DataGuy
Rookie
Posts: 217
|
|
« Reply #2 on: October 20, 2018, 10:58:08 AM » |
|
People are curious about how I'm doing my predictions. I'm still working on the technical details, but I'll give a brief overview. The first part, naturally, is the polls. I average the polls within a certain time frame, adjust them for bias (not house effect, like most use), and in a new twist I also consider trend lines. In other words, who has the momentum? Finally, the polls are converted to numbers that are realistic considering third-party candidates.
The fundamentals basically consider national shifts vs. local shifts and applies them to historical margins. I have chosen not to include fundraising, as I believe that money is much less relevant to the outcome than it used to be. Finally, the model blends in some "expert" ratings and converts them to numbers.
These different factors are weighed appropriately, with polls getting more weight as more are available (although they can never be weighed above 70%). The basic ingredients are similar to 538's "Deluxe" model, which they expect to be the most accurate. However, I apply the data somewhat differently. Indeed, everyone has the same data, but the success or failure of predictions depend on how the data is used.
As for the actual vote tallies, it's worth noting that turnout is difficult to predict, so that part of the model is still incomplete right now.
I intend to post predictions just like this for every Senate race just before Election Day.
|