This is an article that I found at
http://www.pollingreport.com/incumbent.htm and I think it relates well to the subject.
Incumbent Races:
Closer Than They Appear
by Nick Panagakis
How will undecideds vote on election day? Traditionally, there have been two schools of thought about how undecideds in trial heat match-ups will divide up at the ballot box. One is that they will break equally; the other, that they will split in proportion to poll respondents who stated a candidate preference.
But our analysis of 155 polls reveals that, in races that include an incumbent, the traditional answers are wrong. Over 80% of the time, most or all of the undecideds voted for the challenger.
The 155 polls we collected and analyzed were the final polls conducted in each particular race; most were completed within two weeks of election day. They cover both general and primary elections, and Democratic and Republican incumbents. They are predominantly from statewide races, with a few U.S. House, mayoral and countywide contests thrown in. Most are from the 1986 and 1988 elections, although a few stretch back to the 1970s.
The polls we studied included our own surveys, polls provided to us directly by CBS, Gallup, Gordon S. Black Corp., Market Opinion Research, Tarrance Associates, and Mason-Dixon Opinion Research, as well as polls that appeared in The Polling Report.
In 127 cases out of 155, most or all of the undecideds went for the challenger:
DISPOSITION OF UNDECIDED VOTERS
.
Most to challenger 127
Split equally 9
Most to incumbent 19
The fact that challengers received a majority of the undecided vote in 82% of the cases studied proves that undecideds do not split proportionally. If there were a tendency for them to split proportionally we would see most undecided voters moving to incumbents, since incumbents win most elections. Similarly, even accounting for sample error, it’s clear from the chart above that undecideds do not split equally.
For poll users and reporters this phenomenon, which we call the Incumbent Rule, means:
* Incumbent races should not be characterized in terms of point spread. If a poll shows one candidate leading 50% to 40%, with 10% undecided, a 10-point spread will occur on election day only if undecideds split equally (i.e. a 55% to 45% outcome). Since most of the 10 points in the undecided category are likely to go to the challenger, polls are a lot closer than they look – 50% to 40% is likely to become 52% to 48%, on election day. If a poll is a mirror of public opinion, think of an incumbent poll as one in which objects are closer than they appear.
* An incumbent leading with less than 50% (against one challenger) is frequently in trouble; how much depends on how much less than 50%. A common pattern has been for incumbents ahead with 50% or less to end up losing. Final polls showing losing incumbents ahead are accurate. The important question is whether results are reported with an understanding of how undecideds decide.
* Many polls may have been improperly analyzed and reported. Some postmortem accounts of polls have been inaccurate -- many polls remembered as wrong were, in fact, right. It’s only natural to interpret the term "undecided" literally. But as with so many other findings in survey research, data should be analyzed according to what they mean, not what they say.
Undecided about the Incumbent
Why do undecided voters decide in favor of challengers?
It seems that undecided voters are not literally undecided, not straddling the fence unable to make a choice – the traditional interpretation. An early decision to vote for the incumbent is easier because voters know incumbents best. It helps to think of undecided voters as undecided about the incumbent, as voters who question the incumbent’s performance in office. Most or all voters having trouble with this decision appear to end up deciding against the incumbent.
The exceptions we found to the Incumbent Rule help support the theory on why this happens.
Many challengers who did not get a majority of undecideds were recent or current holders of an office equal to the one they were seeking. Voters were equally or more familiar with the challenger’s past performance in a similar office, so the challenger assumed incumbent characteristics. Other exceptions include well-known challengers or short-term incumbents.
Some examples of where more undecideds voted for incumbents or split evenly:
Last year in Minnesota, where Hubert Humphrey III challenged Sen. David Durenberger; and in Nebraska, where Bob Kerrey, the former governor, challenged David Karnes, who had been appointed to his Senate seat. In 1986 in Florida, when incumbent Sen. Paula Hawkins faced ex-Gov. Bob Graham. And in Chicago in 1979, where two-year incumbent Mayor Michael Bilandic split undecided voters with challenger Jane Byrne.
These examples and similar ones account for 17 of the 28 exceptions to the Incumbent Rule that we uncovered. In some of the remaining cases, the incumbent simply turned the race around in the final days. A good example of this is the 1982 Missouri Senate race pitting incumbent John Danforth against Harriet Woods. Other exceptions can be explained by sampling error.
There is an interesting pattern in the polls where most undecideds voted for challengers. In 98 of the 127 cases (77%), the incumbents’ final polls standing was plus or minus four percentage points from the actual election result. The most frequent result was two points gained by the incumbent over the final poll preferences -- 24 cases in all.
In 41 cases, or 32% of the 127, the incumbent ended with less than his stated poll percentage. This means that about one in four of all 155 polls actually overstated the incumbent’s percentage.
Of the 127 challengers who gained more undecideds than did incumbents on election day, 78 gained 10 or more points over their stated poll percentage.
Making allowances for factors stated above, most polls appear to estimate support for the incumbent. All or most undecideds end up with the challenger regardless of the size of the undecideds.
Most troublesome are polls showing an incumbent leading but who ends up losing the election.
Some examples: In Wisconsin in 1986, incumbent Gov. Tony Earl and incumbent Attorney General Bronson LaFollette were ahead in the late polls with less than 50%, but lost by five and seven points, respectively. In 1986, one poll showed Georgia incumbent Sen. Mack Mattingly ahead by 10 points, but he gained only one more point to lose with 49%. In 1984, incumbent Illinois Sen. Charles Percy led with 45% and 49% in final polls and wound up losing the election 48% to 50%. ...
Avoiding Election Day Surprises
The overwhelming evidence is that an incumbent won’t share the undecideds equally with the challenger. To suggest otherwise by emphasizing point spread or to say that an incumbent is ahead when his or her percentage is well under 50% leads to election day surprises.
