My biggest concern about some previous books on infographics (and much of what is available online) relates to their uncritical promotion as brilliant ways of displaying factual information based on complex data. I was therefore pleased to see that in her introduction, Sandra Rendgen, stresses the problems with “Suspicious Data” stating that:

“…general popularisation (of infographics) brings with it a level of denigration, and content-related weaknesses are frequently found in graphic representations”.   

After reading this I was reassured that there was a strong editorial policy in terms of the graphics selected and their associated commentary. This shows and, despite the many pages to be filled, that vast majority of what is featured can be held up as best practice. The essays and associated images at the beginning offer good historical context (data visualisation is nothing new) and the location section alone would make for a good book for cartographers (sad not to see an Swiss mountain maps in there though). I also found the Category section (especially the work of Stefanie Posavec, below) particularly interesting by showcasing a range of examples of visualising more qualitative data.

Everyone who comes by my office has opened the book at least once and been tempted to buy it. For £29 on Amazon (RRP £44) I think you would be mad not to. For those producing data visualisations the book provides some great inspiration for future projects, whilst those who simply enjoy looking at them will not be disappointed.

The explicit message of this map is that surprisingly large distances are covered across the network and that the Central Line rolling stock travels furthest.  It is up to the reader to work out why this may be the case. Slow thinking maps require the viewer to think a little harder. These can range from more complex or unfamiliar ways of projecting the world as demonstrated by this gridded population cartogram of Africa from Benjamin Hennig’s PhD thesis

or the seemingly impenetrable (from a distance at least), but wonderfully intricate hand drawn work of Steven Walter (click image for interactive version).

I have seen bad examples of both slow thinking and fast thinking maps but there is undoubtedly more rubbish in the latter category. I blame the rise of infographics in addition to the increasing ease with which data can be mapped (I note, this latter point has also facilitated many great maps).  It’s not all bad though, much like tabloid newspaper headlines I think clever fast thinking visualisations have required a lot of slow thinking by their creators and are good for portraying simple but important messages. My concern, however, is that slow thinking data visualisations are on the decline, especially online, because they do not grab the attention of potential viewers quickly enough or  a similar impact (in terms of internet traffic) can be achieved with less data processing or, in the case of geography, cartographic flair.  In addition there is a (perhaps legitimate) fear that producing complicated visualisations will intimidate or confuse readers. This latter point is important in the context of the media and is a problem the New York Times Graphics Department (also at the AAG conference) seem to have grappled with. Their approach is to introduce the unfamiliar or complex alongside graphics people are used to. This was done to excellent effect with their 2008 election coverage- look out for the cartogram towards the bottom.

So do the renowned folks at the NY Times Graphics Dept. prefer fast or slow thinking visualisations? I asked them what they think makes a successful map. Archie Tse said what I hoped he would: the best maps readable, or interpretable, at a number of levels. They grab interest from across the room and offer the headlines before drawing the viewer ever closer to reveal intricate detail. I think of these as rare visualisations for fast and slow thinking. The impact of such excellent maps is manifest by the popularity of atlases and why they inspire so many to become cartographers and/or travel the world. The work of David Ismus offers a classic example.

From a distance (above) the road network reveals major population centres, the shading mountains and the colours the depths of water surrounding/ within the US. You want to know more and each time you look something new catches your eye.

In addition Ismus offers something else- he has painstakingly visited (cartographically rather than physically) every part of the map through manual labelling. Most of the maps we see are the product of automated cartography and can therefore make sense computationally but are less intuitive to use. This relates to the final commonality in the most interesting visualisation talks I went to at the AAG – all great maps, fast or slow, “feel right” to those who created them. There is a gut instinct at work that, perhaps, cannot be taught or acted on quickly. So next time you see a map, give it the time it deserves. Is it fast, slow, or the best of both?

The above map (and this one) was produced using R and ggplot2 and serve to demonstrate just how sophisticated R visualisations can be. We are used to seeing similar maps produced with conventional GIS platforms or software such as Processing but I hadn’t yet seen one from the R community (feel free to suggest some in the comments). The map contains three layers: buildings, water and the journey segments. The most challenging aspect was to change the standard line ends in geom_segment from “butt” to “round” in order that the lines appeared continuous and not with “cracks” in, see below.

I am grateful to Hadley and the rest of the ggplot2 Google Group for the solution. You can see it here. From this point I layered the plots using the geom_polygon() command for the buildings and water bodies and my new function geom_segment2() for the journey segments- these were simply the start and end latitudes and longitudes for each node in the road network and the number of times a cyclist passed between them. I have included the code below
 


#Code supplied by james cheshire Feb 2012
#load packages and enter development mode
library('devtools')
dev_mode()
library(ggplot2)
library(proto)

plon2


If I said a country was 1594719800 metres squared it would mean a lot less to you than if I said it was about the size of Greater London (so long as you know about how big Greater London is). For this reason the media tend to report the extent of a flood in relation to the size of the Isle of Wight or Icebergs in relation to the size of Wales (or Luxembourg) so that we can imagine the extent and scale of a disaster or news story. Despite plenty of comment on how ridiculous such comparisons are, and a great website that will convert standard measurements into the fractions or multiples of the size of Wales, I am yet to see a mapped representation of our increasingly standard units of area. The one I produced above is not meant to be definitive, just a starting point to what I hope will be a new system to replace the frankly inadequate* metric measures we are used to.

A much more effective alternative to simply stating an area in terms of its relative size to another area is of course to produce a map.  Geographically correct maps contain most of this information in the first place but they aren’t much good if you want to compare two things at either ends of the World or even the Solar System. With loads of mapping data online it is now easy to start shifting things around and laying them on top of each other in the same way the BBC’s How Big Really? website does.

This is fine if you want to compare a couple of things, but the map gets messy if you want to do more than that. For more complex comparisons you need to start with a fresh map (be careful of the projection) and shifitng everything around to fit on a single page. Doing this can have a big impact as Kai Krause did with his “True Size of Africa” map.

…sparsely populated areas (UK Cities on top of the Highland region of Scotland by Alasdair Rae)…

…and their transport systems such as subways (by Neil Freeman)

and cycle hire schemes (by Oliver O’Brien).

I think they offer a new perspective on the world and use maps as more abstract forms of information visualisation, so lets hope we see them more often to accompany the usual descriptive “relative to Wales” statements.

*I don’t seriously mean this.

I have been using R (a free statistics and graphics software package) now for the past four years or so and I have seen it become an increasingly powerful method of both analysing and visualising spatial data. Crucially, more and more people are writing accessible tutorials (see here) for beginners and intermediate users and the development of packages such as ggplot2 have made it simpler than ever to produce fantastic graphics. You don’t get the interactivity you would with conventional GIS software such as ArcGIS when you produce the visualisation but you are much more flexible in terms of the combinations of plot types and the ease with which they can be combined. It is, for example, time consuming to produce multivariate symbols (such as those varying in size and colour) in ArcGIS but with R it is as simple* as one line of code. I have, for example, been able to add subtle transitions in the lines of the migration map above.  Unless you have massive files, plotting happens quickly and can be easily saved to vector formats for tweaking in a graphics package.

R’s utilisation has been tempered by its relatively sparse documentation and challenging usability. The R community is increasingly aware of this with packages such as DeducerSpatial providing a graphical user interface to some of R’s spatial data functionality. More and more tutorials are appearing and people have been inspired by some high profile maps made with R (see here) so I am confident that it will be increasingly seen as the engine for slightly glossier analysis and visualisation packages.

R can’t do everything- I find handling map projections a bit tricky and its not possible to pan and zoom the maps as they are being created. In some circumstances I can’t do without these functions so I opt for a traditional GIS. Also, for the programmers out there used to the likes of Python and Java, R can have quite a few quirks in its syntax so be patient. Despite it’s flaws, if you have a large data processing and visualisation task R is a great option. It offers a high degree of flexibility in terms of input data formats and with packages such as twitteR, RCurl, and XML it is easier than ever to import online data sources from social media sites and data feeds.  Aside from traditional export formats for the visualisations it has become incredibly simple to export interactive and animated graphics using the googleVIS package or igraph for network visualisations. Such flexibility is invaluable if you are seeking to create a variety of different graphics from a single datasource without having to format it for multiple software packages. The great thing with R is the sense that it still has masses of unrealised potential for  future spatial data visualizations. If you know of any good visualisations or tutorials please leave a comment!

I should also say that if you would like to learn how to do these sorts of visualisations (and more!) come and do our masters course!

*simple might be a slightly optimistic way of thinking about it if you haven’t used R before, but with a bit of practice you will ge there! 

I have spent the last few years investigating the geography of family names (also called surnames). I work with the team who assembled the UCL Department of Geography Worldnames Database that contains the names and geographic locations of over 300 million people in nearly 30 countries (a few of these are yet to be added to the website). My research has focussed on the 152 million or so people we have data for in Europe and they all come from publicly available telephone directories or electoral rolls. I also had access to a historical dataset for Great Britain in the form of the 1881 census.  I have tried to answer two questions:

1. Is it possible to approximately establish the origin of a surname based on its modern day geographic distribution?

2. Are particular surnames more likely to be found together and if so do they form distinct geographic regions?

In the past surname research has involved  lot of manual work to create a detailed history of a particular name. With so many surnames in the database I had to think of some automated ways to do this computationally. The patterns I produce are much more generalised than the manual work- I find broad patterns rather than specific genealogical facts- but they provide useful context for population genetics, migration, historical geography and demography. If you want to find out more about this research here are titles for the papers I have had published in academic journals:

The Surname Regions of Great Britain.

Creating a Regional Geography of Great Britain Through the Spatial Analysis of Surnames.

Identifying Spatial Concentrations of Surnames.

People of the British Isles: A Preliminary Analysis of Genotypes and Surnames in a UK Control Population.

Delineating Europe’s Cultural Regions: Population Structure and Surname Clustering.

For a full list see my UCL academic profile. The left map at the top of the post is from the last paper I listed and shows how the surname regions vary across Europe. The map on the right shows how confident I am of the regions based on the number of times they emerge in the cluster analysis.

Last week I attended a “Beyond 2011” Census event organised by the Prof. Dave Martin and the Office for National Statistics (ONS). The attendees came from central and local government, private companies that utlise census data, and a few universities. The majority there (based on an approximate straw poll) believed that there would not be a census “in its current form” in 2021. This is because census data collection is a very costly exercise and its results are becoming more out of date more quickly as the pace of societal change (through migration, for example) increases.

So, we need to find an alternative but from where? Census data collection has been taking place for centuries and has, until relatively recently,  been our only source of “Big Data” concerning the population characteristics of the UK. Why then, when we have finally achieved the computational and analytical capabilities to efficiently analyse it are we prepared to throw it all away? The reason relates to the fact that we have hundreds of big data sources available now and, if used properly,  we can potentially generate the same levels of insight (and more!) as those provided by a traditional census form.

We live in an era of "Big Data" as Techcrunch (click image) explains

For example the 2011 Census only accurately recorded where people live, not where they work. What good is this for emergency planning in central London? Why not use Oyster Card trip data, or mobile phone usage instead to give daily updates on population movement? Another issue is the fact that we have a more mobile population and we miss a lot of change by only taking decennial snapshots. Other government datasets are updated more often, such as the NHS Patient Register or the Electoral Roll, and can provide an indication of where we are living this year, rather than where we lived 5-10 years ago (I moved house 4 times between censuses). We also, and perhaps controversially, hand over loads of personal data every-time we use a store-card or credit card, log into Facebook, use our mobile phone or surf the web. This has contributed, perhaps for the first time, to many private companies having a much better idea about aspects of the population of the UK than the government. The question is do we want companies to share it for the greater good (or evil depending on what you think of the “big brother” state), or should we let them keep the data and have the government spend more to source it itself? We also have to be sure to count those who don’t feature on private company databases (put crudely, often because they aren’t worth anything to that company) and it is these groups, often the most vulnerable in society, that we are most likely to miss with a non-census solution.

I think there is more than enough data to go around without having to fill in lengthy census forms, the issue is we haven’t worked out how to join it all together yet. Once we solve that problem we then need to work out who we have missed and that is much harder to do without a compulsory census!

If you want to contribute to this debate please fill in the consultation documents. 

When the color and size dimensions are correlated in some way with the tree structure, one can often easily see patterns that would be difficult to spot in other ways, for example, if a certain color is particularly relevant. A second advantage of treemaps is that, by construction, they make efficient use of space. As a result, they can legibly display thousands of items on the screen simultaneously.” (source: Wikipedia)

I think treemaps are becoming an increasingly popular method of visualising multidimensional datasets and there are many ways to create them. This tutorial is going to focus on the use of R and builds on the one published on flowingdata.

Open R and install the following packages

install.packages("portfolio")
install.packages("RColorBrewer")

now load them.

library(RColorBrewer)
library(portfolio)

The next step is to load in the data file we are using. This is an edited version of the London Borough Profiles csv taken from the London Datastore. There are five columns of data. The three we are interested in are “pop”, “earnings” and “party”.

input<-read.csv("http://spatialanalysis.co.uk/wp-content/uploads/2011/08/tree_eg_data.csv")
attach(input)

A treemap generally requires 4 pieces of information:
the item- in this case the London Borough’s or “id”- each will be assigned a rectangle,
a value to scale the size of the rectangle by- in this case the population or “pop”,
a value for assigning the colour- in this case the average earnings per person or “earnings”,
and a broader group to which the item belongs- in this case the ruling political party or “party”.

Armed with this we can simply used the map.market function from the portfolio package (installed earlier) to produce a treemap.

map.market(id, pop, party, earnings, lab = c(TRUE, TRUE), main="London Earnings, Population and Politics")

The output looks OK but I don’t really like the colours. I have therefore edited the code so that a selection of colours can be used using the ColorBrewer palettes. You can either download the code or load it straight away with
source("http://dl.dropbox.com/u/10640416/treemapbrewer.r")

you have now loaded in a new function called “treemap” that does a very similar job to the one above but has a few other options that you can see used below

treemap(id, pop, party, earnings, lab = c(TRUE, TRUE), main="London Earnings, Population and Politics", pal="Oranges", linecol= "dark gray", textcol="white")

The plot above used the “Oranges” palette but there are many more such as “Blues”, “BuPu” and “Reds”. Try for example
treemap(id, pop, party, earnings, lab = c(TRUE, TRUE), main="London Earnings, Population and Politics", pal="Blues", linecol= "white", textcol="black")

When you are happy with the results save the plot as a pdf
pdf("my_tree_map2.pdf")
treemap(id, pop, party, earnings, lab = c(TRUE, TRUE), main="London Earnings, Population and Politics", pal="Oranges", linecol= "dark gray")
dev.off()

or PNG
png("my_tree_map2.png")
treemap(id, pop, party, earnings, lab = c(TRUE, TRUE), main="London Earnings, Population and Politics", pal="Oranges", linecol= "dark gray")
dev.off()

and then you can edit it using image/ vector editing software such as GIMP or Inkscape to get the following result:

I hope that the options to change colours makes for a more interesting treemaps than the standard red/green ones we are used to seeing. If anyone knows how to alter the scale bar so that it does not show values beyond the range of the data it would be great to see how it is done!

The image below shows an assortment of map projections of the UK (and one of Great Britain). These have all been taken from Wikipedia so the level of detail along the coastline varies a little. They demonstrate nicely the effect that different map projections can have on the shape of a country.

As you can see, some of the projections have squashed the UK whilst others have stretched it or changed its orientation. The British National Grid is the best representation because it has been designed specifically for Britain. It is the projection you will see used on Ordnance Survey maps and therefore most printed maps of the UK (it is rarer to find it online). Whilst excellent for Britain, the National Grid projection does not work on a global scale because it would cause massive distortions to the other countries.  Instead, we should apply global projections, which need to be chosen carefully depending on their purpose and the scale of the map being produced. A poor choice of projection can have significant consequences because the relative size of a country on a map matters. Perception of country’s size is a delicate issue: in international politics, for example, countries which appear small on the map fear being overlooked. Indeed, this has long been the argument against the commonly-used Mercator projection, especially with reference to Africa, which appears relatively small on maps of this style. This effect is seen in Kai Krause’s “True Size of Africa”.

However, even this map has been criticised for using an inappropriate projection. The Economist, for example, responded by producing a map that maintains the correct relative areal proportions between each of the countries included (with Gall’s Stereographic Cylindrical Projection). Nonetheless, both maps illustrate the way in which our perceptions of this vast continent have been altered by “mainstream” mapping practices.

The Economist

The most widely used projection (in the Western World at least) is the Mercator Projection, but as the West Wing explains this has a number of flaws that are often overlooked…

If you are interested in where people live, then the above picture may also be misleading. It is possible to alter projections so that the size of the country on the map is influenced by its total population.

Mark Newman

It is clear that cartographers can produce different views of the world. We, as informed consumers of maps, need to be aware of this, to think twice about what we see and to consider how the information would look if projected differently. More importantly, by asking why the cartographer chose the projection they did, we may even be able to learn something beyond what we see on paper.

Following my previous post I have been digging around archive.org for interesting spatial/ geographical related resources. A search for “geographic” yielded a number of back issues of the National Geographic Magazine. They date back as far as 1888 and contain some great images and maps. There are some real gems to be had, such as the “Scenes from every land” series (link to one here) that contain century-old photographs from every continent. I was also amused to see an article titled “Notes about ants and their resemblance to man“. I have shared a few of my highlights here.

The top map is taken from the 1907 edition of “Scenes from Every Land”. It shows a map of the world with its key trade routes. I suspect today’s equivalent would be much more complex such is the nature of contemporary global transport.

My final two favourites are taken from the 1894 edition of the National Geographic Magazine and illustrate climatic variables for the USA. The top shows the mean temperature for the hottest 6 weeks of the year, and the lower map shows the sum of daily mean temperatures above 6 degrees. Despite their age I really like the clarity and detail in these maps.