Kenneth Field, Senior Cartographic Product Engineer with Esri and controversial blogger (to some, at least), felt compelled to write a critique of Simon Rogers‘ Beyoncé twitter map (on occasion of an album release) that was hyped by Time Magazine (in a punny way) as being “flawless” (to be very clear: the “flawless” attribute does not originate from the map author and I don’t get hung up on it, since it’s a pun for the pun’s sake).
I share many of Kenneth’s points in this case though (especially also regarding data quality). I’d thus like to chime in on some as well as add a few of my own:
1. I agree that we see quite a number of not very substantial, link-bait / “map-listicles” type of maps online (“These 20 maps will …” – knock your socks off, or similar). While I definitely can be a sucker for nice visualizations (i.e. I can perfectly enjoy them as entertainment), it sometimes saddens me a bit that not more consideration is given to properly crafting and publishing maps (e.g. basics like specifying data sources, limitations of the data, authors, choosing an appropriate projection, etc.). In the case of this map, for example, the slider shows a certain time and the sub-title states that tweets of December 12–13 have been analysed, but we don’t even know what timezone the time refers to. I tried, but found it quite hard to guess the local times in the map as it’s being animated.
2. The mix of dark basemap and bright purple in the tweet map works well to stimulate the (my) eye and I really like the animated nature of this map, since the temporal dimension adds a lot to the way the spatial patterns unfold. That said, to me the map looks way too nervous when animated. I would have wished for better temporal smoothing, potentially also more spatial smoothing or even aggregation.
3. The parameters for the “density” raster or its mapping into colour space are not well chosen, in my opinion. E.g., looking at the image above eastern US reminds me of what photographers call blown or burnt out highlights.
4. The reason I put “density” in quotes is that to the untrained eye the visualization looks like what we geographers call a typical kernel density estimation, but it isn’t (at least not a good one). I’m not sure if there is a term for this method that paints a semi-opaque disk for every occurrence of a phenomenon and then alpha-blends overlapping disks – faux density surface? (I know, you can classify it as a KDE with uniform kernel, but overwhelmingly we apply density estimation (hint!) to account for uncertainty and thus use a smoothing kernel. Also, that way it is visually way more appealing, in my opinion. I can remember using the faux density surface technique myself, when I tinkered with Processing since it is cheap to implement. It would be great, if the CartoDB folks added the capability for smooth KDEs though (I haven’t checked if it doesn’t already).)
5. Since the map is Web mercator and supposedly the uncertainty attached to the location of tweets is spatially not very variable (at least not systematically), we should technically see the tweet footprints having substantially varying sizes (and shapes) depending on their latitude.
6. Also, because of this it would be worth a thought to have either a reference or some normalisation of some kind.
I’m fully aware that as a professional geographer I may well be strongly over-thinking some of these issues. However, I’m still optimistic that maybe with the occasional map critique we might be able to improve our discipline (and its understanding by its new adopters as well as the appreciation of the wider audience) a bit here and there.
(edit: fixed typos)
3 thoughts on “On the Beyoncé tweet map”
Hey! Great post. Full of interesting things!
It is true that the map could include some very relevant metadata like timezones, totally agree. I guess lot of this is examplained on the text around the maps on the original post, but in these days of easy “sharing without context” is easy to get lost of some vital information like you are describing. I believe better tools to provide contextual and metadata information are key, and in CartoDB we are working towards that.
In terms of “density” you are right, there is likely better ways to obtain a kernel density estimation. In reality those circles you see actually represent a fade-out of the point over time. I am not sure if Simon Rogers actually made use of another variable like number of Retweets, but the size of the circle could represent the impact of that tweet. In general I believe much better results can be obtained by having different styles at different “zoom levels”.
Torque visualizations are done using CartoCSS and extending it with filters to be able to represent multiple frames at the same time with different styles. So like if a tweet happened during the first bucket of time, you can also represent it on 3th, 4th etc with a different style. Thats the bubbles around the dots… is a way to leave traces of the points as time passes, or if not it will be hard to see the “density” over an area and not just dots appearing and disappearing.
On projections… well yes, the never ending discussion… hopefully soon there will be an easy way to provide as performance maps as it is right now with webmercator tiling system. I know other projections could be done tiling too, but well, I guess projections are not currently the biggest priority on web mapping products. I believe that browser vector rendering will make things better and hopefully we will have a much better variety of projections out there :)
Definitely should have wrote a blog post, hehe.
Final comment… there is not much written about animated maps as far as I know. Specially those that you can actually interact with, zoom in, get details, etc… if you guys know of articles or resources exploring time-lapsed or animated maps, I would love to hear about them!
Thanks, Javier, for commenting.
You raise an interesting point: lineage/metadata and ensuring its sharing along with sharing of the actual content. Future tools should properly provide means so that “sharing without context” actually gets hard. Until we are there, we’ll have to rely on authors of maps and visualizations making as much of an effort as possible in providing context.
Thanks for the background regarding Torque. Highlighting the occurrence of tweets (or any other phenomenon) by bright and fading-over-time circles makes sense, in my opinion. I’d just love to see the addition of a more intelligent kernel density estimator in the mix.
Projections _are_ topic of discussion, indeed. My hope, too, is that the advent of things like Proj4js and the strong projection capabilities built into D3.js also helps pave the road for off-the-shelve webmapping projects to provide widespread availability and ease-of-use for custom projections. Personally, I find e.g. kartograph.js (http://kartograph.org) has already some interesting projection options.
Regarding research: I am aware of some research strands around dynamic maps and usability/human-computer-interaction at GIVA labs at University of Zurich (http://www.geo.uzh.ch/en/units/giscience-giva/research), e.g. the Animeye project (http://www.research-projects.uzh.ch/p6389.htm), you might want to look into, or rather: start your exploration of research on these topics from.