How we read maps and dataviz - new research needed?

There's a fairly long academic tradition of looking at how humans interact with maps but, in my view, there is a need to revisit such research in relation to the new wave of digital mapping and dataviz currently available online. Some of it is fantastic and some less so, but this isn't about being critical of the bad stuff. Instead, I'm hoping others will share what they've been doing or what they've seen (via @undertheraedar) to try to understand the effect of new dataviz/mapping on how we perceive/read maps - and what impact this might have on cognition/understanding of underlying issues. 

Early last year I had some discussions about this with a very helpful colleague in psychology at Sheffield - Megan Freeth - and I gave her one of my blog images to test with her eye tracking technology. The results are shown below, in sequence (click to enlarge). I've also put them together in a slide show if you want to download them all at once.

The original 3D image

Scan path from first 10 seconds of map viewing

Scan path for one minute of map viewing

Heat map showing areas focused on most

'Region of interest' analysis

I'm aware that I am probably just not up to date with the kind of research being done in this area but before going further I should say that I am aware of people across the world who have done work in these fields - e.g. Alan M. MacEachren and others at the GeoVISTA Center at Penn State and this study from Brodersen et al at Risø National Laboratory in Denmark - but I'm not aware of what's been done in the last 4 or 5 years in particular to help us understand the effects of new approaches to mapping and visualisation on cognition and perception.

Are we understanding more because of the new wave of mapping and dataviz? Are we understanding less? Are we just enjoying how things look and being wowed by the technology more than we are critically engaging with the underlying content? Has the method become the message?

I'm as guilty as anyone of posting maps and images on twitter and this blog without necessarily thinking too much, though my aim is always to inform and engage - but as the protagonist in David Lodge's Changing Places says, "Every decoding is another encoding" and my visual 'decodings' of spatial data will always be 'encoded' by the viewer in ways I might not have expected - or even want. It's always interesting to see how people interpret things and whether this aligns with what we'd hoped. This perception issue might also come up tomorrow when one of my maps appears in the new HS2 report in the UK - we'll see.

Anyway, thoughts and insights welcome via @undertheraedar.

The Age of Buildings in the City of Chicago

Following my last post, on the geography of New York City, I've been exploring other building-level datasets to see what they can offer us in relation to telling us more about the fabric of the cities we live in. This time, I've focused on Chicago's 'Building Footprints' dataset. It's not nearly as detailed as New York's PLUTO data but it does include variables on (e.g.) number of floors and year built. As with the NYC data, it is not perfect but we can still make good use of it to understand the development of the city and its structure. I've mapped the city using number of floors as a proxy for height and shaded it by building age to produce the following overview (blues = older buildings, reds = newer).

Even bigger version

Besides looking relatively interesting, the above graphic also reveals something about the phased construction of the City of Chicago and - possibly - something more about the data itself. As with the New York City PLUTO data, I produced a chart of the 'year built' column just to give me some idea of its distribution. It looks better than the New York City chart but I'm still not convinced it is 100% accurate (the year built data run from 1852 to 2010). 

Were there really nearly 15,000 buildings constructed in 2006 and only 78 in 2000? Possibly, but it would be good to know more about the accuracy of the data. In total there are 820,154 building footprints in the dataset and there are a range of different columns - which you can read more about in the metadata file. Once again, it's pretty cumbersome to work with in a normal desktop GIS setting but my machine can just about handle it. 

The Geography of New York City

In June 2013, the city of New York released as open data one of the most detailed, fascinating and user-friendly datasets ever. The Property Land Use Tax lot Output (PLUTO) dataset is essentially a record of every parcel of land in the city, what is on it and who owns it - but this is only part of it. See the full PLUTO data dictionary for more on this. Wired said the mapping elite were 'drooling' over it and there have been a few impressive visualisations already but I was keen to look at the data in more detail and then map land use patterns and get to grips with the dataset more generally. So, as an initial experiment, I mapped all 11 land use categories for the whole city in 3D (PLUTO has a field for number of floors so the maps below are extruded on this basis). Click on an image to enlarge and then flick through the images to compare land uses.

I've also put these images in a PowerPoint file in case anyone finds it useful... These visualisations in many ways tell us what many New Yorkers already know but the PLUTO data (n.b. I've used the ready-made MapPLUTO shapefile) offers everyone for the first time the opportunity to explore this open data and examine the geography of New York City as a whole in much more detail. 

Some further information about the dataset. There are 857,879 rows in the complete dataset and the MapPLUTO version has 85 fields so if you want to work with it then you better have a good computer. When you go to the download page you'll notice that the PLUTO dataset is available as one csv file while the MapPLUTO data is split into the five boroughs of New York City. 

This is an amazing resource but it is not perfect - as the Department of City Planning recognise when they say 'PLUTO is being provided ... for informational purposes only'. The data are only as good as the sources, and sometimes when you look closely things seem a little strange. For example, here's what you get when you chart the YearBuilt column for all buildings constructed since 1800 (click to enlarge). It's hard to tell but I reckon that from about 1980 onwards the YearBuilt column is pretty accurate but before that is is something of a best estimate - though I'd be happy to be proven wrong on this!

I'll probably come back and explore this again soon but that's all for now...


Footnote: 0.4% of tax lots and 1.0% of land remains unclassified. I produced the 3D maps in ArcScene and then annotated them in GIMP. I've just done these to explore at a basic level the characteristics of the dataset and the geography of land use in New York City.

Employee Growth in London, 2001 to 2012

The Office for National Statistics has released a new dataset on the number of employees across London's 983 MSOAs. The data are sourced from the Inter-Departmental Business Register (IDBR) and they reveal some interesting trends. Naturally, I had to do a 3D map of this, so take a look at the image below for the obvious growth points...

Massive absolute growth in the City of London and Canary Wharf - and some other central MSOAs in Camden, Southwark and Westminster, but also massive growth in employment in Uxbridge.

Click on the image to enlarge

The number of employees in the City of London increased by 36%, compared to 267% in Canary Wharf and 170% in an Uxbridge MSOA. By contrast, one part of Islington had 78,600 employees in 2001 but only 57,000 in 2012 - a drop of 27%. This area of Islington is immediately north of the City of London and includes Clerkenwell and Finsbury.

If you're interested in this kind of thing it's definitely worth looking at the original dataset.

US Election 2012 County-Level Results

I've been experimenting with the 2012 US election results at the county level* published on the Guardian Datablog and comparing percentages for Obama and Romney. One of the most striking things is how Obama won in DC (over 91.4% of the vote) and in the Bronx, NY (91.2%). The highest percentage for Romney was in King County, Texas where he won 95.9% of the vote. The image below shows these patterns and also includes some information on race, with % Hispanic and % African-American mirroring, to a large extent, the percentages of voters choosing Obama. It's not really that simple of course, but there is a correlation. One interesting nugget here is the difference in total votes won in counties with the highest voting percentage for each candidate. In the 5 counties with the highest Obama percentage, almost 850,000 voted for Obama. By contrast, in Romney's top 5 counties the total was just under 20,000 voting for him.

Full size version here

What does any of this mean? It means that the Republican Party probably need to think about how to do better in cities, with Hispanics and with African-Americans, but they already know that. Romney was very successful in areas where not many people live but not successful enough in major cities. It's all pretty obvious but it stands out more when you look at it on a 3D map!

*(for the continental United States, so no Alaska or Hawaii for now)

Shannon County, South Dakota: Democrat Stronghold

I've talked about 3D mapping here before, and why I like it, so I thought it was about time for another one. Since the US presidential elections have just taken place I thought I'd look at some of the data and make some maps. I'll post some more when I have time but for now I thought this 3D map showing the ratio of Obama to Romney voters at the county level was pretty interesting, not least because it identifies an interesting high point in South Dakota. 

Click here for a full screen version

South Dakota is a Republican stronghold but Shannon County is the exact opposite. In both 2004 and 2008 it had the highest Democratic voting percentage in the United States (over 85% both times) and in 2012 more than 93% voted for Obama. This might not come as a surprise when you discover that Shannon County is located entirely within the Pine Ridge Indian Reservation and that out of around 8,700 registered voters, 6,500 are registered Democrats. 

One advantage of using a 3D choropleth is that you can often differentiate between places within the same class in a way that is impossible with a normal 2D choropleth. It usually makes it very easy to see places that stand about - such as Shannon County - and it is a powerful way to visualise this kind of data. There are down-sides too (visual occlusion being one of them) but the continental United States is a nice shape with nicely divided counties, so it works well there.

Beyond Shannon County, the Obama strongholds with real weight in terms of population are DC and the Bronx. In both cases Obama voters outnumbered Romney voters by more than 10 to 1 - 222,332 to 17,337 in the case of DC and 264,568 to 24,430 in the Bronx.

More on this kind of thing coming in the near future...

Why I Like 3D Maps

I do quite a bit of spatial analysis and mapping in my academic research, and some of it ends up on this blog. Over the past few years I've done quite a few 3D* maps - most recently one of population density in China. A comment by map-guru James Cheshire made me think about the 3D issue, hence this post, which attempts to say a little more about why I like using the third dimension, as it were. Also, there's not much about this stuff online at present. The two images below show population density in Europe at NUTS3 level, with a colour scale running from red (high density) to blue (low density). Click the images to enlarge.

*N.B. Data for some parts of Italy, Germany, and the UK are missing, 

but that doesn't matter for now - this is just an example.

The reason I like the addition of the third dimension with this kind of dataset is that you can tell more about the differences between areas within the same statistical category. Essentially, it adds an additional dimension of information that you can not observe from the conventional 2D map above. This is particularly true of the most high density areas in the first map. There is of course an issue here about the relative size of areas and how this might change the population density of different places but that is a different matter since I'm not in control of NUTS3 definitions! For more on this kind of thing I'd recommend looking at Stan Openshaw's work and for more on the utility (or futility?) of choropleths generally Tobler (1973) is an excellent starting point. Gale and Halperin (1982) is also worth a look.

If we assume that the main purpose of a choropleth map is to present and discover spatial patterns then there is sometimes a strong case for using the vertical dimension and extruding polygons using a z-variable (I do this in ArcScene, in case anybody is interested). However, there are some complications and I don't think it is always appropriate to go 3D. For example, depending upon the spatial structure of your data extruded polygons in one area can obscure those in another. There is also the issue of the different size of areas and the way these might have an impact upon the level of extrusion - i.e. if we used 1km cells for the European population density map it would look rather different in 3D - though this is possibly another artifact of the modifiable areal unit problem, as described by Openshaw. I've patched together three different examples from my blog in the image below, just for comparison. Another option would be to follow the example of Ben Hennig and produce population-weighted cartograms.

When I produce these 3D maps (or visualisations) I'm not trying to create a geographically precise rendering of space but rather I'm attempting to draw attention to variations in a dataset in a way which 2D maps can only do to a limited degree. They are abstractions and simplifications but in terms of understanding the world I find it can be an improvement. There is a little bit about it in this Environment and Planning B paper I wrote but I plan to write more about this in the near future (the'near future' in geological terms of course). 

*Also known as 2.5D in the GIS world, but I'll put that to one side for now...

Population Density in New York City

In some recent posts I've been looking at the issue of population density. I did this for London in December and for the continental United States earlier this month. Given the extremely high population density in Manhattan, I thought it would make sense to take a closer look at New York City. So, I took some publicly available NYC GIS data, some 2010 US Census data and went to work. The result is the 3D map image below...

See this image full size

The mapping here is done at the Census Tract level, of which there are about 2,100. These areas have an average population of about 4,000 though there is some considerable variation between areas in that several tracts contain more than 10,000. The spatial patterns above are fairly obvious and, as expected, Manhattan dominates once again. However, the individual Census Tract with the highest population density is actually in Corona, Queens with a figure of 216,000 persons per square mile*.

There's a lot more information on New York City's 2010 Census results on these pages, from the New York City Department of City Planning...

* N.B. It's important to point out here that these areas are much less than a square mile, but I'm using square miles since it is a conventional measure of population density in urban areas).

United States Census 2010

The 2010 US Census was conducted in April 2010 and already the results are looking very interesting. By the end of 2010 there was a new total population figure for the US, indicating a growth of 9.7% between 2000 and 2010. The total population on the twenty third US Census day was 308,745,538. This is just over double the total population from 1950. For a more up to date population estimate, you can check the US population clock from the US Census Bureau. Because I'm interested in all this, I've produce a graphic which shows population density and some population data for the lower 48 states, at county level. A couple of nuggets here: Los Angeles county has nearly 10 million people and Loving County (Texas) has only 82 people. All other counties lie somewhere in between...

The IMD in 3D

Before I move on from experimenting with the deprivation data from the new English Indices of Deprivation I thought I'd do a 3D version, just for fun. The images below are the results of my experiments. The first image has labels for various places. The second image does not and is also at a higher resolution.



When you add a third dimension certain places stand out more (e.g. coastal areas) but there is a balance to be struck here, as always... That's enough IMD mapping for now!

Manchester Commuter Inflows

A short post today on visualizing commuting flows. I've used some 3D GIS techniques to create a commuting surface for the North West of England (based on wards). Areas with peaks represent high in-commuting. Manchester dominates the North West pattern, as seen below.

I added in flow lines to this map in order to see which 'peaks' were being by-passed on the way to Manchester, and from where. The lines on the map below represent flows of 25 or more from individual wards in the North West. Not a huge number per ward, but it all adds up. When you think of how many people are doing this and how far they come it is significant however...

This is only really a rough draft, but it does communicate quite a bit of information and it tells a familiar story of urban commuting. Click on the map to see it in full size.

The New vs. the Old - Flow Mapping

Back again to a familiar topic - flow mapping. In the past all we had was paper and two dimensions. Now we have e-everything and things can easily be displayed in three dimensions (or 2.5D as we say in the GIS world). The reason for this post is that I'm currently revising some maps for a journal and I have come to the conclusion that some things just can't be effectively displayed in a static, old fashioned manner - they must be made interactive to work properly.

The map below shows about as much as it is possible to show in a traditional geovisualisation of migration. Here I have shown all moves into Manchester (the local authority) between 2000 and 2001, with reciprical links (i.e. where people have moved both in and out along the flow line path) in red, with unique inflows in yellow. I'm busy with other things now, and am still working a lot on the e-learning and screencasting side of things, so time to go...