Abstract
South Korean politics has been haunted by numerous corruption scandals as well as the prosecution of politicians and their cronies for their corruption. Yet despite the prevalence and salience of political corruption, many citizens of South Korea tend to overlook the problem by continuing to support corrupt politicians and administrations. This study defines under what circumstances political corruption shapes citizens' judgment of government and the political system as a whole in South Korea. The results indicate that national economic conditions as a perceptual screen mediate the effect of political corruption on the evaluation of democratic governance.
Acknowledgements
This work was supported, in part, by the University of South Florida System Internal Awards Program.
Notes
Considering that the voting age population in Seoul is around 8.3 million, the sample size of 396 is small, which leaves us a larger margin of error, ±4.9%, than the desired level of ±3%. However, according to Krejcie and Morgan (Citation1970), a 5% margin of error is acceptable for categorical data (see also Bartlett et al. Citation2001). Since most independent and dependent variables included in the empirical analysis here are categorical, 4.9% of margin of error seems to be large, but acceptable.
There is some concern that including all constitutive terms in an interaction model increases multicollinearity, thereby increasing the size of the standard errors, indicating that the coefficient of the interaction term tends to be insignificant. In order to test any multicollinearity problem, we checked pairwise correlations among all the regressors in the models and performed the VIF command in STATA. The pairwise correlation coefficients among the regressors are all statistically significant and less than 0.65, which is high but not high enough to worry about serious multicollinearity. The VIF values for the regressors (less than 10) also indicate that no serious concern for multicollinnearity exists. Some tend to be more conservative. However, we ran ordered logit models with and without variables that have the potential to cause multicollinearity, and the results are stable. Furthermore, Brambor et al. (Citation2006) argue,
Even if there really is high multicollinearity and this leads to large standard errors on the model parameters, it is important to remember that these standard errors are never in any sense ‘too’ large – they are always the ‘correct’ standard errors. (p. 70)
They also maintain that the standard error of marginal effect of X on Y is the interest of a multiplicative interaction model, not the standard errors of the coefficients of constitutive terms (Brambor et al. Citation2006, p. 70). The formulas for the marginal effects of corruption perception and their standard errors are shown in Table 4.
The chi-square from the brant test for each model is 17.07 (p = 0.648), 16.23 (p = 0.702), and 17.69 (p = 0.608), respectively.