Quantization Error

From a comment from by previous post on STV:

this is a feel-good proposal that doesn’t really solve any of the underlying problems.

>> Which are?

There is quantization error transforming the (input signal/citizens’ votes/polygon) to the (output signal/elected representation/pixels). The mathematical forumla used to process (samples/votes/fragments) is your filter kernel. Your (sample/vote/fragment) may contributes to zero or more (values/representatives/pixels).

Filtering signals smooths out your polygon, but it will look like shit if you undersample.

Increasing the size of your polygon does not improve fidelity.

The mathematical solution is to sample at twice the Nyquist Rate. Basically, figure out how often voters’ change their mind and have elections twice as often. Yeah, not likely.

However, increasing resolution (making ridings smaller and having more representatives) or increasing sample rate (having elections more often) guarantees improved fidelity.

If you didn’t follow that, maybe you stoners should have taken more math in university.


About this entry