Micro Level using the WVS

In a next step we are going to use an alternative data source to check whether the previously obtained results remain robust. As discussed, we are using a slightly different proxy for conditional corruption. The WVS provides the possibility to explore a large set of countries and further regions. This also provides the opportunity to explore the relevance of conditional corruption at the macro level. We work with average values within each country using for our dependent variable the 4 point scale (0 to 3). Figure 5-1 shows a relatively strong negative correlation (Pearson r=-0.42), significant at the 0.01 level. Looking at the linear relationship in a simple regression shows that conditional corruption can explain 18 percent of the total variance of the justifiability of corruption.

In general, empirical support for a theoretical foundation depends not only on the validity of the theory but also on the quality of the data. It is not possible to ascertain with survey data whether respondents are truthful in their answers as truth is not observable by the interviewers (Kanazawa 2005). To validate statements one could explore the correlation between respondents’ statements and the Control of Corruption variable at the macro level

using country averages. Figure 5-3 shows an expected positive correlation (Pearson r=0.38) statistically significant at the 0.05

Working with the WVS we are also able to control for the similar independent variables⁷³. Table 5-5 presents the first results. We explore regressions with regional or country fixed effects. Moreover, we provide evidence with and without the income variable.

In addition, we also include sequentially the macro corruption variable Control of Corruption.

In all the specifications the variable Perceived Level of Corruption is statistically significant with marginal effects between 0.6 and 3.5 percentage points. In addition, the macro variable Control of Corruption is also statistically significant with marginal effects close to 6 percentage points. To deal with the social desirability problem we also change the cut-off point (see last specification in Table 5- 1). The values 1 and 2 in the original scale have been coded as 1 and all other values as 0. The coefficient is highly statistically significant, reporting even larger marginal effects than comparable results in specification (20). Thus, we can conclude that conditional corruption is also observable when using alternative data sources. The control variables show similar tendencies. A higher level of political interest is correlated with a lower justifiability of corruption. Risk averse and married people are also less inclined to justify corruption. On the other hand, self-employed individuals are more likely to justify corruption. Similarly, we also observe an age and gender effect. However, the effects of religiosity, urbanization and income are less strong.

73 See definition of the variables in previous footnotes.  

Table 5-5 Conditional corruption using WVS

Dependent Variable: Justifiability Of Corruption

Weighted Probit

Coeff z-Stat. Marg. Coeff. z-Stat. Marg. Coeff. z-Stat. Marg. Coeff. z-Stat. Marg. Coeff. z-Stat. Marg. Coeff. z-Stat. Marg.

(19) (20) (21) (22) (23) (24)

Change of cut-off point

Conditional Corruption

Perceived Level Of Corruption (Plc)

-0.112*** -11.90 -0.035 -0.020* -1.81 -0.006 -0.034*** -2.81 -0.010 -0.035** -2.50 -0.011 -0.043*** -2.95 -0.013 -0.041*** -3.35 -0.009

Control Of Corruption 0.19*** 15.18 0.059 0.19*** 14.09 0.058

Formal And Informal Education

Politicial Interest 0.025*** 3.02 0.008 0.031*** 3.55 0.010 0.021** 2.17 0.006 0.036*** 3.15 0.011 0.030** 2.52 0.009 0.051*** 5.18 0.011

Formal -0.002 -0.64 -0.001 0.003 0.77 0.001 0.006 1.25 0.002 0.005 0.95 0.002 0.010* 1.75 0.003 0.006 1.34 0.001

Demographic Factors

Age 30-49 0.20*** 9.91 0.064 0.20*** 9.20 0.059 0.19*** 8.51 0.059 0.18*** 6.52 0.055 0.18*** 6.43 0.057 0.19*** 8.36 0.043

Age 50-64 0.41*** 15.37 0.12 0.40*** 14.46 0.11 0.40*** 13.50 0.11 0.38*** 10.83 0.11 0.38*** 10.23 0.11 0.39*** 12.87 0.077

Age 65+ 0.57*** 16.11 0.15 0.53*** 14.45 0.14 0.53*** 13.04 0.14 0.55*** 11.50 0.14 0.53*** 10.71 0.14 0.53*** 12.55 0.094

Female 0.14*** 8.89 0.045 0.14*** 8.67 0.044 0.13*** 7.40 0.041 0.16*** 7.58 0.051 0.15*** 6.83 0.048 0.16*** 8.61 0.035

Marital Status

Married 0.10*** 4.99 0.033 0.12*** 5.67 0.038 0.14*** 5.81 0.042 0.13*** 4.64 0.041 0.13*** 4.53 0.042 0.13*** 5.35 0.029

Widowed 0.086** 2.18 0.026 0.084** 2.01 0.025 0.092** 2.04 0.028 0.095* 1.87 0.029 0.072 1.36 0.022 0.098** 2.08 0.021

Divorced 0.020 0.48 0.006 -0.001 -0.02 0.000 0.029 0.64 0.009 0.026 0.49 0.008 0.024 0.43 0.007 -0.0001 0.00 0.000

Separated 0.069 1.22 0.021 0.015 0.26 0.005 0.008 0.14 0.003 0.066 0.87 0.020 0.054 0.69 0.017 -0.0002 0.00 0.000

Employment Status

Selfemployed -0.078*** -2.64 -0.025 -0.088*** -2.87 -0.028 -0.10*** -3.09 -0.032 -0.098*** -2.65 -0.032 -0.11*** -2.70 -0.034 -0.087** -2.59 -0.020 Risk Attitudes

Risk Averse 0.084*** 4.95 0.026 0.071*** 4.03 0.022 0.068*** 3.49 0.021 0.080*** 3.43 0.025 0.075*** 3.07 0.023 0.073*** 3.70 0.016

Urbanization

Urbanization 0.001 0.40 0.000 -0.008** -2.30 -0.002 -0.007* -1.80 -0.002 -0.010** -2.40 -0.003 -0.008* -1.77 -0.002 -0.005 -1.23 -0.001 Religiosity

Church Attendance 0.0002 0.04 0.0001 0.018*** 3.86 0.006 0.017*** 3.40 0.005 -0.003 -0.58 -0.001 -0.005 -0.86 -0.002 0.021*** 4.02 0.005

Economic Situation

Income -0.006 -1.42 -0.002 -0.011** -2.48 -0.004

Regional Fixed Effects YES NO NO YES YES NO

Country Fixed Effects NO YES YES NO NO YES

Pseudo R2 0.038 0.082 0.11 0.059 0.058 0.099

Number of observations 37759 37759 32096 20793 18914 37759 Prob > chi2 0.00 0.00 0.00 0.00 0.00 0.00

Notes: The reference group consists of Age<30, Man, Single/Living Together, Other Employment Status. ***, ** and * denote significance at 1%, 5% and 10%, respectively. Robust standard errors.

Table 5-6 2SLS results (WVS)

Dependent Variable: Justifiability of Corruption Coeff t-Stat. Coeff. t-Stat.

(25) (26)

Conditional Corruption

Perceived Corruption (PLC) -0.044** -2.43 -0.061*** -3.04

Formal And Informal Education

Political Interest 0.007** 2.35 0.003 0.84

Formal -0.001 -0.94 0.000 -0.29

Church Attendance -0.0002 -0.15 -0.002 -1.16

Economic Situation

Notes: The reference group consists of Age<30, Man, Single/Living Together, Other Employment Status. ***,

** and * denote significance at 1%, 5% and 10%, respectively. Robust standard errors.

Table 5-6 presents 2SLS estimations using generalized trust as an instrument for perceived corruption (in line with Table 5-3). Looking at the first stage regressions and the diagnostic tests we can conclude that generalized trust is a good instrument⁷⁴. The results also show that Perceived Corruption (PLC) remains statistically significant, providing therefore further support for previous findings. We report additional findings in Table 5-7 obtained with a filtered PLC variable using previous specifications. Also here we observe that the PLC coefficient is always statistically significant with a negative sign. Thus, even after filtering we can conclude that conditional corruption matters.

74 The WVS does not provide the possibility to consider an index of perceived honesty.  

Table 5-7 Causality discussion filtering with WVS Data Depend. V.: Justifiability of Corruption

(Highest Value = Never Justified) Weighted Probit

Coeff. z-Stat. Marg. Effects

INDEPENDENT V. (see specifications)

Specification (27)

Filtered PLC using specification (19) -0.13*** -12.87 -0.041

Specification (28)

Filtered PLC using specification (20) -0.020* -1.75 -0.006

Specification (29)

Filtered PLC using specification (21) -0.026** -2.08 -0.008

Specification (30)

Filtered PLC using specification (22) -0.035** -2.50 -0.011 Control of Corruption 0.19*** 15.18 0.059

Specification (31)

Filtered PLC using specification (23) -0.043*** -2.95 -0.013 Control of Corruption 0.19*** 14.09 0.058

Notes: Summary of four regressions. The reference group consists of Age<30, Man, Single/Living Together, Other Employment Status. ***, ** and * denote significance at 1%, 5%

and 10%, respectively. Robust standard errors.

We conduct a further robustness test to deal with a potential “social desirability” bias using the EVS and WVS. We run a two-stage approach where the previous estimations were just the first stage. First, respondents decide whether or not to answer that corruption is never justified (“socially correct response”). In a second stage, given the decision to answer something other than the socially correct response, individuals report a value from the remaining scale (1 to 9).

The results are not reported in a table but indicate that our conditional corruption variable is always statistically significant. Despite trying to check the causality relationship one should note that providing a clear causality relationship is quite problematic working with such micro survey data. To some extent we see these results as more precisely estimated partial correlations and not fully precise estimates of a causal relationship (see also Guiso et al. 2003).

In the next stage we are going to explore the importance of conditional corruption at the macro level over time in order to explore hypothesis 2.