Page 105 - Second language development of newly arrived migrant kindergarteners - Frederike Groothoff

Development of narrative ability 105 reliability assured us that we could use the transcriptions for the calculation of NDW, GIS, and MLR. Macrostructure For this study the scoring protocol of the MAIN (Gagarina et al., 2012) was used, but in an extended version. Because there were young participants in this study who just started to learn Dutch, it was difficult to follow the strict MAIN protocol. These pupils were creative, and it was difficult to predict what they would tell in the stories; therefore, the protocol was extended with more examples and guide lines, to make the scoring of the protocol more consistent over the raters. For macrostructure, inter-rater reliability was only calculated for Story Structure, because Structural Complexity was based on the story grammar components of Story Structure and the IST tokens were already included in the reliability of the transcriptions. We again randomly selected 16 stories (four stories per session) for this inter-rater reliability analysis. The author of this dissertation and a trained research assistant both scored the 16 stories independently and the point-to-point agreement for the ratings was 88.7% (range 75.1–100%) with a Cohen’s kappa of .70, which was considered a substantial strength of agreement (Landis & Koch, 1977). The mean reliability in Session 2 was the highest (96.9%) and in Session 3 the lowest (84.4%). Mean Cohen’s kappa scores of the sessions ranged from .49 to .93, meaning moderate to almost perfect agreement. In Session 1, one child did not score any points according to one rater which resulted in a Cohen’s kappa of 0 (the use of only one category does not deviate much from chance), even though the agreement was 81%. This low kappa reduced the overall kappa. When this kappa of 0 was left out of the analysis the overall kappa increased from .70 to .74. This substantial inter-rater reliability assured us that we could continue our analysis with the scorings of the narratives by one rater. 5.4 Results The sample comprised of 164 narratives collected during four sessions over a period of two-and-a-half years. Due to technical difficulties, four stories from four different participants are missing in the data set. The results of the narrative ability development are presented as follows: first, the developmental data from the MAIN is presented to answer the question: How does the narrative ability (in Dutch) of newly arrived migrant kindergarteners develop during the first two-and-a-half years of schooling in the Netherlands in relation to school type? The basis of the analysis is the General Development Model, which includes only Age at which narrative ability was measured as an explanatory variable. Second, child characteristics are introduced into the model. Exposure to Dutch at School will reveal whether exposure has a different influence on the