There was a great deal of talk this year about what Republican operatives claimed would be an extraordinary get-out-the-vote effort in the last 72 hours of this election – an effort that would turn the tide of the election for the GOP. 

In the end the Democrats won control of Congress.  So the Republican tactics obviously at the most basic level did not do what they were supposed to:  create a startling come from behind victory.

But in a technical sense it’s difficult to say that the GOP efforts didn’t have any effect at all. Ad agencies confronted with static sales for a product sometimes argue that sales would have actually been much worse had the advertising program not been in place.  In similar fashion, those in charge for the Republicans may want to claim that the Democrats would have won even more seats in Congress if not for the Republican GOTV campaign.   

Our analysis shows that Republicans still had a modest turnout advantage in this election.  The generic ballot in the House race was slightly more Republican among likely voters than among the larger pool of all registered voters.  But the difference was not as large as we have seen in previous elections. 

I think in part the Iraq war helped neutralize the impact of the more typical GOP turnout advantage.  Opposition to the war most likely moved some Independents into the Democratic voting column.  I think the war also helped provided a strong motivating factor for Democrats, in essence giving them a compelling and emotional reason to vote that might not have been there in previous elections.

Finally, I go back to an important 2004 inteview given by former Bush strategist Matthew Dowd.  Dowd emphasized that the GOP win in the 2002 midterms was perhaps just as much due to the strong pre-existing position of the Republicans (and President Bush) overall in that election as it was to a specific tactic.  Bush’s job approval was in the 60%-70% range going into that election.  It was in the 35%-40% range in this election.  In other words, there is just so much that a get-out-the-vote operation can do.

Good news for those in the pollster profession this year (that includes me, of course).  The pre-election polls in general were quite accurate.  The national generic ballot measures accurately measured the general Democratic orientation in House voting (and USA TODAY/Gallup's historic modeling accurately predicted that the Democrats would have enough strength to gain majority control of the House).   Senate polls as a whole were also very accurate. With one or two exceptions, polls that showed a strong lead for a particular candidate in specific states (e.g., Pennsylvania, Ohio, New Jersey) were borne out by strong victories for those candidates.  Senate polls that showed close, too-close-to-call races were borne out by much closer races in those states.

I was on a radio talk show this morning and the interviewer asked if there were any surprises on Election Day.  I said that there really weren’t.  In general, the picture of the election that we went to bed with on Monday night was the same picture we woke up with on Wednesday morning.

There are a number of methodological challenges faced by survey research today.  As an industry we are constantly attempting to improve and figure out the best scientific procedures for accurately using random samples to estimate the attitudes of a very large population. This election indicates that despite these challenges, the procedures in use today remain effective and continue to give us reliable and valid results.

I caught Time Magazine’s Joe Klein last night on MSNBC saying that at about this time in an election year he gets sick of polls, polls, polls.  I remember President George H.W. Bush in 1992 saying the same thing, complaining about all the polls he was confronted with (almost all showing him losing).  And so on.  This seems to be a relatively normal reaction in an election year. (Of course people are sick of many aspects of the campaign process -- political ads for instance -- by the time Election Day finally arrives).

There is no question that polls are a major part of the pre-election environment. 

But this is not new. George Gallup became famous for conducing  pre-election polls 70 years ago.  Gallup correctly predicted in the 1936 presidential race that Franklin D. Roosevelt was going to be re-elected. His polls generated enormous interest. 

History records many instances of so-called straw polls going back into the 19th century – crude efforts to figure out what voters were thinking before they walked into the ballot box.

There is apparently a human drive to want to know what is happening in an event as important and involving as a national or state election.

I once labeled the process as we know it today as a “continuous election”.  People constantly vote (through polls) in the run-up to the final “real” vote, which is the one that counts.  We monitor the ups and downs of the various candidates as the weeks go by until we finally reach the end when the process stops. 

This is in some ways like a football game.  We follow the score from quarter to quarter, always knowing who is ahead and who is behind, but with the realization that the only score that ultimately will matter is the one at the end of the game. In similar fashion, we follow the progress of candidates and campaigns throughout the fall, understanding that it’s only the final poll at the voting booth which matters.

(We also realize that the score in a football game can change often, and sometimes dramatically.  The same thing happens in campaigns. That’s natural.)

We could of course attempt to run a “blind election” in this country.  That would involve banning all polling and thus leaving us in a situation in which no one in theory would have any idea how well any particular candidate was doing.  A number of European countries in fact have bans on the publication of polls in the last week before an election.

There are a number of reasons why I don’t think this is a good idea (besides the fact that it would very difficult to ban polls on a practical level).   I’m a professional pollster, of course, so my interest in pre-election polls is self-evident.  But more broadly, I think that the people in a democracy should generally be allowed to get the kind of news and information in which they are interested.  People obviously have a great deal of interest in knowing what is going on in election campaigns.  Voters are bombarded with information about the election.  The campaigns dominate the news.  Billions of dollars are spent on advertising and marketing of candidates.  Many Americans are highly emotionally attached to one side or the other of the political process. The consequences of the election can be very significant. 

The campaigns and candidates are themselves spending millions on polling operations.  If we don’t have publicly available polling, the voters themselves are the ones left in the dark.  And that’s not good.   

When all is said and done, the Senate races in four of the six states in which we conducted USA TODAY/Gallup polls this past weekend (Missouri, Tennessee, Virginia, and Rhode Island) were very close – within the margin of error.

What does that mean exactly?   

Here’s one way of looking at it, using Tennessee as an example.  If the population of all voters in Tennessee was truly split down the middle in terms of vote intentions, what would happen when pollsters took a sample from that population?  The sample would certainly not always reflect the true, underlying 50-50 split.  One sample might show a slight lead for the Democratic candidate – Harold Ford, Jr.  Another sample drawn at the same time might show a slight lead for the Republican candidate Bob Corker.  So the three-point lead our latest USA Today/Gallup Poll shows for Corker over Ford does not necessarily mean that if we had been able to interview every voter in Tennessee Corker would have a 3-point lead.  Corker could in reality be even further ahead, the underlying population could be tied, or for that matter Ford could be slightly ahead.

At the same time, when repeated samples are taken from the underlying population and all show the same candidate ahead, that bolsters the case that the candidate is really ahead in the underlying population.  In the case of Tennessee, a number of polls have been conducted over the last week, and each shows Corker in the lead (although leads of differing sizes).  If in truth Ford is really the leader in the underlying population of all Tennessee voters, the chances that multiple, independent samples would all show Corker ahead are small.  So in the current case, the probabilities are higher that had we been able to interview every voter in TN we would find Corker with a lead. (That doesn’t mean that the underlying voter sentiment won’t change before Election Day.) 

But in Virginia, the polls have been mixed, with some showing Jim Webb slightly ahead, others showing George Allen slightly ahead (as ours did), and at least one showing a tie.  This suggests that the underlying population of voters in Virginia is truly split and that the race in that state is truly a toss-up. We really don’t get much of a sense from all of the polls taken in recent days  that one candidate in Virginia is ahead in the underlying population – as of the time period involved in the polls.

When a lead for a candidate stretches to the 10-point range as we found in New Jersey (the Democrat Menendez over the Republican Kean), it does reach the point of statistical significance.  Phrased differently, it would be improbable that a sample of voters showing a 10-point lead for Menendez could have been drawn from an underlying population in which Kean was actually ahead.

The final USA TODAY/Gallup national poll was conducted Thursday through Sunday, Nov. 2-5.  It shows a seven-point Democratic lead over the Republicans, 51% to 44%, on the generic ballot. That’s reduced from the Democratic lead in our previous two polls. 

USA Today and Gallup have delved into these data in some detail.

But here on this Monday morning there are still some interesting questions:

Could the trend (by which the Republicans are gaining on the Democrats in the national generic ballot) continue into Monday and Tuesday, tightening the race still further? 

Answer:  Yes, that’s possible but such a trend is not apparent at this time.  There have certainly been elections in which there appeared to be last minute change as late as the Monday before voting. We did not notice any such trend over the four nights of our interviewing Thursday through Sunday, however.  In other words, it did not appear that there was a continuing trend towards the Republicans on Saturday and Sunday not evidenced in the Thursday and Friday interviewing.  But anything is possible.

Could the vaunted Republican turnout efforts reduce the gap within the likely voter pool still further?

Answer:  Yes, that’s possible.  The Republicans had gained back their more typical turnout advantage in the final Nov. 2-5 poll, with a 7-point disadvantage among likely voters compared to the 11-point disadvantage among all registered voters.  This is a slightly greater difference than in the previous two polls.  The GOP advantage among likely voters has been greater still in other election years.  It’s possible that the Republicans’ 72-Hour Get Out the Vote effort could give Republican House candidates still more of a disproportionate advantage based on turnout.

Could the Democrats actually win the national Congressional vote by 7-points yet still not take the majority of House seats?

Answer:  Yes, that’s possible but not likely.  As we have explained in some detail, estimates of actual seats won and lost based on the national aggregated voter for Congress is developed from historical patterns. (In other words, the best predictor of future behavior is past behavior.)  In theory, a party could get 51% of all votes cast in House elections across the country and win 100% of the seats (if the vote was distributed equally), or get 49% of all votes cast and lose 100% of the seats.  But it usually doesn’t work that way.  The national aggregated vote has been a very good predictor of seats won and lost in previous elections. If the voting this year follows historical patterns, a 7-point lead for the Democrats translates into a majority of the seats for the Democrats.  But more refined redistricting (packing voters of particular political persuasions into specific districts and therefore leaving fewer truly mixed party districts), and more sophisticated get out the vote efforts could mean that historical patterns are no longer as applicable.  We’ll of course have to review that once the election returns are in.

Basically, as the reader of this post has no doubt ascertained, it’s true that a lot of things could in theory happen that would result in the Republicans maintaining their control of the House.  Our analysis is based on the data, the best statistical modeling we have available, and past voter behavior patterns.  And all of that points to the probability that the Democrats will win a majority of House seats.

Here's what we are up to this weekend, which marks the end of the process of conducting our final round of six USA TODAY/Gallup Senate polls. Interviewing in the final two out of the six states in which we are conducting surveys – Tennessee and Missouri – was completed on Saturday, the rest (New Jersey, Rhode Island, Montana, and Virginia) on Friday.   

As interviewing in each state is completed, a complex process unfolds.  The Gallup team headed by Managing Editor Jeff Jones first checks and rechecks the data in a number of ways.  One of these is to compare the demographic composition of each state’s sample to the latest Census figures on region, gender, age, education and race within the state.  If the random sample of the state’s population has significantly different distribution on these variables than Census figures, the sample is adjusted to bring it into line.

The next step is the determination of the pool of likely voters within the state.  As I have explained,  we use a series of questions to identify those voters we believe are most likely to actually turn out and vote in Tuesday’s election. The size of that pool depends on our estimate of the state’s turnout (which varies from state to state). 

We look carefully at the distribution of this likely voter sample across the regions of the state.  In general, it should match the broad pattern of voting in previous elections in that state.  In other words, if the southwest region of Missouri typically represents a certain percent of all votes cast in Missouri in election after election, then our distribution of likely voters in southwest Missouri should reflect roughly the same pattern.  We allow for some differences based on idiosyncratic factors in each election.  But history suggests that turnout is remarkably constant across regions. We adjust each states’s likely voter pool to reflect that if necessary.

Bottom line here:  the overall distribution of the basic, broad sample is reflective of the population of each state, and the regional distribution of the pool of likely voters approximates the regional distribution of votes cast in previous elections.

We then conduct other final checks of the data, including analysis of the raw data anlzyed using different software packages to make sure no glitches have appeared in the processing. 

The data are then crunched in a customized software package which provides a series of compressed cross-tabular tables for each variable.  In Missouri, for example, we can look at each set of responses crossed by the following:  gender, race (white, non-white), age (18-34, 35-54, 55+), region of the state (St. Louis city, St. Louis suburbs, Kansas City area, southwest, north and east), type of community (urban, suburban, rural), education, income, ideology (conservative, moderate, liberal), party identification, when the voters’ mind was made up, (and in Missouri), opinion about the controversial Michael J. Fox ad.

At that point everything is ready to go.

A typical midterm election involves a turnout of only about 40% of the voting age population.  This fact has significant implications for pollsters. Our typical general population sample is designed to represent the attitudes of all adults.  But the attitudes of the much smaller group of adults who actually vote in a midterm election can be significantly different.  That’s where the concept of “likely voters” fits in.

Our final set of Senate polls (now in the field in New Jersey, Missouri, Tennessee, Virginia, Montana, and Rhode Island) includes our standard set of likely voter questions.  We use these questions to assign each respondent a likely voter score.  Someone who gives the most positive answers to the set of likely voter questions gets the highest score.  This would be a person who reports having given a lot of thought to this year’s election, who reports being certain that they are going to vote this year, who has a strong history of voting in previous elections, and who knows where people in their neighborhood vote. Someone who answered more negatively to these questions would get a lower likely voter score.

We then estimate the likely turnout in each state.  Turnout has been historically higher in a state like Montana, and lower in Southern states like Virginia and Tennessee.  The dynamics of the election this year could change that.  We will estimate turnout based on survey indictors within each state’s sample. We then create a pool of likely voters that matches the size of the estimated turnout for each state.  The attitudes and vote intentions of this group is what we report when we talk about likely voters. 

All of this sometimes makes a big difference and sometimes it does not.  In other words, in some situations the attitudes of the group of people who turn out and vote is no different from the group of all registered voters.  In other instances those who vote are significantly different from those who don’t. This can vary from state to state as well from year to year.

So, when we report the results from this latest round of USA TODAY/Gallup Senate polls on Monday, there will be two figures for each race:  the vote estimate among all registered voters, and the vote estimate among those we deem most likely to actually turn out and vote.

Here’s an update on the House race. I mentioned in a previous post the most recent CNN poll (in the field Oct. 27-29), and the Wall Street Journal/NBC News poll (in the field Oct. 28-20).  Both showed double-digit Democratic leads in the generic House ballot.

Two other national polls have been released since then. 

The CBS News/New York Times poll was in the field through Halloween (Oct. 27-Oct 31).  It put the Republicans at 34%, Democrats at 52%.  This is an 18-point Democratic margin, although with a large 14% undecided component.

The Cook/RT Strategies poll (conducted Oct. 26-29) showed an even larger margin:  Republicans at 35%, Democrats at 61% among “most likely voters”.  The same poll had a margin of 39% for Republicans, 52% Democrats among all registered voters.  This type of difference between likely voters and registered voters is an unusual finding. The percent voting Democrat balloons by 9 points when "most likely" voters are considered, while the percentage voting Republican drops 4 points.  (The massive GOP turnout effort waiting in the wings is designed to have the opposite effect:  increasing the Republican vote among those who actually turn out to vote.)

At any rate, neither of these two polls shows signs of a GOP tightening of the House race.  Coupled with other October polling, they show a continuing and substantial Democratic lead in the national vote for Congress.

Could things change?  All of the existing national polling was conducted for the most part prior to the John Kerry “stuck in Iraq” flare-up.  Our latest USA TODAY/Gallup Poll is in the field now, and by the end of the weekend we’ll have an updated view of changes in the House picture – if any.

Our USA TODAY/Gallup Senate Polls are now in the field in six states – Tennessee, Virginia, New Jersey, Montana and Rhode Island.  I’ve written in previous posts about how we draw the samples for these polls. But how do decide what questions to ask? 

Much of the questionnaire is straightforward and based on years of experience.  We start by asking a set of screening questions in order to randomly select a respondent within the household we contact,  to make sure the person we talk to is eighteen years of age and older, and to keep tabs on certain demographic categories being included in the sample. 

Then comes a set of likely voter questions which probe the respondent on his or her past pattern of voting, interest in this election, knowledge of where he or she will vote, and self-reported interest in voting.  I’ll talk more about those in a forthcoming post. But in general we use this information to build a model of those respondents we feel are most likely to actually turn out and vote on Election Day. 

Next are the ballot questions – the heart of the matter.  In this poll we are asking voters for whom they intend to vote  in their state’s U.S. Senate race.  In states where it is appropriate, we ask about the gubernatorial race.  Candidates are identified by party. Respondents who initially say they haven’t made up their minds are asked which way they lean.  Respondents are allowed to volunteer the names of third party candidates for whom they will be voting.   

The survey concludes with a series of demographic and political questions measuring gender, age, education, race and ethnic background, marital status, church attendance, political party identification and ideology.  We use some of the demographic data when we check the final data set against known census parameters and adjust or weight it as necessary.  Others we use primarily for analysis patterns.  (For example, isolating voters who frequently attend church helps us understand the role of religion in the election).   

In this set of state polls, we decided not to ask many additional questions of respondents. One of the most important reasons for that is the tight time frame under which we are operating.  We are fielding six state polls simultaneously, bringing them out of the field, analyzing and processing the data, and getting them ready to publish within a compressed time frame.  Plus, given that Election Day is Tuesday, there is not a lot of room either in USA Today or on Gallup’s website to do heavy analysis and reporting before the voting actually begins. 

We did decide to ask voters in each state when they made their final vote decision.  And, given the huge amount of money spent on television commercials this year, we asked voters in each state to rate the tone of commercials they have seen in their state.  In two states we are asking about the impact of specific commercials:  the Michael J. Fox spot concerning stem cell research in Missouri, and the infamous “Playboy” commercial in Tennessee.

President Bush gave a vote of confidence Wednesday to two of his most high profile – and controversial -- associates:  Vice President Dick Cheney and Secretary of Defense Donald Rumsfeld.    Bush was quoted as saying that “…both those men are doing fantastic jobs, and I strongly support them," and went on to reaffirm that he wanted both to remain in office until the end of his term.

This gesture was widely interpreted as part of the Karl Rove/Ken Mehlman strategy to activate the conservative Republican base in this last week of the election campaign.

Our USA Today/Gallup Poll data would seem to confirm the attitudinal underpinnings of the strategy.

Both Cheney and Rumseld are overall not very popular.  Our last measure of Cheney’s favorable rating puts him at 34% positive, 62% negative.  Rumsfeld comes in at 41% favorable, 45% unfavorable.

But look at what a difference party identification makes!  Among Republicans, Cheney’s favorable rating is an overwhelming 75% positive.  Among Republicans, Rumsfeld’s is 69% positive.  The idea of having President Bush publicly reaffirm his support for the two would seem to make sense in an environment in which the major focus is reaching out to base Republicans.  They like both men.

By the way, there has been some effort on the part of Republicans to use House Democratic leader Nancy Pelosi as a symbol of what a liberal Democratic Congress would look like if the GOP loses control next week.  The presumed assumption here is that Pelosi has a negative image among base Republicans. 

That’s true as far as it goes.  Pelosi is relatively unknown.  A USA Today/Gallup Poll conducted earlier in October showed that only 54% of Americans knew enough about her to give up a rating (which was mixed:  26% favorable and 28% unfavorable). 

Among Republicans, her favorable rating was only 6% and her unfavorable rating was 54%.   Forty-six percent of Republicans didn’t know her well enough to have an opinion.