of violence but not restrictions. Davenport argues that each combination allows government authorities distinct sets of costs and benefits. For example, “restrictions without violence allow government officials to regulate behavior without provoking the negative ramifications associated with state-sponsored violence,” and “violence without restrictions allows authorities to eliminate challengers but avoids the administration, monitoring, and pretense of legality commonly affiliated with civil liberties restrictions” (49). Davenport’s empirical analysis reveals that some factors known to influence repression generally affect certain categories of repression in opposing directions. For example, conflict and development/modernization decrease the probability of a state’s resort to less lethal forms of repressive action, yet these same factors increase the probability of a resort to more lethal repression. He also demonstrates that democracy generally increases the probability that a regime will employ less lethal forms of repressive activity while decreasing the probability that it will employ more lethal methods. As I demonstrate in the measurement section below, I am not confident that the current measures of repression allow us to capture firmly the distinction between lethal and nonlethal dimensions.
Davenport’s debate reflects an earlier debate among scholars who have examined personal integrity abuse and disputed whether this singular category of repression is itself multidimensional or unidimensional. Mitchell and McCormick (1988) have argued that the tools of repression within this subset are substantively distinct, and thus while “arbitrary political imprisonment was certainly reprehensible, resort to torture and killing was a distinct, and qualitatively worse, activity” (484). Poe and Tate (1994) disagreed, claiming that it could “be persuasively argued that the two dimensions postulated by Mitchell and McCormick stem, in reality, from the one dimension that Stohl and his colleagues tap [in their Political Terror Scale]—that both torture/killing and imprisonment are rooted in a regime’s willingness to repress its citizens when they are considered a threat” (855). McCormick and Mitchell (1997) disagree with Poe and Tate, contending instead that “human rights violations differ in type not just amount, such that they cannot be clearly represented on a single scale,” and in particular they argue that the use of imprisonment and the use of torture and killing are “quite different types of government activity, with differing consequences for the victims, differing use of governmental resources and capabilities, and differing costs for the government, both domestically and internationally” (514). They argue that governments or their agents engage in a calculation of costs and benefits when choosing among the tools of repression. However, their empirical analyses demonstrate only thin differences in predictor variables across two types of personal integrity abuse (imprisonment and torture). With one exception, the differences were tied to the level of achieved statistical significance, and most of these measures were ones that have performed thinly or inconsistently in other analyses that cover a longer period; McCormick and Mitchell’s analysis covers only the years 1984 and 1987. The one exception they find, civil war, has been one of the two predictors that have consistently achieved the highest level of statistical and substantive significance in personal integrity models, and civil war fails to achieve statistical significance in only one of the years the authors studied. Poe, Tate, and Keith (1999) note that the similarities Mitchell and McCormick find are “impressive given the differences in design: their use of two dimensions of repression as dependent variables, their not using a lagged dependent variable (as we had), and the necessity of their using smaller (cross-sectional) samples of countries [which] led to much larger standard errors in their reanalysis, making statistical significance more difficult to achieve, a factor that accounts for several of their divergent findings” (299). Poe, Tate, and Keith also raised theoretical concerns about analyzing the components of repression separately, because doing so does not take into account that the behaviors are substitutable policy options, and the fact that the choice to use one tool may either prevent or render unnecessary the use of the other (see Most and Starr 1989, 97–132). For example, killing one’s political opponents eliminates the need to imprison them. Disappearance may also be substitutable for imprisonment and may in many cases actually hide murder. I continue to conceptualize repression of personal integrity rights as a unidimensional phenomenon. As we will see below, Cingranelli and Richards’ (1999b) Mokken scaling of their nine-point physical integrity rights index strongly suggests that this subset of rights is indeed unidimensional.
Measuring Repression
Social scientists seeking to measure state repression or coercive action empirically face significant measurement issues similar to those that challenge all large-N cross-national studies. Mitchell et al. (1986) provide an accurate summary of the challenge: data sets must “(1) have broad coverage across countries and time, (2) be based on multiple sources, (3) be reliable and valid, (4) have intensiveness (depth of coverage), (5) have extensiveness of coverage (multiple indicators), and (6) be sensitive to differences across countries and time” (22). Measurements must capture the scope (the degree of potential harm inflicted), the intensity (frequency and concentration of occurrence), and the range (size and type of target population) of the repression (14–16). The two earliest popular measurements of political repression, Taylor and Jodice’s (1983) State Coercive Behavior (part of The World Handbook of Political and Social Science Indicators) and Freedom House’s Civil Liberties Index, represent significant data-collection efforts, but each fails to some degree to meet these criteria, as I will demonstrate below. Most empirical studies to date, however, have primarily addressed the more severe form of repression, violations of personal integrity (imprisonment, torture, killing, and disappearances) (for example, Poe and Tate 1994; Poe, Tate, and Keith 1999; Cingranelli and Richards 1999a, 1999b; Keith 2002a), and have tended to employ the dominant indicators in the field: the Political Terror Scale (PTS) or the (CIRI) physical integrity rights measures, both of which I believe come much closer to meeting the criteria set forth by Mitchell et al. (1986) than do the two early measures of civil liberties restrictions. The Political Terror Scale was originally developed by Stohl and others (Stohl and Carleton 1985; Gibney and Stohl 1988; Poe 1992; Gibney and Dalton 1996) and is maintained by Mark Gibney (Gibney 2011).
The two standards-based PTS measures are based on assessments contained in the yearly human rights country reports published by Amnesty International and the U.S. Department of State in regard to the occurrence of political imprisonment, execution, disappearances, and torture. The ordered indices range from 1 to 5. The coding categories and their criteria are:
(1) Countries [are] under a secure rule of law, people are not imprisoned for their views, and torture is rare or exceptional…. Political murders are extremely rare.
(2) There is a limited amount of imprisonment for nonviolent activity. However, few persons are affected, torture and beating are exceptional…. Political murder is rare.
(3) There is extensive political imprisonment, or a recent history of such imprisonment. Execution or other political murders and brutality may be common. Unlimited detention, with or without trial, for political views is accepted.
(4) The practices of level 31 are expanded to larger numbers. Murders, disappearances are a common part of life…. In spite of its generality, on this level terror affects primarily those who interest themselves in politics or ideas.
(5) The terrors of level 41 have been expanded to the whole population…. The leaders of these societies place no limits on the means or thoroughness with which they pursue personal or ideological goals. (Gastil 1980, 37, as quoted in Carleton and Stohl 1985, 212–13)1
While the two measures are highly correlated (.81 in the data set here), the country coverage varies. The country coverage in the Department of State-based measure reflects the universal set of independent states, except in a couple of the early years under study here.2 The measure based on Amnesty International reports is less complete because, as a human rights NGO, its mission has been to respond to human rights abuse, and thus it has tended to focus on states with problematic human rights records (Poe and Tate 1994, 869; Poe, Carey, and Vazquez 2001, 655–56). Specifically, the Department of State–based measure tends to cover 20 to 40 more states than the one based on the Amnesty International reports. In previous work we wanted to take advantage of having two measures based on different reports of human rights, so we dealt with this issue of uneven country coverage by substituting Department of State scores for the missing scores in