I have two major pieces of my application to finish.
1. In the list of plans to choose – show a glyph that indicates the relative merit of that plan in each of the three dimensions: population equivalence, moment of inertia and city-splitting. Sort the list by any of those dimensions.
2. When showing the demographic measurements of any region – like district, city or tract, encode the population density as well as the metric (like percent Caucasian).
There are some other details that would be nice – but these are the major ones, and I want to complete them so I can get on with the writing. However, they have some twists and turns. To compare plans – I need more plans. I have imported Brian Olson’s plans and resolved some issues along the way. They are only at the “tract” level, but you can actually see them on Google Maps. That way, you can take advantage of the rich trove of information that those maps offer. CA and TX are too large, but for the others, paste any of these URLs in the search box of Google Maps:
http://www.nautilusone.biz/redistricting/olsonMaps/olsonFL.kml
http://www.nautilusone.biz/redistricting/olsonMaps/olsonMA.kml
http://www.nautilusone.biz/redistricting/olsonMaps/olsonNY.kml
In each case, the districts will show in solid colors. If you “uncheck” a district in the control panel on the left, the underlying geographical details – like roads, city names and aerial photographs will be available. Later I may put in a transparency slider, like Google Earth has. Another way of exploring the district characteristics.
I have tried splitting CA and TX into two pieces, but they still seem to be too big. I may try some other approaches later.
Another potential source for plans is the approach outlined by Span, Gulotta and Kane, in their paper about their Moment of Inertia method. I was able to meet with Dan Kane, and I understand their work and it may be worth a day or two to try implementing their algorithm.
I have the current plans for the 111th COngress, the results of my Cricnione work, recursive splitting on cities and tracts and Brian Olson’s work. I am till hoping to get Georg Clark’s plan for CA. That should be enough for a reasonable range of comparisons.
The other piece to include density when I display the demographic characteristics of a region. Otherwise, given two regions of equal population, a small region will lose significance compared to a large region – even though they are equivalent. I currently use a ColorBrewer scale for the low to high range of any characteristic – with no indication of population density. Below the state average is green, above is purple.
It is defined with RGB values. I have tried translating it to the HSL color-space, and varying the lightness to encode the relative density. I could not get enough contrast between different values. I have been exploring the Lab color space, and that seems more promising. The image below shows a circle of color – with an *a value indicating the relative blue-green/magenta component, and *b indicating the relative blue/yellow component. Both of these axes come close to matching the eye’s channels for visual perception. Then the vertical dimension is lightness.
I found this applet tool for generating related values of Lab colors, and also translating them to the RGB space – Graphic CIE L*a*b* Calculator.
You can choose a and b values with the bottom and right sliders, and then vary the lightness with the left slider. After some experimenting, I decided to keep b fixed at 20 and vary a from -80 to +79. I would use that to represent how much a measure differed from the state average. Then I varied the lightness from 25% to 75%. Tha would encode the relative density of that region, relative to the state average. Results below seem OK for now. I may refine them later. Initially, I varied the colors manually, but found some good formulas here – first translate L*ab vlaues to the XYZ space, and then to the RGB space.
Here is my current color scheme.
Other possibilities include different starting ending colors – using a scheme based on varying the b factor – or one moving diagonally across the color circle, simultaneously varying both a and b. I need to improve my color generating function and then try it on my real range of demographic values.
