Chapter 3. Quantitative Analyses
3.3 Empirical results
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
3.2 Methods and Estimation Techniques
To test my hypothesis about the relationship between political corruption and EU accession, I employ discrete-time survival models (Box-Steffensmeier and Jones 2004). I use this method because there are time-varying covariates in the model and the data are structured as country-year. This assumes there are discrete intervals in time, even though a country can end up with being an EU member or candidate at any point during the year. Because the dependent variable is a dichotomous variable, the survival model in my analysis is basically a logistic regression that also takes time into consideration. Because each country is measured repeatedly (each year), the standard errors are likely not independently and identically distributed (Woolridge 2003). To address this “group effects” problem, I robustly cluster standard errors by country (Zorn 2006).
3.3 Empirical results
Table 2 displays the results of three discrete-time models using different samples. Model 1 considers all samples, Model 2 considers the sample between 1990 and 2007, and Model 3 considers the sample between 2008 and 2018. As can be seen, the coefficients of political corruption are not statistically significant for Model 1 and Model 2. However, political corruption has a statistically significant coefficient in Model 3, indicating that after 2007, a country is less likely to obtain EU membership or EU candidacy if it has a high level of political corruption. Figure 1 shows the predicted probabilities of EU accession based on different scenarios of political corruption. The results show that the probability of EU accession decreases with the level of political corruption. Overall, the findings in Model 3 and Figure 1 provide a strong support for the hypothesis.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Table 2. Logit Models of Political Corruption and EU Accession
Variable
Model 1 (All Samples)
Model 2 (Pre-2008
Sample)
Model 3 (Post-2007
Sample)
Political Corruption 1.477 0.511 -4.327**
(1.097) (1.892) (2.026)
Logged GDP Per Capita 0.942*** 1.439** -9.046**
(3.304) (0.631) (3.780)
Level of Democracy 0.220 0.277 4.082**
(0.236) (0.541) (1.666)
Muslim Population (%) 0.003 -0.049 0.025**
(0.013) (0.031) (0.011)
Total Population 0.021** 0.027 -0.739**
(0.010) (0.016) (0.313)
Constant -12.989*** -17.736*** 41.472**
(3.765) (6.732) (18.761)
Pseudo R squared 0.162 0.240 0.349
Number of Countries 19 14 7
Number of Observations 188 137 51
Notes:
Robust standard errors in parenthesis. *p<0.1; **p<0.05; ***p<0.001
The dependent variable: 1=obtained EU membership or obtained EU candidacy; 0=otherwise.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Figure 1. Political Corruption and Predicted Probabilities of EU Accession after 2007
Moreover, a T-test for comparing the equality of coefficients of political corruption between Model 2 and Model 3 shows that the difference in the effects of political corruption is
statistically significant at p<0.1 level (see Table 3). In other words, while the level of political corruption matters for the probability of obtaining an EU membership or candidacy after 2007, it does not matter much for the probability of obtaining an EU membership or candidacy before 2008. In short, this finding supports the hypothesis.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Figure 2. Political Corruption in Western Balkan Countries 2003-2018
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Table 3. Consistency of Results
Variables Pre-2008
Sample
Post-2007 Sample
T-Test Statistic Political Corruption Insignificant Significant 1.745
Logged GDP Per Capita Significant Significant 2.736
Level of Democracy Insignificant Significant -2.172
Muslim Population (%) Insignificant Significant -2.250
Total Population Insignificant Significant 2.444
Note: The T-test statistic must have an absolute value of at least 1.96 to be significant at p<0.05 and of at least 1.65 to be at p<0.10.
The analyses in Table 2 demonstrate interesting results of the control variables. Regarding the effects of the level of democracy on EU accession, the coefficients of this variable are statistically insignificant in Model 1 and Model 2. However, the level of democracy has a positive and statistically significant effect on EU accession for the post-2007 sample. The T-test in Table 3 shows that the effect in Model 3 is significantly different from the effect shown in Model 2. This finding indicates that holding other variables constant, a more democratic country is more likely to obtain EU membership or candidacy after 2007.
Model 1 shows that a country with a higher level of economic development is more likely to obtain EU membership or Candidacy, supporting the argument made by Katchanovski (2011). The result of economic development for Model 2 also shows similar pattern.
Surprisingly, the result of economic development for Model 3 is very different from the results in Model 1 and Model 2. For the post-2007 sample, a country with a higher level of economic development is less likely to obtain EU membership or candidacy.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Another surprising result is about the percentage of population that is Muslim. The result in Table 2 show that the coefficient of this variable does not reach statistical significance in Model 1 and Model 2. However, for the post-2007 sample, a country with more Muslims tends to have a higher probability of obtaining EU membership or candidacy. This result contradicts with the findings of Katchanovski (2011). Last, the results in Model 1 (all samples) demonstrate that larger countries are more likely to join the EU, while Model 3 shows that smaller countries are more likely to join the EU after 2007.
Robustness checks
To make sure that my results are not driven by the particular coding of my dependent
variable and independent variable, I conduct two tests to check the robustness of my findings.
In Model 4 and Model 5, I estimate logit regressions using regime corruption as an alternative independent variable. Data for this variable are obtained in Varieties of Democracy (V-Dem), the regime corruption variable is operationalized in a scale from 0 (low) to 1 (high), and it measures the extent of political actors using political office for private or political gain, more specifically to focuses on political actors that hold position in political offices and are involved in acts that relate to the concept of neopatrimonialism. The index is a combination of various indicators that concern the executive, the legislative and the judiciary; those indicators are executive embezzlement, executive bribes, legislative
corruption and judicial corruption. Moreover, this variable is also closely related to political corruption index that is also used in this study (Coppedge et al. 2019). The correlation coefficients of political corruption and regime corruption in my dataset are 0.96 for the pre-2008 sample and 0.89 for the post-2007 sample, respectively. The results show that while regime corruption has no statistically significant effect on the probability of EU accession for the pre-2008 sample, it has a negative and statistically significant effect on the probability of EU accession for the post-2007 sample. This finding again suggests that corruption matters for explaining EU accession after 2007.
‧
Logit Models Ordered Logit Models
Regime Corruption -0.108 -3.066**
- -
Constant -17.212*** 34.658**
- -
Robust standard errors in parenthesis. *p<0.1; **p<0.05; ***p<0.001
The dependent variable for Model 4 and Model 5: 1=obtained EU membership or obtained EU candidacy;
0=otherwise.
The dependent variable for Model 6 and Model 7: 2=obtained EU membership; 1=obtained EU candidacy;
0=otherwise.
In Model 4 and Model 5, I use a different dependent variable for operationalizing EU accession. Specifically, I construct an ordinal variable that it is coded 2 when a country obtained the EU membership, 1 when a country obtained the EU candidacy, and 0 otherwise.
I estimate ordered logit regressions on this new dependent variable. As can be seen in Model 6, the coefficient of political corruption is statistically insignificant. In contrast, the results in Model 7 suggest that political corruption has a negative and statistically significant effect on
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
the probability of EU accession. This finding suggests that the likelihood for a country to join the EU after 2007 decreases with its level of political corruption. Overall, the analyses in Table 4 show that my hypothesis is supported, and results are robust across different model specifications.
‧
國立 政 治 大 學