Merging statistics and geospatial information, 2014 projects - Norway
This article forms part of Eurostat’s statistical report on Merging statistics and geospatial information: 2019 edition.
Final report 28 February 2017
Full article
Problem
Quality of life perception surveys can be used to determine attractive urban areas. However, they are a relatively expensive and time consuming task and generally result in aggregated data for the whole of a city, rather than more detailed information on specific neighbourhoods within a city. Statisticians have considered testing alternative data sources — such as statistical registers and georeferenced data — to see if these can be used to make the information gathering process more effective and efficient.
Objectives
The general objective of this project was to combine relevant statistical registers and geo-referenced data in order to identify attractive urban areas. Its aim was to develop an innovative procedure for assessing how changes in population and land use in urban settlements relate to quality of life parameters.
Method
The first step was to describe the quality of data sources and the possibilities for combining these into a dataset for measuring the attractiveness of urban areas. From this a conceptual model of the data structure and data format for a dataset of urban area attractiveness was developed.
A methodology for producing a dataset on the attractiveness of urban settlements was developed. Housing prices (total price and price per m²) of sales in 2014 were used as proxies for attractiveness, as they reflect supply and demand and are a numerical representation of some kind of attractiveness. Intrinsic characteristics of a dwelling such as its area, condition and so on also influence price but are less likely to determine a neighbourhood’s attractiveness.
This project investigated whether there is a variation in house sales prices dependent on location within (not between) nine Norwegian cities, seeking to explain this by correlating price and place with factors related to the dwelling and its location.
In total, six categories of variables were considered: the dwelling itself; distance to geographic entities (such as the city centre, recreational areas, or a coastline); distance to buildings often providing services (such as schools, universities, or restaurants); intensity-environment (such as noise levels or hours of sun); population characteristics within 250 m radius (such as average income, education level, migration); and employment (number of employees within a 5 or 10 km radius). The indicators relating to a dwelling’s sales price, floor space (in m²) and age were available from real estate sales data, which were then combined with data from the geo-referenced property register in the Cadastre to get the XY coordinates of the centroid for each property. A large range of sources were used to provide data for the variables covering the five categories related to the location (rather than the dwelling itself).
In general, migration within a city was problematic as an indicator for measuring attractiveness, as the availability of housing in an area is quite often more related to urban planning issues than the area’s attractiveness. New dwellings are not necessarily built in the city’s most attractive areas, for example due to space issues and urban planners may wish to offer more affordable housing. Equally, new building permits are not necessarily located in the areas of a city that are perceived to be the most attractive. As such, migration and building permits were not selected as explanatory variables.
An ordinary least squares regression analysis was performed for the two dependent variables (total price and price per m²) and the adjusted R-squared calculated for each of the explanatory variables that were to be tested.
Results
The results of the analysis showed that the explanatory variables (other than those related to the dwelling itself) explained slightly more of the variation in total house prices than the price per m². Furthermore, they explained more of the variation in prices for cities of at least 150 000 inhabitants than for cities with fewer inhabitants.
Education levels and household income were found to be strong indicators of variations within cities for the total price of a dwelling. These were particularly important in Oslo indicating that it is more socioeconomically divided than other cities. Equally, floor space (in m²) was also found to be a strong indicator of total price variation. Five other variables were found to be significant for determining total prices in Oslo, but to a lesser extent: the distance to restaurants, the city centre and to water, the mean age of the population, and the age of a building (in years or those dwelling simply built before the Second World War). No other variables (that were tested) were found to be significantly important in Oslo for the total price variation. Notably, the three perception survey variables (schools and higher education establishments, health services and public transport) were not significant and similar results were found for these variables in the other eight cities studied; it appears that the distance to these services is generally short enough throughout the cities that it does not significantly affect total sales prices. Equally, access to recreational areas, noise levels and employment opportunities were not significant within individual cities.
For the price per m², income level was not found to be a significant variable, whereas floor space and education level remained significant. There was a correlation between price per m² and variables related to distance, notably the distance to the city centre as well as the distance to restaurants (reflecting local centres) and higher education establishments (although many are centrally located). However, the distance to hospitals was not conclusive as a measure of attractiveness, suggesting that they are not necessarily perceived as an attractive neighbour and/or that they are not located in attractive areas.
Based on a regression analysis, two attractiveness indices were calculated for several cities, one based on the total sales price and the other based on the price per m². This was done by combining the coefficients calculated from data on house sales with data for the full housing stock in these cities using the geo-referenced building register in the Cadastre. The predicted data for the individual buildings were then averaged within 500 m x 500 m grid cells to classify each grid cell within a price decile: the top decile shows the 10 % of grid cells with the highest average predicted prices (total or per m²) and the bottom decile the 10 % of grid cells with the lowest average predicted prices. The results for Oslo are shown in Figure 2.
Direct access to
Methodology
External links