Analyses

2.5.1. Learners’ interaction

The researcher and the rater examined eight conversation samples for the interactional features namely turn takings, target-like utterances, non-target like utterances (also called trigger), negative feedback, modified output. Then, using simple

65

percentage calculation for inter-rater reliability, 100% agreement was reached for the negative feedback categories and 97% for modified output.

The frequency of the interactional features in the first-year students’ and the second-year students’ interactions were also compared to see whether there were any differences between the two groups. However, as they were so low, no statistical analysis was performed.

2.5.2. Individual differences

The analyses for this study were carried out on the writing score as the outcome variable. First an independent t-test was used to see whether the difference in the writing scores of Group 1 and 2 is significant. Due to the nature of the variables (continuous, categorical, binary), two types of analyses were used to explore the relationship between the independent variables (gender, class rank, motivation, age of acquisition, initial writing score) and the outcome. All analyses were carried out for each group (1 and 2) separately. A one-Way between subjects ANCOVA was used with class rank (fixed factor, ordinal variable), initial writing proficiency (covariate, continuous variable) and the outcome writing scores (dependent, continuous variable). A linear regression analyses was also run for the two groups separately with gender, initial writing proficiency, motivation and age of acquisition as predictors and the writing score as the outcome.

2.5.3. L2 development

The writing scores were processed using the analytical software SPSS 22.0. To get an overview of learners’ development, descriptive analyses were first carried out. The scores from the participants in every session were averaged and compared based on groups and gender to see the overall development of the groups. Then, to determine if there was any significant progress of the learners’ writing scores, the pre-post approach was employed. For the pre-and post-scores, the average scores of the first three writings (pre) and the average scores of the last three writings (post) were used. Then, to get a better observation of the learners’ progress, the average scores of the middle three writings (mid) were also used in the analysis. By averaging these three scores, we hoped to avoid the effects the different topics and the missing data on the overall scores. To test the normality of the distribution of the data, Kolmogorov-Smirnov test was performed.

Then, Levene’s test was also carried out to test the homogeneity of the data. When the data were normally distributed and homogenous, then ANOVA and independent t-test

66

were performed. In contrast, when the data were not normally distributed and not homogenous, the data were analysed using non-parametric tests namely Mann-Whitney and Kruskal-Wallis tests.

In variability analyses from a CDST perspective, the trajectories of individual learners are inspected visually to see if scores go up or down rapidly from one session to the other or if there are major shifts. Visual inspection may be aided with min-max graphs or polynomial trend lines. (cf. Verspoor et al. 2011 for various techniques.) However, visual inspection indicated that none of the learners showed changes over timeafter the first few months. Also at the group level, the pre-post test showed that there is not much change over time. Therefore, no further variability tests were conducted. In the group analysis, variability for each learner was operationalized as the coefficient of variation (CoV), in line with Verspoor and de Bot (2021), but they also point out that this measure may be inadequate as it does not take time into account. They recommend that the Standard Deviation of Differences be used instead.

2.5.4. Pidginization

With the help of one other rater, the researcher examined in detail the writing samples of 20 learners for pidginization features. To select the samples, the learners’

holistic scores in the writing tasks were correlated with the group average. The learners with the strongest correlation coefficient with their group’s average scores were then selected. During the categorization process, discrepancies were discussed between the researcher and the rater until agreement was reached. Each pidginization feature was marked and counted. The percentage of the number of the features from the total number of words in each text was calculated. For the first step, a pre-post analysis of this ratio was carried out to see whether the learners improved in the sense that they produced fewer pidginization features overtime. For this step, we used the average of the ratio of session 2 and 3 for the pre-score and the average of session 17 and 18 for the post score. We did not use the first session since the topic of the first session is self-introduction, which apparently was very easy for the students and consisted mostly of well-memorized phrases. This was indicated with the fact that they produced significantly fewer pidginization features in this first session. Results of Group 1 to Group 2 were also compared to see whether there were any differences between the groups. We assumed that Group 1would improve overtime while Group 2 would remain stable. Finally, we

67

also counted the number of occurrences of each pidginization feature to see which features are more common in the learners’ L2. We also compared the features found in Group 1 and Group 2.