Cargo cult science, armchair empiricism and the idea of violent conflict

Abstract

In this paper I develop a philosophical and methodological critique of ‘rationalist’—or ‘social scientific’—approaches to the study of civil war and violent conflict, especially the work of Paul Collier and David Laitin. Rationalist scholars purport to develop universal explanations for the outbreak and the protracted duration of violent conflict using econometric techniques and rational choice reasoning. My critique of their ‘scientific method’ can be summarised in two propositions: first, the scientific approach is considered to be a ‘cargo cult’ science—the cult being universal law-like causalities underpinning social phenomena. Second, most empirical research in this tradition is based on doubtful statistics enriched with anecdotal evidence rather than by empirical field work. Hence, rationalist scholars largely conduct an ‘armchair empiricism’ instead of immersing themselves in the complex nature of social mechanisms in a specific space-time context. I briefly sketch out an alternative approach based on critical realism and how such a research programme could work in practice for studying the political economy of violent conflicts.

Notes

Valuable comments from Konrad Hagedorn, Martin Petrick, Dan Richter and Christine Schenk on earlier versions of this essay are gratefully acknowledged. I would also like to thank Alexander Austin for discussing with me some of the philosophical issues upon which this essay builds long before it was actually written. The usual disclaimer applies.

1 The purpose of this article is not to discard the work of rationalist scholars per se, since their studies have provided some important insights into the study of violent conflict. What I want to criticise is their epistemological standpoint and their lack of willingness to concede other approaches in empirical enquiry as complementary and standing on equal footing. They tend to advocate a hierarchy, where ‘narratives’ may be useful as anecdotal data to formulate hypotheses, which then need to be subjected to rigorous scientific testing.

2 Cf. Tambiah (1986: 20 – 27; 1992: 100).

3 This argument surely overlooks the fact that those triggering rebellion on the Tamil side originated mostly from Jaffna. These were youngsters who had lost hope of acquiring what they desired: white-collar employment that only the state sector could offer on a large scale. The Sri Lankan government undermined these desires with its university and state sector admission policies. This means that those already in the state sector were not the ones who suffered most, but rather those who were not yet in it. Hence the language issue played an important role in the discourses of Tamil grievances.

4 Recent anthropological research in Sri Lanka has emphasised the ‘hybrid identities’ of its political and social elites and the diversity of interests, cultural patterns and religious identities within the main ethnic communities (Jeganathan & Ismail 1995; Silva, 2002; Tiruchelvam & Dattathreya, 1998). In the advance of the political conflict between Sinhalese and Tamils, each constituency tended to hide away the contradictions within its own community and to emphasise the dividing lines between the ethnic groups in order to define their own ‘ethnic’ identity.

5 See, among others, Peebles (1990), Peiris (1991) on the land issue. More detailed accounts of the history, structural causes and perceptions of the ‘ethnic’ conflict and civil war in Sri Lanka can be found in Spencer (1990); Rotberg (1999); Mayer et al (2003).

6 Similarly, we can interpret the Sinhalese youth uprisings (jvp riots) in 1971 and 1987 – 88 as being triggered by internal inequalities and perceived marginalisation of an educated youth in the rural areas which did not have a fair chance of reaping the benefits from a university graduation with a white-collar job (Mayer et al, 2003).

7 However, there have been serious doubts even within economics over whether or not econometrics is up to standard in providing an instrument for hypothesis testing in the sense of critical rationalism. Petrick (2004), after reviewing a number of critiques from within the economics field (Brandes, 1989; McCloskey, 1985; Schor, 1991), concludes that they ‘expected far too much of econometrics to be the fundamental benchmark for the falsification of theories … the methodological standpoint of critical rationalism becomes untenable for economics. Furthermore, even if the demand on econometric analysis is more modest, a good deal of openness and honesty is required to make any results credible’ (Petrick, 2004: 11, original emphasis). See also Hausman (1994) and Lawson (1997).

8 Several versions of the Collier – Hoeffler model were circulated as unpublished manuscripts, working papers of the World Bank, etc long before their eventual publication. See, for overviews of the different versions and their subsequent alterations, Cramer (2002) and Ross (2004).

9 Initially, the focus was on explaining the outbreak of civil wars, while later studies also looked into what causes the protracted duration of civil wars, and the dynamics of warlordism. See in this research tradition (or paradigm) the work of Azam and Hoeffler (2002), Berdal and Malone (2000), Fearon and Laitin (2003), Skaperdas (2002), Collier et al (2004) and Fearon (2004). See the special issues of the Journal of Peace Research, 39 (4) and 41 (2), Journal of Conflict Resolution, 46 (1) and 49 (4) and, featuring some studies more critical of the greed – grievance debate, the Journal of International Development 15 (4).

10 It needs to be noted that Collier has meanwhile given up his initial strong claims about the role of grievances (Collier et al, 2003), and other scholars in the rationalist and quantitative tradition have helped to differentiate the various components of greed and grievance factors and how these relate to sociopolitical dynamics that trigger violence or sustain it. Some authors have challenged the statistical validity of the original hypothesis, especially the link between primary commodities and the incidence of civil war (Fearon, 2005; Lujala et al, 2005; Ross 2004). However, the general research strategy remains the same—quantitative cross-country statistics that ‘test’ alternative propositions or use different datasets.

11 Cited in Austin (2004: 146) from J Fitzgerald, ‘Coolth—contemporary cargo cults’, 1999, at http://www.coolth.com/cargo.htm

12 Weber (1991).

13 The latter term is used in analogy to Brandes' (1989) criticism of the research practice in agricultural economics which he labelled ‘armchair economics’.

14 While positivist scholars may acknowledge this point, I argue that the methodological approach of falsification via statistical inference focuses on the first part (If A, then B), but there are no clear methodological guidelines to develop the explanations of the process that produces B from A. Often there is a large gap between statistical data analysis and the conceptual explanations, which seem frequently to be ad hoc.

15 It is, in fact, exactly this point that King et al (1994) conflate when they argue for their case study methodology, because they think in representative survey design. They are concerned about statistical generalisation and validity even for small numbers of cases. It is exactly here where the fundamental difference between positivistic concepts and critical realism should become clear. The key point is that cases are not ‘sampling units’ and should not be chosen for this reason (Yin, 1994). Analytic generalisation means that one uses a previously developed theory as a template with which to compare the empirical results of the study.

16 He illustrates this with reference to positivistic survey research, which is constrained by context effects. An interview is a social context, embedded in other contexts, ‘all of which lend meaning to and are independent of the question itself’ (Burawoy, 1998: 12). One can standardise the question, but not the respondent's interpretation of the question. In the rationalist studies of violent conflict, researchers seem to escape these context effects, because they largely rely on country-level statistical data sources. However, the continuing debates even within the rationalist tradition about the quality of data (Cramer, 2002; Mack, 2002; Ron, 2005; Ross, 2004) underline the contextual differences in data compilation, as well as the social and political construction of specific sorts of data by governments and scientists.

17 See also footnote 10 on this, where some references are given for studies that challenge the original empirical claims of the Collier – Hoeffler model.