Insight into tPP’s Coding Process through Case Study: Dylann Roof

This continues our series of student reflections and analysis authored by our research team.

Equally important to the research the Prosecution Project (tPP) yields is the process by which our researchers input cases to the database to ensure the reliability and validity of tPP’s findings. Student members of tPP code for a total of 40 variables per case in addition to other descriptive information such as the specific charges, a narrative of the case, and so on. These variables account for information such as defendant demographics, circumstances of the crime, court results, and more. For better transparency and to promote a better understanding of tPP’s methods, this weeks blog update breaks down the coding process of these 40 variable each case undergoes twice, with two researchers individually coding then comparing their results for accuracy.

To best elucidate this process, take the example of the 2015 Charleston church shooting. Dylann Roof was convicted of 33 counts of federal hate crimes for the murders of nine African American churchgoers and the injuring of another, and after a brief manhunt, was arrested the next morning (DOJ, 2017).

The first variable tPP codes for is the date, which may be of the crime or arrest for older cases with fewer digital records, however tPP’s preference is for the date of the most recent indictment. So for Roof’s case, the date is coded as 06/22/2015, when the federal grand jury returned an indictment, and the second variable, date descriptor, is marked as ‘indictment’ (DOJ, 2015). The next four variables consist of the defendant’s full name, and then separately their first name, family name, and any aliases in order to make searching for defendants in the database easier and to prevent any duplication of cases under aliases. Roof’s full name is Dylan Storm Roof, with no known aliases. Next is co-defendant; given Roof acted alone, this is simply coded as ‘no’.

Next is the reason for inclusion. One or more of ‘obvious socio-political aims’, ‘serves to support organized political violence’, or ‘state speech act’ must be cited. For more information on the inclusion process, see Taylor Maddox’s update , but for these purposes, Roof’s case is coded as having obvious socio-political aims due to the racial motivation of the murders and self expressed desire to start a race war, but lacks any state speech describing it as terroristic or any affiliation to organized violence (DOJ, 2015).

The next set of variables cover the name of the court case, jurisdiction, and the country, city, and state in which the attack took place. While Roof faced both federal and state charges, these are individually coded, so Roof appears twice in tPP’s database, once per indictment. Focusing on the federal charges in this example, these are ‘United States of America v. Dylann Storm Roof’, ‘Federal’, ‘United States’, ‘Charleston’, and ‘South Carolina’.

The next variables concern the target of the attack. ‘People versus property’ describes the intended target, which may be either of these, both (e.g. the September 11th attacks), have no direct target (e.g. charges of providing material or financial support to terrorist organizations), or be unknown or undeveloped (e.g. defendants stopped early by law enforcement). Roof’s target is coded as ‘people’. Next, ‘target: what?’ describes the physical site of attack, and ‘target: why?’ describes the motivation. For Roof, these are ‘religious site’ because the shooting took place in a church, and ‘racial’.

Three variables tPP codes for describe the defendant’s relevant affiliations. ‘Ideological affiliation’ includes such options as jihadist, various national separatist movements, and numerous rightist and leftist ideologies. ‘Affiliation with a Foreign Terrorist Organization’ refers to association with organizations specifically designated by the U.S. State Department, and is answered simply yes or no. Finally, ‘group affiliation’ specifies the FTO or domestic organization the defendant may belong to. In this case, Dylann Roof is coded as ‘white supremacist/neo-Nazi’, has no FTO affiliation, is coded as having ‘no affiliation’ with any group.

‘Previous similar crime’ is used to denote defendants with past convictions. “Similar” in this sense may mean, though is not limited to, a past crime utilizing the same of the next two variables: tactic, which is the method of the crime, and tactic group, which is simply more broad and allows for wider categorization of incidents. It could also be a similar ideological motivation, target, etc. Roof’s tactic was ‘shooting’, making the tactic group ‘firearms’. He had no previous conviction for a similar crime. Continuing with variables describing the attack itself , there is number killed, which was nine, and number injured, which was one (DOJ, 2015).

The next variables concern the court proceedings: court decision (often the jury verdict, or denotes a plea bargain or dropped charges), length of jail sentence, life sentence, death sentence, additional sentencing, and plea. Length of jail sentence is simply the number of months sentenced, and life and death sentences are usually coded as either 0 for none or 1 if applicable but may increase to reflect multiple life or death sentences. Roof, facing the death penalty, plead not guilty in his federal case, making the plea variable ‘not guilty’ (CBS, 2015). The court decision however is ‘guilty’ due to the jury’s verdict, and while both length of jail sentence and life sentence are 0, death sentence is coded as 1 (DOJ, 2017). The additional sentencing variable covers such factors as supervised release or various enhancements, such as the hate crime addition in Roof’s case.

The Prosecution Project also codes for if the particular case involved an informant (e.g. FBI confidential human sources) or not , and if the attack was ‘carried through or stopped’, and if applicable, stopped by whom. Roof’s case involved no informant, and is coded as carried through due to his apparent success in killing a number of people, only being arrested the next day.

Finally, eight variables code for the defendant’s demographic information. Most straightforward are age and sex at the time of the offense, which for Roof are 21 and male. His ethnic group is coded as white/caucasian, and his religion as Christian. Roof’s citizenship is coded as American born citizen. Veteran status seeks to account for either civilian or military history and type of discharge or current service status at the time of the offense. More specifically the next variable, combat veteran status, is coded as either yes or no depending on if the defendant has been deployed to a combat zone. Dylann Roof is coded as ‘civilian’ for the former and with ‘no’ under the latter.

Lastly and slightly more complicated is the demographic variable ‘Other status’, which very simply put seeks to categorize the defendant as someone a typical jury would see as ‘American’ or not, which can be influenced by their name, ethnicity, citizenship, and religion. For more information on tPP’s use of Othering, see Athena Chapekis’ blog post. Given his demographic information, Roof is coded as non-othered.

Finally, having all of these variables coded for and having both researchers in agreement on each, the resources this information was drawn from are added to the dropbox and Dylann Roof’s case is ready for analysis!

– Kayla Groneck


Department of Justice, U.S. Attorney’s Office, District of South Carolina. (2017, January 10). Federal Jury Sentences Dylann Storm Roof to Death [Press release]. Retrieved September 24, 2018, from

Department of Justice, Office of Public Affairs. (2015, July 22). Attorney General Lynch Statement Following the Federal Grand Jury Indictment Against Dylann Storm Roof [Press release]. Retrieved September 24, 2018, from

CBS. (2015, July 31). Not guilty plea in federal court for accused Charleston shooter. Retrieved September 24, 2018, from

Policy Spotlight

This continues our series of student reflections and analysis authored by our research team.

On September 18, 2018, the State Department released its new ceiling of 30,000 refugee admissions for the upcoming fiscal year (Malley & Pomper, 2018). This is the lowest level in the history of the 40-year old refugee resettlement agency, with the average acceptance rate at around 80,000. While the Trump administration did not directly cite national security concerns as a reason for the reduced ceiling, the new refugee admission policy is merely another reiteration of a nearly two-decades long trend in which immigrants, refugees, and asylum seekers were denied entry to the United States, many on the basis of national security.

Prior to 2001, it appears that the issues of immigration and national security were unlinked. Major domestic terrorist attacks such as the World Trade Center Bombings in 1993 and the Oklahoma City Bombing in 1995 led to changes in the internal organization of domestic intelligence agencies, passage of the Anti-Terrorism and Effective Death Penalty Act of 1996 (A.E.D.P.A.), and passage of the Omnibus Counterterrorism Act of 1995 to outlaw fundraising for Foreign Terrorist Organizations (FTOs). Interestingly, prior to the attacks of 9/11, no Congressional push for decreased immigration or refugee admissions cited a major terror attack as a basis for the bill. In fact, following the Nairobi and Dar es Salaam U.S. Embassy Bombings in 1998, President Clinton signed a budget bill that supported pro-immigration measures, such as allowing 50,000 Haitian refugees to stay in the country and increasing the number of visas granted to foreign professionals (Ojito, 1998).

Following the terror attacks on September 1st, 2001, an implicit connection seems to be made between national security concerns and immigration/refugee admissions. This makes logical sense as the 9/11 hijackers did enter the United States using various types of visas (Friedman, 2017). However, following the changes to security protocols in the wake of the attack, only one individual out of all domestic jihadist attackers would have needed a visa to enter the United States at the time of their attack (Peçanha & Lai, 2015). Additionally, the New York Times released a report in 2015 stating that the threat of foreign terrorists coming through the refugee program was lower than even the risks of routine travel (Peçanha & Lai, 2015). The movement away from foreign terrorists (meaning individuals who are not born in the U.S. or naturalized citizens) and towards domestically radicalized individuals is in direct contrast to an increase in the number of Congressional bills seeking to limit refugee and immigration admissions while citing specific domestic terrorist attacks, such as the call for halts on Syrian refugees following the San Bernardino Attack of 2015 and the Pulse Nightclub Attack of 2016 (Peçanha & Lai, 2015). Additionally, executive orders like President Trump’s travel ban, unveiled in 2017, aimed to reduce immigration and refugee admissions from countries that were believed to pose a threat to national security and/or that were affiliated with terrorist activity.

While research examining the relationship between immigration/refugee policies and national security concerns has been conducted by institutions such as New America, the Atlantic, and Only Through US (Bergen, Ford, Sims, & Sterman, n.d.; Friedman, 2017), the Prosecution Project can help to shed light on this issue by examining domestic terrorism across all ideologies. The majority of these reports exclusively focus on jihadist terrorism in the wake of 9/11, without comprehensively including threats from other ideologies and organizations, such as right wing extremists, white supremacists, and eco-terrorists. If the goal of national-security-related policies is to prevent further attacks on American soil, research to support these policies must examine all manners of domestic threats. Furthermore, research of this kind can help determine whether the connection between national security and immigration/refugee policy is still relevant at this moment in time.


Sarah Moore is a senior team member of the Prosecution Project and a former intern at Only Through US.  


Friedman, U. (2017, January 30). Where America’s Terrorists Actually Come From. Retrieved September 20, 2018, from

Malley, R., & Pomper, S. (2018, September 18). Trump’s Refugee Fiasco. Retrieved September 20, 2018, from

Ojito, M. (1998, November 1). THE 1998 CAMPAIGN: IMMIGRANTS; Once Divisive, Immigration Now a Muted Issue. The New York Times. Retrieved from

Peçanha, S., & Lai, K. K. R. (2015, November 25). The Origins of Jihadist-Inspired Attackers in the U.S. The New York Times. Retrieved from

Another update from tPP!

This continues our series of student reflections and analysis authored by our research team. 

Last week we had another busy meeting in order to get all of our ducks in a row before we delve into analysis. We addressed some issues with coding, discussed ways to publicize our efforts, and talked about things that we as individuals can focus on within the data.

One thing we spent a large chunk of time on was addressing some issues that have come to light concerning the coding variables – both the variables themselves and the options (i.e. coding values). For one, as the project has gone on the variables that we are coding have changed; some have been added while others have been adjusted for the values that we’ve found. This has caused a discrepancy in how cases are being coded. Some of the older cases that were coded and verified at the beginning of the project are now missing variables, such as the “other” variable (see Athena’s blog post)! These cases in the database as complete, when they are in fact currently missing information.

Similarly, there was a slight lack of communication amongst the team throughout the coding process causing other variables to remain uncoded or incorrectly coded. “Charges,” for example, proved too difficult to fit into a pre-set list of options, so it was decided that instead the exact charges for each case should be copied into the database.

This same challenges with team-wide communication also lead into the discussion that was held on the variable “other.” The “other” variable was added later on in the project, so it was a two-faced problem. Part one, a good portion of cases simply never were coded for the variable. Part two, there was a less-than-clear definition of what “other” meant when it was introduced.

We attempted to address this as a team and ran into the issue that calling someone an “other” is inherently something that is different person to person. The original idea was to capture the mindset of your typical American jury, but we had trouble even coming to a consensus on what that looked like. As a team, we also struggled with the ethics of trying to take a subjective determination such as “other” and turn it into a variable. After all, who are we to decide what someone who doesn’t fit in looks like? We spent a good portion of the class debating what the function of the variable is, and to what extent we could make assumptions about other people’s mindsets when coding. If we as a team can’t find a middle ground on how to code for “other,” then why are we trying to? And with the number of assumptions that we are making in order to code for the variable, is it even worth coding for? In the end, we decided that it does have merit to help see generalized data trends, but it was a long discussion that led us there.

To wrap up, we tried “live coding” as a class for the first time this week! In an attempt to show a more real-time example of how the coding process is completed, we coded a few cases as a class with the hopes to upload a video of our team work soon!

– Doty

On coding for the variable of ‘othered’

This continues our series of student reflections and analysis authored by our research team.

One of the most difficult topics as far as coding that has come up during the course of our project is one that is fundamental to the question we are attempting to answer: What does it mean for a person to be “othered”?

The Prosecution Project is a project whose foundational goal is to compile prosecutions of acts of terrorism, political violence and
extremism from all ideological backgrounds to compare how different people/groups are treated in the United States justice system. An notable pattern prior to the commencement of our research was that white, American-born, Christians are not frequently charged with “terrorism. ” We sought not only to highlight this trend, but to look at the details of sentencing and the patterns that may exist dependent on a perpetrators ideology, tactic, target, severity of attack, and demographic details of the person themselves.

Initially, we were planning to look at some of the demographic differences
by coding for ethnicity, and using that as our variable of interest (tPP does code for a defendant’s age, sex, ethnicity, religion, veteran status, citizenship, as well as external factors such as ideology, group affiliation, previous arrests).  What we soon realized is that ethnicity is exceedingly nuanced and almost impossible to determine through secondary literature, court documents, and/or photographs. We wondered if instead religion would better capture the differences we were interested in, but soon decided that it didn’t provide us with as much information about the perpetrators as we though it would.

Instead, we had the idea to create a variable that took into account all
of the variables we coded for that contain traits usually used to define a
person who is understood as a “true American”: ethnicity, religion, and citizenship. Essentially, we decided that anyone who did not fit the cookie-cutter image of a white, American-born Christian has the potential to be
considered “other” in the eyes of not only society, but the legal system,
as a jury in a court of law is meant to be reflective and representative
of the society.

The team has had a series of pretty intense discussions about the ethics of this method for determining what we refer to as “othered status.” One concern raised was that a white, Christian-born, American-born man does not truly experience what it is to be othered in society, even if he, for example,
converts to Islam. He still maintains his white American-born identity. To
allow the team to work with an example that this situation may apply to,
we pulled up Christopher Cornell’s case in our database.

Chris Cornell was an American-born white man who converted to Islam and
insisted to be called by his Muslim name, even though all news articles
and legal proceedings continued with his given name. We applied the
question: would your average Joe look at this man and say “yep, that’s an
American.” The answer, almost unanimously, was no.

As problematic as it is, the fact that Cornell grew out his beard and
presented himself as Muslim is probably enough in the United States for
the average person to not consider him a thoroughbred American boy. It is through this understanding of American society that we determined our criteria that must be met to code the “othered” variable as “yes.” From our codebook:

€€‘Other Status’ = Using the decision tree below mark the cell as either:  

  1. OTHERED: The defendant is marked as othered if they meet any of the following criteria
    1. Does the defendant have a name not readily understood as European?
    2. Is the defendant Muslim or a Muslim convert?
    3. Is the defendant an immigrant from a non-Western/European country?
    4. Is the defendant non-white racially as an ‘average person’ would read them (i.e. not passing as white)?
    1. The defendant is marked as white, non-foreign born, Judeo-Christian and a non-jihadist (i.e. can pass as a white, American-born, Christian/Jew)

The fundamental question is: Would your average, mid- to low-educated person in the United States look at this person and see them as an “American”? Using this question, we have decided, most accurately determines whether a person has been othered in a way that has the potential to affect their treatment in the United States Criminal Justice System.

Athena Chapekis is a senior sociology major at Miami University and senior team member and data analyst at the Prosecution Project

Rhetoric of Terrorism

This continues our series of student reflections and analysis authored by our research team.

Future posts will focus on gender, religion, race and ethnicity, the role played by federal informants in sentencing, denaturalization as a response to terrorism,  deciphering the distinction between hate crimes and other forms of bias-motivated violence, and a host of other topics. Stay tuned!

This week, I researched the 2016 Malheur Wildlife Refuge Occupation outside Burns, Oregon. A large group of armed militants took control over the federally-owned land through intimidation and threats, and preceded to occupy the refuge for over a month (Levin, 2016). Though the occupiers did not injure or kill anyone, this was not a victimless crime; federal employees at the refuge were unable to do their jobs for a month, local schools were shut down for a week, and the expansive police response cost Oregonian taxpayers a conservative estimate of $3 million (Levin, 2016).

I read twenty-two news articles and two court documents while researching this case; not once was the word “terrorist” used to describe the occupiers. 

The FBI defines a “federal crime of terrorism” as “an offense that is: (i) calculated to influence or affect the conduct of government by intimidation or coercion or to retaliate against government conduct; and (ii) is a violation of federal statute” (Federal, 2008). The Malheur occupiers inhibited the federal government’s ability to operate at the refuge for over a month through armed intimidation, which constitutes as terrorism according to the FBI’s definition. So why were the occupiers never described as terrorists?

I do not have a conclusive answer, but I suspect the word “terrorist” was never used because all but one of the twenty-six arrested occupiers were white, Christian, American-born citizens. They were anti-government ideologues, not “jihadists”. The perpetrators were described as “activists,” “occupiers,” and “militiamen” throughout the articles surveyed while researching the case. None of these words really evoke  the sense of fear and havoc that the occupiers instilled in the Burns, Oregon community. None of these words reflect the fact that the occupation falls under the FBI’s definition of a federal crime of terrorism.

The tempered rhetoric used to describe the Malheur occupiers reflects a phenomenon I have come across during my research where white, anti-government perpetrators are rarely described as terrorists. They are portrayed in the media as patriots who love their country, but fear their government. At worst, they are characterized as erratic, mentally unstable gun-lovers. But rarely terrorists.

My involvement with the Prosecution Project has taught me that language and rhetoric have the power to shape how our team and the public understand terrorism and political violence. Rhetoric shapes cultural attitudes, and cultural attitudes shape can sometimes shape how crimes of political violence are charged and sentenced. I worry that if we use terms like “activist” and “lone wolf” to describe violent perpetrators who terrorize Americans and try to hold the American government hostage, sentencings for crimes like the Malheur occupation will be reduced and the rule of law will become diluted.

As I move forward with the project, I hope to move beyond anecdotal evidence and use statistical analysis to better understand the correlation between ideology, race, religion, and sentencing outcomes.

– Nikki Gundimeda

Works Cited

Federal Bureau of Investigation. (2008, December 16). Domestic Investigations and Operations Guide. Retrieved from

Levin, S. (2016, February 24). How much did the Oregon standoff cost taxpayers? Millions, says early estimates. The Guardian. Retrieved from

Greetings from tPP!

Beginning today, tPP will feature weekly blog posts from our student research team on a variety of topics. Stay tuned!

Greetings from tPP!

Hello, and welcome to the first blog update of the semester. We’ve only met once thus far as a class; we have one three-hour class every Thursday afternoon as well as an optional two-hour meeting on Tuesday for those who are available. The class period Thursday demonstrated how helpful having a full class period together is going to be. With just this first meeting, we were able to clear the air of a summers’ worth of housekeeping and establish the plan for the completion of coding all available cases.

We’re currently in the home stretch of coding these final cases before we begin analyzing the current data. Before tomorrow, the goal is to have added every person who qualifies into the database. This includes new cases that have made the news this summer as well as older cases we have found by combing through lists of people charged with terrorism. We understand that it is impossible to be fully comprehensive, but are hoping to get as large of a data set as possible before we start to pick apart patterns in the data.

While the final search for new cases is being completed, the majority of the team has been focusing on coding the rest of the database, both new cases and the case backlog. Much of this process is the standard coding procedure – each case is coded by two people, who then compare their results before entering it in the database as complete. Some cases are missing a few variables or awaiting charges and sentencing, while others need to be reviewed before being entered as a case that falls within our parameters.

A small group met Tuesday in order to work out some of the kinks in cases that have been up for debate in order to finalize the data available. The decision was made to hold cases to a stricter rather than a more lenient standard; if the case is in a grey area, we will be more willing to exclude it than include it. This was chosen so that our data and results truly represents what we say it does.

Wish us luck as we plow the final stretch of coding!