Hell Hath No Fury…
George W. Bush’s news conference today comes at a low point in his presidency; in the most recent (July 8) national survey, conducted by Gallup with USA Today, just 29 percent of Americans approve of the president’s overall performance in office, while 66 percent disapprove. Bush has been under 30 percent approval in four of the last six national news polls.
He’s at a career low in the Gallup poll; among postwar presidents only Harry Truman, Richard Nixon and Jimmy Carter have gone lower (22, 23 and 28 percent, respectively). Bush has been below 50 percent approval in ABC News/Washington Post polls continuously for the last two and a half years, since January 2005. Truman was the only postwar president to stay that low for longer (three years).
Bush now holds one of the steepest declines in popularity on record – more than 60 points, owing to the fact that he peaked at 92 percent in an ABC/Post poll shortly after 9/11. The only larger fall in support was Truman’s, from a high of 87 percent approval to a low of 22 percent; third was Bush’s father’s, from a high of 90 percent to a low of 33. Naturally, these reflect not only how low these presidents went, but also how high.
Nixon was hammered by scandal, Truman and Bush’s father by economic discontent, Carter by the double-punch of a bad economy and the Iran hostage crisis. For Bush it’s all about the war in Iraq. In Gallup data 62 percent call the war a mistake; in our own last ABC/Post poll 61 percent said it was not worth fighting – and majorities have said so continuously since December 2004.
Just 22 percent (Gallup) think the surge of U.S. forces is making the situation any better. Most want the deployment of U.S. troops reduced and eventually ended. Seven in 10 in Gallup’s poll favored an April 1 deadline, but in fact when and how to withdraw U.S. forces is not fixed in public attitudes, given compunctions about leaving Iraq in chaos (see Monday’s blog item).
Other matters don’t help – e.g., 66 percent say Bush should not have intervened in the Scooter Libby case. But in terms of public opinion, this president’s problems are all about Iraq; his approval rating and views of whether the war was worth fighting have correlated since April 2003 at a near-perfect .93. Indeed, Bush’s approval year-to-year, declining sharply as the Iraq war has ground on, resembles nothing so much as Lyndon B. Johnson’s as the nation became enmeshed in Vietnam 40 years ago. Plainly, in politics, hell hath no fury like an unpopular war.
Email
USS Gabrielle Giffords Honors Courage
Mitt Romney's Full Speech at CPAC 2012
Although there’s undoubtedly a relationship between Bush approval and Iraq, that 0.93 correlation looks suspiciously high. Did you make that calculation on month-to-month (or poll-to-poll) *changes* in opinion or just the raw values of approval and “Iraq was worth it” percentages?
Posted by: Chris G | July 15, 2007, 3:23 pm 3:23 pm
Chris, we based the correlation on the raw values in each poll we’ve conducted in which both questions were asked (most recently listed at http://abcnews.go.com/images/US/1040a3PoliticsandtheWar.pdf, Q1 and Q9). I think it’s fair to say that correlating values rather than change scores is the standard approach in our field. In this case in particular we’re looking at the ongoing relationship between these two variables, not the relationship between the change in each one. Correlating change seems less straightforward conceptually, as well as increasing the variance. If there’s a case for using change scores here, I’m all ears. But I think the correlation we’ve presented is meaningful, and yes, remarkable.
Posted by: Gary Langer | July 16, 2007, 1:24 pm 1:24 pm
Public opinion tends to reflect an accumulation of changes over time, just like the stock market or population, right? So using Pearson’s correlation in the way you did actually violates several fundamental assumptions in that test (data points are not independently and identically distributed, or “stationary”). Restricting analysis to month-by-month changes is one way of dealing with this (“differencing” or “detrending” in time-series analysis jargon).
For example, generate two completely random, independent vectors of 30 numbers between 0 and 1 (e.g. in Excel just rand()). Then take the cumulative sum of each vector. I guarantee you that you will find that these cumulative sums will, on average, have a correlation well over 0.9. But the only thing they have in common is that they are both going up. Countless processes in the world show some kind of positive or negative trend over time. Just because two of them happen to have the same trend doesn’t mean they are related statistically (let alone causally).
Obviously Iraq and Bush approval are related, but if a statistic doesn’t account for possible violations then it can be grossly exaggerated. I’ve analyzed similar data, but month-by-month, and find a correlation of about 0.65. There’re of course many other ways of asking about the link between approval and Iraq without factoring in time. What I did just asks how strongly changes in opinion are related.
I think packages like SPSS usually have built-in functions for time-series analysis, and pollsters should use them instead of just applying standard statistics. There are dozens of resources online, here’s a good one:
http://statistik.mathematik.uni-wuerzburg.de/timeseries/
Sorry to be a crank
Posted by: Chris G | July 16, 2007, 7:36 pm 7:36 pm