[EM] Hare (aka IRV) versus STAR

[Kristofer Munsterhjelm]

On 2024-04-11 11:57, Michael Garman wrote:
>  >> Greens & Bernie nearly always CW.  Republicans consistently, always, 
> finishing at the  very bottom. That would take a lot of sampling bias.
> 
> And I'm willing to bet there's a lot of sampling bias in who takes an 
> online STAR poll.

For what it's worth, this paper argues that the Condorcet loser in 2016 
was Trump and, while the error margins are greater, the winner might 
have been Gary Johnson.

https://journals.sagepub.com/eprint/G2XRIJW6VSQA9PHDGSRC/full

To quote:

> The Condorcet winner, if one existed, could have been either Johnson
> or Clinton—not Stein—with the evidence slightly favoring Johnson,
> but not with anything approaching statistical significance.

The same paper says that Johnson is to the right of Trump, so this seems 
a bit strange if politics are one-dimensional. But they also say, about 
trying to fit an 1D model to the data:

> Easily the highest percentages in the table are those for CSJT/TJSC,
> which are all above 70%. That spatial ordering has Clinton at one end,
> Trump at the other end, with Stein (next to Clinton) and Johnson (next
> to Trump) in the middle. (...)  Thus, based on this analysis, the
> centrists seem to be Johnson and Stein, not Clinton and Trump.

and more generally, about the suitability of a 1D perspective:

> Corresponding values for our PCA of the ANES policy questions show a
> much lower effect of the first dimension, and thus suggest greater
> dimensionality, or perhaps a fracturing of the shared understanding
> of the political space, for voters compared to legislators. The
> percentage of total variability accounted for by the first dimension is
> only 27.7,  and the ratio of the second eigenvalue to the first is 0.29.

Table 12 furthermore suggests that if Johnson *were* the CW, IRV would 
have denied him victory.

I know too little about third party candidates to comment further, but 
it's an interesting indication of things not being clear cut.

-km