|
|
|
|
Campaign 2008 Entries
ABC News polling expert Gary Langer discusses un-leaned and leaned Party ID
Rasmussen changes Party-ID-weighting method -- party template is now updated each week, based on past six weeks' totals
A look from Pollster.com at partisan composition trends from states where voter registration is by party
Nate Silver at FiveThirtyEight has a nice posting on weighting for Party ID
Pollster.com's Mark
Blumenthal weighs in on "Party
ID Wars" for this year's presidential campaign
Sample for Obama-McCain Wisconsin Poll Raises Partisan Composition Issue
by Alan Reifman
The newly released (June 12, 2008) UW Political Science/WisPolitics poll of Wisconsin "probable voters" raises some issues, in my mind at least, of whether Democrats were oversampled. In a state that has gone Democratic in the 2000 and 2004 presidential elections by only the slightest of margins, Democrat Barack Obama topped Republican John McCain by an eye-opening 13 percent in the recent poll (50-37). True, the timing of the poll right after Hillary Clinton's suspension of her campaign and announcement of support for Obama may have given him a "bounce." However, the party composition of the surveyed sample leads me to think that Obama's lead likely is artificially high.
As shown in the table below, the recent sample consisted of 14% more self-identified Democrats than Republicans, 38% vs. 24% (column 1). These are responses to party-identification questions, not candidate preferences, although the two tend to be congruent (i.e., self-identified Democrats tend to vote for the Democratic candidate, and same for the GOP). The party ID gap in favor of the Democrats widens to 16% when we eliminate categories other than Democrat (D), Republican (R), and Independent (I), in order to make things more comparable with other polls (column 2). The above-linked article announcing the results of the recent Wisconsin poll notes the virtual parity in D and R identification in the '00 and '04 elections. There is no question that Democrats have been building a substantial party ID edge over the Republicans over the last two years or so, both nationally (see the section immediately below on this page) and in Wisconsin. But has the gap grown this large? In the 2006 Badger State governor's race, the Democrats held a 4% advantage (column 3).
| Original Proportions 2008 Wisconsin Poll | Modified Proportions 2008 Wisconsin Poll* |
Exit Poll from 2006 Wisc. Governor's Race |
|
| Republican | .24 | .26 | .34 |
| Democratic | .38 | .42 | .38 |
| Independent | .29 | .32 | .27 |
|
*Readjusted to exclude other, no preference, don't know, and refused on party ID item, thus making R, D, and I sum to 1.00. |
|||
My advice to supporters of a candidate has always been to look at polls as
pessimistically as possible. Any Obama supporter who accepts a 14%
party ID edge at face value -- an optimistic or even reckless stance, in my
view -- runs the risk of being blindsided by polls that are (likely) much
closer later on. In accordance with this philosophy, I would like to
re-weight the recent Wisconsin poll as if the partisan composition of
the surveyed sample had matched that from the 2006 governor's race. To
do this, we also need to know how Obama and McCain did in the poll among the
partisan subgroups. Fortunately, the above article reports the
following:
Party polarization is a striking feature of Wisconsin opinion at this time with 87 percent of Democrats and 89 percent of Republicans saying they would vote for their party’s nominee if the election were held today. “For all the talk about a new politics, the electorate is sharply split,’’ said [UW professor Charles] Franklin. Among independents, Obama holds a nine-point lead, 46 percent to 37 percent for McCain with 17 percent undecided.
Here's how we re-weight, for Obama first:
We can say the Democratic share is 38 "units" big, of which Obama captures .87. Multiplying the two figures yields almost exactly 33%. The Republican block is 34 units big, of which .11 is for Obama (I'm assuming anyone not among the .89 of Republicans for McCain is for Obama). These numbers multiply out to 3.74%, which we'll round to 4%. Finally, among the Independents (27 units big), Obama takes .46, yielding 12.42%, which we'll round to 12%. The three obtained products (in blue) sum to 49%. This number is virtually identical to the 50% support figure in the recent Wisconsin poll; this is not surprising, as Democrats comprised 38% of the sample in the poll and in the exit poll from 2006.
Now let's do McCain:
Again, the Democratic share is 38 "units" big, of which McCain gets .13 (assumed from what's left after Obama's support ). Multiplying the two figures yields essentially 5%. The Republican block is 34 units big, of which .89 is for McCain, multiplying out essentially to 30%. Finally, among the Independents (27 units big), McCain wins .37 yielding 10%. The three obtained products (in red) sum to 45%. Thus, positing that self-identified Republicans comprise 34, rather than 24, percent of the Wisconsin electorate makes the Obama-McCain contest much closer, 49-45.
As a final aside, I took a summer statistics course at the University of Michigan (my graduate school alma mater) in 1985 from the aforementioned Dr. Franklin.
UPDATE: Dr. Franklin addresses the
issue of party ID in Wisconsin, on his
own blog (once the new page comes up, you may have to scroll down to
find that particular posting). Apparently, the shift in recent years
to a sizable Democratic edge is pretty robust.
Trends in Partisan Composition
(Party ID) of the Electorate
***
Ipsos/AP Poll, March 3-5, 2008
D 39-R 24; with leaners, D 52-R 35
***
Politico.com, September 18, 2007, "GOP Identification Drops"
Party identification — or affiliation, as
it’s also called — measures whether people consider themselves to be
Republican,
Democratic or independent. In 2004, according to [Public Opinion
Strategies'] merged polling data (think of it as one huge data set
compiled over the course of a year), Republicans were at rough parity with
Democrats in party ID, trailing them by three percentage points.
At this point in 2007, they trail Democrats by seven.
Other, nonpartisan national surveys show a similar disinclination to
identify with the GOP.
***
Pew Research Center for the People and the Press
March 20, 2008
"Fewer Voters Identify as Republicans:
Democrats Now Have the Advantage in 'Swing' States"
March 22, 2007
"Trends in
Political Values and Core Attitudes: 1987-2007:
Political Landscape More Favorable to Democrats"
Party ID-Related Information Compiled During the
2006 Midterm Election Campaign
Dr. Reifman's summary of the party ID/sample weighting issue in the 2006 elections is available at Pollster.com.
End-of-the-year 2006
Gallup report
on party ID at the national and state levels
(discussed
here and
here at Pollster.com)