Ipsos Watch

Posted by John Rosenthal

Is Ipsos sensitive to bad publicity? The latest AP-Ipsos Presidential Approval data would suggest so. The headline on yesterday’s AP story presenting the results of the poll reads: “Bush’s Approval Rating Rebounds Slightly”. I’m not sure that “slight” is the word for the movement that the Ipsos data in fact describes. The latest Ipsos poll shows the President with a 42% approval rating; the previous one from early November, with a 37% rating. This seems rather like a quite substantial rebound. It represents indeed the largest single month movement in the Ipsos Presidential approval numbers for all of 2005. (See table below.)

But, in any case, the apparent “improvement” in the President’s approval rating, like the apparent deterioration that preceded it, might be more an artifact of the volatility of Ipsos’s dubious polling methods than of any real volatility in public sentiments. As I noted in my recent article on the AP-Ipsos partnership on Tech Central Station, Ipsos’s dire November Presidential approval numbers were based on a sample in which 51% of registered voters identified as Democrats as opposed to only 40% who identified as Republicans. Perhaps the most remarkable detail amidst Ipsos’s latest data [pdf-file] is that now all of a sudden the party identification figures show a more plausible 47%-44% Democrat-Republican split. In a discussion thread on my TCS article on the Free Republic site , FR contributor “dvwjr” posted a very useful synthetic table of Ipsos presidential approval data for 2005. With his kind permission, I reproduce the table in slightly modified form below with the December figures added in. 

AP-Ipsos Presidential Approval Data, 2005

  7-Dec 9-Nov 2-Nov 5-Oct 16-Sep 6-Sep Aug July June May April Mar Feb Jan
Rep 44% 40% 40% 40% 41% 42% 39% 42% 40% 42% 41% 39% 39% 43%
Dem 47% 51% 49% 48% 49% 48% 48% 51% 50% 49% 48% 47% 52% 49%
Ind 9% 9% 11% 12% 10% 10% 13% 7% 10% 9% 11% 14% 9% 8%
                             
Reg 78% 78% 80% 78% 77% 76% 79% 77% 80% 79% 82% 78% 76% 80%
n-Reg 22% 22% 20% 22% 23% 24% 21% 23% 20% 21% 18% 22% 24% 20%
                             
App 42% 37% 37% 39% 40% 39% 42% 42% 43% 47% 44% 48% 45% 49%
Dis 57% 61% 59% 58% 57% 59% 55% 56% 55% 51% 54% 50% 54% 49%

 
The table shows that Ipsos’s apparent oversampling of Democratic voters in November was not a temporary blip. From February up until the latest December poll, the Ipsos sample shows a party identification gap persistently pushing 10% and sometimes exceeding it. The most astonishing figure is the 52%-39% Democrat-Republican split for February – this coming just three months after a general election that saw the Republicans win the White House and increase their majorities in both houses of Congress!

 
If there is an anomalous finding in the series of Ipsos data on party identification, it consists precisely of the latest, more plausible, December numbers. And whereas Presidential approval ratings might well show considerable volatility from month to month in response to particular events, it is, of course, far less credible that party identification – presumably a matter of long-term convictions regarding questions of principle – should do so. Does anyone at Ipsos care to explain?

Comments are closed.