10 replies to this topic

[Hans van der Maarel]

I'm busy producing a map for a client that shows quantative data, per postal code region. Apart from the obvious method of using proportional point symbols, I'm also going to try out this method

 screenshot_kleingeld.jpg   177.74K   214 downloads

So what's everybody's opinion on this?

Once I'm done with the test, I'll show him both and let him take his pick. Haven't started the proportional point symbols yet, but I'm expecting some problems there, given that the values in this test area range from 700 to 96000 and most of the symbols will be concentrated in relatively small areas.

-- edit --

This is just a screenshot from Illustrator, not production quality yet...

[loximuthal]

Do the different sized squares represent different orders of magnitude (1s, 10s, 100s, etc.)? I think my 10-year old son would have fun interpreting this map since they have recently worked with symbolizing numbers this way recently in school. Though not on maps; I don't think I've seen this kind of symbolization on a map before.

It could be a bit misleading, given that the size of the symbols increases more slowly than the data reprensented. Adjacent areas with values of 9 and 10 might give the initial impression that the area with 9 had more than the area with 10. I think I would have a hard time getting the overall picture of the data distribution. I'm also finding myself distracted by the irregular outlines of the blocks.

That's a tough problem, trying to display data over two orders of magnitude, and of course the large values always seem to cluster in the small areas

[Hans van der Maarel]

Do the different sized squares represent different orders of magnitude (1s, 10s, 100s, etc.)? 

<{POST_SNAPBACK}>

Yes they do. 100s, 1000s and 10000s. Forgot to mention that (obviously it will be in the legend).

I agree, it doesn't make it easy to get an easy overview of the data, but I think it makes it easier to read the actual values, as opposed to proportional point symbols.

The people who this map is for are not experienced map users, so that's another reason why I think this method might work just a bit better.

[Kartograph]

Dear Hans!

Finally back from the US. What a trip.
I see your problem. According to Bertin using different scales for symbols is not a good idea. Biggest problem is, that the map user will not realize the differences in scale. Even if he decodes it with the legend, he would be better off just having the plain numbers. The whole point with graphics is to show things instead of telling them, isn´t it?

But I know the problem with data like this, sometimes you just can´t help.
If you really want to use the Kleingeld-Methode, how about colour coding the different denominations? Would be easier to differentiate.
You could also re-arrange them to match our numbering system e.g. thousands-hundreds-tens-ones.

without wax,

Andreas

[Hans van der Maarel]

Andreas,

I'm not too keen on color coding, as there may have to be a second level of information added to the map.

This map indicates rental homes owned by organisations that are clients of clients of my client (yes, complicated). My client also would like to see a breakdown between homes that are equipped with a certain product and homes that aren't. I'm hoping that that could work out with the Kleingeld method (btw, is it called like that in English too?), but I think I'd have to go for proportional pie-charts.

Due to the data range (800 - 96000), I think I'll stick to my 100-1000-10000 system.

I'll see about adding a pie-chart sample soon.

[Hans van der Maarel]

Resurrecting this one...

The requirements have changed a bit. I need to map the total number of rental homes in an area (which, in the case of big cities, can easily approach 300,000+) and the number of homes in that area which use either one of the two products sold by my client (which is in some cases 0 or 1...)

After adding up some numbers, it looks as though the 300,000+ in an area is an exception, rather than the rule. All of the other test areas are a factor 10 smaller, so I'll focus on those numbers and come up with a special solution for the really big ones.

So... did some thinking and experimenting and came up with 3 possible ways to map this:

 experiment_symbolen.jpg   18.24K   144 downloads

Any thoughts? I like the first one, but it may be confusing.

Fortunately the areas are now larger, so this will cut down on visual clutter.

Edited by Hans van der Maarel, 05 January 2006 - 10:23 AM.

[Dennis McClendon]

At the risk of having my cartographic license revoked, I wonder if this should REALLY be a map. With such problems of data scale, I think I would do the data display in a chart, which is then linked somehow to a map. So I might have a little row of circles representing number of housing units coupled with squares representing product use. Nearby I'd have the map of districts. The two could be linked by callout lines, by color (if few enough) or by numbering. In all cases, I'd repeat the name of the district on both the map and the chart, because it has meaning to readers that does not require going to the map at all. (In other words, a Hollander seeing the chart for Haarlem wouldn't need to look at the map to see where Haarlem is.) With data districts this large, the map is not really useful for seeing geographic patterns; it's just a helpful way to organize the district data.

[Hans van der Maarel]

Dennis,

I was wondering about that too... I think this is borderline mappable

One reason to do this on a map rather than a map+chart is that the final map will actually be used to plan reseller activity per geographic area. This can still be done through a chart, but having it all in one map has its advantages.

I've told the client I need to experiment with this for a bit and that I'd come up with a sample next week. I like the map+chart idea and I'll definately give it a try.

[Martin Gamache]

Hans,

I think Dennis brings up a good question.

If using one map I think your preference for option 1 is mine also. Would the pie/circle size vary with the number of rental units?

The best map only solution would be small multiples, with three or four maps: rental units, product one usage, product 2 usage and no product usage. This would allow you to see intra-class spatial patterns more easily.

Doing some spatial analysis (autocorrelation) may reveal inter-class relationships better than any map/visualisation method.

mg

[Hans van der Maarel]

Martin,

Yes, the pie-charts are proportional.

This is supposed to be a series of 12 maps (one per province) so I'm not sure about the multi-map solution.

[Martin Gamache]

Hans,

Having made large series of statistical maps, I can say that the first one is the hardest to make, once the template is made it is just a matter of applying the data with minor tweaks. I would think that a template page with 4 maps would be quite easy to design...just some ideas. Probably a bit more work, but I think it would be worth it and not to forget that it may be more difficult to design and implement the 1 map design with all these variables, and 12 province is really not that much...imagine 50 states!