labeled
as Affective Performance Reaction and Reaction to Evaluation. The above results suggest
that Low Self-confidence is a major component of the WAT.
Together with Burgoon and Hale’s (1983) finding that the WAT could be factor analyzed
into three dimensions representing Discomfort or Ease in Writing, Enjoyment of Writing,
and Rewards of Writing, Shaver (1990) argued that contrary to Daly and Miller’s (1975)
proposed unidimensional structure, the WAT represents a multidimensional construct,2
comprising three dimensions of L1 learners’ attitudes toward writing. However, the
dimensionality of the WAT is still subject to debate because two-factor solutions have been
obtained in other studies, as reported in Bline, Lowe, Meixner, Nouri, and Pearce (2001).
Y.-S. Cheng / Journal of Second Language Writing 13 (2004) 313–335 315
2 A unidimensional conceptualization of anxiety treats anxiety as a unitary, global construct. A unidimensional
measure of anxiety contains no subscales and produces only one single summed score. In contrast, a multi-
dimensional conceptualization of anxiety defines anxiety as being composed of several different but intercorre-
lated facets or dimensions. Each facet or dimension of the anxiety construct can be seen as representing a separate
construct. But at a more abstract level, these facets or dimensions are all integral parts of the more global anxiety
construct. Developed from this perspective, a multidimensional measure of anxiety comprises several subscales
designed to measure the various facets/dimensions of anxiety. The items on each subscale can be summed to get a
score representing the degree of reaction in each facet/dimension of anxiety.
These conflicting results not only highlight the need to clarify the issue of how many and
what dimensions are embedded in the WAT but also question the construct validity of the
WAT. It should also be noted that Shaver (1990) treated the WAT as a measure of attitudes
toward writing, a construct broader in scope than writing anxiety. Indeed, results of the
aforementioned content and factor analyses suggest that treating the WAT as a specific
measure of writing anxiety may be problematic.
In their report on two L1 studies that examined the predictive validity of the WAT,
Richmond and Dickson-Markman (1985) held a similar view to McKain (1991) regarding
the nature of the WAT. The two studies reported in Richmond and Dickson-Markman
(1985) upheld the predictive validity of the WAT because the WAT was found to be a
significant predictor of L1 achievement (i.e., the American subjects’ performance on the
English section of the American College Test) and L1 writing quality (i.e., the subjects’
performance on a 15-minute essay). However, they doubted that the predictive power of the
WAT came from anxiety. Based on the finding that low and moderate levels of writing
anxiety, as measured by the WAT, did not produce significant differences in reports of state
anxiety, Richmond and Dickson-Markman (1985) noted that the WAT might be ‘‘a measure
of self-confidence in ability to write rather than anxiety about writing’’ (p. 259).
In the form of a subscale, self-confidence has been included in sport anxiety measures
such as Competitive State Anxiety Inventory-2 (Martens, Burton, Vealey, Bump, & Smith,
1990) and the Anxiety Rating Scale-2 (Cox, Robb, & Russell, 2000). However, according
to the theory from which these sport anxiety measures were developed, self-confidence is
an individual difference construct that is separate from the construct of anxiety, which is
conceptualized to be multidimensional and consist of two components: cognitive and
somatic anxiety (Craft, Magyar, Becker, & Feltz, 2003). Cognitive anxiety refers to the
mental aspect of anxiety experience, including negative expectations, preoccupation with
performance, and concern about others’ perceptions; whereas somatic anxiety refers to
one’s perception of the physiological effects of the anxiety experience, as reflected in
increased ‘‘autonomic arousal and unpleasant feeling states such as nervousness and
tension’’ (Morris, Davis, & Hutchings, 1981, p. 541). This approach is quite different from
that taken for the WAT, which does not attempt to distinguish between self-confidence and
anxiety or separate cognitive anxiety from somatic anxiety.
As a result, use of the WAT may create some difficulty in teasing apart the conceptual or
causal links between anxiety and self-confidence, a subject of much research and
discussion in the field of L2 learning (Horwitz et al., 1986; MacIntyre, Cle´ment, Do¨rnyei,
& Noels, 1998; Oxford, 1999). The finding of a close relationship between general second
language anxiety and self-perceived competence in L2 has provided some insight into ways
of reducing anxiety in the L2 classroom (Cheng, 2001; MacIntyre et al., 1997;
Onwuegbuzie, Bailey, & Daley, 1999; Price, 1991). However, similar results found in the
studies of L1 or L2 writing (e.g., Cheng, 2002; Daly & Wilson, 1983; Pajares & Johnson,
1994) become more difficult to interpret. The writing anxiety scale (i.e., the WAT) used in
these studies to examine the relationship between writing anxiety and self-confidence
contains many items related to self-confidence in writing. Such attempts thus become a
potential tautology.
One possible solution to this problem is to develop a measure of writing anxiety that
does not confound writing anxiety with self-confidence or beliefs about one’s writing
Y.-S. Cheng / Journal of Second Language Writing 13 (2004) 313–335316
ability. To achieve such a goal, McKain (1991) devised an L1 writing anxiety measure by
drawing 12 items from the WAT and Holland’s (1978) Writing Problems Profile. The 12
items were chosen on the basis that at least three of four independent raters coded them as
being related specifically and narrowly to ‘‘anxious feelings’’ associated with writing, but
not other aspects of writing such as ‘‘behavior (e.g., avoidance of writing) or cognition
(e.g., beliefs about one’s writing ability)’’ (McKain, p. 81). Deliberately excluding items
related to writing self-efficacy beliefs and enjoyment of writing, McKain’s writing anxiety
measure, the Writing Anxiety Questionnaire (WAQ), was shown to be an improvement
over the WAT in terms of content validity and construct validity, despite their similarity in
predictive validity.
Like the WAT, the WAQ defines L1 writing anxiety as a unidimensional, global
construct and does not contain any subscales. However, as previous studies suggest,
empirical evidence does not always support scale developers’ categorization and definition
of the scale items. For example, Russell and Cox (2003) found that the word ‘‘nervous’’
could be interpreted as being related to somatic anxiety or cognitive anxiety by different
people. In a factor analysis of potential cognitive and somatic anxiety items drawn from
existing anxiety scales, Morris et al. (1981) found that although some of the items, as
expected, loaded differentially on either the worry factor (i.e., cognitive anxiety) or the
emotionality factor (i.e., somatic anxiety), many of them did not. Likewise, the proposed
unidimensional structure of the WAT has been challenged by several factor analyses of the
WAT, as reviewed earlier. Therefore, the presupposed unidimensional structure of the WAQ
requires validation via factor analytic methods. In addition, because the WAQ, like the
WAT, was not developed specifically for L2 learners, its applicability in the L2 context is
questionable.
More importantly, accumulating evidence has indicated the promise of multi-
dimensional conceptualizations of anxiety in investigating the antecedents of anxiety as
well as the effects of anxiety on different aspects of human behavior or intellectual
performance (Morris et al., 1981; Smith & Smoll, 1990). For instance, following a
multidimensional approach and conceptualizing test anxiety as consisting of two
components—worry and emotionality—Morris and Liebert (1973) showed that the
antecedents of the two components of test anxiety differed. To be specific, the threat of
electric shock aroused merely emotionality when one worked on an intellectual task, but
failure feedback at the same situation aroused worry only. On the other hand,
Deffenbacher (1977) found that the two components of test anxiety varied in their impact
on intellectual performance. That is, worry alone correlated with Miller Analogies Test
performance among graduate school applicants; emotionality, however, produced
debilitative effects only in the high-worry group. By the same token, when conceptualizing
pain-related anxiety as consisting of four components: cognitive anxiety, somatic anxiety,
fear, and escape/avoidance, Bishop, Holm, Borowiak, and Wilson (2001) found that
somatic anxiety was the only significant predictor of pain tolerance among participants
with tension headache. Obviously, the significant relationships between various facets of
anxiety and human behaviors reported above would have been obscured by the use of a
unidimensional conceptualization and measure of anxiety. No wonder more and more
researchers advocate a multidimensional approach to anxiety conceptualization and
measurement.
Y.-S. Cheng / Journal of Second Language Writing 13 (2004) 313–335 317
From the multidimensional perspective, anxiety is not a unitary, unidimensional
phenomenon but involves various response dimensions. Various multidimensional
measures of anxiety have been developed in fields such as test anxiety (see Morris et
al., 1981; Sarason & Sarason, 1990), speech anxiety (see Fremouw & Breitenstein, 1990),
and sport performance anxiety (see Smith & Smoll, 1990). In these measures, anxiety
symptoms are grouped along several relatively independent dimensions, including
somatic/physiological (e.g., upset stomach, pounding heart, excessive sweating, and
numbness), cognitive (e.g., worry, preoccupation, and negative expectations), and
behavioral (e.g., procrastination, withdrawal, and avoidance). However, a measure of
second language writing anxiety that takes into consideration the multidimensional nature
of anxiety is still wanting. In light of the problems discussed above, the purpose of this
study was to develop a self-report measure of second (here foreign) language writing
anxiety grounded in both L2 learners’ reports of anxiety experiences and the
multidimensional conceptualization of anxiety. Further, this study presented preliminary
information regarding the construct and criterion-related validity of the measure.
Method
Participants
Three groups of EFL students majoring in English in Taiwan participated in the present
study. Only English majors were recruited due to the concern that students otherwise might
not have sufficient English writing experiences to provide rich information regarding their
writing anxiety experiences. First of all, to help generate an initial pool of L2 writing
anxiety scale items, 67 undergraduate and graduate students (59 females and 8 males,
ranging in age from 19 to 35 years, with a mean age of 22 years) from three intact classes in
the English department of two universities were recruited to fill out an open-ended
questionnaire that asked about their anxiety exper