Exactly, Ellen! 

NPR involved its listeners in the polling process a few years ago.  We were encouraged to ask follow-up questions.

My participation in this experiment taught me the extent to which respondents answer inaccurately (often answering quickly to avoid showing ignorance of the subject or adherence to a popular point of view) and then give their real response to the follow-up questions.   The one I remember the most went something like this:

"Do you trust government?"  "Hell, no!"  Follow-up:  "Who do you mean by 'government?'  Is it Congress?  Is it the guys in offices in places like the Agriculture Department?  Or is it, say, the President or the Governor?"  

Gradually the respondent eliminated all those and finally admitted that she trusted government but not 1) hypothetical crooked politicians, or 2) arrogant, unhelpful phone answerers in specific government offices. 

By the end of the conversation, she had persuaded herself that "hating government" was a stupid concept and that she depended a lot on government.

By living in DC, I have learnt that people draw conclusions which are influenced by their mood and filtered through preconceived notions-and even more surprisingly, through rumor and innuendo. They rarely ask questions-couching most opinions in the form of declaratives. Heaven forbid of you question more than two-for then you are being (ahem) 'uncool' or posing an 'annoyance'. 
Forget about trusting the government; we don't trust ourselves since we rely on other people glib opinions as to how we should live. 
Beyond two questions is just ohhhhhh sooooo mentally exhausting. Ohhh! I'm tired already thinking about it. 

By the end of the conversation, she had persuaded herself that 'hating government' was a stupid concept and that she depended a lot on government.----
Well if you ask questions which beg the conclusion, I'm not sure why you would believe that you are promoting critical thinking. I don't think there is a cognitive psychologist or logician who wouldn't be able to see through that ruse. 

Price of gasoline affects poll ratings because that is where it hurts the public.

That's so far over my head as to be completely out of my intellectual orbit.  And I'm embarrassed to admit to how very pedestrian my concerns are.

For example, in this study Eichenberg looked at 141 surveys of public attitudes about the use of force in Bosnia taken over 11 years (1992-2002).  During that lengthy time period the public's knowledge of and feelings about "Bosnia" can be presumed to have changed based upon its acquisition of facts such as the expulsion of the Croats, the siege of Sarajevo, the massacre at Srebrenica, the taking hostage of UN peacekeepers, the arrival of I-4 and then, S-4 peacekeepers, etc.  And these are but highlights during the 11 year period Eichenberger selects.

How do we know Eichenberger has properly adjusted the survey responses to account for the highs and lows of an irregularly changing level of emotion and knowledge which formed the public's basis for its expressed opinions?

It would be nice, if your explanation of the methods actually worked.  The fancy stuff he is doing is risky (unless the polls occur at even intervals), but it based on some relatively simple assumptions.  The main assumption is that the interventions (shifts between one type of condition and another) are appropriately modeled using an ARIMA intevention variable.  That is why he is using ARIMA. 

ARIMA is appropriate ONLY if the polls themselves occur at equal intervals, a highly dubious assumption for political polls, but perhaps he is solving this by pretending that anything that happens in a week as happening in the same time unit.

The intervention variable (in this case) is similar to what you would find in a regression model called a dummy variable.  It has a value of zero for the non-intervention period and 1 for the intervention period.  That is how he makes the line on his graph shift location.   If you look back at his graph, you will notice this shifting in the purple and red areas.  If he gets statistical significance (sorry, techie-talk) when tossing in the variable for the intervention at the right location, then he is justified in keeping it in his model.  Presumably he has.

The large number of surveys is required for different reasons.  First, it is required because ARIMA is invalid without a large number of observations.  Second, there is a hint in the literature, that while one person might misjudge the meaning of a survey question, the misjudgments of many people will tend to washout leaving a relatively accurate response if (1) questions are phrased in many ways and (2) many people respons.

Now, for his error.  He claims the light blue dots are just a "decay" of the red and purple periods back to the grey period.  However, he fails to note that the blue period has a sharp coincidence with the deliberate White House effort to restore its credibility.  To call it mere decay when there is another more likely explanation at hand is specious.  

Moreover, if its decay, why doesn't it look like some gradual shift?  This is a successful campaign.  There is a campaign going on at the same time.  So, the idea that its decay is silly.

Now, the fact that its the results of a campaign is actually good news.  How long can they keep it up?  With this monsterous lift campainging almost every day, Bush has won back 6 points.  He can't do it forever.... 

DMO,

Is it your position that the studies done by polysci profs such as Eichenberg's are exercises of limited value?  that they deploy marvelous methods of statistical analysis ("She blinded me with science") in search of an elephant and wind up giving birth to a mouse?

Or to put it another way, are surveys of public opinion -- usually a mile wide and an inch deep -- of any particular import? 

Professor Eichenberg,

The different approval rating that you have found for the war when people are asked about casualties is striking. It suggests to me that a process of compartmentalization is at work more than some rough cost/benefit analysis that was mentioned as a possible explanation in your article, Victory Has Many Friends. Consider the observations made on page 173 of that article:  

The single exception to the reticence of the public to involvement in internal conflicts is the popularity of regime change in Iraq. When survey questions mention that the purpose of the intervention is to remove Saddam Hussein from power, support levels increase by 17 percentage points. This may be because regime change in Iraq was seen as instrumental to U.S. security interests (and thus perceived as foreign policy restraint), or perhaps it resulted simply from the personalization of the conflict against a vicious dictator.

Unlike the lead up to the Balkan Wars of the Nineties, the deep divisions in Iraq were glossed over by the Administration during their promotion of the invasion. There was much talk of the reconstruction being financed by Iraqi oil and the grateful people who would cheer when the yoke of Saddam Hussein was finally thrown off. Maybe the approval spike you observed wasn’t a reflection of a consensus on U.S. security interests or the personalization of Saddam Hussein but rather the personalization of the Iraqi people.

As the war drags on, the simplistic talking points that the Administration used at the beginning of the conflict have been abandoned in the face of harsh realities. But the fluctuations you have charted as a response to positive or negative reports from Iraq may not have so much to do with the U.S. public guessing what conditions are really like in Iraq as permiting or restraining the belief in the New Iraqi.

The ideal New Iraqi is important to a lot of the initial supporters of the war because only he/she can crown us as Liberators.