Last August, U.S. Congressman Duncan Hunter was indicted for misuse of campaign funds for personal benefit. The Justice Department alleges that Hunter conspired with his wife, whom he appointed campaign manager, to steal from his campaign to support their lavish lifestyle: the campaign spent $15,000 on airline tickets and hotel rooms for Hunter’s children and relatives, a $14,000 Thanksgiving trip to Italy, and for other expenses like $700 for seven adult and five children’s tickets to see “How the Grinch Stole Christmas.”
Though Representative Hunter’s conduct is only now being investigated, the allegations of improper spending go back to 2009, and many of the expenses now under scrutiny were detailed in his campaign’s filings with the Federal Election Commission (FEC). FEC filings are public records, readily available and searchable (via simple keyword searches on the FEC’s webpage) to anyone interested in looking. For example, Representative Hunter is a “vaping” enthusiast (even smoking his e-cigarette in Congress). Using the FEC’s webpage and a simple search for the words “cigar,” “smoke,” and “tobacco,” I found that Representative Hunter’s 2015-16 campaign expenditures include hundreds of dollars of spending at a cigar lounge, smoke shop, and tobacco company in his home district. Similar search results through the FEC website show all sorts of eyebrow-raising transactions.
So why weren’t the problems detected earlier? The problem, in cases like this, is not that the FEC doesn’t have enough information to identify suspicious activity—it’s that it has too much information. The FEC has massive amounts of data, making the detection of fraud a needle-in-the-haystack problem. The FEC relies largely on complaints and referrals to guide its enforcement process, with the result that enforcement remains anemic. In 2017, for example, the FEC levied administrative fines in 215 matters totaling under $2 million, despite having data on 23.4 million line-item disbursements and 34.5 million individual contributions, not even counting electioneering communication transactions and the massive data on political action committees (PACs).
Waiting for referrals, or screening data by hand, is not an effective way for the FEC’s roughly 300 employees to detect corruption or fraud in campaign finance. There are no silver-bullet solutions; fraud detection is a fundamentally difficult, especially when fraudsters take steps to cover their tracks. But there are some steps the FEC can take to better monitor fraudulent expenditures to identify suspicious cases early on:
- Automated flagging of suspicious transactions. The FEC can employ a blacklist or automatic flagging system for certain types of vendors and/or transactions. When a campaign filing reports that a candidate like Representative Hunter is spending up to $1,000 at a time at a cigar lounge, this should trigger an audit. The creation of something as simple as a keyword flag could vastly expedite the narrowing process for which candidates need additional monitoring. However, the FEC should keep its screening algorithms confidential, as the public disclosure of the keywords that trigger additional scrutiny could make it easier to circumvent the system through false reporting.
- Comparison between candidates. Many corrupt or fraudulent transactions can’t be identified simply through the identity of the vendor. For example, candidates regularly spend money on food and drinks for donor events, so the same reported transaction (a large expenditure for catering) might be a legitimate campaign expense, but could also be an illegitimate use of campaign funds for personal benefit. The FEC could, however, use comparisons between candidates to more accurately identify potentially aberrant behavior. By matching on geography, party, incumbency, and amounts of spending, a simple algorithm could pick out circumstances when certain candidates behave very differently from the norm. Unusual spending does not necessarily imply illegal spending, but it does suggest the need for additional scrutiny, so an algorithm that incorporates comparisons among similar candidates could be very effective in targeting audits and maximizing the effectiveness of the FEC’s limited oversight capacity.
- Bounty program. Government agencies are notorious for their limited computational expertise. But as FEC data are already public, the FEC could implement a bounty program, paying individuals or companies for finding fraud among candidates, individual contributors, or PACs, when their information leads to an administrative fine. Bounty programs work well because they only pay out in circumstances where bad behavior is uncovered. This type of data monitoring is already being undertaken for Medicare, where data privacy is a much greater concern. It is unclear what techniques private companies would use on the FEC data, but fraud detection has become a well-developed field in the era of e-commerce, and a paid program would reward those with the best ability to detect fraud. To avoid over-enforcement or random enforcement, private entities participating in the bounty program should be required to provide specific information, for example by identifying specific suspicious transactions or patterns. Furthermore, to avoid either random or politically-targeted reporting, participants would have to prove that they perform sufficiently well (perhaps by debarring reporting entities with too many false positives in their track record). The FEC already has the capacity to levy administrative fines, and a privatized bounty program would vastly expand its capacity for screening, while the administrative process such as fines and appeals can stay firmly within the bureau’s control.
The case of Duncan Hunter is not just a case of brazen corruption by one congressman. Rather, it underlines the failings of the FEC to make use of the data it demands and collects, even when a cursory examination would have raised red flags. By applying simple computational techniques, or outsourcing the problem to those capable of using more complicated ones, the FEC can detect and deter corruption among candidates.
This was super interesting to read! I did not know that US had such a detailed mechanism of reporting campaign costs (not so common, by they way!). I have one question. It seems that the politicians you are mentioning felt very confident in disclosing the dubious costs you are listing here. Does this mean they are so sure that nobody will ever look at these reports, they really don’t care? Or is there more to it – and they think they could defend those costs in many cases?
Hi Jetson. Completely agree with Ruta. This was a fascinating read. I have a follow-on question to Ruta’s (and some thoughts), regarding the legal framework in which the FEC operates. I don’t know anything about election law, but is it possible that the politicians you mention don’t care about disclosing these dubious expenditures because they aren’t actually illegal? Or to put it another way, were Duncan Hunter’s campaign expenditures at the cigar bars necessarily illegal? I suspect the answer depends on what the explanation is for those expenditures. In that case, if such expenditures can be justified or explained away, I wonder if there isn’t more of a role for advocacy organizations to play here. A transparency org with a decent algorithm might be able to find these types of expenses that, even if not strictly illegal, certainly look fishy, at which point naming and shaming could be an effective tool at curbing these kinds of practices. Seems like opposition research teams would be able to make effective use of this database as well.
Thank you for the post! This is a fascinating topic. So much of the time the answer to campaign finance concerns is a call for more disclosure, but this post effectively shows how the paralysis that can result from too much of a good thing.
A couple of questions: first, how would you ensure the bounty program wasn’t turned into a partisan instrument, or do we just assume that both parties would have an equal shot at targeting candidates on the other side, neutralizing partisan concerns? Second, would the FEC actually be allowed to keep the screening algorithms confidential, and if so, should they? I completely get the benefits (preventing circumvention), but it seems that allowing those algorithms to operate unchecked could have drawbacks, too. Algorithms in this case are not so contentious as those proposed for use in policing, for example, but there is still some pretty considerable (and probably pretty healthy) public distrust of their use.
Finally, I second Jason’s point that there is a purpose for these measures beyond FEC enforcement – even where these expenses could be explained away (hard to imagine for the cigars, but fine) having advocacy orgs use the data to name and shame could effectively target transactions that, while not illegal, are at least unpalatable. This becomes all the more true if it were applied at the state level, where in a few states, personal use of campaign funds is still allowed.
I second your point, Natalie! I was wondering whether the current bounty programs are effective across the board, or if you tend to see partisanship at play. Further, once that information is uncovered what, if any, disclosure requirements would be in place for the FEC?
Jetson, what an interesting topic—always appreciate reading about your discoveries with datasets! Inspired by your post and the FEC link you provided, I ran a few searches through the database on my own, and in addition to being a needle-in-the-haystack problem because of the sheer amount of data available, I also think that part of the problem is a lack of mechanism to read between the lines of what some of the disbursement descriptions imply. I’m not sure how even an algorithm can detect whether what seems innocent—“meeting food and beverage,” for instance—is actually personal spending. And I worry that an algorithm that can pick up on keywords will cause disbursement descriptions to become even more vague, and will caution spenders to use merchants with names that do not raise red flags.
Additionally, when standalone transactions seem like they may be within the norm of what is considered appropriate, especially in their amounts, I find it difficult to imagine how an algorithm would piece together a series of unsuspicious transactions to develop a narrative of inappropriate spending.
Jetson, again, a really interesting article here. One possible (rather rudimentary–but common-sense) solution that I thought about to avoid as many instances of the “false positive” problem is as follows: for a given expenditure (e.g., food/beverage) could the reporting requirements be tweaked so that the candidate must detail how many people for whom the food/beverages (or in this case, cigars) were provided for? In other words, it seems that many of the largest food/beverage transactions would, naturally, be for larger groups of people. But if it was required that the number of people was reported for each expenditure, you could isolate expenditures by their cost per person–which probably gets at the “lavishness” issue more easily.
Great post! It was very interesting to learn that this data is publically available. It really seems like an untapped resource. On Signa and Natalie’s point, I am not so sure that letting opposition researchers use the bounty system would be such a bad idea. Why not let politicians constrain each other? They are the ones with the most incentive, time, and money to do so and will likely be very effective. Of course, there are concerns around parties reporting their opposition just so that they can say that their opponent is under an FEC investigation. Your ideas around debarring entities that report too many false positives could be one way to curb that. Another way to ensure accurate reporting would be to require reporters to point to specific evidence from the data and their own research to support their findings. While there are some risks, I think they can be mitigated and the added pressure of being scrutinized by your well-resourced opponent could actually incentivize campaigns to be less corrupt.