5.4.1 selecting and preparing data for analysis
The total amount of video data in the EvTech and CASCATE projects reached hun-dreds of hours. It was therefore necessary to carefully select the data that would be analysed. Original article I, based on the data from the EvTech project, utilised ap-proximately 172 minutes of video. First, sessions that included the three action stations that were of interest to the analysis of this original article (LEGO constructing, symbol matching, dance mat playing) were chosen. This resulted in 9 sessions out of a total of 18. The number was then brought down to 6 by choosing sessions that were evenly timed between autumn 2009 and spring 2010 to allow meaningful comparisons. This sampling method can be referred to as purposeful random sampling (Patton, 2002): the data selection was done systematically in advance of any knowledge of the results.
However, as Patton (2002) highlights, it is important to understand the purposefulness of such sampling as it does not aim for representativeness, unlike random sampling in quantitative research.
The logic for selecting the data for the rest of the original articles was different.
It was done using unsystematic purposeful sampling, which according to Patton (2002, 230) refers to ‘selecting information-rich cases whose study will illuminate the ques-tions under study’. Original article II utilised approximately 46 hours of video data from the CASCATE project. The video material included all the children with ASD or autistic features in the project, and was examined for instances of interactional trouble. The analysis was further narrowed down to sequences in which children with ASD initiated the management of trouble. The article includes data extracts from the storytelling and Kinect playing stations.
The data for original article III included approximately 168 minutes of video data from the LEGO constructing station. The analysis in this article focused on the use of pointing, and this particular activity that consisted of examining a model on the screen and assembling one using plastic bricks, created natural opportunities to use pointing. The chosen 168 minutes of video covered the participants’ time spent at the LEGO constructing station during autumn 2013.
The selection of data for original article IV was restricted by the available material;
the analysis in this article utilised the eye tracking data, approximately 232 minutes of which was available. This was combined with the video material that was recorded when the children wore the eye tracking glasses at the Kinect playing station.
All the original articles include purposefully chosen data excerpts that involve either more general descriptions of the events (article I) or detailed transcripts (articles II, III, and IV). In article I, this resembled what Patton (2002) calls operational construct sampling: the data excerpts were chosen based on a coding scheme developed to cap-ture specific behaviours related to joint attention. All the articles also involved the use of typical case sampling (Patton, 2002) in an attempt to demonstrate how the interactions often unfolded. Despite not being utilised in the published articles, the analyses (in articles II, II, and IV) involved the use of deviant case sampling (Patton, 2002) in order to examine unusual evolvement of interactions. For instance, in the analysis made for original article IV, there was an instance where a child, after successfully catching a virtual object, shifted his gaze from the game screen to an adult co-participant and laughed out loud. Based on previous similar instances, one would have expected the co-participant to join the child’s laughter to share the success in the game playing. In-stead, the co-participant verbally redirected the child back to the playing activity. Here (and after examining similar deviant cases) it became obvious that the co-participant treated the timing of the child’s attempt to share his success as problematic: the game was still ongoing. Such investigations can thus facilitate the understanding of both the unusual and usual.
The ELAN multimedia annotation tool (developed by the Max Planck Institute for Psycholinguistics) was used in preparing the data for analysis. ELAN allowed for the synchronisation of the video files so that recordings from the same situations could be played side-by-side. Usually there were two camera views from each situation (excluding original article I). Eye tracking data was exported as MP4 video files with scan path visualisation using the SMI BeGaze software. Scan path visualisation shows how gaze moves temporally, visualising the order of saccades and fixations (raw data without fixation filter was used). For original article IV, the eye tracking data and video recordings were all imported into ELAN and synchronised.
5.4.2 categorisation approach
The categorisation approach was applied for examining gaze in children with ASD (original articles I and IV) and educators’ pointing gestures (original article III). Firstly, original article I utilised a coding scheme drawing on content analysis to capture gaze-related behaviours from video recordings. Here content analysis was first used for qualitative analysis by mixing inductive and deductive approaches (see Elo &
Kyngäs, 2008; Tuomi & Sarajärvi, 2009): the categories in the coding scheme were partially derived from the data and partially drew on previous literature on joint attention behaviours in ASD (see section 6.1 for a more detailed description of the development of the coding scheme). The coding scheme was then utilised to quantify the observed behaviours, representing quantitative content analysis (see Neuendorf, 2002). ELAN was used for this categorisation phase. After the categorisation, PASW Statistics software was utilised for statistical analyses to compare the frequency counts of the observed behaviours in the three different contexts. The tests used were the non-parametric Kruskal-Wallis test, followed by pair-wise comparisons with the Mann-Whitney U test. Bonferroni corrections were employed. The analysis continued with a qualitative content analytic approach (broadly understood) in which the identified gaze behaviours were contextualised by transcribing the situations in which they oc-curred. This transcription did not follow the conventions of CA.
Original article III utilised quantitative content analysis to identify educators’
pointing gestures from the data based on a simple predefined coding scheme by two independent coders for the purpose of inter-rater reliability. However, instead of any statistical tests, here the content analytic phase was followed by a CA examination in the multimodal frame. The categorisation approach was also used for gaze behaviours in original article IV to assess where, how often, and for how long children with ASD look at the environment in terms of AOI’s, representing quantitative content analysis (Neuendorf, 2002). This coding was made frame-by-frame to record the exact timings of the gaze shifts. The AOI’s included people (i.e. looking at any person in the envi-ronment), which were further specified as educators (i.e. teachers and special needs assistants), researchers (i.e. researchers who were present at the weekly activity group sessions), children (i.e. other children who participated the activity group sessions) and non-present people (i.e. people who were not present at the Kinect playing station).
This was followed by a further specification of whether the children gazed at others’
head area or other body parts. No differentiation was made between looking at others’
eye or mouth region as such a categorisation is likely to yield inaccurate results when live eye tracking is used (Falck-Ytter et al., 2015). This categorisation was made using ELAN. The results of the categorisation were further quantified for distributional examination that was mainly conducted for illustrative purposes.
5.4.3 conversation analytic approach
The analyses in original articles II-IV draw on CA. Rather than relying on context-free, predefined analytical categories, CA starts off with ‘unmotivated looking’ with the intent of identifying analytically and interactionally interesting phenomena (Psathas, 1995). CA progresses inductively and can bring out new discoveries that one did not initially plan to research. As Peräkylä (2004, 295) has put it, ‘this unpredictability arises from the inductive character of the conversation analytic enterprise; it causes
both the fundamental difficulty and the exceptional fascination of conversation ana-lytic research’. In this thesis, the inductive approach meant viewing the recordings multiple times in order to identify interactionally relevant phenomena for detailed analysis. While my theoretical knowledge on concepts such as ‘joint attention’ was inevitably present, they were suspended from guiding the analytical observations made (in original articles II, III, and IV).
Specifically, this thesis utilises multimodally informed CA, which refers to the ap-preciation of all the resources that can be used in interaction (e.g. speech, gaze, ges-tures). In practice, the video recorded interactions were first reviewed multiple times by replaying the material, occasionally in slow motion. When an analytically interest-ing phenomenon was observed, a rough transcript was produced. The analysis then continued to both identify other interesting phenomena and to collect more instances of the interesting phenomenon first observed. In particular, my attention was directed towards sequences in which the children and their social partners were responsive to each other. The careful examination of these led me to further concentrate on gaze, gestures, and other multimodal resources, and on the occasional challenges in their use. When a decision was made on which phenomenon to focus on, more careful multimodal transcripts were produced to enable the study of sequentially organised actions. These transcripts show how talk, gaze, and other multimodal conducts move in relation to each other or during silent intervals. Talk was transcribed using the Jef-fersonian conventions (Jefferson, 2004), and gaze using Goodwin’s (1981) notations.
Furthermore, in original article III, pointing gestures were transcribed paying careful attention to the gesture phases (Kendon, 2004, 111-113). For a ‘first-round analysis’, ELAN was used to annotate the data based on initial observations of what might be analytically important. The annotations were made frame-by-frame to determine the timings accurately. These initial annotations were later reconstructed using CA con-ventions. At this phase, the transcripts were clarified and ‘cleaned’ from details that were not analytically relevant and, thus, would have been distracting for a reader. Such details often involved multimodal conduct such as hand and upper body movements that did not accomplish anything of interest in an interaction under study. The talk in Finnish was idiomatically translated into English using bold typeface. As the children’s talk was not always grammatically correct, the translations were occasionally ‘rough’.
The manner in which CA is utilised in this thesis can be conceptualised as applied CA; more specifically, it includes features of foundational, communicational, and in-stitutional applied CA (see Antaki, 2011). In foundational applied CA, the tools of CA are imported to a new discipline to encourage a rethinking of some of the discipline’s basic foundations (Antaki, 2011). In this thesis, CA allows us to reconsider the indi-vidualistic view of social interaction in ASD often taken in psychology, and specifically in ASD research based on the biomedical paradigm. In communicational applied CA,
‘disordered’ talk or behaviour can be understood by providing a complementary or alternative CA examination (Antaki, 2011). In this study, such a CA take on the inter-actions between children with ASD and their social partners might even challenge the view of ASD as a disorder of social communication and interaction. Institutional applied CA refers to investigating institutional activities, such as interactions during a doctor’s visit, in an attempt to understand how interaction is organised in such contexts (Antaki, 2011). The focus is not on everyday interactions. As the interactions in this thesis occurred in institutional settings (see section 5.2.2), institutional applied CA is inevitably present here. However, the institutional nature of the interactions is not the core focus of the study and is thus not emphasised.