At the outset of the thesis and before the experiments were conducted predictions were made regarding the variables. Specifically:
The primary hypothesis is that in a variety of situations participants who are using systems that sense and transmit information related to emotion will view the situation they are placed in as more unethical, invasive, uncomfortable, hindering, immoral, suspicious, and unfair when compared to a control.
What we found was that only in the Quiz Experiment were these hypotheses partially true. In the control and charity gains motivators, significant differences were observed in reports on the respectful-invasive variable. In only the Control motivator a significant difference was observed in reports on the immoral-moral variable. In the Charity Gains condition a significant difference was reported on the hindrance-help variable. In these limited cases some of the original hypotheses were true: subjects reported their situation as less respectful (p=.03), less moral (p=.05), and more of a hindrance (p=.02).
Elsewhere, the hypotheses were not observed to be true and in some cases the opposite of what was expected was shown to be true. Specifically, in the Control motivator in the Interview Experiment, interviewees when paired with their interviewer reported the situation as significantly more ethical, more respectful, and comfortable when they were using sensors than when they were not.
These results seemingly contradict earlier findings in which affective computing systems presented without an ethical contract were reported as tending toward invasion of privacy [reynolds2004CHI]. To what might we attribute this disagreement?
It is quite possibly the case that the experience of live systems may be responsible for this difference in opinion. As evidence for this consider the following: if participants for all experiments are pooled and a Wilcoxon test is performed on the variable "with sensors that collect information about emotion" to "without sensors that collect information about emotion" we see a significant difference between the sensor and no sensor groups (p=.02). Those who did not experience sensors express a preference for using systems "without sensors that collect information about emotion," which is consistent with the contract findings. However, those experiencing systems "with sensors that collect information about emotion" have a more positive opinion. This suggests that the ecological validity of the questionnaire surveys was indeed lacking. In this more realistic approximation of using affective systems, participants express different opinions.
It may also be the case that the control participants (who were all from Good Company) had nothing to hide, so perhaps their ability to act ethically was projected onto their views of the technology. This may be related to the demeanor of the participants, one of the social dimensions discussed above.