3.3 Chapter Summary
4.1.1 Sampling Strategy and Data Sources
I test my argument on a sample of all cases of interim government that followed at least one year of intrastate armed conflict since 1989 and that terminated in elections by 31 December 2012.32 In line with the vast majority of existing research, I operationalize intrastate armed conflict as a contested incompatibility between a government and at least one rebel group “where the use of armed force ... results in at least 25 battle-related deaths in one calendar year” (cf. Chapter 1 and Wallensteen and Sollenberg, 2001).
Four theoretical and methodological issues motivate my choice for this obser- vation period. Firstly, and concerning the left margin of 1989 as a delineation of this period, the end of the Cold War enabled international engagement in war-torn states to a degree that ever more included the reform of state insti- tutions and the promotion of democracy. This empirical development closely links to the role of interim governments as instruments for peacebuilding, not least because such governments often convene to prepare war-torn states for elections (cf. Paris, 2004; Sisk, 1993). Secondly, intrastate conflicts and their resolution attempts before 1989 often took the form of proxy wars, were strongly influenced by relations between the superpowers, and thus followed different dy- namics compared to internal conflicts in the post-Cold War period (cf. Kalyvas and Balcells, 2010). Thirdly, many data collections on conflict characteristics start with information collected from 1989 onwards, so my choice also concerns issues of data availability. Fourthly, because I want to understand the effects of interim governments on the stability of post-interim peace, only analyzing interim governments that terminated by December 2012 allows me to assess this duration aspect of my dependent variable for at least two consecutive years (as of June 2016, conflict data is available until December 2014).33
This strategy results in a sample of 62 instances of interim government, fully listed in Table A.1 in the Appendix. To visualize the distribution of interim gov- ernments across the world, Figure 4.1 additionally plots these governments on a map that shows how interim governments are chiefly an African and Asian phe-
32Because I do not include any interim governments that enter my study already in place
in 1989, I circumvent the issue of left truncation.
33I argued in Chapter 1 that I understand interim governments as institutions that terminate
in elections. The only exception I make in my data set is Sudan’s interim government that ended in a popular referendum in 2011, and that I keep in my sample because of its influence in the qualitative debate on power-sharing interim government (e.g. Johnson, 2008; Zambakari, 2013). I run a robustness check in which I exclude the case from the sample. If an interim government convened between 1989 and 2012 but did not hold elections by December 2012; or if it convened but did not hold elections within 15 years, I drop a respective case from my sample. This concerns three cases of interim rule: (1) Afghanistan’s 1992 Peshawar Accords transferred power to an interim council, but elections never took place as the Taliban seized power in 1996; (2) the May 1993 Memorandum of Settlement for India’s Bodoland conflict called for an Interim Bodoland Executive Council, but elections scheduled for November 1993 never transpired; and (3) the interim government set up following a military coup in Mali in 2012 only ended in elections in 2013, well after my December 2012 cut-off point.
Figure 4.1: Map of Interim Governments in the World
Notes: Map based on the data described in this chapter. Countries that experienced interim governments following intrastate armed conflict are shaded in dark gray.
nomenon, reflecting the predominance of internal conflicts on these continents (cf. Pettersson and Wallensteen, 2015). It is important to note that previ- ous studies on interim government – as well as more generally on institutions in war-torn societies – have often used sampling strategies that only consider institutional configurations negotiated in peace agreements (e.g. Hartzell and Hoddie, 2003; Jarstad, 2010; Lyons, 2005), while I do not restrict my empir- ical analysis to this criterion. Firstly, an important empirical argument for extending my sample beyond cases of peace agreements is that a more inclusive sampling strategy does not select against some of the four interim government models as put forward by the Shain and Linz (1995) typology; a typology that as I have demonstrated in Chapter 2 forms the conceptual basis of most present- day interim government research. For instance, it is possible to imagine that restricting my analysis to interim governments after peace agreements would likely result in a biased sample where power-sharing governments comprised of the signatory parties to an agreement are over-represented. Secondly, this strat- egy would also overlook prominent cases of interim government that were not called for in peace agreements but that have strongly influenced the qualitative academic debate on the topic, such as the 2001-04 Afghan Interim Authority (AIA), or the rule of UNTAET in East Timor from 1999 to 2001.
To delineate a sample of interim governments that goes beyond peace agree- ments, I relied on a number of existing data sets as well as on information from qualitative data sources. In a first step, I identified all intrastate con- flict episodes using the Armed Conflict Dataset provided by the Uppsala Con- flict Data Program (UCDP) and the Peace Research Institute Oslo (PRIO), v. 4/2014 (Gleditsch, Wallensteen, et al., 2002; Pettersson and Wallensteen, 2015).
Figure 4.2: The Duration of Interim Governments 0 5 10 15 0 1 2 3 4 5 6 7 8
Interim government duration (in years)
No . of inter im go v er nments
I then coded whether an interim government convened after a case experienced at least one year of intrastate conflict according to the UCDP/PRIO data, consulting four existing data sets. Firstly, I referred to the Polity IV Annual
Time-Series Dataset that captures regime patterns and changes in all sovereign
states with a population of over 500,000 inhabitants since 1800. Polity IV also reports “transition periods,” defined as phases in which new institutions are planned and implemented (Marshall et al., 2014). Secondly, I relied on the Au-
thoritarian Regimes Dataset that includes a measure for “transitional regimes,”
defined as temporary institutions with the purpose to realize a transition (Hade- nius and Teorell, 2007; Wahman et al., 2013). Thirdly, I consulted the National Elections across Democracy and Autocracy Dataset (NELDA) that offers data on all elections from 1960 to 2006 and that captures if a country was “ruled by ‘transitional leadership’ tasked with ‘holding elections”’ (Hyde and Marinov, 2012). Fourthly, Jarstad (2010) presents the Post-Accord Elections (PAE) data set on power-sharing interim rule between the signing of peace accords and first post-accord elections (1989-2004).
Three aspects explain why it proved insufficient to only rely on these data sets. Firstly, my theoretical definition of what constitutes as interim government does not translate one-to-one to the conceptualizations of these data sets that are often more restrictive in their definition of an interim government. For instance, I have in Chapter 1 defined interim governments as the institutions with executive and legislative power between a disintegration of an old regime
Table 4.1: Sampling and Coding Sources
Source Period
Polity IV Dataset 1989-2014
Authoritarian Regimes Dataset 1989-2010
NELDA Dataset 1989-2006
PAE Dataset 1989-2004
UCDP Conflict Narratives 1989-2014 Africa Yearbook (Afrika Jahrbuch) 1989-2014
Asian Survey 1989-2014
BTI Reports 2003-2014
Freedom House Reports 1998-2014 Economist Intelligence Unit 1996-2014
and first elections. Wahman et al. (2013) however use a further coding rule and only consider interim governments lasting fewer than three years, which would inter alia exclude Burundi’s 2000-05 interim government from my sample. The case of Burundi has yet inspired a vast number of qualitative analyses on power- sharing interim government in particular (e.g. Curtis, 2007; Lemarchand, 1994). Figure 4.2, depicting interim government duration by length in years, moreover shows that it is not unusual for interim governments in war-torn societies to last longer than three years. Secondly, many of the cited data sets select against some of Shain and Linz (1995)’s four models of interim government. For instance, although it is not explicitly stated in their codebook, Hyde and Marinov (2012) do not include interim periods with high degrees of international authority in the NELDA data set, such as the rule of UNTAET.
Thirdly, only using information provided by existing data sets does not al- low me to consider if interim government was really the result of armed con- flict and meant as an instrument for its resolution. For instance, according to Hyde and Marinov (2012), Bangladesh has had repeated interim governments since the early 2000s, and UCDP/PRIO data report an internal conflict in Bangladesh in 2005-06, thus simply matching these two data sets would have included Bangladesh as a case of post-conflict interim government in my sample. But caretaker interim governments are formalized in Bangladesh’s constitution and appear regularly before elections to create a political environment in which voting can take place without influence by an outgoing regime. Similarly, Hyde and Marinov (2012) code an interim government in Thailand in 2006-07, a period in which Thailand also saw ongoing conflict. But fighting took place be- tween the government and secessionist insurgents in southern provinces, whereas the interim government was not put in place to solve this conflict over territory, but resulted from a coup d’état against then-Prime Minister Shinawatra.
To deal with these issues of ambiguity, I complemented the data taken from the existing regime data sets with qualitative information to delineate my sam- ple – and, in a second step, to code my independent variables (cf. below). I used
sources that are comparable over time and cases. These include, firstly, Ber- telsmann Transformation Index (BTI) country reports that biannually assess political transformation in independent countries with a population of over two million inhabitants since 2003, except consolidated democracies (Bertelsmann Foundation, 2014). All BTI reports include standardized sections on issues such as institutional performance, power to govern, and elections. Secondly, Free- dom House’s “Freedom in the World” country reports compare political rights and civil liberties in all independent countries and in selected disputed terri- tories since 1998 and address issues such as the nature of government or the role of political opposition and civil society (Freedom House, 2014). Thirdly, conflict narratives that also review the institutions put in place for conflict res- olution are available for every intrastate conflict reported in the UCDP/PRIO Armed Conflict Dataset in the UCDP “Conflict Encyclopedia” (Uppsala Conflict Data Program, 2014). Fourthly, country reports of the Economist Intelligence Unit (2014) provide detailed accounts on political and economic developments worldwide since 1996, sometimes on a monthly basis, although reports vary in frequency across countries and time. Finally, for cases in Sub-Sahara Africa, I also used the annual Africa Yearbook (“Afrika Jahrbuch” until 2003) that covers political developments in the region (Mehler et al., 2014); and for cases in Asia, I relied on the bimonthly academic journal Asian Survey published by the University of California Press, which covers contemporary politics in South, Southeast, and East Asia since 1961. Table 4.1 displays that while more in- formation is available for more recent cases of interim government, all years in my observation period are covered by several independent sources (“period” describes the portion of the observation period covered by each source).