Social Indicators Research

, Volume 117, Issue 3, pp 1069–1087

Respect for Civil Liberties During the Third Wave of Democratization: Presenting a New Dataset

Article

DOI: 10.1007/s11205-013-0391-y

Cite this article as:
Møller, J. & Skaaning, SE. Soc Indic Res (2014) 117: 1069. doi:10.1007/s11205-013-0391-y

Abstract

The literature on state repression has increased voluminously in recent decades. However, First Amendment-type civil liberties have not received much attention compared to neighbouring freedoms such as electoral rights and physical integrity rights. This neglect is arguably related to the lack of high-quality disaggregated measures. In this paper, we present a new dataset on respect for civil liberties—the Civil Liberties Dataset (CLD)—which includes indicators on freedom of expression, freedom of assembly and association, freedom of religion, and freedom of movement for 207 countries in the period 1976–2010. An assessment of the levels, development, sequencing, and correlates of civil liberties not only reveals a number of interesting patterns, it also shows that the CLD has a competitive edge vis-à-vis extant measures. We conclude that the CLD opens new avenues for research on state repression in general and civil liberties in particular.

Keywords

Civil liberties Dataset Measurement validity Developments Sequencing Correlates 

1 Introduction

In recent decades we have seen an explosion in studies of democratization and human rights developments. However, the increase in empirical analyses of electoral rights and physical integrity rights1 (see Berg-Schlosser 2007; Munck 2011; Carey et al. 2010; Davenport 2007a) has not been matched by similar attention to First Amendment-type civil liberties (Davenport 2007a: 1). These liberties have mostly been treated as parts of other overarching concepts, such as liberal democracy, and when they have in fact received separate attention it has been in the guise of a composite concept of civil liberties. Although we have relatively clear indications about the general development in the repression of First Amendment-type civil liberties compared with the development in the repression of electoral rights and physical integrity rights (e.g., Davenport 2004; Diamond 1999), we therefore know little about the particular developments in repression levels for the different kinds of civil liberties, i.e., the freedoms of expression, assembly, association, religion, and movement. Furthermore, we do not have systematic evidence showing whether strong patterns exist in the sequencing of these civil liberties. This situation is unfortunate considering the widespread violations of civil liberties, the consistency with which they have been championed by social movements in modern times, and their prominent position in liberal political theory, international human rights conventions, and national constitutions (Berlin 2002; Ishay 2004; Keith 2002; Hathaway 2002).

The lack of separate attention to the specific civil liberties is at least in part a consequence of the lack of disaggregated, cross-temporal data. Most extant measures of civil liberties—such as Freedom House’s Freedom in the World survey2—do not make fine-grained, disaggregated data publically available and often they are characterized by a very limited coverage of countries, years, and/or rights3 (Skaaning 2010). In this article, we introduce a new dataset on government respect for civil liberties (the Civil Liberty Dataset, henceforth CLD), which covers four First Amendment-type rights: freedom of expression, freedom of assembly and association, freedom of religion, and freedom of movement. The scope of the dataset is all independent countries (2074) in the period 1976–2010.

Considering the measurement and inter-coder reliability issues endemic to social science debates (e.g., Adcock and Collier 2001; Munck and Verkuilen 2002; Munck 2009), the availability of alternative measures is almost always valuable for the research community. However, we explicitly use the one extant dataset that actually contains disaggregated indicators of several civil liberties—the CIRI Human Rights Database (see Cingranelli and Richards 2010)—to show what is gained by the availability of CLD. First, the CLD indicators only correlate moderately with the CIRI, which indicates that they capture something partly different. Second, the CLD indicators perform at least as well as the CIRI equivalents with regard to scalability and different types of validation. Third, the CLD provides a larger coverage in terms of country-years and somewhat more fine-grained distinctions.

An added value of our interrogation of the CLD is the identification of a number of interesting empirical patterns. We demonstrate that the more political civil liberties (i.e., freedom of expression and freedom of assembly and association) are generally repressed to a higher extent than the more private civil liberties (i.e., freedom of religion and freedom of movement). Furthermore, based on a Mokken scale analysis we identify a four-step empirical hierarchy of the following form: freedom of expression is generally repressed at least as much as freedom of association and assembly, which is generally repressed at least much as freedom of movement, while freedom of religion is only rarely repressed more than any of the other liberties. In addition, we reveal some notable temporal dynamics. The developments in global repression levels of the four civil liberties are characterized by a partial convergence over time, especially because respect for freedom of expression and freedom of assembly and association increase remarkably at the end of the Cold War. However, for all four kinds of civil liberties the post-1989 increase is smaller than that for electoral rights, corroborating the notion of a liberal deficit of recent democratization processes. Finally, variables measuring the regime form, the level of modernization, dominant religion (Protestant or Muslim), oil production, and the end of the Cold War are strong predictors of a composite civil liberties index. These findings demonstrate both the soundness of the CLD and some of the potential which the dataset offers for future research on civil liberties.

2 The Civil Liberty Dataset

The CLD includes measures of civil liberties “understood as certain freedoms to perform actions that individuals might wish to perform, which (it is thought) the state should not restrict” (Waldron 2003: 195). These freedoms are also known as First Amendment-type rights (Goldstein 1978: 30–31; Davenport 2007a: 2). More particularly, the CLD includes indicators on freedom of expression, freedom of assembly and association, freedom of religion, and freedom of movement for 207 countries in the period 1976–2010. In the creation of the CLD, only the actual practices of states and their agents have been taken into account in the assignment of scores. Accordingly, neither merely formal-legal guarantees nor citizens’ missing abilities to make use of the freedoms due to lack of initiative, commitment, financial means, etc. have influenced the coding of the following civil liberties:
  • Freedom of expression To what extent do citizens, groups, and the press have the right to hold views freely and to seek, obtain, and pass on information on political issues broadly understood without being subject to actual limitations or restrictions?

  • Freedom of assembly and association To what extent do citizens have the right to gather freely and carry out peaceful demonstrations as well as to join, form, and participate with other persons in political parties, cultural organizations, trade unions, or the like of their choice without being subject to actual limitations or restrictions?

  • Freedom of religion To what extent do citizens have the right to have and change religion or belief of own choice and alone or in community manifest their religion or belief in practice, worship, observance, and teaching in private or public as well as proselytize peacefully without being subject to actual limitations or restrictions?

  • Freedom of movement and residence To what extent do citizens have the right to settle and travel within their country as well as to leave and return to their country of own choice without being subject to actual limitations or restrictions?

The source for the coding is the US State Department’s Country Reports on Human Rights Practices,5 a set of annual country reports that provide systematized information on violations of civil liberties and other human rights over several decades. After the first reports, issued in the mid-1970s, were met by criticism, the scope, quality, and independence of the information improved significantly (McNitt 1988: 97–98; Innes 1992), and the “results of careful, critical examinations over the years … tend to agree that the annual State Department Reports are an invaluable source accurately reporting on the conditions of most of the countries most of the time” (Poe et al. 2001: 651). Each of the four indicators has been coded based on a four-point scale.6 The points denote situations where the respective civil liberties are severely restricted (1), fairly restricted (2), modestly restricted (3), and not restricted (4). The four points are anchored in an overall distinction between respective ideal typical characteristics of liberal, semi-liberal, illiberal, and anti-liberal regimes (cf. Merkel and Croissant 2000; Linz 2000 [1975]), with the two intermediate categories inserted symmetrically between the endpoints.7

How does the CLD compare with the two most widely employed measures of civil liberties, provided by Freedom House (FH) and David Cingranelli and David Richards (CIRI)? The civil liberties index constructed by Freedom House suffers from a number of shortcomings8 (Munck 2009; Coppedge et al. 2011; Skaaning 2010; Giannone 2010), most notably a blatant lack of disaggregate scores for the years 1972–2004. With respect to measuring First Amendment-type rights this is a crucial weakness because the composite index bundles together an incoherent set of freedoms, such as academic freedom, free trade unions, independent judiciary, protection from political terror, the right to establish private businesses, and the absence of economic exploitation. No similar problems burden the impressive CIRI Human Rights Dataset, which includes indicators for no less than seventeen different human rights in the period 1981–2010. Among these are five indicators for First-Amendment type rights: freedom of speech, freedom of assembly and association, freedom of foreign movement, freedom of domestic movement, and freedom of religion. In terms of disaggregated indicators, the CIRI dataset is obviously superior to previous datasets. But with respect to the actual scoring of countries it has several limitations compared to our dataset. First, in the case of freedom of speech and freedom of assembly and association, scores for 33 small countries are only available for the years after 2000.9 Second, inter-coder reliability tests have solely been performed on the coding of 1 year (i.e., 2004). Third, the CIRI indicators are only measured on three-point scales.

To illustrate, Cingranelli and Richards (2008) assign scores to their freedom of speech indicator by determining whether “Government censorship and/or ownership of the media (including radio, TV, Internet, and/or domestic news agencies)” is (0) Complete, (1) Some, or (2) None. One should make the critical observation that censorship and ownership of the press is at most a subset of the freedom of speech. But what concerns us most here is the relatively big jump from ‘some’ censorship to ‘complete’ censorship, which means that the CIRI coding rules are unable to tease out important differences among countries with low levels of freedom of speech. Similar problems affect the coding rules for “Citizens’ rights to freedom of assembly and association”. The three-point scale here distinguishes between (0) Severely restricted or denied completely to all citizens, (1) Limited for all citizens or severely restricted or denied for select groups, and (2) Virtually unrestricted and freely enjoyed by practically all citizens. The code of 0 once again collapses interesting differences, such as whether the freedom of assembly and association is very or next to completely restricted. Furthermore, the distinctions between no censorship, some censorship, and complete censorship and between restricted, limited, or virtually unrestricted freedom of association and assembly do not seem to be anchored in more general theoretical distinctions. They rather tend to be of a more pragmatic nature, which is also reflected in the fact that the linguistic qualifiers used to denote the different scores differ from right to right, something that makes it problematical to carry out comparisons of absolute repression levels across the different liberties.

3 Co-Variation with Alternative Indicators

Before examining the extent to which the differences in definitions and coding procedures lead to different rankings, we first report the correlations among the four CLD indicators: the two ‘political’ liberties of the freedom of expression and freedom of assembly and association and the two more ‘private’ liberties of freedom of religion and freedom of movement. As illustrated in Table 1, the correlations within each of these categories is higher than that between the categories, a disjunction we return to when investigating the construct validity of the CLD below. More generally, the pairwise correlation coefficients between the four CLD indicators reported in Table 1 are high but not on a level where it makes no analytical sense to distinguish between them.
Table 1

Correlation between civil liberty indicators

 

Freedom of expression (CLD)

Freedom of speech (CIRI)

Freedom of assembly and association (CLD)

Freedom of assembly and association (CIRI)

Freedom of religion (CLD)

Freedom of religion (CIRI)

Freedom of movement (CLD)

Freedom of domestic movement (CIRI)

Freedom of speech (CIRI)

0.71

       

Freedom of assembly and association (CLD)

0.77

0.64

      

Freedom of assembly and association (CIRI)

0.69

0.61

0.79

     

Freedom of religion (CLD)

0.55

0.50

0.54

0.53

    

Freedom of religion (CIRI)

0.49

0.45

0.49

0.48

0.72

   

Freedom of movement (CLD)

0.64

0.55

0.62

0.57

0.54

0.47

  

Freedom of domestic movement (CIRI)

0.57

0.51

0.58

0.58

0.52

0.49

0.66

 

Freedom of foreign movement (CIRI)

0.48

0.44

0.48

0.44

0.41

0.37

0.61

0.51

Freedom of expression and belief (FH)

0.67

0.64

      

Freedom of expression (BTI)

0.65

0.63

      

Associational and organizational rights (FH)

  

0.68

0.54

    

Association/assembly rights (BTI)

  

0.66

0.54

    

Freedom of religion (Marshall)

    

−0.72

−0.67

  

Entries are correlations coefficients (Kendall’s tau-b), pairwise deletion

This is as one would expect, but to what extent are the aforementioned similarities and differences between the CLD and CIRI datasets reflected in the co-variation between, on the one hand, the respective indicators and, on the other hand, between each of these and alternative indicators based on expert evaluations? Regarding the former question, we expect the correlations between the CLD indicators and their counterparts among the CIRI indicators to be relatively high as they are meant to measure the same civil liberties and as the scores are based on the same source, although not extremely high because the coding schemes differ somewhat. It turns out that the CLD and CIRI indicators correlate in the range of 0.61–0.79 (see Table 1). We find the lowest pairwise correlations between the freedom of movement indicators, which was to be expected since CIRI provides separate indicators for domestic and foreign movement. The highest correlation is found between the freedom of assembly and association indicators.

These pairwise correlations are a little lower than what we would have expected. Needless to say, the correlations are of course not low in an absolute sense but compared to pairwise correlations between neighboring concepts such as measures of democracy (cf. Casper and Tufis 2003), they are relatively low. This observation holds, a fortiori, when considering that the CLD and CIRI indicators are thematically identical and measured using the same source (the annual Country Reports on Human Rights Practices) whereas the democracy measures differ substantially both in terms of defining attributes and data sources. This indicates that the CLD and CIRI datasets capture different aspects of the empirical variation in civil liberties10 and that the addition of the CLD is therefore not superfluous.

Indeed, these differences raise the question why the two sets of indicators are so relatively far apart? To push at this, we have identified the country-years where we find the biggest discrepancy (above 40 points on indices ranging from 0 to 100) between a composite CLD-index and a parallel CIRI-index, which we present below. As Table 2 shows, Israel is the country that turns up most often (five out of the fourteen cases of extreme differences). In all five instances, the differences are caused by very low CIRI scores but moderately high CLD scores. A partial reason for this disjunction is perhaps that the CIRI seemingly includes the occupied territories on the West Bank and Gaza into their assessment of Israel. We contend that this clouds the relatively high respect for civil liberties in Israel proper and that it illustrates some problems with the CIRI-codings. More generally, we have revisited the Country Reports on Human Rights Practices for each of the country-years included in Table 2. In general, we feel confident in concluding that these descriptions lend more support to the CLD scores than to those of CIRI.
Table 2

List of the countries with the largest disjunction of CLD and CIRI scores

Country

Years

Difference

CLD-index

CIRI-index

Israel

2008

58.33

58.33

0

Congo, DR

2005

48.33

58.33

10

Israel

2009

48.33

58.33

10

Ukraine

1992

48.33

58.33

10

Yemen

2002

48.33

58.33

10

Israel

1981

45

75

30

Israel

1982

45

75

30

Israel

1984

45

75

30

Indonesia

1992

41.67

41.67

0

Colombia

2000

−41.67

58.33

100

Mozambique

1985

−41.67

8.33

50

Nigeria

1982

−41.67

58.33

100

Nigeria

1983

−41.67

58.33

100

Romania

1991

−41.67

58.33

100

Next, we have correlated the CLD and CIRI indicators with indicators based on expert assessments that are supposed to reflect respect for the same civil liberties but with much lower coverage in terms of years and/or countries. For the years 2005 onwards Freedom House (2013) has—as part of the Freedom in the World survey—released the yearly subcomponent scores for Freedom of expression and belief and Associational and organizational rights, and the Bertelsmann Transformation Index (2013) has provided disaggregate scores for Freedom of expression and Association/assembly rights on a biannual basis. As for freedom of religion, we use the Religious freedom scale (covering 2006) presented in Marshall (2007). From Table 1 we see that the CLD and CIRI indicators are generally highly correlated—the former somewhat more than the latter—with these alternative indicators, which rely on multiple sources and expert assessments.

4 Expectations About Levels, Sequencing, and Trends

We move on to examine the construct validity of the CLD and to compare the results with those that come out when substituting the CLD indicators with the CIRI indicators. To do so, we need clear expectations about repression levels across civil liberties and time (see Adcock and Collier 2001).

4.1 Levels

As mentioned above, the four civil liberties included in our dataset can be divided into the more political/public civil liberties of freedom of expression and freedom of assembly and association, on the one hand, and the more private/personal civil liberties of freedom of religion and freedom of movement on the other. Generally speaking, we expect the political liberties to be repressed more than the private liberties. This is based on the received consensus that political liberties threaten the powers that be more than more private liberties do. Indeed, the political liberties of freedom of expression and freedom of association and assembly are often treated as defining attributes of democracy (Dahl 1989; Beetham 1999; Diamond 1999; O’Donnell 2007), which reflects that the government and state agents in autocracies normally have more direct interest in curtailing these compared to less political liberties. This expectation is in agreement with the theoretical claim supported by numerous empirical findings that, as summarized by Davenport (2007a: 7),

… governing authorities should respond with repression to behavior that threatens the political system, [and] government personnel … When challenges to the status quo take place, authorities generally employ some form of repressive action to counter or eliminate the behavioral threat; in short, there appears to be a “Law of Coercive Responsiveness”.

In terms of the actual levels, this means that we expect the repression of freedom of religion and freedom movement to be lower, or at least not higher, than the repression of freedom of expression and freedom of assembly and association throughout the period under investigation.

4.2 Sequencing

The expectation concerning repression levels only implies the existence of a general hierarchy between the more political and the more private civil liberties. But we expect that there exists a more elaborate four-step hierarchy in the sequencing11 of civil liberties. Regarding the general distinction between the two clusters of rights, it follows from the reasoning presented above that we expect government respect for freedom of movement and the freedom of religion to be at least as high as for the freedom of expression and the freedom of assembly and association. However, in addition to that, we hold that it is likely that freedom of expression is violated before equivalent violations of freedom of assembly and association. Autocrats have always fought open criticism of their rule, and restrictions on freedom of expression often indicate that a government intends to assault other fundamental rights. To quote Gearon (2006: 129): “If truth is the first casualty of war, freedom of expression is the first target of the totalitarian”.

We also expect a stepwise sequence to be found between freedom of movement and freedom of religion as the latter is simply the least political of the four civil liberties. Tellingly, many right-wing authoritarian regimes have presented themselves as defenders of religion. And even left-wing totalitarian regimes such as the Soviet Union, where the dominant ideology condemned religious practice at large and arguably made up a substitute political religion (Maier 1996; Maier and Schäfer 1997), tended not to suppress freedom of religion more than any of the other freedoms. Against this background, we set out to analyze whether there is a manifest, four-level stepwise hierarchy (double monotonicity) in the fulfillment across the full spectrum of civil liberties. If identified, such a hierarchy would approximate the logic of a perfect simple order scale, that is, a Guttman scale (see Bailey 1994; Schuur 2003, 2011). Notice the qualifier ‘approximate’. Guttman scales are deterministic, and we instead examine the issue using its probabilistic sibling, viz., a Mokken scale analysis (Mokken 1971).

4.3 Trends

We expect the more general hierarchy between the two kinds of civil liberties to be in existence over the entire period covered by the CLD (1976–2010). However, we also expect that a certain convergence has occurred over time. We base this proposition about temporal developments on a dynamical factor pointed to in many contemporary studies of political regime change. The collapse of bipolarity in 1989–1991, following the breakdown of communist regimes and the demise of the Soviet Union, led to an international climate propitious to democratization (Carothers 1999; Levitsky and Way 2010; Boix 2011). This dynamical factor works both in terms of the general Zeitgeist (Linz and Stepan 1996) and the foreign policy of especially Western democracies (Levitsky and Way 2010). The consequence of this has been a genuine explosion in the global number of democracies piggybacking the onset of the post-1989 liberal world order (Berg-Schlosser 2009; Møller and Skaaning 2013a). But though a rising tide lifts all boats, we expect that the Western liberal hegemony have had some important asymmetrical effects across the attributes of electoral competition and the four kinds of civil liberties.

First, the average level of democracy scores is likely to have increased more than the general respect for any kinds of civil liberties because the spread of electoral competition has, as pointed out by a series of observers, only to a limited extent been accompanied by increased respect for civil liberties and the rule of law (see Zakaria 1997; Diamond 1999, 2002; O’Donnell 2007, 2010; Merkel et al. 2006; Møller and Skaaning 2011). This means that we should have witnessed a significantly higher decrease in average repression of electoral rights after the 1989–1991 juncture than in the average repression of civil liberties. Second, as democracies—even when first and foremost electoral—are likely to repress political liberties much less than autocracies (cf. Davenport 2007b: 488), we expect a partial convergence in the levels of repression across our two categories of civil liberties over time. While the freedom of expression and freedom of assembly and association are expected to be repressed more than the freedom of religion and freedom movement throughout the entire period, the magnitude of this difference should decrease after 1989–1991, with the onset of Western liberal hegemony. For this reason, we expect the hypothesized developments to be abrupt rather than gradual. This means that we expect jumps in both the electoral rights and civil liberties scores around this juncture, albeit jumps that are higher for electoral rights than for more political civil liberties and higher for this kind of liberties than for more private liberties.

5 Global Developments, 1979–2010

To what extent are the actual repression levels and the developments in repression levels in line with the expectations formulated above? Figure 1 depicts the global average scores for the individual CLD indicators, rescaled to range from 0 (lowest respect) to 100 (highest respect).
Fig. 1

Development in average respect for civil liberties and electoral rights (democracy), 1979–2010

The years 1976–1978, measures for which are part of our dataset, have been excluded due to the selective country coverage characterizing the very first years of the Country Reports on Human Rights Practices (see the description of the dataset above). Figure 1 shows that freedom of expression and freedom of assembly and association have, as expected, been violated more than freedom of religion and freedom of movement over the entire period. Moreover, Fig. 1 also exhibits a sudden increase in respect for, first and foremost, the political liberties at the end of the Cold War, meaning that the gap between the two sets of rights has decreased over time. Nonetheless, even after the spike in 1989–1991 the absolute differences have remained salient. To compare these trends with developments in the respect for electoral rights, we employ the democracy measure included in the Democracy-Dictatorship Dataset (DD) of Cheibub et al. (2010).12

Figure 1 shows that the 1989–1991 juncture produced an abrupt increase in the respect for electoral rights (here measured by the percentage of democratic countries), which was followed by more gradual improvements. Hence, electoral rights have experienced a higher relative jump than even the more ‘political’ civil liberties of the freedom of speech and assembly/association, which is in line with the widespread observation that many recent democratization processes have been characterized by a ‘liberal’ deficit. To further probe the nature of the changes, we have investigated if the mean score of the listed indicators in the post-Cold War period differ significantly from that of the previous period. The differences for both electoral rights and the civil liberties are indeed statistically significant. The magnitude of the differences follow the earlier described hierarchy as the increase is by far the largest with respect to the share of democracies (62 %), followed by freedom of assembly and association (40 %), freedom of expression (18 %), freedom of movement (12 %), and freedom of religion (2 %). These results also come out when substituting the CIRI indicators for the CLD indicators. The following magnitudes of difference can be reported for the CIRI-indicators: freedom of assembly and association (42 %), freedom of speech (17 %), freedom of domestic movement (14 %), freedom of religion (6 %), and freedom of foreign movement (1 %).

6 The Sequencing of Civil Liberties

Whereas differences in repression levels across different rights were clearly identifiable from the graphs in Fig. 1, we need to interrogate the data further to see if they also lend support to the existence of a more particular sequencing in the respect for civil liberties. In Table 3, we report the means for each of the four CLD-measures over the entire period of 1976–2010.13 A mean score of 1 (on the original scores) would indicate that the civil liberty in question was severely violated everywhere in all years. Conversely, a mean score of 4 would indicate that the civil liberty in question was not restricted at all across space and time. In the table, the indicators have been ordered according to their mean value from low to high.
Table 3

Mean, standard variation, and H coefficients for the CLD indicators, 1976–2010

 

Mean

Standard deviation

H coefficient (Loevinger’s)

Freedom of expression

2.53

0.963

0.82

Freedom of assembly and Association

2.69

1.126

0.80

Freedom of movement

3.23

0.810

0.75

Freedom of religion

3.38

0.783

0.70

H scale 0.769, N = 6,035

The ordering suggests that for a particular country-year, a high score on any indicator is matched by at least the same score on the indicators listed below this indicator in the table. Hence, the numbers suggest that, in most cases, governments respect the freedom of religion and the freedom of movement at least as much as the freedom of assembly and association, which they respect at least as much as the freedom of expression. In addition, the differences in the averages and in the standard deviations (i.e., the frequency distribution) indicate that if the indicators are unidimensional, one needs to apply a method that is able to capture systematic differences in ‘difficulty’. Mokken Scaling Analysis provides just such a way of testing the presence of a systematic hierarchy (Mokken 1971; Schuur 2011) as it “is a nonparametric probabilistic version of Guttman scaling” (Schuur 2003: 139), i.e., a probabilistic cumulative scale. The scaling analysis tests whether countries do indeed cross thresholds on the four civil liberties in an ordered/cumulative way, meaning that we should only rarely find instances of other sequences. It does so, in the first place, by producing an H coefficient which indicates the homogeneity/scalability of such a cumulative scale. This score is achieved by comparing the number of deviations from a Guttman scale with the expected number of deviations (based on a random distribution), given the hierarchy between indicators indicated by the mean values. Running the scaling analysis, we also obtain H coefficients for each of the scale items (the four civil liberties), which show the homogeneity of each item with the rest of the items in the scale.

Mokken’s (1971) rule of thumb says that H > 0.5 indicates a strong scale. In an attempt to scale physical integrity rights (i.e., disappearances, political killings, torture, and political imprisonment), Cingranelli and Richards (1999) find a total H value of 0.59 and H coefficients for the individual items between 0.55 and 0.64. The results reported in Table 3 show that the CLD data is characterized by a very high H scale coefficient (0.77), with the coefficient for the four scale items ranging from a low of 0.70 (freedom of religion) to a high of 0.82 (freedom of expression). Based on these H coefficients, the cumulative scale is strongly one-dimensional and—allowing for its probabilistic nature—characterized by a unique way to reach any combination of attributes if these are awarded a particular score. In other words, such a unidimensional scale contains information about both the level and sequence of government respect for civil liberties. The scale runs from a low of 4 (all civil liberties severely restricted) to a high of 16 (no restrictions on any of the four civil liberties). As already indicated, these aggregate scores tell us much more than that, say, Chile in 1990 (with a score of 13) ranks higher than, say, Burma in 2010 (with a score of 6). From the overall scale scores (the level of respect for civil liberties in general), we can (probabilistically) infer the particular level of respect of each of the four civil liberties. To exemplify, an aggregate score of 13 means that the country in question scores 4 on freedom of religion and 3 on the three other civil liberties.

In spite of the very high H coefficients, the established scale does not take the form of a perfect hierarchy, however. In Table 4, we report the actual sequence of government respect for the different civil liberties that finds most support in the data.
Table 4

Sequencing of respect for civil liberties: Mokken scale predictions of patterns

Score

Freedom of expression

Freedom of assembly and association

Freedom of movement

Freedom of religion

4

Severely restricted

Severely restricted

Severely restricted

Severely restricted

5

Severely restricted

Severely restricted

Severely restricted

Fairly restricted

6

Severely restricted

Severely restricted

Fairly restricted

Fairly restricted

7

Fairly restricted

Severely restricted

Fairly restricted

Fairly restricted

8

Fairly restricted

Severely restricted

Fairly restricted

Modestly restricted

9

Fairly restricted

Severely restricted

Modestly restricted

Modestly restricted

10

Fairly restricted

Fairly restricted

Modestly restricted

Modestly restricted

11

Fairly restricted

Modestly restricted

Modestly restricted

Modestly restricted

12

Fairly restricted

Modestly restricted

Modestly restricted

Unrestricted

13

Modestly restricted

Modestly restricted

Modestly restricted

Unrestricted

14

Modestly restricted

Modestly restricted

Unrestricted

Unrestricted

15

Modestly restricted

Unrestricted

Unrestricted

Unrestricted

16

Unrestricted

Unrestricted

Unrestricted

Unrestricted

The pattern indicates that for all values of the scale, none of the private liberties are repressed more than any of the political liberties, and—consistently across the scale—freedom of religion is not repressed more than freedom of movement. However, freedom of assembly and association changes from severely restricted to fairly restricted after the freedom of expression. Table 4 also reveals some additional exceptions to the stepwise hierarchical pattern as freedom of religion is predicted to be unrestricted when freedom of expression is still fairly restricted, and freedom of assembly and association only changes from severely restricted to fairly restricted at the scale value of 10, although this, according to the logic of the proposed simple order scale, should already happen with a combined score of 7. Taken together, the results thus indicate some noteworthy violations of a stepwise difficulty (double monotonicity) in the fulfillment across the full spectrum of civil liberties. In spite of these misfits, the very high coefficients of homogeneity (H) lend strong support to the scalability of the indicators. Based on the empirical patterns, it is thus possible to aggregate the civil liberties scores into a unidimensional scale, running from 4 to 16. This can be done because Mokken scaling uses the proper aggregation rule given the hierarchical nature of the data, as opposed to using aggregation procedures that assume similarity in means and standard deviations.

Re-running the Mokken scaling analysis with the CIRI indicators, we find the same general hierarchy between the political liberties and the more private liberties. However, we do not find nearly as systematical a stepwise hierarchy. This is also reflected in the H-coefficients for the CIRI indicators, which range between 0.58 and 0.69. The overall coefficient of scalability is 0.65.14 We take this not only to further corroborate that the two datasets are not identical but also as an indication of the higher construct validity of the CLD data considering the theoretical expectations about the existence of such a hierarchy.

7 Correlates of Civil Liberties

As a last step of assessment of the construct validity, we have identified a number of conditions, which—according to extant theory and previous empirical analysis—should be able to account for much of the variation in general respect for civil liberties. A short description of the explanatory logic of each of the five identified variables is followed by a short description of the respective operationalizations.
  • Democracy There is a broad consensus that democracies repress civil liberties to a lesser extent than autocracies (Davenport 2007b: 488). The core issue is legitimacy. Democracies have an ‘electoral’ legitimacy which autocracies do not, meaning that the latter are forced to resort to coercion to a higher extent (see Davenport 2007a: 10–11). Moreover, the vertical accountability created by elections tends to put restrictions on the incumbents’ exercise of power. We use the aforementioned DD constructed by Cheibub et al. (2010) to measure democracy as it provides a minimalist measure of democracy, which covers virtually the same country-years as CLD and, most importantly, as it is not based on definitional criteria and empirical indicators pertaining to civil liberties (cf. fn. 12 above).

  • Communism Repression of civil liberties is likely to be higher in communist regimes than in other regimes, including other kinds of autocracies. For totalitarian versions of communism, which entirely destroy pluralism, this is true by definition (Linz 2000 [1975]). However, even with non-totalitarian versions of communist regimes, we would expect higher repression levels because such regimes “give ideological and practical priority to the community (sometimes embodied in the state) over the individual” (Howard and Donnelly 1986: 808; see also Dallin and Breslauer 1970). Constructing a dummy for communist regimes, we rely on their self-reference as communist, the advantage of which is that—analogous with the point about using an electoral definition of democracy—repression of civil liberties is not part of the definition or operationalization.

  • Modernization Modernization theory has continuously been identified as a cause of democracy and civil liberties, (e.g., Inglehart and Welzel 2005; Poe et al. 1999). However, Huntington (1968) and O’Donnell (1973), among others, have suggested that the relationship is non-linear. More particularly, intermediate levels of modernization are likely to be associated with the highest levels of repression for two reasons. First, because rulers face significant demands for freedom from below but are able to answer with repression rather than concessions. Second, because the political institutions are too weak to channel popular demands which therefore stoke instability, create open conflict, and—as a consequence—are conducive to poor performance with regard to civil liberties. We operationalize the modernization variable using the index of socioeconomic modernization constructed by Teorell (2010).15 In order to be able to capture if the suggested relationship is non-linear, we also include a squared term of this variable.

  • Oil Rents The so-called ‘resource curse’ theory argues that large-scale dependency on the extraction of natural resources, particularly oil, has negative consequences for civil liberties (Ross 2012; DeMeritt and Young 2013). Relevant mechanism here include that oil-induced economic growth does not advance the cultural and social changes that spur popular demands for liberty, that oil revenues make it possible to create a repressive apparatus which can be used to keep a lid on opposition forces, and that oil rents short-circuit the historical connection between revenues and rights. To measure the variable oil rents, we employ the data on oil rents per capita (in 1,000$, logged) collected by Ross (2008).

  • Protestant Liberal democracy and fully developed notions of human (natural) rights first arose in Protestant societies (Clarke 1994; Witte 2007). This connection has been attributed to the Protestants being “less authoritarian, more congregational, participatory, and individualistic” (Lipset 1994: 5). Protestantism has also been emphasized as important for civil liberties outside of the countries in North-western Europe and their settler colonies. According to Woodberry (2012), Protestants proselytes have played an important role in the increased compliance with liberal rights in many developing countries by promoting and spreading doctrines of religious liberty, education, mass printing and newspapers, and civil society organizations. We have created a dummy variable to measure this variable, distinguishing countries with a Protestant plurality or majority population (1) from the rest (0).

  • Muslim Governments in societies with Islam as the predominant religion have been shown to violate civil liberties to a higher degree than other governments (Inglehart and Welzel 2009; Fish 2002). Several reasons for this result have been suggested. First, it has been argued that Islam is negatively associated with emancipatory values (Inglehart and Welzel 2009). Second, especially in the Middle East and North Africa, autocratic elites have been successful in using popular resentment of ‘Western imperialism’ to ward off the pressure for liberalization from international organizations and their Western sponsors (Lewis 1990). Meanwhile, Western leaders have been vary to pressure these regimes for fear that regime breakdown would pave the way for Islamist alternatives, and have therefore often preferred autocratic stability to the destabilizing effects of political openings (Levitsky and Way 2010: 41; Callaway and Matthews 2008). We have created a dummy variable to measure this variable, distinguishing countries with a Muslim plurality or majority population (1) from the rest (0).

Finally, we also include a dummy variable measuring the end of the Cold War (with 1991 as the pivot), a juncture we have already shown to be related with developments in the civil liberties scores.

Based on the hierarchy identified in the sequencing analysis, we create an aggregated (simple addition) CLD-index of civil liberties and an aggregated CIRI-index based on the five resembling indicators: both have been rescaled to range from 0 to 100. Table 5 reports the results when entering the seven variables listed above in a common model and subjecting it to OLS panel regression analysis16 with the CLD and CIRI civil liberty indices interchangeably used as dependent variables. We have run the analysis with and without fixed effects17 and with and without an identical coverage of country-years. These alterations do not influence the general picture much.
Table 5

Results from regression analysis of explanatory factors and the civil liberty indices

 

CLD-index

CIRI-index

CLD-index

CLD-index

CIRI-index

CLD-index

Random effects

Random effects

Random effects + common coverage

Fixed effects

Fixed effects

Fixed effects + common coverage

Constant

59.366***

(1.419)

59.345***

(1.840)

60.298***

(1.479)

56.721***

(0.801)

55.909***

(1.230)

56.155***

(0.860)

Democracy

17.811***

(0.599)

18.234***

(0.908)

14.956***

(0.645)

17.127***

(0.604)

16.491***

(0.934)

14.070***

(0.653)

Communist regime

−33.425***

(1.014)

−34.382***

(1.567)

−34.440***

(1.115)

−32.422***

(1.031)

−33.771***

(1.629)

−33.453***

(1.139)

Modernization indext-1

−4.019***

(0.614)

−1.941*

(0.899)

−2.146**

(0.689)

−8.483***

(0.752)

−8.154***

(1.291)

−7.094***

(0.902)

Modernization indext-12

1.316***

(0.212)

0.910**

(0.315)

0.970***

(0.227)

1.868***

(0.221)

1.832***

(0.349)

1.650***

(0.244)

Oil rent per capitat-1 (logged)

−0.013

(0.106)

−0.172

(0.157)

−0.113

(0.115)

0.006

(0.115)

−0.023

(0.183)

−0.107

(0.128)

Protestant

10.489***

(2.273)

10.734***

(2.502)

9.188***

(2.349)

   

Muslim

−16.892***

(2.042)

−20.829***

(2.502)

−18.877***

(2.117)

   

Post-Cold War

0.607

(0.383)

−1.367*

(0.556)

1.377**

(0.399)

2.452***

(0.411)

0.786

(0.612)

3.022***

(0.428)

R2

0.594

0.562

0.617

0.311

0.271

0.318

Observations (countries)

5,048

(183)

3,961

(182)

3,961

(182)

5,048

(183)

3,961

(182)

3,961

(182)

Note: *P < .01, **P < .01, ***P < .001. Unstandardized betacoefficients with standard errors in parentheses. One-sided significance tests

The variables generally behave as expected. Democracy and communist regime have a statistically significant effect on the violation of civil liberties, in opposite directions. Likewise, the findings also suggest that the dominant religion in the form of Protestantism or Islam is important for contemporary levels of civil liberty: the former tends to increase respect for civil liberties, whereas the latter generally tends to decrease it. The squared modernization term turns significant for both indices, indicating that the relationship between socio-economic development and civil liberty is U-shaped. Regarding the supposedly negative impact of oil rents, the relationship is only significant for the random effects model. The most surprising finding, however, is the negative (random effects) or insignificant (fixed effects) relationship between the post-Cold War variable and the CIRI-index. Based on our rather clear expectations with regard to this factor, these results go some way towards supporting that the CLD data have a competitive edge compared to the CIRI data. Notice, furthermore, that although most of the coefficients differ little, the standard errors are almost consistently lower when the CLD-index is used as dependent variable than when the CIRI-index is employed—and vice versa with regard to the R2. This indicates that there is more ‘noise’ in the CIRI data.

8 Conclusions

In this article, we have presented a new, global dataset on civil liberties, the CLD. We have used it to reveal some strong patterns in state repression. Our analyses have demonstrated that—for all years between 1979 and 2010—political liberties have generally been repressed more than private liberties. Moreover, based on a Mokken scaling analysis, we have identified a stepwise hierarchical pattern across the four civil liberties in the period 1976–2010. In general, freedom of expression is repressed at least as much as freedom of association and assembly, which is repressed at least much as freedom of movement, while freedom of religion is not repressed more than any of the other liberties. Despite some interesting exceptions to this pattern, the result indicate that the civil liberties tap into a common, underlying dimension, which makes it possible to construct a cumulative scale. Finally, we have assessed the extent to which variations in a composite CLD-index can be accounted for by some widely accepted correlates of civil liberties.

The results generally lend support to the quality of our data. More specifically, vis-à-vis the one extant dataset containing disaggregated scores for First Amendment-type civil liberties—the CIRI Human Rights Database—the CLD fares at least as well (and in several respects somewhat better) with respect to scalability and construct validity. Furthermore, though the indicators are thematically identical and coded on the basis of the same source, the two datasets seemingly capture somewhat different empirical aspects of the global levels of respect for civil liberties as documented by the relatively low inter-index correlations. We thus conclude that the introduction of the CLD presents a two-fold advantage. First, recourse to two measures beats recourse to one for the simple reason that it makes it possible to carry out robustness tests of findings about descriptive patterns, causes, or consequences of civil liberties. Second, the CLD has a higher coverage of country-years than the CIRI data, and therefore makes it possible to marshal more data in analyses of state repression in general and civil liberties in particular.

We urge scholars to use the composite CLD scale or the disaggregated CLD scores to shed more light on the violation of civil liberties since the third wave of democratization took off in the late 1970s. A first attempt to pursue some of the issues that have been discussed in the literature has been carried out in this article. To single out one particular result, we have used the CLD to show that the end of the Cold War produced a spike in both electoral rights and civil liberties but that this abrupt increase has been lopsided in that the average respect for electoral rights increased much more than the average respect for civil liberties, and among the civil liberties, improvements in respect for political liberties were higher than for the more private liberties. These findings go to show that a disaggregated, comparative approach to regime change is necessary to understand important dynamics when it comes to governments’ repression of fundamental rights.

Footnotes
1

In the human rights literature mostly understood as freedom from disappearances, political killings, torture, and political imprisonment.

 
2

At least for the period until 2005. Since then, the sub-component scores of the Freedom in the World survey have been made publically available.

 
3

E.g., the Countries at the Crossroads, Freedom of the Press, and Nations in Transit datasets compiled by Freedom House and the Bertelsmann Transformation Index provided by the Bertelsmann Foundation.

 
4

The high number of countries is partly a consequence of splits and mergers of a few units over the period. The only sovereign unit not covered by the CLD is the Vatican State.

 
5

The only exception is the coding of the US (not covered by these reports), where we have relied on the country narratives produced as part of the Freedom in the World survey by Freedom House.

 
6

In transforming the information into the scores constituting the dataset, at least two independent coders (trained graduate students well-versed in comparative politics and, regarding the (post-)communist and Latin American countries for the period 1977–2003, also Svend-Erik Skaaning) assigned scores to all the country-years. In case of the Latin American and (post-)communist countries 1979–2003, disagreements were settled by discussions among the coders, and for the remaining country-years a third coder was authorized with the final judgment in the case of disagreement. For a more detailed description of the coding standards and the full dataset, please visit: www.ps.au.dk/dedere. Regarding inter-coder reliability tests, these have been run for all four freedoms for all years. The percent agreement ranges from 0.73 (freedom of assembly/association) to 0.78 (freedom of movement). The agreement tends to increase somewhat over the period. The Cohen’s kappa is 0.60 for the freedom of religion scores, 0.64 for the freedom of movement scores, 0.64 for the freedom of assembly/association scores, and 0.68 for the freedom of expression scores. In all cases, this is within the range of what Landis & Koch (1977) regard as good (substantial) agreement.

 
7

On this point, we are inspired by Munck’s (2006) arguments about the need to establish scale scores, and particularly the end-points of scale scores, based on explicit theoretical distinctions.

 
8

For example, the underlying coding scheme is merely a checklist that has undergone several changes over the years, and it does not specify the meaning of the different scores.

 
9

Partly for this reason, the number of country-years covered by CIRI’s First Amendment-type indicators ranges between 4,748 and 5,363, while it is 6,035 for all the CLD indicators.

 
10

It is, of course, also possible that much of the difference is due to problems with coder reliability and/or bias.

 
11

For a similar use of the notion of sequencing, see Cingranelli and Richards (1999).

 
12

Importantly, the DD-measure is not based on conceptual attributes denoting or empirical indicators measuring civil liberties. More on this later.

 
13

The reason we include the years 1976–1978 in this analysis is that the sequencing should be the same over the entire period in all countries. Also, omitting these early years hardly alters the results.

 
14

It drops to 0.59 if we replace the indicator on domestic movement with that on foreign movement.

 
15

The index is based on eight indicators: (1) industrialization (output of non-agricultural sector/GDP), (2) education (gross secondary school enrollment ratio, (3) urbanization (urban percentage of total population), (4) life expectancy at birth (in years), (5) the inverse of infant mortality rate (per 1,000 life births), (6) the log of GDP/cap. (current US dollars), (7) radios/cap., (8) televisions/cap., and (9) newspaper circulation/cap. The index values are computed by taking the factor scores “and then using imputation on the regression line with all nine indicators as regressors” (Teorell 2010: 164–165).

 
16

According to Schuur (2011: 95–96), one can treat Mokken scales as metric scales in statistical analyses.

 
17

We are unable to include the Protestant and Muslim dummies into the fixed effects analyses due to a lack of diachronic variation in these variables.

 

Copyright information

© Springer Science+Business Media Dordrecht 2013

Authors and Affiliations

  1. 1.Department of Political ScienceAarhus UniversityAarhus CDenmark

Personalised recommendations