CHAPTER 2: WHEN SURVIVORS SPEAK UP: A QUANTITATIVE
3. Data and Methods
To address the puzzle of what motivates governments to promise and pay reparations, I introduce a new dataset that I have built using original data on reparations. It was necessary to gather my own data because pre-existing datasets that include information on reparations do not distinguish between promises and payments, do not include data on victim group strength, utilize different inclusion criteria, and, with the exception of Powers and Proctor (2015), they do not
focus specifically on reparations. As a result, there has been very little quantitative work on reparations, and the work that does exist has neither tried nor been able to assess the interplay between victim-group characteristics and reparations decisions. This means that I am
constructing not only an original theory, but original counterarguments to that theory, as well. The literature simply does not examine the phenomenon I am investigating in my dissertation, and therefore one important contribution of this dissertation is that it provides an explanation as to why governments promise and pay reparations promises. Given that this theory will lay a foundation for future work on the political, social, and economic dynamics of reparations, I spend the bulk of my time focusing on my theory about governments’ reparations decisions rather than developing and discounting alternative explanations to that theory.
The novelty of this research also means that in order to assess what factors influence a government’s likelihood of promising and paying reparations, I needed to gather my own data on reparations promises, reparations payments, and the independent variable that is the key to my theory: the strength of victims’ organizations. My dataset includes information on the reparations promises and payments made in response to dictatorships and internal conflicts in Europe,
Central Asia, and Latin America that occurred between 1939 and 2006 and resulted in
widespread, systematic human rights abuses. My dataset includes an important fraction of the universe of positive cases: 107 cases from Europe (29 post-conflict cases and 78 post-
dictatorship cases), 20 from Central Asia (12 post-conflict cases and 8 post-dictatorship cases), and 35 from Latin America (21 post-conflict cases and 14 post-dictatorship cases). I selected these regions because there is more documentation available on abuses committed in these regions, not only in terms of accessible government publications, legislation, and websites, but also in terms of scholarly work and articles published in the popular press. Furthermore, my
language skills in English, German, and Spanish enabled me to search both primary and secondary sources in these regions much more easily and thoroughly than I could for cases in Africa and Asia. The unit of analysis is the victim group in a given abusive episode.
For example, the dataset contains multiple cases of West German reparations promises and payments made to Holocaust victims: Jewish German Holocaust victims constitute one case, Romani German Holocaust victims form a second case, and German Jehovah’s Witnesses present a third case. All of these cases involve reparations paid to German citizens for egregious human rights abuses committed during the same time period by the same German government, but subsequent West German regimes did not promise and pay reparations to these groups on the same schedule. West German Chancellor Konrad Adenauer promised reparations to Jewish Germans in 1951, and the West German government began paying these reparations in 1953. German Jehovah’s Witnesses, in contrast, were theoretically legally eligible for reparations under the same 1953 law that initiated reparations payments to Jewish Germans, but they did not receive reparations payments until 1997.
The West German case provides an opening to discuss the way I coded successor states, as well. Technically, the West German government did not abuse its citizens during WWII, because West Germany did not exist. However, West Germany became the legal successor state to Nazi Germany, and so the West German government inherited all of Nazi Germany’s
reparations obligations. Similarly, both the Czech Republic and the Slovak Republic are legal successors to Czechoslovakia, and so they both inherited Czechoslovakia’s Communist-era and WWII-occupation-era reparations responsibilities. Thus, when these states were legally
established, they enter the dataset with the obligation to promise and pay reparations to victims of the crimes committed by their respective predecessor states.
I classified a country as a dictatorship when it was scored as not democratic for a given
time period on at least two of three regime classification scales that I consulted.16 I opted for this
coding scheme because regime classification scales do not always agree on when a country qualifies as a democracy. By coding countries as dictatorships only when multiple scales agreed that a country was in fact experiencing an autocratic period, my coding is robust to the
definitional peculiarities of any single classification scale. This also means that current autocratic countries can be (and are) included in the sample due to having failed to pay reparations for previous periods of internal conflict, abusive dictatorial regimes, or present abuses. Thus, my analysis is not just of democratic countries, but of countries of all regime types that have
experienced either an internal conflict or dictatorship in the past. Currently abusive governments are included in the analysis and are coded for having promised/not promised and/or paid/not paid reparations not just for the abuses that are ongoing, but also for abuses that were committed by
previous governments.
As another hypothetical example, let us imagine that a country experienced abuses at the hands of an autocratic regime, transitioned into democracy, transitioned out of democracy, and acquired a non-democratic government that promised and paid reparations. This country would be included in the dataset for all of those time periods, insofar as coding victim group strength was possible for the years in question. It would also be classified as a post-autocratic case, because the abusive period in question was one of autocracy. This does not imply that the case is not currently autocratic (or, in the case of a post-conflict coding, that it is no longer in a state of violent internal upheaval); it simply identifies the abusive period for which reparations ought to
be paid as having been a period of autocracy rather than internal conflict. If this same country experienced an internal conflict at a different point in time, a separate case would be added in the dataset and coded as post-conflict to track the reparations promises and payments made for that specific abusive episode.
The internal conflicts consist of cases in the Correlates of War Dataset included on the
list of intra-state wars in which the government was a participant17 and cases in the UCDP/PRIO
Armed Conflict Dataset of state governments fighting internal armed conflicts against one or more domestic opposition groups (Gleditsch et al. 2002; Sarkees and Wayman 2010). I also included cases in which the CIRI data and Political Terror Scale averages out to 3 or more, indicating that citizens are frequently subjected to government violence and political
imprisonment, and where the CIRI Human Rights Dataset records a government as having no respect for citizens’ rights to freedom from disappearance, extrajudicial killing, imprisonment, and/or torture (Cingranelli et al. 2014; Gibney et al. 2017).
When coding reparations promises, I looked for laws, peace agreements, news reports, and official statements in which the government guaranteed citizens that reparations would be paid for abuses committed in a certain dictatorship or conflict. To compile this information, I consulted the UCDP Peace Agreements database, the Notre Dame Peace Accords Matrix, the University of Ulster Transitional Justice Peace Agreements Database, and the UN Peacemaker Database, as well as countries’ own databases of past legislation, domestic and international news outlets, field reports from human rights organizations such as Human Rights Watch and Amnesty International, and books and scholarly articles about each case. I coded this dummy
variable as 1 when there was evidence of a reparations promise and as 0 when there was no sign that the government had made an official reparations promise.
I collected the information on reparations payments from similar sources. In addition to reading primary and secondary literature on reparations and on each conflict and dictatorship included in the dataset, I consulted national budget reports, government websites, reports on domestic reparations programs, statements from victims’ organizations, news articles, and reports from organizations such as the International Center for Transitional Justice, Amnesty International, Human Rights Watch, and the United States Institute of Peace. For each case, I searched for data in both English and the applicable local language. I did not code reparations as paid unless I found documentation confirming that at least a subset of the victims had received some sort of governmental aid that qualifies as reparations.
Reparations can take many forms besides the traditional cash payment model. These run from measures that are employed almost universally, such as educational vouchers and
healthcare, to less common methods, such as free public transportation and admission to state- run museums. Reparations can also be communal rather than individual. Peru has employed communal reparations in addition to individual financial compensation and healthcare measures, as the government funds local development projects in towns where the entire community was affected by the Armed Internal Conflict. Communal reparations can also be made in the form of symbolic reparations. In contrast to material reparations, which come in the form of money, goods, and services, symbolic reparations are much more abstract. Examples of widely-used forms of symbolic reparations include state-sponsored memorials, changes to educational curricula, and government-funded initiatives to locate, exhume, and identify victims’ remains.
Although individual, communal, material, and symbolic reparations are all valid forms of reparations, I constrain my analysis to individual material reparations. From a theoretical
standpoint, there is good reason to believe that the political dynamics behind symbolic reparations may be too different from those of material reparations to permit meaningful comparison. First of all, symbolic and communal reparations generally impose much smaller financial costs on governments than material reparations. Consequently, the government’s resistance to reparations claims and the intensity of victims’ political struggle for acquiring these types of reparations, particularly symbolic reparations, is likely to be greatly diminished.
Second of all, projects that politicians like to refer to as being symbolic and communal reparations may not actually be reparations at all. Regardless of a politicians’ assertions, should we view a government as having made symbolic reparations when it adds a paragraph about historical crimes to a textbook when the next scheduled textbook revision date rolls around, builds a commemorative statue without survivors’ input, or publicly acknowledges the site of a mass grave? Communal reparations encounter similar difficulties in terms of validity. Although some of Peru’s communal reparations development projects have been genuinely reparative, many were excuses for politicians to say they were paying reparations when actually they were completing infrastructure and development projects that had already been scheduled. Thus, due to the heightened political dynamics behind individual material reparations, as well as their concrete and easily measurable nature (which makes data collection both more feasible and more reliable), I have opted to exclude symbolic and communal reparations from my quantitative analysis.
Some of the greatest obstacles and greatest rewards in this project center on the data that I have gathered on the characteristics and behavior of victims’ groups’ efforts to acquire
reparations. Measuring the strength of victims’ organizations presents a genuine challenge, both due to the difficulty of deciding precisely what a measurement of ‘strength’ entails and the scarcity of data on characteristics of victim groups. Consequently, I look to work in the fields of transitional justice and social movements to assist me in creating a coding scheme for victim group strength.
I start with the model given by Conor O’Dwyer (2012) in his study of gay rights activists in post-communist Poland. He evaluates the robustness of activist networks by assessing three characteristics. The first of these is the activist network’s density, or the number of active
organizations involved in the cause. The more groups that are involved, the larger the movement and the better able it is to pursue pro-LGBT+ efforts in multiple arenas. The second
characteristic is the coordination between groups in the activist network; that is, to what extent are different groups in the movement cooperative or competitive with each other in terms of resources, goals, and projects. The third and final characteristic is the “capacity to engage in political lobbying,” which at the low end of the spectrum simply requires groups to define themselves as political, whereas highly robust groups could file lawsuits, train and field electoral candidates, and be closely involved in writing relevant laws (15).
Next, I look to Hugo van der Merwe and Maya Schkolne’s (2017) discussion of civil society and transitional justice to be sure that, when I looked at the transitional justice landscape of a given country, I searched for all types of relevant groups and advocacy efforts. Van der Merwe and Schkolne list eight types of civil society actors that are generally engaged in transitional justice work: Religious organizations, human rights NGOs, peacebuilding NGOs, psycho/medical NGOs, gender justice NGOs, community-based organizations/victim
they define eight ways that these organizations can pursue transitional justice: “mobilizing action; targeted advocacy; monitoring and transparency; official support; public engagement; service provision and victim support; peace building, reconciliation and development; and truth telling, commemoration and memorialization” (229).
The same sources with information on reparations promises and payments in a particular case often had information on the extent to which victims’ organizations existed, were pushing for reparations payments from the government, were collaborating with other civil society organizations, and were attracting attention and sympathy from society in general. Although sometimes non-governmental organizations or scholars would explicitly refer to victims’
organizations as being unified, vocal, and decisive in acquiring reparations, my coding decisions more often required triangulating the information I could find about the number of victims’ organizations that existed at the time (O’Dwyer’s density criterion), victims’ organizations’ self- reports of their activities, goals, and collaborative efforts, both present and historical (O’Dwyer’s coordination criterion and capacity for political action criterion); contemporary news reports, and outside assessments about the strength and reach of the human rights sector of civil society (all three of O’Dwyer’s criteria). I also examined to what extent the victims’ rights movement contained organizations in van der Merwe and Schkolne’s eight categories of civil society and how involved they were in pursuing the eight types of pro-transitional justice work listed in van der Merwe and Schkolne’s typology. I then used this information to score the strength of
victims’ rights organizations, focusing on how much pro-reparations pressure they placed on the government via their network density, coordination, and political lobbying.
Due to the scarcity of such data, it was impossible to create an explicit typology, and the coding is necessarily comparative rather than rigidly defined. Although some of the cases in
question have generated an extensive scholarly literature—for example, the case of German Jews after the Holocaust, the case of political victims of Pinochet’s dictatorship in Chile, and the case of political victims of Argentina’s military junta—most cases received much less attention. Given that reparations did not start becoming an international norm until fairly recently, it was particularly difficult to find data on the historical cases. However, consulting sources in local languages usually yielded enough data for me to make an informed assessment of the state of victim groups in those contexts. When data was too sparse to be reliable, I dropped the case from my analysis.
Another difficult element of this project was collecting data on victim groups for each year from the end of the abusive period up through the first year that reparations were paid. Although part of the power of the analysis lies in its cross-temporal nature, which allows me to identify what factors are present or absent in promise and payment years as opposed to years where no reparations activity occurs, finding victim group data for a span of thirty to forty years was not always possible. I was usually able to find data on the state of victims’ groups during the first few years after the abuses and for three to four years directly preceding reparations promises and payments, but, depending on how long it took for a government to promise and/or pay reparations, I could not always gather data on the intervening years. Luckily, civil society groups—in this case, victim’s groups—are unlikely to drastically gather or lose strength from one year to the next. Thus, I was able to use the data from well-documented years to extrapolate the victim strength for the two to three years before and after in cases where those preceding and subsequent years were not as well documented. This approach helped to fill in some, though not all, of the gaps in the data.
For all cases with sufficient amounts of data on victim group strength, I scored this strength on a scale from 0-5, with 0 indicating that victims’ rights groups were nonexistent and 5 indicating that they existed, comprised a large number of people (density), were unified in their pursuit of reparations (coordination), and concertedly pressured the government for reparations (capacity for political action). A score of 1 might indicate that, while victims’ organizations existed in this year, they did not express any desire for reparations, or it could mean that victims’ organizations were requesting reparations, but they were so small, disorganized, and/or plagued by infighting as to be inconsequential. When victims’ organizations existed and desired
reparations but were disorganized, small in number, and/or internally divided, I scored their strength as a 2. To score as a 3, victims’ movements needed to comprise more than one
organization with a moderately-sized membership (approximately 50 or more members), most of these organizations needed to agree with each other about wanting reparations even if they disagreed about what kinds of reparations should be paid, and they needed to be organized