Either the method embodies assumptions that hold only in particular contexts (low undecided shares, two-parties), or it is truly general. If it’s the former, the correct course is to specify the domain of applicability and the assumptions clearly, not to proclaim universal adequacy as they do.

My debates with analysts in the field take the form of

"But it works fine for Australia and Africa!"

"It's close enough for United States!"

The proper response is... So?

The point is that people who understand how these systems work would not claim that these projections are perfectly or (near-perfectly) representative of reality - but in political data, they do.

So I'm trying to fix it.

Different projection systems make different assumptions that make different sacrifices.

Again, whether Mercator vs Robinson or Spread vs Proportion, these simplifications have some worthwhile application. Maybe.

Depending on your projection method of choice (literally, Spread or Proportion) you get varying, conflicting outputs for what represents "accurate"

Simplification sometimes has merit - but the field's experts do not understand or appreciate the limitations of this approach.

This is almost perfectly analogous to what we're dealing with in poll data.

Two and three-dimensional poll data is simplified to this one dimensional line segment: Candidate A, Candidate B.

Undecideds? Project them.

Third party voters? Ignore them.

The "poll data" maps that look like this.

See also: the Senate

I would LOVE to talk to you about old polling data -

US only or other countries too?

Have they always been reported similarly, or have they changed in how undecideds are reported?

I worked/ am working on a few things that overlap - would love to look at reliable, older data