An increasing popularity of big data analyses in anthropology in recent years is most evident in the emergence of a new field called computational anthropology. Even though computers have been used in anthropology at least since mid 20th century and even though computational social sciences are a well-established field, there are few anthropological researches that use computational approaches for data analysis. The first to point to the usefulness of computers as research tools for anthropology were Lévi-Strauss, who wanted to efficiently organize basic units of myths, and Edmund Leach, who was looking for general structures of societies. In Slovenia in 1970, Helena Ložar-Podlogar used punch-card system to record material on marriage rituals, and in 1991 Jurij Fikfak designed a digitized information system of Göth topography with his co-workers. Nevertheless, most studies remained predominantly qualitative, with quantitative data used only as secondary material.
In the last decade we witnessed methodological and epistemological progress in inclusion of computational analyses in anthropological research. Integration of qualitative and quantitative approaches stems from the tradition of mixed methods, where specific approaches include cooperative descriptions of visualizations (i.e. »ethno-mining«), calibration of analytical parameters and interpretation of findings with ethnography (i.e. »stitching«), designing a new conceptual field from both methods (i.e. »blending«), and explicit interdisciplinary methodological experimentation (i.e. »hybrid methodology«).
In this dissertation, I propose a novel methodology called circular mixed methods, which draws from both approaches and promotes continuous reformulation of hypotheses and findings based on data analysis and ethnography. I show the advantages and shortcomings of circular mixed methods on the analysis of workspace sensor data, explore the possibilities of intertwining ethnography with data mining and propose a methodological framework for applying computational techniques in anthropology.
Circular mixed methods stem from the tradition of interdisciplinary researches in anthropology, specifically from mixed methods. If general mixed methods combine qualitative and quantitative aspects sequentially or in parallel, the focus in circular mixed methods in on circularity. Circularity signifies continuous transversing between the two approaches, where the impetus for a change in the technique comes from the problem in the data. Once the extraction of new information from the data is exhausted with one technique, another is used, which gives a novel view of the data.
The circular mixed methods are demonstrated on the case of the Faculty of Computer and Information Science (FCIS) building. The complex, where the FCIS building is located, regularly records over 20,000 input-output signals, among which I have chosen room temperature and occupancy, energy use and air quality for the analysis. There were over three million data points in total, which I have aggregated to 15-minute intervals. Repetitive patterns in the data where identified with visualizations, frequentist techniques, and clustering. In 2018, we installed air quality sensor in certain rooms, with the aim of improving work space parameters. By installing the sensor, which reflected air quality with the color of the light, the frequency of opening windows increased. The intervention proved useful, but only if it is designed in a user-friendly way. Sensor wasn't very responsive, which frustrated the users, who stopped trusting the sensor as the time went by.
I used ethnography to identify people's reactions to the air quality sensor. In the first few weeks most users took the warning of poor air quality seriously and instantly responded (i.e. opened the window). When the sensors failed to reward the user with favourable feedback for while, the user ceased to trust the device and got increasingly frustrated. The initial response was joking about the sensor, then anger, then finally the user simply ignored the sensor. Technological solutions have to be designed in such a way to be easy to use and fit the users' expectations, otherwise they will stop using them sooner. The easiest way to develop such a product is by cooperating with the end users and following the principles of co-creation. To enhance data analysis I have developed a system for interview transcript segmentation. The structure of the interview is determined by questions and answers, which we have to consider when analysing text. Therefore I tested six algorithms for interview segmentation, where question-answer pairs are taken into account when determining the thematic structure of the text. It turns out none of these approaches work. The best one is a simple algorithm, which joins question-answer pairs if the second question contains less than five words or it doesn't end with a question mark. The segmented interviews were then clustered with hierarchical clustering, which successfully joined the segments with a similar topic. This approach made it easier to compare the answers of different research participants.
The results of interview analysis were once again supplemented with ethnography. The main take way is that technological solutions are a welcome tool, but they have to enable some level of control for the user. Concurrently, »smart« buildings should be designed by end users' requirements from the get go, as subsequent modifications are costly or are even impossible to do. And when the users feel like the solution isn't appropriate, yet is vital for their work and well-being, they will hack and modify it to their needs.
With sensor data analysis I show how quantitative approaches, specifically machine learning, data mining and text mining, successfully complement qualitative approaches, that is participant observation and interviews. Such mixing of methodologies can be just as fruitful in other research contexts. Predictive models can be used for automatic labelling of archive images, as I show on the case of hayracks. Mixed methods also work well in the industry; I present the development of a user interface for a media portal, where the interface was designed in cooperation with the end users. By combining machine learning and focus groups the team was able to design the final product, which fit the needs of the users and the requirements of the client.
Quantitative approaches in anthropology don't reject standard ethnographic approaches, they enrich and enhance them. Computational techniques are specifically appropriate for analysis of large data sets, longitudinal and simultaneous phenomena, and for preliminary field research and generating research questions. Ethnography, on the contrary, explains and makes sense of the discovered patterns, places the results in social and cultural context, and generally supplements the findings with rich descriptions. The main advantage of circular mixed methods is that they create a recurrent research loop, which ensures an additional perspective for each data source and enables a holistic insight into the studied phenomenon.
|
