I mentioned in an earlier article that I was not going to be posting the “predicted” electoral numbers for a bit because of the volatility that comes with shifting poll numbers.
For example, on the 19th, we would have had Obama winning Florida and Ohio because of the Quinnipiac results that were posted. On the 23rd, we would have had McCain winning in those states because of the Rasmussen Reports polls that were posted. Our algorithm was simplistic as could be in that we only looked at the latest poll and crowned the winner of that poll the winner of the state.
So a comment was raised by a poster on that article.
“How about using some form of a weighted average with more recent polls receiving more emphasis, as most sites seem to be using? It would greatly reduce the volatility and increase the reliability of the general election numbers.” – Jack.
I intend to have a weighting approach but I don’t feel that it is the appropriate time right now to do that. Here is my justification for thinking this way.
Imagine Pollster A has Obama down by 5% in mid-May but Pollster B has Obama up by 6% in June. Since poll B is earlier, it would make more sense to weight that a bit more.. and I would end up with something that was along the lines of a dead heat. But the reality would be that it is not a dead heat. The poll back in May would have been at a time when it was still uncertain that Obama was going to be the Democratic nominee and thus have lower numbers. I may be able to get away with weighting Poll A at a very low rate but imagine the same scenario occurred in another state with Pollsters A and B but with slightly different numbers but with Poll A being conducted in early June. The justification to include Poll A would exist for this state but the average lay-person would wonder why I chose to include Pollster A for one state but not for another.
It is true that the date is important but I want to remove all the potential biases from the reporting of these results… That gives you, the consumer of polls, the power to break the information down to how you would like.
The media as well as many of the other polling compilation sites would have a standard average of the last 3 polls. Not sure that is very accurate either. Consider the latest Florida polls where Obama came back from 16% behind. Should we be including those numbers into the mix? As we get closer to the election and start getting gobs of data points
So what I have come up with… is to report these numbers in a couple of different ways.. One would be pollster centered. Where we only look at the polls from one particular pollster and see the electoral map looks per pollster. It only really makes sense to do this for Survey USA and Rasmussen Reports right now as those are the pollsters with the most data points. This will make the pollsters accountable for their results if, by chance, their numbers are not jiving with those of the other pollsters.
We will also do a raw average and a weighted average of the more recent polls for each state just as the poster mentioned that we should. So long as all these results are reported in the context of the other data that we mention here, I think that would be the ideal and least biased way to do so.
Expect this feature to be on our site by the end of this week… Cheers!