The data used here are based on population register data for the entire Norwegian population registered as resident on 1 January 2011, provided by Statistics Norway. The results are compared to corresponding figures from Sweden, Denmark, the Netherlands and Belgium, as described in Andersson et al. (2018). To facilitate comparative analyses, we have aimed to make the data as similar as possible across countries. The process of harmonizing the national data sets is documented in Nielsen et al. (2017) and was the aim of the ResSegr project.Footnote 1
The k-nearest neighbours approach to measuring segregation is well suited for comparative analyses, as it provides a comparable definition of a neighbourhood; the k-nearest neighbours of each individual. This partially circumvents the Modifiable Areal Unit Problem (Hennerdal and Nielsen 2017) by allowing for a comparison of residential patterns that do not rely on administrative borders. Further, the neighbourhoods are scalable, allowing us to study segregation at both the macro-level (k = 51,200), the micro-level (k = 200) and at intermediate levels. However, a drawback of this method is that the geographical size of each neighbourhood is determined by the local population density. Thus, the geographical area that is considered a “neighbourhood” is highly variable and may become very large at high k values, particularly in less densely populated areas in Norway and Sweden.
The Norwegian data are based on a 100 × 100 m grid covering the entire country, excluding unincorporated areas. We first calculate the total number of individuals and the number of non-European immigrants in each populated grid cell. Non-European immigrants are defined as people born in a non-EU28/EFTA country to two foreign-born parents. Using the specialized software Equipop (Östh 2014), we calculate the proportion of non-European immigrants among the k-nearest neighbours of each grid cell, producing a data set consisting of the composition of the egocentric neighbourhoods of each grid cell at different scale levels. The scale levels used here are k = 200, k = 1600, k = 12,800 and k = 51,200. In the analyses, these values are weighted by the population count of each grid cell. For k = 51,200, a grid of 400 × 400 m cells was used in order to circumvent a technical problem. Table 1 provides descriptive statistics for the grids.
Table 1 The gridded population, descriptive statistics, 2011. As mentioned above, the k-nearest neighbours approach produces neighbourhoods that are comparable in terms of population size, but highly variable in geographical size. This is clearly shown in Table 2, which summarizes the geographical size of neighbourhoods at k = 200 and k = 51,200. Norwegian neighbourhoods at the micro-level of k = 200 are roughly similar to those found in the other countries up to the 50th percentile. However, the area covered by many Norwegian neighbourhoods at this scale level is much larger than the areas of neighbourhoods in Belgium, Denmark and the Netherlands. In Norway, 10% of the population live in places where we have to draw a circle with a radius of approximately 1.5 km or more in order to encompass their 200 nearest neighbours. At the macro-level of k = 51,200, the Norwegian neighbourhoods are much larger in size than those of Belgium, Denmark and the Netherlands across most of the distribution, but they are comparable in size to Swedish neighbourhoods.
Table 2 Size of individualized neighbourhoods in Belgium, Denmark, the Netherlands, Sweden and Norway, radius in metres (percentiles based on population count), 2011. Measures of Segregation
Concentration
A concentration measure of segregation is obtained through Equipop, which calculates the proportion of non-European immigrants among the k-nearest neighbours of each grid cell. Weighted by the number of residents in each cell, the percentiles of the distribution of these neighbourhood compositions correspond to the percentile distribution of all individuals’ neighbourhood composition. The interpretation of the percentile values is straightforward; if, for instance, the 10th percentile is 1%, 10% of the population resides in neighbourhoods where 1% or less of the population are non-European immigrants.
Representation
Our measure of the representation of non-European immigrants is calculated from the percentile distribution of the concentration of non-European immigrants in a fashion identical to that in Andersson et al. (2018). Thus, non-Europeans are overrepresented in a percentile bin if the value is above 1, and under-represented if the value is below 1 (Andersson et al. 2017, 2018; Hennerdal and Nielsen 2017).
Dissimilarity Index
We calculate the DI for each k level based on the percentile distribution of our concentration measure, in the same fashion as Andersson et al. (2018). The DI is an aggregate measure of over- and under-representation that will be zero in the case of perfectly even representation, and one if the non-European population is perfectly segregated from the rest of the population.