• No results found

Chapter 3 Research Methodology

3.6 Data Analysis Procedures

3.6.1 Quantitative Data Analysis

The quantitative data were analysed by means of both descriptive and inferential statistical procedures. The reliability of the questionnaires was also computed.

3.6.1.1 Descriptive and Interferential Statistical Analyses

The collected data were entered into the SPSS programme for descriptive and inferential statistical analyses.

Descriptive statistical analysis: The participants’ responses were presented in terms of

percentages, mean scores and standard deviations for each item, each subscale and the whole scale in Part B, Part C, and Part D. For calculating the participants’ attitudes towards each item in the sections of beliefs (Part B) and motivation (Part C), responses for strongly disagree and disagree are categorised as disagree (D), neither disagree nor agree as neutral (N), and agree and strongly agree as agree (A).

For calculating the participants’ attitudes towards each item in the strategies section (Part D), responses for I never do this and I seldom do this are categorised as seldom (S), I sometimes do this as neutral (N), and I usually do this and I always do this as often (O).

Inferential statistical analysis: A parametric test (an Independent Samples T-test) was

performed to ascertain any possible changes in these three learner variables and to examine the extent of these changes among the participants upon and after their arrival at the EMI University. The means of each individual item, each subscale, and the whole scale were compared between the Upon- and After-arrival questionnaires when performing the Independent T-Test. The Cronbach alpha level was set at 0.05. However, when performing Independent T-Test, Items B1 and B13 were excluded because these

two are ordinal data, which fail to meet the requirements for executing a parametric test (Dörnyei, 2007).

3.6.1.2 Reliability of the Questionnaire

This section presents the results of the internal consistency reliability estimates of this questionnaire (Table 3.5). Reliability coefficients, as used here, indicate the degree to which a subscale is internally consistent or reliable. All of these reliability estimates are Cronbach alphas. Cronbach alpha can be interpreted as the percentage of reliable or consistent variance in each instrument. For the current questionnaire, the reliability of the subscale of “difficulty of language learning” in the scale of the BLL could not be calculated, because Items B1 and B3 of this subscale are rank order statements while the remaining items are multiple-choice items.

Table 3.5 Reliability of the Language Learning Questionnaire

Part Content Number Items used Alpha

(Survey I) Alpha (Survey II) Part B BLL 15 B1-B15 .5061 .6345 Difficulty 4 B1, B2, B7, B13 \ \ Nature 6 B3, B4, B5, B6, B8, B9 .3218 .4240 Autonomy 5 B10, B11, B12, B14, B15 .5316 .6512 Part C MLL 19 M1-M19 .7017 .7352 Intrinsic interest 2 M1, M2 .7585 .6471 Immediate achievement 2 M3, M4 .7593 .6617 Going abroad 2 M5, M6 .4518 .5110 Individual development 3 M7, M8, M9 .5799 .4925 Information medium 2 M10, M11 .3737 .4070 Important others 2 M12, M13 .6713 .6875 Learning situation 6 M14, M15, M16, M17, M18, M19 .6449 .5405 Part D LLS 30 S1-S30 .8333 .8250 Memory 5 S1, S2, S3, S4, S5 .4117 .5050 Cognitive 6 S6, S7, S8, S9, S10, S11 .5862 .5642 Compensation 3 S12, S13, S14 .3856 .3334 Meta-cognitive 8 S15, S16, S17, S18, S19, S20, S21, S22 .7518 .7270 Affective 4 S23, S24, S25, S26 .5421 .5212 Social 4 S27, S28, S29, S30 .7404 .5077

As revealed in Table 3.5, the reliability for “Nature of language learning” in Survey I was a low .3218 and the reliability for “Autonomy in language learning” was a medium .5316. The reliability of the overall BLL items in the questionnaire was .5061, acceptably high. With regard to the MLL items in the questionnaire, the reliabilities of their seven subscales ranged from a relatively low .3737 for “Information Medium” to a relatively high .7593 for “Immediate Achievement”. The reliability of the overall MLL items in the questionnaire was .7017, reasonably high. As for the reliability of LLS in the questionnaire, its six subscales fluctuated from a relatively low .3856 for “Compensation Strategies” to a relatively high .7518 for “Mata-cognitive Strategies”. The reliability of the overall SLL items in the questionnaire was .8333, fairly high. In Survey II, the reliability for “Nature of language learning” was a relatively low .4240 and the reliability for “Autonomy in language learning” was a medium .6512. The reliability of the overall BLL items in the questionnaire was .6345, acceptably high. As for the reliability of MLL in the questionnaire its seven subscales ranged from a relatively low .4070 for “Information Medium” to a medium .6617 for “Immediate Achievement”. The reliability of the overall MLL items in the questionnaire was .7352, satisfactorily high. With regard to the LLS items in the questionnaire, the reliabilities of their six subscales fluctuated from a relatively low .3334 for “Compensation Strategies” to a relatively high .7270 for “Meta-cognitive Strategies”. The reliability of the overall SLL items in the questionnaire was .8250, fairly high.

It can be observed from the above table that the overall reliabilities of the MLL and LLS scales in the two surveys and that of the BLL scale in Survey II are all above .60, while that of the BLL scale in Survey I is .5316. According to Dörnyei (2007), a Cronbach alpha of 0.60 is an acceptable level for quantitative research in applied linguistics. Judging from this criterion, the scales of this questionnaire, except for BLL, could be considered to be relatively high in terms of reliability coefficients. The relatively low Cronbach alpha of BLL may be explained as follows: the BLL items were designed by drawing sources from the BALLI. The BALLI was reported to have a low alpha ranging

from .53 to .71 in the literature (Ahn & Yang, 2009). This low alpha might be caused by its inadherence to the convention of using multi-item scales, a prerequisite for the measurement of a questionnaire’s internal consistency reliability (Sage, 2011).

Another issue to be addressed is that, as revealed in the above table, when each scale is dismantled into subscales, the reliabilities of these subscales become smaller. This might be caused by the smaller number of items for each subscale. The number of items which comprise a scale tends to also decide the internal consistency reliability of the scale (Dörnyei, 2007). Generally, the more items a category contains, the higher the reliability estimate is. Therefore, when each scale is broken into subscales, fewer items are included in each of them, which might have resulted in smaller alphas.

More importantly, the low alphas might be engendered by the complexity of the SLA process (Dörnyei, 2007). The complex nature of the SLA process makes L2 researchers “typically want to measure many different areas in one questionnaire” (Dörnyei, 2007, p.207). Therefore, it is impractical for them to utilise long scales. In addition, a very long scale might be time-consuming for participants to use. Therefore, it is reasonable to expect to encounter low Cronbach alphas in quantitative research of applied linguistics.