In the New York Times today, Edmund Andrews mentions that the firm Global Insight is predicting the GOP will only get 46 percent of the presidential vote:

“From a political point of view, the job market is not going to provide a lot of comfort at election time, especially for Republicans,” said Nariman Behravesh, chief economist at Global Insight.

Drawing on a model that correlates past economic performance and past election results, Global Insight is predicting that the Republican candidate for president in November will receive only 46 percent of the vote.

Such models have often been wrong before. But Republicans are keenly aware that Mr. Bush received little credit from voters for the last five years of good economic growth.

If Global Insight is predicting Republicans will get 46 percent of the *two-party* vote (as these models typically do), the result they are projecting would be comparable to Bill Clinton's win over Bob Dole in 1996.

However, Democrats shouldn't get so cocky. Lane Kenworthy, a professor of sociology and political science at the University of Arizona, notes that one of the least ad hoc presidential forecasting models -- the "bread and peace" model of Douglas Hibbs, which combines war deaths with weighted growth in per capita real disposable personal income -- actually projects a close election narrowly won by Republicans based on Kenworthy's calculation of the "bread" figure at the end of 2007.

Here is its track record to date (the regression line excludes the outliers of 1952 and 1968):

Here are three 2008 scenarios Kenworthy plots (2008a is based on income through the end of the year, 2008b assumes the GOP is punished for deaths in Iraq, 2008c assumes a continuing economic slowdown through November) -- as you can see, they are each extremely close to the magic 50% line:

The Yale economist Ray Fair also has a well-known election forecasting model. In his latest update, he projects a GOP two-party vote share of 48 percent. However, Fair writes that "it should be noted that the forecasts of inflation from [his econometric model] are more pessimistic than the current consensus view and the forecasts of real output growth are more optimistic." (By contrast, if you project -2% economic growth, his model predicts a GOP presidential vote share of 45.4 percent.)

Of course, as Kenworthy notes, we shouldn't take any of these forecasts too seriously. A model is a representation of how things worked in the past; there's no guarantee of future performance, especially in an election in which no incumbent president or vice president is on the ticket. Also, each model has specific limitations that are worth keeping in mind. For instance, in terms of Hibbs's model, Kenworthy correctly points out that growing income inequality means that per capita income has become an increasingly unrepresentative measure. (The main critique of Fair's model is that the specification has been frequently modified for ad hoc reasons to maximize fit, as Hibbs discusses at some length in an unpublished paper.)

Still, the Hibbs projection in particular is a reminder that Democrats shouldn't get cocky, especially if John McCain consolidates his hold on the GOP nomination tomorrow night. This election may be closer than anyone expects...

I've just posted an analysis of the implications of my Bread and Peace model of US presidential elections at:

www.douglas-hibbs.com/Election2008/2008 Election.mht

Douglas Hibbs

Posted by: Douglas Hibbs | March 12, 2008 at 12:58 PM

2008 Election Model

Monte Carlo Electoral Vote Simulation

TruthIsAll

http://www.geocities.com/electionmodel/2008ElectionModel.htm

The Election Model utilizes the latest state and national polls to determine the winner of the True Vote.

Of course, it assumes that the election is held on the date of the most recent polling update.

In projecting the True vote, the implicit assumption is that the election will be fraud-free.

But the recorded vote is never equal to the True Vote.

The evidence is overwhelming: millions of mostly Democratic votes are uncounted in every election.

Therefore, the Election Model cannot accurately project the official, recorded vote.

If it did, then the polling data was wrong and/or the projection assumptions were invalid.

In 2000, 110.8 million votes were cast, but only 105.4 million recorded,

In 2000, 125.7 million votes were cast, but only 122.3 million recorded.

Why should 2008 be any different?

It is a certainty that millions of votes will be uncounted.

It is also a certainty that more than one million Democratic voters will be disenfranchised.

Can we expect that DRE touch screens, without a verifiable vote count, will not be rigged?

Can we expect that central vote tabulating software will not be tampered with?

Obama’s True Vote (T) will be reduced by uncounted (U), switched votes (S).

The recorded vote formula is: R = T - U - S (not including the disenfranchised)

The bottom line is that Obama will need a landslide to overcome the multiple levels of fraud.

The Election Model does not use historic econometric time-series data.

Interest rates, employment rates, commodity prices, consumer confidence are important factors.

But don’t expect political pundits to quantify the fraud factor in their regression models.

The 2004 Election Model: Confirmed by the Exit Polls

The pre-election aggregate State polls confirmed the pre-election National Polls.

The pre-election polls matched the unadjusted State exit polls and the 12:22am National Exit Poll.

Bush won the recorded vote with 50.7% and had 286 EV.

Approximately 3.4 million votes were uncounted.

The Monte Carlo electoral vote simulation indicated that Kerry would win 51.0% with 337 EV (51.8% of the 2-party vote).

The National projection model indicated that he would win 50.9% (51.6% of the 2-party vote).

The aggregate pre-election state polls matched the national polls.

Kerry led the aggregate final state pre-election polls by 47.7 - 47.0%.

The aggregate state polling average closely matched the national 18-poll average (47.3-46.9%).

Unadjusted state exit polls indicated that Kerry won by 51.8-47.2%.

The Election Model State projection matched the unadjusted exit poll aggregate to within 0.8%.

The 12:22am National Exit Poll update of 13047 respondents indicated that Kerry won by 50.8-48.2%.

The NEP margin of error was 1.12%, assuming a 30% exit poll “cluster” effect.

The Election Model projection matched the NEP to within 0.1%.

Professional pollsters allocate undecided voters to the challenger, especially if the incumbent is unpopular.

Bush had a 48% approval rating on Election Day.

The Gallup Poll allocated 90% of undecided voters to Kerry.

Harris and Zogby gave him 67-80%.

The Election Model assumed that Kerry would capture 75% of undecided voters.

A few naysayers still argue that polling analysis cannot prove that the 2004 election was stolen.

But a careful analysis of the pre-election and exit polls provides powerful evidence that it was.

They claim the early exit polls were wrong and that the Final National Exit Poll was correct.

But the Final was forced to match the recorded vote.

And it was proven to be mathematically impossible.

It assumed that 4 million more Bush 2000 voters turned out to vote in 2004 than had actually voted in 2000.

They also argue that the pre-election polls favored Bush; the data is provided in the tables below.

The unweighted state poll average favored Bush, but Kerry led in the aggregate weighted average.

They claim that Bush was leading the national polls, but the data indicates a virtual tie.

To believe that Bush won, you must believe that all pre-election and unadjusted exit polls were wrong.

And that only the Final Exit Poll, which was forced to match the recorded vote, was correct.

Election Model Methodology

It actually contains two independent models:

a) Monte Carlo Electoral Vote Simulation - based on the pre-election state polls.

b) National average model - based on the latest national polls.

In the state model, the average weighted poll share is calculated. The vote shares are projected by adjusting the polls for the allocation of the undecided voters. In the Monte Carlo simulation, 5000 election trials are executed to calculate the expected electoral vote and win probability.

A powerful feature is the built-in sensitivity analysis. Five scenarios of undecided voter allocation are executed to project state and national vote shares, electoral votes and win probability.

Political scientists generally use one of three methods to project election outcomes.

The first method analyzes historical economic data: growth, jobs, inflation, etc. using regression analysis models to predict the popular vote. Unfortunately, these models lack precision; they often use limited, outdated time series data. The data doesn’t reflect current news and information which affect voter psychology and preference.

The second method tracks national polls to project the winner of the popular vote.

The third method tracks the latest state polls in order to project the popular and electoral vote.

The Election Model uses current state and national polls. The only projection assumption is in the allocation of undecided/other voters. Historically, undecided voters have split at least 2-1 for the challenger. So if a poll has the race tied at 45-45, then allocating a 60-40% split of the undecided 10% derives a 51-49% projected vote share.

The winner of the popular vote will almost certainly win the electoral vote. But that might not be the case if the winning margin is less than 0.5%.

Polling Mathematics

An advantage of national polling is its relative simplicity. If the polling spread exceeds the margin of error (3% for a typical 1000 sample) then the leader has at least a 97.5% chance of winning the election assuming a) it was held that day and b) the poll is an unbiased sample. But that’s just the probability for a single poll.

If three independent national polls are done on the same day, that is approximately the equivalent a single poll of 3000 sample size, and the theoretical MoE is 1.8%. Assuming that the average split is 52-48%, there is a 95% probability that the leader will receive between 50.2% and 53.8%. But since there is a 2.5% probability that his vote share will exceed 53.8%, there is a 97.5% probability of winning at least 50.2%.

The MoE is 1.96 times the standard deviation, which is a statistical measure of the variability of polling observations. The standard deviation and projected vote share is input to the normal distribution function in order to determine the probability of winning a vote majority.

To calculate the expected EV from state polling data, the final vote is projected. State polls typically sample 600 voters, so the state MoE is 4% compared to 3% in National polls. The probability of winning each state is calculated based on the projected state polls.

For example, in the case of a 50-50 projection, each candidate has a 50% probability of winning the state. For a 51-49 split, the leader has a 69% chance of winning; for 52-48, the probability is 83%; for 53-47, 93%; for 54-46, it’s 97.5%.

Monte Carlo Simulation

In the simulation, 5000 election trials are run to determine the probability of winning 270 Electoral Votes. The probability is the number of election trial wins divided by 5000.

In each trial, a random number (RND) is between 0 and 1 is generated for each state. The RND determines who wins the state. For example, assume the poll leader has a .60 probability of winning the state. If the RND is less than 0.60, he wins the state’s EV; otherwise, it goes to the other candidate. The process is repeated in 5000 trials. The total number of trial wins is calculated for each candidate.

The probability of winning the electoral vote is the total number of trial wins divided by 5000.

The expected (mean) electoral vote is just the average EV.

An advantage of the simulation method is that the effects of minor shifts in individual state polls are minimized since the expected EV is the average of 5000 simulations and not just a single snapshot.

Using independent national and state polling models provides a mathematical confirmation of each method. This reduces the margin of error, so that we have more confidence in the results.

In summary, the Election Model projects the latest national and state polls after adjusting for the allocation of undecided voters. The probability of winning each state is calculated. A Monte Carlo simulation of 5000 election trials is then executed (using the individual state probabilities) to determine the expected final electoral vote and win probability.

Posted by: TruthIsAll | June 06, 2008 at 10:00 AM