This Land Is My Land! Large-Scale Land Acquisitions and Conflict Events in Sub-Saharan Africa

ABSTRACT

The paper provides evidence that large-scale land acquisitions in Sub-Saharan Africa raise the likelihood of experiencing outbursts of organized violence. This evidence is achieved through a spatially disaggregated approach, to account for local characteristics, and a quasi-experimental research design, implemented to overcome limitations due to missing geo-referenced information about land deals. Domestic acquisitions are particularly significant in explaining organized violence outbreak, suggesting that the national concentration of power among economic elites hinder social stability. Further, the institutional framework, which forms the background for LSLAs, is a prominent determinant of organized violence occurrence. In particular, more egalitarian and inclusive societies are less prone to conflicts, even in the presence of LSLAs. When land deals are finalized to the cultivation of agricultural commodities for biofuels, the effect is less evident and violence partially depends on climate conditions. On the other hand, land deals still in implementation stage seem to narrow organized violence, possibly due to both local control of areas under investments and socio-economic effects of the deals. Results also show the existence of significant spatio-temporal dependence path, since events of organized violence tend to be recurrent and to persist in space, feeding ‘neighbouring’ effects of proximity and local patterns of violence concentration.

KEYWORDS:

• Large-scale land acquisitions
• agricultural investments
• natural resources
• conflict events
• Africa

JEL CLASSIFICATION:

• F21
• O13
• Q15
• Q34
• R14

Acknowledgments

Earlier draft of this paper has been presented at the 16 Jan Tinbergen European Peace Science Conference, Università Cattolica del Sacro Cuore (Milan, IT), on 20–22 June 2016; at the International Peace Science Society workshop on Environmental Security, University of Notre Dame (South Bend, USA), on 20 October 2016; the Land and Poverty Conference, the World Bank (Washington, USA), on 20–24 March 2017; and the 3rd FLARE Meeting, Stockholm University (Sweden), on 29 September to 2 October. We warmly thank participants for their comments and observations. We would like to thank Simona Beretta, Jampel Dell’Angelo, Jonas Nordkvelle, Kerstin Nolte, the editor and the anonymous referees for their valuable comments and observations. The findings, interpretations and views expressed in this paper are entirely those of the authors; equal authorship implied.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. The definition of ‘unused’ land is highly debated: shepherds, nomadic herdsmen and indigenous communities often traditionally access lands without clearly recognized rights, due to a common lack of rights definition, especially for these groups. The term ‘unused land’ might therefore hide traditional uses relevant for social dynamics.

2. As result, large-scale land acquisitions are generally labelled as ‘land grabbing’ events and part of a new ‘global land rush’ (Anseeuw et al. 2013).

3. It does not imply that organized violence is necessarily oriented to land issues (and thus violent events should be understood as land conflicts only); rather, this study is meant at assessing whether LSLAs might contribute in fuelling broader dynamics of violence.

4. Data are accessible to http://landmatrix.org/en/.

5. In October 2016, an updated version of Land Matrix was released: in particular, doubtful cases (with inadequate information) and double entries were removed; and some deals were enriched by characteristics previously missing. However, the number of large acquisitions in Sub-Saharan Africa with geo-referenced data did not substantially change. (see Figure 1)

6. Paired t-tests (or Mann-Whitney test for non-normal distributions) confirm that the two sub-samples are not statistically different at standard level of significance.

7. Members of the Land Matrix coalition provide the best available data at local level through media reports, company resources, research papers and policy reports, government sources and personal information collected on the field.

8. For complete model of analysis, the Vuong test confirms that a zero-inflated negative binomial specification better predicts our response variable than a standard negative binomial specification (z = 1.37; Pr>z = 0.0846).

9. Same logic for the number of total fatalities applies.

10. One-sided Events show some spatial correlation as well (Moran’s I test coefficient: 0.053; p-value:0.000).

11. The ‘geography’ of this paper is composed by many ‘islands’ (as they are called in the spatial econometrics jargon): namely areas selected by specific criteria (in our case, individual cells where LSLAs are reported plus contiguous neighbouring cells) with a reduced contiguity between them. This fragmentation substantially reduces the scope of an explicit modelling of spatial dependence.

12. Relying on the best estimate provided, 80 cells record at least one conflict-related death. In other words, once transformed in a binary variable, Events and Total_deaths do not perfectly overlap. As expected, also the number of conflict-related fatalities shows some grade of spatial autocorrelation (coefficient: 0.041; p-value: 0.000).

13. Although this measure of organized violence is less characterized by spatial autocorrelation, a dependency pattern can still be detected (coefficient: 0.014; p-value: 0.000).

14. The term agrofuels refers to any agricultural commodity – such as sugar cane, maize, wheat, barley, palm oil and soybeans – produced with the intention of producing bioethanol or biodiesel.

15. Coefficient of correlation: 0.6352, significant at 1%.

16. In our sample, domestic deals occurred in Angola, Cameroon, Nigeria, Tanzania, Uganda and Zambia.

17. This covariate is labelled Minerals in the text and tables, whereas oil or natural gas deposits are labelled Hydrocarbons.

18. A rainy season is defined as the three consecutive months in which it on average rained the most during a year in any cell.

19. Results are available upon request. We also verified the effect of including the foodcrops_deal variable as robustness check. It bears similar sign and significance level, as expected; and diagnostic for different models are substantially identical.

20. For hydrocarbons, the same relation holds referring to the occurrence of conflict events also.

21. As further robustness check, we applied a probit estimation strategy, achieving very similar results in magnitude, sign and significance. Given such consistency, we gave preference to the logit estimation to keep logical coherency between models, since the link function applied in the ZINB estimations is a logit indeed. We also tested whether modelling the spatial dependency found in the dependent variables would have affected the outcomes. Thus, we run all model specifications by spatial probit estimations with spatial autocorrelation of the dependent variable (through MLE). The effects of the spatial autocorrelation generally appear significant. Above all, domestic LSLAs remain a powerful predictor of organized violence likelihood at the local level. Results available upon request.

22. It should be noted that the particular ‘geography’ of this paper – resulting from the location of LSLAs and, indirectly, by their neighbouring contiguous cells – does not include, almost by definition, broad territories in several countries, thus potentially reducing ex-ante the effect of minerals deposits on violent events.