Semi-structured interviews
In order to understand affective states experienced during the experiment, several interview techniques and frameworks were embedded into the interview process and questions. Mauss and Robinson in their review article on measures of emotion suggest that self-reports of emotion are generally more valid when applied to the currently experienced emotions or directly after an event. Interviews were therefore conducted right after the experiment. The interview structure followed the Spatio-temporal thread of the McCarthy and Wright’s Technology as experience framework – the participants were guided to relive the experience from the beginning to the end. This was combined with Petitmengin’s interview method which aims to elicit precise descriptions of participants’ subjective experiences. Following this method, participants’ attention was first stabilised by setting expectations about the length of the interview and its focus (the interaction between the participants, not the video content). Then, they were guided to reconstruct and describe the interactions between them during the experiment. Building on the related research of social presence and ambient social TV, measures of social presence and co-presence from the Networked Minds Social Presence Inventory which normally serve as questionnaire scales, were adapted and incorporated into the interview to explore the sense of co-presence between participants. Lastly, at the end of the interview, the participants were asked to reflect on the experience as a whole following Norman’s three-tier design framework which includes visceral, behavioural and reflective levels. Norman argues that only the reflective level requires conscious interpretation. It is, therefore, suitable to be explored in an interview.
Microsoft Emotion API – Facial expression analysis
I used the MS Emotion API to analyse facial expressions of the participants captured during the experiment. The API returns eight basic emotion values. Although sounding promising the results were disappointing. When manually comparing the videos, and detected emotions, there were obvious glitches (e.g. returning surprise when a participant touched her mouth). The data overall did not correlate well with manual assessment and is not therefore reported. Although a valuable lesson was learnt – the technology is still not ready to supplement research.
Tools I used
- Two laptops running letsgaze.com
- NVivo 11 (for interview transcripts and thematic analysis)
- Microsoft Emotion API
- Exploratory.io (An R-based data science tool to analyse emotions)
Identified Affective States
Six overarching themes
The participants did not find the experience very enjoyable, which reflects in the six overarching themes identified:
- Privacy intrusion
- Situational objective self-awareness
- Boredom
- Divided attention
- Obligation to interact
- Expecting emotion reciprocity
Each theme was examined and affective states were extrapolated based on literature review findings, participants‘ self-report and analysis of the videos from the experiment. Identified affective states were confirmed with the participants in a follow-up session in which a video of the experiment was re-played.
Affective states taxonomies
Before assigning specific names to identified affective states, I reviewed three seminal affective state taxonomies:
- Plutchik’s three-dimensional model
- Rusell and Barrett’s Core affect model
- Ortony, Clore and Foss’s Affective Lexicon
Identified affective states
Based on all of the above, I identified the following affective states:
- Alertness
- Anxiety
- Boredom
- Disappointment
- Discouragement
- Encouragement (negative valence)
- Overwhelmedness
- Vigilance
If you are interested in a detailed description of each state, please have a look at the study report below.
