|
this guy says Kerry will win by a huge margin on November 2. He makes a lot of good points.
Election Analysis: Why Bush is Gone I don't really want to turn this into a political blog, so this will likely be my last election post. It also fits in with my prior posts on Futures vs Polling. If I offend anyone, it's because of my socially insensitive Aspie side.
I wanted to go beyond the superficial reporting of polls in the media and produce a fairly predictive analysis. (The best way to improve your analytical skills is to make a prediction using best available evidence and then test it your assumptions against reality. If I am wrong when the time comes, you can leave a comment of ridicule.)
My American History professor in 1988 indicated almost a month before the election that Bush 41 would handily defeat Michael Dukakis, because historically an opponent could not recover a trailing campaign in so late a stage in the game. He was right. I too am going to make a prediction one month in advance.
(In 2000, I used statistics to predict the election. I concluded the opposite of what actually happened--that Gore would win the electoral vote, and Bush would win the popular vote, by performing a weighted average of all poll results I could find. I missed my prediction on two states--New Mexico and Florida--both results hinged on just a few hundred votes. Also, nearly every poll in my data had Bush out in front in the popular vote. Here's my 2000 election calculations. I messed up the variances in the spreadsheet, so the probability estimates were wrong. )
If you have been following one of the electoral college sites, either Electoral-Vote.com or ElectionProjection.com, you may have been lead to believe that Bush was or is currently leading. Neither site, though, adjusts the electoral college count using the historically validated 50% rule, which I'll mention later.
My post will predict, despite what current polls and futures indicate, that, even without the first debate fiasco of Bush, the election will go handily to Kerry. Surprisingly, the electoral count will NOT even be close, barring an act of God or Osama. I believe this conclusion will all make sense in hindsight.
I believe that campaign insiders know this and this also explains the especially hard-hitting, negative election tactics of the Republicans this year, and the more subdued approach of the Democrats. Bush just recently signed a tax cut bill in a tight county in Iowa, and a number of election specific bills have raced through Congress in the past month.
My analysis looks at several trends such as cellphone usage, higher activism, larger turnout, and historical performance of polls.
The Fifty Percent Rule
From the following article, The Big Five Oh, Guy Molyneux convincingly makes the case the polls should be looked at as a referendum of the president, who is running for re-election. In the last four presidential elections involving an incumbent, this 50% rule was extremely accurate as you can see in the article.
A president should be polling at least 50% (49% is borderline), because historically the overwhelming number of undecideds actually opt for the challenger. No president has won reelection with numbers below 50%. The challenger's poll numbers are not as important as the challenger will typically pick up all the remaining slack. Why is this so? First, undecided voters are typically undecided, because they are disenchanted with the incumbent, and upon voting, typically asks themselves "Should the president be re-elected?" Second, incumbent presidents have instant name recognition, whereas the challenger may not be well-known to pollee. Some pollees may indicate a preference for the incumbent before learning more about the challenger. These pollees roughly cancel out the number of undecideds that stay with the incumbent.
As a result of this rule, the spread between an incumbent and a challenger can easily shrink by 5 or 6 percentage points. Because many of Bush's poll numbers in the battleground states were quite a bit off of 50% even before the debate, this is likely to be devastating.
Not only are the battleground states likely to fall but many of his supposedly solid red states he won from 2000 are also in danger, being under 50%, even before the debate; hence, my earlier conclusion that the election will not be close.
Distorted Polls
In 2000, an overwhelming number of polls had Bush out in front in the popular vote, though Gore won it by half a million votes. Divergences between traditional polling methods and election results are becoming more pronounced. The leading numbers which Bush garners are likely biased for him due to these divergences. Already, a few expert pollsters like Zolby are calling the election for Kerry.
Much of this discrepancy can be explained by increasing cell phone usage, which is distorting the results of phone polling. It is illegal for pollsters to call cell phone numbers, because cell phone users incur costs of each call. Thus the sample pool excludes many cell phone users.
Cell phone usage surpassed landline usage at the end of 2002, and many users now only use cell phones. There are major demographic differences between cell phone users. Cell phone usage is primarily an urban phenomenon (a majority of metropolitan residents own cell phones), because of wider coverage and greater need, and also tends to be concentrated in younger populations. Both of these environments have disproportionately greater Democratic representations. In 2000, Gore pulled in voters by a 3 to 1 margin in large cities of over 500,000 and 3 to 2 margin in urban areas of over 50,000 to 500,000. In contrast, small towns and rural areas gave bush 60% of their votes.
The net result is that the makeup of the polling sample pool have experience an inversion in the makeup of Democrats and Republicans. I believe the Gallup/USA Today/CNN pools shifted from a makeup of 39% Democrats and 35% Republicans in 2000 to 40% Republicans and 31% Democrats in 2004. Also in that makeup, the Democrats tend to be more conservative-leaning; Democrats in the sample are far less supportive of their candidate than Republicans are of Bush. I have also noticed that a large percentage of pollees believe that Bush will win, which one could interpret the sample pool as being drawn from mostly Republican communities.
This skewed makeup may also explain why Kerry received little to no boost from the Democratic Convention, while Bush receive a double-digit boost from the Republican Convention. It may explain why Edward's inclusion has not drawn his state North Carolina into the Democratic camp in the polls. It may also have created, along with the 2000 election, the perception of a very divided nation as well as stronger than expected support for war. More importantly, it could have robbed the Bush administration of an accurate measure of voter sentiment to guide his policies.
The Republican campaign are contending that the new makeup reflects a major shift of political affiliations during an Republican administration (or, equally likely in my opinion, the impact of the new conservative media outlet, Fox News Channel.) It is desirable for them to mention that, as positive press from poll results, can increase momentum and influence votes. However, such a dramatic shift hasn't been observed in the past, and, instead, as a counterargument, the party distributions are not matching up with actual state-reported registrations and there appear to be dramatically more new registrations for the Democratic party.
Activism and Increased Registrations
Polls also don't count the million of voters overseas. Voter registrations overseas are significantly higher this year, and there is generally the view that these voter, being international-minded, are against the Bush administration.
The methodology polls use also tend to miss first-time voters, which are expected to be far greater due to the closeness and bitterness of the last election and the anti-war sentiment. The Democratic party has been more successful at mobilizing voters, especially with the help of unaffiliated activists. (While I am not sure if this is an apples and oranges comparison, total donations to Democratic presidential candidates were over 50% higher than for the single Republican candidate, George Bush.)
In Ohio, Pennsylvania and Florida, the three biggest battleground states, there have been an dramatic upsurge in voter registrations. New registrations have been one-sided. In all three states, growth in new registrations are primarily in urban areas, which are more liberal and Democratic strongholds. New registrations in Democratic areas in Ohio jumped 250% from 2000 versus 25% in Republican areas. New registrations in Florida went up 60% in the strongest Democratic areas versus 12% in the heaviest Republican areas. In Oregon, another battleground state, Democrats outnumber Republican in new registrations from two to one, with 9% voting for the first time.
Even in a Bush stronghold like Oklahoma, in which polls indicate that Bush is favored 61% to 29%, the 60% increase in new registrations for 2004 are evenly distributed between Democrats and Republicans, very much out of proportion to traditional political composition of the state, which is even more Republican than Texas.
Debate Denial
There is somewhat of a denial in the media and the Republican campaign that Bush's performance was devastating. What I have been reading is that there was no knockout punch or that Bush will probably still win (based on the flawed poll numbers). No knockout was needed; there was a self-implosion. I don't think recovery is possible.
What is a knockout punch, anyway? While the Quayle-Betsen debate (You're no Jack Kennedy), the Bush-Dukakis (Cold Liberal) and the Carter-Reagan Debate (Are you better off now than you were four years ago?) are probably considered to have served knockout punches, those debates were much less dramatic and received less publicity than this debate. Sometimes, it takes actual election results for a consensus to emerge and the issue to be clarified.
Futures
Current futures indicate a Bush win, even after the debate debacle. I don't think futures are accurate until perhaps the last few days, though they have consistently had predictive performance in pre-2000 presidential elections, but missed the popular vote in 2000. I think, since unsophisticated individuals are likely involved, the underlying behavior is dictated more by psychological rather than analytical expectations, such as the name recognition/incumbency bias. Their conclusions are probably based on the same flawed polls. It's not as rational as the stock market, where most share are owned by institutional investors and detailed data is gathered and analyzed by experts. In addition, since futures can be traded anytime, bets may be made on expected price swings due to intermediate events rather than the final outcome.
Conclusions
What we have are weak poll numbers, that are already distorted to the president's base, and major demographic elements of the population underrepresented (mobile phone users) or not represented at all (overseas and first-time voters). Even in polls using registered voters, which appear to be shifting to Kerry's direction, the underlying sample pool are still based on households with landline phones.
Based on current polling data, I see Bush losing all the battleground states and all the Gore states. I also see him losing a number of his red states from 2000 like VA, WV, TN, MO, and LA. In the popular vote, Kerry could easily take 55%, this is because the president's poll numbers are hovering in the high 40's, which through the 50% rule offers Kerry a slight majority, and Kerry could gain a few additional percentage points due to distortions in the polls and new registrations.
|