Household Income and Relationships with Different Power Entities as Determinants of Corruption

This article adds to the corruption literature by identifying factors influencing Bangladeshi farm households’ probability of experiencing corruption in different service sectors. The econometric results show that households’ probability of being exposed to corruption can largely be explained through their income and their relationship with different power entities. The direction of the relationship between income and corruption vary across services. Relatively rich households have a higher probability of experiencing corruption in sectors such as education, health and electricity. These households are less likely to experience corruption in local government and agricultural extension services. The results here are contrary to the common trend in corruption research that addresses households’ aggregate corruption experiences. Households with relationships with different power entities have a lower probability of experiencing corruption than their counterparts without these types of relationships.


Introduction
When interacting with different public service entities, some households experience corruption and some do not. There are also differences in the forms of corruption experienced by households. Such differences under the same policy and institutional setup can possibly be explained through different household and community level characteristics. The available literature on the micro-level determinants of corruption identifies several factors responsible for households' corruption experiences. These factors include income, education, location, gender, marital status, profession, attitude and perception towards corruption, trust network, city size, etc. (Anik, Bauer, & Alam, 2013;Čábelková & Hanousek, 2004;Herrera, Razafindrakoto, & Rouband, 2007;Hunt, 2004;Mocan, 2008;Shaw, 2009;Swamy et al., 2001;Torgler & Valev, 2006). The literature aggregates households' experiences gained through interactions across sectors. For example, Anik et al. (2013) collected households' experiences with ten different sectors, but converted experience into a single Household Income and Relationships with Different Power Entities as Determinants of Corruption dummy variable to identify corruption determinants. Mocan (2008) used the International Crime Victim Survey (ICVS), in which individuals were asked to identify their bribery experiences with government officials without specifying any sector or service. Though the literature provides important insights about microlevel corruption determinants, this aggregation might be misleading. Because demand for services varies across households based on differences in socio-economic characteristics, assuming that a specific type of household is equally vulnerable to corruption in all sectors may be inappropriate. Furthermore, household level characteristics may influence households' probability of experiencing a particular form of corruption (Anik et al., 2013), and a particular service may be more prone to a specific form of corruption.
In the literature, the most pronounced determinant of corruption is a household's economic status, generally measured by income. Rich people have a higher probability of experiencing corruption, specifically bribery (Hunt, 2004;Mocan, 2008;Torgler & Valev, 2006). The rich have higher demand and hence higher interactions with service entities, which increases their probability of exposure to corruption. To a corrupt official, wealthy households are a better potential source for extracting bribes than those with less wealth. Furthermore, as the rich have relatively less marginal cost for a bribe of the same amount than the poor, they might be assumed to be less reluctant to enjoy services through bribery. However, an alternative hypothesis can be offered here. Demand for services such as relief, social safety net programs, etc. decreases with an increasing income level and people belonging to lower-income deciles may become more vulnerable to Moreover, concerns about perception biases can lead to doubts about the acceptability of these studies. The corruption literature is primarily concerned with the impact of corruption on different macro-level economic or development indicators, primarily due to the unavailability of micro-data versus the relatively easy access to aggregate-level corruption data (Mocan, 2008). Due to the secretive nature of corruption, in most cases it is difficult to find direct witnesses of corruption, especially when it creates a win-win situation (Johnston, 2000).
In this article, we try to address some of these limi- The entire paper is divided into five sections. This introductory section is followed by the literature review section. The third section contains the data and the analytical procedure, and describes the sampling techniques, data and econometric models. The next section offers the results and discussion, and the findings of the study are presented and discussed. Finally, the results are summarized in the conclusion.

Household Income and Relationships with Different Power Entities as Determinants of Corruption
In cross-country analysis, they found that corruption is less severe in countries in which women make a relatively higher contribution in parliament, senior bureaucratic positions and the labor force. Similar findings are noted in the works of Alatas et al. (2006), Mocan and Rees (2005), Mocan (2008), Torgler and Valev (2006). Mocan's (2008) analysis using data from 49 countries also showed that high income individuals, people living in large cities and those with more education have a higher probability of being exposed to corruption. He also mentioned several country level characteristics that influence corruption probabilities.
These include a country's unemployment rate, average education, and the strength of its institutions.
Through ordered probit models, Torgler and Valev (2006) investigated the determinants of differences in the justifiability of corruption in eight Western European countries. Exploring the relationship between age and corruption was these researchers' primary interest.
They found strong age effect (the changing attitudes of the same cohort over time), but no cohort effect (differences in attitudes among similar age groups in different time periods). Due to the higher degree of social norms, more educated and married people have lower justifiability of bribery. Among different economic classes, the highest economic class has the lowest justifiability of corruption. As the marginal utility loss (wealth reduction) when caught and penalized for corruption is lower for individuals with a higher income, these individuals are more likely to accept bribery. Herrera et al. (2007) also found the incidence of petty corruption more common among richer households in Peru. They estimated that bribery as a share of food expenditure is higher for households belonging to the upper economic classes.
Dividing the sample into different income groups, Hunt (2004) (2003) found a positive correlation between firm profit and bribe amount. Čábelková and Hanousek (2004) explored the role of perception as a determinant of corruption by analyzing 2600 Ukrainian respondents' opinions. Among different professional groups, they found businessmen and peasants to be more willing to engage in bribery.
As businessmen are more likely to extract a benefit from bribery than others, their willingness to bribe is also higher. To justify Ukrainian peasants' higher willingness to bribe, they referred to Harrison (1985).
Peasants tend to believe that things that exist in this world are finite/limited in quantities, similar to their farm area and land productivity. Hence, they believe less in their own capability and in effort as a source for betterment. For peasants, grabbing a bigger piece of the social pie is the easiest and most comprehensible path to betterment. Bribing dishonest government officials to obtain a higher share of the social pie is therefore a more practiced option. The study also emphasized the media's role in controlling corruption by creating perceptions. If media broadcast that there is a higher level of corruption than actual, some citizens may gain the impression that bribery is required in government offices. Ultimately, some people might be encouraged to bribe.

Data and Empirical Model
Sampling Techniques and Data: The empirical analysis is based on information collected from 210 Bangladeshi farm households belonging to six villages in six different districts of the country. Among these households, 99.1% received service from at least one sector during July 08-June 09. The highest proportion of households received service from the education sector (78.6%). The education sector is followed by health (66.2%), electricity (48.6%), agricultural extension (42.1%) and local government (41.0%) in terms of number of service recipients (Table 1). A multi-stage sampling technique was applied to select the respondent households. The first three stages were for selecting the survey areas (e.g., selection of districts, upazila 1 and villages), whereas respondent farmers were select-    Analytical Model: The earlier discussion shows that notable differences exist in households' corruption experiences across sectors, e.g., a household experiencing corruption in one sector may or may not experience corruption in other sectors. Forms of corruption also vary across sectors. We try to address these differences through some individual and community level factors.

Forms of corruption
To analyze households' corruption experience (did the household face corruption in that sector during the referenced period or not) in binary terms, we use the probit model approach. For each sector or service, a probit model is developed. As the number of service recipients varies across sectors, so does the number of observations across models. A household's corruption experience in a sector is the dependent variable in the model developed for that specific sector. A probit model is the most commonly used econometric tool in the microeconomic literature exploring corruption determinants (e.g., Alatas et al., 2006;Čábelková & Hanousek, 2004;Hunt, 2004;Hunt & Laszlo, 2005;Shaw, 2009;Swamy et al., 2001;Torgler & Valev, 2006 where P r is the probability; i y is the i th household's corruption experience, binary in nature and varying across models; Φ is the cumulative distribution function (CDF) of the standard normal distribution; methods (Bascle, 2008;Hahn, Hausman, & Kuersteiner, 2004;Murray, 2006 One may also suspect endogeneity problems between a household's corruption experiences and different relationship variables. Past experiences with corruption and/or a future expectation of facing cor-ruption may encourage a household to develop relationships with power entities. In our models, we consider a household to have a relationship with a power entity only when the household claimed to have their 'members and/or close relatives and/or friends' within the power entity. Even after having such a person in a power entity, a household was not treated as having a relationship unless the household claimed that its request of the power entity was somewhat obligatory.
A relationship that induces obligation cannot be developed in a short period of time. When a household faces or expects to face corruption, it may try to develop a relationship with the power entity. However, such a relationship may not imply any obligation by the power entity. The power entity may or may not feel the need to do something for the household. In our case, by mentioning the word 'obligatory' as a criterion to identify 'close relatives and/or friends' , we limited the scope of inclusion for any short-term relationship.  Note: a A household is identified as having a relationship only after satisfying the following conditions:

Measurement technique
• The household has family member(s) and/or relative(s) and/or friend(s) as public representatives or working in the institution(s) or organization(s); these persons hold some influence over the service delivery entities. • When households make requests to the aforementioned persons, these persons will feel obligated to carry out the request and have the ability to carry it out. It may not always be possible for a household to try to develop relationships with a power entity when it faces corruption in the service sector, mostly due to time constraints. A household must decide whether or not to pay bribes while waiting in the service queue. It is less likely that a household returns from the queue and tries to develop a relationship with the power entity; e.g., while admitting a member to the hospital, if a household is asked for a bribe, it is hard to believe that instead of paying the bribe, the household will return from the hospital and try to develop a relationship with a power entity to avoid bribery. Hence, we believe there are enough grounds to ignore simultaneous causality between relationship variables and households' corruption experiences. However, in the longer term, simultaneous causality might be an issue. Because we have cross-sectional data, it is beyond the scope of this study to analyze such possibilities. This might be a new arena for future research.

Results and Discussions
Descriptive Statistics of the Explanatory Variables  (Table 5).

Factors Influencing Households' Probability of Experiencing Corruption: Household level determinants of corruption experiences in different sectors
are presented in Table 6. We tried alternative model The estimated relationship between income and corruption probability in education, health and electricity are consistent with the findings of Hunt (2004), Herrera et al. (2007), Torgler and Valev (2006) and Mocan (2008). The literature has mentioned several reasons for a rich household's higher probability of being exposed to corruption. These reasons include higher demand for services and hence more interactions with service delivery entities, higher ability to pay bribes, and a relatively lower marginal cost for a bribe of the same amount.
However, these arguments become insufficient when explaining the inverse relationship between income and corruption in local government and ex-  Unlike in other sectors, in extension services there was no incidence of bribery. A higher ability to pay bribes increases rich households' probability of facing corruption (Herrera et al., 2007;Mocan, 2008;Torgler & Valev, 2006). The result here can be more specifically    corruption in sectors such as education, health and electricity. The same variable significantly reduces a household's probability of being exposed to corruption in local government and agricultural extension services. The significant impact of the income variable highlights officials' tendencies to maximize bribes.
Relationships with different power entities reduce a household's probability of being exposed to corruption. This inverse relationship is an outcome of the officials' tendency to avoid conflict with the power entities in anticipation of the implicit form of Hunt's (2004)  3 The survey selected extensive rice producing areas because agricultural extension services (e.g., training, demonstration plots and advisory services, etc.) are greater in these areas. Furthermore, a relatively high share of the agricultural subsidy goes to these areas. 4 We performed the Amemiya-Lee-Newey overidentification test for the validity of the used instruments (Lee, 1992). The test statistic is insignificant, which argues for the validity of the instruments. The Smith-Blundell test was performed to test the endogeneity of the income variable (Smith and Blundell, 1986 (Wooldridge, 2002). Failure to reject the null hypothesis means that the error terms in the structural equation and instrumented regression are uncorrelated, and therefore following the IV approach is not necessary.