Mission Accomplished: The Wartime Election of 2004

The dynamic has the form of a second-order autoregressive process, with the following parameter estimates and standard errors for elections from 1828 to 2004: \( {\hbox{VOTE}}(t){\hbox{ }} = \mathop {49.3}\limits_{(.81)} + \mathop {.48}\limits_{(.12)} \,{{\hbox{VOTE}}(t - 1)}{\hbox{ }} - \mathop {.51}\limits_{(.11)}\,{{\hbox{VOTE}}(t - 2)} .\quad N = 45\quad {\hbox{SER}} = 5.4 \)

The elections of 1812 and 1816 could not be considered for the estimation of the autoregressive vote model since popular voting was not widespread enough then.

Gallup and the NYT/CBS polls use the following wording: “Do you approve or disapprove of the way [Name] is handling his job as president?” The WSJ/NBC polls ask the following: “In general, do you approve or disapprove of the job [Name] is doing as President?”

NYT/CBS polls always ask and Gallup almost always asks the following: “Do you approve or disapprove of the way [Name] is handling the economy?” In this time period (1990 through 2004), Gallup has asked two variations of this question. The first, used through March 1992, asked: “As I read off each problem, one at a time, tell me whether you approve or disapprove of the way [Name] is handling that problem... Economic conditions in this country.” From April 1992 through February 1993, Gallup asked the following: “Now please tell me whether you approve or disapprove of the way [Name] is handling some specific problems facing the country... the economy.” The WSJ/NBC polls ask: “Do you generally approve or disapprove of the job [Name] is doing in handling the economy?”

The NYT/CBS polls ask the following: “Do you approve or disapprove of the way [Name] is handling foreign policy?” Gallup has used two different questions in this period. The first, used until 1992, asked: “As I read off each problem, one at a time, tell me whether you approve or disapprove of the way [Name] is handling that problem... Foreign policy.” From 1992 to 2004, Gallup asked: “Do you approve or disapprove of the way [Name] is handling foreign affairs?” The WSJ/NBC polls ask: “In general, do you approve or disapprove of the job [Name] is doing in handling our foreign policy?”

The algorithm for aggregating polls drops the last observation, so our analysis covers each of the series until November of 2005.

A two-tailed t-test for the difference between the coefficients produced a test statistic of −0.194, which has a p-value of 0.846 (267 degrees of freedom).

The test of the difference between these two coefficients produces a t-statistic of 3.255, which corresponds to a p-value of 0.002 (54 degrees of freedom).

This also includes a second-order autoregressive process for the error term.:

The “war opposition” measure was constructed by an aggregation of results from three polling organizations. The CBS News/New York Times question asked: “Looking back, do you think the United States did the right thing in taking military action against Iraq, or should the U.S. have stayed out?” The Pew Center question asked: “Do you think the U.S. made the right decision or the wrong decision in using military force against Iraq?” And the Gallup polls asked: “In view of the developments since we first sent our troops to Iraq, do you think the United States made a mistake in sending troops to Iraq, or not?”

All variables of the approval model have been treated in level form, not in differenced form. Diagnostic checks have confirmed this choice. What is more, model estimates for the variables in first differences do not suggest any drastic revisions of the results in level form.

The 9/11 rally, so our estimate, had a life span of 69 months, using the following equation: (Impact in 9/01 + Impact in 10/01) × (Decay ^X ) = 0, where X equals a number of months since the final impact.

Note that, while the Bush approval series begins in January of 2001, the time series for war opposition and casualties only begin in March 2003. Setting values for these two series equal to zero for the time points prior to that might raise objections. It could be argued that this step is bound to overestimate the effect of casualties on approval by “holding” casualties at zero when in fact they do not exist. To address this objection, we tried several alternate measures of casualties. Specifically, we used (1) total War on Terror (including Iraq) casualties, which begin to be reported in October of 2001; (2) Iraq casualties with the mean casualty figure imputed for all time points prior to March 2003; and (3) Iraq casualties imputing the minimum casualty figure for those months. Regardless of casualty measurement, the results were remarkably consistent: casualties begin to have a negative effect on approval only when Iraq opposition crosses a threshold around 53%. We also estimated the model on a post-Iraq sub-sample, beginning in Mach 2003. Again we find the threshold is around 53%.

Our measure of war opinion tends to overstate the extent of opposition. It was derived by subtracting the proportion supporting the war from 1.0, and thus includes those without opinion as well.

The question in the 2004 exit poll asked, “In general, does John Kerry mostly say what he believes, or say what he thinks people want to hear?” A quarter of the Democrats and slightly over half of the Independents chose “say what he thinks people want to hear.”

The estimates for the logit model are provided in Table 4. It must be noted that the 2004 exit poll did not ask all voters the same set of questions. Most questions were put only to subsets of the sample. The large subset that was asked about the Iraq invasion and current conditions in Iraq was not given the questions about whether the Iraq war improved the long-term security of the United States, or whether the war in Iraq was part of or separate from the war on terrorism, each of which was asked in separate and smaller subsets of the exit poll.

In contrast, Gelpi, Feaver, and Riefler (2006) report a vote probability for the “wrong & well” type of voters (“Pottery Barn” in their terminology) that favors Kerry by a good margin. But in their analysis this group is also much larger than it is in the exit poll, while the anti-war segment is much smaller. What may account for the discrepancy is the possibility of movement from one group to the other since the time before the election when Gelpi et al. conducted their survey and Election Day. The Gelpi et al. vote estimate for the other conflicted group (“Noble Failure”) is consistent with ours.