Treemaps for Exploring Spatial and Temporal Variation in House Prices
This demo is a part of our contribution (download abstract) to the GeoViz Digital City Workshop in Hamburg, 3-5 March 2009. It demonstrates how 2D ordering and layout in treemaps can be exploited to explore spatial and temporal patterns in data. London property sales over the past eight years are used as a case study.
Interactive demo
Our interactive applet allows variation in house sales and prices in London over the past eight years to be explored with an interactive treemap.
« Click the image to run the application.
The layout of the treemaps can be changed, introducing increasing amounts of geographical space: standard squarified, use of temporal ordering, use of spatial ordering and the use of more conventional maps. The effect of thee various orderings can be evaluated with this applet.
All the treemaps use size to represent relative number of sales and colour to show different characteristics of the property (orange is used for average price).
Data and hierarchy
The data are for residential property sales and are sourced from the UK Land Registry. We have summarised the individual transactions by London Borough (administrative region), year and month. The treemap present this as a hierarchy, the depth of which can be changed using the three "hierarchy" buttons. Note that to explore the data more fully, alternative hierarchies are required (not supplied in this demo).
Controls
- To zoom drag up/down with the right mouse button
- To pan, drag with the left mouse button
- To reset zoom, press 'r'.
- To toggle on/off of highlighting corresponding areas, press 'h'
- Mouseover areas to see details
Labels' visibilities are dependent on zoom level. Months will be labelled when zoomed in far enough.
Size
In all cases (other than the map layout view, see below), size corresponds to the number of sales.
- Borough: Wandsworth has the most sales over the eight year period (the largest area in the treemap) and the City of London has the least (contains very little residential property).
- Years: There are similiar numbers of sales in each year so sizes are similar. The most obvious is 2003 for City of London (need to zoom in).
- Months: Again, there are similiar numbers of sales in each month.
Colour
The six "colour" buttons control the variables by which the treemap is coloured:
- Aboslute number of sales (purple);
- Average price (orange)
- Standard deviation of price (green)
- Coefficient of variation of price (red)
- Deviation from number of 'expected' sales (blue=fewer; red=greater), using the chi-statistic. Expected values were assumed to be:
- At the borough level: total sales (all borough and all years) weighted by the number of residential properties in each borough
- At the year level: an eighth of the total number of sales for that borough (there are eight years)
- At the month level: a twelfth of the total number of sales for that year in that borough
- Deviation of average price 'expected' average price (green=fewer; red=greater), using the chi-statistic. Expected values were assumed to be:
- At the borough level: average price for all boroughs and all years
- At the year level: average price for that borough
- At the month level: average price for that year in that borough
The maximum and minimum values of the colour scheme can be altered by dragging the max/min values on the legend. This is helpful for characteristing variation where local variability is low.
Using the chi-statistic values, variations in the number of sales and average price are apparent that are not obvious from the absolute values. For example, this is the case for yearly and monthly variations in the number of sales.
Layout
Use the six layout buttons to switch the layout of the treemap with increasing degree of spatial layout (from left to right):
- standard squarified (boroughs with the largest and smallest numbers of sales are apparent and the 'squarified' nature make relative sizes easier to compare, but spatial and temporal patterns are not apparent.
- adds temporal ordering, with years ordered in vertical strips (2000 on the left to 2007 on the right) and months as horizontal strips (January at top, December at bottom). Temporal patterns by borough are revealed (note that there is no layout different for hierarchy level 1)
- adds spatial ordering, positioning boroughs in their approximate geographical configuration.
- adds spatial positioning which shows the true geographical positioning of boroughs, but results in high visual occlusion (since areas are not sized geographically.
- a Gastner cartogram view sized by number of sales (for the borough level only).
- a map, the only layout which does not size areas by the number of sales.
Acknowledgements
The house price data are Crown copyright and are reproduced with the permission of Land Registry under delegated authority from the Controller of HMSO. The spatial data are Crown Copyright/database right 2008 from an Ordnance Survey/EDINA supplied service. The applet is based on Processing. Thanks for Andrew Crooks from CASA for obtaining the house price data.