Seite - (000100) - in Knowledge and Networks

92 were deleted (less than 5 % of no votes or abstentions). The resolution-based selec- tion was motivated by the presence of very specific resolutions on Israel for which only Israel and the United States voted against or abstained. These two states are clearly peripheral in the UNGA, and including these votes would not provide any supplementary information. The second selection, based on member states, com- prised those states that were often absent and not able to vote. Keeping the threshold used in previous studies, I chose to delete states that did not participate in at least 30 % of the votes. Failing to omit these states would produce a group lacking politi- cal consistency. The final tables included 158 states and 145 resolutions for the 42nd session (1987–1988) and 178 states and 68 resolutions for the 63rd session (2009– 2010). These basic measurements are congruent with the general trend in the UNGA in this period: a rising number of member states (with the addition of former Eastern-bloc countries in the 1990s and many small states in the 2000s) and a diminishing number of resolutions put to a vote. Several network approaches can be used to study voting behavior in the UNGA. The most common method is to create a multibipartite state-resolution graph and to transform it into a state-state similarity matrix with range values from 0 (two states always vote in a different way) to 100 (two states always vote the same way). One issue concerns the threshold process because finding a value suitable for different sessions seems challenging. The solution proposed by Beauguitte (2011) was to choose the same statistical threshold for all similarity tables in order to allow comparison. However, transformation from continuous to discrete values remains quite unsatisfactory. Although some authors proposed this approach in the 1960s (Lijphart, 1963), multivariate analysis (mainly principal component analysis) soon became the canonical method of handling this data. An alternative option that allows keeping link weights is to adopt a variation of the CONCOR method adapted for valuable matrices; the classification procedure groups together states that have the same relational profile. Figure 5.1 shows the regional structure based on voting behavior before the end of the Cold War (42nd session, 1987–1988), and Fig. 5.2 reflects the situation in 2010 (63rd session, 2009– 2010). The considerable cluster inertia is quite surprising but confirms results obtained in previous research (Voeten, 2000). Even today, there is one Northern bloc in opposition to one Southern and Eastern bloc. In 2009 the East appeared smaller because eastern European countries began to behave like western ones— adhesion or application to the EU—but the main fracture between developed and less developed countries continued to be the dominant pattern. The voting pattern of Turkey, one of the few countries showing a marked change, was becoming more and more like that of western countries. The cluster inertia is also related to the structure of resolutions voted upon in the UNGA. The Israeli–Palestinian conflict is the sub- ject of more than one third of all resolutions voted upon in each session. Regarding this issue, the West strives for a balanced stance, whereas other members share a pro- Palestinian approach, which partly explains their cohesiveness. In 2009 there appeared a tiny group of small developing island states that tended to vote like Southern-bloc states, except when resolutions concern global warming. L. Beauguitte