Quantitative Data Analysis and

Interpretation

Approved Essay Title: Explaining Political Instability in a

Quantitative Cross-National Study (QDA)

2007-2008

Word Count: 5202

Page 1 of 29
Abstract:

This essay reports statistical correlations and associations between key variables and
political instability. The variables selected represent internal and external, as well as
economic and political, factors which allows conclusions to be drawn as to their
relative importance for the explanation of political instability. This is done through
statistical techniques such as regression analysis. This essay finds that economic
factors and internal factors seem to have the most relevance for the analysis of the
correlates of political instability. The essay concludes by highlighting possible issues
with the analysis including the measurement of variables and cause and effect.

Part 1 – Introduction

The central research question of this essay is concerned with the question of

why political instability happens, placing the analysis in a cross-national context. The

issue of political instability is important because of the problems, both social and

economic, which political instability brings to the people of a country which is being

deeply affected by it. Indeed, as Shaw notes, “Prospects for regional development

recede as conflicts both escalate and proliferate” (2003, p. 489). An understanding of

political instability could lead to developments of policy strategies aimed at

preventing these problems when it arises, and aimed at preventing its occurrence.

Without an accurate theoretical understanding of why and how instability is generated

even the best meaning action may fail. The issue of political instability is also

theoretically interesting because of the numerous and disparate variables which have

been suggested as explanatory in this field. Just some of the explanations offered for

political instability include; economic inequality (Lichbach 1989, passim), poor

Page 2 of 29
regional relations (Brown, 2001, p. 16), the impact of crime (Brown, 2001, p. 18) and

ethnic fractionalisation (Gurr, 1993, p.161). More research is obviously needed in this

area in order to determine what the relationship actually is and which factors are more

important. In this essay the relative importance of political and economic factors will

be assessed as well as the relative importance of internal and external factors,

concluding that economic and internal factors seem most important respectively.

The dependent variable of this study is the level of political instability 1. The

independent variables are; wealth, level of democracy, ethnic fractionalisation and the

number of border states with any kind of major conflict.

I – Research expectations:

Wealth and political instability

We may expect, before analysis, a negative relationship between the wealth of

the people of a country and political instability. Theoretically there are at least some

reasons to suppose that this is the case, mostly relating to what might be termed the

opportunity costs of violence. Przeworski and Limongi used income per capita to

show that democratic regimes were most vulnerable to collapse when the level of

income was low absolutely (1997, p. 161). Taking this as a starting point, the

relationship between the overthrowing of democracy and political instability seems

intuitively obvious; when democracies are overthrown violence and repression often

follows. This gives us a theoretical reason for assuming that wealth and political

instability as defined above are related for the same reasons that income per capita

1
The definition of political instability which is used here will follow Kaufman in his definition of the
term “political stability”, although will obviously focus on the negative side. A full explanation of this
term can be found in World Bank (2007) and the Univariate analysis section in this work.

Page 3 of 29
and democratic death are related. Here no assumption is made as to the source of the

violence, whether it is initiated from above, by elites, or below, by mass level forces.

This theory, though, only explains why violence becomes an option which can be

taken; the reason why violence is chosen still needs to be explained. Przeworski and

Limongi suggest that the choice is likely to be made by those who seek to enlarge

their own share of the income distribution whilst the marginal costs are lowered, and

thus so is the risk (1997, p. 166). This is largely in line with Shaw’s observations of

conflicts within Africa; increasing conflicts over diminishing resources to acquire (see

Shaw, 2003, p. 488). Similarly, Gurr noted that objective conditions like poverty were

crucial grievances used by elites to mobilise people for rebellion (1993, p. 189).

Whether the effect which Przeworski and Limongi noted still applies strongly is an

open question, Bermeo argued that wealth is “not a necessary condition of democratic

durability” and as such we may question if wealth still has explanatory force in terms

of political instability (2003, p. 169).

Level of democracy and political instability

It has been observed that when a country becomes a “full” democracy civil

wars are an extremely rare, if not completely absent, phenomena (see Hegre et al,

2001, pp. 33-34). This might be because, as Bermeo claimed, that people are more

likely to engage in violent competition, rather than peaceful and democratic

competition, when the costs of the former are reduced compared to the latter (2003, p.

164). Semi-democracies have been regarded as bringing stabilising characteristics to

countries and it has been contended that they can even prevent conflict in societies by

allowing demands for at least some popular consultation whilst at the same time

Page 4 of 29
allowing the present ruling elites to maintain power which they would be unwilling to

give up (see Case, 1996, p. 457). Despite this we should not necessarily expect the

relationship to be a simple linear one. It has also been seen that countries which are

harshly authoritarian have fewer civil wars than those which are in the intermediate

stages of democratisation (Hegre et al, 2001, p. 33). These analyses are also likely to

hold true for other forms of violence which fall short of civil war, including ethnic

violence (see Snyder, 2000, p. 310). This in itself is surprising if Bermeo was correct.

One explanation sees potential conflicts between ethnic groups, or between different

communities, as having already existed but laying “dormant” because of the

repressive nature of authoritarian regimes (Byman and Van Evera, 1998, p. 33).

Another explanation sees national elites who do not wish to abandon their own

political power creating conflict and nationalism to allow them to keep their power

(Snyder, 2000, p. 36). Indeed, there seems to be considerable amounts of qualitative

data to back this claim (see Mansfield and Snyder, 2005, pp. 169-227). The exact

expectation is then unclear and will alter depending on whether we accept Case’s

analysis or that of Mansfield and Snyder.

Near-by conflicts and political instability

It is important to note that political instability, as defined by Kaufman, can

often be caused by forces external to the country in which the problems are manifest.

External factors operating at the mass-level include “spill-over” or “diffusion” of

conflict from other states (Brown, 2001, p. 16). These factors can operate regardless

of domestic problems, or absence thereof. To focus exclusively on internal factors,

then, will lead to an incomplete analysis and may even create misleading results.

Page 5 of 29
Spill-over of conflict could be created either by “swarms of refugees” or key radicals

moving operations due to conflict (see Brown, 2001, p. 16). Foreign armies can also

play a key role in causing political instability including, but not limited to, leading

direct military assaults on the government (Shaw, 2003, p. 491). Similarly “diffusion

and contagion” of conflicts has been shown in previous research to be a factor in

explaining conflict, although not a particularly strong one in all cases (Gurr, 1993, p.

189). Theoretically, then we should expect to see a relationship between the proximity

of foreign conflicts or foreign hostile regimes and the amount of political instability

which is found within a country.

Ethnic fractionalisation and political instability.

The degree to which a country is ethnically homogenous has been suggested

as a possible explanatory factor in many aspects of political instability. Ted Gurr

notes that most countries have seen conflicts relating to the “terms of incorporation

for ethnic minorities…” (1993, p. 161). Whilst, as Gurr notes, this does not have to be

violent conflict we at least have some reason for expecting to see more conflict in

countries which have a higher degree of ethnic heterogeneity (see Gurr, 1993, p. 161).

This theory runs counter to Collier and Hoeffler who argue that fractionalisation will

increase the costs of recruiting to a rebel force and will therefore reduce instances of

civil war, and so political instability (2000, p. 8).

Page 6 of 29
The Null Hypothesis

As always the null hypothesis is a statement of no difference and can be formulated

as;

- There is no difference in terms of political instability between countries with

differing: levels of wealth, levels of ethnic fractionalisation, number of geographically

close states with major conflicts, or levels of democracy.

This then means that the hypothesis can be formulated as stating that there is a

difference between the levels of political instability in countries, and those countries’:

levels of wealth, levels of ethnic fractionalisation, number of geographically close

states with major conflicts, or levels of democracy. Showing significance in a

relationship between the dependent variable and any one of the independent variables

disproves the null hypothesis relating to that variable and shows that there is a

difference between countries level of political instability which is related to the

independent variable. Importantly significance does not tell us the direction of the

relationship, and this must be discovered by an ex post analysis, which will be

computed if necessary.

The cases are 191 countries from the global indicators data set. These

countries represent a population within the field of study. Whilst this means that they

are technically not a sample it will still be useful to treat them as a sample of the

wider world throughout time. Hopefully this will mean that we can say with greater

confidence that the results found are generalisable into the future as well as having

current time-period empirical validity.

Page 7 of 29
II – Univariate Analysis

Concept and measurement of political instability

Political instability, the dependent variable, will be measured using Kaufman’s

political stability rating. This means that the relationships which are found

(positive/negative) will actually be the inverse of the relationship we are concerned

with. For example if increasing GDP is found to have a positive relationship with

political stability it has a negative relationship with political instability. The choice for

this variable is because Kaufman’s variable captures the important elements of the

likelihood of politically motivated violence in a way which no other variable is likely

to do. For example, data on the number of politically-motivated deaths in a country

may well accord ‘positive’ scores to countries in which non-fatal political violence is

often used as a measure of coercion. The year which will be selected to measure

political stability will be 1998. Kaufman’s political stability is a largely subjective

measure based on ordinal level assessments from experts and respondents based on

their assessment of the country they are concerned with regarding political violence

(Kaufman et al, 2007, Appendix A). The ordinal sources used to create the political

stability variable are weighted according to the reliability which Kaufman et al.

believes that they have and then aggregated (World Bank, 2007). The variable

measures countries against each other in order to construct a value which can be given

to each, such as that the mean is approaching 0 and the standard deviation is

approaching 1. Because of the way the variable is constructed it is an inherently

imperfect measure of what we are actually trying to measure. Whilst possibilities of

Page 8 of 29
error are obviously very important here it is important to note that the variable will

still be of use for creating comparisons within the same variable (see Kaufman et al,

2007, p. 15). The variable itself is ostensibly ordinal however because it is continuous

and constructed to be used as an interval-ratio level variable it will be used as such.

A histogram representing the distribution of the countries which are ranked

using the political stability variable is included below (see chart 1 below). As can be

seen data is only available for 163 countries in the world. Missing values are not

explained by the World Bank nor Kaufman, however they are likely to be caused by

lack of available data. It is important to note though that the missing values do not

appear to be just limited to the most authoritarian regimes, or even all of the most

violent ones, as can be seen by the inclusion of North Korea and the Democratic

Republic of Congo respectively.

Chart 1:

Histogram showing distribution of Kaufman's political stability variable

25

20
Frequency

15

10

Mean =-0.0118
Std. Dev. =1.00383
N =163

0
-2.00 0.00 2.00

Kaufman's political stability

Concept and measurement of wealth

Page 9 of 29
 The wealth of the people of a country is a concept which can be measured in

many different ways, income per capita being just one. GDP per capita is another,

related, way of measuring wealth. Per capita income and GDP per capita are likely to

be related in a strong way such as that theoretically one could be operationalized in

place of the other, as Bermeo does (2003, p. 169). If either of these are used the

comparative rankings of countries are likely to be the same. Neither of these are

measures of absolute wealth of countries, this is important because it avoids the issues

relating to extremely large but poor countries having a high GDP compared to

extremely small yet rich countries. Within the global indicators data set there is data

available for GDP per capita for 1997 in 1987 US$. Whilst the year does not perfectly

correspond to the year which is used for the dependent variable it is likely that the

GDP of countries will not have varied significantly over a one year period. This is of

course an untested assumption and could introduce the possibility of error, however in

the absence of perfect data it does not seem like a particularly egregious assumption.

This histogram below (chart 2) displays the variable distribution graphically;

Chart 2:

Page 10 of 29
A histogram representing distribution of GDP per capita measured in 1987US$ per
country

80

60
Frequency

40

20

Mean =4322.134
Std. Dev. =6940.6476
N =134

0
0.0 5000.0 10000.0 15000.0 20000.0 25000.0 30000.0
GDP per capita in 1997 (measured in 1987US$)

As is immediately apparent the values of this variable are not normally distributed.

This is because of the few countries which have a comparatively high GDP per capita

introducing a positive skew. This is the reason for the extremely high standard

deviation. I will however treat this variable as if it was normally distributed for the

purpose of this essay, although I am aware of the issues this creates. Data is available

for 134 countries. As can be seen from the chart, the standard deviation is

comparatively high.

Concept and measurement of external conflicts

Page 11 of 29
 The types of external conflict which impact upon the functioning of a regime

and create political instability seem likely to be major conflicts within countries which

are geographically close. If this were not the case the “swarms of refugees” problem

may not arise (see Brown, 2001, p. 16). Similarly it is only likely to be major conflicts

which force insurgents or guerrillas to move operations. In both these cases the

countries in which the problems manifest themselves do not have to be directly

involved in the conflict. In the case of foreign powers attacking a country

geographical proximity is less important. Unfortunately no data for the belligerence of

foreign regimes external of geographic proximity is available. This means we are

focussing almost exclusively on what Brown referred to as “bad neighbourhoods”,

that is to say focussing on mass-level external problems (2001, p. 15). The variable

which will be used to measure this concept, then, is the number of border states with

any major conflict during the period 1990-1999 2. The variable will be recoded into a

nominal level dummy variable assessing if a country is in a “bad neighbourhood”; this

is because the available data makes no attempt to judge neither how serious (beyond

an undefined level of “major”) a conflict was nor how long it lasted for. Given this

fact, a country which is bordered by six countries which have had “major” yet short

lived conflicts which caused little trouble internationally will get a higher score than a

country bordered by one country which has had one conflict that destroys much of the

underlying infrastructure of that society. Obviously the latter poses the greater danger

but could be missed if we simply focussed on the numerical results. To avoid

problems like this the variable will be created as having values of “yes” for being in a

bad neighbourhood, which is to say has more than one boarder state with a major

2
It is important to note that this variable would almost certainly include countries which go to war
against each other but are geographically close.

Page 12 of 29
conflict during the period, or “no”, for not having. The underlying source of this

variable covers years past 1998 when the political instability was measured, but this

may not be a problem. If a major conflict seemed likely to begin within a year it

would not be unexpected to see refugees escaping the probable violence, and so some

effects might be detected then. Similarly problems can take a long time to be resolved

and even a conflict in a bordering state in 1990 might still be having an impact in

1998.

The distributions of this variable after recoding are listed below (Table 1);

Table 1:

Concept and measurement of level of democracy

This concept is represented in the global indicators data set as the variable

“Democracy-Autocracy, mean (1990-99)”. This is measured by subtracting the

commitment to authoritarianism from their commitment to democracy (see CIDCM,

2006). For this variable I shall recode the data into a three category ordinal variable

representing how democratic the country is using categories of “democracies”, “semi-

democracies” and “autocracies”. This variable is a mean throughout time and as such

Page 13 of 29
does not give us exact data for 1998 to compare on equal terms with the political

instability variable. The benefit of this, though, is the effect of “averaging out” of

“random fluctuations of the variable” (Parvin, 1973, p. 288). This is an advantage for

which foregoing some equality of the variables may be worthwhile. The variable is

represented in the bar chart below (see chart 3);

Chart 3:

Bar Chart Representing Distribution of Values of "Autocracy" "Semi-Democracy" and

"Democracy"

80

60
Count

40

20

0
Autocracy Semi-democracy Democracy

N=157

As can be seen, the largest grouping is “democracy” with autocracy and semi-

democracy each having roughly similar distributions.

Concept and measurement of fractionalisation

Page 14 of 29
 Fractionalisation refers to how diverse the character of a country is. There are

several important ways of measuring this concept including but not limited to; racial

fractionalisation, linguistic fractionalisation and religious fractionalisation. These

variables are often used in conjunction. Indeed, it had been common to use

“ethnolinguistic fractionalisation” as a variable; however this tends to miss the

subtleties between ethnicity and language which make them sometimes congruent and

others not (Alesina et al, 2003, p. 159). For example language may be an important

part of ethnicity in Africa but in the United States it is unlikely to be taken as such

(ibid.). Because of this issue, this essay will use Alesina et al.’s ethnic

fractionalisation variable from the global indicators data set. This variable represents

both racial and language variables, although the emphasis is on the latter (ibid.

pp.159-160). This variable is therefore a composite variable which can range between

0 and 1 to describe the level of ethnic fractionalisation, with higher numbers meaning

greater fractionalisation. The level of measurement is ostensibly interval-ratio. The

following histogram shows the distribution of this variable (see chart 4);

Chart 4:

Page 15 of 29
 A histogram representing the distribution of ethnic fractionalisation scores

14

12

10
Frequency

2 Mean =0.439376
Std. Dev. =0.2565464
N =188

0
0.0000 0.2000 0.4000 0.6000 0.8000 1.0000
Ethnic Fractionalisation

The distribution seems bi-modal, showing two fairly distinct peaks. I will however

treat the data as the distribution was normal.

III – Bivariate analysis

Political instability and GDP per capita

Chart 5:

Page 16 of 29
 This scatterplot (chart 5) shows that there is a positive relationship between a

country’s GDP per capita and result on the political stability variable. The relationship

is significant at the 99.9% level. The regression line has a Y intercept of -0.384,

meaning that the regression line crosses the Y axis at a political stability score of -

0.384. This means that if a country had 0 GDP per capita in 1997 we could expect it

to have a political stability score of -0.384. The gradient of the regression line is

0.0000882. This means that for each unit change in GDP per capita we can expect to

see 0.0000882 of a unit change in political instability. Whilst this change appears very

small it is due to the large numbers which we deal with GDP per capita in and the

ostensible 5 unit range of Kaufman’s political stability variable. This allows us to

predict values for political stability by using the formula “Y’=-0.384+0.0000882 X”

where X is any value of GDP per capita and Y’ is the predicted political stability

value. However because we are primarily concerned with political instability the

result of this analysis is that increasing GDP per capita has a negative effect on

political instability.

Page 17 of 29
 For this model both t and F report a significant relationship. The F value and t

values reach significance levels of 99.9% which tells us our regression model is

significantly better at predication than using the mean value, as well as having a slope

which is different from 0 (Field, 2005, pp. 154-156). This allows us to reject our null-

hypothesis and conclude that there is a difference in terms of political instability.

The R of the relationship represents the correlation between the two variables

and was 0.637. The R² of this relationship was 0.406. This means that we can account

for 40.6% of the variation in political instability by knowing the country’s GDP per

capita. Following Healey, we can interpret these two figures as showing a strong

positive relationship (Healey 2005, p. 404).

Political instability and ethnic fractionalisation

Chart 6:

A Scatterplot showing political stability against ethnic fractionalisation

2.00
Kaufman's political stability

0.00

-2.00

R Sq Linear = 0.156

0.0000 0.2000 0.4000 0.6000 0.8000 1.0000

Ethnic fractionalisation

This scatterplot (chart 6) shows a negative relationship between political

stability and ethnic fractionalisation. That is to say as countries get more ethnically

Page 18 of 29
fractured they get less stable or, of more concern for the present research question,

they get more unstable. Here the Y intercept, the value which we would expect on the

political stability variable if ethnic fractionalisation was 0, is 0.719. The gradient is -

1.562, meaning that for each unit change in ethnic fractionalisation we would expect

to see a fall of 1.562 units on the political stability variable. It is important to note,

though, that ethnic fractionalisation runs completely from 0 to 1 so in fact the slope

isn’t as steep as it might at first seem from that number. Similarly to the analysis of

GDP per capita and political stability above, both F and t scores report a significant

relationship at the 99.9% level, allowing us to reject the null hypothesis. As such we

can reject our null hypothesis and say that there is a difference between a country’s

rating on the political stability variable based on differing values of ethnic

fractionalisation. The R was 3.95 which, following Healey, we can interpret as a weak

to moderate relationship (Healey, 2005, p. 404). The R² was 0.156; this means that

15.6% of the variance of the political stability variable was explained by the amount

of ethnic fractionalisation.

Political instability and bad neighbours

For the following two bivariate analyses the dependent variable has been

collapsed because in it’s former state as an ostensibly interval-ratio level variable it

contained too many possible values or “scores”. Assessing this nominal level variable

we can see a moderately strong3 relationship between the variables, but one which is

significant at the 99% level. This significance allows us to reject the null hypothesis

3
Obviously Cramer’s V does not have an exact interpretation in terms of the strength of the
relationship, however here I am following Healey in his interpretation (see 2005, p. 342)

Page 19 of 29
and state that political instability and being in a bad neighbourhood is related. The

results from the Cramer’s V analysis are presented below (table 2);

Table 2:

Looking at a cross-tabulation table we can see that the relationship is such that if

countries are in a bad neighbourhood the likelihood of them being stable themselves

will be reduced; and so the likelihood of them being unstable increased.

Political instability and semi-democracy

The results shown below (table 3) demonstrate that there is a relationship

between semi-democracy and political instability. The relationship is significant at the

99% level which is enough for us to be confident in rejecting the null hypothesis.

Following Healey we can say a Gamma of 0.455 represents a moderate relationship

(2005, p. 368).

Table 3:

Page 20 of 29
The gamma also shows direction, showing increasing democracy levels are positively

related to political stability or negatively to political instability.

Treating the variables in this way however assumes linearity, which we should

assume not to be the case from our theoretical assumptions. Taking the original

continuous data from the data set and measuring against the non-compacted political

stability variable we can see a curvilinear relationship which we expected in theory,

with an R² of 0.337, meaning 33.7% of the variance is explained by this variable (see

chart 7);

Chart 7:

Page 21 of 29
 Expanded political stability against expanded autocracy/semi-
democracy/democracy

Kaufman's political stability 2.00

0.00

-2.00

R Sq Quadratic =0.337

-9.0 -6.0 -3.0 0.0 3.0 6.0 9.0

Autocracy/semi-democracy/democracy

IV – Multivariate Analysis

Multivariate analysis will create a model which includes all of the variables

which have been used in this essay to construct a more complete model. Here all of

the variables are treated as if they were interval-ratio level. This will allow us to

assess the relative importance of each variable when they are being controlled for the

effects of each other. Cases were excluded “pair wise” to maintain a higher number of

cases for analysis, despite noting criticisms of this method by Field (2005, p. 183).

The total model summary is shown in table 4.

Table 4:

Page 22 of 29
Here we can see that the R², which is to say the amount of variance on the dependent

variable which is explained by the independent variables, is 0.482 or 48.2%. The

adjusted R² value tells us how much variance would be explained by our model if the

model had been derived from the population which the sample was drawn from. As

was stated above the global indicators data is being treated as representing a time-

sample to aid in generalisations, even though strictly speaking it is a population in

itself. The Durbin-Watson score is one of the assumptions which must be met for a

sample regression to be applicable to the population; here testing for independent

errors (Field, 2005, p. 170). A value of between 1 and 3 is good enough, although 2 is

the best result for this test (ibid.). Because the value is close to 2 we can assume that

at least this requirement has been met 4. The R for the model is .694 which, following

Healey, shows a strong relationship between the four variables and the dependent

variable (Healey, 2005, p. 404). The ANOVA results show that the model is

significant, meaning that the model predicts the results significantly better than using

the mean values (see Field, 2005, p. 189).

Table 5:

4
Obviously there are more assumptions which must be met in order for us to be completely satisfied
that we can generalise, a full list of which can be found in Field, 2005, pp. 169-170. Some of the
assumptions, such as linearity, have already been questioned in the above research however this will
probably offer a reduction in our model’s predictive capacities so at worse serves to make us more
cautious than we might be otherwise. Other assumptions will not be tested, but will be assumed to have
been met.

Page 23 of 29
The above table (table 5) allows us to analyse which variables have the biggest impact

upon our ability to predict the outcome of the dependent variable and to state the least

squares regression line. The regression line can be stated as;

Y’ = -0.456 + (-0.455)X1 + (0.0000704)X2 + (0.311)X3 + (-0.113)X4

Where Y’ is the predicted value on the political stability variable, X 1 is ethnic

fractionalisation, X2 is 1997 GDP per capita, X3 is whether a country is

autocratic/semi-democratic or democratic5, and X4 is whether the country has bad

neighbours.

The direction of the relationship shows ethnic fractionalisation and whether or

not the country was in a bad neighbourhood had a negative relationship to political

stability. This means that if the countries were in a bad neighbourhood or were

ethnically fractured political stability would decrease, and so political instability

would increase. As the level of democracy and GDP per capita increases political

5
Because for this multivariate analysis linearity has to be assumed, it will be. This may weaken the
effect of this variable in the final analysis. The compacted form of the data described in the univariate
analysis is used here.

Page 24 of 29
stability increases, which means increasing democracy and GDP per capita results in

political instability decreasing.

The standardised beta values represent the strength of the relationships

between each independent variable and the dependent variable in a way which makes

them directly comparable. The unit used is standard deviations and tells us how many

standard deviations the dependent variable will change if there is a one standard

deviation change in the independent variable (Field, 2005, p. 193). As can be seen the

country’s level of GDP has the biggest impact upon the political stability variable, and

so on political instability. The autocracy/semi-democracy/democracy has the next

biggest. Ethnic fractionalisation represented a lower effect. Whether or not the

country in question was in a “bad neighbourhood” represented the lowest impact. Not

all variables when measured as part of a model were still significant at the 99% level,

although GDP and autocracy/semi-democracy/democracy were. The significance of

increasing ethnic fractionalisation was under the 90% level, which is too low for us to

accept that it is making a significant effect. Whether or not the country was in a “bad

neighbourhood” only resulted in significance at just over the 50% level, which is far

too low for us to say with confidence that this variable is making a significant

contribution to the model.

V – Discussion

The results, and especially the importance of per capita GDP, seem to suggest

that “bad domestic problems” are the key factor within the explanation; here this

refers to mass-level internal issues (see Brown, 2001, p. 15). Similarly, other domestic

issues are no doubt of importance (how democratic the country is, for example). The

Page 25 of 29
only external issue that was analysed was the issue of bad neighbours, but this pointed

to a far weaker relationship. Obviously it is too early to say that domestic issues are

far more important for political instability than external issues but the results are

suggestive of this. To test this proposition more fully it would be useful to create

variables to deal with the magnitude of geographically close conflicts and variables to

deal with conflicts against the country in question which arose from a non-bordering

state.

The low importance of ethnic fractionalisation was surprising, although the

direction seems to lend support to Gurr’s analysis (1993, p. 161), contra Collier and

Hoeffler (2000, p. 8).

The results also seem to be suggestive of the interpretation that economic

factors are better predictors of political instability than political factors. However

there is a concern that the full relationship is actually;

“political instability  reduced development and investment  reduced wealth”.

An analysis which is consistent with the findings of Shaw (2003, p. 489), Collier and

Hoeffler (2000, p. 2) and Alesina and Perotti (1996, p. 1203), although contrary to

Przeworski and Limongi (1997, p172). Choosing 1997 for the year to measure per

capita GDP and 1998 for political stability, and thus introducing a small time lag,

does not fully avoid this problem because of the possibility of unstable countries

staying so for significant amounts of time. This could perhaps be resolved by a small-

n qualitative study, attempting to discover which change precedes which.

Page 26 of 29
 There were several other concerns with this research which if more time and

resources were available would need to be addressed. Firstly Kaufman’s political

stability variable is far from a perfect measure because it is based on perceptions of

the likelihood of violence rather than on actual violence; so the USA’s score dropped

sharply following the attacks on the 11 th of September 2001, even though the actual

likelihood of politically motivated violence stayed largely constant (World Bank,

2007). Whilst the data in question did not cover this time period similar effects could

potentially occur during the time period under study in different countries. Secondly,

most of the data used represented a single year within each given country. Whilst this

isn’t necessarily a problem, Parvin rightly noted that “Representing the average value

of several years as one point has the benefit of averaging out a large portion of the

transient or random fluctuations of the variable” (1973, p. 288). It may be advisable in

future research, then, to attempt to achieve this averaging. Thirdly no attempt was

made to deal with the relationship between income inequality and political instability.

This was partly because of issues relating to how to measure and evaluate the

institutions which society has to manage inequality and partly because of the poor

quality of the data available (see Weede, 1981, p. 641 and Cramer, 2003, p. 397).

Finally to fit the assumptions of the multivariate analysis linearity had to be assumed

even though we should not expect this to be the case. Indeed, chart 7 seemed to

provide some persuasive evidence to suggest that this certainly isn’t the case. In a

more complete analysis the curvilinear assumptions should be built into the

multivariate analysis in order to see the variable’s true weight.

Page 27 of 29
Bibliography

Web sources

CIDCM (2006), http://www.cidcm.umd.edu/polity/data/variables.asp, accessed on

21/1/08

World Bank (2007), http://info.worldbank.org/governance/wgi2007/faq.htm, accessed

on 10/1/08

Journal Articles

Alesina, A. at al (2003) “Fractionalization” Journal of Economic Growth 8. pp. 155-

194.

Alesina, A. and Perotti, R. (1996) “Income distribution, Political Instability and

Investment” European Economic Review 40. pp. 1203-1228

Bermeo, N. (2003) “What the Democratisation Literature Says – Or Doesn’t Say –

About Postwar Democratisation” Global Governance 9. pp. 159-177

Byman, D and Van Evera, S (1998) “Why they fight: Hypotheses on the causes of
contemporary deadly conflict”, Security Studies 7, pp. 1 - 50

Case, W. (1996) “Can the "Halfway House" Stand? Semidemocracy and Elite Theory
in Three Southeast Asian Countries” Comparative Politics, 28. pp. 437-464.

Collier, P. and Hoeffler, A. (2000) “Greed and Grievance in Civil War” World Bank
Policy Research Working Paper 2355. pp. 2-44.

Cramer, C. (2003) “Does inequality Cause Conflict?” Journal of International

Development 15, pp. 397–412.

Gurr, T (1993) “Why Minorities Rebel: A Global Analysis of Communal

Mobilization and Conflict Since 1945”. International Political Science Review, 14.
pp.161-201

Hegre, H. et al. (2001) “Towards a Democratic Civil Peace? Democracy, Political

Change and Civil War 1816-1992” American Political Science Review 95. pp.33-48

Kaufman, D. et al. (2007) “Governance Matters VI: Aggregate and Individual

Governance Indicators 1996-2006” World Bank Policy Research Working Paper 4280

Lichbach, M. (1989) “An Evaluation of "Does Economic Inequality Breed Political

Conflict?" Studies” World Politics, 41. pp. 431-470

Parvin, M. (1973) “Economic Determinants of Political Unrest: An Econometric

Approach”, Journal of Conflict Resolution 17, pp. 271-296

Page 28 of 29
Przeworski, A. and Limongi, F. (1997) “Modernization: Theories and Facts” World
Politics 49. pp.155-183

Shaw, T. (2003) “Regional Dimensions of conflict and peace building in

contemporary Africa” Journal of International Development 15, pp. 487–498

Weede, E. (1981) “Income Inequality, Average Income, and Domestic Violence” The
Journal of Conflict Resolution, 25. pp. 639-654.

Books and Edited Volumes

Brown, M. (2001) “The causes of Internal Conflict” in Brown et al (eds.) Nationalism

and Ethnic Conflict. London: MIT press. pp. 3-25

Field, A (2005) Discovering Statistics Using SPSS (second edition). London: Sage.

Healey, J (2005) Statistics: a tool for social research (seventh edition). London:
Thompson

Mansfield, E. and Snyder, J. (2004) Electing to Fight: Why Emerging Democracies go

to War. London: MIT Press

Snyder, J. (2000) From Voting to Violence: Democratisation and Nationalist Conflict.

London: Norton.

Page 29 of 29