Pollster.com

Articles and Analysis

 

Obama's Dog Days of Summer?


The first two weeks of August have not been good for Barack Obama. As we said last week, McCain's "Celebrity" ad--along with his campaign's subsequent attacks on Obama--blunted any momentum Obama may have gotten from his overseas trip and kept this thing close. Given that all of the internals (right direction/wrong track, generic congressional ballot and party ID) and the Illinois senator's huge "intensity gap" support an Obama/Democrat blowout, it is astounding that he is underperforming as much as he is. The electorate, of course, is still in flux, as a large segment of voters is still undecided (or switchable); it is clear, however, that Barack Obama has not come nearly far enough to close the sale on this election.

It is worth repeating what we said last week: the McCain attacks on Obama are working. And when you look at how long it took Obama's team to respond to the "Celebrity" spot (nearly two weeks) you can't help but be reminded of John Kerry in the summer of 2004. Voters don't just look at the candidates and their issue positions, they look at how the candidates run their campaigns and make decisions, and Obama has looked awfully soft in his response to this charge. Combine this with his tepid statements on the Russia-Georgia crisis and you have the makings of a legitimate campaign swoon. Plain and simple, the McCain team has been winning the earned media battle for the last two weeks.

This is a difficult election to classify because there is no incumbent president or vice president in the race, but it might be helpful to look at past elections to give us some guidance. While there are multiple ways to categorize elections, in every presidential election the two sides try to make the election hinge on some mix of referendum and personality (including the policies the candidate stands for). The winning campaign is the one that better succeeds in establishing its frame and making the case for it.

Some elections are more of a referendum and some are more about personality/issues. For example, 1980 and 1992 were clearly referendum elections. In both cases the electorate decided that things were bad and the alternative was acceptable (Reagan in '80 and Clinton in '92). In each case the referendum on the current administration worked for the challenger. In 2004 the direction of the country was poor but voters decided that the alternative (Kerry) was not acceptable. Kerry's referendum on Bush failed.

This year is clearly a referendum on Bush and the direction of the country and, as we have said before, if Obama is viewed as acceptable to a majority of voters he will win this election. Whichever side does a better job of framing the debate will win. Right now McCain is doing a good job of framing the debate as "this guy (Obama) isn't ready to lead." Obama's basic change thematic might be enough on its own because things are seen as so bad, but to improve his "referendum" position he needs to do a better job of tying Bush and McCain together.

So, at its core, this election is about Obama's ability to make this a referendum and him the acceptable alternative to the current course. Therefore, a McCain strategy to make Obama unacceptable is his only winning course of action. Any other strategy would be political malpractice. Contrast ads work. Anyone who says McCain has gone too negative too early has never been involved in a political campaign. In 2004, the Bush team started running attack ads against Kerry in March. Of course, that year there was a Democratic nominee much sooner but it shows that it makes sense to start defining your opponent in July and August.

Gaps Galore

According to a July Wall Street Journal poll there is both a "generation gap" and an "intensity gap" in the 2008 Presidential race. We have seen this in our own polling and in polling by other media outlets, as well. In this particular WSJ survey Obama leads McCain among 18-34 year olds by 24 points (55% to 31%). Among those 65 years of age and older McCain led Obama by 10 points (51% to 41%). There is also an enthusiasm or intensity gap between Obama's and McCain's vote with almost half (44%) of Obama voters saying they are enthusiastic about their candidate and only 14% of McCain voters saying the same. Inevitably, then, we have some questions that will be answered come November:

  1. What percentage of the 2008 electorate will be 18-29 year olds? If their raw vote totals are up but their share of the electorate remains the same then the impact is less. According to the VNS and NES exit polls, in 2000 and 2004 they represented approximately 17% of the vote. Yes, the raw vote total for 18-29 year olds increased significantly in 2004 but so did other age cohorts. If the 2008 youth vote share increases (to say 20%) and Obama improves upon the Kerry vote among this cohort (54%) then he will be tough to beat. If, however, the percentage of 18-29 year olds remains at about 17% and Obama does only marginally better than Kerry did with this group (let's say he wins that share of the electorate with around 56-58%) then I don't see the youth vote having as much of an impact. (An aside: according to Pollster.com contributor Charles Franklin, who uses more reliable Census CPS turnout data, the young did actually increase their share of the electorate, but not by an impactful margin. As he says, "Perhaps we will indeed see another rise, as we did in 2004. But unless something truly unprecedented occurs, no one can win on the young alone.")
  2. How much of the intensity/enthusiasm gap is due to Obama's overwhelming lead among 18-34 year olds? There is no doubt that the enthusiasm level among McCain's core vote needs to improve-and Obama's lead here is an important ingredient for driving likely voters--but I am not sure that the enthusiasm difference isn't being artificially inflated by the youth vote (a cohort that doesn't historically vote in overwhelming numbers). Again, time will tell.

National Horserace Observations

As the below chart indicates, the race remains close. From a macro level, Obama was poised to blow this race open in mid- to late-June when most polls had him up by double-digits. Since that time we have seen the gap close. Also, take a look at the numbers in March when the Reverend Wright story broke. Yes, Obama was still engaged in a primary battle and Clinton voters were not likely in the fold yet, but clearly that news story was a staggering blow and it showed up in his head-to-head numbers with McCain.

horserace aug 13.png

 

Comments
werty83:

Steve,

You credit the celebrity ad with blunting Obama's momentum, but it looks to me like McCain has tanked in his support since it came out.

____________________

dwschwartz:

Your entire hypothesis is contradicted by the fact that just today (or maybe yesterday) Pollster moved a significant number of electoral votes from McCain to undecided.

____________________

Alan Abramowitz:

There is not a shred of evidence in the polling data to support Lombardo's contention that McCain's attack ads have weakened Obama. In fact, the Gallup tracking poll shows Obama gaining strength in recent days. This reads like pure Republican propaganda.

____________________

1magine:

Close? WTF are you talking about. Who shows this race as close? 2% is not really close on a national level. 3-5% is an astounding blowout; especially given a highly divided country. To remind you: 1% = a 70% chance of EC victory. 2% = 90% chance of EC victory. 3% = 97% liklihood of an EC victory. What swing states is McCain ahead in? VA? CO? OH? NH? No, No, No, no- heck the fact that red - red states like VA MO, IN, & Co are even in play let alone showing a "slight?" lead for BO speaks volumes.

From the day HRC suspended her campaign till today - exactly what has changed in the polling numbers? 3-5% then, 3-5% today. Which given JM attempts to frame BO, that is remarkable and quite frankly the OBVIOUS point of the overseas trip. Not to buoy that day's poll numbers, but to lay a floor underneath them. Which it seems was done successfully.

BO -is ahead. Only distorted minds see this as a loss. "He's not nearly as far ahead as he should be..." It reminds me of the $8K JM spent creating an ad showing how BO used injured troops as a media prop on his overseas trip. Even after the Pentagon told him not to go. It wasn't used because BO didn't go. Instead the alternative ad was used claiming he ignored our injured troops. (PS who is using injured troops as a prop?) In your warped sense of the world, everything that surrounds this man (BO)is wrong. And so a lead is losing, winning isn't winning by enough... And that is sad, that you can not escape your own darkness long enough to think rationally about the numbers right in front of you - On a web site designed for the rational use of numbers to determine the state of various races!!!

Lies, no matter how loud are lies.

Truth, no matter how soft is truth.

____________________

Y'know, I expect better from Pollster than this. I come to this site for the data and intelligent analysis, not dumb analysis whose basic point is completely contradicted by the data he presents at the end of the article.

McCain's celebrity ads have utterly failed to move Obama's poll numbers and, if anything, seem to have pushed his own downward.

____________________

jsh1120:

Mr. Lombardo makes as good a case as possible for his nominee. I, too, am inclined to doubt the effect of the "celebrity" line of attack in the long run, but it's clear that the GOP believes in it. Possibly because the strategists see no other alternative and adhering to the Atwater/Rove "attack the other guy's strengths" approach seems the only viable strategy.

Personally, I'm inclined to be skeptical about the entire "Obama is underperforming" narrative pushed by the GOP and the media (for different reasons.) Rather, it looks more and more as if much of the electorate has taken the summer "off" from the campaign. Much to the frustration of (us) political junkies. We were suckered into believing the elections was at the top of everyone's priority list by the Democratic primaries and now find it unimaginable that a significant portion of the electorate hasn't maintained their interest.

In such a climate it's hardly surprising to see GOP voters "come home," especially when the Republican campaign has fed their appetites for attacks on Obama. Likewise, it's not surprising to see the likely voter models revert to a "non-campaign" state where GOP voters' "habit" of voting is a more important component than campaign enthusiasm.

The overall "drift" (excluding short term factors) seems to be slightly in Obama's favor. I suspect that the Obama campaign is happy to see such stability at this point, knowing that enthusiasm can only be sustained for so long and that peaking in November is far better than peaking in August.

Obamaphiles' hopes for a landslide are probably unrealistic, but they always have been. I've always thought that a three to five point gap in the two party popular vote is the safest bet. I see no reason (yet) to change that guess.

____________________

David_T:

The last Gallup tracking poll to be taken entirely in July showed Obama and McCain tied 44-44. Today's Gallup tracking shows Obama leading by six points.

The last Rasmussen tracking poll taken entirely in July showed Obama leading McCain by one point. Today's Rasmussen tracking poll shows Obama leading by two points. If the celebrity ads, etc. have helped McCain, the August tracking polls sure don't seem to indicate it.

I have a feeling that if in the end Obama defeats McCain by two points (and wins in the Elctoral College as well) Steve Lombardo will say that it was "really" a victory for McCain because Obama should have done better!

____________________

Allen:

Nice try, Steve.

The race has remained essentially unchanged since late June, with Obama firmly in the lead.
See http://election-projection.net/

Obama will be happy to take three more months just like July, and then when November comes, cruise to victory.

____________________

bmcfar01:


1magine,

2% is close when you have 10%+ undecided.

____________________

bmcfar01:

Allen,
No one is "firmly in the lead" if they are under 50%. Obama hasn't polled at 50% or more in over a month, and before that, it was very sporadic.

His achilles' heal is his inability to "close the deal" (I've heard that phrase so much it makes me sick). His challenge is going to be get the 10%+ to break decisively for him.

____________________

bmcfar01:

(clarification)

10%+ =undecideds.

____________________

Lombardo's statement that the race has narrowed is true. The problem is that national data alone contain too much fluctuation to make the trend apparent.

Instead, see the meta-analysis of state polls at http://election.princeton.edu/history-of-electoral-votes-for-obama/. The meta-analysis uses dozens of polls at once to generate an Electoral College-based estimate. It gives a far more accurate snapshot than a national polling average. This is obvious when you look at its history over time, which moves up and down much less than the national average shown here.

The history plot there clearly shows that since June there have been several major swings in the Obama-McCain matchup. The most obvious one is the uptick for Obama after Hillary Clinton's withdrawal from the primaries in early June. Now, in August, the same measure is starting to show a decline in Obama's support. The fall may not yet be complete, since state polls in August are conducted infrequently, leading to a delayed response in the electoral vote index.

I do think Lombardo has made an excellent point: given Bush's low approval ratings, Obama has clearly not met his full potential. I'm not sure that the Obama campaign's pushback has been slow so much as weak. The sense of vigor and fun from earlier in the year doesn't seem to be there. For instance, Paris Hilton's comeback was brilliant - why hasn't the Obama campaign grabbed onto that?

____________________

Allen:

Hello Dr. Wang,

You are correct of course that national poll data is meaningless--you have to look at the individual states to determine the status of the race.

Since late June, the only two significant states where the race has tightened are NH and MI. On the other hand, certain other states such as CA, MA, CT and NJ have made it clear since late June that they are solidly Obama. These two factors offset and the net result is that the race has remained essentially unchanged since late June.

There is a possibility that the race has narrowed slightly in August, but the jury is still out on that. Until more polling data comes in, it is impossible to make a claim that early August is significantly different than July.

The problem with the analysis at Princeton is that is includes only three polls for each state. For this reason, the results are just as likely to be looking at noise as actual voter sentiment. The analysis at election-projection.net includes up to 7 polls, with newer poll weighted more than older polls, so it is less sensitive to noise.

Another problem with the analysis at Princeton is that it only reports electoral votes. On November 5, the difference between 300 and 270 electoral votes will be pretty meaningless, while the difference between 268 and 270 electoral votes will be enormous. For that reason, if you really want to know that state of the race, you have to look at the probability of each candidate breaking 270 electoral votes. election-projection.net also reports that figure.

Finally, while not related to the above, it should be noted that the confidence margins in the Princeton results are simply too high. It does not appear that these results account for any potential sources of error other than sampling error. The results at election-projection.net include a margin for non-sampling error, and also a margin for potentially correlated poll errors, i.e., errors that could cause the election results to be higher than the polls across several states, or at least across the battleground states where it matters the most. If the Princeton analysis included reasonable values for non-sampling and correlated errors, the results would have a wider distribution of electoral votes and correspondingly lower confidence of the leading candidate achieving 270 electoral votes. In other words, it would end up looking almost exactly like the results reported at election-projection.net

Best regards,

Allen
http://election-projection.net

____________________

brambster:

I do believe that McCain going negative helped blunt the impact of Obama's Middle East and European tour. I don't think it swung any extra support to McCain, it just stopped it from flowing towards Obama. There were certainly some polls that were suggesting a surge for Obama just after, but they quickly sunk back to the expected range.

This was really before the Britney/Paris attack ad. I don't understand how such an ad has any effect, nor the Obama response ad. These ads only serve as talking points for partisans.

____________________

werty83:

Professor Wang,

The problem with looking at the state meta-analysis in order to interpret changes in the race is that it doesn't match up well w/ current events. Because of their infrequency right now, we can't say that the drop you show in August is necessarily related to the events of August or the events of four weeks before when the last polls were put out.

____________________

werty83, I agree that there's a problem with poll infrequency. However, the issue is not whether the drop is linked in time to August events - you can examine the graph closely and see that it is. It's just that the response is delayed in time, what engineers and scientists call "filtered."

I think the main issue is that we don't know how large it will be yet. My sense is that the peak effect won't be that large since national polls don't show a reversal of the margin.

____________________

Sam Wang:

Allen's comments are not entirely accurate. The meta-analysis takes into account both sampling error (as a minimum error) and inter-poll variability present in the data.

In addition, the stability of the meta-analysis shows that it is not sensitive to fluctuations in the manner he describes. This is evident from inspecting the graph.

The question of correlated errors is an interesting one mainly if one is projecting far enough that there is a chance of movement in one direction or the other. The point here is to present a high-resolution snapshot of where things are. The future is harder to see. As Niels Bohr and Yogi Berra both said, "Prediction is hard, especially of the future."

____________________

richard pollara:

Allen: I am not quite sure I understand how gains that Obama has made in states that he was already likely to win offset a tightening in NH and Michigan.

____________________

Allen:

Hello Dr. Wang,

You wrote:
The meta-analysis takes into account both sampling error (as a minimum error) and inter-poll variability present in the data.

First, the sampling error is a minimum, but the actual error is always going to be larger, and an analysis that does not take this into account will overestimate the level of confidence.

Second, if you had enough polls, and if the polls were all unbiased such the expected value of the poll results in each state was always equal to the actual election outcome, then a confidence estimate derived from the poll variability would be an accurate figure. The assumption in bold however is never going to be true. No matter how many polls you look at, biased errors cannot be seen by looking only at the poll results, and therefore the model should include an actual confidence estimate that is larger than the poll variability.

Third, the potential for correlated poll errors also has a significant effect on the confidence of the outcome. Correlated errors simply means that an aspect of methodology that causes the poll results to skewed toward candidate A in one state is likely to make the poll results skewed toward the same candidate in other states. For example, if the likely voter screens or assumed party turnout is wrong in one state, it is likely to be wrong by a similar margin in other states. These are potentially correlated errors. Some other sources of potentially correlated errors are sampling bias, non-respondent bias and non-truthful answers (for example, some voters say they will vote for Barr even though they have decided to actually cast their ballots on election day for McCain). All of the potential sources of correlated errors have nothing to do with projecting the results into the future; they have everything to do with assessing the accuracy of the current estimate.

The effect of modeling errors as correlated vs. uncorrelated is also significant. Suppose there are two key states in the election. One candidate is slightly behind in both states, and you estimate he has a 25% chance of winning each state. If you treat these states results as independent, then his probability of winning the election is 6% If however the poll errors in those states were perfectly correlated, he would either win both with probability 25%, or win neither with probability 75%. The difference between these two probabilities is large: 6% without correlated errors and 25% with perfectly correlated errors. The actual result is somewhere in the middle, partially uncorrelated and partially correlated. However, the more events you string together (win state A + state B + state C + ...), the more unlikely the uncorrelated results becomes, while the correlated result does not change. An election strings together approx. 18 of these events (the battleground states), and therefore the results including correlated errors do differ significantly from the results using only uncorrelated errors. This can easily be seen using the correlation control in the Interactive Presidential Election Probability Calculator at http://election-projection.net/interactive.html

Finally, it should be noted that overestimating the level of confidence is "worse" than erring on the side of caution. The Princeton results show an approximately 99% level of confidence that Obama is currently ahead, but a more realist figure, taking into account the potential for other errors in the polls including correlated errors, is approx 90%. This is the result shown at election-projection.net

Best regards,

Allen
http://election-projection.net

____________________

Allen:

@ richard pollara
"I am not quite sure I understand how gains that Obama has made in states that he was already likely to win offset a tightening in NH and Michigan."

Because these states have so many electoral votes that even a small chance McCain could win them effects the estimate of him winning the election. You can demonstrate this using the Interactive Presidential Election Probability Calculator at http://election-projection.net/interactive.html

____________________

Joseph Marshall:

Stepping aside of all the high powered number crunching for a moment, I think something important is being missed. Let's ask the question this way:

What does John McCain need to do to win this election?

First of all, he has to achieve a higher popular vote count than the 42-43% he's been floating at in the Gallup Daily. He needs to pull up a minimum of 8-9 percentage points to win. Since there are only 10-11% uncommitted at the moment, McCain would have to win the overwhelming majority of these voters.

Second, to win the EC he needs 270 votes. Take the state by state trend estimates on this site and only 236 of these are within his immediate reach. This assumes he wins every state where he currently has any lead whatever, no matter how slim.

So he needs to find 34 more electoral votes among states where he is currently not leading. To do so he must win an absolute minimum of three more states, one of which must be either Ohio or Michigan.

Based on this same standard of "immediately in reach" Obama has already far exceeded 270.

This gives us some basis for evaluating Mr. Lombardo's pithy remark:

"Therefore, a McCain strategy to make Obama unacceptable is his only winning course of action. Any other strategy would be political malpractice."

The implication of this is clear. Mr. Lombardo doesn't believe that John McCain can win this election on John McCain's own merits. Based on the strategy they have so far used, McCain's campaign seems to believe this, also.

Further, I have yet to see any of the talking heads step forward and say that McCain CAN win on his own merits, so I presume the pundocracy is in agreement with Mr. Lombardo as well. Nor have I seen any speech of McCain's that indicates that he himself believes he can win on his own merits.

I don't think he can either.

This is why McCain is currently losing. By nominating someone who is not capable of convincing people to vote for him, rather than against the other fellow, the Republican Party has started out ready to lose rather than ready to win. They probably will lose.

There are several compelling reasons why "making Obama unacceptable" is a lackluster strategy at best. First he clearly believes in himself, and has no trouble communicating that he belongs in the White House on his own merits.

Second, in public he is an engaging and likable guy who is perfectly comfortable with crowds of people, no matter what the size--probably because he does believe in himself.

Third, and most important, his swift rise to prominence has left him without obvious vulnerability from past political blunders. So to make people dislike Obama, the campaign has to simply make stuff up.

Fiction is a talent. It demands that you entertain. Neither the polls nor the commercials suggest as yet that anybody in the McCain campaign has even as much talent as Paris Hilton.

Until they find some talent, making Obama unacceptable will be as hard, if not harder than trying to convince hundreds of thousands of people that John McCain actually a good choice for President rather than the least worst choice.

____________________

Joseph, I agree both on the number-crunching and on why the campaign has taken the tone that it has. It seems to me that the McCain camp's main option is to make the celebrity/white girl/Antichrist attacks even more explicit. I'll guess that this month we will see their high-water mark in polls.

In 2004, if one looks at the numbers at http://synapse.princeton.edu/~sam/history_12oct.jpg (sorry, Joseph, impossible to resist!), the major swings of opinion happened after the following events, in order of time: Fahrenheit 9/11 release, adding Edwards to the ticket (probably wouldn't work as well this time), the party conventions, the Swift Boat attacks, and the first debate. So if I am wrong and things stay close, the debates may be deciding.

Assuming no crazy surprises...

____________________

richard pollara:

Allen: Just wondering if that is the same election projection model that gives McCain a less than 10% chance of winning and only a 40% chance of winning if he were to win Ohio and Michigan? Think anyone out there will give you 10-1 on McCain or 3-2 for Obama assuming he loses Ohio and Michigan. If you know such a person, please put him in touch with me!

____________________

Allen:

@ richard pollara

The answer to that question depends on what assumptions you are rolling into the question. Are you asking what is the probability of McCain winning the election if he were to win OH and MI on November 4? Furthermore, I'm wondering what application or purpose you see for this question, i.e., how would you use the answer? Would you use it to determine which states Obama and McCain should fight the hardest over, or would you use it to place bets on various scenarios that might happen on November 4, or would you use it for some other purpose? This is important because the assumptions built into both the question and the answer need to be consistent with how the information will be used. Without knowing how you plan to use the information, it is not possible to know if the answer is valid.

____________________

richard pollara:

Allen: The point I was making is that while Election Projection is a fun site to play with, it does not accurately predict the likely outcome of the general election. On a post I made a couple of weeks ago I said that it was a bad sign that Obama was spending resources in places that were not likely to matter in the fall. It made me think that McCain's campaign had a better grasp of the electoral map than Obama's. Obama's campaign (IMO) would be better served by pulling out a few history books than playing around with Election Projection. Of course when you position yourself as a post-historical figure, the past no longer matters...

____________________

Hello Richard,

election-projection.net makes no attempt to predict the outcome on November 4. As it states on every page: "If the election were held TODAY...". To further quote from the Methodology at http://election-projection.net/methodology.html#Results:

---------

ELECTION-PROJECTION.NET answers the hypothetical question "Who would win if the election were held today?" The results give the probability of each candidate winning and the distribution of expected electoral votes if the election were held today.

The results can and will change as public opinion changes toward one candidate or the other. ELECTION-PROJECTION.NET does not attempt to predict any of the following: How far and in what direction public opinion may shift between now and the election day; how many voters will change their minds between now and election day; whether undecided voters will "break" more toward one candidate or the other (it essentially assumes undecided voters will vote the same as the decided voter); or how many new and previously unaccounted voters will enter the picture.

The net result is a accurate snapshot of the current state of the race. It allows you to immediately see who is "ahead" right now, and how changes in public opinion influence the likely outcome of the election.

In the days just prior to the election, the numbers reported here will be one of the best available predictions of who will win. In the early days however, it should be recognized that things can and probably will change between now and November 4.

---------

There are some important distinctions between sites like election-projection.net and 538.com. 538.com is attempting to somehow model the myriad of unpredictable events that might happen between now and election day. Good luck. It should be noted that an election is nothing like baseball. In baseball, the same players do the same things over and over--pitch, swing, run and catch, thousands of times--and their performance is measured each time. The season is the result of a series of independent contests, stretched over time. In politics, the "season" has only one game, the players are different each year and the moves are constantly changing. The prediction methods that work in baseball simply do not apply to elections.

The important thing to understand though is the difference in how the results can be used. The results at 538.com are suitable for purposes such as placing bets on what might happen on Nov. 4. The results at election-projection.net can be used as a starting point to make a prediction, but the unknown events between now and Nov. 4 should first be factored in. The results at 538.com can be used to make statements like "The mostly likely outcomes are XYZ", and "If McCain wins the election, there is a 70% probability that he won Ohio."

The results at 538.com cannot be used however to make the following statements: "The most important states are XYZ"; "McCain and Obama should focus their resources on XYZ"; "If McCain wins X, then he has a 70% probability of winning the election". The reason 538.com cannot be used to make those statements is that each of these statements imply independence between the states. When we ask "If McCain wins X", we are implying that this is independent of all other states. Similarly, when we say the candidates should focus their resources on XYZ, we are implying that this is independent of the other states. The model used by 538.com breaks the independence assumption, because it assumes the states will move together, in demographically similar groups. If OH is up, MI, PA, etc, will also be up. Using 538.com's results, you can accurately say that "If McCain wins OH and demographically similar states, then he has a 70% probability of winning the election", but you cannot say "If he only won OH, independent of the other states...". Similarly, you could use the results at 538.com to say that the candidates should focus their resources on OH and demographically similar states, but the results say nothing about what would happen if they focused resources only on Ohio, independent of other states.

In contrast, the results at election-projection.net give you a current snapshot of the election, and for that reason are more useful for tracking the current state of the race because they do not muddy the waters by projecting the #'s forward to Nov. 4. In addition, election-projection.net does not model dependent movement of the states. For that reason, the results are suitable for asking all of the questions above that assume independence.

In summary, the results at 538.com are suitable for:

- predicting the outcome on Nov. 4
- examining the relative probability of various outcomes on Nov. 4

while the results at election-projection.net are suitable for:

- tracking the current state of the election
- determining which states are most likely to effect the outcome
- determining where the candidates should focus their resources
- determining conditional probabilities that assume independence

That is why I stated that it is important to understand how you intend to use the information before you can know if the answer is correct.

Also, one note about 538.com's predictive ability: no where does the 538.com model include factors for the amount of resources the candidates are expending in each state. It assumes all states demographically similar to Ohio will move the same, even if the candidates are expending no resources in those other states on organizing, advertising and campaigning. If you believe that "campaigns matter", then this essentially means that 538.com's results ignore on of the most important factors. For this reason, I think the political markets like Intrade are probably most better predictors of the Nov. 4 probabilities than 538.com

Finally, you may believe the Obama campaign is wasting resources where it has no chance of winning, but (a) I assume they have better info and better models than either of us, and (b) there are other reasons to expend resources on a state such as party building or forcing the opposing candidate to expend resources defending the state. The Obama campaign has proven itself to be so well run that I am not going to second guess anything they do.

Best Regards,

Allen
http://election-projection.net

____________________

richard pollara:

Hi Allen: A couple of points. The snapshot analogy gets bandied about quite often for polls and other predictive tools. I would respond that Polaroid is in snapshot business and not Survey USA, Insider Advantage or Election Projection. These people are all in the business of telling us the future and like soothsayers of the past their methods are suspect and their results often wrong (excluding the Ides of March of course). A better qualifying statement would be, "if the election were held today and the data, which we know to be unreliable and our methods, which we know to be unsound, are true, then this is the likely result (plus/minus a margin of error). It would seem to me any predictive tool which gives John McCain less than a 10% chance of winning is suspect.

Regarding the "betting" markets: They are notoriously inaccurate and really are predictors, not of the outcome but of what people think.. other people think.. the outcome will be. I pointed out in early March that Obama was a 4-1 favorite to win Texas even though the race was essentially tied. If you have misspent as much time as I have at Casinos you will realize that gamblers are second only to stock pickers in herd mentality. Pick a long line and get in it should be their slogan.

The one thing that I do think has been predictive this election cycle has been demographics. Find a voting pattern in a group (say white women over 65) and that pattern will likely repeat itself. I don't pretend to understand the methodology that Nate Silver is using (either for baseball or politics) but it became clear to me in early February that looking at the percentage of African American voters in a state and the income of the state's white voters was much more predictive than all the polls combined.

Perhaps I am being too hard on the pollsters but there is an assuredness about their results which is so contrary to the facts that I think their place in politics and the media needs to be reevaluated. I have mentioned this before, but the Zogby release on the day of the California Primary was an egregious example of the unhealthy alliance between pollsters and the media. Tom Brokaw's grave announcement that Hillary was going down to a double digit defeat in California bordered on vote rigging.

As for my prediction, I think this race is essentially a toss-up with the winner of Ohio most likely to prevail. It worries me that Obama has not spent the resources (financial or intellectual) to connect with voters in that part of the country. The talk on August 16 may be about Butte and Alexandria, but I think on November 4, we will be focused on late returns coming in from Youngstown and Dayton. It is hard to say President McCain, but I guess I would give it a better than 10% chance...

____________________

Allen:

These people are all in the business of telling us the future

There I think you are wrong. The outfits you list are all in the business of telling us the current state of the race. I can understand your desire to know what will happen on Nov. 4, but no one can tell you that. I do agree with you however that the methods of many pollster are questionable, even more so the media-run national polls. I do not agree though that the Obama campaign has not spent enough resources on OH. The swing voters are simply not ready to make up their minds yet (why should they), and the Obama campaign has been doing what it needs to do--laying the ground work by introducing Obama to the voters and setting up a ground operation. The things you are looking for will be be happen starting next week and continuing through election day. Also, while OH is at the moment the most important state in the election, it would IMO be a mistake to put all the eggs in one basket. Of course, campaigns, like athletic teams, always seem to be judged mostly in hindsight--if you win, your strategy was brilliant, and if you lose, your strategy was a wrong :-)

____________________



Post a comment




Please be patient while your comment posts - sometimes it takes a minute or two. To check your comment, please wait 60 seconds and click your browser's refresh button. Note that comments with three or more hyperlinks will be held for approval.

MAP - US, AL, AK, AZ, AR, CA, CO, CT, DE, FL, GA, HI, ID, IL, IN, IA, KS, KY, LA, ME, MD, MA, MI, MN, MS, MO, MT, NE, NV, NH, NJ, NM, NY, NC, ND, OH, OK, OR, PA, RI, SC, SD, TN, TX, UT, VT, VA, WA, WV, WI, WY, PR