States’ interests at international climate negotiations: new measures of bargaining positions

Abstract

To advance empirical research on international environmental institutions, new data on national positions at the international climate change negotiations are introduced. The observations cover more than 90 countries at two historical moments of climate change decision making: the pre–Kyoto Protocol enforcement (2001–2004) and the post–Kyoto Protocol (2008–2011) meetings. Data were collected from different types of written text. Coding entailed a qualitative (dictionary-based) content analysis and a quantitative text analysis. By systematically exploring these new data, I offer a ‘map’ of national preferences at the United Nations Framework Convention on Climate Change (UNFCCC). I also propose a discussion of the dimensions of conflict and policy competition over 10 years of climate negotiations.

Keywords:

• climate cooperation
• climate change negotiations
• bargaining positions
• UNFCCC

Acknowledgements

This work is based on my dissertation, funded by the German Academic Exchange Service and written at the University of Konstanz. I thank the Environmental Politics editor and two anonymous referees for very useful and constructive comments. I am also grateful to Gerald Schneider, Fabio Franchino, Katharina Holzinger, the 2011 EITM Europe team, and participants of the 2012 ISA and EPSA conferences for valuable feedback. The article refers to data and an appendix that are available at www.federica-genovese.com.

Notes

^1. I refer to issues as ‘areas’ where nations have specific preferences (Hix and Crombez 2005). Issues can overlap with topics. I use these two concepts interchangeably, as both are related to the UNFCCC negotiation agenda points. Note that these are not intended as ‘dimensions’. I discuss how issues (and thus topics) relate to the dimensions further below.

^2. The fact that the NCs are compulsory submissions lowers the risk that the data are missing in a systematic (non-random) way. It is realistic to believe that some NCs can be copied across countries. This is also because part of the international bargaining may take place before the general assembly, i.e. in the coalition meetings before the observed negotiation rounds. However, I follow the general bargaining literature and assume that positions in national documents are sincere (Laver et al. 2003).

^3. A UNFCCC provision for failed consensus exists, but countries never actually use it.

^4. The 2001 conference refers to the mid-year Bonn meeting in July 2001, which for several special circumstances represented a true full convention (Dessai 2001).

^5. The COP06 and COP14 agreement texts represent the status quos for the two bargaining periods. The COP10 and COP17 agreement documents, by contrast, provide the two respective final outcomes.

^6. Although ideally one should study position papers, the fact that the NCs are more technical reports is not a problem for the sake of comparing similar content.

^7. Arguably countries may anticipate or delay the NC submission for strategic reasons. However, I do not have prior information or theoretical reasons to assume that countries may be deliberately submitting the NCs late.

^8. The mentioned NCs are in English, French, or Spanish. Ukraine and Russia submitted NCs in 2010 only in Russian. Since I do not have access to Russian-speaking assistants, I have not included them.

^9. Although the CMP studies parties, the coding approach has migrated to the analysis of political discourse in other decision-making organisms (e.g. parliament speeches) with significant success (Benoit and Laver 2012).

^10. The manual coding strategy is the following: the text is parsed into units (i.e. periods or short paragraphs). I screen the text units to search for words that carry relevant content. The issue-based dictionary facilitates the ‘screening’. In cases where the text unit is indeed connected to an issue, I assign it to such an issue. If instead the text unit does not refer to an issue, reflecting some other topic that is not relevant for this work, it is left uncoded. This type of coding does not rely on the sentence-by-sentence sequence. This implies that coders’ priors do not affect the quality of the coding, and the generated values are independent and identically distributed (i.i.d.) (Benoit et al. 2012).

^11. Note that, due to randomly missing observations, I have performed linear imputations.

^12. Weiler (2012) and colleagues, for example, have information on the repetition of issues on UNFCCC legal texts, but only cover issues discussed at the 2009–2010 negotiations.

^13. Quinn’s estimator (2004) models the variable:

where Λ is a matrix of factor loadings, φ_i is the vector of factor scores, ε_i is the error term, and x*_ij is the vector of latent responses associated with the elements of X:

where j are  the indexed responses, i are the observations, c is the categorical indicator that identifies whether a variable is ordinal, and γ is a collection of cut-points that tend to infinity. The analysis was performed with the MCMCpack in R. For details, see online Appendix.

^14. Scree test plots obtained from the principal component factor analysis with varimax rotation further support the strong loadings on Factor 1. Figures in the online Appendix show that the eigenvalue of the first component is noticeably higher than the rest. It is therefore safe to presume that this measure captures the major latent variation of national preferences.

^15. Note that I use quantitative text analysis only on the NCs, as the agreement texts are too few for an accurate estimation.

^16. As with other quantitative text analysis software, Wordfish works only with texts in the same language. Hence, I kept the English documents and dropped the French and Spanish ones. The samples include 65 and 63 NCs for the 2001–2004 and 2008–2011 respectively.

^17. This inference holds robustly to Wordfish estimations with other reference documents and on different text subsamples (details in the online Appendix).

^18. The results of the ordinary least squares regressions are (for both periods) R² = 0.5, β = 0.58, p = 0.000, S.E. = 0.07. The significance also holds when dropping outliers.