Thursday, July 31, 2003

Maybe I'm beating this to death, but I want to clarify something I said last night about the ability of the Iowa Electronic Markets to see the future.

In today's New York Times business section, UCal Berkeley business professor Hal Varian specifically cites IEM's presidential vote-share betting pool and says that pools of this kind have

provided somewhat better forecasts than polls right before the election — and they provide much better (and less volatile) forecasts several months before the elections. Thus, markets do best exactly where the public opinion polls and expert opinion polls are weakest.

But this, I think, is based on a misreading of the data.

Go here (IEM's page for the 2000 White House vote-share pool), click "Price History," then check the numbers in August, September, and early October. Now compare polls for the same period at Polling Report. Months before the vote, IEM did have Bush closer to Gore than (most) polls did. But, by election eve, five of the seven polls listed by Polling Report had Bush up by a hair, or Gore up by a hair -- while IEM's betting pool had Bush up by 4.5%.

What that says to me is that IEM had a consistent, if slight, pro-Bush skew. At IEM Bush was even with Gore before he was even in most professional polls; then Bush was up by five at IEM when, in most polls (and, ultimately, the actual popular vote), he wasn't up at all.

I don't think this is right-wing bias. I think the consistently negative mainstream-media coverage of Gore skewed the results. The IEM betting pool, as I noted last night, utterly blew the 2002 congressional race, predicting a GOP House and a Dem/Independent Senate -- because, I think, the same mainstream press that loathed Gore in 2000 consistently told us in 2002 that that's how the congressional races would turn out.

So why do we believe IEM sees the future? And why should we believe that the Pentagon's death pool would have seen the future?

No comments: