Home Communication Public Opinion, Transatlantic Relations and the Use of Force
Available poll data
In order to approach the problem of measuring the impact of situational factors, we proceed in a number of steps. Unfortunately, conventional surveys seldom make a concerted effort to allow us to do the kind of analysis that we need. Hardly any polls use questionnaires that aim at exploring systematically the interaction between (some of) the various factors in our model, let alone all of them, and allow a multivariate analysis. When, for instance, mission support, perceived success and tolerance of casualties all decline monotonously, as was the case with the war in Iraq, it remains difficult to disentangle which is causing which.
To compensate for this weakness we made an effort to test the model in another way.
First, it was decided to select four different cases of military conflict which are both relatively recent and international in terms of involvement, military or otherwise. We decided to select the air intervention against Serbia over Kosovo in 1999, the terrorist attacks of 2001 and the military actions taken or considered in that context, the military actions against the Taliban group in Afghanistan since 2001, and the war against the regime of Saddam Hussein in Iraq since 2002. Other earlier and later conflicts to consider, given the degree of international involvement and the availability of comparable poll data, could have included the Gulf War against Iraq of 1991 and the intervention in the civil war against the regime of Gaddafi in Libya. These were left out in order to limit our ambitions. They are obvious candidates for a future enlargement of this study. No data from before 1999 and beyond 2003 were included in the analysis.
The next step was to collect as many outcomes of polls on these conflicts as we could find. We used various available data sources and archives and availed ourselves of the possibilities of the Internet, which increased tremendously during the period considered here. We tried to collect poll data from as many countries as we could manage.
Since this was to be our dependent variable, each poll had to have at least one question measuring support for the use of military force, either by one's own country or more generally. Poll questions, not individual respondents constitute our unit of analysis. Only questions that referred to support for the use of force were included in the data set, not, for instance, questions on the likelihood of particular outcomes.
This search produced a final data set of over 3,000 poll questions that were asked once or more often (and sometimes quite frequently) and constituting a time series. In the latter case each poll was taken as a separate unit of analysis, however.4
Our dependent variable consists of the degree of support for the use of military force. Various kinds of wording are being employed for this purpose: 'support' versus 'oppose', 'agree' versus 'disagree', 'justified' versus 'not justified', 'right' versus 'wrong'. Since there turns out to be little variation among these question in degrees of support all such questions were included in order to increase the statistical power of the analysis (Eichenberg, 2006: 5).
The figures for the dependent variable (support score) is not only based on those questions for which we have a straight yes-no answer in the dependent (support) variable, but also those questions which originally had more than two answer categories. The original multiple answers have been recoded and dichotomized. Average percentages for 'yes' are used as the support score in both combined cases.
Is our data set a complete and/or representative collection? Most probably there are smaller or larger lacunae. Although an effort has been made to be as thorough and complete as possible in collecting the data, it is not unlikely that a number of polls have been held that were overlooked. These were therefore not included in the analyses, although they could have strengthened the analysis. More seriously, although unlikely, it could be that the absence of data would not be random but of a systematic nature and thus bias our results. We have no suspicion that this might indeed be the case, but cannot rule out that these missing data would have thrown a different light on the issues discussed in this and the next chapter.
Looking at the origin of the data, the first thing to note, however, is that while polling on international issues has become more frequent in recent years in general, there is still relatively more polling going on in the United States compared to Europe. Moreover, efforts to collect data and make them publicly available is also better developed in the US.5
However, despite obvious differences in coverage, the data set that we work with is probably the best that could be collected at the present time. Table 6.1 gives the results of our data collection, by country (group) and conflict issue.
Table 6.1 Total number (N) of observations, by issues and countries
Table 6.2 Support for the use of force in four international cases (average percentages and differences from mean scores)
The support score is an average of the support percentage across all relevant cases. Questions with multiple answers have been dichotomized.7. The 'Other' category includes a set of more than 40 countries for which polling data were available.
For a first cut, the countries where the questions had been asked were grouped into three clusters: United States, Europe and Other countries.6 While the latter two are, of course, a mixed lot, and further disaggregation may be instructive, this created large enough groups to allow meaningful statistical conclusions. They allow us not only to compare views in the US with those in Europe, which is and remains the major purpose of this study, but also to put this comparison into a wider, worldwide perspective.
|< Prev||CONTENTS||Next >|