The ‘Methodology’ section of published articles tends to present research as a smooth process, whereby investigators move neatly from research design, to data collection, analysis and finally ‘write it all up’. This blog post provides a ‘behind the scenes’ for our project, ‘Belong Anywhere: The Politics of Airbnb in Occupied Territories’. The project is funded through a GDC seed grant, and was also boosted through two Urban Lab apprenticeships, through the UvA Urban Studies masters.
The starting point for the project is Airbnb’s slogan: ‘belong anywhere’. Several studies have already questioned the extent to which the company lives up to this promise, highlighting widespread discrimination against potential guests from minorities and racialised groups. This research focuses primarily on cities in US and, to a lesser extent, Europe, and accepts Airbnb’s premise that it is indeed desirable for everyone to feel ‘at home’ during their travels. In historical Palestine, however, the question of who belongs where is arguably more fraught, as the claims that Jewish people ‘naturally’ belong has been leveraged by the Israeli state as justification for the Occupation and for exclusive political and civil rights based on religion and ethnicity. Our project, then, explores how Airbnb’s potential capacity to generate a sense of belonging plays out in this context.
Many studies on Airbnb rely on quantitative (spatial) analysis of listings to identify common patterns in the platform uptake and usage, as well as its impact on housing stock and rental prices. Alternatively, researchers have used interviews with guests or hosts to consider which social groups use the platform, how and why. By contrast, we were mostly interested in examining how people engage with Airbnb as a digital artefact: How do hosts communicate through the listings? How does the platform afford certain people the ability to belong in particular places through hosting? Our focus was thus on the meanings communicated through the listings themselves.
Our first idea was to use topic-modelling. Topic modelling is a statistical model of natural language processing, which identifies clusters of words that tend to co-occur in a given text corpus. This technique is more commonly used by businesses to automate data organisation, but it has also been used to examine, for example, Airbnb reviews. We were curious to see if topic modelling could be used to inductively generate initial codes for our listings or show us that certain topics were more prevalent in certain areas rather than others. What we had not considered is that large portions of any listing are rather mundane and generic, describing, for instance, the check-in procedures, or the availability of spare linen. Although it is standard practice in topic modelling to remove words from a corpus which don’t add much meaning and produce so-called noise, which drowns out meaningful word relations (such as ‘and’, ‘if’ and ‘the’), removing commonly occurring noisy words such as check-in or bedroom would significantly slim down our corpus. Thus, we came to realise that this kind of computational approach would not work very well for our type of texts. Mainly because it works best with larger textual databases and our corpus, as consisting of 21639 words in 164 listings (prior to removing noise), was not large enough to produce useful results from topic modelling. Therefore, we chose to go a different route through which we could qualitatively analyse the listings in a systematic way.
Figure 1: Our provisional coding schema
Next, we decided to perform a content analysis of the listings, using Yuval-Davis’ framework of belonging to develop a coding scheme. Yuval-Davis distinguishes between three levels: identities and attachments, social positions and values. The first level foregrounds the importance of collective as well as personal narratives for the construction of belonging; the second level hints at the existence of structural forces and power relations that are beyond one’s choice and control; finally, the third level points to the role of shared ethical and political values in defining who is accepted or not within a group. For each level, we developed a number of codes (Figure 1), and operational definitions for each code. For example, we decided that sentences through which a host shared information about themselves, their families and their reasons for hosting would be coded as Identities & Attachments | Personal Narratives. We performed two cycles of coding, examining 164 listings. Yuval-Davis’ framework gave us a lens through which to look at belonging, and consider different ways through which claims to belong can be made. What is more, this approach also allowed us to produce a good overview of common themes and trends within the listings, mitigating our biases. For example, we expected religious and national identities to be very present in the listings: coding helped us to see that in fact most listings contained almost no information in this vein. At the same time, we also found this coding approach quite limiting: on the one hand, we found that Yuval-Davis’ framework wasn’t always helpful for our case, and on the other hand we also noticed that, by directing our attention to particular paragraphs, this coding methods made it difficult to consider listings in their entirety.
We thus attempted to take a less structured but more holistic approach to the analysis, focusing on the interpretation of entire listings. We did this in a workshop setting, so that we could contribute our respective insights: knowledge of the region, knowledge of Airbnb as a platform, and knowledge of the specific dataset through the steps above. In this phase, we stepped away from our framework and coding scheme, and discussed each listing individually. This helped us to put words into context, consider the relations between sections in the listings, and notice contradictions and changes in tone; last but not least, this step also highlighted the importance of the photos included in the listings and their relation to the written text. This third approach proved very productive, but also time-consuming, to the point that we realised we need to significantly scale down our ambitions and lower the number of listings we will analyse.
And so, even though the road itself was not straight forward, we have ended up with an appropriate and effective methodological approach that is sensitive to our data’s specificities… By trying out different approaches and being reflexive we were able to better understand the characteristics of the texts and familiarise ourselves with them. This led finally to an overall approach comprising of multiple heterogeneous cycles of organisation and interpretation ranging from machine based to human methods of identifying patterns, specifics, and unique cases. Perhaps what this goes to show is that texts in digital space, specifically on a platform claiming universal belonging on geopolitically contested ground, create narratives which work on the reader and the context of their creation in complex ways which can’t be grasped with one method alone. More importantly, what becomes apparent is the fact that there is no one essence of a text or true meaning because each approach will give a different answer. Thus, when writing up a methods section for a research paper, in presenting a coherent journey, a lot of knowledge and insight is lost as the reader doesn’t gain access to the ‘behind the scenes’ and the trials of triangulation.
Author Bios
Jelke Bosma is a PhD candidate in the Department of Media Studies and the Centre for Urban Studies at the University of Amsterdam. His PhD project looks into dynamics of value and place related to short-term rentals in Amsterdam and Berlin hosted on Airbnb. He has a background in urban studies and his research interests include platform urbanism, housing, and urban theory.
Alexandra Knight is a Research Master student in Urban Studies at the University of Amsterdam. Her current thesis topic combines social reproduction theory and abolition theory, providing a lens to study neighbourhood collective actions which aim to prevent youth from criminality in Stockholm and Rotterdam. Alexandra’s research interests surround meaning making processes and the impacts of neoliberalization on the socio-spatial complex that is the city.
Valentina Carraro is Assistant Professor at the Department of Human Geography, Planning and International Development Studies, University of Amsterdam. Her research and teaching sit at the intersection of digital and political geography, with a focus on how digital technologies and practices reconfigure geopolitical relations.
