October 14, 2008
Understanding Current Presidential Race PollsBy Elwin Tobing
Almost all, if not all, current presidential race polls are pointing to a comfortable lead for Obama, varying from 11% points (Gallup) to 5% points (Zogby). Some perhaps wonder whether the poll numbers really capture the reality. Let us examine Gallup daily tracking polls, a widely respected pollster.
The Basis: Party ID
The biggest issue in this presidential race polling is to figure out the party ID. Put simply, if you walk on the street, what is the chance that you will meet a Democrat, a Republican, or an independent? This party ID is used to weight the responses of the respondents in order to obtain the aggregate result in the daily tracking polls for Gallup or other pollsters. Putting too much weight on Democrats will result in a poll that is skewed to Obama, and likewise, a weight that is skewed toward Republican will give McCain an advantage. While it is crucially important, there is hardly any consensus on this weight among pollsters. In fact, they are using different approaches.
For instance, Rasmussen constantly adjusts their party ID numbers so their daily tracking polls portray an accurate view of the electorate as a whole. Rasmussen establishes the weight based upon survey interviews with a separate sample of adults nationwide completed during the preceding six weeks. For example, for polling data released during the week of September 21-27, 2008, the partisan weighting targets used by Rasmussen Reports will be 39.0% Democratic, 33.5% Republican, and 27.5% unaffiliated. Taking the average of the September gives a distribution of 39.13%, 33.07%, and 27.8% for Democrat, Republican, and unaffiliated, respectively. It is a gap of 6.06% points Democrat's advantage.
With an unpopular Republican administration and the war in Iraq, it is generally believed that the Democrats enjoy an advantage on party ID in this election. But how big is really it?
In the first quarter of 2007, Gallup found Democrats enjoyed a "favorable political environment" where a 51.9% responded "leaned" toward Democrat while only 39.6% "leaned" toward Republican. That is a 12.3% gap. However, the gap has decreased in the last eight months. Based on the average from February to September 2008, the gap in party ID between Democratic Party and Republican Party has declined to 8% points.
A similar trend also found by the Pew Research Center. In 5,566 interviews with registered voters conducted by the Pew Research Center, the gap has decreased from 15% in the first two months of 2007 to 9% in the first two months of 2008; 36% identify themselves as Democrats, while 27% as Republicans.
Reasonably, party ID can change as people can change their mind. But should pollsters always adjust the weight to get a clear picture of people's preference toward a certain political party? It does not make sense to believe that people will change their party ID every day, even every week. If that happens, all of these polls would be virtually meaningless.
To avoid this issue, Pollster John Zogby uses a relatively constant weight by taking last two presidential exit polls as an indicator. His polls in 2004 used a party weight of 39% Democrat, 35% Republican and 26% Independent as indicated by the exit polls from the 1996 and 2000 presidential elections.
Based on the historical track record for the last ten years in terms of self-identified party affiliation from actual exit polls, the distributions are as follows: 38.4% Democrat, 35.8% Republican, 26.0% Independent, with only 2.6% point advantage for Democrat. And if we specifically use the exit polls from the last four presidential elections, the results are as follows: 37% Democrat, 35% Republican, 28% Independent.
Which one of these surveys really captures the true distribution? PEW's results tend to overestimate the Party ID for both Democrat and Republican, particularly the former. Take the year of 2004 for instance. The exit polls indicate that each Democratic Party and Republican Party had 37% of Party ID, while PEW's survey indicated that Democratic Party enjoyed a 6% point advantage, 47% to 41%. In 1992, it also shows a 5% points advantage for Democrat while exit polls indicated that none of the two Parties enjoyed any advantage in terms of Party ID. Hence, I am inclined to reject PEW's survey and exclude it from subsequent analysis.
For sensitivity analysis, I use the following scenarios for Party ID:
PID1: Average Gallup and Rasmussen (38.565%, 31.535%, and 29.9% for Democrat, Republican, and Independent). This is a solid advantage for Democrat.
PID2: Average Gallup, Rasmussen and past four president elections (38.04%, 32.69%, 29.27%). This is an advantage for Democrat.
PID3: Average past presidential elections (37%, 35%, 28%). This is a slight advantage for Republican.
Another big question regarding this party ID is, how often does it change? Specifically, what factors can produce a significant change (greater than 2% points) in the distribution? For instance, the current economic crisis seems to drive the undecided voters toward the Democratic Party. But will it net more than 2% points? Absent any other game-changer issue, the real Party ID will very likely fall into the above range, from 2% to 7% to the Democratic Party's advantage.
Within Party ID and Cross-over support
Another key data needed to do analysis is the within-party ID. The best data available so far is from the National Election Studies, stretching back to 1952 elections. Focusing on the elections since 1992, the averages of the 7 party ID are as follows:
Strong Democrat (18%)
Weak Democrat (17%)
Independent Democrat (15%)
Independent Independent (10.75%)
Independent Republican (12.25%)
Weak Republican (13.25%)
Strong Republican (12.75%)
There are many interesting features that can be drawn from this data. First, the Party ID gap between Democrat and Republican (11% points) is significantly higher than that of the exit polls from the last four presidential elections (1.7% points). For this reason, I will not use this as an indicator for within party ID. Secondly, the distributions of Party ID are 35% Democrats, 26% Republicans, and 38% Independents, with about 1% apolitical. Of the 38% Independents, 39.47% are leaning Democrats, 32.22% leaning Republican, and 28.29% stay Independent. This shows that the trend of Party ID among the Independents almost mimics that of all voters. Third, the within party ID shows that the distribution among strong Democrat, weak Democrat, and Independent Democrat (I will interpret this as conservative Democrat) is roughly 36%, 34%, and 30%, respectively. For Republican, the distribution is roughly 32%, 35%, and 33%. Both are reasonable and will be used as a benchmark.
As an alternative, I set a 60% weight for Liberal/Moderate Conservatives and 40% for Strong Conservatives. Next is to figure out the distribution for the Independents. I will construct it based on the last two months Gallup aggregate data. Among 57% of the pure Independents, 32% of them support McCain and 25% of them support Obama. I assume this also mimics the not-so independent Independents which gives a small edge for Republican. As for Democrat, the above distribution will still be used Hence, the alternative within party ID of the Independents is as follows:
Leaning Democrat: (25%/57%) × (1-Independent Party ID) × Independent Party ID.
Leaning Republican: (32%/57%) × (1-Independent Party ID) × Independent Party ID.
Pure Independent: Independent Party ID × Independent Party ID.
Sensitivity Analysis: Results
For raw data used for this analysis, see Gallup weekly aggregate.
1. Benchmark Scenario #2 is somewhat close to Gallup polls. This indicates that Gallup is very likely using a 5% to 6% point difference in Party ID for the Democratic Party's advantage and the within Independent Party ID mimics the general voters. Alternative Scenario #2 is somewhat close to Zogby polls. For instance, during the last two weeks, Zogby reported that Obama led McCain by 1% and 2 % points, respectively. Sensitivity analysis on Alternative Scenario #2 gives the same results.
2. Focusing on Scenario #2, when minor changes on the within party ID are made, the results are considerably different. Here, McCain becomes more competitive (Table 6).
3. If the party ID follows the past presidential elections, the race, until the first week of October, is still very close.
4. If the party ID follows the past presidential elections and the Independents tilt to McCain by 2.5% points (Alternative, Scenario #3), McCain is essentially up in the polls. This scenario, that McCain is leading among the Independents by about 2.5% points, is not something that is unattainable. Gallup surveys show that in the first and second weeks of September McCain led Obama among the Independents by about 9% to 12% points. However, in the first week of October, Obama led McCain by about 2%-3% points. The reversal trend is indeed dangerous for McCain. To be competitive or win the election, it is necessary for McCain to have an advantage by at least 3% points among the Independents.
5. Obama's increased poll numbers since September 1, 2008 are mainly due to two reasons. First, about 5% of the moderate and conservative Democrats switched their support from McCain to Obama. Secondly, about 9% of the liberal and moderate conservatives switched their support from McCain to Obama. Obama enjoyed a huge increase in support from the liberal/moderate conservatives, from 10% in the first week of September to 19% in the first week of October. Meanwhile their support for McCain decreased from 85% to 76% during the same period.
6. The results above indicate that to win the election, McCain still needs to convince his own traditional supporters while at the same time needs to convince a small portion of the moderate and conservative democrats, which could be Hillary's supporters, to switch back to him.
The 2000 CNN/USA TODAY/GALLUP Polls
Gallup's current presidential race polls are somewhat similar to those of Bush-Gore 2000. Entering the second week of October 2000, about a month before the Election Day, Al Gore was trailing Bush badly in the CNN/USA TODAY/GALLUP polls. About three weeks before the Election Day, Al Gore was behind by about 7% points. And a week before the E-day, Al Gore was still behind by 5% points (See below).
Week Bush Gore
9/7/00 - 9/15/00 43 48
9/16/00 - 9/22/00 42 49
9/23/00 - 9/30/00 46 45
10/1/00 - 10/7/00 44 46
10/8/00 - 10/14/00 47 44
10/15/00 - 10/21/00 49 42
10/22/00 - 10/28/00 48 42
10/29/00 - 11/4/00 48 43
11/5/00 48 43
ELECTION RESULT 48 48
As everyone knows, Al Gore got 48% of the votes.
Could the same thing happen this year? Maybe. Today's election, in some respects, is similar to that of 2000. None of the candidates is incumbent. Gore was part of the party in power (Clinton) and Bush was a candidate with a reform slogan. Change Clinton with Bush and "reform" with "change", you will get the idea.
The question is, can McCain still get at least 48% of the votes? Possibly.
With Alternative Scenario #2, it is more plausible, taking into account the Bradley Effect, and coupled with Gore's experience in 2000, McCain still has a very good chance to win this election.
Perhaps this is the reason why voter fraud in battleground states, so far committed by the far left organization ACORN, has already kicked in.
Elwin Tobing, Ph.D. lives in Fullerton, California.