Political importance and its relation to the federal prosecution of public corruption

Abstract

In the US, federal prosecutors are appointed by the president, confirmed by the Senate, and have significant discretion over which cases they choose to take to court. Federal prosecutors handling an overwhelming majority of corruption cases invites the possibility of political influence in the monitoring of corruption. Additionally, political disparities across states may result in differences in corrupt behavior. Using individual case level data, I examine the effect political factors have on federal corruption cases, with an emphasis on states that are an important focus in the next presidential election. I find that corruption convictions tend to be higher in politically important states. This effect seems more significant when Democratic administrations are in power. In addition, it seems that these effects are relevant only for corruption crimes labeled as “federal”.

1 Introduction

Political corruption, defined broadly as the “misuse of public power for private or political gain”, is ubiquitous.¹ However, given its secretive nature, measurement of political corruption is inherently difficult. Often researchers rely on indirect measures of corruption that come from surveys. In the United States, however, the number of corruption convictions per-capita is used as a proxy for state-level corruption (e.g., Goel and Nelson 1998; Glaeser and Saks 2006; Leeson and Sobel 2008; Alt and Lassen 2008). While this may be an improvement upon survey measures, there are two caveats that must be considered when using conviction rates as state-level corruption measures.

First, assuming that conviction rates are an accurate representation of state level corruption, there are important political differences across states that can influence corrupt behavior. If political differences also affect the outcome of interest, studies will suffer from omitted variable bias by ignoring these disparities. Second, the monitoring and enforcement mechanisms used to control corruption are susceptible to political influence. Voigt and Gutmann (2015) find that independence in the judiciary and in prosecutorial decisions is key to having a non-corrupt judiciary. Unfortunately, true independence regarding corruption cases seems to be lacking in the U.S., making convictions a potentially inaccurate measure of corruption.

The goal of this paper is to examine the effect political factors have on federal corruption convictions, with a particular focus on the political importance of the state in which the case is located. In this paper, a state is considered politically important based on a combination of three factors: (1) the closeness of the presidential election; (2) the number of electoral votes allocated to the state; (3) the easiness of swaying voters within that state.² This is the first paper to consider how corruption convictions systematically differ across states and time due to political factors using individual case file data.

A politically important state will likely have a larger number of corruption convictions for two reasons. First, the returns to corrupt behavior may be higher in politically important states, incentivizing corruption. If this is the case, then political importance is a potential omitted variable in previous state-level corruption analyses. Second, politically important states tend to be the focus of upcoming presidential elections. If an increase in highly publicized corruption convictions improves the image of the current administration, we may see more corruption convictions in these states even if corruption levels are constant across the U.S.

Swing states are seen in the popular press as being the fastest growing (Leubsdorf 2016).³ Thus, if there are additional benefits to being a politician in a state with higher than average growth, the returns to corrupt behavior may be higher as well. Additionally, non-politicians may be incentivized to engage in corruption in these same states. The ‘grease the wheels’ hypothesis of corruption suggests that corruption can be used to circumvent excessive regulations (Dreher and Gassebner 2013; Dutta and Sobel 2016).⁴ However, creating this type of mindset can lead to a vicious cycle of corruption and encourage unproductive entrepreneurship (Sobel 2008). If individuals see battleground states as an opportunity to bring a beneficial politician into office, they may be incentivized to engage in corrupt acts to achieve this goal. This increase in corrupt behavior on the part of both politicians and non-politicians will translate into more corruption convictions if convictions represent corruption.

Additionally, a state’s political importance has been shown to influence variety of political matters ranging from federal budget allocation decisions (Larcinese et al. 2006) to Federal Emergency Management Agency disaster declarations (Garrett and Sobel 2003). These political outcomes can have direct consequences on broader outcomes that are usually the focus of empirical corruption studies (e.g., per-capita income). If ignored, this implies that political importance is a relevant omitted variable, resulting in biased estimates.

Corruption convictions are also susceptible to influence from the current administration. In the US from 1986 to 2012, nearly 94% of all public corruption cases were handled in federal court (Cordis and Milyo 2016).⁵ This is important for two reasons: (1) appointment of US Attorneys and promotion of US and Assistant US Attorneys (AUSAs) tends to be politically motivated (Nyhan and Rehavi 2017); and (2) discretion in the decision to pursue specific cases lies mainly with the prosecutor (Gordon and Huber 2002; Shotts and Wiseman 2008; Gordon 2009; Rehavi and Starr 2014; Nyhan and Rehavi 2017). Combined, these two factors invite a significant amount of political influence over convictions, especially in states that are important to win in an upcoming election.

Federal prosecutors are appointed by the president, and confirmed by the Senate. Thus, federal prosecutors have a clear connection to the political party in power. Their likelihood of promotion to the federal judiciary has been shown to be partisan dependent as well (Nyhan and Rehavi 2017), further increasing their ties to the current administration. Federal prosecutors, however, do not handle a majority of cases themselves as most cases are assigned to the AUSAs.

Assistant US Attorneys may or may not seek promotion within the DOJ. Many AUSAs leave the DOJ entirely once they have gained enough experience to obtain a higher paying private sector job (Boylan and Long 2005). However, the AUSAs handling corruption cases are still likely sympathetic to the political party in power’s interests, regardless of their general career objectives. U.S. Attorneys can strategically assign cases to AUSAs with political interests in mind (Nyhan and Rehavi 2017). It is possible that corruption cases are disproportionately assigned to AUSAs seeking to move up within the ranks of the DOJ, where the AUSA’s superiors award promotions based on political factors (Nyhan and Rehavi 2017).

This potential for political influence would be less meaningful if there was no room for discretion. The prosecutor has the power to decide whether to pursue a case or dismiss all charges. Indictments of corruption typically result in a conviction.⁶ A declination, or dismissal of charges, is therefore the only sure way a defendant can escape a conviction. Additionally, the administration can alter the priority of case types, pushing certain cases to the front of the queue. For example, the FBI lists public corruption investigations as its top priority.⁷ This emphasis on corruption cases may be strongest in politically important states. The possibility of influence, combined with discretion, can affect the handling of corruption cases.

Political influence coming from the current administration may come in two, non-mutually exclusive, forms: (1) a partisan bias in the timing of case filings and the choice of defendants being convicted for corruption; (2) or an overall increase in corruption convictions to make the presidential administration appear strict on crime. While both influences likely factor into prosecutorial decision-making and the case’s priority, the focus of the paper will be on the latter form.

Since a corruption case can irreversibly harm a politician’s reputation, we may see that prosecutors are more willing to prosecute political opponents than political allies. Gordon (2009) finds evidence of this as federal prosecutors during the Clinton and (George W.) Bush administrations were more willing to file weaker cases against political opponents than political allies. Similarly, Nyhan and Rehavi (2017) find that opposing party defendants are more likely to be charged before an election than afterwards. However, a majority of defendants in both the Gordon (57%) and the Nyhan and Rehavi (2017) (75%) sample cannot be identified with either party. Thus, even if this form of bias is present it accounts for a small percentage of cases overall. The second form of bias is more general and likely more meaningful.

Corruption is a highly publicized crime, which will increase voter awareness regarding law enforcement effectiveness. This will be especially true immediately following the case disposal as the DOJ typically issues a public statement the same day the case is decided. This statement details the outcome of the case and the crime(s) involved. The Federal Bureau of Investigation (FBI) helps distribute these statements for corruption cases specifically by providing links to all DOJ press releases concerning public corruption.⁸ The media can easily obtain of this information and share it with the public. Importantly, since most defendants are not partisan identifiable the media coverage will likely be non-partisan. The administration may use the non-partisan publicized nature of these convictions to influence voter perception on crime.

Through the Transactional Records Access Clearinghouse (TRAC) data warehouse (TRACFED) I obtain individual case level data from administrative records of federal prosecutors’ workloads across each of the 94 judicial districts within the United States from 1993 to 2010. Using this data, I estimate the effect of the state’s political importance on federal cases of corruption across each state.⁹ I use this data to expand upon the current literature in several ways.

First, I examine how political influence over corruption cases systematically differs across the US states using case level data. Bologna (2016) explores the impact of political influence on corruption convictions using data from the US Department of Justice’s Public Integrity Section (PIN) annual Report to Congress on the Activities and Operations of PIN. The author finds that corruption convictions per-capita is dependent on the state’s level of political importance using the PIN data.¹⁰ However, since PIN data is constructed using a survey, it is unclear that corruption convictions truly vary systematically. The first major contribution of this paper is to show that political factors influence the true number of corruption convictions.

Additionally, this paper explores systematic differences in case dismissals as well using this TRAC data. We may see an increase in convictions simply because there are more cases. Alternatively, an increase in convictions may come from a decrease in dismissals. While this paper cannot empirically separate an increase in convictions from an increase in corruption per se, we can see if dismissals tend to move in the same direction as convictions.

Another contribution of this paper is to show that the systematic differences are present when using measures of corruption convictions and declinations that are not scale dependent. Many factors affect the number of corruption cases a state receives. While scaling output by population partially addresses these differences, there are likely omitted variables affecting the number of case disposals that cannot be appropriately controlled for. Additionally, if convictions and declinations are changing in the same direction, it is important to know if these changes are proportional. Therefore, I look at the effect of political importance on corruption convictions as a percentage of total cases disposed of. Additionally, I see if this estimated effect is unique to corruption cases.

A final contribution is to show that case disposals tend to be higher in presidential election years.¹¹ Political factors may not only cause corruption estimates to differ across states, but it also may cause corruption estimates to differ across time. This relates to the idea that retention creates incentives for officials to engage in policies that create a political business cycle. This includes both administrative and prosecutorial retention incentives. In the case of state attorneys, there is evidence that the probability of conviction tends to be higher, and the probability of dismissals tend to be lower, in the year an attorney faces re-election (Dyke 2007). US and Assistant US Attorneys do not face re-election incentives, but an election year signifies the end of term for US Attorneys and may signify promotion possibilities for AUSAs. To the extent that US and Assistant Attorneys are judged according to their case disposals (e.g., convictions), we would expect disposals to rise in election years. This will be especially true if the current administration values an increase in corruption convictions and can influence case priority.

As with the effects of political importance, it should be noted that increases in corruption convictions during election years could be unrelated to prosecutorial incentive bias. As political business cycle theory suggests, there may be a significant number of policies changes during election years. It is possible that these policy changes result in changes in the number of corrupt acts consequently altering the number of corruption convictions, regardless of prosecutorial and administration incentives. However, the finding does suggest that there is something unique occurring during election years that should be considered in subsequent analyses.

Overall, the evidence of this paper indicates that corruption convictions tend to be higher in politically important states. These results are robust to a variety of specifications and seem to be more significant in democratic administrations. In addition, it seems that political importance matters most for corruption crimes labeled as “federal”. While it is impossible to empirically distinguish between actual corruption and corruption convictions given current data availability, this paper highlights the impact political factors have on corruption conviction measures. In the very least, researchers should consider including a measure of political importance in future analyses and should be cognizant of election year effects.

A further review of the literature is presented in Sects. 2 and  3 presents the data; Sect. 4 describes the empirical methodology and results; Sect. 5 concludes.

2 Prosecutorial incentives

The goal of this paper is to show that political factors can affect corruption convictions. Corruption convictions may differ across states and through time due to differences in actual corruption levels, prosecutorial incentives, or administrative queue manipulation.¹² It is impossible to distinguish between each theory empirically given current data availability. However, understanding how prosecutorial incentives may affect corruption convictions is necessary, assuming the true level of corruption and ordering of cases are held constant.

There has been a significant amount of research exploring the incentives facing both US and Assistant US Attorneys. As noted in Boylan (2005), among others, common elements for prosecutorial evaluation include an attorney’s conviction rate, number of indictments, and sentence length. Consequently, much of the existing research explores how specific factors influence each outcome. However, much of this research focuses on non-political factors, and assumes constant output across time.

Boylan (2004) finds that lower salaries for US Attorneys increase turnover, and subsequently lowers output.¹³ This indicates that US Attorney jobs are competing with jobs in the private sector and that prosecutors respond to salary incentives as expected. Additionally, while Boylan (2005) finds that the average prison sentence length associated with a US Attorney during his/her term is positively related to favorable career outcomes overall, this effect was not constant across career choices. He finds that the conviction rate and the number of indictments have no significant effect on career outcomes. Boylan and Long (2005) examine how differences in local labor markets across judicial districts affect the behavior of AUSAs specifically. They find that Assistant US Attorneys tend to take more cases to trial, as opposed to plea bargains, in districts with high private salaries. In these districts, AUSAs seek government employment to gain the trial experience necessary to obtain a higher paying private sector job.

However, as noted above, these studies do not consider how political differences across locations affect output or how prosecutorial output varies over the course of a term. It is quite possible that the relevant maximizing criteria differs across districts for political reasons. It is also possible that the prosecutor’s end of term output differs from their full-term average.¹⁴ In both cases, we will see that these differences are even more probable when considering corruption cases.

If voters value an increase in convictions, the administration may place a heavier weight on the prosecutor’s conviction output when evaluating their performance if the prosecutor is located within a battleground area. This is even more likely with corruption cases considering their high profile nature. Some US and Assistant US Attorneys aspire to move up in the political world, while others do not. For attorneys with political aspirations, they have a clear incentive to handle corruption cases in a way that benefits the party in power. Even those handling corruption cases with no political aspirations tend to be sympathetic to the goals of the current administration. As noted above, Nyhan and Rehavi (2017) find evidence that prosecutors do tend to exhibit partisan bias when it comes to the handling of corruption cases. If prosecutors are loyal, or at least sympathetic, to the party in power we can expect them to reallocate their output according to the wishes of the current administration.

If re-election incentives result in locational differences in prosecutorial output, they likely result in time differences as well. More specifically, if the administration is concerned about keeping their party in power they may place a heavier weight on the performance of prosecutors during election years. This idea derives from political business cycle theories. Political cycles arise from the traditional principle-agent problem with informational asymmetries. Political figures in power are inclined to alter policy to influence outcomes visible to voters in a way that maximizes their party’s vote count. To the extent that voters do value an administration that is tough on crime, we may see an increase in prosecutorial output during election years. We may also see that output increases during election years because prosecutors are concerned about future employment opportunities in general. Regardless of whether the career aspirations driving prosecutorial actions are political or personal, the differences in output in election years will be more probable with corruption cases. While private firms may not care about the content of corruption cases per se, they will likely be more aware of the prosecutor’s quality due to the publicized nature of corruption cases.

Theory outlining the mechanisms through which federal prosecutors respond to retention and future employment incentives is scarce. However, we can utilize existing theories regarding response mechanisms of state-level prosecutors to re-election incentives. The consensus in the state-level prosecutorial literature is that prosecutors do respond to re-election incentives in the sense that prosecutorial output is affected by re-election. More specifically, there is empirical evidence showing that re-election pressures result in more cases going to trial (e.g., Bandyopadhyay and McCannon (2014; 2015a)). However, the appropriate mechanism behind this result has less agreement.

Bandyopadhyay and McCannon (2015b) note that there are two competing theories that can explain why more cases go to trial at the state level during election years. The first is that prosecutors exert more effort. The second is that, while effort exertion is constant, prosecutors begin pursuing jury trials in place of plea bargains. The intuition behind this latter theory is that a trial is used as a signaling mechanism to the state attorney’s constituents, but trials require more resources than plea bargains. Bandyopadhyay and McCannon (2015b) find evidence supporting the latter theory. Using district-level caseload data from North Carolina, they find that as case backlogs tend to increase, disposals tend to decrease, in election years.¹⁵ They find that the decrease in disposals is caused by a reduction in dismissals. Similarly, Dyke (2007) uses individual level data to estimate the effect election years have on individual case outcomes finding that the probability of conviction is higher and the probability of dismissal is lower in an election year. Thus, state prosecutors tend to increase their visibility and austerity in election years.

The goal of this paper is not to distinguish between the effort exertion theory and trial signaling theories outlined in Bandyopadhyay and McCannon (2014; 2015b) at the federal level. Both theories are plausible. This paper focuses on the effect of various pressures on corruption case disposals alone. The aim of this section is simply to understand why prosecutorial incentives may result in an increase in corruption convictions. Prosecutors may increase corruption case disposals without increasing their case backlog or increasing their exerted effort. They may simply make corruption cases their main priority over alternative cases to increase their own visibility. Alternatively, they may increase corruption disposals during election years via increased effort exertion. Testing which theory is more appropriate at the federal level is left to future research.

3 Data

The data for this paper can be classified into three groups: measures of prosecutorial output, measures of political influence, and basic controls. Sources, a brief description, and summary statistics of each variable are given in Tables 1, 2 and 3. The remainder of this section will discuss the construction and sources of each data group in turn.

Table 1 Dependent variable names, brief descriptions, and sources^c

Table 2 Basic summary statistics of dependent variables

Table 3 Names, brief descriptions, and summary statistics for independent variables

3.1 TRACFED administrative records data

The corruption measures are constructed using administrative records of case files obtained from the Transactional Records Access Clearinghouse (TRAC) data warehouse (TRACFED). This data contains information on each case in the federal prosecutors’ workload including the judicial district in which the case is being handled, the Department of Justice program category (e.g., “official corruption”), the lead charge associated with the case, and the date of the case’s referral. In addition, if the outcome of the case has been determined, the dataset also includes information on the date case was disposed of, the outcome of the disposal (e.g., guilty, immediate declination, etc.), and if applicable, sentencing information. Using this information, I develop several measures of prosecutorial output. I combine all of the information on cases within each judicial district in each state to make several state-level measures of prosecutorial output that is comparable to state-level measures of political importance.

I first examine political influence over prosecutorial output per-capita. The number of corruption convictions per-capita is commonly used as a proxy for actual corruption. Therefore, it is important to test if this measure systematically varies across states for reasons that may be unrelated to actual corruption. Therefore, I calculate the number of corruption convictions in each state per 100,000 people (Conviction Per-Capita). Additionally, I calculate the number of corruption declinations in each state per 100,000 people (Declinations Per-Capita) to see if the change in convictions comes at the expense of fewer dismissed cases.¹⁶ All measures are dated according to the disposal date.

The disposal date is used over the filing date primarily because I am interested in the variation of actual convictions, as annual convictions per-capita are commonly used as an indicator for annual corruption. This is not to say that the timing of the case filing is unimportant. Nyhan and Rehavi (2017) find evidence that prosecutors base corruption indictment timing decisions on election dates. However, the variation in the timing of case filings due to elections found in their study is generally contained within a year. Thus, these effects tend to aggregate out when studying annual data.

Additionally, there is reason to believe that actual convictions matter more for prosecutors than indictments alone, even though they have more control over the timing of the latter.¹⁷ Boylan (2005) finds that US Attorney’s tend to maximize total prison sentences for all convicted defendants rather than the number of indictments or the conviction rate (percent of indictments resulting in conviction). Since total prison sentence length is a function of the number of convictions a prosecutor obtains, it is quite possible that prosecutors are also maximizing the number of cases that result in convictions.¹⁸ Thus, if convictions increase during election years it may be because prosecutors are incentivized to increase their disposals. Alternatively, convictions may increase because the current administration alters case priority or because corruption truly increases in election years as discussed above. These effects are empirically indistinguishable.

Increases in convictions during election years, regardless of the source, may or may not be driven by increases in case filings in that same year. For example, prosecutors may reallocate their time at the end of their term in an effort to close ongoing corruption cases. To this end, prosecutors may purposely shorten or extend case length to have more convictions occur when they are most beneficial to their career. The administration may also help quicken the pace of ongoing cases near the end of the term. If this occurs, we may see an increase in convictions without an increase in case filings. Alternatively, it is possible that we see a spike in convictions during election years because prosecutors begin ramping up their case filings. The administration may also encourage the prosecutors to file more cases in election years. Thus, we may see a spike in both filings and convictions during election years. Regardless of the underlying mechanism, an increase in convictions during election years indicates that there is something inherently different going on in these years that needs further exploration.¹⁹

The length of time between referral and disposal can span across years and consequently across terms and prosecutors. It is not clear that cases filed under the guidance of a U.S. Attorney that previously held office would be weighted the same as cases initiated by prosecutors presently serving their term. Interestingly, this is something that is not explicitly acknowledged in the literature concerning federal prosecutorial output. For example, Boylan (2005) studies U.S. Attorney output without specifying when the cases were initiated. It is also not clear that cases initiated by a previous attorney would be susceptible to the same pressures described above. A finding of this paper is that it is important to distinguish between cases which were received and disposed of within a single presidential term and those which were not. Cases handled entirely within a single presidential term are likely controlled by, or under the guidance of, a single US Attorney.

In the data, I distinguish between these cases by including only convictions and declinations where the referral and disposal occurred within a single presidential term. As a robustness check, I compare this with convictions and declinations that occurred during the presidential term, regardless of when the referral was made. Political influence seems to matter more for cases handled entirely within a single presidential term. Therefore, all other measures will include only controlled cases for brevity. Examining controlled cases alone allows this paper to focus on the output specific to, or under the guidance of, a single U.S. Attorney and a single presidential term.

An issue with the population scaled measures is that they can inaccurately account for state size.²⁰ For example, if politically important states have a larger number of lower-level law enforcement officers per-capita, the U.S. Attorney’s office may receive more case referrals. Thus, they may have more corruption convictions simply because there are more lower level law enforcement officers and have nothing to do with political motivations. Therefore, I look at the percentage of total cases disposed of within each administration that are convictions (Corruption Conviction Percent of Disposals). Thus, the number of disposals is controlled for and the effect cannot be driven by a sheer volume of cases.

This proportional measure has the added benefit of revealing how convictions change in proportion to dismissals. If dismissals and convictions are systematically changing in the same direction, it is important to know if their changes are proportional. For example, if I find that both convictions and declinations are higher in election years, I only know that disposals are higher. However, if I find that convictions are proportionally higher than dismissals in election years, this indicates that a case is more likely going to be disposed of with a conviction than a declination in the final year of a president’s term.

It is important to note that finding results using these proportional measures do not eliminate the possibility of results being driven by a change in true corruption. It is quite possible that politically important states see proportionately more cases being decided with convictions than declinations because there are more truly guilty defendants. This will likely be the case if a “grease the wheels” mindset is instilled in these same areas (Dreher and Gassebner 2013; Dutta and Sobel 2016), encouraging entrepreneurs to engage in unproductive activities (Sobel 2008). However, these proportional measures allow me to test if the level changes come from a change in the number of case referrals alone.

Lastly, if political pressure is driving prosecutorial output, we should see a disproportionate focus on corruption cases. Political importance of the state should have a smaller effect on the convictions for non-corruption related crimes. Regardless of political influence, the volume of non-corruption convictions is much greater than convictions concerning corruption. Thus, if I want to compare the magnitude of effects I need to develop a measure that is comparable across corruption and non-corruption related crimes. To do this I calculate the percentage of disposals resulting in convictions for all cases that do not involve corruption (Other Conviction Percent of Disposals), which is analogous to the proportional corruption measure described above. To test if the difference in coefficients is meaningful, I look at how the percent of convictions for corruption cases relative to non-corruption cases varies across states and time as well (Ratio of Corruption to Other Convictions).

3.2 Political influence

A president is more likely going to be more concerned about his or her administration’s reputation in states that he or she is on the verge of winning in a presidential election. This importance will be amplified the number of electoral votes allocated to that state. Further, if voters within the state are known to be easily swayed, the administration has even more of an incentive to boost their reputation in that state. These are the states where it may be most beneficial to be corrupt. In this section, I develop an index that captures all three aspects of political importance.

Following Garrett and Sobel (2003), I construct a measure of electoral importance that captures two of the above characteristics of politically important states. First, I use a formula to calculate the “closeness” of each election. This formula is defined as follows:

$$Closeness \, of \, Election_{s,t} = \, 1{-}4 \times \left( {Percent \, of \, Votes \, for \, Democrat_{s,t} {-}0.50} \right)^{2}$$

(2.1)

where s indexes the state, t indexes the year, and Percent of Votes for Democrat represents the fraction of votes in the presidential elections that are for the democratic candidate.²¹ Note that the maximum of this formula is 1, indicating that exactly 50% of the votes are for the democratic candidate and the election is extremely close in that state. While, the minimum of the formula is 0, indicating that exactly 100% (or 0%) of the votes were for the democratic candidate and the election was dominated by one party or another in that state.

However, since not all states count the same in presidential elections, Closeness of Election needs to be weighted by electoral votes. Thus, I multiply Eq. (2.1) by the number of electoral votes allocated to the state resulting in the following:

$$Garrett \, \& \, Sobel_{s,t} = Closeness \, of \, Election_{s,t} \times Electoral \, Votes_{s,t}$$

(2.2)

where s, t and Closeness of Election are defined as above and Electoral Votes represents the number of electoral votes allocated to the state at the time of the presidential election. Equation (2.2) is the electoral importance variable used in Garrett and Sobel (2003). However, since this paper is interested in the ability of the executive to influence corruption convictions in order to sway voters, it is likely that regardless of the closeness of the election, some states are much easier to sway than others. Thus, similar to the electoral importance measure used in Young et al. (2001), I multiply Eq. (2.2) by the standard deviation of the percentage of votes for the democratic candidate from 1993 to 2012 for each state.²² Thus, the electoral importance variable used in this paper is given by the following:

$$Political\,\,Importance_{s,t} = Garrett\,\,\&\,\,Sobel_{s,t} \times Standard\,\,Deviation\,\,of\,\,Votes_{s}$$

(2.3)

where s, t, and Garrett and Sobel are defined as above and Standard Deviation of Votes represents the standard deviation of the percentage of votes for the democratic candidate over the entire sample period for each state. Taken together, this measure says that states that have close elections, with more electoral votes, and a population that is easier to sway are more politically important to the president.

Though political importance is clearly central to the goal of both political parties, it may be the case that one party exhibits more of a bias than the other. For example, though Gordon (2009) finds evidence that federal prosecutors under both the Clinton and (George W.) Bush administrations exhibited partisan bias in their choice of defendants to take to trial, he finds that the bias is understated in the Republican Bush administration and overstated in the Democratic Clinton administration. Given this finding, it is reasonable to expect the effect of political importance to be different depending on the political party in power, at least in this sample. Thus, in addition to looking at the effect of political importance alone, I see if this effect differs depending on the political party in power. Specifically, I develop an indicator that equals one if it is Democratic administration and 0 otherwise (Democratic President). I then interact this indicator with the political importance measure. This will give some indication if Democratic administrations exhibit more or less bias overall than Republican administrations. I also include a dummy variable indicating whether the president is in his or her first term (First Term).

If the goal of political influence over corruption cases is to improve the reputation of the current administration we may see that corruption cases in election years are handled differently than in non-election years. Additionally, prosecutors may be motivated to increase corruption convictions nearing the end of their term due to non-political retention incentives. Therefore, in all regressions, I include an indicator variable equal to one for election years and zero otherwise. I also test whether the election year effect varies according to the state’s political characteristics, finding no evidence of such variation.²³ The effect of the election year seems to be constant across states. This suggests that it may simply be non-political retention incentives driving prosecutors to boost convictions during election years, however further analysis would be required to test this assumption.

Additionally, corruption acts may generally increase in election years due to the nature of political business cycles, resulting in increased corruption convictions. However, much of the political business cycle comes in the form of federal aid (e.g., Kriner and Reeves 2015), which is controlled for in this analysis.²⁴ Additionally, a time trend is included to capture any general trends present in the data. Even so, it is important to note that any increase in disposals found during election years may be due to real changes in corrupt acts rather than prosecutorial incentives or other political motivations. One obvious increase in corruption convictions that may drive this result is an increase in election related crimes. Therefore, as a robustness check I estimate the relationship after removing all election related crimes, finding no change in the results.

Lastly, while increasing corruption convictions overall can benefit the current administration by making that political party appear to be tougher on crime, it may be more or less beneficial to do so in states that have a governor that is of the same political party as the president. It may also be more difficult to influence corruption cases in states controlled by the opposing party. To capture this possibility, in this robustness checks I include a variable equal to one if the governor is of the same political party as the president and zero otherwise (Governor Same Party as President). In these specifications, I also include a three way interaction, in addition to all of the relevant two way interactions, between Governor Same Party as President, Political Importance, and Democratic President in the robustness checks as well.

3.3 Basic controls

In addition to the political variables as described above, I include several controls that are common in the literature and have been found to influence corruption conviction rates specifically. First, I include three variables that are intended to capture the federal government’s involvement in each state (DOJ Wage, DOJ Miscellaneous Aid, and Federal Aid). One way the federal government can influence federal corruption convictions is by decreasing (or increasing) the resources the federal investigation agencies or courts have access to in that state (Alt and Lassen 2014). To control for this I include per-capita federal government expenditures on salaries and wages and federal aid allocated to US Department of Justice programs in each state (DOJ Wage; DOJ Miscellaneous Aid).

The federal government can also influence activities in general by providing federal aid to the state. To control for this, I include federal aid per-capita given to each state (Federal Aid). Aid in general has been found to impact corruption in the international arena (e.g., Tavares 2003; Okada and Samreth 2012), so it is likely for this to be an important factor here as well.

In general, a democratic society is likely less corrupt than a nondemocratic one (Treisman 2007). A major reason for this is that politicians in democratic countries can be held accountable for their actions through the voting process. Translating this theory to areas within the US, a state population that is more politically active is also more likely to be more aware of and report corrupt activities which could potentially alter the number of corruption conviction in each state. To control for this I include the percent of the population that votes each presidential election (Voting Percentage).

In addition to these federal government and political variables I also control for standard determinants of corruption averaged over the presidential term. I include percent of the adult population with a bachelor degree or higher (Bachelors), population density (Density), log of total population (Population), and log of personal income per-capita (Income) as in Glaeser and Saks (2006).²⁵ In addition, I include the median age of the population (Median Age) to reflect any differences in states that vary by age that are not captured by the other variables in the model. Lastly, I include regional dummy variables according to the ten Census regions and a linear time trend (Time Trend).

4 Empirical methodology and results

This paper uses a least-squares panel data analysis to estimate the effect political factors have on prosecutorial output. All results presented in Table 4 through Table 10 give the summarized marginal effects of the main variables of interest. All regressions include Election Year, First Term, DOJ Wage, DOJ Miscellaneous Aid, Federal Aid, Voting Percentage, Density, Population, Income, Bachelors Degree, Median Age and Time Trend and regional effects as controls unless otherwise noted. Full results available upon request.

Table 4 Comparing the marginal effect of Political Importance for Democratic Presidential Administrations versus Republican Presidential Administrations

4.1 Democratic versus republican administrations

The first section of results looks at how political influence differs based on the political party in power using the following equation:

$$\begin{aligned} Prosecutorial\,Output_{s,t} = &\, \alpha_{0} + \beta_{0} \,Political\,Importance_{s,t} + \beta_{1} \,Democratic\,President_{t} \\ & + \beta_{2} Importance\,Democratic\,President_{s,t} + \beta_{3} Election\,Year_{t} + \theta X_{s,t} + \, \gamma_{t} + \lambda_{r} + \, \varepsilon_{s,t} \\ \end{aligned}$$

(3.1)

where s, t, and r index the state, year, and region, respectively; Prosecutorial Output is one of the prosecutorial output variables described in the previous section and summarized in Table 1, Political Importance measures the state’s political importance; Democratic President indicates the if the president’s political party is Democratic; Importance × Democratic President is the interaction between Political Importance and Democratic President; Election Year indicates if the year is an election year; θ is a vector of coefficients that correspond to the basic controls included in X as described above and summarized in Table 3; γ represents linear time trend; λ represents Census region dummies; and ε is the error term. Regional dummies are chosen over state fixed effects because the main variable of interest, Political Importance, does not vary greatly over time and is overpowered by state fixed effects (see e.g., Glaeser and Saks 2006).²⁶

The first measures of output examined are corruption convictions and declinations per-capita (Convictions Per-Capita and Declinations Per-Capita) (results given in Table 4). As a simple demonstration, I calculate these results using both controlled and uncontrolled cases. As expected, controlled cases seem to be more effected than uncontrolled cases. The results indicate that election years tend to increase output when considering only controlled cases, with a negative but insignificant effect on uncontrolled cases. Similarly, while the effect of political importance does not change signs, it is only robustly significant for controlled cases.

The results may have the broader implication that prosecutors are more concerned about their own cases than cases inherited from the previous U.S. Attorney. The election year results specifically may indicate that current corruption cases are given priority over older cases in election years, whereas older cases tend to be dealt with earlier on in the term. Similar to the ideas presented in Bandyopadhyay and McCannon (2015a, b), rather than increasing their effort, prosecutors may reallocate their time to signal effectiveness. Prosecutorial effectiveness may be determined by the prosecutors’ ability to finish their own cases, even if this comes at the expense of reducing disposals overall. Alternatively, if convictions accurately represent corrupt acts, it may suggest that election years do increase immediate corrupt acts, but the resulting convictions are only a small portion of the overall convictions. Uncovering the exact cause behind the difference in results is beyond the scope of this paper, as it would require a thorough understanding of how prosecutors value cases they did not initiate. This is an interesting avenue for future research in prosecutorial motivations.

The controlled results are as anticipated. For both declinations and convictions, the election year effect is positive. Additionally, political importance seems to be positively related to corruption convictions and negatively related to corruption dismissals. In all cases of significance, political influence seems to be driven by Democratic administrations. Thus, there tend to be more corruption convictions and less corruption dismissals in politically important states when a Democrat is in office.

As mentioned above, this level effect may be explained away by an increase in case referrals if case referrals also systematically vary across states and time. To control for this, I look at the percentage of case disposals that result in a conviction (Corruption Conviction Percent of Disposals). These results are also presented in Table 4. As the results suggest, while federal prosecutors convict and decline more cases overall in election years, it seems that they are increasing convictions by a larger amount. In addition, it seems that politically important states experience a higher percentage of convictions and a lower percentage of declinations.

However, this does not necessarily mean that the focus is more on corruption cases relative to other cases. It could simply mean that prosecutors in general tend to increase their output nearing the end of their term. Additionally, it could mean that prosecutors in politically important states face pressure to increase convictions and reduce dismissals in general. There may be nothing specific about corruption cases.

To test whether the focus tends to be more or less on corruption cases, I examine how the election year and state’s level of importance affect the percentage of cases disposed of that resulted in convictions for non-corruption crimes (Other Convictions Percent of Disposals).²⁷ I then look at the ratio of the percentage of corruption cases resulting in conviction to the percentage of non-corruption cases resulting in conviction (Ratio of Corruption to Other Convictions). Both sets of results are presented in Table 4. This will give some information on the relative focus of prosecutors in politically important states.

Election years are associated with an increase in the percentage of cases resulting in convictions for non-corruption related crimes as well. This effect is slightly smaller in magnitude than the election year effect on corruption related crimes. However, as indicated in the last row of the table, the difference in magnitude is not statistically significant.

The percentage of the case disposals resulting in conviction for non-corruption related crimes, however does not seem to be affected by the state’s political importance. While the signs of the coefficients match the results using the analogous corruption related measure, the coefficients are no longer statistically insignificant. Additionally, the differences in the coefficients are statistically significant. The percentage of corruption cases resulting in conviction is higher than the percentage of non-corruption cases resulting in conviction in politically important states. Again, the effects of political importance seem stronger in Democratic administrations.

4.2 The importance of the political party of the governor

The second section of results looks at how political influence differs based on the political party in power as before, but now also considers how this effect changes based on the political party of the state’s governor. Thus, I add in two additional two-way interactions (Political Importance interacted with Governor Same Party as President and Democratic President with Governor Same Party as President) and one three-way interaction where Political Importance is interacted with both Governor Same Party as President and Democratic President. Table 5 presents the summarized marginal effects for each output variable as before, however the estimates are now broken down by political party of the governor as well.

Table 5 Comparing the marginal effect of Political Importance for Democratic Presidential Administrations versus Republican Presidential Administrations with state governors of the same party as the president and of the opposing party

Similar to the results presented in Table 4, corruption convictions are higher in politically important states, while corruption declinations are lower, though this latter relationship is again statistically insignificant. Additionally, corruption cases are more likely to be disposed of with a conviction in politically important states. Similar to the results in Table 4, we see that these effects are stronger in Democratic administrations.

Interestingly, the effect of political importance on the level of convictions under democratic administration seems to be constant across governor parties. Suggesting that the political party of the governor matters little in terms of the sheer number of convictions. However, in all cases of statistical significance the effects of political importance become stronger when the governor is of the Democratic party. The response is the largest in magnitude when the president and the governor are both Democrats. This may suggest that it is easier to increase convictions in states with a party in power that is sympathetic to the current administration. Thus, the administration must not be too concerned with corruption cases hurting their political image. This is likely because most defendants are not partisan identifiable. The remainder of the results will include all interactions included in Table 5.

4.3 Federal versus state, local, and other corruption

The third section of results looks at how the effect differs based on the type of corruption considered. As mentioned above, the broad program category of public, or official, corruption includes seven subcategories, four categories pertaining to federal corruption and three separate categories for state, local, and other corruption. Since federal corruption is more likely associated with the federal government it seems likely that federal prosecutors will place the most effort into increasing federal corruption convictions. Thus, we should see a stronger effect of political importance on federal corruption cases than cases pertaining to the remaining three categories of state, local, and other corruption. Tables 6 and  7 show that this is exactly what happens. Political importance increases corruption convictions, and likewise decreases corruption declinations, for federal corruption but has no effect on the other types of corruption.

Table 6 Comparing the marginal effect of Political Importance for Democratic Presidential Administrations versus Republican Presidential Administrations; for corruption falling in the program categories of federal only

Table 7 Comparing the marginal effect of Political Importance for Democratic Presidential Administrations versus Republican Presidential Administrations for corruption falling in the program categories of state, local, and other only

4.4 Robustness checks

4.4.1 Election crimes

One major concern with the results summarized above is the possibility that politically important states experience more election related crimes and thus tend to experience more corruption convictions. More generally, all states may experience a boost in corruption convictions during election years if there is an increase in election related crimes. In both cases, the increase in corruption convictions would simply indicate that the federal prosecutor is efficient at doing his or her job, not that there is any political bias.

Therefore, using the Justice Department’s Federal Prosecution of Election Offenses (Donsanto and Simmons 2007) guide and the lead charge of the crime codes given in the TRAC data, I exclude crimes considered to be either election crimes or campaign financing crimes (results in Table 8).²⁸ As can be seen in the table, there are no significant differences between results in Table 5 and Table 8, indicating that election crimes are not the driving force behind the relationships found above.

Table 8 Comparing the marginal effect of Political Importance for Democratic Presidential Administrations versus Republican Presidential Administrations excluding crimes involving federal elections and campaign financing

4.4.2 State fixed effects

Though it is unlikely that reverse causality is an issue as political importance of a state is fairly constant overtime, there may be a concern of omitted variables driving results. Traditionally, one way to address this issue is to use state fixed effects, however, state fixed effects overpower political importance since neither vary over time. To get around this, I first estimate the results using Political Importance as before with fixed effects and show that much of the statistical significance disappears as expected (Table 9). I then exclude the number of electoral votes and easiness of swaying voters from the importance measure and include state fixed effects (Table 10). In this latter case, political importance now varies over time making regressions with fixed effects more meaningful.

Table 9 Comparing the marginal effect of Political Importance for Democratic Presidential Administrations versus Republican Presidential Administrations with state fixed-effects included

Table 10 Comparing the marginal effect of Political Importance (excluding electoral votes) for Democratic Presidential Administrations versus Republican Presidential Administrations with state fixed-effects included

As shown in Table 9, all statistical significance of political importance goes away for corruption convictions and declinations per-capita when including state-fixed effects in place of regional fixed effects. In addition, political importance is no longer statistically related to the percentage of disposed cases involving corruption crimes that resulted in convictions or declinations.

I re-estimate the results using the Political Importance measure where electoral importance and easiness of swaying voters are excluded from the measure. When doing so, I see that the results become more similar to the results summarize din Table 5 (see Table 10). The difference is that now we have less overall significance in Table 10 than we did in the main results presented in Table 5. This is likely due to the fact that all three components of political importance matter, not just the closeness of the election.

5 Conclusions

Political corruption, and the monitoring of it, are serious concerns as both are reflective of the quality of the current administration. To study political corruption in the US, researchers often use corruption conviction rates as measures of state-level corruption. The issue with examining US state-level corruption, though, is that there are important political differences across states that influence corruption convictions, but are often ignored.

States that are important to win in a presidential election likely see more corruption convictions for two reasons. First, the incentive to be corrupt is likely higher in these states. Assuming corruption convictions represent true corruption, this will result in higher corruption convictions. Second, the enforcement of corruption via federal corruption convictions is susceptible to political influence. Given the highly-publicized nature of corrupt crimes, the current administration may have an incentive to influence the handling of corruption cases in battleground states.

Additionally, there are reasons to expect corruption convictions to vary across time even when the true amount of corruption is held constant. Election years often represent the end of a term for a U.S. Attorney. Therefore, we may see an increase in corruption convictions during election years as attorney’s attempt to increase their employment prospects. Alternatively, we may see an increase in corruption convictions in election years because the current administration wants to appear strict on crime. If the attorney aspires to move up in the political world, they will adhere to the administration’s wishes. If not, the administration may still alter conviction timing indirectly by shifting the priority of case types.

This paper uses individual case file data through the Transactional Records Access Clearinghouse (TRAC) data warehouse (TRACFED) to examine the effect political factors have on federal corruption convictions and declinations, with a particular focus on the political importance of the state in which the case is located. In this paper, a state is considered to be politically important if it is a focus of the next presidential election.

Overall, the evidence of this paper indicates that federal prosecutors tend to convict more individuals in politically important states. In addition, convictions account for a higher percentage of case disposals in these same states. Thus, the increase in convictions is not driven exclusively by an increase in case referrals. These results are robust to a variety of specifications, including the elimination of election related crimes, and seem to be more significant in democratic administrations. The results also seem to be more significant under both administrations when the state governor is of the Democratic party. Lastly, it seems that the political factors only affect corruption crimes labeled as “federal”; they do not seem to matter for state or local corruption crimes.

While it is impossible to distinguish between true increases in corruption and corruption convictions, this paper highlights the significance of political importance and election year effects. To this end, future research should explore alternative measures of corruption across the US States. Using these alternative measures it would be interesting to see if the results found here are something specific to corruption convictions, or to corruption in general.

Additionally, this paper highlights several avenues of future research concerning prosecutorial behavior in general. The election year effect is suggestive of prosecutors reacting to employment incentives. However, this paper is unable to distinguish between these specific types of incentives and general administrative case manipulation. Further, even if prosecutorial motivations drives this result, it is unclear what mechanism prosecutors are using to respond to these incentives. At the state-level, there is disagreement between effort driven output increases (Gordon and Huber 2002) and signaling mechanisms (Bandyopadhyay and McCannon 2014; 2015a). Understanding how prosecutors respond to incentives is important as they represent the chief law enforcement officers of their district.