Footnote 75, Chapter 6, Page 161:

The cases that are quantitatively analyzed for this book were compiled through reference to the MAR dataset, rather than datasets constructed through reference to fractionalization data. Douglas Rae’s fractionalization measure, known as F, indicates the “degree to which a society is split into distinct groups,” by telling “the probability that any two randomly selected voters disagree in their voting choice” (Esteban & Ray, 2008, p.166; Wildgen, 1971, p.234). Analysis of Lijphart’s definition of the term, “plural,” indicates that it is not evidenced by ideological difference. Therefore, it is inappropriate to choose this book’s cases according to F because it does not differentiate between ideological and other forms of fractionalization. A measure of ethnolinguistic fractionalization, known as ELF, focuses on group characteristics that fit Lijphart’s definition of “plural” but does not incorporate the second component of this definition, which requires intergroup potential antagonism. ELF “measures the probability that two randomly selected individuals from the entire population will be from different groups” (Cederman & Girardin, 2007, p.174). Like the MAR dataset, construction of ELF data requires identification of these groups. Fractionalization measures “vary significantly,” apparently because division of populations into such groups requires “many subjective decisions” and emphasis on certain types of divisions is less appropriate for depiction of some societies (Driessen, 2008, p.23, 22; Fearon, 2003, p.215; Cederman & Girardin, 2007, p.174; Lind, 2007, p.123). The MAR dataset is used for this book’s quantitative analysis because, although it has these group identification problems, its groups are identified according to whether they are “at risk,” and thus potentially antagonistic, rather than the extent to which they constitute identifiable categories. The ELF’s focus on diversity rather than potential antagonism is illustrated by the fact that that “most of the literature fails to find any significant evidence of ethnic fractionalization as a determinant of conflict” (Esteban & Ray, 2008, p.164). Laitin and Daniel Posner (2001, p.15) observe that there are many groups which are “culturally distinct from their neighbors but that are irrelevant as political actors in their own right.” Like a number of other scholars, they advocate constructing a fractionalization index that “reflects the groups that are actually doing the competing” (Laitin & Posner, 2001, p.15). Joan Esteban and Debraj Ray (2008, p.163-165) observe that fractionalization and polarization have profoundly different effects on conflict levels. Quantitative analysis conducted by Jose Montalvo and Marta Reynal-Querol (2005, p.812) suggests that the incidence of civil wars is more strongly influenced by polarization than by fractionalization. Since polarization seems to imply the sort of potential antagonism which characterizes Lijphart’s “plural” societies, arguably it would make sense to use polarization data to construct the list of cases for this book’s quantitative analysis. MAR data have been used for this purpose instead because comparable polarization data are not available for the sort of large sample of countries analyzed here. Daniel Posner (2004, p.849) provides a versionof ELF that focuses on politically relevant groups but data for it currently is available for sub-Saharan Africa. Since consociational elements are not commonly used throughout the world, maximizing the number of cases for this book must be prioritized if its statistical findings are to be of value. Alberto Alesina (2003, p.178) also analyze data indicating polarization but the unavailability of adequate data indicating distances between groups necessitated the treatment of all such distances as equal within their dataset. Compared with the MAR dataset, which indicates the existence of “Minorities at Risk,” this polarization data would be less helpful for identifying cases of “plural” societies containing potentially antagonistic groups because it does not seem to enable recognition of the more hostile intergroup relationships. The decision to use the MAR dataset to identify cases for this book is confirmed through reference to the precise types of available data constructed to represent fractionalization, ethnolinguistic fractionalization, and domestic polarization.