Modern spatial data analysis has been made rather simple with the development of GIS software that allows researchers to easily map months of data points in a matter of minutes. With companies which operate primarily in a digital format, detailed data can usually be found in abundance. We use ArcMap 10.6.1 to perform all of our analysis.

For this study, we collected seven months of listings data from Airbnb ranging between October 2019 and April 2020. In order to explore the effects of COVID-19, it was important to study the state of Airbnb in Vancouver prior to the outbreak as a baseline comparison. The data was separated into listings per month and initially collected from Inside Airbnb as comma-separated values (CSV) before being converted into point vector data with provided coordinates. In addition to the coordinates in decimal degrees, each listing provided information on which neighbourhood it was located in, the unit type, nightly rate, minimum stay duration, and reviews. Due to our primary goal being to examine the change in Airbnb listings in Vancouver over time, only location data was deemed especially relevant.

Our study area was the City of Vancouver, and the city boundaries were collected from the City of Vancouver Open Data Catalogue. A key spatial exclusion would be of the University Endowment Lands due to the fact that there were no data points for the area. We elected to use census tracts taken from Statistics Canada for enumeration areas as they allowed a balance between having distinct inter-city regions and being general enough for patterns to be visible. In order to isolate CTs for British Columbia, we entered a definition query for all objects with BC’s province ID (59). With the scale of this study being hyper-local, the most appropriate map projection would be the Universal Transverse Mercator 10 (UTM10) based on the North American Datum of 1983 (NAD 1983) because of its high angular fidelity which creates an accurate representation of Vancouver.

As a secondary observation, we were also interested in exploring any relationships between relative affluence in a region and Airbnb listings. The University of Toronto’s CHASS Data Centre provided median household incomes from the most recent 2016 census for each census tract. A spatial join with the CT boundaries allowed the income data to be visualized as a thematic choropleth map. CTs were classified into 5 classes using Jenks Natural Breaks in an effort to create organic groupings of neighbourhoods.

Considering our goal to examine change in distribution of Airbnb listings, we chose to perform a series of hotspot analyses in order to determine areas of high occurrences. An analysis was conducted for each of the seven months of data and compiled into one animation which visualizes the change in listings distribution over the course of the coronavirus pandemic as has been recorded so far. As defined by ESRI, ArcMap’s Hot Spot Analysis tool calculates the Getis-Ord Gi* statistic for each point in a dataset. Each feature is considered within the context of neighbouring features. For a feature with a high value to be considered statistically significant in the analysis, the neighbouring features must also have high values (ESRI). We use the ModelBuilder tool in ArcMap to conduct the hotspot analysis for each month of data and create the resulting output which animates the changes over time (Figure 1). This style of analysis is appropriate for our purposes because it effectively visualizes the spatial distribution of our data points and with a temporal variable added, we are able to see change as it happens.

In order to conduct our analyses, each month of point data was joined to the base census tract layer and the feature count per CT was used to determine hotspots. To supplement the hotspot analysis, data was also provided in tabular format in the output. These tables detailed the exact changes in listings over the study period separated into the unique unit types. Additional observations were made by examining the change in hosts with multiple listings over the study period.

Figure 1: Our hotspot analysis workflow on model-builder