In conjunction with a project on state capacity and repression, I’ve recently been exploring the specific allegation data from Courtenay Conrad and Will Moore‘s Ill-Treatment & Torture (ITT) Data Collection Project. ITT’s specific allegation data are interestingly different from existing human rights data sets like the one (full disclosure) I co-direct, the CIRI Human Rights Data Project. Whereas CIRI provides a yearly country score based on a standards-based ordinal coding of the level of government respect for its citizens’ right to be free from torture and ill-treatment, the ITT data provide information on all allegations of government torture or ill-treatment made by Amesty International. As such, the unit of observation for the ITT data is the individual Amnesty International torture allegation. Further, for each allegation, the ITT specific allegation data contain information about the type of abuse, the number of victims, the agents responsible for carrying out the abusive act(s), and the victim type, as well as many other topics. For more information, the codebook can be found HERE, and a forthcoming Journal of Peace Research article introducing the data, written by Courtenay Conrad, Jillienne Haglund, and Will Moore, can be found HERE.
Figure 1. Total AI Torture Allegations, 1995 – 2005
As is my wont, I decided to get to know the data by making a few maps. In doing so, I stumbled on some potentially interesting patterns. First, the map above shows the total number of torture allegations that Amnesty International made against each state over the time period from 1995 to 2005, i.e. the time period covered by the ITT data. As one might suspect given the research on Amnesty International’s information efforts (e.g. Ron, Ramos, and Rodgers 2005; Hendrix and Wong 2013), the number of allegations does not neatly match up with the level of torture in each state, as AI’s strategic considerations lead it to focus its attention more on some states rather than others. This problem can be partly remedied by taking into account the approximate size of the affected population in each allegation, which is provided by ITT. However, as Conrad, Haglund, and Moore point out, due to the strategic bias in Amnesty’s reporting as well as the impossibility of observing all torture events, even this method is likely to result in a biased undercount of the number of torture events that actually took place in a given country over this time period. There are several ways of attempting to deal with coverage bias in such measures; indeed, Patrick Brandt and John Beieler each recently posted on ways to deal with similar issues in GDELT. For the purposes of this blog post, the maps below will simply display the various attributes of torture allegations as a proportion of the total allegations made against the state. With that in mind, you may want to remember to refer to the map above regularly if some states seem to perform particularly poorly in one area or another, as it could be the result of having been the target of relatively few allegations overall (for instance, see Mongolia).
Figure 2. Type of Torture as a Proportion of Total AI Torture Allegations, 1995 – 2005
(Click the image to enlarge)
The series of maps displayed in Figure 2 show the proportion of total AI torture allegations made up by the four types of abuse distinguished in ITT: scarring torture, stealth torture, ill-treatment, and unstated torture. It should be noted that abuses are not mutually exclusive; that is, a single AI allegation may contain information on several types of abuse. As such, the proportions for any individual state will often not sum to 1. Overall, it appears that stealth torture, i.e. torture that is more difficult to detect after the fact because it leaves no scars on the body, is, on average, less common than the other types of torture and ill-treatment. Indeed, the average state received only about 22 allegations of stealth torture from 1995-2005, whereas the average state was accused of ill-treatment approximately 48 times, scarring torture about 56 times, and unstated torture around 39 times.
Matching up with work done by Darius Rejali, highly developed states, and particularly highly developed democracies, appear to be more likely to engage in higher proportions of stealth torture than states at lower levels of development. The United States, Canada, Norway, Sweden, Finland, Belgium, Chile, South Africa, and South Korea have all been the targets of fairly high proportions of stealth torture allegations. That said, stealth torture also appears to be relatively more common among Asian states, particularly China, Kazakhstan, Uzbekistan, Myanmar (Burma), Laos, and Vietnam. On the other hand, scarring torture makes up a much lower proportion of allegations in these states. Instead, the proportion of allegations made up by scarring torture is much higher in many Sub-Saharan African states, like Angola, Mozambique, Central African Republic, Niger and Ghana, as well as in Eastern European states, such as Bulgaria, Romania, Hungary, and Slovakia. Finally, it seems we know the least about the type of torture used in North African and Middle Eastern states, with unstated torture accounting for a large proportion of allegations in Algeria, Libya, Egypt, Saudi Arabia, Syria, and Iran.
Figure 3. Agency of Control as a Proportion of Total AI Torture Allegations, 1995 – 2005
(Click the image to enlarge)
The ITT data also provide information on the agency of control, i.e. the “domestic institution and/or agent(s) … responsible for a given allegation of torture” (Conrad and Moore 2011: 12). Figure 3 maps the various agencies of control as a proportion of each state’s allegations. As with the various types of torture and ill-treatment, agencies of control are not mutually exclusive; a single torture allegation may name several government agencies as responsible for engaging in abuse.
Across Figure 3, a few interesting patterns stand out. While Canada, Australia, and European states are the targets of a relatively moderate number of torture allegations overall, it would appear that the majority of those allegations target torture and ill-treatment carried out by the police. On the other hand, prison abuse appears to account for a relatively high proportion of allegations in the United States, China, and Brazil; unsurprisingly, these states rank 1st, 2nd, and 4th, respectively, in total prison populations. The map of immigration abuse shows that these allegations primarily apply to highly developed democracies, like the U.S., Germany, Italy, Australia, South Korea, and Japan, possibly due to the high demand for admittance into those countries. Abuse by intelligence agents appears to account for a relatively high proportion of allegations in Middle Eastern and North African states, like Egypt, Jordan, Israel, Yemen, and the United Arab Emirates. Intelligence abuse also accounts for a high proportion of allegations in South Korea and Mongolia, although those states were subject to relatively few allegations overall. Finally, the states for which we see high proportions of military and paramilitary abuse align fairly well with the states highlighted in Figure 4 below, which shows the states in which armed conflicts were fought between 1995 and 2005 according to the PRIO Conflict Site data developed by Johann Hallberg.
Figure 4. Locations of Armed Conflict, 1995 – 2005
(Conflict Sites in Red)
Of course, all of the patterns that I see in these maps could disappear upon controlling for other factors, and there are surely many patterns that I have missed. Primarily, I wanted to draw attention to the wealth of information present in the ITT data. I have only focused on the number of allegations, the type of abuse, and the agency of control; the ITT also contains information about the identities and numbers of victims, whether the abuse occurred within the state or abroad, whether the abuse was investigated, and many other factors. Overall, these data should open the door to many studies of phenomena that we previously could not investigate crossnationally, and I hope we see more projects like the ITT in the future.
*NOTE: All maps in this post were made using Weidmann, Kuse, and Gleditsch’s CShapes GIS Vector Dataset, Version 0.4-2, in ArcMap 10.0.