Condorcet is just "if a candidate would beat all others head to head, then that candidate is the winner".

But the voting community insists on immediately jumping to all the crazy tie-breaking algorithms you need to implement if there is a "loop" (Smith Set).

My opinion has always been that those two concepts should be severed. All the "flaws" of Condorcet actually only apply to when tiebreakers are used. The flaws do not apply when there is a Condorcet Winner.

(The Smith Set isn't hard to understand either. It's just "these candidates beat all others head to head.")

And when there isn't a Condorcet Winner, that is actually valuable information. It means that the population was not ready to decide. It may warrant a later runoff, giving the population more opportunity to research the remaining Smith-Set candidates, and decide through additional debates or whatever else. At an alternative, people can always choose to apply any old method for the tiebreaker - even IRV! Condorcet-IRV (where IRV would only be applied to the Smith Set) strictly dominates IRV-only, in every way.

A cynical take would be that IRV has gotten traction in the US precisely because it does not seriously threaten the existing power structure. It might nibble away on a few edges, but if you look at all the countries and cities with IRV, they are not obviously better than those still stuck on FPTP.

You also cannot have meaningful estimates until all results from every district are in. How long it took in New York to get any results might leave people disillusioned and opposed to anything but FPTP.

Using Borda and Condorcet is also clearly a straw man, as AFAIK nobody is actually pushing for those in practice. The real runners seem to be IRV, Approval and STAR. Of those IRV is clearly the most compared to evaluate.

IMI they are all pretty simple and in an ideal world it wouldn't matter that much. However, we need to be highly aware that as soon as a major election gets won with anything but FPTP, we'll see the losing party "flood the zone with shit" to undermine the election results and the method. We've already seen the beginning of this after the election in Alaska. We need to keep it as simple as possible, otherwise the shit will drown everything else out.

This is always an interesting point to me. I often hear arguments that STAR and Approval are too difficult for the voters and far more complex (there's plenty of comments here suggesting that). While I disagree, that's a matter of opinion (I think both are trivial fwiw). But on the other hand, how votes are tallied is much harder to argue. Ordinal systems use a while loop where you pop an element each iteration and reallocate votes. Cardinal methods are sum then argmax. It is a joke to think that voters would understand Ordinal methods better than Cardinal methods with respect to how votes are tallied.

I'm puzzled by this. You can do it with pencil and paper in a few minutes? It's simple addition. You can easily build a website that visualizes each step if you think addition is difficult for people. It's trivial to compute in a spreadsheet.

> might leave people disillusioned

This is speculative; if it is true, you can find evidence by talking to, e.g. Alaskan voters. (IOW, Citation Needed.)

> clearly a straw man

I'm simply painting a picture by recounting my experience of what happened in the last thread on HN where voting systems came up, not making a strong claim about these specific systems being the runners up. My point still stands if you substitute any other systems, in this thread for example https://news.ycombinator.com/item?id=33111242. The point I was making had nothing to do about the merits of those specific systems; the point was that a big technical can of worms gets opened, and the average voter has no interest in digging in to compare N solutions on M technical axes, Arrow's theorem, etc. This is what people mean when they say "alternate voting systems are complicated".

Not quite. The tallying algorithm is `while (no candidate > 50% of votes): pop min(candidates), reallocate_votes(min(candidates))`. On the other hand cardinal methods are `argmax(sum(candidates))`. For comparison, FPTP is `argmax(sum(candidates))` too. The only difference is that FPTP is a one-hot vector. I'd argue that the average person intuitively understands cardinal methods without any training.

I do wonder if many people actually care to work through that step; in all of the public discourse around the Alaska vote, it's simply presented as a table with the reallocations already summed. For example, https://fivethirtyeight.com/features/what-democrats-win-in-a....

The general algorithm of "reallocate the votes from the candidate that is being eliminated" is conceptually simple, even if it's annoying to tabulate. I do stand by my claim that it would be very easy to build a website that can visualize this process without getting bogged down in the tabulation. See every site that's giving analysis on the Alaska results for examples.

I want to see actual UX studies on this; for example has anyone polled in Alaska to see whether people understand the result?

I agree that it isn't hard, but I'm glad we both agree it isn't trivial.

> I do wonder if many people actually care to work through that step

I don't think most people care to be honest. But I think what does matter is recount events. I often point to what happened in Arizona as a perfect example. It would be substantially easier to create a narrative that an election is rigged if the tallying is more complicated. Malicious people like to hide in complexity and will often oversimplify a complex topic. Because of this, I think one of the most important factors in creating a secure election is ensuring that the tallying system is simple. While malicious actors will always act maliciously, we can reduce their effectiveness through this simplicity.