I do not have experience with duration, but change in coding unit can greatly affect your statistical values, making it impossible to compare with other studies. We showed the vast difference caused by doing the same statistical test based on character, word, or larger unit (Rau & Shih, Evaluation of Cohen's Kappa, section 4.5). More importantly, since characters are not coded individually, using them as the basis for a statistical test would be invalid.

In one part of the research project DiaMaNt (Kreis & Brunner, 2019), we are investigating the quality of classroom discussions between four settings: discussions of student teachers with (1) their mentor teachers, (2) their university-based educators, (3) their lecturer in research in mathematics education and (4) their peer (student teacher in the same class).

For the conversation analysis, we want to find out how much was talked about a topic of conversation. We work with MAXQDA and code transcripts. MAXQDA does not automatically transfer coding from the transcript to the video file. Therefore, we do not have any information about how long certain topics are talked about. However, MAXQDA can determine the number of characters in the transcript assigned to a category. Now we wanted to find out to what extent the percentages of code coverage of transcripts and video correspond.

We compared the code coverages of transcripts with those of the video track in three videotaped classroom discussions with student teachers and mentor teachers. The data came from the research project Content Focus Coaching (e.g. Kreis, 2012). First, the transcripts were coded along a category system using qualitative structuring content analysis (Mayring, 2010). Subsequently, the codings were transferred to the video track. We found very little difference between the percentage code coverages. The differences between the transcript and the video ranged from 0.01% to 3.9% per category.

The small differences show that the transcribed text passages respectively the transcription rules cover the time required in the interview well. This can be explained, among other things, by the fact that the mentoring dialogue are formalised (mentor teacher = expert; student teachers = novice). The subject of the conversations (action of student teachers pre, during and post the lesson) is also predefined. In such formalised conversations, there are few overlapping and competing conversational contributions. It would be necessary to examine how code coverage differs in less formalised conversations.

Literature

Kreis, A. (2012). Produktive Unterrichtsbesprechungen: Lernen im Dialog zwischen Mentoren und angehenden Lehrpersonen. Bern: Haupt.

Kreis, A. & Brunner, E. (2019). Lerngelegenheiten für Lehrstudierende im sozialen Netzwerk Praxisfeld aus allgemein- und mathematikdidaktischer Perspektive (Projektantrag). Bern: SNF.

Mayring, P. (2010). Qualitative Inhaltsanalyse. Grundlagen und Techniken (11.). Weinheim: Beltz.

I wrote the CLASS program to accomplish your goal. See http://class.wceruw.org/index.html & http://class.wceruw.org/class.html.

Good luck.

Martin Nystrand