3 ms·
The likelihood numbers cannot be tested empirically. The experiment only gets run once. In other words, there is no error term for the likelihood numbers. I'll
by robg 18y ago
The likelihood numbers cannot be tested empirically. The experiment only gets run once. In other words, there is no error term for the likelihood numbers.
I'll be looking at Electoral Votes and the popular margin.
There the difference between pollster and 538 is the most interesting to me. The former is a simple average of all polls while the latter uses weighted averages. Individual pollsters might get closer but within their margins of error.
- jedc 18y agoBut the markets (and fivethirtyeight projections) exist for every state. Together they're a decent dataset, though still a bit light. I, too, am very interested in comparing betting (prediction) markets versus poll-based probabilistic forecasts. I wrote more about this here: http://blog.mercury-rac.com/2008/11/03/election-tuesday-what-to-expect-from-the-prediction-markets/ http://blog.mercury-rac.com/2008/11/03/election-tuesday-what...