Registered: 1182202224 Posts: 100
Reply with quote #1
What are your thoughts about bricks? We invite you to post your comments here.
Registered: 1136226912 Posts: 247
Reply with quote #2
Displaying data on maps is challenging (with issues such as "area size bias" and such). I usually find it useful to display the data in several different ways to get a more complete picture.
The bricks idea is interesting - I think it is easier to count & compare the number of bricks, than comparing the size of bubbles (and of course with bubbles, some people tend to compare the height/diameter/radius of the bubbles, rather than the areas). But perhaps they'd be even easier to count if the bricks are all stacked into a single 'bar'(?) ... and if you want to compare bars, it's best to plot them all against the same axis in a bar chart rather than a map. And now we've come full circle! :) Also, if some of the areas are very small (such as the NorthEast US), seems like the groups of bricks are likely to bump into each other - and when one group of bricks touches another group, it might give the impression they are 1 group (rather than 2). Whereas pie charts don't generally have this problem. Also, the bricks seem dangerously close to unit charts ... which 'are for kids', right? ;) http://www.perceptualedge.com/articles/visual_business_intelligence/unit_charts_are_for_kids.pdf Interesting idea though - I think it's got some potential!
Registered: 1135986598 Posts: 838
Reply with quote #3
Bricks function differently than unit charts in that bricks can be subitized (preattentively counted) but unit charts must be consciously counted, which involves a slower, attentive process. Stacking the bricks in one dimension to make them look like bars would eliminate our ability to decode and compare them preattentively via subitization and shape perception. Also, stacking them to form bars would make it more difficult to fit them into small spaces, which is a concern, as you point out. Fitting bricks into a small space is not significantly more challenging, however, than fitting bubbles. __________________ Stephen Few
Registered: 1357735344 Posts: 2
Reply with quote #4
Did you consider to use triangles (instead of bricks) to come to a decimal system (eg 10 bowling pins stacked to a pyramid) or binary approach (16 zones within such pyramid) ?
V V V V V V V V V V __________________ Thomas
Registered: 1366981110 Posts: 18
Reply with quote #5
What do I think of bricks? Well, I implemented some bricks in Tableau.
To do this, I created some custom shapes. In Tableau it would be very difficult to implement shapes in the complete way suggested by Stephen – it would be very hard to deal with outliers in the way he suggests. However, this is a starting point and
. you can download my workbook here What do I think of bricks? Appropriate only for “known” geospatial dimensions They might only be appropriate where the geospatial dimension is already familiar. Using bricks on US States is effective because we already know which state is where. However, using bricks when the location is less familiar will not be as effective. When we use circles, we instinctively understand that the centre of the circle represents the exact location of the mark. But where’s the centre of the brick? Is it the centre of the 9-mark grid, or the brick itself? Occlusion is a big problem Look at Florida above. In the circles chart. there are lots of overlapping circles but I can still get a very good idea of the size of each city. With the bricks, however, the picture is extremely confused. Is the problem significant enough that it warrants a new type of display? I’ve said before that every visualisation is a compromise: the purpose in visualising data is to reveal patterns. Details should be available on demand. Also, the circle is a comfortable metaphor for the majority of people. Bricks will require training, or clear instructions, to understand them. This is fine if you are in, say, a closed corporate environment, where you can reach everyone, but not good if you are publishing on the web to a wide audience. This is related to what psychologists call schema and is why it is sometimes preferable to use pie charts rather than, say, stacked bars: familiarity means people will engage more quickly.
Registered: 1366983092 Posts: 1
Reply with quote #6
@sfew, some interesting ideas. I think the overlap is a pretty serious issue, much more so than with bubbles -- (see Andy's post).
I'm also not buying the distinction between bricks and unit charts. Moreover, in your examples you've got the bricks corresponding to ranges of values -- so I have to count the bricks and *then* convert that to a range. Seems like a lot of room for cognitive error there? Most people are going to assume 6 bricks = 6 units of X. I do agree that bubbles are less than optimal for mapping things.
Registered: 1366984190 Posts: 1
Reply with quote #7
<grr, post lost while typing, short version instead: >
In some cases, it is useful to encode a true zero differently rather than "lumping it in" with the lowest band. I think a good option here would be to use an empty brick, ie one with a border and no fill. A state with no sales at all may require a very different response than one which just has low sales (that response may be to ignore it if we expected the zero, but not if we usually have sales there).
Registered: 1135986598 Posts: 838
Reply with quote #8
Thomas -- I didn't think of using triangles to build larger triangles. I like the fact that they could be used to build a neat combination of 10. Offhand, I'm not sure that the shapes formed with lesser sets of values than 10 would work as well as bricks, however, but I'll play with triangles a bit to see.
acotgreave (Andy) -- I definitely agree that bricks don't work then they overlap, which I mentioned as a downside in my article. The fact that circle intuitively mark a location based on their center is a slight advantage of bubbles over bricks, but only slight, because a set of bricks has a center also, just not one that is as precisely located. Regarding the unfamiliarity of bricks compared to circles, every new form of data visualization is unfamiliar in the beginning. This was true of bars, lines, and even bubbles. When an existing form of display suffers from significant problems and a potentially better method that can be easily learned, the question we should ask is "Will a minute or two of training be offset by improved effectiveness?" When I introduced the bullet graph in 2005, my answer to this question was "Yes." Perhaps "Yes" is the correct answer regarding bricks as well. We'll see. cingraham -- The distinction between bricks and unit charts is based on a distinction between subitization and counting. To confirm that differences between bricks of various values can be perceived through the perceptual process of subitization, experimental research is needed. The hope that they work in this manner is based on perceptual theory derived from previous research. To decode brick, you don't need to count the bricks and then convert to a range, you merely match the brick set based on subitized count and shape to the corresponding brick set in the key. To decode the value of a bubble, you must match the bubbles size to the corresponding size in the key, which is perceptually more difficult. AdamV -- Distinguishing a value of zero from the absence of a value is not a problem unique to bricks, of course. There is no standard way to do this with bubbles either. It would certainly be possible in either case to create a special mark, such as an empty brick, as you proposed, to handle this. Good suggestion. __________________ Stephen Few
Registered: 1280427427 Posts: 21
Reply with quote #9
I agree that the biggest difficulty to displaying information using bricks (as mentioned) is overlapping.
This technique to display information spatially has potential but warrant some improvements. I don't know how so I am pointing out an example that's currently being used to display meteorological data (i.e. weather maps). To learn more on hour meteorologists display weather data, you can read more at http://en.wikipedia.org/wiki/Station_model For this discussion, I pick "wind barb" since it's most relevant. And here's the convention to represent wind speed. - Each half barb (or flag or barb) represents 5 - Each full barb (or flag or barb) represents 10 - Each pennants (or filled triangle) represents 50 Because wind is a vector, the symbol above is also combined with a long shaft which points to the direction FROM which the wind blows. (for the purpose of this discussion, you can ignore the direction). Using the wind speed convention above, one can represent lots of information in a compact manner. I'd compare it to Roman numerals but that system uses 7 and this one only require 3. In the example of bricks, there is only 1 symbol being used to display. In term of training to understand each system, I believe bricks require no training. In the case of wind bards, some training is required and likewise, Roman numerals would require the most training. In several examples provided by Steve and others, I think the bricks method of encoding information could work. But in practice, when plotting spatial information, we would face the overlapping problem (probably more often than we want to admit). In comparison, the wind bard display MINIMIZES the overlapping issue quite well (although it doesn't eliminate it especially when plotting too much stations in a small area). See an example from NCAR attached below ( http://weather.rap.ucar.edu/upper/upaRAOB_500.gif) When looking at the map, ask yourself whether you can see the pattern (areas of lower wind speed and higher winds speed). It would be interesting to hear thoughts. So my question is can we use a combination of bricks and possibly something else to encode information? Attached Images
Registered: 1366981110 Posts: 18
Reply with quote #10
Stephen - your point about the bullet chart is totally right. I find that people are now very comfortable reading bullets; they certainly weren't in 2005. I am also aware that an expected first reaction to anything new is "Ugh - you want me to *change*. No way!" I tried to assess the bricks in a way that is open to change. Over time, we could well see these being normalised.
Occlusion remains my biggest concern of the ones I outlined. Andy
Registered: 1366989712 Posts: 1
Reply with quote #11
Another issue I see that hasn't been raised yet is that my brain keeps trying to turn the brick shapes into buildings -- factories, office towers, mosques. This attempt at pattern recognition interferes with my ability to interpret the bricks as data. Because circles are uniform and atomic, they don't create that extra cognitive load.
Registered: 1135986598 Posts: 838
Reply with quote #12
Actually, what you're experiencing is a benefit of bricks. The shapes that are formed by bricks can be associated with patterns (meanings) that make them more recognizable and memorable. __________________ Stephen Few
Registered: 1330960889 Posts: 57
Reply with quote #13
Bricks function differently than unit charts in that bricks can be subitized (preattentively counted) but unit charts must be consciously counted, which involves a slower, attentive process. Stacking the bricks in one dimension to make them look like bars would eliminate our ability to decode and compare them preattentively via subitization and shape perception." I'm not convinced about this last sentence, at least not for a 4 brick scale. I don't think I'd actually need to count "one, two, three, four" every time there was four bricks. Much beyond that I might need to do so. How about having four bricks stacked for every case and then only adding fill colours where necessary? So instead of having 3 bricks for 50-75% you'd have four with three filled and one unfilled. I'm fairly sure I wouldn't need to count anything then.
Registered: 1135986598 Posts: 838
Reply with quote #14
In theory, arrangements of up to nine bricks as a 3x3 matrix can be perceived without counting. We know that quantities of up to three or four can be subitized (recognized without the conscious act of counting). We also know that rows of objects properly delineated are preattentively seen as groups. By combining groups of quantities up to three as I've done with bricks, in theory quantities up to nine can be preattentively discerned. Ease of distinguishing the quantities is further assisted by the fact that we can preattentively discern simple shape differences, which I believe includes the shapes of bricks as I've distinctly arranged them to encode each quantity from one to nine. My hope that bricks can be decoded and compared much more rapidly than bubbles is based on a good theoretical foundation, but we can't conclude that it works without subjecting bricks to experimental research. __________________ Stephen Few
Registered: 1240321507 Posts: 46
Reply with quote #15
As much as I'm intrigued by Stephen's "bricks" to better encode/display geospatial data, I don't quite get the value of these kinds of visualizations, specifically if the data visualized turns out to be a single shape (of whatever dimension) that reflects a single measurement for the State of xxxx. (This type of visualization seems to be a step up from the over-generalized US presidential "Red State" vs. "Blue State" maps.) Wouldn't a table with horizontal bar graphs, sorted descending or ascending, offer a more useful view? (Perhaps the bars could be colored to reflect various regions of the country.)
Truth be told, I don't have experience developing geospatial visual solutions, so I'm speaking truly as a novice in this regard. I find myself drawn to maps that plot data at the most granular level, and then allow for the user to select a zoom level (i.e., Here's what the data looks like across the country vs. here's what the data shows at the city level.) Simon Rogers of The Guardian published an article back in November 2011 that mapped every road fatality in the USA between 2001 and 2009. Different types of casualties are encoded (car, pedestrian, bicycle...) From a view of the country, the visualization is a mashup of the data, but drilldown to the state, and even further to the city/street level and the data becomes quite specific. Granted, Roger's visualization doesn't offer the reader any mechanism to size up the data visually the way Stephen's bricks can offer. Stephen's bricks remind me of dominoes: a quick glance at the size and shape tells me the magnitude of a value quite quickly. But, that's the point, "magnitude". Are geospatial visualizations meant to reflect data that are specific or to reflect comparable magnitudes? (e.g., Doesn't a radar weather map easily convey understandable information about the severity of a storm across a region?) In the end, I'm not holding a firm position on the subject -- just seeking better understanding. --Pete Attached Images