Ana Petrović, Maarten van Ham (TU Delft), David Manley (University of Bristol)
What is the relationship between your neighbourhood and your individual income? This research question was addressed in this pilot project (completed fall 2019).
To investigate this, the researchers used register data for the entire population of the Netherlands. The data were geocoded into 100m by 100m grid cells and aggregated at 100 different spatial scales. Based on that information, the researchers created concentric circles around each individual cell, with radii ranging from 100m up to 10km. For each spatial scale, two contextual characteristics – the share of residents with a non-Western background and the share of low-income earners – were calculated and linked to individual characteristics such as income. The results differ widely across different spatial scales, showing the importance of measuring spatial contextual characteristics at various spatial scales.
Computers with at least 64 GB of working memory were needed to analyse those large data sets. Because this project involved various combinations of data, a cluster of 25 nodes with 24 cores and 64GB of working memory each was deployed. In total, 600 processor cores were applied, a huge increase in capacity in comparison to laptops, which usually have 2 or 4 cores. This cluster allowed the analysis to be completed in a week and a half.
Performing the analyses in the Remote Access environment of Statistics Netherlands (CBS) would have taken between four months and two years. Processing time was thus considerably shorter with the ODISSEI Secure Supercomputer (OSSC). For researchers, this meant that more variables covering longer periods of time could be used for the analysis of spatial contextual effects.