11-2 Election Day
My morning estimate of Kerry 50.3 is based largely on a single (Marist) poll begun and completed yesterday, so it does not have the virtue of averaging which is the essence of this approach. The previous day estimate Bush 50.7 should be more accurate, and it is awkward that the two estimates differ by a full point and predict different winners. What we seem to know for sure is that it will be close.

The final electoral college estimate, Bush 278 to Kerry 260 reflects that final day poll and the decision rule that doing less well nationally than Gore 2000 (50.3 compared to 50.4!!!) assigns all the red states, including Florida to Bush. This is obviously a knife edge decision. This is exactly the 2000 result except that the red states have gained 7 electoral votes from the blue states.


11-1 Election eve
It's 50-50, but the final two days are driven by only two pols (both with kerry leading). On the last day for which there are multiple sources it is Bush 50.4 and Kerry 49.6.

Today's trends are a wash, with Fox, Marist, CBSNYT trending toward Kerry, and ABC/WP, Zogby, and Rasmussen moving toward Bush.  All movements are small.

My electoral college projection stills signals Bush because any Kerry performance less that Gore in 2000 necessarily produces that result. At this point the abundence of state polls is better evidence than a national projection.  From the national number Bush would win FL, OH, NH, WI, NM, IA, and OR.


10-31 10:30AM

Electoral College:
My electoral college estimate is consistently different than others and consistently points to a big Bush edge.  That is for two reasons: (1) It is a national estimate, entirely driven by (a) the current national estimate (actually the filtered one) and (b) all 538 votes are allocated to the two candidates with no "toss-up" category. On (1) any national number smaller that Gore's  2000 standing (50.4) predicts a Democratic loss, and Kerry has always been below that number since the Republican convention. I make no allowance that the race might be different in the battleground states, as it appears now to be, than nationally.  On (2) an estimate of 50.01 for either candidate puts that state in his column, where realism would predict about a 50-50 shot instead. There are about 120 electoral votes that switch from side to side for a movement of about 1 percent in the national estimate. Currently my estimate assigns virtually all of the battlground to Bush (OH, FL, WI, OR, IA), where state polls suggest toss-ups in OH and FL and Kerry leads in WI and OR.

For the final day estimate, I will use the daily estimate and not the filtered one, because the latter lags behind trends.

------------------------------------------------------------------------------------------------------------------------------------------------------------------

As a reality check and because I worry that my estimate heavily overweights the Rasmussen numbers relative to others I present here simple averages for the last week based, as always for me, on the first day the study was in the field.

Day   Date       Bush      Kerry     Bush lead
Sat     10-23     47.4      45.9       1.5
Sun     10-24     47.3      46.4       0.9
Mon    10-25     47.9      46.0      1.9
Tue      10-26    46.8      46.4      0.4
Wed    10-27    48.5      45.4       3.1
Thu      10-28    47.3      47.0       0.3
Fri        10-29    45.8      47.0     -1.2 (based on Democracy Corps(D) and Fox polls only)

These leads are smaller than those I have estimated, except for one case where I had Kerry ahead, and the cause is overweighting Rasmussen. They are also substantially smaller than reported averages on the web, the reason being that virtually all such reports are based on likely voter samples. These estimates include registered as well as likely samples. For a year in which turnout is a wild card, I have little confidence in likely voter screens, however well intentioned.

For my final day estimate I will report two numbers, both with and without including Rasmussen.

------------------------------------------------------------------------------------------------------------------------------------------------------------------

10:30 5:50PM
Tracking Polls
Zogby:  Kerry gains 1.1 and takes small lead (0.3)
TIPP:   Bush gains 1.5 and moves from tie to lead
Rasmussen: Kerry gains 1.2 and trails by 0.8
Fox: Kerry moves from -5 to -2 among likely voters, remains tied among registered voters.
WP/ABC: Kerry moves from -3 to -1 among likely voters (RV unavailable)

The Osama bin Laden tape:
Kerry gains in 4 of 5 tracking polls in the field after the announcement (Zogby, Rasmussen, Fox, and WP/ABC). It probably indicates no effect of the news, but is consistent also with a move toward Kerry..

Other national polls
Newsweek:  Bush leads by 6 (LV) or 3 (RV)

10-29 Friday
Latest day results are driven entirely by an early release Fox poll showing anywhere between tie and Bush +5, depending upon sample typle and question.

Democracy Corps (D): Bush  46 Kerry 49
Zogby: Bush  46.2  Kerry 45.4 (47-47 with leaners)
Rasmussen: Bush 48.7 Kerry 46.7
TIPP  tie
WP/ABC  Bush +3 (LV) or +2 (RV)


10-27
Zogby moves from Bush +2.6 to Bush +0.5.
Rasmussen moves from tie to Bush +1.7, effectively back to where it was before the weekend (false?) trend toward Kerry.
TIPP moves from Bush +6 to Bush +4
WP/ABC moves from Kerry +2 to Kerry +1

10-26 (PM)
Rasmussen moved back sharply from Kerry +2 to tie.  A two point movement is normal for smaller samples, but atypical of the N=3,000 Rasmussed 3 day samplings. The weekend movement looks more like sampling fluctuation than trend. Since the apparent trend appeared in WP/ABC data also, that yet-to-be-released number (5:00PM EDT) will help resolve the issue.

 (AM) The day begins with Zogby reporting no change (Bush +2.6). The question of trend toward Kerry over the weekend is unresolved.

10-25 (PM)
WP/ABC released 10/25, covering 10/21-10/24 picks up a trend toward Kerry.  It shows Kerry +1 (LV) or +2 (RV) after reporting Bush leads of 6 and 7 during the last few days.

TIPP is moving toward Bush at the same time. Because the TIPP and Zogby tracking polls are not positively correlated with the latent series estimate, they contribute nothing to it.  Both are very short in a period of generally no movement and so the lack of correlation is not particularly meaningful.

Gallup is reporting Bush 51 to Kerry 46 (Likely) and 49 to 47 (registered). This both appears to establish a significant Bush lead among likely voters and shows a movement toward Kerry over the larger lead reported a week ago. The earlier survey had a Bush lead of 8 among likely voters and 3 among registered.

The afternoon estimate puts Kerry ahead for the first time since August.  This is driven by a strong two day movement in the Rasmussen tracking poll. Some other polls fielded at the same time suggest a trend toward Kerry (e.g., Washington Post/ABC, Gallup), but others (Zogby) do not. So the Rasmussen gain might be sampling fluctuation. What is more solid, however, is that Kerry has closed the gap to the point where one day fluctuations can put him in the lead.

10-24
(late) Today all current releases moved toward Kerry.  That is Zogby, Rasmussen, and Washington Post tracking polls.  The TIPP poll, however, moved toward Bush, but is a day older than the others.

(PM) Like the downward move two days ago, the upward movement in this afternoon's estimate is largely driven by one result, a Rasmussen poll showing a tightening of the Bush lead to 0.4.  Things are edging closer and the wild volatility of earlier polls seems to have eased.  Most recent releases show a small Bush lead or a (nonsignificant) Kerry advanatge.

10-22
The big move downward this afternoon is entirely a function of a relatively small move in the Rasmussen numbers, to which this estimate is overly sensitive.


10-21
The Harris organization newest results provide some perspective on likely voters screens (and some rare candor about procedures.)  Harris employed two different likely voter screens.  In the first, they simply declared that voters who asserted that they were registered and "absolutely certain" to vote were likely voters.  These respondents  produce a two point (48-46) Bush lead.  Then Harris also tried its traditional screen, which drops from this group anybody who was eligible in 2000 and did not vote.  That produces an 8 point Bush lead!  The truth probably lies in between.  Since turnout is likely to be higher in 2004 than in 2000, the traditional screen probably underestimates liklies. But too respondents do fib a bit about vote likelihood.



10-20
My estimate slides back and forth, within the day, between a tie (morning) and a clear Bush lead (afternoon), based on reporting schedules.  The latter uses more data and is therefore better.

Today's numbers:
Fox  Bush +5 (without Nader)
Fox  Bush +7 (with)
Democracy Corps (a Democratic firm) Kerry +3
Zogby Bush + 0.5
Rasmussen Bush +1.4

On Real Movement in the Campaign:
I am of the view that movement is hugely exaggerated in media treatments.  Looking at all the polls and requiring real movement to show up in most,
(1)  My off-the-top estimate is that Swift Boat campaign and Democratic convention netted out to zero.  Since they were going on at the same time, we can't tease out intervention effects.  My guess, from analysis of previous convention bounces, is that Swift Boat cost Kerry 4-5 points and the convention gained about the same amount.
(2) Bush gained 4-5 points from the GOP convention.
(3) Kerry gained back 2-3 points from the three debates, most movement coming after the first, leaving
(4) Bush ahead 2-3 points going into the final two weeks.

Most other claims for trends, coming almost daily, are simply capitalizing on chance variation in the very small samples that characterizes current horse-race polling.  As I have written, "we can't help ourselves."

10-18
Back to normal update schedule.

Both Zogby and Rasmussen agree for once today, both calling a tie.  ABC/WP will be the tie breaker.

Gallup, as usual, is all over the place, now reporting an 8 point Bush lead.  I've come to the conclusion that its clients, USA Today and CNN, prefer fantasy data to hype the news over good estimates. Stay tuned.  It will probably wobble back and forth some 20 points before election day, at which point Gallup will bet its reputation on a realistic final estimate. That one I will believe.

Jim Stimson