91

State Repression and Domestic Democratic Peace

importance of political order for subsequent state repressive activity but

also assists me in addressing any temporal dependencies that exist within the

data.23

Seeking to understand and communicate how different explanatory fac-

tors in¬‚uence the diverse categories of repression (across equations), I cal-

culate the predicted probabilities of achieving a particular combination of

restrictive and violent repressive behavior, given a movement of a speci¬c

independent variable from its minimum to maximum value (holding other

variables at a speci¬c value, normally the median). Predicted probabilities

are useful because standard statistical results merely provide statistical sig-

ni¬cance but do not help with understanding substantive in¬‚uence. Proba-

bilities, by contrast, help us assess the substantive impact of an independent

variable on state repression such that we can understand the magnitude of

relevant in¬‚uences. This is the technique applied throughout the remainder

of the book.

Results are also communicated in another way. As discussed earlier, I

assess the in¬‚uence of some aspect of democracy on repressive lethality as

the former is increased from its minimum to its maximum. While in¬‚uential

in getting a general understanding of relationships, this does not consider

the fact that governments do not start at the same level of repression. Some

governments might have previously been using coercive behavior that was

low in political violence but moderate in restrictions (at level 2) “ Turkey

between 1976 and 1979, Haiti in 1986, or Ecuador between 1994 and 1996 “

whereas others might be high in violence but moderate in restrictions (at

level 8) “ Iran between 1977 and 1978, Brazil between 1990 and 1996,

Haiti in 1994, South Africa between 1989 and 1994, and Rwanda in 1993.

These are very different contexts and represent extremely different scenar-

ios within which democracy is expected to have an impact. To address this

issue of diverse starting points, I consider the most likely value of current

repression (that is, at time t), given a speci¬c value of previous repression

(that is, at time t ’ 1).

Understanding the Base Model

With an ordered probit model, the relative importance of political order

and development/modernization for state repression is provided below.

23 At present, there are no simply ways to estimate ordered probit models with a more rigorous

estimation and correction for time-serial problems.

92

Data and Methodology

Although both explanations ¬nd support across types of repressive behavior,

the political order argument is generally more robust.

In Table 3 of Appendix 1, I incorporate all relevant independent vari-

ables discussed earlier in an attempt to understand variation in repressive

behavior. From the analysis, one can identify several things. First, all vari-

ables increase the lethality of coercive action except economic development/

modernization and the difference from regional repression, which have ne-

gative in¬‚uences. Second, diverse aspects of political order reveal the same

empirical ¬nding: threats to government increase the lethality of repressive

behavior and prior coercion increases the likelihood of subsequent appli-

cations.

For a clear understanding of exactly how these explanatory factors in¬‚u-

ence diverse strategies of repression, I present predicted changes in proba-

bilities (that is, the likelihood that each repressive category will be achieved

given a shift in the variable of interest from its minimum to its maximum).

To simplify the discussion of these results, I have estimated the effects of

lagged repression, setting the value of this variable to 5.24

To read Figure 3.1, one simply looks at a speci¬c category of repression

(that is, 1“9, along the x- axis) with reference to the independent variable

of interest (identi¬ed on the legend along the top or at the right side of the

¬gure). One then identi¬es the corresponding change in the probability

of achieving this category derived from moving the relevant explanatory

variable from its minimum to its maximum.25 If the value for the speci¬c

category is positive, then, given an increase in the independent variable, the

likelihood of a particular combination of repressive behavior is increased.

Correspondingly, if the value is negative, then, given an increase in the

independent variable, the likelihood of a government using a particular

strategy of repression is reduced.

Almost immediately, one can see from Figure 3.1 that there are impor-

tant differences between the explanatory variables associated with political

order and those associated with development/modernization in terms of

24 Recall that the foregoing analysis suggested that dummying out the different values of the

lagged dependent variable was the most appropriate strategy for measuring this variable.

In order to calculate the result for the statistical procedure employed here, however (in

a relatively straightforward manner), I decided to select a value of repression near to the

mean.

25 Within the previous model, each of the values was investigated individually, but in order to

acquire a simple and straightforward understanding of the general impact of these variables,

I employed this somewhat coarse methodological technique.

93

State Repression and Domestic Democratic Peace

GNP/Capita and Population

Change in Categories of Repression

Log(GNP/capita) Log(Population)

0.3

0.25

0.2

0.15

0.1

0.05

0

1 2 3 4 5 6 7 8 9

-0.05

-0.1

-0.15

-0.2

Categories of Repression

Figure 3.1 (a) Changes in Probability for Maximal Changes in Base Model Vari-

ables

Conflict Variables

Change in Probability of Repression

Violent Dissent Civil War Interstate War

0.15

0.1

0.05

0

1 2 3 4 5 6 7 8 9

-0.05

-0.1

-0.15

-0.2

-0.25

Categories of Repression

Figure 3.1 (b)

94

Data and Methodology

Repression Variables