Monday, October 15, 2012

How I Make Projections?

In the days leading up to the 2004 Presidential race I took my love for politics, math, and making predictions and merged them all together. While I had been closely monitoring the polling and trend lines for months leading up to the election, I began to look more into the numbers in order to make my own opinion on the presidential race. Of course we seen great success and no doubt some good luck as we correctly guessed the outcome of each state and thus what the final tally for the electoral college would be. I used the same exact system in 2008 and while I correctly predicted a comfortable win for Obama I did miss the outcome of three states. I do not pretend to have some type of secret knowledge or formula which makes my projections error-proof. However, I am also aware that no formula, polling agency, or political insider can ever have an error proof system. The biggest reason I do this in addition to the reasons cited above is that I like to have a good idea on election day which way things will probably go. Here are the answers to the most common questions about my projections...

  • Aren't polls terribly inaccurate and practically worthless? The biggest mistake I think people make is putting too much stock in one or two polls and then judge polling accordingly. I never put too much confidence in one poll and would rather look at what the consensus or average of what the reputable polling companies are reporting. I have found that when anyone finds a poll they really like they have a tendency to believe it, but when they find polls that are reporting the opposite they quickly discard it as being false or biased. Polling companies are critiqued heavily on their accuracy and thus they care about their credibility. Historically, the consensus or average of the polls together give a very good picture of what is going to happen. Let us look back to 2008 and consider the real clear politics final polls and average of what the popular vote would be between Obama and McCain. Notice the average of all the polls came within 0.3 of the final results.
Final Results------52.945.6Obama +7.3
RCP Average10/29 - 11/3----52.144.5Obama +7.6
Marist11/3 - 11/3804 LV4.05243Obama +9
Battleground (Lake)*11/2 - 11/3800 LV3.55247Obama +5
Battleground (Tarrance)*11/2 - 11/3800 LV3.55048Obama +2
Rasmussen Reports11/1 - 11/33000 LV2.05246Obama +6
Reuters/C-SPAN/Zogby11/1 - 11/31201 LV2.95443Obama +11
IBD/TIPP11/1 - 11/3981 LV3.25244Obama +8
FOX News11/1 - 11/2971 LV3.05043Obama +7
NBC News/Wall St. Jrnl11/1 - 11/21011 LV3.15143Obama +8
Gallup10/31 - 11/22472 LV2.05544Obama +11
Diageo/Hotline10/31 - 11/2887 LV3.35045Obama +5
CBS News10/31 - 11/2714 LV--5142Obama +9
Ipsos/McClatchy10/30 - 11/2760 LV3.65346Obama +7
ABC News/Wash Post10/30 - 11/22470 LV2.55344Obama +9
CNN/Opinion Research10/30 - 11/1714 LV3.55346Obama +7
Pew Research10/29 - 11/12587 LV2.05246Obama +6
  • Is their bias in polls? Remember, polls do not simply call a bunch of random people and say here is what is going to happen. They all have specific methods that they use to determine what percentage of each party to call, gender, race, and geographic locations. I do not believe any credible polling company decides to manipulate their findings to favor a candidate. However, their can be a real bias when determining what methods to use. In other words, they believe one party is going to show up more heavily on election day and they then poll a heavier percentage of people within that party. Some polls take in consideration an "enthusiasm gap" in determining what percentage of each party to call. This is done by asking the ones they call two questions. "How excited or enthused are you to vote for..." and "How likely are you to vote for..." Then they weight their poll accordingly. While I believe in the use of enthusiasm to weight party voting I also believe that this is where the bias comes in as companies can have a tendency to over react to what they are finding. I do weight or adjust polls according to their over all past accuracy when making projections.
  • Do you ever throw out polls entirely? Almost never. I just take into account their leanings and weight the findings accordingly. The only time I would ever throw out a poll is when their findings are all over the place in comparison the polling average. In other words their polls are always outliers and their seems to be no rhyme nor rhythm in which way they go.
  • What does trend lines mean? Which way the race is trending at the moment. In other words has one candidate been seeing consistent steady gains leading up to that moment. Trend lines are only accurate by comparing the same identical polling companies. This is one of the first things I look at in every poll; How does this poll compare to their last poll? Of course the polls need to be taken within one to two weeks of each other to give a more accurate picture.I do look at trend lines in making my projections and will give up to one percentage point gain if their appears to be a strong momentum for one candidate.
  • Do you take into consideration past results? Yes, but mainly this is utilized when their are little polls done in a particular state. After I make adjustments to the polls according to their accuracy and past leaning and average them together I ask myself these questions? "Are these results logical in terms of past results?" and "Are these results consistent with the general state of the race" Most of the time these questions do not make much change in my projections unless their is little polling to go by, because the polling is almost always consistent to these two areas. I would generally only adjust the average adjusted polling average by a maximum of one percentage point.
  • How much of your projections are formula and how much is your personal assessment? While I go strictly by my formula, which I do not change in the middle of a race, their are times that conjecture has to be used within the frame work of the formula. I would say that 75% of my projections are based on raw data and 25% are based on conjectures.
  • How much time does this take? Since I know my formula basically by memory the projections only takes within 15-30 minutes. However, the analyzing of the polls can also take some time on heavy poll days. My final projections which will include a margin of victory column for the swing states will require more detailed study and take a few hours.
  • Are the confidence percentages done by formula? No
  • How do you decide whether a state is a toss-up, leaning state, or safe for one candidate? Simply by my confidence level with one additional rule of thumb. I am never more then 65% confident of the outcome of a toss-up and more then 85% confident for a leaning state. But, I never consider a toss-up to have more then a 2 1/2 point margin and a leaning state to have more then a 4 point margin.
  • Is your formula a secret? Not really, since it requires an element of personal assessment and would thus be impossible to replicate.

No comments:

Post a Comment