Control Variables

half” as 0, and “common,”; “very common”; “more than half”; and “all” coded as 1.

3.2.2 Independent Variable

The main independent variable for the empirical analysis is the level of anti-corruption performance, operationalized as how much efforts that the CICIG had made to fight against corruption in Guatemala. The presence of the government’s anti-corruption efforts suggests specific actions taken against corrupted and corruption dynamics. In Guatemala, the press releases of the CICIG used to be an essential information source for the society to follow the anti-corruption agenda. The information about corruption include 1) the number of individuals identified as involving corruption crimes but not charged with sentences or fines, and 2) those that are charged with sentences or fines. Most of the individuals under CICIG’s investigations involved issues of corruption, impunity, and bribery.

I coded the data using CICIG’s annual reports on a year basis. The variables are lagged by one year to avoid endogeneity. For instance, all the individual respondents for the 2012 LAPOP survey are assigned the value of the CICIG anti-corruption performance in 2011. I expect that other things being equal, an individual is more likely to perceive corruption when the total number of individuals under CICIG’s investigation of corruption is higher. I also examine how anti-corruption performance affects corruption perception by decomposing the independent variable. Specifically, I expect that an individual is more likely to perceive corruption when more individuals are identified as involving corruption crimes but not charged with sentences or fines. I also expect that an individual is more likely to perceive corruption when more individuals are identified as involving corruption crimes charged with sentences or fines.

3.2.3 Control Variables

Education Attainment. The years of education an individual has affected her perception of corruption. For instance, Melgar et al. (2010) and Birdsall, Kenny, and Diofasi (2018) show

‧

‧

more able to process such information, they tend to perceive more corruption. Therefore, education attainment is controlled for its possible impacts on the perception of corruption.

Political Ideology. As Navia et. al (2019) and Julnes and Villoria (2014) show, political ideology can influence the perception of corruption. Specifically, rightist-leaning individuals tend to be more skeptical about the state than those in the left that center their beliefs on the role of the state. It is expected that those in the right-wing will perceived more corruption than the rest. The LAPOP surveys coded political ideology using a 1 to 10 scale. I follow Navia et al. (2019) to construct three dummy variables for right-wing (7-10), left-wing (1-3), and centrist (4-6). I include the dummy variables for right-wing and left-wing in the models, with no response as the reference group.

Media exposure. Another critical factor that affects the perception of corruption is the frequency individuals read the news. In general, mass media in democracies tend to report corruption scandals. Therefore, when an individual has more exposure to media, he or she will be likely to receive information about corruption, and thus this person will perceive more corruption (Weyland 1998). The LAPOP surveys coded the frequency of following news in mass media as a categorical variable. In my analysis, I generated five dummy variables for every day, every week, every month, and rarely. I include all of the dummy variables except

“never read news,” which is excluded as the reference group. I expect that individuals that read news more frequently are more likely to perceive corruption.

Economic Outlook. Navia et al. (2019) argue that the relationship between an individual’s outlook toward national economic performance and the perception of corruption was shaped by the evaluation of this person’s future income. It is expected that an individual that considers the national economic situation as better in the past twelve months will perceive less corruption than the rest. Moreover, a person with a positive evaluation of personal economic situation will be more tolerant of corruption, and therefore, perception might not be affected. Sociotropic and egotropic economic outlook are coded as trichotomous variables in my empirical analyses.

‧

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

Presidential Approval. Gómez-Vilchiz’s (2012) study shows that as the perception of corruption increases, the presidential approval decrease. In my analysis, I believe that presidential approval might have a negative impact on the corruption perception. Using the LAPOP survey data, I coded presidential evaluation as trichotomous variable. I include presidential approval and presidential disapproval in the models, and leave the regular approval as the reference group. Table 1 displays all of the variables used in the models.

Table 2 provides summary statistics for all the variables used in the analyses.

‧

國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

‧

Table 1 Description of Variables Used for the Empirical Analyses

Variables Variables Description

Dependent variable Perception of corruption

coded 1 if the response is very generalized or somewhat generalized; 0 if the response is little generalized or not generalized.

Independent Variables

Total number of CICIG’s investigations Total number of individuals under CICIG’s investigation of corruption

Number of uncharged investigations Number of people identified as involving corruption crimes but not charged with sentences or fines

Number of charged investigations Number of people charged with sentences or fines because of committing crimes of corruption

Control Variables

Female 1 for female respondent; 0 otherwise

Rural 1 for respondent who lives in rural area; 0

otherwise

Age Age of the respondent.

Country economic outlook Sociotropic evaluation, coded 1 if the response is better; 0.5 if the response is same; 0 if the response is worst.

Individual economic outlook Egotropic evaluation, coded 1 if the response is better; 0.5 if the response is same; 0 if the response is worst.

Ideology left

Ideology left if the respondent give 1 to 3 answer;

0 otherwise

*The values none and 1– 10

Ideology center

Ideology center if the respondent give 4 to 6;

‧

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

Ideology right

Ideology right if the respondent give 7 to 10 answer;

0 otherwise

*The values none and 1– 10

Education Attainment Respondent total years for education attainment

Presidential approval coded 1 if presidential approval;

0 otherwise

Presidential disapproval coded 1 if presidential disapproval;

0 otherwise

Presidential regular coded 1 if presidential regular;

0 otherwise

News everyday coded 1 if respondent follow news every

day ; 0 otherwise

News week coded 1 if respondent follow news every

week;0 otherwise

News month coded 1 if respondent follow news every

month;0 otherwise

News rarely coded 1 if respondent follow news rarely;;0

otherwise

‧

國

立 政 治 大 學

‧

N a tio na

l C h engchi U ni ve rs it y

‧

Variable Mean Standard

Deviatio

Total number of CICIG’s investigations 119.4 49.63 32 178 Number of uncharged investigations 93.35 50.92 21 165

Number of charged investigations 26.02 22.6 1 57

Female 0.504 0.5 0 1

Presidential disapproval 0.215 0.411 0 1

Presidential regular 0.513 0.5 0 1

‧

Because my dependent variable is coded as dichotomous, the main models are estimated using probit. I also estimated logit models for robustness checks. The predicted probabilities were generated using CLARIFY (Tomz, Wittenberg, & King 1999; King, Tomz, &

Wittenberg 2000). My data has a pooled cross-sectional structure for the years 2010 to 2019.

The results from the models on the perception of corruption can be seen in Table 3.

As expected, in Model 1 and Model 3, the total number of CICIG investigations has a positive and statistically significant effect on the perception of corruption among Guatemalans. It indicates that the government’s anti-corruption effort is an important determinant of perception of corruption in the country from its begin in 2008 to the end in 2018. In the Model 2 and Model 4, I decompose the total number of CICIG investigations into two variables. Both coefficients for the number of charged investigations and the number of uncharged investigations are positive and statistically significant. This finding suggests that the number of people identified with corruption with sentences or without them increases the likelihood that an individual perceives corruption in the country. More importantly, the results demonstrate that the number of charged investigations has a more significant coefficient than that for the number of uncharged investigations. This finding suggests that the number of charged investigations makes citizens more likely to perceive corruption.

Table 3 also demonstrates exciting findings for the control variables. Among Guatemalans, older generations are more likely to perceive corruption. Also, those that consider that the country’s national economic performances are negative are more likely to increase the likelihood of perceived corruption. However, the coefficient of egotropic economic outlook does not reach statistical significance. Education attainment has a positive and statistically significant coefficient for the perception of corruption. Specifically, higher educated individuals tend to perceive corruption. This finding suggests that highly educated individuals

‧