Applying clustering to analyze opinion diversity

In empirical software engineering research there is an increased use of questionnaires and surveys to collect information from practitioners. Typically, such data is then analyzed based on overall, descriptive statistics. Even though this can capture the general trends there is a risk that the opinions of different (minority) sub-groups are lost. Here we propose the use of clustering to segment the respondents so that a more detailed analysis can be achieved. Our findings suggest that it can give a better insight about the survey population and the participants' opinions. This partitioning approach can show more precisely the extent of opinion differences between different groups. This approach also gives an opportunity for the minorities to be heard. Through the process significant new findings may also be obtained. In our example study regarding the state of testing and requirement activities in industry, we found several significant groups that showed significant opinion differences from the overall conclusion.