Monday, October 15, 2012

Detailing my Formula on Projecting Ohio

It is no mystery that no republican has ever won the presidency and lost Ohio and I still see that being the case in this election unless Wisconsin swings into Romney's column. Let me illustrate or detail my formula a little bit more. Since Ohio is such a critical states lets examine it further.

1. First I look at all the polls that meet the qualifications of realclearpolitics...
RCP Average
10/5 - 10/13
--
--
48.3
46.1
Obama +2.2
10/12 - 10/13
880 LV
3.3
51
46
Obama +5
10/10 - 10/10
750 LV
4.0
48
47
Obama +1
10/6 - 10/10
1313 LV
2.7
45
46
Romney +1
10/7 - 10/9
994 LV
3.1
51
45
Obama +6
10/5 - 10/8
600 LV
4.0
47
48
Romney +1
10/5 - 10/8
722 LV
3.5
51
47
Obama +4
10/5 - 10/8
808 LV
3.5
45
44
Obama +1

2. Now comes the tricky part...Adjusting polls according to past accuracy and their relationship to the average in come cases. This is done in two different ways. Polls that I consider reasonably accurate I slightly adjust based on past leanings. -OR-Those polls that have very questionable methods that have more clear bias or those that are clear outliers from the norm will be adjusted by comparing their poll results to the partially adjusted results. For example:

  • PPP- Is probably the most accurate democratic pollster and has a slight democrat bias. But, since they are within three points of the initial average they will only be reduced one point from the average. Obama by 5 now becomes Obama by 4.
  • Rasmussen Reports- Is a non-official republican poll that has good accuracy, but generally only in presidential races or when their is not a large enthusiasm gap, since they use the same party weighting in every poll. They do have a slight republican bias but since they are within three points of the average their is only an addition of one point for Obama. Obama by 1 becomes Obama by 2.
  • Gravis Marketing- If their is one polling company that I could yet decide to throw out entirely it is this one. They are new in this election cycle so has no track record, but their numbers are normally all over the place and have shown generally unrealistic margins. In this case since they are showing only a three points gap from the average we will treat them as a pollster with questionable findings and compare their poll to the adjusted average at the end and take the median of the two.
  • NBC/WSJ/Marist- They are an extremely reputable polling company that generally show slight bias for the democrats, but in this case they have a very questionable weighting scale within the poll. They polled 40% democrats, 29% republicans, 29% independents. Compare these numbers to the lopsided turnout in 2008 in which 39% of the voters were democrat and 32% of the voters were republican, which was higher for dems and lower for the GOP then typical. This poll is predicting the largest voter turnout for democrats in modern history and the lowest turnout for republicans in modern history. They likely way over compensated what they seen as a enthusiasm gap for democrats. I personally think we are going to see more average or normal turnouts by party in this election. Therefore, I treat this poll the same way I do the Gravis poll and that means we will have to compare this finding to the adjusted average at the end.
  • ARG- The American Research Group is one with very mixed and more unreliable results throughout the years. They also have on average a more republican leaning. In this case since it is basically within three points of the initial average we will be taking the median of this poll to the initial average and therefore, Romney by 1 becomes Obama by 0.5.
  • CNN/Opinion Research- A poll with a long track record of slight democratic bias. Therefore, we discount Obamas margin by one. Obama by 4 becomes Obama by 3.
  • SurveryUSA- A long reputable polling agency with no discernible bias. This means that their polls can sometimes slightly favor democrats and sometimes slightly favor republicans. No adjustment is needed  in this case.
Adjusted Numbers are...
  • PPP-Obama by 4
  • Rasmussen- Obama by 2
  • ARG- Obama by 0.5
  • CNN/Opinion Research- Obama by 3
  • SurveryUSA- Obama by 1
Adjusted average of these 5 polls are Obama by 2. Now lets add in the two other polls in strong question.
The median of Obama by 6 in the NBC poll and the adjusted average of Obama by 2 makes this poll become Obama by 4. Gravis Marketing Romney by one becomes a virtual tie.

Therefore the final adjusted polling average is still Obama by 2. Basically the exact same as the polling average of RCP. Not always is this the case and sometimes it can swing the race up to a point or two depending upon the track record of the polls that are done.

3. Now we look at which way the state is trending. For this we use the initial numbers compared to their previous findings, assuming the poll is no more then roughly three weeks told. 
PPP- Moved towards Obama by 1
Rasmussen- Unchanged
ARG- Moved towards Romney by 2
Gravis Marketing- Moved towards Romney by 2
NBC/WSJ/Marist- Moved towards Romney by 2.
Therefore, we see a slight trending towards Romney and will give him a .5 bounce. Obama by 2 becomes Obama by 1.5. (In cases of one candidate trending either .5 or 1 point will be added to the final adjusted polling average.

4. Now the first crucial test...Is Obama by 1.5 realistic with the over all state of the race? Yes. Virginia and Florida are both extremely close and slightly favor Romney at the present, but generally they are more republican in past elections then Ohio. I could give several more examples which furthers my assessment. No adjustment is needed.

5. Lastly, the second crucial test...Is Obama by 1.5 realistic to their past voting trends? Definitely. I think every one knows this state in close elections show very similar numbers to the overall popular vote. No adjustment is needed here as well.

No comments:

Post a Comment