Hacker News Reader

Comment by MartinCron | Hacker News Reader

Bringing up the Coleman/Franken election reminds me of my particular pet peeve. When the margin of victory is less than the margin of error, it's impossible to know the will of the people. One hundredth of one percent is not within any reasonable margin of error.

View on HN

Sure, and in scenarios like that, perhaps the best thing to do is to flip a coin.

Instead of going to all the expense of flipping a coin, though, you could just take the person who seemed to get the most votes (after you've counted them all really hard to make sure you're within the margin of error). Just an arbitrary rule, no biggie.

perhaps the best thing to do is to flip a coin

Or have a runoff election, or have instant-runoff voting. Having elections determined by real or statistical coin-flips undermines the (important) story of self-rule.

You can't get rid of edge cases by moving the edge.

Yes you can. If there is a tie in the UK the result is chosen on a cut of the cards.

http://news.bbc.co.uk/1/hi/uk_politics/election_2010/england...

The candidates agree, and the winner and loser both accept the outcome, works for me.

I'm confused by your reply. The grandparent comment was saying "You should never flip a coin", the person you replied to said (roughly) "It's impossible to avoid all situations where you can't measure the winner, you have to either flip a coin or just take the person who happened to come out ahead in the vote count".

Your link to a place where they used a coin flip isn't disagreeing with him (nor really the grandparent, who wasn't claiming that you can't use a coinflip, but rather that you shouldn't use a coinflip).

Maybe not, but you can lessen the real-world impact of edge cases by moving the edge.

And if that is too close as well?

In the end, if one candidate got 100,000 votes and the other got 100,001 then to the extent that we randomly select between the two I'm not sure that it really makes a difference which one wins. In this case the will of the people is that the two candidates are equally good.

"The greatest value of free elections is in all of the out-of-equilibrium outcomes that, because of the regularity of free elections, never come close to happening."

In this case the will of the people is that the two candidates are equally good

That's a very positive way to look at it. The problem of reduced legitimacy persists, though. The disgruntled Coleman supporters will probably always have a sneaking suspicion that the election was stolen from them, which tends to poison actual discourse.

Sneaking suspicion? After the 2000 election you would be looked down on for calling Bush president in many demokratic circles.

Yeah, we aren't willing to admit it, but every voting system has a margin of error. Even 1 in a thousand voters will have problems with completely unambiguous questions.

I applaud your statistical sensibility, but I have a question – you're specifically looking for the margin of victory to be bellow the margin of error – but the margin of error of what, exactly?

Are you expecting the vote to be a proxy for the entire population of the country/state/whatever? Perhaps the entire population of eligible voters? Or, do you view it as simply the preference of those who took the time to vote?

Personally, I see little harm in disregarding the intentions of those who could vote but choose to not vote.

Issues of systemic inequalities in access to voting aside, I have little problem with viewing an election the task of accurately counting the votes of those who actually made it to the poling place. If you accept that concept, then the margin of error is extremely low – especially with electronic voting systems.

You're right. But for 3,000,000 votes cast, the margin of error is about 6 hundredths of one percent. At the 99% confidence level, it's .074% variance.

What are you defining as "the" margin of error?

The margin of error of a poll is the margin of error at the 95% confidence level (two sigma). The standard "margin of error" when you are talking about polls.

http://en.wikipedia.org/wiki/Margin_of_error#Different_confi...

In reality, you should apply a finite population correction, which would make the "margin of error" a bit smaller if you get a substantial percentage of the population voting:

http://en.wikipedia.org/wiki/Finite_population_correction#Co...

Reply on news.ycombinator.com