How Google Can Predict Elections | Hacker News Reader

How Google Can Predict Elections

pierrevalade.com

16 pointsway6612 comments

This sounds like one of those phenomenons that suffers when too many people find out about it.

The more it's written about, tested, and the more people in the disgruntled minority (or majority) with what appears to be the statistics on a given Monday --- the less accurate this'll likely be on Tuesday... rest of the time.

Reminds me of Asimov's "psychohistory". He conceived of a science capable predicting future events with probabilities but two of its foundational principles are that 1) the population sample is sufficiently large 2) the population is unaware that they are being observed or unaware of the variables/mechanisms used to predict their future.

Part of what made his writing interesting was that this science was very believable.

Those google numbers are 30-day averages, so it's possible to get plenty more than 100%. Looks like coincidence.

In the US now we have:

    michele bachmann 14
    herman cain      36
    mitt romney      11
    ron paul         34
    rick perry       33

All that said, I will be watching Insights and Trends now to see how it pans out. It might be coincidence, but it's potentially very interesting coincidence.

I tried it for the GOP presidential primary.

  mitt romney	   11
  rick perry	   33
  herman cain	   36
  michele bachmann 14
  ron paul	   34

So I guess it's Cain or Perry with Romney not standing a chance?

Also:

  perry	73
  obama	67

  romney  9
  obama	 72

In MA:

  scott brown        0
  elizabeth warren  31

Seems like selection bias to me. Voters who search for candidates are a subset of all voters. Plenty of voters (probably a shocking majority) watch the commercials and that's the extent of their research.

Ron Paul got better results than Perry, but he's not in your top two? If anything this demonstrates a major flaw in using this technique to predict elections

(as much as I wish Ron Paul got the nomination)

Google is getting thought samples from the brains of a billion people. Whether this particular method is valid or not, I expect they can make such predictions, and more. Flu trends, elections, real estate by region, box-office openings, popular conflict ("why are Tutsis so..."), traffic, stock prices in some cases ("xbox red light", "how to buy krispy kreme stock")...

Seems to be a coincidence for me.

If you do it only for Hollande and Aubry, now. The results are the same... Aubry 29 Hollande 39, difficult to prdict a second turn with more than 30% not turning up.

We'll see next sunday if it's a coincidence or no.

It's very much a coincidence; amongst the "top searches" for "hollande", "la hollande" comes third.

"la hollande" is the country Netherlands, and has absolutely nothing to do with the man of the same name.

The results are so close it is almost scary. It does make sense that the most popular candidate will get the most searches.

And pollsters just had their job made easier, until people figure out how to play it.

Yes, how?

Reply on news.ycombinator.com