Kerry Will Win. Part III. And It Could Be Big.
I apologize for my absence, dear readers, for the
last several days. Last week I succumbed to a very bad head cold, the worst
I've suffered in many, many years. Only now am I digging out from under it, and
even as I write the after-effects still linger.
You would never guess from the mainstream media, but the trends in this election are moving decisively against President Bush. Bush remains consistently under 50 percent in the vast majority of published polls. Because undecided voters break against incumbents, Bush's final number is likely to be the same number he is getting currently in the polls -- 46 or 47 percent (maybe 48 if he catches a break or two).
The math is startlingly simple. For example, take a poll which has Bush and Kerry tied at 47. Throw in 2 percent for "other" which leaves 4 percent undecided. Apply the incumbent rule, and predict that the undecided voters will break 3 to 1 against the incumbent (I've read in some places that the anti-incumbent break is as great as 4 or even 5 to 1, but we'll try to be conservative here).
Result: Kerry 50, Bush 48.
We've heard about it all over the internet - incumbents under 50 percent are considered vulnerable, and this is the reason why. The farther an incumbent is under 50 percent, the harder it is to pick up enough undecideds to get over the 50 percent mark. There are plenty of cases where an incumbent under 50 percent but with a clear lead in the polls still lost on election day (for example, Paul Wellstone's defeat of Rudy Boschwitz in 1990).
An incumbent tied at 47 with a challenger going into election day is almost certain to lose.
John Kerry is in a much stronger position to win today than Al Gore was four years ago. At this point in 2000, Bush had a clear lead in the polls of 4-6 points (IIRC, something like 46-42). There was no incumbent in that race, so undecided voters had no fixed impression of one candidate, and could have easily broken for Bush. As it was, undecideds broke for Gore in 2000, giving him a popular vote victory.
A net four point swing to Gore in the last week of the 2000 race turned a 46-42 deficit into a 48.4 - 47.9 popular vote win. A four point swing to Kerry in 2004 (which is far more likely because of the incumbent rule) turns a 47-47 tie into a smashing 51-47 victory.
Applying the incumbent rule to the state-by-state polls really illustrates Bush's weak position going into Election Day. Most of the sites that track state level polls do not apply the incumbent rule in assigning undecided voters. That's why a site like electoral-vote.com shows Bush ahead today with 285 electoral votes -- because, for example, it assigns Ohio (polling 46-45 for Bush), Iowa (polling 47-46 Bush) and Florida (48-47 Bush) to Bush when in point of fact Kerry will win all three states if undecided voters break in his direction.
Last week I took all the state-by-state polls and applied the incumbent rule to assign undecided voters, to see which candidate was likely to win.
My result: Kerry 311 electoral votes, Bush 227.
I am not alone in this conclusion. Sam Wang, a professor at Princeton, has been tracking state-by-state polls for several weeks and running statistical analysis to see who will win the election. In the last couple of weeks, he has split his analysis into two categories: who's ahead among decided voters only (excluding undecideds), and who's ahead if you assign undecided voters according to the incumbent rule.
As of this morning, if undecided voters are NOT counted, Professor Wang estimates that Bush would win the election 279 electoral votes to 259 for Kerry.
However, allocating undecided voters 3:1 against the incumbent, Professor Wang estimates Kerry would win 307 electoral votes to Bush's 231.
Andrea Moro, from the University of Minnesota, has been running statistical simulations of the election **without** applying the incumbent rule. As of today's state-by-state polls, summarized at 2.004k.com, John Kerry wins 76 percent of 100,000 simulated elections, and George Bush only win 22 percent (2 percent of the elections end in a 269-269 Electoral College tie).
All of this is great news for Kerry, but it is based on a large assumption: that the polls are accurate.
In order to generate meaningful numbers from the raw data, polls have to make some assumptions about how many people are likely to vote. Different polls have different methodologies to determine who is and isn't a likely voter, but one of two things are true about these methodologies: (1) either they are based on past voting behavior, or (2) they have a track record of accurately predicting likely voters in previous elections.
What is this election is different from past elections? Suppose that a substantial number of people who have not traditionally shown up to vote decide for various reasons to vote this year? Would the existing polls account for this phenomena? Mostly the answer is no. The polls assume that this election will be very similiar to previous elections, and apply their mathematical formulas accordingly. For example, the partisan breakdown of the 2000 election according to exit polls was 39% Democratic voters, 35% Republican voters, and 26% independent. Zogby, for example, applies this ratio when he normalizes his raw data; and I believe the Rasmussen tracking poll does as well.
Suppose, however, that Democrats are far more motivated to vote this year than they were in 2000. Wouldn't a poll based on 2000 turnout statistics underestimate the likely vote for the Democratic candidate in 2004? Yep, it would.
In fact, relative partisan turnout is the true wild card of this election cycle. There are good reasons to believe that Democratic voters are highly motivated this year, especially compared to 2000. And, if new voters and youth voters tend to favor John Kerry (as seems to be the case), if more of them show up in 2004 then polls based on 2000 turnout statistics will underestimate Kerry's likely support.
Let's apply some turnout factors to the state-by-state polls and see what it does for John Kerry.
Kerry's current lead: 311-227 (Kerry wins Ohio and Florida).
Increase Kerry's margin by 1 percentage point: Kerry picks up West Virginia (5 EV) and Arkansas (6 EV), for a 322-216 lead.
Increase Kerry's margin by 2 percentage points: Kerry picks up North Carolina (15 EV) and Virginia (13 EV) for a 350-188 lead.
Increase Kerry's margin by 3 percentage points: no change
Increase Kerry's margin by 4 percentage points: Kerry picks up Arizona (10 EV) and Colorado (9 EV) for a 369-169 lead.
This analysis is based on state-by-state polls that are 4-5 days old.
Before I did this analysis, I would have guessed a 2-3 point "high turnout" bump for Kerry. That results in a 350-188 Electoral College victory for Kerry.
That's just too high, I think. Let's keep Virginia and North Carolina off Kerry's board, and I'll stick with a 322-216 Kerry victory as my prediction.
edited 10/26/2004 5.34 pm CDT -- edited to fix bad math
You would never guess from the mainstream media, but the trends in this election are moving decisively against President Bush. Bush remains consistently under 50 percent in the vast majority of published polls. Because undecided voters break against incumbents, Bush's final number is likely to be the same number he is getting currently in the polls -- 46 or 47 percent (maybe 48 if he catches a break or two).
The math is startlingly simple. For example, take a poll which has Bush and Kerry tied at 47. Throw in 2 percent for "other" which leaves 4 percent undecided. Apply the incumbent rule, and predict that the undecided voters will break 3 to 1 against the incumbent (I've read in some places that the anti-incumbent break is as great as 4 or even 5 to 1, but we'll try to be conservative here).
Result: Kerry 50, Bush 48.
We've heard about it all over the internet - incumbents under 50 percent are considered vulnerable, and this is the reason why. The farther an incumbent is under 50 percent, the harder it is to pick up enough undecideds to get over the 50 percent mark. There are plenty of cases where an incumbent under 50 percent but with a clear lead in the polls still lost on election day (for example, Paul Wellstone's defeat of Rudy Boschwitz in 1990).
An incumbent tied at 47 with a challenger going into election day is almost certain to lose.
John Kerry is in a much stronger position to win today than Al Gore was four years ago. At this point in 2000, Bush had a clear lead in the polls of 4-6 points (IIRC, something like 46-42). There was no incumbent in that race, so undecided voters had no fixed impression of one candidate, and could have easily broken for Bush. As it was, undecideds broke for Gore in 2000, giving him a popular vote victory.
A net four point swing to Gore in the last week of the 2000 race turned a 46-42 deficit into a 48.4 - 47.9 popular vote win. A four point swing to Kerry in 2004 (which is far more likely because of the incumbent rule) turns a 47-47 tie into a smashing 51-47 victory.
Applying the incumbent rule to the state-by-state polls really illustrates Bush's weak position going into Election Day. Most of the sites that track state level polls do not apply the incumbent rule in assigning undecided voters. That's why a site like electoral-vote.com shows Bush ahead today with 285 electoral votes -- because, for example, it assigns Ohio (polling 46-45 for Bush), Iowa (polling 47-46 Bush) and Florida (48-47 Bush) to Bush when in point of fact Kerry will win all three states if undecided voters break in his direction.
Last week I took all the state-by-state polls and applied the incumbent rule to assign undecided voters, to see which candidate was likely to win.
My result: Kerry 311 electoral votes, Bush 227.
I am not alone in this conclusion. Sam Wang, a professor at Princeton, has been tracking state-by-state polls for several weeks and running statistical analysis to see who will win the election. In the last couple of weeks, he has split his analysis into two categories: who's ahead among decided voters only (excluding undecideds), and who's ahead if you assign undecided voters according to the incumbent rule.
As of this morning, if undecided voters are NOT counted, Professor Wang estimates that Bush would win the election 279 electoral votes to 259 for Kerry.
However, allocating undecided voters 3:1 against the incumbent, Professor Wang estimates Kerry would win 307 electoral votes to Bush's 231.
Andrea Moro, from the University of Minnesota, has been running statistical simulations of the election **without** applying the incumbent rule. As of today's state-by-state polls, summarized at 2.004k.com, John Kerry wins 76 percent of 100,000 simulated elections, and George Bush only win 22 percent (2 percent of the elections end in a 269-269 Electoral College tie).
All of this is great news for Kerry, but it is based on a large assumption: that the polls are accurate.
In order to generate meaningful numbers from the raw data, polls have to make some assumptions about how many people are likely to vote. Different polls have different methodologies to determine who is and isn't a likely voter, but one of two things are true about these methodologies: (1) either they are based on past voting behavior, or (2) they have a track record of accurately predicting likely voters in previous elections.
What is this election is different from past elections? Suppose that a substantial number of people who have not traditionally shown up to vote decide for various reasons to vote this year? Would the existing polls account for this phenomena? Mostly the answer is no. The polls assume that this election will be very similiar to previous elections, and apply their mathematical formulas accordingly. For example, the partisan breakdown of the 2000 election according to exit polls was 39% Democratic voters, 35% Republican voters, and 26% independent. Zogby, for example, applies this ratio when he normalizes his raw data; and I believe the Rasmussen tracking poll does as well.
Suppose, however, that Democrats are far more motivated to vote this year than they were in 2000. Wouldn't a poll based on 2000 turnout statistics underestimate the likely vote for the Democratic candidate in 2004? Yep, it would.
In fact, relative partisan turnout is the true wild card of this election cycle. There are good reasons to believe that Democratic voters are highly motivated this year, especially compared to 2000. And, if new voters and youth voters tend to favor John Kerry (as seems to be the case), if more of them show up in 2004 then polls based on 2000 turnout statistics will underestimate Kerry's likely support.
Let's apply some turnout factors to the state-by-state polls and see what it does for John Kerry.
Kerry's current lead: 311-227 (Kerry wins Ohio and Florida).
Increase Kerry's margin by 1 percentage point: Kerry picks up West Virginia (5 EV) and Arkansas (6 EV), for a 322-216 lead.
Increase Kerry's margin by 2 percentage points: Kerry picks up North Carolina (15 EV) and Virginia (13 EV) for a 350-188 lead.
Increase Kerry's margin by 3 percentage points: no change
Increase Kerry's margin by 4 percentage points: Kerry picks up Arizona (10 EV) and Colorado (9 EV) for a 369-169 lead.
This analysis is based on state-by-state polls that are 4-5 days old.
Before I did this analysis, I would have guessed a 2-3 point "high turnout" bump for Kerry. That results in a 350-188 Electoral College victory for Kerry.
That's just too high, I think. Let's keep Virginia and North Carolina off Kerry's board, and I'll stick with a 322-216 Kerry victory as my prediction.
edited 10/26/2004 5.34 pm CDT -- edited to fix bad math
Posted: Tue - October 26, 2004 at 11:02 AM
Quick Links
Who In the Hell is
Greg Abbott?
Categories
Feedback - Send Me E-mail
greg (at) gregabbott.org
Calendar
| Sun | Mon | Tue | Wed | Thu | Fri | Sat |
Visitors
XML/RSS Feed
Archives
Interesting Reading
Entry Content
Comments powered by
Statistics
Total entries in this blog:
Total entries in this category:
Published On: Jan 20, 2005 11:32 AM
Total entries in this category:
Published On: Jan 20, 2005 11:32 AM