Thursday, 23 July 2009

SuperFreakonomics update

When Stephen Dubner announced a contest to guess the number of Google hits there would be for "SuperFreakonomics" on November 3, my thought process went something like this:
  1. There are about 11,000 results now.
  2. There are 1.3 million for "Freakonomics".
  3. As the SuperFreakonomics launch comes closer, it will be discussed more and more.
  4. But it is unlikely to get close to the total for Freakonomics - partly because many SuperFreakonomics pages will also have the word Freakonomics in them; partly because it's a sequel; partly because Freakonomics is a cultural phenomenon, surely beyond the original expectations, and the chances of Levitt and Dubner hitting two home runs in a row are low (nothing personal, guys, just stats).
  5. So somewhere in the 50,000-300,000 range is probably about right.
  6. I have no accurate way to predict where it will fall within that range
  7. But - just like The Price is Right - success in this contest does not correlate precisely with accuracy; instead, it depends on being closer than any other entrant. There is a subtle difference. Winning an English auction does not arise from getting closest to the real value of the item; it comes from bidding $1 higher than the second highest bidder.
  8. So I analysed the figures submitted by all other contestants - fortunately this was not a sealed-bid contest - and picked the number - 88,782 - that gave me the highest probability of winning. According to my own little predictive model, of course.
However, at the time I did have a fear about the consequences of running this contest. My concern was that it is potentially quite easy to game.

Let's say that you bid 10 million. It's 15th October and the deadline is approaching - and Google still only shows 75,000 results. But you're desperate to win that signed copy of the new book.

So you start posting about SuperFreakonomics on your blog. And other people's blogs. And perhaps you hire a linkfarm company to spread the word around around their automated content generation systems.

Pretty soon you could probably generate another 10,000 hits or so.

And if there are a hundred other contestants who also bid high, then soon enough there are a million new results.

Your chances to win are thus substantially improved - until the guy who guessed 500 million realises and starts doing the same thing.

This gaming is all one-way - there's no way to reduce the number of pages on the Web so only those with high guesses are incentivised to participate.

Still, I relied on the likelihood that $50 worth of schwag would not be worth the effort for most people, and that gaming would be minimal. Thus I stuck with my guess, in the expectation that the increase would be slow - perhaps with an acceleration towards the launch date, but not an overwhelming one.

The last thing I expected was that the number of hits would reduce.

So imagine my astonishmentTM when I searched a few days ago to find only 9,600 results. A glitch in the Google data refresh? Nope: this afternoon there are only 6,360. It is shrinking by the hour, as a slightly earlier search returned 6,400.

Is SuperFreakonomics dying out? Will there be only ten hits left by November? Or even - using the Buiter/Mankiw/Sumner Taylor rule extrapolation - minus two thousand?

I can only assume Dubner and Levitt's promotional scheme is backfiring and that the gods - or Google, their representative on Earth - somehow resist the attempts of mere mortals to generate searchable publicity through clever non-monetary incentives. Does this mean social media is dead? Is this the final revenge of neoclassical rational economics? Or are all those predictions of deflation just belatedly coming true?

Update 30th July: Down to 5,180 now.

1 comment:

Anonymous said...

The number of results appear case-sensitive, which is surprising.

I'm betting that the Google algorithm adapts for attempts to manipulate results nowadays, though, simply by refusing to list results which look like spam. It only needs to explicitly list 99 pages of 10 results each, after all; accuracy in the number of claimed results is not a priority.