The Federal Review Composite Poll™ is an attempt to answer these questions. It is a way of averaging the results of various polls. That’s why we call it a Composite of all available polls. Once we have the Composite result, we try to figure out what this might mean to the Electoral College. If the Composite Poll is a prediction of the national vote, how does this work on a state-by-state level? We try to answer that, too, in the Federal Review Electoral Vote Projection. Here are the assumptions and methodology of the poll. Feel free to email us with comments.

Composite Poll

• In each week’s composite poll, we use only those polls with a survey period ending during the week immediately preceding the Composite Poll.  But we don’t completely ignore the polls that make up the prior week’s Composite Poll.  The results from the immediately preceding are included in the calculations as though it was a separate poll.  By using the previous week’s Composite Poll result, we expect that swings to the polling results from short term news events or from the inclusion of fewer polls will be mitigated and real trends can be discerned.

• The commonly used polls include Zogby, FoxNews/Opinion Dynamics, CBS/New York Times, NBC/Wall Street Journal, CNN/USAToday/Gallup, ABC /Washington Post and Newsweek.  An additional poll may be added (AP / Ipsos, Pew, for example) may be included, especially during weeks in which few polls are newly updated.  Because Rasmussen Research uses unique automated technology and was the most incorrect poll in 2000 (off by 4.5%), originally were not going to include it, but we are now including it, but at a reduced weighting. We are also using a 7 day average of Rasmussen to eliminate the wild swings in the poll.

• In addition to the polling data, the latest closing price in the Iowa Electronic Market is used as though it were a constituent poll.  This will become more problematic the closer we get to the election, because the Iowa Electronic Market does not predict the final vote, but does indicate who a financial market believes will win.  It’s inclusion helps to bring in the opinions of the many market participants who are betting with real dollars.

• A weighted average formula is applied to the weekly polls.  Greater weight is accorded to polls based on the number of people surveyed.  Thus, polls with a lower margin of error (and a greater number of survey participants) are weighted more heavily than those with a higher margin of error.

• In addition, “likely voter” polls are weighted more heavily than “registered voter” polls, which, in turn, are weighted more heavily than “all adults” polls.

• Where a poll includes Nader as a choice, that poll is used.  For states in which Nader is not a ballot choice, an adjustment will be made in the Electoral Vote analysis once that data is more readily available.

• Thus the basic formula to determine Kerry’s support is as follows, where Vn is the number of people surveyed, Weight is a factor based on whether the poll is likely voters, registered voters or all adults, and KerryPoll is the level of support for Kerry in the applicable poll: 

((Vn*Weight*KerryPoll₁) + (Vn*Weight*KerryPoll₂) + (Vn*Weight*Poll₃) . . . (Vn*Weight*Poll_n) / (AggVn*AggWeight*AggKerryPoll)) = Initial Composite Result

• Once we have applied our formula to determine the initial Composite Result, we then allocate the percentage of those who are undecided.   In the event that the President’s job approval rating exceeds his job disapproval rating by less than 10 points (for instance, if job approval is 54 and disapproval is 46), then the undecided vote is divided based on favorability and unfavorability ratings of Bush and Kerry.  If Bush’s favorable rating is more than 10 points higher than his disapproval rating, then the undecided vote is allocated based on the Initial Composite Result.  Thus, if Bush’s disapproval rating is high relative to his approval rating, the undecided vote is allocated to the candidate who is rated more favorably than unfavorably.  As of April 13, this will result in a greater allocation of undecided voters to Kerry, which is consistent with the conventional wisdom that undecided voters break toward the challenger.

• Once the undecideds are allocated, we call this the Composite Poll Result.

Electoral Vote Projection

• To determine who will win each state based on the Composite Poll Result, we apply the Composite Poll Result to each state’s margin from 2000.  Thus, if Bush won a state by 1% in 2000, but Kerry leads in the Composite Poll Result by 2%, then this analysis predicts Kerry to win that state by 1%.

• To determine the base margin between the candidates for each state, we use the two-party popular vote margin in 2000 as a starting point.  This is the 2000 Base Margin.

• The 2000 Base Margin is then adjusted to allocate the Nader and Buchanan votes.  While it is popular wisdom that Nader votes would have been Gore votes and Buchanan votes would have been Bush votes in 2000, it is likely that Nader and Buchanan voters would have voted for neither Bush nor Gore.  In order to ensure that the Composite Poll and Electoral Projection is not biased toward my choice for President, I have adjusted the 2000 two-party vote margin by adding 65% of the Nader vote to Gore’s total and 65% of Buchanan’s vote to the Bush total.  This results in a baseline national popular vote margin that would have been in Gore’s favor by 0.68%, which is only slightly higher than the actual two-party vote margin.  The difference, however, is more pronounced in states with higher votes for these third-party candidates.  Florida, for instance, would have been won by Gore with a 0.4% margin.  This is the 2000 Adjusted Base Margin.

• The 2000 Adjusted Base Margins for each state are further adjusted based on changes in voter registration since 2000.  I believe it is reasonable to assume that if voters moved from Democrat to Republican since 2000, it is likely that the voter changing his registration supported Bush in 2000, therefore, only 20% of the registration change is factored into the adjustment.  Thus, if Gore’s 2000 Adjusted Base Margin for a state is 0.4%, and there was a net change of 0.72% of registrations in favor of the Republicans, the 2000 Adjusted Base Margin for that state is adjusted 0.14% (0.2 x 0.72%) in favor the Republicans.  Thus Kerry would begin with a margin of 0.25% in that state.  This is the situation with Florida.  (A Pew poll of party identification showed an increase of 6% of Floridians who see themselves as Republicans.  We have used actual voter registration data from each state).

• Once the margin between candidates in 2000 is adjusted for the Nader/Buchanan vote and changes in party affiliation, we have the 2004 Base Margin.

• Generally, the 2004 Base Margin for each state is increased or decreased by the margin revealed by the Composite Poll Result to determine who we project to win the electoral votes of each state. 

• I have not accounted for Maine and Nebraska, which do not apportion their electoral votes in a winner take all fashion.  If Maine goes Republican or Nebraska Democrat – or get close, I will note the possibility.

• But the projected popular vote by state is not based only on the 2004 Base Margin plus the Composite Result.  It will be further adjusted to take into account any recent state poll.  For example, if our analysis says Kerry would win a state by 5.8%, but a recent poll shows him leading in that state by 10%, then our result is adjusted.  In this case, we would report a Kerry lead in that state by 7.9%.  This is the case with Michigan in the April 13 Composite Poll.   We don’t accept the recent poll at face value because it is subject to statistical error and if it doesn’t conform to the national composite result, then it should be mitigated.  For instance, if Kerry is shown with a lead in the national polls, is it reasonable that Bush would lead by as many as 6 in Pennsylvania, a state he lost in 2000? We recognize that the electorate can change, so state-by-state polling will be accorded greater weight as we get closer to the election.

• In any state where the margin of victory is less than 6%, that state is categorized as a toss-up (say, Kerry 52%, Bush 48%).  Lean states are states in which a candidate leads by between 6% and 12%, and Locks are states where a candidate leads by more than 12%.

• The base margin for Tennessee has been increased by 2% in favor of Bush, since no “favorite son” is on the ballot.  Bush 41 won Tennessee in 1988 by 16%.  Clinton/Gore won by 5% in 1992.  Even though the Clinton/Gore national popular vote margin increased by 3% in 1996 over the result in 1992, their margin of victory in Tennessee dropped to 2.4%.  That trend is reflected in the Electoral Vote Projection.

• There’s one more number that the analysis can yield.  Once we have taken into account all of the various state results predicted by the Composite Poll and Electoral Vote Projection, we can recalculate the two-party national popular vote numbers, based on each state’s results.  We’ll call this one the Electoral Vote Popular Vote.

• Punditry:  If I don’t believe the math, I will apply my own judgment as an observer of electoral politics.

A Note About Other Projections

Many of you are aware that with the debut of this week's Composite Poll and Electoral Vote Prediction, we're joining several others in the Electoral Vote prognostication business. Other very good prediction sites include those at Electionprojection.com and Dales' Electoral College Breakdown.

Federal Review didn't join the fray in order to offer an alternative, better projection than those two, just a different one. All three of these projections present valid and useful information in evaluating the race. Electionprojection analyzes head-to-head polls, job approval polls, and right-track/wrong-direction polls to determine where the popular vote might go, and then applies its results to each state's 2000 outcome to determine a potential electoral vote outcome for 2004.

Dales' Electoral College Breakdown focuses primarily on state-by-state polling without regard to the overall national popular vote and then applies analysis to to the state-by-state polls to determine the trends for each state, which are then categorized as tossup, slight advantage, lean, strong advantage and safe. He also includes detailed analysis and historical polling numbers for each state. Quite a burdensome task, but one I am very greatful he has undertaken.

Federal Review's effort grew out of our composite poll from 2000 -- a way of averaging weekly polls and smoothing out margin of error swings. But I realized that the national popular vote I was tracking doesn't tell you who will win, as the 2000 outcome reminded us all. So, I've combined my composite poll result with a state-by-state mathematical analysis that takes into account state-by-state polling. I don't account for the job approval and right-track/wrong-direction numbers in the detail that Scott's Electionprojection does, and I don't provide as detailed an analysis of the state-by-state polling as Dales' Electoral College Breakdown does. Thus, if state-by-state polling is dead on the money, Dales' will be "more right" than me, and if job approval and right-track numbers factor heavily into how people actually vote, then Electionprojection is going to be "more right" on the national popular vote.

Sure, I tried to address what I see as problems in the other two analyses. Dales' analysis relies heavily on polling that is infrequent and applies his own analysis. I think his analysis is objective and fair, but I'd have a hard time believing Bush would lose, say, Tennessee, despite what the numbers told me. That's my bias and why I have the "numbers" make those decisions for me. Although I think it is possible for Bush to win Pennsylvania but lose Florida, this outcome is impossible in Electionprojection's analysis, and there's historical data that supports Electionprojection's result, but not mine.

While I expect some friendly comparison of results between these sites in November, I don't see them as competitive, but as more information for us political junkies.

And if you see all three sites calling things the same way, then you can bank on that result.

I'll also note another site: Dave Leip's Atlas of U.S. Presidential Elections. Dave has a prediction page that allows users to post what they think will happen in the Electoral College. Federal Review's page on Dave Leip's site is here. He then aggregates the predictions and gives you an idea of what people believe will happen. That may be just a useful as any of our projection sites and I urge you to check it out.

Finally, I can't find any comprehensive projection sites run by Democrats, liberals, Kerry-supporters, whatever. Election Projection, the ECB and Federal Review's predictions may have pro-Bush bias, even though we all try like hell to avoid it. It would be interesting to see a detailed analysis from the other side that tries to avoid bias, too. Let me know if you find one.

FEEDBACK: We'd love your feedback. Please let me know if you have any comments on this methodology. Does something not makes sense? Are my assumptions wrong? E-mail me at winstonATfederalreview.com. Of course, replace AT with @.