In summary, in the present study on foreign language anxiety, the components which served as aggregate variables from this instrument are the ought-to self/own, ought-to self/other, ideal self, motivated learning behavior, language use anxiety, language class anxiety, language learning attitude. It must be noted here that the language learning attitude component from this questionnaire was used in later analyses as opposed to the attitude component derived from the FLCAS due to psychometric considerations: the attitude component of the motivation questionnaire had a higher reliability coefficient associated with it (α = .88). The aggregate scores were calculated from the mean of each scale for each respondent. These were used in later analyses as observed variables.
& Furuoka, 2006) or learners from similar cultural background studying different foreign languages (Horwitz, 1988; Yang as cited in Kuntz, 1996; Diab, 2006).
The second problem stems from the first in that the groups of items claimed to revolve around one particular issue actually tap into other areas of language learning that at best overlap but are not identical. This is the case of the following sample items that aim at measuring learners’ beliefs of utilizing learning strategies according to Horwitz (1988). It can be seen that item7 It is important to speak English with an excellent pronunciation and item21 I feel timid speaking English with other people do not fit into any taxonomy of learning strategies, but seem to report on beliefs about language production and language anxiety, yet Horwitz grouped them under the heading of learning and communication strategies.
In spite of these problems, and Kuntz’s (1996) critical review, the use of these latent constructs still persists in research studies using the BALLI as a data collection instrument. Table 15 below summarizes five BALLI studies and the theoretical categories data analysis was based on. Horwitz (1988) in her descriptive analysis consistently kept to the original categories (except for items 31 and 32). Yang (as cited in Kuntz, 1996), Nikitina and Furuoka (2006) and Rieger (forthcoming) with the help of principal components analysis aimed to confirm the existence of Horwitz’s (1987) latent factors, but their results are quite varied.
Yang (as cited in Kuntz, 1996) conducted her study in the Taiwanese context with the participation of 505 English language learners. The data collected with the BALLI was analyzed with principal component analysis. The author found evidence only for the existence of beliefs about language aptitude as a latent dimension of BALLI, but not the other four components. Three new dimensions however did
Table 15
Componential Structures of BALLI as Published in Previous Studies Component Horwitz
(1987, 1988)
Yang (cited in Kuntz, 1996)
Nikitina &
Furuoka (2006)
Rieger
(forthcoming)
Language
learning aptitude
1 2 6 10 11 16 19 30 33
2 8 10 19 11 24 30
19 11 1 6 30 33
Language learning difficulty
3 4 5 15 25 34
Self-efficacy 4 5 13 16 21 6
24 6 4 34
Nature of language learning
8 12 17 23 27 28
Importance of formal study 17 22 23 25 28 34
12 17 23 28
Learning and communication strategies
7 9 13 14 18 21 22 26
9 14 26 (36 37)
Motivation 20 24 29 31 32
High value 7 9 12 18 29 20 33
29 20 24 24 29 32 (39)
Note: All items are referred to with numbers as in the questionnaire used in this study based on Horwitz (1987). The additional items added by other researchers are in brackets.
emerge: beliefs about self-efficacy, the importance of formal study, and beliefs about the high value of learning English.
Nikitina and Furuoka (2006) in their Malaysian study found that nine items of the 34-item instrument loaded onto four of the factors conceptualized by Horwitz (1987), albeit only two items were found to explain the latent constructs of language aptitude, language learning difficulty, and learning strategies. Motivation was defined by three items. The rest of the items, as Nikitina and Furuoka (2006) reports, were excluded from the analysis because they loaded onto more than one dimension.
Rieger’s (forthcoming) investigation is interesting in that the research study was conducted in Hungary with the participation of English and German majors. The author here too examined the validity of the theoretical construct with principal component analysis. Rieger added five more items to BALLI with the intention of gaining information on learners’ beliefs about culture, their attitude to communicating with non-native speakers and their beliefs about learning through using authentic materials. Rieger identified the dimensions of language learning aptitude, the nature of language learning, learning and communication strategies, and motivation with three or more items loading onto each factor. On the other hand, only two items loaded on the language learning difficulty factor, which questions its reliability.
The primary intention of this part of the present data analysis was to aggregate the information provided by the respondents on the BALLI. For this reason, principal components analysis was used based on the conceptual framework of Horwitz (1987 and 1988) and dimensions that were identified by Yang (as cited in Kuntz, 1996).
The principal component analysis based on the original categories yielded the results as depicted in Table 16. Each factor was identified by at least three items. The variance these items explained in each component was rather low in all cases except for the dimension of motivational beliefs. The determinants approached the value of 1 and the KMO value did not exceed .6 which meant that there were probably high partial correlations detected among the variables indicating that they do not explain the component adequately.
Table 16
Componential Structure of BALLI
Component Item loadings Variance explained
Det. KMO Bartlett’s test significance Language aptitude balli1 .726
balli6 .481 balli16 .767
44.91% .904 .537 .001
Language learning difficulty
balli3 .632 balli5 .735 balli15 .682
46.84% .890 .580 .001
Nature of
language learning
balli12 .695 balli27 .594 balli28 .690
43.74% .932 .564 .001
Learning strategies balli14 .604 balli18 .695 balli26 .737
46.40% .894 .572 .001
Motivation balli29 .611 balli31 .842 balli32 .809
57.91% .640 .588 .001
Note: The analysis was based on the original categories of Horwitz (1987)
The scales calculated based on these components further underpinned the assumption that these dimensions are not measured reliably with these variables (Cronbach's alpha values ranged from .35 to .62) (see Table17); therefore, an aggregate score would not provide meaningful information for further analysis.
Table 17
Descriptive Statistics of the BALLI Scales Based on Horwitz (1987) Component Reliability
(α)
M Variance SD Item
N
Language aptitude .38 9.19 3.70 1.92 3
Language learning difficulty
.43 11.41 3.38 1.84 3
Nature of language learning
.35 11.72 2.62 1.62 3
Learning strategies .40 11.76 2.49 1.58 3
Motivation .62 13.07 3.12 1.77 3
As another possibility, the categories found in Yang’s study were also tested.
In the Hungarian context of high school students the items in Yang’s dimension of language learning aptitude loaded onto four variables, those in the importance of formal study and beliefs about the high value of learning English factor loaded onto three-three components. Surprisingly, when investigating the dimension of self-efficacy beliefs, results showed acceptable item loadings that explained 56.10% of the variance, with a KMO approaching 1 and a low determinant value. What is more, having computed the scale from the items that loaded onto the factor, its reliability also proved acceptable at α = .80 (see Table 18 and 19). Thus, the next observed variable included in the later analysis of language anxiety and other individual variables was self-efficacy beliefs as measured by items from the BALLI.
Table 18
Principal Component Analysis of the BALLI Based on Yang’s Self-efficacy Scale Component Item loadings Variance
explained
Det. KMO Bartlett’s test significance Self-efficacy balli4n .664
balli5 .835 balli13 .724 balli16 .789 balli21n .721
56.10% .232 .828 .001
Note: After the item number, the letter ‘n’ refers to a key reversed item.
Table 19
Descriptive Statistics for the Self-efficacy Scale Component Reliability
(α)
M Variance SD Item
N
Self-efficacy .80 16.79 13.89 3.73 5
In summary, despite the importance of investigating learners’ beliefs about language learning, the most widespread instrument that measures this individual variable apparently does not seem to possess psychometrically adequate properties. It seems that due to problems of construct validity and reliability, the same dimensions of beliefs could not be identified across different studies. Many times, this is attributed to the distinct context of these studies in which they were carried out.
Nevertheless, these weaknesses do not allow for the meaningful comparison of results or for their meaningful interpretation. In the present case, only one scale of a previous study was replicated: namely, self-efficacy beliefs. In later analyses of anxious learners’ beliefs and discrepancies between teachers’ and their learners’ beliefs about language learning primarily this scale was used. For a more detailed understanding,
following previous studies, learners’ and teachers’ responses to individual BALLI items were also examined.