Causes of Corruption  
By

Larry Pounders and Michele Hites


       
The Public Integrity Section of the Department of Justice puts out an annual report to congress addressing public corruption.  In this report the department lists the number of convictions per state for crimes involving abuse of public trust.  Our study is focused on the causes of public corruption and as such we found this data to be invaluable.  To begin our study we did a preliminary literature review which yielded only a handful of documents on point most of which dealing with corruption on a broader scale.  One report of particular similarity was released by the Corporate Crime Reporter January 16, 2004 and written by the national press club of Washington D.C. titled “Public Corruption in the United States.  This report addressed Public integrity laws and their effects on public corruption finding little if any correlation.  We followed suit looking more specifically at Freedom of information laws and their effects on corruption but found support for the findings of the paper written by the National Press Club.  We found little correlation in laws controlling openness of government and corruption convictions.  We quickly abandoned this pursuit after our first few attempts yielded nothing of significance.  We began to stray from the other literature and look for ways to improve the existing methods of approaching the issues and began searching for new causes to compare and new variables to attempt to control.

    We reworked our data to create, what was to us, a new dependant variable.  We recognized initially that depend variable, corruption convictions, needed to be modified to make it more fitting for our endeavors.  First off we won’t to control for state size.  Other papers have used convictions per one-hundred thousand persons.  We worked up what we felt would be a more accurate approach.  We found U.S. Census data which listed the number of public officials per state.  We combined the public corruption conviction data and the data showing the number of public officials to create a dependant variable which ended up being the number of corruption convictions per one-hundred thousand public officials.

    I want to take a moment to caveat. Our data covers only those corrupt officials actually convicted by federal persecutors.  This means that our model does not encompass unprotected corruption or corruption prosecuted by state and local officials under state and local law.  We are looking for ways to improve our model by controlling for state and local impacts on federal convictions but we have yet to discover such a control.

    While there are aspects uncontrolled in this model which may have marginal effects on the overall accuracy of the model we feel that such uncontrolled events are small and insignificant although we continue to pursue methods of controlling these issues.  We have controlled for many issues we feel will most seriously cause a disconnect in corruption convictions and actual corruption.   Primarily, we control for size of government which we feel is paramount and judicial philosophy which we will discuss next.

    Starting from our newly configured dependant variable we wanted to further control the model for jurisdictional differences.  These differences would include alteration in judicial interpretation and judicial tendencies which would impact conviction rates and prosecution rates.  To control for this we decided to include Federal Appellate districts as independent variables leaving out the 5th circuit as a control.  We did this to prevent precedence handed down from appellate districts, the highest regional federal courts who would have final say on matters prior to Supreme Court of the land which would apply equally to all states covered in the model, from creating a disconnect between corruption and conviction levels.  Interestingly while working with the numbers at times it appeared that a handful of districts may have had significantly lower than the others districts which may indicate some precedence found in those districts impacts corruption in a desirable way or else conversely prevents convictions for corrupt officials.  Simultaneously it is completely possible that these effects are entirely coincidental or caused by regional and other co-linear issues.  That said, this may be an interesting topic of investigation for future projects.      Next, we wanted to control our model for the eventuality that corruption may prevent corruption from coming to light while also breeding more corruption.  We included an independent variable which bought historic corruption into the model.  For this we combined past corruption in much the same way as we did for our dependant variable and included that as an independent variable.  This variable had significant impact on corruption convictions as we expected but with seemingly small impact as far as absolute numeric values are conserved.  Historic corruption has coefficient of .0225 and a T statistic of 1.07.      Since openness laws had little impact we began to search for policies which are generally considered to be corruption fighting policies but which where not covered in the literature we found in our preliminary review of the topic.  We decided to look at the impacts of campaign finance.  We found no scientific significance as limitations on corporate contributions was only significant to the about the 80th percentile with a P value of .197 however to our surprise this number was negative indicating the possibility that corruption could actually increase as governments limit corporate contribution.  Again we feel as though this would be an interesting and important issue for further investigation though we couldn’t completely address it in this paper.

    Our flagship variable is our party dominance.  To create a party dominance variable we gathered data from The Book of the States.  We subtracted the total number of democrats from republics and dividing that number by the total number of officials in the houses of congress in each state.  We then took the absolute value of the resulting figures so as to represent party dominance without respect to which party dominated.  Our data for this variable spanned from 1996 through 2004.  The resulting variable was representative of the relative control of any party over the applicable time frame without regard to which party dominated.  The variable had natural limits of zero and two with zero representing perfectly divided government during the subject period while a score of 2 would represent total domination with one party controlling ever open seat.  Party dominance did turn out to be very significant in corruption convictions among the state and gave a T statistic of 2.283 and coefficient of 2.407.  We prove here that in more evenly divided government convictions for corruption go down while convictions for corruption are high when one part has solidified power.      Our model as a whole was surprising to us.  Our final regression resulted in a Chow of 3.404 with a significance of 0.0018.  Our R² was .551.  These numbers show that our model as a whole explains a little better than 52% of corruption at a scientifically significant level.  This was better than expected for a cross-sectional model which is less inclusive than a time series which we intend to undertake in the near future.

    In summation the model as a whole was a good start in our study of this topic and breaths incite into the causes of corruption.  We found Statistical proof that party dominances impacts corruption.  We further found that based on convictions in the United States there is a statistically significance correlation between historic corruptions and present corruption.  Finaly, we opened for discussion the possible impact of the judicial system, campaign contributions and freedom of information legislation on corruption in government.

Liberty, power, philosophy, Freedom, Taxes, Government , Immigration, Corruption