From oil to democracy,
the modernization mechanism
evaluated
Didi de Jong | 10896325 University of Amsterdam
Master Thesis Political Science | International Relations Word count 13.521 | Final version | 24 June 2016 Supervisor | dr. F. Boussaid
Preface
The following thesis spurs from a combination of genuine personal interest and a lack of detailed research on this specific topic. Ever since following an introductory course in political economy in the pre-master programme, anything on the intersection between politics and economics has interested me greatly. Throughout more courses and my weekly reading of the economist, I found that my interest lies with topics such as development of which economics are an important factor. When I first came across the resource curse, it struck me as a contradictory topic, why would a country with so much easy income still not be able to develop a strong economy?
Continued interest in this lead me to my research topic. The formulation of a research question came rather easy in consultation with my supervisor. The process of defining methods, finding data and everything else, however, was not as easy. I initially planned to make this a mixed-methods project. However, due to time constraints and a lack of data, this was scaled down to a quantitative study. Even though I was disappointed, I feel like what I’ve been able to put on paper using just quantitative data still has the possibility to contribute to the literature in a significant way.
Overall, I’ve enjoyed the process of writing this report a lot. Even though at times it was frustrating and I got stuck several times, it was a very valuable experience. I hope you enjoy reading this thesis as much as I’ve enjoyed writing it.
Acknowledgements
I would like to thank Farid Boussaid, my supervisor, for his guidance and feedback throughout the project. There were many moments where I was stuck, and he got me back on track. I would also like to thank Marijn Landman en Merel Hendriks, with whom I had valuable conversations on methods and I could share the endless frustration of having to write a large research project.
Abstract
The purpose of this study is to provide insight into the relationship between oil and democracy by taking a closer look at the modernization mechanism which is argued to connect the two. An in-depth examination of this mechanism has not been done to date. The modernization mechanism proposes that oil leads to less democracy through a lack of economic diversification, which leads to an underdeveloped education sector, because of which populations lack certain norms and values which would otherwise induce democracy. To provide more detailed insight the study adopted a correlational design, using OLS simple and multiple regressions with different operationalizations of the mechanism.
The study evaluated two causal mechanisms. The first was the classic modernization mechanism. The results suggest that this mechanism holds, even though problems of reverse causality cannot be ruled out completely. The second causal mechanism concerned the effect of female education as a mediating factor. Contrary to expectation, Oil-richness seems to influence female education positively, indicating that oil could lead to more democracy through the education of women.
Table of contents
1. Introduction ... 7
2. Literature review ... 9
2.1 Education and democracy ... 9
2.2 Oil and democracy... 11
2.3 Oil, economic diversification and democracy ... 11
2.4 Oil, education and democracy ... 13
3. Theory ... 13
4. Methodology ... 16
4.1 Research approach ... 16
4.2 Case selection ... 16
4.3 Operational definitions and instruments ... 17
4.4 Methods ... 21
4.5 Assumptions and limitations ... 21
5. Multiple regressions: what explains variance in democracy? ... 24
5.1 Descriptive statistics ... 24
5.2 Multiple regression models causal mechanism one ... 25
5.3 Multiple regression models causal mechanism two ... 33
6. Simple regressions: the relationship broken down ... 36
6.1 The overall connection between oil and democracy ... 36
6.2 Oil-richness, economic diversification and democracy ... 36
6.3 Education measured in years of attainment ... 36
6.4 Oil and the gender gap in education ... 37
6.5 Self-expression values ... 38
6.6 Education measured by quality ... 39
6.7 Simple regressions summarized ... 41
7. Conclusion ... 43
7.1 Introduction ... 43
7.2 Summary of findings ... 43
7.3 Theoretical implications ... 45
7.4 Practical implications ... 46
7.5 Limitations and alternative explanations ... 47
7.6 Suggestions for future research ... 48
To what extent does Oil-richness deter democracy through the modernization mechanism? ... 49
9. Reference list ... 50
A. Appendix A: Descriptive statistics ... xlvi
B. Appendix B: Multiple regressions causal mechanism one ... xlviii
a. Self-expression values ... xlviii
b. Education in years of attainment ... liv
c. Years of attainment and self-expression values ... lxiii
d. Quality of education ... lxvi
e. Quality of education and self-expression values ... lxvii
C. Appendix C: Multiple regression models causal mechanism two... lxxiii
D. Appendix D: Simple regressions ... lxxv
a. Overall relationship between Oil-richness and democracy ... lxxv
b. Relationship between Oil-richness and economic diversification ... lxxv
c. Relationship between economic diversification and democracy ... lxxvi
d. Relationship between education in years and democracy ... lxxvii
e. Relationship between Oil-richness and education in years ... lxxx
f. Relationship between economic diversification and education in years ... lxxx
g. Relationship between self-expression values and democracy ... lxxxi
h. Relationship between Oil-richness and self-expression values ... lxxxiv
i. Relationship between economic diversifaction and self-expression values...
...lxxxiv
j. Relationship between educational quality and democracy ... lxxxv
k. Relationship between Oil-richness and educational quality ... lxxxvi
l. Relationship between economic diversification and educational quality. lxxxvii
m. Relationship between educational quality and self-expression values ... lxxxvii
E. Appendix E: dataset considerations... lxxxix
a. Cases excluded from all analyses ... lxxxix
b. List of countries included in the multiple regression analysis for causal
mechanism one ... lxxxix
F. Appendix F: datalabels ... xc
List of figures and tables
Figure 1: overall causal mechanism ………...………... 11
Figure 2: specified causal mechanism one………...………...…………... 11
Figure 3 specified causal mechanism two ………...………...…... 12
Figure 4: causal mechanism specified with variables used ……...………...….. 42
Figure 5: repetition of causal mechanism one ...43
Figure 6: adapted causal mechanism two ... 46
Table 1: correlation scores of educational quality measures and educational attainment measurements...………... 22
Table 2: correlation scores of educational quality, values and educational attainment measurements with democracy.…...………... 25
Table 3: multiple regressions of Oil-richness, self-expression values and control variables on Freedom House democracy indicators....………... 27
Table 4: multiple regressions of Oil-richness, self-expression values and control variables on Freedom House democracy indicators ………... 28
Table 5: multiple regressions of Oil-richness, self-expression values, total attainment and control variables on Freedom House democracy indicators ... 30
Table 6: multiple regressions of Oil-richness, educational and control variables on Freedom House Political Rights and Civil Liberties ... 31
Table 7: multiple regressions of Oil-richness, ratio total attainment and control variables on Freedom House democracy indicators ………... .34
Table 8: multiple regressions of Oil-richness, ratio total attainment and control variables on Freedom House democracy indicators ………... 35
Table 9: single regressions of Oil-richness on Math, Science and Language scores ... 40
7
1. Introduction
“Fast cars whizz around, malls are full of expensive luxuries and cranes dominate the skyline. But scratch the shimmering surface of the Gulf and you soon find countries hurting from the low oil price, currently around $40 a barrel. Growth is slowing and unemployment is
rising. Policymakers even dare utter a three-letter ‘t’ word until recently taboo: tax” (The economist 2016: Para 1.) With oil-prices at an historical low, the world’s eyes are once again turned to oil-states. The economies of the bigger oil producers are crumbling while some of the smaller Gulf states are making moves towards diversifying their economies or turning to tourism to fill up their drying up reserves. Their governments are feeling the consequences of being reliant on natural reserves. It is a well known phenomenon called the resource curse at play. Oil has been proven to have negative consequences for states in terms of economic development, democratization and conflict.
Most developed and prosperous countries in the world today are democracies; most oil-states are not. States that are highly resource dependent are more likely to have autocratic governments and extractive institutions and are less likely to become democratic. But why is this? Many have tried to answer the question of what leads countries to develop into democracies and why there seems to be a high number of oil-states that fail to do so.
Democracies share certain characteristics apart from their geographical concentration and shared cultural heritage. Most are highly developed welfare states, have great education systems and are so-called knowledge economies. Many scholars propose that education is highly important when it comes to becoming a democracy. Democracies are argued to emerge through the so-called modernization mechanism: economic development and diversification leads to a demand for highly skilled people. Because people are more highly educated, they will have different norms and values. Those norms and values are more conducive of democracy than any other political system, and thus populations will demand democratization.
In contrast to democracies, oil states have populations that are less well educated (Gylfason 2001). This is a possible explanation for the lack of democracy amongst them. The relationship between oil, education and democracy has been researched multiple times. The results are generally confirmative. Ahmadov (2013) finds evidence of a negative relationship
8 between oil, education and democracy and Barro (1999) finds that being an oil state or not influences the relationship between education and democracy in a negative way. How this mechanism works, though, remains unclear. Even though many studies have looked at the negative consequences of oil for democracy, see for example Ross (2001); Ahmadov (2013), the relationship between these variables has only been shown to exist within large n studies, in which the variables included either education or oil as a dummy measure.
As said, no study to date has systematically explored how the causality between oil, education and democracy runs exactly and in detail. This study therefore aimed to answer the question: to what extent does Oil-richness deter democracy through the modernization mechanism? This thesis thereby aims to contribute to the study of the relationship between oil and democracy by focusing on the role played by modernization, and by extension the salience of education.
The purpose of this correlational design was to first quantitatively assess what combination of variables explains the incidence of democracy best by using multiple regressions. On the basis of this initial exploration, tentative conclusions were drawn. The second part of the analysis focused on breaking down the relationship into small steps, to show how causality runs in detail.
In the following report, first of all an overview of the most important authors within the research fields of the resource curse, education and democratization will be given. In the following section, their theories and research will be used to formulate causal mechanisms to be tested through regression analysis. Thirdly, a comprehensive description of methods and data used will be given. This is followed by a results section, which is split into two parts. The first part of the results section will focus on the multiple regressions with the aim to explain the occurrence and levels of democracy with different combinations of variables. The second part will focus on breaking down the causal relationship through which oil leads to less democracy. Lastly, the results and analysis will be summarized in the conclusions section.
9
2. Literature review
The following section is split in four parts: firstly, the literature concerning education and democracy will be addressed. The different mechanisms by which education is said to lead through democracy will be introduced and the significance of female education will be addressed. Secondly, the relationship between oil and democracy is described and the mechanisms by which more oil theoretically leads to less education. Thirdly, the economic consequences of oil will be reviewed in light of the literature on the modernization mechanism. Lastly, the research in which the relationship between oil, education and democracy has been mentioned before will briefly be summarized.
2.1 Education and democracy
Many authors have written about the relationship between education and democracy. The classic argument is put forward by Lipset (1959) and has since been called the modernization thesis. Simply put, the modernization thesis forwards the idea that as a country becomes more prosperous economically; social developments such as an educated population will make it more likely to become democratic. Lipset (1959: 80) concludes “if we cannot say a high level of education is a sufficient condition for democracy, the available evidence does suggest it becomes close to being a necessary condition in the modern world”.
Since Lipset's famous article, extensive research on the relationship between education and democracy has been conducted. Most authors find high correlations and significant evidence of a positive relationship linking education and democracy (Barro 1999; Cueresma & Abbasi-Shavazi 2010). Although others argue that with different statistical methods a different result is reached (Acemoglu et al. 2005), they have since been refuted (Castelló-Climent 2008). Overall, the described relationship seems to hold for a variety of measures and operationalizations of both democracy and education.
The mechanisms through which education is said to democracy have been tested by multiple scholars. Glaeser, Ponzetto and Schleifer (2007) propose that the relationship works through a socialization mechanism. As people become more educated, they become more participatory in a great many associations and organizations. This leads to them becoming politically active, leading to an increased pressure for democracy. The logic is simple: “a poorly educated populace is easier to oppress over long periods than is a well educated population that knows full well what is lacking” (Gylfason 2006: 4).
10 Another, already mentioned above, much-researched mechanism is modernization. Modernization theory proposes that economic developments lead to an increase in democracy through the changing of mass values. These values change because of education, the fact that economic development leads to a more diversified economy, better educated population and a ‘knowledge society’ in which populations have more democratic values (Inglehart & Welzel 2010).
Inglehart and Welzel make use of the World Values Survey and European Values Survey to prove which types of values influence democracy. In an important article, Inglehart proves that self-expression values are more important than attitudes towards democracy in terms of predicting democratic change (2003: 54). Amongst self-expression values are political action, tolerance of homosexuality, trust in others, happiness and materialism. The materialism/post-materialism index measures the extent to which an individual is occupied with material or survival in turn for self-enrichment.
Others use demographic determinants to argue that as a population ages and the fertility rate drops, meaning that an older cohort of the population is better educated, the pressure for democracy will rise (Cueresma & Abbasi-Shavazi 2010). The same authors also find evidence that an increase in female education is very likely to increase democratization.
The significance of female education for democracy can again be explained by the modernization mechanism: more participation of women in jobs, because of more economic diversification, will lead women to be more educated and thus increase pressure for democracy through changing values. The ratio male-female education is therefore important for seeing how much democratic pressure there is. There has, however, been very little research establishing why female education has such an effect on democracy.
There is an issue of reverse causality present within the theory. Some scholars have argued for the reverse relationship: democracy increases female education (Silova & Magno 2004; Brown 2004). Likewise, as mentioned above, it has been argued by Acemoglu et al (2005) that democracy could lead to higher and better education for both women and men. Democracy does lead many states to develop into welfare states. And more generally, higher levels of development are accompanied by higher levels of education.
11
2.2 Oil and democracy
Oil has been shown to have many effects: it is bad for economic development, makes for less democracy and increases civil war. More recently, work has been conducted showing that oil increases patriarchy (Ross 2008; 2012). In other words, Oil-richness makes for less equality between men and women.
Not unlike the study of education and democracy, a lot of research has been done on the relationship between oil and democracy. It has more or less been established that oil hinders democracy (Ahmadov 2013). Ahmadov conducted a meta-analysis of almost all large-n studies concerning the link between oil and democracy. He concludes that “this study … found that there is a small, in meta-analytic terms, but nontrivial negative association between oil and democracy across the globe” (2013: 22). However, the causal mechanisms which are theoretically available to explain this link have not been researched extensively.
Ross (2001: 328) tests three possible explanations: the rentier effect, the repression effect, and the modernization effect. The rentier effect “suggests that resource-rich governments use low tax rates and patronage to relieve pressures for greater accountability” (Ibid: 328-329). The repression effect implies that oil-rich governments are able to build greater security forces and thus repress their population more effectively. And lastly, the modernization effect, which postulates that because of oil, the social and cultural changes that produce democracy do not take place.
In his analysis, Ross finds tentative evidence for all three mechanisms. Ahmadov, on the other hand, does not. He only analyses the modernization effect because of the available data and he finds that only one out of three predictors of the modernization mechanisms holds: education. The theoretical argument for the modernization effect (or anti-modernization effect) of oil is as follows: because populations do not get incentives to diversify their economic activities and to get educated to do so, the economy will not modernize and modernization is a prerequisite for democracy. The evidence for this hypothesis, however, is slim (Ross 2001).
2.3 Oil, economic diversification and democracy
As mentioned above, oil is said to deter economic development. This is the link by which oil is said to influence democracy through modernization mechanisms. There is a multiplicity of economic outcomes oil is said to affect. According to Ross (1999), the most
12 common explanations are the declining terms of trade for oil-exports – the simple fact that their prices fall over the long run -, the instability of oil-prices – and the economic volatility that comes with it -, poor economic linkages between resource and non-resource sectors – the lack of investment of resource money in other sectors – and the Dutch Disease – the negative economic effects of oil-booms.
Most of the explanations surveyed in Ross’ literature review (1999) are related to the price volatility of oil-exports. Because of this price volatility, negative economies effects take place. When prices go up, the oil-sector crowds out the other sectors, leaving them undeveloped. When prices go down, economic growth falters and leaves no room for the development of other sectors because of the lack of funds. Even though it has also been argued that it is precisely this volatility of oil-regimes that makes democratization more likely (Haggard & Kaufmann 1995), this will not be taken into account.
Overall, it is clear that oil-rich countries fail to diversify their economies away from oil. It is a clearly observable pattern, and even though economies such as Kuwait and the United Arab Emirates are praised for their attempts to diversify their economies (Herb 2009), their oil-exports still make up a large part of their GDP.
Even though a lack of economic diversification and economic volatility seem inherent to oil-economies, this is not clearly linked to democracy. What is, however, clearly linked to democratization, is the above cited argument by Lipset (1959): economic modernization leads to democratization. The above explained modernization mechanism once again comes into play. Diamond (1992) retested the Lipset hypothesis using updated methods of analysis. He finds the same results, although he does stress that economic development is not a prerequisite for democracy. Democracy can also lead to economic development.
Gylfason takes political diversification and economic diversification as two processes that occur simultaneously, and tries to show their effect on economic growth. “The pattern shown suggests a direct relationship between economic and political diversification. The [findings] suggest that liberalization from excessive reliance on natural resources goes along with increased political freedoms and vice versa. Put differently, natural capital tends to crowd out social capital and vice versa” (2006: 5).
If political diversification and economic diversification go hand in hand, but economic growth leads to democracy through the modernization mechanism, it is easy to see why
13 economic diversification is also important to democracy: diversification is a highly important aspect of economic growth (Hesse 2006). And, as argued in case study research on the United Arab Emirates, a good education system is a prerequisite to economic diversification. If there are no people to fill the new positions, diversification will not succeed (Muysken & Nour 2006).
2.4 Oil, education and democracy
Finally, both in the field of study that concerns itself with oil and democracy and the field of study which concerns itself with education and democracy, the modernization effect is a predominant explanation. In both fields of study, whenever one or the other variable was included, it had a significant negative sign. Barro (1999) includes a dummy variable on whether or not a country is an oil-country in his analysis of the relationship between education and democracy. The dummy is negative and significant. As mentioned above, Ahmadov (2013) finds the same relationship over a large number of studies.
Gylfason (2001) shows that several educational indicators such as expenditure on education and years of schooling for girls are negatively related to resource-income. So it seems that oil has a negative influence on education, which in turn has a negative influence on democratization. In other words, oil seems to negatively influence education, and as a consequence, the values that make a country democratic or not. Oil deters modernization, which leads to less chance for democracy. However, it is not clear how Oil-richness actually leads to less education or how Oil-richness influences values.
3. Theory
As mentioned above, the causality from more oil to less democracy is most likely to run through the modernization mechanism: countries with more oil are less likely to diversify their economies and less likely to educate the female cohort within their populations. This leads to fewer incentives for people to be more educated and a less well developed education system. This, in turn, leads to a lack of values that are conducive of democracy and the country becomes less likely to democratize. Therefore, on the basis of the above explained theory, an overall causal mechanism can be specified:
Figure 1: overall causal mechanism
More oil
Lack of
modernization
Less pressure for
democracy
14 This causal mechanism depicts the overall hypothesis: more oil leads to fewer values conducive of democracy, which in turns means that a country will not democratize. The modernization mechanism in turn, could run through two different causal paths. First, modernization could fail to occur through a failure to diversify the economy. Oil-richness leads to less economic diversification because it will provide a government with enough funds to support its population without having to create high economic diversification and high-value added sectors. Because of this lack of economic diversification and the jobs it brings with it, the education system will remain underdeveloped because there is no need for highly educated people. This will lead populations to be less educated, and a lack of education leads to a lack of the values and societal pressures that ultimately cause a country to become a democracy.
This operationalization of the modernization mechanism is depicted below, and will serve as the first subsidiary hypothesis in this thesis: more oil leads to less economic diversification, which in turn leads to an underdeveloped education system. Therefore, values conducive of democracy will be less salient within the population, which decreases the chance a country is a democracy.
Figure 2: specified causal mechanism one
Secondly, as argued above, more oil could lead to less education for women, because of the same problem with economic diversification. It is unclear within the literature how this happens, but it could have something to do with Ross’ recent argument that Oil-richness increases patriarchy. Women therefore will go to school less and be educated worse than men. It has been shown that the more females are educated, the higher the chance for democracy is (Cueresma and Abbasi-Shavazi 2009). Although this second causal mechanism is slightly less well founded, it will be tested throughout the research.
Figure 3: specified causal mechanism two
This causal mechanism, in which the lack of democracy stems from a lack of female education, is shown below. It will serve as the second subsidiary hypothesis in this thesis:
More oil Less economic diversification Less education conducive of Less values democracy
Less chance for democracy
15 more oil leads to less education for women, the lack of which leads to a lower chance for a country to be democratic.
16
4. Methodology
In the following, firstly the research approach and design will be explained. Secondly, case selection will be discussed. Thirdly, the section will list the operational definitions of the variables tested, where the data was obtained from, and how it was manipulated to be included in the dataset for this study. The reliability of the data sources will be discussed briefly. Lastly, the specific statistical methods used will be explained and the assumptions and limitations of the methods will be discussed.
4.1 Research approach
This study employed a quantitative approach. A quantitative research approach is “an approach for testing objective theories by examining the relationship amongst variables. These variables, in turn, can be measured, ..., so that the numbered data can be analyzed using statistical procedures” (Cresswell 2014: 5). Within this field, there are multiple subtypes. The research design that was used in this particular study is often dubbed a correlational design, a type of non-experimental design (Cresswell 2014: 12). Correlational research is a “non-experimental method because … [it] lacks manipulation of an independent variable which is under the control of the experimenter and random assignment of participants is not possible” (Johnson 2001: 5). This means that the variables are observed as they occur, without manipulation. Within correlation research, the researcher looks for association between variables through the use of a multiplicity of correlational statistical methods, such as regression analysis (Ibid).
4.2 Case selection
Case selection was done on the basis of available data. Since the research concerned the testing of a hypothesis for states, the population is all states. The world bank aims to reports data for all states in the world, so their country list was taken as the basis for the dataset. Thirteen states were excluded from the dataset on beforehand, because they did not have any data on any indicator. These states can be found in appendix E.a.
The sample used for the different models differs on the basis of available data in the datasets used. Before running a set of models for the different operational definitions of the modernization mechanism, all cases with missing variables were excluded from the analysis. Therefore, across models that employ the same indicators, which are compared in the results section, the n has been kept consistent. The exact sample size per set of models is given
17 below. Both oil rich and oil poor countries were used in the regression analysis. This was done to make sure that if a relationship between oil and education exists and the modernization mechanism is valid this counts for countries with all levels of oil-income. If this did not hold, oil cannot be an explanation for lack of education.
All data are from the year 2010, for which the data of most datasets is available. 2010 was selected because it is recent, before the drop in oil-prices and a year in which the consequences of the 2008 economic crisis are already less hard-felt and economies are naturally growing again.
4.3 Operational definitions and instruments
As can be seen in the two proposed causal mechanisms in chapter 3, a total of 5 terms need to be operationally defined the first causal mechanism and one extra for the second causal mechanism. The terms that needed to be defined for the first causal mechanism were: Oil-richness, economic diversification, education, values conducive of democracy and democracy. For the second causal mechanism female education had to be operationally defined. The regression models were controlled for GNI per capita as a measure level of economic development and Islam as a dominant religion as a measure of cultural difference. The terms will be defined below in the above mentioned order and the datasets the data are taken from are discussed directly below the operational definition.
Oil-richness can be defined in multiple ways, and is often measured by taking the oil export to GDP ratio to capture the importance of external dependency on oil. This study follows Ross (2001) and uses the same oil export GDP measure. This way, the variable will capture the oil dependency of a given country’s economy, which is relevant to the modernization mechanism. The data was obtained from the World Bank’s trade dataset, in which Petroleum income as a percentage of total merchandise exports is reported. The measure was constructed by multiplying petroleum income as a percentage of merchandise exports with total merchandise exports and dividing this number by total GDP. The data in the World Bank database comes from national account data and is reliant on the reliability of national data collection (World Bank 2016).
Economic diversification was measured by a variable representing the value added to economies by things other than oil. Economies with high economic diversification will have high value added in the services and manufacturing sectors, and less value added in
18 agriculture and industry. Value added by industry and agriculture are excluded, because these are both low value added activities and industry includes income from natural resources. The final economic diversification variable is a sum of two World Bank variables: Value Added by manufacturing (% of GDP) and Value Added by services (% of GDP).
The second part of the modernization mechanism, education, can be made operational in different ways. Within this thesis, two different operationalizations of education were used to see which one has the most explanatory power. Education was made operational by a variable expressing total years of attainment and by a variable that tries to capture the quality of education.
The first measure of education, which is usually positively correlated with democracy and will therefore be tested against oil is education measured by years of attainment (Barro 1999). The data was obtained from the Barro-Lee dataset on educational attainment, which is available online. The dataset reports years of attainment for population over 15 and over 25. For this study, the years of attainment of the population above 25 was taken, because people above 25 can still be following tertiary education. The dataset is constructed on the basis of national census data (Barro & Lee 2010) and is reliant on the reliability of data collection by national governments.
The second way to operationalize modernization through education is to look at education quality, in other words, how well education within a given country performs. To test whether educational quality is affected by oil, the study makes use of the TIMMS and PRILS datasets. TIMMS and PRILS are two surveys conducted across countries every four or five years. They measure math, science and language abilities of fourth and eighth grade children. The data is collected at a national level using a survey provided by the research institution. Sampling procedures are provided to the national handler of the survey, and are checked by the research institution. Overall, the procedures are defined detailed to ensure reliability across national samples. The data used in this thesis are the country averages reported in the overall assessment reports for fourth grade children, as these were more complete.
The third part of the modernization mechanism is values conducive of democracy. According to Inglehart (2003) the most important determinants of democracy are self-expression values. He defines these as being: political action, happiness, trust in others, justifiability of homosexuality and post-materialism. These self-expression values are highly
19 correlated with democracy, as mentioned above (Inglehart 2003). The most comprehensive datasets of values are the European Values Survey and the World Values Survey, which are administered repeatedly over a number of years. The surveys cover a great number of topics, but will only be analyzed following Inglehart (2003), who defines the above mentioned 4 items (political action, happiness, trust in others and Justifiability of Homosexuality) and one index variable (post-materialism) as being highly correlated with democracy. The items used in the dataset to measure these self-expression values can be found in Appendix A.
In order to obtain these data from the European Values Survey (EVS) and World Values Survey (WVS) datasets, which are not available in aggregated format, the two datasets were combined. The five items which are used in this research were coded so that higher values presented more self-expression. In other words, when the score on one of these variables goes up, the score on democracy should be higher. The data were then averaged for each country year. And finally the data of wave 4 of the WVS and wave 5 of the EVS were filtered out. The reliability of the EVS and WVS datasets are ensured by random sampling procedures on a national basis. All surveys are conducted by trained interviewers. The questionnaire itself has detailed procedures noted on it, so the interview is highly scripted and consistent across participants.
For democracy the Freedom House dataset were used. Freedom house measures freedom in the world by scoring countries on two dimensions: Civil Liberties and Political Rights. Freedom house defines the two as
“Political rights enable people to participate freely in the political process, including the right to vote freely for distinct alternatives in legitimate elections, compete for public office, join political parties and organizations, and elect representatives who have a decisive impact on public policies and are accountable to the electorate. Civil liberties allow for the freedoms of expression and belief, associational and organizational rights, rule of law, and personal autonomy without interference from the state” (2012: Para 3).
The effects of education and Oil-richness were calculated for both the Civil Liberties and the Political Rights indexes. Within these indexes, a higher number stands for a lower score on the two dimensions of democracy. The freedom house defines democracy as electoral democracy, and to be rated as such, a country needs to have the following four characteristics:
20 “a competitive, multiparty political system; universal adult suffrage for all citizens; regularly contested elections conducted in conditions of ballot secrecy, reasonable ballot security, and in the absence of massive voter fraud, and that yield results that are representative of the public will; significant public access of major political parties to the electorate through the media and through generally open political campaigning” (Ibid: Para 5).
Logistic regressions were run for their thus constructed binary measure of democracy. The coding is done by a team of scientist and regional experts to ensure reliability.
Within the second causal mechanism which includes female education, the same operationalization of Oil-richness and democracy was used. Following Cueresma and Abbasi-Shavazi (2009), a highly correlated measure of female vs. male education with democracy is a ratio variable of total years of education. Female education will therefore be made operational as a ratio-variable in which total years of education for female population over 25 is divided by total years of education for male population over 25. The data is obtained from the above mentioned and explained Barro-Lee dataset on educational attainment.
Within the regression analysis, next to the above mentioned variables, a number of control variables are used. Important control variables are economic development and Islam. Oftentimes, the argument is made that culture affects democracy, especially religion. To counter this argument, a variable expressing Islam as the dominant religion was be added to the analysis. Economic development equally is often said to influence the democratic outcome significantly. Therefore, to counter the argument that it is all about the development of a country, a control variable for this explanation was added as well.
The two are made operational as GNI per capita and Islam as a dominant religion, since these have been shown to respectively have a high correlation with education and oil-dependency. GNI per capita is taken from the world bank database. The variable Islam was calculated from the World National Religion Dataset. It is obtained by coding countries where total percentage of Islam is higher than total percentage other religion as 1 and others as 0. An overview of all used variables can be found in appendix A.
21
4.4 Methods
This study employed multiple statistical methods. Firstly, multiple regression and logistic regression was were to see what different operationalizations of the modernization mechanism explain the occurrence of democracy best. In other words, regression analysis was used to establish which set of variables best predict democracy when put together. The second phase of the study looked at simple regression models across variables to establish likely causality and see what operationalizations of the modernization mechanism can be said to be correlated with all other indicators: democracy, economic diversification, self-expression values and Oil-richness.
Regression is a statistical method which “uses a formula (usually a straight line) to approximate how the expected value (the mean) for [the dependent variable] changes at different values of [the independent variables]” (Agresti & Franklin 2014: 583). In multiple regression SPSS reports B-slopes, which show in what direction and to what extent certain variables influence the regression line. The software also reports the multiple correlation, which “describes the association between [the independent variable] and a set of explanatory variables in a multiple regression model” (Ibid: 637). The multiple correlation is denoted R2 and can be interpreted as the amount of variance explained in the dependent variable by the independent variables. The other method used, logistic regression, is another regression method in which the outcome variable is binary. The model predicts the chance that with certain values the outcome will be either one of the two possible outcomes (Ibid: 665).
4.5 Assumptions and limitations
To yield reliable results, the data need to comply with a number of assumptions. The assumptions for regression analysis are that data is gathered using randomization and the population values are normally distributed. For multiple regressions an additional assumption is that the sample size should be at least 10 times the number of explanatory variables (Agresti & Franklin, 2014: 363). Randomization was not possible because of the nature of the unit of analysis: the population of all countries in the world is so small, random sampling is not possible. The population values are also not distributed normally for all variables. However, regression is generally robust to a violation of these assumptions. The assumption of sample size is more problematic, since because of missing data, it is not satisfied everywhere.
22 The sample sizes for the different sets of multiple regression models are as follows: the sample size is 63 for all models including self-expression values, 109 for models including educational attainment and 32 for models including measures of educational quality. For the second causal mechanism, the sample size is 118. One can see that in the models including educational quality, the assumption is violated as soon as more than 3 predictors are put in the model and for self-expression values for models with over 6 predictors. However, this mostly influenced the results in that there was a lack of significance for these models so reliable conclusions could not be drawn.
Another threat to significance is multicollinearity between the predictors. Multicollinearity means that two independent variables are so highly correlated, neither of them has a significant effect on the dependent variable. In order to counter this problem, for each analysis multicollinearity diagnostics were run in SPSS. For two operationalizations, it turned out to be a problem. For educational quality, measured by the TIMMS and PRILS datasets the correlations between the measures were too high, as shown in the table left below.
Correlations with Math Correlations with Total attainment
Science 0.95*** Primary attainment 0.87***
Language 0.91*** Secondary attainment 0.91***
Tertiary attainment 0.77***
Table 1: correlation scores of educational quality measures and educational attainment measurements amongst
themselves, *= significant at the 0.1 level, **= significant at the 0.05 level, ***= significant at the 0.01 level
Reliability analysis yielded a Chronbach's Alpha of 0.98, meaning the three items could be reliably put together to form one scale. Since this would have limited the number of cases significantly (from 54 to 38) math was taken to in itself represent the quality of education.
The same problem occurred for total attainment, primary attainment, secondary attainment and tertiary attainment, also visible in the table right above. Cronbach's alpha for these variables is 0.83, meaning these variables could reliably be put together to make a scale. However, since total attainment is a sum-variable of primary, secondary and tertiary attainment, it was taken as the variable to measure all three in one.
Another threat to the reliability of the results is reverse causality. As already mentioned in the literature background, it is possible that better education is a consequence of democracy, rather than a cause for democracy. To counter the possibility of claims in this direction, the simple regression chapter assesses every step within the causal mechanism. The association between all variables was tested, so a causal claim can be made more reliably than
23 just on the basis of multiple regressions. Unfortunately, OLS regressions do not imply causation, neither in their simple form or in their multiple form. A real way to deal with the issue of reverse causality was therefore not employed in this research. This limitation will be discussed further in the conclusion.
24
5. Multiple regressions: what explains variance in democracy?
As mentioned above, the results section will consist of several chapters. First off all, a short description of the used variables will be given, noting the most important characteristics. Secondly, multiple regression results will be shown, to see what variables make up the best prediction on democracy, Political Rights and Civil Liberties. The multiple regression analyses start with the second part of the modernization mechanism, self-expression values, which is analyzed separately. The two different operational definitions of education: education as measured by quality and education in Years of Attainment are analyzed afterwards. Education in Years of Attainment and education measured by quality are both also analyzed in models including the self-expression values. This analysis will be followed by a preliminary conclusion citing the most important findings. In the next chapter of the results, simple regressions will be used to show how the relationship between oil, education and democracy works more specifically and how the causality runs.
5.1 Descriptive statistics
A table of the used variables and their most important characteristics can be found in the appendix (Appendix A). Several variables have surprising attributes. Oil-richness, for example, is divided with M = 7.91 and a high standard deviation: σ = 15.40. The standard deviation of the two measures used in the Freedom House index (civil liberties & political rights) differs by more than one. Civil Liberties is divided with M = 3.26 and σ = 1.18, whereas Political Rights is divided with M = 3.34 and σ = 2.16. Because all the variables come from different datasets n ranges between 43 and 202. As mentioned above, the different regressions are run for the available n for the variables used, because the number of cases that have data available for all variables is too small to be valuable for regression analysis: n = 26. All the above mentioned mediating variables have been tested to actually be associated with democracy, by the use of simple correlations. If variables are not associated, they cannot be a cause. The results can be found in the table on the next page. As can be seen, almost all used variables are significantly correlated with democracy. The exception is Trust in Others, which has a low correlation with all three measures of democracy employed.
25
Freedom House Democracy (0-1)
Freedom House Civil Liberties Freedom House Political Rights Educational quality Math scores 0.55*** -0.56*** -0.55*** Science scores 0.57*** -0.56*** -0.57*** Language scores 0.57*** -0.65*** -0.67*** Self-expression values Happiness 0.26** -0.37*** -0.33*** Trust in Others 0.06 -0.22 -0.16 Signing a petition 0.47*** -0.59*** -0.57*** Justifiability of Homosexuality 0.46*** -0.63*** -0.61*** Materialist/ Post-materialist (12-item) 0.31*** -0.34*** -0.36*** Educational Attainment Average years of total
attainment 0.42*** -0.57*** -0.51***
Average years of primary
attainment 0.45*** -0.56*** -0.50***
Average years of
secondary attainment 0.30*** -0.46*** -0.42***
Average years of tertiary
attainment 0.28*** -0.42*** -0.38***
Ratio total educational
attainment 0.34*** -0.34*** -0.31***
Table 2: correlation scores of educational quality, values and educational attainment measurements with democracy; *= significant at the 0.1 level, **= significant at the 0.05 level, ***= significant at the 0.01 level
5.2 Multiple regression models causal mechanism one
5.2.1 Self-expression values and democracy
The first multiple regression models use the variables Oil-richness, Economic Diversification, Political Action, Happiness, Trust in Others, Justifiability of Homosexuality and Post-Materialism and control for GNI per capita and Islam as a dominant religion. In Models 1-10 in appendix B.a., the effects of these variables on democracy (0-1) are measured using logistic regression. Models 1-5 present the separate effects of the self-expression values on the outcome variable. When combined with Oil-richness and controlled for GNI per capita and Islam as a dominant religion, the variables Political Action (Model 1, B = 6.630, P < 0.01, R2 = 0.65) and Trust in Others (Model 3, B = -7,534, p < 0.05, R2 = 0.57) are significant contributors. Trust in Others has a negative sign, contrary to the hypothesis that higher scores on the self-expression values contributes positively to democracy.
26 In model 6 (R2 = 0.603), the effects of all self-expression values are combined to predict democracy, the B-values for Political Action (B = 4.611, p < 0.05) and Justifiability of Homosexuality (B = 1.206, p < 0.05) are significant. When in model 7 (R2 = 0.625) the control variables are added, the significance of Justifiability of Homosexuality disappears. When all predictors are combined in model 10, this yields an R2 of 0.681.
The same regressions have been run for the Political Rights and Civil Liberties indexes and are presented in models 11-30 in appendix B.a. Once again in the multiple regressions with outcome variable Political Rights the variables Political Action (Model 11, B = -1.172, P < 0.1, R2 = 0.48) and Trust in Others (Model 13, B = 3.029, p < 0.05, R2 = 0.52) are significant contributors when combined with Oil-richness and controlled for GNI per capita and Islam as a dominant religion. When combining the self-expression values in model 16 (R2 = 0.52, F(5,58) = 12.104, p < 0.01) predicting Political Rights Political Action (B = -1.672, p < 0.01), Trust in Others (B = 3.688, p < 0.01) and Justifiability of Homosexuality (B = -0.541, p < 0.01) have significant effects. Trust in Others once again has an effect contrary to the hypothesis that more trust means more rights. When adding the control variables in model 17 (R2 = 0.56), Political Action loses its significance, and GNI per capita (B = -4.931, p < 0.05) contributes significantly. When adding Oil-richness in Model 18 (R2 = 0.52, F(8,55) = 9.981, P < 0.01) the variance explained does not rise much (0.002). When adding Economic Diversification in Model 20 (R2 = 0.59, F(9,54) = 5.721, P < 0.01) the significance of GNI per capita falls away and the significance of Political Action comes back.
The results for the same models run with outcome variable Civil Liberties have roughly the same results as the models run with outcome Political Rights. Model 30 (R2 = 62, F(9,54) = 9.590, p < 0.01) with the highest explained variance is once again the one combining the variables Oil-richness, Economic Diversification, Political Action, Happiness, Trust in Others, Justifiability of Homosexuality, Post-Materialism, GNI per capita and Islam. The biggest difference with the equal model for Political Rights (Model 20) is that Economic Diversification does not seem to contribute significantly to Civil Liberties, as can be seen in table 3 below.
The models with the highest explained variance are depicted in table 3 below. As can be seen, Political Action is a significant contributor in all models. For the models with quantitative dependent variables, Trust in Others and Justifiability of Homosexuality are also significant contributors.
27
Model 10 Model 20 Model 30
Outcome variable: Freedom House Democracy (0-1) Outcome variable: Freedom House Political Rights Outcome variable: Freedom House Civil Liberties
B (se) Beta B (se) Beta B (se) Beta
Oil-richness x1 -0.036 (0.149) 0.965 -0.025 (0.035) -0.079 -0.011 (0.028) -0.045 Economic Diversification x2 0.097 (0.075) 1.102 -0.037* (0.035) -0.239 -0.025 (0.016) -0.195 Political Action (signing a petition) x3 6.971** (3.304) 1065.535 -1.268* (0.653) -0.319 -0.934* (0.516) -0.286 Happiness x4 0.960 (3.131) 2.612 1.026 (0.988) 0.128 0.999 (0.781) 0.151 Trust in Others x5 -5.215 (4.686) 0.005 4.291*** (1.242) 0.401 3.096*** (0.982) 0.351 Justifiability of Homosexuality x6 0.282 (0.870) 1.326 -0.324* (0.178) -0.355 -0.292** (0.141) -0.389 Post-Materialism x7 3.172 (4.038) 23.862 -0.998 (1.185) -0.097 -0.246 (0.936) -0.029
GNI per capita x16 0.000
(0.000) 1.000 -2.667E-5 (0.000) -0.226 -3.086E-5 (0.000) -0.317 Islam x17 -0.327 (1.461) 0.721 -0.099 (0.622) -0.016 -0.096 (0.491) -0.019 (Pseudo) R2 0.681 0.587 0.620 Intercept -18.489 5.721 3.397 N 63 63 63
Table 3: multiple regressions of Oil-richness, self-expression values and control variables on Freedom House
democracy indicators; *= significant at the 0.1 level, **= significant at the 0.05 level, ***= significant at the 0.01 level
5.2.2 Education measured by years of attainment
The second group of multiple regression models measures the contribution of educational attainment to democracy. Models 32-35 in appendix B.b. measure the impact of Total Attainment (Model 32, R2 = 0.43), Primary Attainment (Model 33, R2 = 0.45), Secondary Attainment (Model 34, R2 = 0.39) and Tertiary Attainment (Model 35, R2 = 0.37) combined with Oil-richness and controlled for GNI per capita and Islam as a dominant religion. In all models Oil-richness was a significant contributor (p < 0.01). In model 32, Total Attainment contributes positively to democracy (B = 0.289, p < 0.01) and Oil-richness is negative (B = -0.118, p < 0.01). The same counts for Model 32, in which Primary Attainment is positive (B = 0.359, p < 0.01) and Oil-richness negative (B = 1.22, p < 0.01). Secondary and Tertiary Attainment (models 33 and 34) do not have the same significant contribution. This can be seen in table 4 below. The same models have been repeated including Economic Diversification as a variable (Models 39-42). Economic Diversification is a significant contributor in Models 36 (B = 0.087, P < 0.01) and 37 (B = 0.061, P < 0.05) but
28 loses its significance in Models 39 which includes Total Attainment and 40 which includes Primary Attainment.
Model 32 Model 33 Model 34 Model 35
Outcome variable: Democracy (0-1) Outcome variable: Democracy (0-1) Outcome variable: Democracy (0-1) Outcome variable: Democracy (0-1) B (se) B (se)
B (se) Beta B (se) Beta B (se) Beta
Oil-richness x1 -0.12*** (0.04) 0.89 -0.12*** (0.04) 0.88 -0.12*** (0.04) 0.87 -0.13*** (0.04) 0.88 Total Attainment x11 0.29*** (0.11) 1.34 Primary Attainment x12 0.66*** (0.23) 1.93 Secondary Attainment x13 0.36* (0.20) 1.43 Tertiary Attainment x14 0.76 (0.80) 2.15 GNI per capita x16 0.00 (0.00) 1.00 0.00 (0.00) 1.00 0.00* (0.00) 1.00 0.00** (0.00) 1.00 Islam x17 -1.18* (0.62) 0.32 -0.84 (0.66) 0.43 -1.42** (0.60) 0.24 -1.32** (0.60) 0.27 Pseudo R2 0.43 0.45 0.39 0.37 Intercept -1.15 -2.065 0.041 0.566 N 109 109 109 109
Table 4: multiple regressions of Oil-richness, self-expression values and control variables on Freedom House
democracy indicators; *= significant at the 0.1 level, **= significant at the 0.05 level, ***= significant at the 0.01 level
The same models have been run with outcome variables Political Rights and Civil Liberties. In Models 44-47 (appendix B.b.) the results are visible for the regressions run with Political Rights as outcome variable and the variables Total Attainment (Model 44, R2 = 0.50), Primary Attainment (Model 45, R2 = 0.49), Secondary Attainment (Model 46, R2 = 0.47) and Tertiary Attainment (Model 47, R2 = 0.44) combined with Oil-richness and controlled for GNI per capita and Islam as a dominant religion. Oil-richness, GNI per capita and Islam as a dominant religion are significant in all models (P < 0.05). These models were re-run with Economic Diversification as extra variable, as were the logistic models. Economic Diversification contributes significantly to all models but model 51, but does not change the overall results. The B-values have the expected positive and negative contributions to Political Rights, in line with the hypotheses.
The models with outcome variable Civil Liberties have roughly the same outcomes. In Models 56-59 (Appendix B.b.) the results are visible for the regressions with the variables Total Attainment (Model 56, R2 = 0.62), Primary Attainment (Model 57, R2 = 0.61),
29 Secondary Attainment (Model 58, R2 = 0.58) and Tertiary Attainment (Model 59, R2 = 0.56) combined with Oil-richness and controlled for GNI per capita and Islam as a dominant religion. The B-values of all types of attainment are significant (P < 0.1) and Oil-richness and the control variables also contribute significantly to the models. In model 60-66 Economic Diversification is added as a variable. It contributes significantly, except in model 63 which include total attainment and 64 which includes Primary Attainment. Other than that it does not change the results, the B-values have the expected positive and negative contributions to Civil Liberties, in line with the hypotheses.
On the basis of these models, we can say that total education and primary education are the most important when it comes to predicting democracy. They are more important than level of development, since GNI per capita is not significant in these models. In the models with secondary and tertiary education, GNI per capita becomes significant. This means that the explanatory power of level of development is more important to democracy than secondary or tertiary education. However, the variance explained is much higher for the models including primary and secondary education.
5.2.3 Education measured by years of attainment and self-expression values
The total causal mechanism runs through economic diversification, education and self-expression values, therefore regressions were run including all these variables as predictors. Models 67-70 (Appendix B.c.) show the overall regressions for Total Attainment, Economic Diversification, Political Action, Happiness, Trust in Others, Justifiability of Homosexuality, Post-Materialism and the two controls GNI per capita and Islam as dominant religion. In the most complete model 70 (Pseudo R2 = 1.000), none of the predictors has significance. This could have something to do with the number of cases (n = 62).
The same models were run for Political Rights and Civil Liberties. Models 73-74 in table 4 below show the results for outcome variable Political Rights. In the most complete model 74 (R2 = 0.54, F(10,52) = 5.483, p < 0.01), only Political Action (B = -1.416, p < 0.1) and Trust in Others (B = 3.330, p < 0.01) are significant contributors. The results for Civil Liberties are visible in models 75-78. In model 78 (R2 = 0.64, F(10,52) = 8.250, p < 0.01) there are different significant contributors than in model 74, which includes the sample variables. Trust in Others (B = 2.404, P < 0.05), Justifiability of Homosexuality (B = -0.267, P < 0.05) and total attainment (B = -0.0163, p < 0.05) are the significant variables in this model. The models can be found in table 5 on the next page.
30
Model 73 Model 74 Model 77 Model 78
Outcome variable: Freedom House Political Rights Outcome variable: Freedom House Political Rights Outcome variable: Freedom House Civil Liberties Outcome variable: Freedom House Civil Liberties
B (se) Beta B (se) Beta B (se) Beta
Oil-richness x1 -0.02 (0.03) -0.07 -0.02 (0.038) -0.05 -0.01 (0.03) -0.04 0.00 (0.03) 0.01 Economic Diversification x2 -0.02 (0.02) -0.15 -0.02 (0.02) -0.13 -0.02 (0.02) -0.12 -0.01 (0.02) -0.06 Political Action x3 -1.54** (0.63) -0.44 -1.42* (0.73) -0.40 -0.89* (0.47) -0.30 -0.61 (0.54) -0.21 Happiness x4 0.85 (0.98) 0.12 0.92 (1.05) 0.13 0.55 (0.74) 0.09 0.58 (0.78) 0.10 Trust in Others x5 3.28*** (1.21) 0.35 3.33** (1.25) 0.36 2.36** (0.91) 0.30 2.40** (0.93) 0.31 Justifiability of Homosexuality x6 -0.31* (0.16) -0.37 -0.28 (0.18) 0.34 -0.33*** (0.12) -0.47 -0.27** (0.13) -0.38 Post-Materialism x7 -0.42 (1.12) -0.05 -0.55 (1.19) -0.06 -0.04 (0.84) -0.01 -0.23 (0.89) -0.03 Total Attainment x1 1 -0.11 (0.01) -0.16 -0.01 (0.11) -0.14 -0.18** (0.07) -0.31 -0.16** (0.08) -0.27 GNI per capita x1
6 -0.00 (0.00) -0.01 -0.00 (0.00) -0.21 Islam x1 7 0.03 (0.65) 0.01 0.316 (0.49) 0.06 R2 0.542 0.544 0.633 0.642 Intercept 4.932 4.530 5.013* 4.362 N 62 62 62 62
Table 5: multiple regressions of Oil-richness, self-expression values, total attainment and control variables on
Freedom House democracy indicators; *= significant at the 0.1 level, **= significant at the 0.05 level, ***= significant at
the 0.01 level
5.2.4 Education measured by quality
The quality of education is measured by the variable Math scores. Models 79-80 (Appendix B.d.) show the results of the logistic regressions of Oil-richness, math-scores, GNI per capita and Islam as dominant religion with the outcome variable democracy. Models 81-82 shows the results of a regression with the same variables for Political Rights and models 83-84 for civil rights.
In the logistic regressions presented in models 79-80 the B-coefficients are not significant. In model 79, in appendix BD, one can see a multiple logistic regression analysis of Oil-richness on democracy, controlled for GNI per capita. Even though without the control and for a larger n this relationship was significant, the B-coefficients of neither predictor are significant. In the other models, the results are slightly more interesting. . In model 81
31 (F(29,3) = 6.816, p < 0.01) Oil-richness is used to predict Political Rights, controlled for GNI per capita
Model 81 Model 82 Model 83 Model 84
Outcome variable: Freedom House Political Rights Outcome variable: Freedom House Political Rights Outcome variable: Freedom House Civil Liberties Outcome variable: Freedom House Civil Liberties
B (se) Beta B (se) Beta B (se) Beta B (se) Beta
Oil-richness x1 0.09* (0.04) 0.28 0.09* (0.05) 0.29 0.06* (0.03) 0.27 0.06* (0.03) 0.27 Quality of education x8 -0.01 (0.01) -0.12 0.00 (0.00) GNI per capita x16 -0.00***
(0.00) -0.63 -0.00*** (0.00) -0.57 -0.00*** (0.00) -0.61 -0.00*** (0.00) -0.62 Islam x17 0.67 (1.23) 0.08 0.56 (1.25) 0.07 0.96 (0.89) -0.16 0.97 (0.92) 0.16 (Pseudo) R2 0.422 0.430 0.424 0.424 Intercept 3.454*** 5.907 2.830*** 2.604 N 32 32 32 32
Table 6: multiple regressions of Oil-richness, educational and control variables on Freedom House Political Rights
and Civil Liberties; *= significant at the 0.1 level, **= significant at the 0.05 level, ***= significant at the 0.01 level
and Islam as a dominant religion. Both Oil-richness (B = 0.085, p < 0.1) and GNI per capita are significantly associated with Political Rights (B = -0.00069, P < 0.01). As Oil-richness goes up and GNI per capita goes down, Political Rights go down. In all models (81-84) for the outcome variables Political Rights and Civil Liberties, Oil-richness has a significant influence (p < 0.1). Math scores, however, do not significantly influence the outcome. All overall models are significant (F(4,28) > 5, p < 0.01).
5.2.5 Education measured by quality and self-expression values
As mentioned before, the total causal mechanism runs through economic diversification, education and self-expression values, therefore regressions were run including all these variables as predictors. In this case, the operational definition of education is the quality, which is measured by math scores. The following regressions will consider a set of independent variables: math scores, Political Action, Trust in Others, Justifiability of Homosexuality and Post-Materialism, controlled for GNI per capita with the three dependent variables: Freedom House Civil Liberties, Political Rights and their measure of democracy.
Models 85-92 (appendix B.e.) present the logistic regressions for the five self-expression values and math scores. None of the predictors is significant in the models, no matter what variables are added. The explained variance does not vary across models (Pseudo
32 R2 = 1.000). Because of this lack of significance, the results cannot be interpreted. The most likely explanation for the result is the fact that within the n of 32, only four non-democracies are included, which violates the requirements for logistic regression (both outcomes should be above n = 15).
The multiple regressions models using Freedom House Political Rights and Civil Liberties as dependent variables are slightly more interesting. In models 93 through 100 with dependent variable Political Rights, the same variables are tested as in models 85-92. Within the sample, Oil-richness isn’t significantly associated with Political Rights when self-expression values are added to the model. We do see, however, that the explained variance (R2) goes up with every variable added. Model 92 (F(10,22) = 4.989, p < 0.05) has the most variables and the highest explained variance (R2 = 0.704). Almost all the B-coefficients have the expected sign, indicating negative or positive association with the outcome variable. Three are significant: Economic Diversification (B = -0.075, p < 0.05), Political Action (B = -1.638, P < 0.1), Trust in Others (B = 4.686, p < 0.05) and Justifiability of Homosexuality (B = -0.415, P < 0.1). Trust in Others has the opposite effect from the hypothesis.
The multiple regression models used to predict Civil Liberties look similar. In models 101-105, when the self-expression values are added to Oil-richness, Oil-richness’ B-coefficient loses its significance. In almost all models, GNI per capita is a significant contributor, but its significance disappears once more variables are added to the model. All models are significant (F(4,28) > 3, P < 0.05), indicating that the variables together predict a significant part of the variance on the outcome variable. Model 105 (F(10,22) = 3.958, p < 0.05) has the most variables and the highest explained variance (R2 = 0.653) as was the case with the same variables for Political Rights. The largest difference with model 92 is that Economic Diversification and Political Action lose their significance. They are less important to Civil Liberties than to Political Rights.
As mentioned above, the logistic models with democracy (0-1) as outcome variable, lacked significance. This lack of significance can be explained by three problems, linked to the dataset. Firstly, when in multiple regressions none of the B-coefficients are significant; this can often be explained by an issue of multicollinearity. Multicollinearity occurs when the variables used are highly correlated with each other, meaning they are so similar and overlapping, neither of them significantly predicts outcomes. Another problem could be the size of the dataset, due to missing data and the use of different datasets. An n of 32 is hardly
