How can I effectively display different types of worker coverage on a map?

[DaveC426913]

OK, I'm stuck on a problem implementing this map.
(It'll be built in HTML with JavaScript and CSS and is interactive but that's just context - this is really about map-colouring.)

The function of the map is to help users see at-a-glance what regions/counties of the province are serviced by three types of workers (PSWs, Doulas and RNs). (I will have rollovers and drill downs to provide another level of detail, but that's not exactly "at-a-glance").

The below map examples are at 100% scale. You can see by the size of the areas that I have very little room for details. That's why I'm preferring block colouring.

Here is a (bad) atttempt:

County Simcoe is serviced by a PSW, county Toronto is serviced by a Doula and county Halton is serviced by an RN.

Counties served by more than one type are a combination of the colours:
County Peel is serviced by a Doula and an RN (green, blue)
County Dufferin is serviced by a PSW and an RN (red, blue)
County Durham is serviced by a PSW and Doula (red, green)**
County York is serviced by all three (red, green, blue)**.
** note that these are actual mixes of the colours at 40% opacity, so R and G don't make yellow and RGB doesn't make white.

Anyway, you can see how this doesn't work - it's not at all apparent what, say, purple represents.

I tried this too, where the three colours are visible and don't overlap:

but it's pretty hideous ain't it?

I'd like to be able to incorporate density of coverage at the same time:

So, Burlington has multiple RNs - much more than Toronto.

Suggestions?

 

[DaveC426913]

Technically, this works without the legend getting out-of-hand...
I'm just not sure the colours are differentiable enough.

 

[pbuk]

I think even with perfect colour vision this is too great a cognitive load: the brain is just not used to separating 3 orthogonal axes from a colour map.

How about separate the colours into sectors so that in each area from 12 to 4 o'clock you display red at the appropriate opacity for the density of Doulas, from 4 to 8 o'clock green at the appropriate opacity for the density of RNs etc. You may need to experiment with an algorithm to calculate centres for the 'pies'. Also there are probably better colours than red, blue and green.

 

[pbuk]

Akternatively a single (grey?) shade for density covering the whole of each area with a radar chart or star chart for numbers or ratios of each type of carer.

 

[pbuk]

Also consider a non-geographic representation of the areas like this one for the US states: https://www.thedataschool.co.uk/chris-meardon/hex-map-how-to-why-to-blogs-about-one-dashboard

 

[DaveC426913]

How about separate the colours into sectors so that in each area from 12 to 4 o'clock you display red at the appropriate opacity for the density of Doulas, from 4 to 8 o'clock green at the appropriate opacity for the density of RNs etc. You may need to experiment with an algorithm to calculate centres for the 'pies'.

Yeah, I tried pie pieces. It was nightmarish-looking.

Also there are probably better colours than red, blue and green.

I played with CMY as well. Not as intuitive as RGB.

I am pretty confident that RGB - and the secondaries C, M and Y - are by far the easiest to identify. They are all colours we're intimately familiar with.

Any other three primaries will produce a rainbow with strongly overlapping segments, for example: green, yellow-green (is it green? is it yellow?) and yellow.

 

[DaveC426913]

I'm going to move forward with this until I come up with something better:

I've just added some functionality to make it much easier: the legend is comprised of hot spots to turn colours on and off:

Unfortunately, it doesn't facilitate a density parameter (yet).(I guess I may have to actually reinstall an editor after years. Notepad is a bit clunky for this.)

 

[pbuk]

Yeah, I tried pie pieces. It was nightmarish-looking.

Shame. How about 3 vertical or horizontal bars with relative quantity represented by relative area?

Any other three primaries will produce a rainbow with strongly overlapping segments, for example: green, yellow-green (is it green? is it yellow?) and yellow.

I was thinking of different colours for separate display to overcome the different perception of equal intensities of each of R, G and B. I agree mixing anything other than primaries is going to be even worse.

 

[DaveC426913]

Shame. How about 3 vertical or horizontal bars with relative quantity represented by relative area?

See post #1, pic #2. Hideous.

 

[pbuk]

I've just added some functionality to make it much easier: the legend is comprised of hot spots to turn colours on and off:

Well 2-way blends are easier to grok than 3-way I suppose, but it still doesn't work for me.

See post #1, pic #2.

Too many stripes - I was thinking of exactly 0, 1, 2 or 3 for each region.

 

[DaveC426913]

Too many stripes - I was thinking of exactly 0, 1, 2 or 3 for each region.

Surely you don't mean small counties have small bars, while huge counties have huge bars?

 

[DaveE]

See post #1, pic #2. Hideous.

I kind of liked that one. It doesn't look great, but it's easy to grok the meaning, easy to understand the legend. When I see purple, I just don't think of red + blue. You may have to choose between pretty and easy to understand.

How about different colored dots instead of stripes. This would be particularly good if you had more detailed geographic data than by county boundaries. You could separate the county boundaries from the service area boundaries. Kind of like a radar plot of rain superimposed on the map.

Anyway, you won't satisfy everyone, including yourself.

 

[pbuk]

Surely you don't mean small counties have small bars, while huge counties have huge bars?

Yes that's exactly what I mean: if coverage per unit area or head of population is represented by opacity/intensity what is the problem?

When I see purple, I just don't think of red + blue.

Me neither - and even worse for red + green = brown at low intensities, yellow at high intensities.

How about different colored dots instead of stripes.

Yes, (non-overlapping) dots could be good.

 

[DaveC426913]

. When I see purple, I just don't think of red + blue

THats OK the legend makes it clear

 

[Tom.G]

Here is a brain-shift:

Instead of colors use, use a single letter, P, D, R, for the service provider.

Make the letter, uppercase, block, with the size larger than the names.

If more than one provider, proportionally vary the letter size to reflect relative applicability. (probably 3 different sizes, possibly 4)

The letter color could be chosen to contrast with the region-color they are embedded in. (Although the region coloring is no longer needed.)

Where possible, readability could be slightly improved if the letters are consistently either above or below the region names.

Cheers,
Tom

 

[Office_Shredder]

You don't have to actually mix colors. You just need 7 distinct colors for the 7 different mixes of services, and you can pick those 7 colors to be as visually distinct as possible.

 

[DaveC426913]

You don't have to actually mix colors. You just need 7 distinct colors for the 7 different mixes of services, and you can pick those 7 colors to be as visually distinct as possible.

Can you imagine the legend though?

[ ] PSWs
[ ] Doulas
[ ] RNs
[ ] PSWs and Doulas
[ ] Doulas and RNs
[ ] RNs and PSWs
[ ] PSWs, Doulas and RNs

 

[Office_Shredder]

You could still draw the Venn diagram and just color it appropriately if that's what you're worried about

 

[Jarvis323]

I prefer the striped one. It's actually pretty easy to interpret at a glance.

The coloring blended version is absolutely terrible in terms of function.

I think you should make sure that the encodings of all of the colors in the map are clearly represented in the color map.

If you choose to mix colors, then rather than just blending one color over the other using transparency, it would probably be better to just take a linear combination of the colors (e.g. 0.5 * c1 + 0.5 * c2 ), and ideally in a more perceptually uniform color space than RGB. You might want to also consider colorblind friendliness.

 

[sysprog]

(I guess I may have to actually reinstall an editor after years. Notepad is a bit clunky for this.)

I suggest that you try NotePad++.

 

[jim mcnamara]

FWIW - your scheme is not generally applicable. Period. You may not see it as a problem. Your call.

Colour (color) blindness (colour vision deficiency, or CVD) affects approximately 1 in 12 men (8%) and 1 in 200 women in the world. In Britain this means that there are approximately 3 million colour blind people (about 4.5% of the entire population), most of whom are male.

https://www.colourblindawareness.org

British site, but it obtains worldwide.

 

[Ibix]

Suggestion: use the bounding box of the county name as a single bar stacked barchart to give the ratio of your three practitioner types (so each name has a background that looks like a standard issue three colour national flag). Use a greyscale (or other single colour) shading of the county polygon to denote the total number (or density or whatever) of practitioners.