Dynamic spatial panel estimates of war contagion

• Date: 01 January 2018
• DOI: http://doi.org/10.1111/twec.12522
• Author: Fabrizio Carmignani,Parvinder Kler
• Published date: 01 January 2018

ORIGINAL ARTICLE

Dynamic spatial panel estimates of war contagion

Fabrizio Carmignani

|

Parvinder Kler

Griffith University, Brisbane, Queensland, Australia

1

|

INTRODUCTION

A cursory look at the chronology of wars in the post-World War II era reveals three facts.

1

First,

war is (still) a frequent occurrence worldwide. The average country in the world has been at war

(be that an internal civil conflict or an international war) for approximately 12% of the time since

1960. This is equivalent to a cumulative total of almost a thousand years of war across the globe

in the last six decades. Second, conflicts tend to cluster geographically. The unconditional proba-

bility that a country is at war when at least one of its neighbours is also at war (excluding cases

when two neighbours are fighting against each other) is close to 0.7. Third, the “at war”or “at

peace”status of a country is persistent over time, albeit not necessarily permanent. With the excep-

tion of a few countries that have been at peace without interruption since 1960, most nations have

experienced at least one transition from peace to war or vice versa. On average, the unconditional

probability that the status of a country in the current year is the same as its status in the previous

year is 0.9. This unconditional probability that the status in the current year is the same as the sta-

tus 10 years ago is 0.8. Our purpose is to bring these three facts together to understand the spatio-

temporal dynamics of war prevalence.

2

There is already a voluminous body of research on the determinants of war, with several papers

that specifically focus on war prevalence (see, inter alia, Elbadawi & Sambanis, 2002; Reynal-

Querol, 2002; Miguel, Satyanath, & Sergenti, 2004; Besley & Persson, 2008; Nunn & Qian, 2012;

Cunningham, 2013; Couttenier & Soubeyran, 2014). Most papers in this line of work deal with

either the temporal dimension (i.e., the relationship between current status and previous status) or

the spatial dimension (i.e., the spillover of war from a country to its neighbours). However, there

is very little research that models the two dimensions jointly (see Franzese, Hays, & Cook, 2016),

possibly because of the methodological difficulties involved in the estimation of equations that

include both temporal and spatial lags of the dependent variable. Our first contribution is therefore

to explore possible approaches to the estimation of a spatio-temporal model of war prevalence. We

identify an approach that is econometrically sound and computationally accessible given the type

1

Here we refer to the chronology available from UCDP/PRIO data set (see Gleditsch, Wallensteen, Eriksson, Sollenberg, &

Strand, 2002).

2

Prevalence is the amount of war that is likely to be observed at a given time. Prevalence therefore arises from the combina-

tion of war initiation (onset) and war duration. A large number of existing studies focus on either onset or duration. How-

ever, as noted by Elbadawi and Sambanis (2002), understanding prevalence is equally important in a policy perspective. In

the literature, the word “incidence”is occasionally used as a synonym of prevalence.

DOI: 10.1111/twec.12522

126

|

©2017 John Wiley & Sons Ltd wileyonlinelibrary.com/journal/twec World Econ. 2018;41:126–142.

of panel data that are currently available for research on armed conflicts. To operat ionalise this

approach, we construct a non-conventional measure of war prevalence. We then report new evi-

dence on contagion effects accounting for the autoregressive nature of war prevalence and for the

potential endogeneity of other control variables. As a comparison, we also report (in an Appendix)

estimates obtained from alternative estimation approaches, including standard models where the

dependent variable takes a more conventional form.

In empirical terms, the notion of war encompasses several different situations. First of all,

wars differ in terms of their intensity, for instance measured by the number of battle related

deaths. Second, and perhaps more important, there are two fundamentally different types of war:

internal and interstate. An internal conflict occurs between the government of a state and one or

more internal opposition groups (i.e., a civil war). An interstate conflict instead occurs between

two or more states.

3

Most existing papers focus exclusively on internal conflict since this has

become much more frequent than international conflict after the end of World War II (see Blatt-

man & Miguel, 2010). However, it is not always clear as to when a conflict can be clearly sta-

ted to be internal or interstate and may in fact be part of a regional war complex, or indeed be

subject to potential errors in definition (Gersovitz & Kriger, 2013). Our second contribution is to

look at a comprehensive sample of wars and see how results vary depending on intensity and

type.

Our results can be summarised as follows. First, we do not reject the hypothesis that both

temporal and spatial effects significantly impact upon the prevalence of war . Second, we find

evidence of a significant “spatio-temporal”interaction effect only when excluding contempora-

neous spatial lag of war. Third, after separating war by type and intensity, both temporal and

spatial effects remain significant, with stronger impacts on high-intensity conflict. Fourth, the

spatial effect is stronger in magnitude for interstate wars whereas the time effect is stronger

for civil wars.

The rest of this paper is set up as follows. Section 2 discusses various estimation strategies.

Section 3 presents in more detail our preferred approach and the data/variables used. Section 4

contains results and discussions, followed by concluding remarks in Section 5. The Appendix

reports additional results obtained from alternative estimation approaches and model specifications.

2

|

ECONOMETRIC ESTIMATION OF A DYNAMIC MODEL

WITH SPATIAL EFFECTS

We are interested in the estimation of the following general model:

yt¼qyt1þdðWytÞþXtbþet:(1)

The model is written in vector form for a cross-section of observations at time t. Therefore y

t

is a

N91 vector consisting of one observation of the dependent variable for every generic country

i(i=1, ....., N) at time t(t=1,....,T), X

t

is an N9Kmatrix of control variables (Kbeing the

total number of control variables), Wis a N9Nnon-negative matrix of known constants

3

The UCDP/PRIO codebook (http://www.pcr.uu.se/digitalAssets/124/c_124920-l_1-k_codebook_ucdp_prio-armed-conflict-da

taset-v4_2013.pdf) separately identifies two other types of war: (i) internationalised internal conflict, which occurs when a

foreign state intervenes on one side of a civil war and (ii) extrasystemic conflict, which occurs between a state and a non-

state group outside its own territory (e.g., a colonial war).

CARMIGNANI AND KLER

|

127