Analysis of World Governance Indicators

From Teachwiki
Jump to: navigation, search

Top.jpg

Introduction[edit]

Governance matters. This analysis will give a short introduction to the Worldwide Governance Indicators of the World Bank. It captures all 6 indicators which are listed below. It will be checked how these indicators are connected to a political classification of governments published by The Economist. This classification is named "Democratization Index". This will help to structure the WGI (Worldwide Governance Indicators) to provide a deeper view.

Data[edit]

The Worldwide Governance Indicators (WGI) are composite indicators of six dimensions of governance, covering 212 countries and territories between 1996 and 2006 (1997, 1999, 2001 missing). They are based on hundreds of individual indicators of governance drawn from 33 data sources produced by 31 different organizations (visit http://www.govindicators.org for detailed informations).

  • Voice and Accountability: Voice and Accountability measuring the extent to which a country’s citizens are able to participate in selecting their government, as well as freedom of expression, freedom of association, and a free media (Kaufmann et al., 2007).
  • Political Stability and the Absence of Violence: It measures perceptions of the likelihoood that the government will be destabilized or overthrown by unconstitutional or violent means, including domestic violence and terrorism.
  • Governmental Effectiveness: GE measures the quality of public services, the quality of the civil service and the degree of its independence from political pressures, the quality of policy formulation and implementation, and the credibility of the government’s commitment to such policies.
  • Regulatory Quality: RQ measuring the ability of the government to formulate and implement sound policies and regulations that permit and promote private sector development.
  • Rule of Law: Rule of Law measures the extent to which agents have confidence in and abide by the rules of society, and in particular the quality of contract enforcement, the police, and the courts, as well as the likelihood of crime and violence.
  • Control of Corruption: CoC measures the extent to which public power is exercised for private gain, including both petty and grand forms of corruption, as well as ”capture” of the state by elites and private interests.

The Democracy Index (DI) has been published by the Economist Intelligence Unit (http://www.eiu.com) 2006 and in the beginning of 2007. It summarizes a survey which has been made to estimate the state of democracy within political systems. The index consists of five categories (59 questions); electoral process and pluralism, functioning of government, political participation, political culture, civil liberties. Political systems are characterized to be a full democracy, a flawed democracy, a hybrid regime or an authoritarian regime. For a detailed description: http://www.economist.com/media/pdf/DEMOCRACY_INDEX_2007_v3.pdf.

Statistical Analysis[edit]

Preliminaries[edit]


In order to provide an overview of selected variables, lots of variables will be excluded from this analysis. They are left in the data set due to their expected importance as regards content in further analysis. So, only the following key-variables will be selected: Voice and Accountability Indicator 2006, Political Stability Indicator 2006, Governmental Effectiveness Indicator 2006, Regulatory Quality Indicator 2006, Rule of Law Indicator 2006, Control of Corruption Indicator 2006. First, the data set is ordered with respect to the Human Development Index 2005 to set a reference value. With this step, a first step in interpreting the results can be made (If the characteristic variable for ordering changes it will be noted). First, some descriptive statistics will be provided:

The units in which governance is measured follow a normal distribution with a zero mean, a minimum of -2.5 and a maximum of -2.5 (Kaufmann et al. 2007). These boundaries correspond to the 0.005 and 0.995 percentiles of the standard normal distribution. For a handful of cases, individual country ratings can exceed these boundaries when scores from individual data sources are particularly high or low. This implies, no information of the change in global governance can be extracted from the indicators. Instead, only the performance of single countries or groups in comparison to others can be extracted.

Descriptive Statistics[edit]

Below you can find some introductory descriptive statistics.

Descriptive Statistics I
Descriptive Statistics II

The table below shows the correlations between the indicators.

Descriptive Statistics II

Detection and Identification of Outliers[edit]


Stem-and-Leaf Diagram[edit]

The stem and leaf will not be used extensively in the analysis. To explain shortly how it works and what information can be extracted, the Control of Corruption Indicator 2006 (CC2006) will be used. The next Figure gives an impression how the variable is distributed. As noted earlier, all units used for the indicators are normal distributed with [−2.5; +2.5] being the 0.005 and 0.995 percentiles of a standard normal distribution. One can see that the aggregated indicator is not perfectly normal distributed due to the number of observations. It is skewed to the right. Each leaf stands for 2 observations. This results in fractional leafs (&) which can not be put either in one or the other group. In fact, there are two values being lower than −1.4 and one value greater than 2.4.

Stem-and-Leaf Plot: Control of Corruption 2006

One can see that although the underlying data is normalized, the distribution is slightly skewed to the right. Taking a look on the table which show the 5 largest and lowest values, one gets the following results:

Extreme Values: Control of Corruption 2006

Boxplot[edit]

The boxplot is another tool which helps to get an impression about the data. As the authors of the Word Government Indicators normalize all the data, single boxplot would not give any insights. Thus, the plots for all variables are shown in the next figure. All variables except Voice and Accountability 2006 (VA2006) and Political Stability 2006 (PS2006) are skewed to the right. That means, the larger fraction of countries perform less than the mean of the specific indicator. The following analysis is far more interesting. Every single indicator has been grouped by the groups of the Democracy Index 2007 (DIg07). The circles represent observations which lie between FUL ± 1.5df and FUL ± 3df where df = FU − FL, the the range where 50% of the data are in. All observations marked with a star lie outside the FUL ± 3df borders.

Boxplot: World Governance Indicators

By looking on the individual boxplots possible outliers become visible.

Voice and Accountability[edit]

In the next figure no outliers can be identified. The boxes are small in comparison with the following grouped boxplot. Both seems due to the compuation of the DI2007. The VA2006 and the DI2007 seem to have similar underlying characteristics. The medians of all groups seem to differ considerably. Whether they are significantly different needs to be checked.

Boxplot: Voice and Accountability
Political Stability[edit]

The general pattern of the locations of the boxplot is similar to the other indicators. The difference is, that the data within the groups is more spreaded. Again, the boxblot for the functioning democracies is smaller, thus the variance of the data within that group is smaller in comparison to the other. The medians of the authoritarian and the hybrid regime are very close. The only outlier is Nicaragua in the NAN group.

Boxplot: Political Stability
Governmental Effectiveness[edit]

Singapore is the only country which stands well outside as an extreme value is the hybrid regime group. The medians of all groups seem to differ considerably. Whether they are significantly different needs to be checked.

Boxplot: Governmental Effectivenss
Regulatory Quality[edit]

Again Singapore is an exception in his group. The regulatory quality lies outside the FU + 1.5df border of the boxplot. In the same group Venezuela represents a negative exception. It’s regulatory quality is characterized as an outlier in this group. The medians of all groups seem to differ considerably. Whether they are significantly different needs to be checked.

Boxplot: Regulatory Quality
Rule of Law[edit]

As expected, Singapore is again an extreme value. In addition, Quatar represents a positive outlier as a Authoritarian Regime.

Boxplot: Rule of Law
Control of Corruption[edit]

The next Figure shows the CC2006 grouped by the DIg07. This is far more interesting than the previos boxplot. Kuwait, Quatar, Oman, Bhutan and the United Arab Emirates are authoritarian regimes with outlying good performance in control of corruption. Looking in the next group, it is not very remarkable that Hong Kong is an outlier in ”Flawed Democracy”. The functioning democracies seem to represent a homogeneous group. In the last group, Singapore is an extreme value, thus the small harbour performs incredible good in the control of corruption. The medians of all groups seem to differ considerably. Whether they are significantly different needs to be checked.

Boxplot: Control of Corruption



Tests for Outliers[edit]

The general preliminary assumption for tests for outliers is the normal distribution of the variables. In case of the WGI2006 this assumption is satisfied, as already mentioned, by the computation procedure of the authors of the indicators.

Summary[edit]

The following table summarizes the results. ”x” the rejection of being an not an outlier, ”-” denotes a decision in favour of being not an outlier and ”.” characterizes cases where the test was not performed. The table shows us that the boxplot is no good tool for outlier-detection. For the case of Singapore being an outlier, it was quite obvious from the boxplot. For the other cases, the boxplot show that they are outside the FUL ± 1.5df range. The tests show that this is no reliable criteria of being an outlier.

Outliersummary.png
Grubbs Test[edit]

The Grubbs test checks, whether the biggest or the smallest value is an outlier or not. H_0: x(1) is not an outlier vs. H_1: x(1) is an outlier

or

H_0: x(n) is not an outlier vs. H_1 : x(n) is an outlier

where x(1) is the smallest and x(1) the largest observation. The corresponding test-statistics can be calculated as follows:

T_1=\frac{\bar{x}-x(1)}{s}

or

T_n=\frac{\bar{x}-x(n)}{s}

The boxplot for the ungrouped indicators do not show any outliers. If one performs the test on these indicators, the result is very far from being different. The highest test statistic is 2.59 whereas the critical value for a significance of 95% is 3.21 and for 99% 3.6. The table below shows the results. Far more interesting will the results of the test for the DI2007 grouped indicators be. The WGI2006 is as above grouped by the DI2007. The grouping do not change the necessary assumption of normality. Due to that assumption only outliers detected by the boxplot will be checked.

Nicaragua in the grouped boxplot for the Political Stability Indicator is the first outlier but as it is part of the NAN group it is of no use to check whether it is an outlier in this analysis.

The first important case is Singapore as a hybrid regime in the governmental effectiveness indicator. The country is marked as an extreme value in the boxplot for Governmental Effectiveness. There are 27 observations in the group of hybrid regimes. The critical values are T(30,0.95) = 2.75 or T(30,0.99) = 3.10. The test statistic for Singapore is T(S) = 3.8759. Thus, the small harbor must be considered as an outlier as an hybrid regime in governmental effectivenss.

Also in the indicator of regulatory quality Singapore role as a hybrid regime needs to be checked. The critical values are the same as above. The test statistic for Singapore is T(S) = 3.665. According to that, Singapore is an outlier also in this category. Venezulea is the other outlier marked in the boxplot for Regulatory Quality (grouped). The test statistic for Venezuela is T(V) = 1.708. So, the country is marked in the boxplot as an outlier, in constrast, the Grubbs test rejects the result, the null hypotheses of Venezuela being an outlier must be rejected on a 95% and 99% confidence level.

In figure which shows the boxplot for the Rule of Law Indicator two outliers are identified, Quatar as an authoritarian regime and again Singapore as hybrid regime. The Grubbs test identifies Singapore also in the indicator of the rule of law as an outlier with T(S) = 3.7787. The critical values are the same as above. The group of authoritarian regimes consists of n = 53 observations. Thus, the critical values differ from the previous group of hybrid regimes with n = 27. The test statistic of quatar is T(Q) = 2.3196 leads to the decision that H_0 can not be rejected, thus the country is not an outlier, taking T(50,0.95) = 2.96 and T(50,0.99) = 3.34 as critical values.

The last grouped indicator in which outliers needs to be checked is the Control of Corruption. The boxplot identifies Singapore and Haiti as outliers in the group of hybrid regimes. The Grubbs test do not rejects Haiti to be an no outlier (F(H) = 1.4162) but do reject Singapore to be no outlier with F(S) = 4.25. Hong Kong is identified as an outlier as flawed democracy (n = 51; T(50,0.95) = 2.96; T(50,0.99) = 3.34). The Grubbs do not reject H_0 with F(H) = 2.8819. By choosing a slightly lower significance level, say 90% might result in the rejection of H = 0. The boxplot shows in the group of the authoritarian regimes a bunch of outliers. The largest outlier are the United Arab Emirates. The critical values are the same as above. The test statistic is F(UAE) = 2.643. This value do not reject the null hypothesis of the country being no outlier. The result is probably again sensible to the significance level. Taking that result the test needn’t to be performed for Bhutan, Oman, Quatar and Kuwait. As the United Arab Emirates is not an outlier, they are neither one.

Dixons r-Statistics[edit]

This test is as well as the Grubbs test only valid under the assumption of normality of the data. This is not violated as stated above. The test hypotheses are the same as in the Grubbs test. The test statistics will be computed as follows:

r_{kg}(1)=\frac{x(1+k)-x(1)}{x(n-g)-x(1)} for k=1,2; g=0,1,2

r(kg)(1) tests whether x(1) is not an outlier. g denotes the number of potential outliers with large values, k the number of potential outliers with small values. The test statistic which test whether x(n) is not an outlier is

r_{gk}(n)=\frac{x(n)-x(n-g)}{x(n)-x(1+k)} for k=1,2; g=0,1,2.

First, test Singapore for being an outlier (large value) in governmental effectiveness as an hybrid regime. The critical value is r(1,0;0.99;25) = 0.36. The test statistic is r(1,0) = 0.535. This value do reject H_0 that Singapore is not an outlier. Second, test Singapore for being an large and Venezuela being an small outlier as hybrid regimes in the category of regulatory quality. The critical value is now r(1,1;0.99;25) = 0.39. Singapore has been rejected of being no outlier with a test statistic of r(1,1) = 0.4884. The null hypothesis for Venezuela of not be an outlier was not rejected with a test statistic of r(1,1) = 0.0988. Third, the test for Singapure being an outlier in the rule of law category as an hybrid regime gives the test statistic of r(1,0) = 0.5149 which rejects the H_0. Quatar is also tested for being a positive outlier in the group of authoritarian regimes in this category. The H0 is not rejected on an 99% significance level (r(1,0;0.99,30) = 0.34; r(1,0) = 0.068). Fourth, Hong Kong is suspected to be an positive outlier in the category of control of corruption as an flawed democracy. The critical value is r(1,0;0.99;30) = 0.34 and the test statistic is r(1,0) = 0.133. Therefore the null hypothesis can not be rejected, Hong Kong is not an outlier. Fifth, the United Arab Emirates are suspected to be a positive outlier as an authoritarian regime in the same category. Compare the test statistic r(2,0) = 0.116 with the critical value r(2,0;0.99,30) = 0.40 shows that the H_0 can not be rejected. Therefor, the result do hold for all the remaining suspects.

David-Hartley-Pearson-Test[edit]

As the previous test, the David-Hartley-Pearson-Test assumes normality of the considered data. The test statistic can be computed as follows:

T=\frac{R}{s}=\frac{x(n)-x(1)}{\sqrt{\frac{\sum{(x(i)-\bar{x})^2}}{n-1}}}.

SPSS provides R and s. The null hypothesis of the biggest or the smallest value is part of the sample will be rejected if T > Q(n;1−a). First, Singapore is suspected to be an outlier in the category of governmental effectiveness as a hybrid regime. The test supports that as T = 5.3018 is larger than the 95% and the 99% critical value Q(30;0.95) = 4.89 and Q(30;0.99) = 5.25. Second, Singapore is also suspected to be together with Venezuela outliers as hybrid regimes in regulatory quality. As T = 5.423, the value with the bigger distance to the mean must be deleted, thus Singapore is not a part of the sample. Singapore is also suspected to be an outlier as an hybrid regime in the indicator of rule of law. Comparing T = 5.434 with the critical values above, leads to the result that the H_0 will be rejected. In the same indicator but in another group, Quatar (authoritarian regime) is identified as an outlier in the boxplot. The DHP test do not reject the H_0 in this case (T = 3.75). If one do the test for the last indicator, the same results appears as in the test before.

Robust Estimators for Localization and Variation[edit]

Lestimators.png

The weighting constant for the Huber’s M-Estimator is 1.339, for Tukey’s Biweight 4.685, for Hampel’s M-Estimator 1.7, 3.4 and 8.5 and for the Andrews’ Wave 1.34*pi�.

Mestimators.png

Tests for Distribution[edit]


The authors of the WGI2006 transformed the units of indicators as such, that they are X\sim N(0;1) distributed. Based on that assumption the tests previously done can be performed. Now, by testing for the distribution of the variables, the results of the analysis below will differ from the expected as the largest number of observations included is 177. One can expect that the test results become better as N increases (difference between grouped and not grouped indicators). This clearly shows the law of large numbers in practical applications. Again the indicators will be grouped by the DI2007. As the Jarque-Bera test for normality is not includes in SPSS, a short review of the method will be given to let the reader examine the calculations. The test is based on the skewness and kurtosis of a distribution. Note that SPSS always prints the deviation from the normal distribution of the kurtosis.

H_0: The sample is normal vs. H_1: The sample is not normal

The test statistic is computed in the following way: JB=\frac{n}{6}\left(S^2+\frac{(K-3)^2}{4}\right)\sim\chi^2_2.

Voice and Accountability[edit]

Variable Grouping N Skewness Kurtosis K.-S. Test (Asymp. Sig. 2-tailed) SK Test J.-B. Test (p-value)
VA2006 - 177 0.013 -0.948 0.922 (0.363)* 19.28 (0.000) 7.069 (0.03)
VA2006 AR 53 0.004 -0.469 0.521 (0.949)* 0.54 (0.762)* 0.873 (0.65)*
VA2006 FlD 51 0.040 -0.829 0.477 (0.977)* 3.27 (0.195)* 1.921 (0.38)*
VA2006 FuD 28 0.449 -1.206 0.967 (0.307)* 5.74 (0.056)* -
VA2006 HR 27 -0.164 -0.739 0.522 (0.948)* 1.13 (0.568)* -

{*} Significant on 5%

On average, the tests show normality of the VA2006. Only the Skewness and Kurtosis and the Jarque-Bera test reject the null hypothesis of normality of the not grouped VA2006. The power of the tests for the grouped indicator are very low as the number of observations are small. The histogramms and Q-Q/P-P plots of the variables can be seen below.

Histogramm: Voice and Accountability 2006

Political Stability[edit]

Variable Grouping N Skewness Kurtosis K.-S. Test (Asymp. Sig. 2-tailed) SK Test J.-B. Test (p-value)
PS2006 - 177 0.013 -0.948 1.084 (0.191)* 9.76 (0.008) 7.694 (0.02)
PS2006 AR 53 -0.239 -0.438 0.670 (0.760)* 1.04 (0.593)* 1.251 (0.54)*
PS2006 FlD 51 -0.396 -0.577 0.632 (0.820)* 2.58 (0.275)* 2.311 (0.31)*
PS2006 FuD 28 -0.020 -0.544 0.517 (0.952)* 0.36 (0.836)* -
PS2006 HR 27 0.306 -0.127 0.362 (0.999)* 0.54 (0.763)* -

{*} Significant on 5%

The decision is not obvious for PS2006. The Kolmogorov-Smirnov test do not reject the H_0 of normality but the SK test and the Jarque-Bera test do. The power of the tests for the grouped indicator is very low as the number of observations is small. The histogramms and Q-Q/P-P plots of the variables can be seen below.

Histogramm: Political Stability 2006

Governmental Effectiveness[edit]

Variable Grouping N Skewness Kurtosis K.-S. Test (Asymp. Sig. 2-tailed) SK Test J.-B. Test (p-value)
GE2006 - 177 0.531 -0.568 1.140 (0.148)* 10.55 (0.005) 10.800 (0.00)
GE2006 AR 53 0.724 -0.356 0.924 (0.361)* 4.84 (0.089)* 4.753 (0.09)
GE2006 FlD 51 0.461 -0.599 0.781 (0.575)* 3.02 (0.221)* 2.686 (0.26)*
GE2006 FuD 28 -0.584 -0.693 0.675 (0.752)* 2.87 (0.238)* -
GE2006 HR 27 2.208 8.187 0.895 (0.400)* 20.60 (0.000) -

{*} Significant on 5%

Again, the Kolmogorov-Smirnov test (sample normal distributed) provide another result than the SK test and the Jarque-Bera test (sample not normal distributed) with regard to GE2006. Most of the grouped indicator test results support the H_0 of normality except the SK test for the hybrid regimes. But, the power of the tests for the grouped indicator are very low as the number of observations are small. The histogramms and Q-Q/P-P plots of the variables can be seen below.

Histogramm: Governmental Effectiveness 2006

Regulatory Quality[edit]

Variable Grouping N Skewness Kurtosis K.-S. Test (Asymp. Sig. 2-tailed) SK Test J.-B. Test (p-value)
RQ2006 - 177 0.097 -0.689 1.021 (0.248)* 7.33 (0.026) 4.188 (0.12)*
RQ2006 AR 53 0.230 -0.293 0.553 (0.920)* 0.66 (0.718)* 0.916 (0.63)*
RQ2006 FlD 51 0.190 -0.270 0.550 (0.923)* 0.44 (0.803)* 0.727 (0.70)*
RQ2006 FuD 28 -0.832 -0.184 0.678 (0.748)* 3.80 (0.149)* -
RQ2006 HR 27 1.714 6.258 0.992 (0.279)* 16.18 (0.000) -

{*} Significant on 5%

According to the Kolmogorov-Smirnov test, the indicator and all subgroups (DI2007) are normal. The SK test and the Jarque-Bera test deliver different result on RQ2006. The SK test rejects H_0 in contrast to the JB test which do not reject H_0 of normality on a significance level of 5%. The latter mentioned test find mostly not enough evidence to reject the null hypothesis of normality in the grouped indicator except the SK test on the hybrid regimes. But, the power of the tests for the grouped indicator are very low as the number of observations are small. The histogramms and Q-Q/P-P plots of the variables can be seen below.

Histogramm: Regulatory Quality 2006

Rule of Law[edit]

Variable Grouping N Skewness Kurtosis K.-S. Test (Asymp. Sig. 2-tailed) SK Test J.-B. Test (p-value)
RL2006 - 177 0.821 -0.163 1.784 (0.003) 12.96 (0.006) 11.931 (0.00)
RL2006 AR 53 0.880 -0.142 1.001 (0.269)* 6.07 (0.048) 6.352 (0.04)
RL2006 FlD 51 0.538 -0.642 1.030 (0.239)* 4.14 (0.126)* 3.493 (0.17)*
RL2006 FuD 28 -0.766 -0.805 1.029 (0.240)* 4.41 (0.111)* -
RL2006 HR 27 1.919 7.284 0.829 (0.498)* 18.25 (0.000) -

{*} Significant on 5%

All tests reject the H_0 of normality for the ungrouped RL2006. The Kolmogorov-Smirnov test do not reject the null hypothesis for the grouped indicator. The SK test do not reject the H_0 for flawed and functioning democracies but it do reject for authoritarian and hybrid regimes. The Jarque-Bera test do only not reject for the flawed democracies. For the authoritarian regimes the JB test results in the same decision as the SK test. But, the power of the tests for the grouped indicator are very low as the number of observations are small. The histogramms and Q-Q/P-P plots of the variables can be seen below.

Histogramm: Rule of Law 2006

Control of Corruption[edit]

Variable Grouping N Skewness Kurtosis K.-S. Test (Asymp. Sig. 2-tailed) SK Test J.-B. Test (p-value)
CC2006 - 177 0.183 -0.163 1.785 (0.003) 14.14 (0.000) 19.576 (0.00)
CC2006 AR 53 0.995 0.334 1.083 (0.192)* 7.52 (0.023) 7.806 (0.02)
CC2006 FlD 51 0.671 0.310 0.790 (0.561)* 4.47 (0.107)* 3.397 (0.18)*
CC2006 FuD 28 -0.437 -0.898 0.718 (0.681)* 3.15 (0.207)* -
CC2006 HR 27 2.997 12.853 1.181 (0.123)* 27.82 (0.000) -

{*} Significant on 5%

The performed tests show that CC2006 is not normal distributed. Only some of the subgroups show normality. But, the power of the tests for the grouped indicator are very low as the number of observations are small. The histogramms and Q-Q/P-P plots of the variables can be seen below.

Histogramm: Control of Corruption 2006

Comparison of Parameters of Independent Samples[edit]


The analysis is only meaningful if it is ensured that the samples to compare are independent. Therefor it is not allowed to compare the score of single groups of observations between the indicators of the WGI2006. There are huge correlation between them. A country with a good CC2006 ranking, say Iceland, will with no doubt have a good RL2006 score and so on. One can not assume that independent characteristics are responsible for both results. For example, it is impossible to argue for independence of the rule of law and the political stability score as both criteria strongly depend on each other. Furthermore, an analysis of means of the indicators over time will not bring any useful result as the scores are expected to be at least serially correlated over time. This is also true for different groups over time. The grouping variable will again be the verbal description of the DI2007. One possible and very interesting task will be to compare the different means of the subgroups of a single governance indicator. The assumption of independence of the underlying data seems sustainable. It could be interesting whether there exists a significant difference between authoritarian and hybrid regimes with respect to their control of corruption performance, for instance. Another analysis will the comparison of variances of subgroups to check for homogeneity. In the first step, some important parameters, for each grouped indicator respectively, will be collected and an error bar chart for the means of each subgroup (with 95% confidence intervals) helps to get a graphical illustration of the ongoing hypotheses which needs to be tested. To test for the means of two independent samples, a t-test with the following test-statistic will be used: T=\frac{\bar{x}_1-\bar{x}_2}{\sqrt{\frac{n_1+n_2}{n_1\cdot n_2}\frac{(n_1-1)s_1^2+(n_2-1)s_2^2}{n_1+n_2-2}}}\sim t_{(n_1+n_2-2)}

If the number of observations is each sample are sufficiently large (>30) one can approximate the t by the N(0;1)-distribution. The H_0 tests for the equality of means against H_1.

Voice and Accountability[edit]

Error Bar: Voice and Accountability 2006
Variable Grouping N Mean Std. Error for Mean 95% Conf. Interval for Mean Variance
PS2006 AR 53 -1.158 0.066 [-1.291;-1.024] 0.234
PS2006 FlD 51 0.280 0.071 [ 0.138; 0.421] 0.254
PS2006 FuD 28 1.343 0.054 [ 1.233; 1.454] 0.081
PS2006 HR 27 -0.421 0.077 [-0.580;-0.262] 0.162

There are significant differences between the different political regimes groups within the Voice and Accountability indicator. The mean of the functioning democracies is again the largest one. No test is needed to proof this. Authoritarian and hybrid regimes First, the equality of variances is tested. Than, the t-test for the equality of the means can be done:

t-test for means of subgroups of VA2006 grouped by DI2007
Null Hypothesis Alternative Hypothesis p-value Test decision**
AR=HR AR\neqHR 0.000 H_0 rejected
FlD=HR FlD\neqHR 0.000 H_0 rejected
AR=FlD AR\neqFlD 0.000 H_0 rejected

{* equality of variances rejected by Levene's test on 5% significance level}

{** on 5% significance level}

In this case there is no equality of means at all can be detected. This is not always the case for Authoritarian and 
Hybrid Regimes. Every subgroup perform significantly different from the other. The ranking is the same as for the other  
indicators. Functioning Democracies on the first, Flawed Democracies on the second, Hybrid Regimes on the third and
Authoritarian Regimes on the last rank.

Political Stability[edit]

Error Bar: Political Stability 2006
Variable Grouping N Mean Std. Error for Mean 95% Conf. Interval for Mean Variance
PS2006 AR 53 -0.632 0.118 [-0.868;-0.395] 0.737
PS2006 FlD 51 -0.126 0.112 [-0.350; 0.099] 0.637
PS2006 FuD 28 0.940 0.066 [ 0.805; 1.075] 0.121
PS2006 HR 27 -0.543 0.144 [-0.840;-0.246] 0.563

There are significant differences between the different political regimes groups within the political stability indicator. The mean of the functioning democracies is again the largest one. No test is needed to proof this. Authoritarian and hybrid regimes First, the equality of variances is tested. Than, the t-test for the equality of the means can be done:

t-test for means of subgroups of PS2006 grouped by DI2007
Null Hypothesis Alternative Hypothesis p-value Test decision**
AR=HR AR\neqHR 0.621 H_0 can not be rejected
FlD=HR FlD\neqHR 0.015 H_0 rejected
AR=FlD AR\neqFlD 0.000 H_0 rejected

{* equality of variances rejected by Levene's test on 5% significance level}

{** on 5% significance level}

The result shows that the authoritarian and the hybrid regime are far away from guarantee political stability. Both 
are equally bad. Flawed democracies do better than hybrid and authoritarian regimes. That shows, that regimes can not
ensure their often quoted initial justification of stability. Societies with political and social participation
possibilities can guarantee better political stability, although the systems are guided by far more actors than one
can expect in regimes.

Governmental Effectiveness[edit]

Error Bar: Governmental Effectiveness 2006
Variable Grouping N Mean Std. Error for Mean 95% Conf. Interval for Mean Variance
RL2006 AR 53 -0.753 0.089 [-0.932;-0.575] 0.419
RL2006 FlD 51 0.072 0.095 [-0.118; 0.262] 0.458
RL2006 FuD 28 1.496 0.106 [ 1.277; 1.714] 0.316
RL2006 HR 27 -0.417 0.130 [-0.684;-0.150] 0.456

One can see on the right hand side that, again, Functioning Democracies have the most effective governments. Authoritarian Regimes perform very bad, but the significance of the difference to the Hybrid Regimes needs to be tested subsequently. Next, the significance of the differences will be tested. First, the equality of variances is tested. Than, the t-test for the equality of the means can be done:

t-test for means of subgroups of RL2006 grouped by DI2007
Null Hypothesis Alternative Hypothesis p-value Test decision**
AR=HR AR\neqHR 0.034 H_0 rejected
FlD=HR FlD\neqHR 0.003 H_0 rejected
AR=FlD AR\neqFlD 0.000 H_0 rejected

{* equality of variances rejected by Levene's test on 5% significance level}

{** on 5% significance level}

The equality of means of the indicators for Governmental Effectiveness is broadly rejected. The rejection of equal
performance of Authoritarian and Hybrid Regimes do not hold on an 1% significance level.

Regulatory Quality[edit]

Error Bar: Regulatory Quality 2006
Variable Grouping N Mean Std. Error for Mean 95% Conf. Interval for Mean Variance
RL2006 AR 53 -0.779 0.103 [-0.986;-0.572] 0.564
RL2006 FlD 51 0.139 0.102 [-0.065; 0.344] 0.531
RL2006 FuD 28 1.298 0.080 [ 1.133; 1.463] 0.180
RL2006 HR 27 -0.333 0.115 [-0.568;-0.097] 0.355

One can see on the right hand side that, again, Functioning Democracies have the most reliable law and order system. Authoritarian Regimes perform very bad, although they claim the opposite for themselves. Flawed Democracies are only a little better than hybrid regimes. Next, the significance of the differences will be tested. First, the equality of variances is tested. Than, the t-test for the equality of the means can be done:

t-test for means of subgroups of RQ2006 grouped by DI2007
Null Hypothesis Alternative Hypothesis p-value Test decision**
AR=HR AR\neqHR 0.290* H_0 rejected*
FlD=HR FlD\neqHR 0.000* H_0 rejected*
AR=FlD AR\neqFlD 0.000 H_0 rejected

{* equality of variances rejected by Levene's test on 5% significance level}

{** on 5% significance level}

The authoritarian and the hybrid regime perform equally bad in legal system. Flawed democracies do better
than hybrid and authoritarian regimes. That shows again, even weak democracies are more successful 
than any form of dictatorship.

Rule of Law[edit]

Error Bar: Rule of Law 2006
Variable Grouping N Mean Std. Error for Mean 95% Conf. Interval for Mean Variance
RL2006 AR 53 -0.704 0.096 [-0.897;-0.509] 0.496
RL2006 FlD 51 -0.138 0.090 [-0.318; 0.042] 0.411
RL2006 FuD 28 1.472 0.097 [ 1.272; 1.671] 0.264
RL2006 HR 27 -0.533 0.120 [-0.779;-0.286] 0.388

One can see on the right hand side that, again, Functioning Democracies have the most reliable law and order system. Authoritarian Regimes perform very bad, although they claim the opposite for themselves. Flawed Democracies are only a little better than hybrid regimes. Next, the significance of the differences will be tested. First, the equality of variances is tested. Than, the t-test for the equality of the means can be done:

t-test for means of subgroups of RL2006 grouped by DI2007
Null Hypothesis Alternative Hypothesis p-value Test decision**
AR=HR AR\neqHR 0.290 H_0 can not be rejected
FlD=HR FlD\neqHR 0.000 H_0 rejected
AR=FlD AR\neqFlD 0.000 H_0 rejected

{* equality of variances rejected by Levene's test on 5% significance level}

{** on 5% significance level}

The authoritarian and the hybrid regime perform equally bad in legal system. Flawed democracies do better
than hybrid and authoritarian regimes. That shows again, even weak democracies are more successful 
than any form of dictatorship.

Control of Corruption[edit]

Error Bar: Control of Corruption 2006
Variable Grouping N Mean Std. Error for Mean 95% Conf. Interval for Mean Variance
CC2006 AR 53 -0.607 0.092 [-0.791;-0.423] 0.447
CC2006 FlD 51 -0.143 0.090 [-0.324; 0.038] 0.413
CC2006 FuD 28 1.560 0.130 [ 1.293; 1.826] 0.472
CC2006 HR 27 -0.529 0.128 [-0.792;-0.267] 0.442

The error bar on the right hand side indicates significant differences between the different regimes. The authoritarian and the hybrid regime seem not to differ very much with respect to the control of corruption. Countries with flawed democracies do a little better. Whether they perform significantly better needs to be tested. Functioning democracies have the best control mechanisms against corruption. First, the equality of variances is tested. Than, the t-test for the equality of the means can be done:

t-test for means of subgroups of CC2006 grouped by DI2007
Null Hypothesis Alternative Hypothesis p-value Test decision**
AR=HR AR\neqHR 0.621 H_0 can not be rejected
FlD=HR FlD\neqHR 0.015 H_0 rejected
AR=FlD AR\neqFlD 0.000 H_0 rejected

{* equality of variances rejected by Levene's test on 5% significance level}

{** on 5% significance level}

The authoritarian and the hybrid regime perform equally bad in fighting corruption. Flawed democracies do better
than hybrid and authoritarian regimes. That shows, even weak democracies can fight corruption more successful 
than any form of dictatorship.

Conclusion and Perspective[edit]

So far the results show that the WGI and the Democracy Index are connected. The grouping by the DI how for every indicator similar results. That means, Functioning Democracies perform best, Flawed Democracies second best, Hybrid and Authoritarian Regimes worst. It was shown that this ranking holds with only a few exceptions. Consequently, the data basis of both indices must be checked. This hasn't be done here as the data collection is rather extensive and needs for a own work. One can watch the websites and the underlying papers for details. To interpret the results with respect to the data collection method could be interesting.

References[edit]

  • Kaufmann, D., A. Kraay, and M. Mastruzzi (2006): “World Governance Indicators 1996 - 2006,” World Bank Institute, World Bank Development Economics Research Group, http://www.govindicators.org, downloaded 5.11.2007.
  • ———(2007): “Governance Matters VI: Governance Indicators for 1996-2006,” Working Paper No. 4280, World Bank Policy Research, http://ssrn.com/abstract=999979, downloaded 18.12.2007.
  • Lambsdorff, J. G. (2001): “How Corruption in Government Affects Public Welfare - A Review of Theories,” Discussion Paper 9, Center for Globalization and Europeanization of the Economy, http://www.icgg.org/corruption.research contributions.html, downloaded 1.11.2007.
  • Oxley, H., J.-M. Burniaux, T.-T. Dang, and M. d’Ercole (1997/II): “Income Distribution and Poverty in 13 OECD Countries,” OECD Economic Studies No. 29, OECD.
  • Wegner, S. (2007): “An Explanation for the Increase of Social Transfers and the Gini-Coefficient in the Age of Globalization: An Investigation for Selected Industrialized Countries,” Seminar paper, Humboldt University Berlin.

Appendix[edit]

List of Abbreviations[edit]

  • AR - authoritarian regime
  • CC - Control of Corruption Indicator (opional: specific year added)
  • DI - Democracy Index (opional: specific year added)
  • FlD - flawed democracy
  • FuD - functioning Democracy
  • GE - Governmental Effectiveness Indicator (opional: specific year added)
  • HR - hybrid regime
  • PS - Politcal Stability Indicator (opional: specific year added)
  • RL - Rule of Law Indicator (opional: specific year added)
  • RQ - Regulatory Quality (opional: specific year added)
  • VA - Voice and Accountability Indicator (opional: specific year added)
  • WGI - World Governance Indicator (opional: specific year added)