Appendix 1 – Data and supplementary analysis
This Appendix presents further information about the secondary data used in the main report. It provides details of the data source used, how it was validated and filtered, and how the research team interpreted it. Potential limitations of the data are also discussed. In addition, a number of important questions arising from the secondary data analysis are answered, in relation to technical questions such as 'what is a listing?' and analytical questions such as 'where is the greatest concentration of Airbnbs in Scotland?'. An additional breakdown of Airbnb listings by individual council area is provided. This is followed by more detail on the location of Airbnb listings in the four council areas within which the case studies are located, with respect to socio-economic composition.
In May 2019 Murray Cox, of Inside Airbnb, was commissioned by Indigo House to provide the research team with Airbnb data for the whole of Scotland. This process involved collecting information for each Airbnb listing in Scotland using a well-established automated collection process. This is typically referred to as 'web scraping', 'web harvesting', or 'web data extraction', and it is the process whereby live data is collected from a website and collated, often for further analysis. The data were therefore provided by a third party, but sourced directly from Airbnb's website.
For each Airbnb listing the research team had the full URL, the property type (e.g. 'Cottage', 'House', 'Apartment', 'Tent', 'Campervan', 'Castle'), the room type (e.g. 'Entire home/apt', 'Private room') and a large number of other variables including the advertised price per night, when the calendar was last updated, full text descriptions of each listing and many more. The latitude and longitude of each listing were also included, so that each listing could be located. The raw data was provided as an aggregated text file (in csv format) of online Airbnb listings from Scotland as advertised online in May 2019.
Validation and filtering of data
The dataset used in this research covers the whole of Scotland. It is important when attempting to understand the rise of short-term lets (STLs) in Scotland (or elsewhere) to be as objective as possible when analysing and interpreting data. An important part of this is filtering and validating the data. The veracity of data used to explore the rise of STLs has been challenged in the past and for this reason a deliberately cautious approach was taken here.
Previous research by members of the research team demonstrates that this kind of data closely resembles data published by Airbnb already in the public domain. That being said, the secondary data analysis in this report covers only volume and locations of STLs across Scotland, while the qualitative analysis conducted by the research team seeks to understand their impact and provides richer insights in relation to the experiences of local residents, STLs hosts, community actors and businesses in five locations.
The reporting of secondary data above (Chapter 3) relates to location (where listings are located), volume (the number of listings in each location), type (full property, shared property, shared room), and a comparison of these elements to the number of dwellings in any given area. The complete Inside Airbnb dataset for 2019 had 35,474 listings for Scotland (see Table 1). However, there is no way to be completely sure that a listing on Airbnb is 'active' in the sense that it is currently available for STLs or is likely to be in the near future. The approach to identifying an 'active' listing involved including only those listings where the host had updated their property availability calendar in the last six months. This resulted in a figure of 31,887 for Scotland. Upon further analysis it was found that three of these listings were actually just across the border in England, so the final figure for the total number of Airbnb listings in Scotland as of 19 May 2019 was 31,884.
Table 1: Analysis of Airbnb listings by last calendar update
|Calendar updated||Listings (19 May 2019)||Cumulative total||% of all active||% of all listings|
|2 days ago||640||16,104||50.5||45.4|
|3 days ago||1,059||17,163||53.8||48.4|
|4 days ago||950||18,113||56.8||51.1|
|5 days ago||836||18,949||59.4||53.4|
|6 days ago||384||19,333||60.6||54.5|
|1 week ago||2,411||21,744||68.2||61.3|
|2 weeks ago||2,467||24,211||75.9||68.2|
|3 weeks ago||1,563||25,774||80.8||72.7|
|4 weeks ago||1,053||26,827||84.1||75.6|
|5 weeks ago||790||27,617||86.6||77.9|
|6 weeks ago||665||28,282||88.7||79.7|
|7 weeks ago||366||28,648||89.8||80.8|
|2 months ago||1,143||29,791||93.4||84.0|
|3 months ago||851||30,642||96.1||86.4|
|4 months ago||583||31,225||97.9||88.0|
|5 months ago||365||31,590||99.1||89.1|
|6 months ago||297||31,887||100.0||89.9|
|More than 6 months ago||3,587||35,474||111.2||100.0|
Source: Inside Airbnb, May 2019
The figure of 31,884 was then compared to data in the public domain published by Airbnb. According to Airbnb's UK Insights Report 2018 (p. 30) there were 31,000 'active' listings in Scotland in 2018 so, taking into account continued growth since 2018, the research team are confident that the final figure is a fair reflection of the current state of the sector in Scotland. Individual figures for Scottish council areas published by Airbnb in their UK Insights Report 2018 state that there were 2,700 'active' listings in Glasgow and 10,500 in Edinburgh in 2018, so at the local level it also appears the data used here are accurate. It must be borne in mind, however, that it is likely to be some seasonal variation in the number of listings, hence the figure of just under 10,000 for Edinburgh reported by the research team may not reflect listings which came online after May 2019, ahead of the annual Fringe Festival.
The veracity of 'scraped' data in particular has come under scrutiny, possibly because the term implies a lack of rigour or due process. This may be the case if one were seeking to make inferences on money generated per host, occupancy rates or other information not in the public domain. However, in this analysis the research team mainly looked at volume (number of listings) and location of listings (where active listings are located). Part of the secondary data analysis also involved validating the data by batch checking listing URLs to make sure they were still active at the time of analysis (July 2019). The research team also conducted some validation of the data as part of the secondary data analysis phase of the study. This involved manually checking 400 listing URLs, 100% of which were exactly as described in the raw data file. For this, a random sampling method was then used to investigate listing URLs in more detail to check the veracity of the information contained for each listing. This process confirmed that the data provided matched live listings advertised on the Airbnb platform.
Another part of the process involved geographic validation. Each listing has a location attached (latitude and longitude), but these are known to be randomly displaced by up to 200 metres by Airbnb. However, the analysis demonstrated that in the majority of cases the geographic displacement was in fact much less than this, particularly in densely populated urban areas like central Edinburgh and Glasgow. In many cases the displacement was less than 20 metres. The research team found that in rural locations (particularly in Skye) the level of geographic displacement was sometimes greater, and in some cases exceeded 200 metres.
However, since wards were used as the lowest level of analysis and all listings fell within individual wards, this did not give rise to concern at the national level. At ward boundaries in Edinburgh or Glasgow, and other dense urban areas, a small proportion of listings were likely to have been displaced in or out of individual wards, but on the whole this analysis suggests the impact of such displacement is minimal. Based on further investigation, a figure of up to 5% for geographic displacement was arrived at. Further analysis using smaller geographic areas would be less reliable and that is why it has not been attempted here.
Some further manual checking was conducted in relation to the accuracy of room type, property type, location and availability data. This involved inspecting a random sample of 150 listing URLs to make sure the data matched the original spreadsheet. In every case, it did.
As a final piece of validation, and particularly for properties with multiple listings within them (e.g. as with one house in the Highlands) the team compared the location of a listing (from the URL provided with from Inside Airbnb) to its location on Google Street View, plus external photographs of the listing. This Google Street View survey also confirmed other important local details, such as the emergence of key boxes in Edinburgh. This was achieved by comparing historic Google Street View images from when it was first captured in 2008, to the most recent imagery online today.
Limitations of the data
It is important to be clear on three crucial points of interpretation in relation to the secondary data. Fundamental to these points is that an individual Airbnb listing does not necessarily equate to an entire property and, even where it does, more information on its availability would be required in order to fully understand the impact on the wider property market.
First, the analysis presented here allows one to say something about the percentage of dwelling stock in Scotland that STLs have a presence in. The question of whether this represents an overall net loss of longer-term rental stock is not possible to discern from this data alone since it is impossible to tell whether properties are let occasionally, or permanently as STLs.
The second point is that some private rooms, listed separately on Airbnb, may be located in the same dwelling, though it appears that this is the case for 6% of listings at most. The research team arrived at this figure by identifying all active Airbnb listings with exactly the same geographic coordinates and then looking at individual property characteristics for these 1,900 listings in order to estimate how many listings were in the same building. However, without access to raw data from Airbnb or other platforms it is impossible to say for certain which of the listings in the same geographic location are in the same dwelling (and not just in the same building with identical coordinates).
The third issue relates to entire homes or apartments and the fact that some listings do not relate to traditional dwellings and include tents, camper vans, caravans, yurts, boats and pods. Based on the analysis presented above, it appears that these account for just under 2% of all listings across Scotland. This figure was arrived at after conducting a more in-depth analysis of property type (e.g. 'house', 'apartment', 'tent', 'yurt', 'camper van'), which is one of the fields included in the secondary data provided to the research team by Inside Airbnb. The research team has not excluded these from the analysis on the basis that it is impossible to be certain which of them would qualify as a 'dwelling' from a statistical point of view. Such a decision would require further qualitative analysis on a listing-by-listing basis.
Further questions arising from secondary data analysis
It is important when looking at secondary data on the subject of STLs to be as careful as possible, with respect to interpretation. After thorough analysis of the most recent data for Scotland, it is useful to consider what the data does, and does not, allow one to infer. These are presented here as a short series of questions and answers.
Where is the greatest concentration of Airbnbs in Scotland?
By total volume, the answer to this question is Edinburgh City Centre ward (2,710 listings), and in particular the Old Town. By rate of penetration (i.e. compared to dwelling count), the answer is Skye (18.6% penetration rate for all listings, 11.4% for whole properties). If one looks at individual square kilometres, the highest density is in an area of the Old Town in Edinburgh, in which one single square kilometre area has 812 active Airbnb listings. It is not possible to derive a penetration rate for individual square kilometres because data on dwellings are not available at this level of granularity.
What is a 'listing'?
It is important to understand that there are many different types of property listed on Airbnb. A 'listing' refers to a single web address which advertises accommodation for visitors. This may be a full property, a shared property or a shared room. In this sense, the word 'listing' is a different term for what people may refer to as 'an Airbnb' or a 'short-term let'. So, a listing may be an entire property, but it could also be a single room in a shared house. It would be wrong to assume all listings are individual properties. But it is also important to remember that some listings are not in traditional dwellings at all, with a small number of listings being for pods, tents, camper vans or boats. However, the vast majority of listings are in traditional dwellings such as flats or houses and the majority (69%) of these are for entire properties.
Do many hosts have multiple properties?
The simple answer to this question is that, according to the analysis presented above, 45% of Airbnb hosts appear to have more than one listing. However, there are some complicating factors here. Typically, a 'host' might be thought of as being an individual with a spare room or property that they rent out for part of the year. Based on the qualitative and quantitative analysis conducted here, this is sometimes true, and is an important part of the sharing economy. Yet it is also the case that some Airbnb listings are advertised not by individual hosts but by intermediaries acting on behalf of individual owners. For example, some large holiday cottage companies use Airbnb to advertise listings, from which many are likely to be owned by individuals. Thus, the proportion of Airbnb listings owned and operated by individual hosts as opposed to property professional cannot be definitively calculated.
How long are properties available for?
This is an important question but it is not possible to provide a definitive answer based on the data available. However, it should be noted that even if a property is only listed on Airbnb for a fixed number of days per year (e.g. 180) it may be available for short-term let on other platforms.
Does the presence of an Airbnb listing mean that a property has been permanently removed from the private rented sector?
The intensity of Airbnbs in some areas of Scotland, combined with the qualitative data presented elsewhere in this report, suggests that there has been some switching from long-term to short-term rentals. In Skye and in central Edinburgh, this switch appears to have been quite profound. However, a more precise answer to this question cannot be determined from the analysis of secondary data currently available.
How many Airbnbs are there in Edinburgh?
The secondary data analysis conducted here has arrived at a figure of just under 10,000 active Airbnb listings for the City of Edinburgh, as of May 2019. If all listings are included, even those with no calendar updates in the past 6 months, or recent reviews, the figure is 12,600 exactly for May 2019. However, the analysis of active listings suggests a figure of 6,622 entire properties, 3,314 private rooms and 58 shared rooms on Airbnb within the boundaries of the City of Edinburgh – a total of 9,994. The Edinburgh data from May 2019 forms part of the wider Scotland dataset on Airbnb listings used in this report, but there is also a more recent set of data for the City of Edinburgh alone, published in June 2019.
If one looks at more recent data for Edinburgh on Inside Airbnb (25 June 2019), there are a total of 13,245 listings. This raw figure includes all listings, regardless of availability or calendar updates. There are a number of ways to filter the data, including:
- Excluding those listings with zero calendar availability in the next 365 days
- Excluding those listings with no reviews: a) from 1 January 2018 to May 2019, b) in the past 12 months, or c) ever
- Including only those listings where the calendar has been updated in the last 6 or 12 months.
It is also possible to use a combination of the above criteria to filter the data, yet no matter what approach is taken the number of listings deemed to be active is always substantially lower than the headline figure.
The table below shows the effect of using these different filtering methods. The analysis presented in this report uses calendar update in the last six months as a proxy for identifying an 'active' listing, but of course it is possible to use other methods, each of which has its own merits. The highest figure in the table relates to listings with a calendar update in the past 12 months, but once listings with no availability in the next 365 days are removed the figure drops from 11,734 to 9,592 for Edinburgh. So, even if as of June 2019 there were more than 13,000 Airbnb listings for Edinburgh, it appears that the figure for active listings remains substantially lower, and likely to be in the region of 10,000.
Table 2: Different filtering methods for identifying 'active listings'
|Filter criteria||Number when filtered||Total unfiltered listings||% of listings retained|
|Exclude listings with zero availability in next 365 days||9,714||13,245||73.3%|
|Include only those listings reviewed in 2019 or 2018||9,787||13,245||73.9%|
|Include only listings with a calendar update in the last 12 months||11,734||13,245||88.4%|
|Include only listings with calendar update in the last 6 months||10,713||13,245||80.9%|
|When listings with no reviews, ever, are removed||11,213||13,245||84.7%|
|When listings with no reviews in the past 12 months are removed||9,414||13,245||71.1%|
|Calendar updated in the last 12 months + removing zero availability in next 365 days||9,587||13,245||72.4%|
|Calendar updated in the last 6 months + removing zero availability in next 365 days||9,385||13,245||70.9%|
|When listings with no reviews, ever, are removed, but including only those with calendar availability over the next 12 months||10,081||13,245||76.1%|
Source: Inside Airbnb, May 2019
2. Additional analysis
Analysis of Active Airbnbs by type and Local Authority area
A breakdown of active Airbnb listings is provided here on a council-by-council basis. Given the potential importance of local authorities in the administration of STLs, it is useful to understand any differences in relation to the mix of entire home versus individual room listings. The highest figure for entire homes or apartments was in Dumfries and Galloway (83.5%) and the lowest was in Midlothian (47.8%) – the only area in Scotland where private rooms are the main type of Airbnb listing.
Table 3: Active Airbnb listings by council area, May 2019
|Council Area||Entire home/apt||Private room||Shared room||Total||Max category|
|City of Edinburgh||66.3||33.2||0.6||100||Entire home/apt|
|Glasgow City||65.9||33.8||0.3||100||Entire home/apt|
|Argyll and Bute||76.0||23.3||0.7||100||Entire home/apt|
|Perth and Kinross||79.3||20.4||0.3||100||Entire home/apt|
|Dumfries and Galloway||83.5||16.3||0.2||100||Entire home/apt|
|Scottish Borders||76.5||23.2||0.3||100||Entire home/apt|
|Aberdeen City||64.0||34.8||1.1||100||Entire home/apt|
|East Lothian||77.2||22.8||0.0||100||Entire home/apt|
|Na h-Eileanan Siar||77.4||22.0||0.6||100||Entire home/apt|
|South Ayrshire||78.7||21.3||0.0||100||Entire home/apt|
|North Ayrshire||75.5||24.5||0.0||100||Entire home/apt|
|Orkney Islands||66.9||33.1||0.0||100||Entire home/apt|
|Dundee City||56.7||41.3||2.0||100||Entire home/apt|
|Shetland Islands||72.4||27.6||0.0||100||Entire home/apt|
|South Lanarkshire||67.0||32.5||0.5||100||Entire home/apt|
|West Lothian||57.8||42.2||0.0||100||Entire home/apt|
|North Lanarkshire||67.3||32.7||0.0||100||Entire home/apt|
|West Dunbartonshire||61.7||38.3||0.0||100||Entire home/apt|
|East Ayrshire||68.8||31.2||0.0||100||Entire home/apt|
|East Dunbartonshire||60.3||39.7||0.0||100||Entire home/apt|
|East Renfrewshire||61.1||38.9||0.0||100||Entire home/apt|
Source: Inside Airbnb, May 2019
Airbnb listings by SIMD16 decile
An important question about the presence of STLs, and their recent rapid growth, is where they are located. On the one hand, the data show that there is significant concentration of a high number of Airbnb listings in a relatively small number of areas. Yet this does not say anything about the characteristics of the areas themselves. In order to shed more light on this issue, the location of all 31,884 Airbnb listings from May 2019 was cross-referenced with data from the Scottish Index of Multiple Deprivation (SIMD) 2016 in order to understand the extent to which they are clustered in more or less deprived neighbourhoods.
In the four local authority areas within which the five case studies are located, there is some variation (Figure 1). In Edinburgh, more than a quarter (26.6%) of all active Airbnb listings are located in the least deprived decile. Edinburgh's most deprived decile contains 1.5% of all listings (a total of 146). In Glasgow, the most deprived decile has more Airbnb listings (14.5%) than any other individual SIMD decile. In Highland, most Airbnbs are somewhere in the middle of the SIMD rankings, with very few in either the most or least deprived decile or quintile. In Fife, the pattern is more similar to Edinburgh, with the least deprived areas having the majority of these kinds of STLs.
Figure 1: Airbnb active listings by SIMD decile in case study council areas (where decile 1 is most deprived and decile 10 for the least deprived)
3. Data from other online accommodation platforms
This secondary data analysis, which forms only one part of this research, is based on data regarding the Airbnb platform. Airbnb are the leading online provider of STLs, but of course they are not the only online accommodation platform to offer this: Booking.com, TripAdvisor, tripping.com, FlipKey, HomeAway, and a range of other similar websites also offer STLs. Researchers have no way of knowing what proportion of the short-term lettings sector Airbnb accounts for but at the very least it would appear from the data analysed here, in addition to the results of the qualitative research, that it accounts for a very high proportion of it across Scotland.
Traditional self-catering accommodation also forms part of the sector as the analysis of hosts from the Airbnb data demonstrates that several traditional holiday companies use Airbnb as a route to market. For example, analysis of secondary data by the research team showed that 950 Airbnb listings (3.0% of the total) were advertised by Cottages.com on Airbnb, and 743 by Sykes Holiday Cottages (2.3%).