A Statistical Exploration Of School Shootings

This article comes courtesy of guest authors Michael Keith and Victoria Roberts. Keith and Roberts both graduated from Florida State University with Masters of Science degrees in Applied Economics. An expanded version of this piece can be read on Keith’s blog. 

On the afternoon of February 14, 2018, the country began to be informed of a school shooting that had occurred in Parkland, Florida. As the days passed and details unfolded, the image became clearer of the  perpetrator and his death toll. The 18 year old, recently dropped out of the high school he victimized, had killed 17 people and wounded another 17. The national discourse that followed centered around the same questions it always does after such events. Why did this happen? What drove this young man to a point where he could do something so evil? How can we prevent this from happening again?

Undoubtedly, these are important questions. And there’s a reason the debate surrounding school shootings is so heated on every side—people are genuinely revolted by this kind of thing and want to see it stop.

But, there is another set of questions that rarely get asked. And maybe only a statistician or an economist  would think to ask them—what are the commonalities across school shootings? What things can be measured and what things can be predicted? Knowing the motivation of the shooter and what drove him to do something so awful is one thing. We can make good guesses about that and do so logically, but ultimately, we may never know the final answer. However, there are things that can be better understood and knowing them may lead to some mitigation of the phenomenon.

Data Analysis

The Washington Post maintains a page online that gives us access to important data—hard numbers. Numbers about each school shooting since Columbine in 1999—the type of weapon used, the level of poverty in the victimized school, whether a resource officer was present at the time of the shooting, how the weapon was obtained, the age and gender of the shooter, etc. Using this repository as well as a set of econometric skills combined with statistical inference, we analyzed school shootings to provide a set of answers we rarely hear. Again, these aren’t conclusions that will solve all our problems, but perhaps they could make up the basis of how we think about the issue.

A more detailed explanation of the methods employed in this analysis can be found on my blog, here and here. This is the process we followed.:

• Collect data
• Decide what to model
• Decide how to model it
• Write a model
• Test and measure the accuracy of the model
• If necessary, write other models to validate initial findings
• Share conclusions

You’ll notice in these steps that there is no room to put personal spin or opinion on things. Statistical models use numbers that can’t be obtained by an outside source and plugs them into statistical algorithms that are well established empirically to make the best predictions possible. Although the conclusions can be interpreted subjectively, there is no subjectivity in the process (as long as believable assumptions are made about the methods—which, in this case, is something you can and should check on my blog).

Summary of the Data

Some of the simpler insights that the data revealed are the following:

• At least 89% of school shootings were committed by a male (or multiple males)
• At least 61% of the shootings were committed by individuals under 18, usually current or former students
• 35% of the shootings occurred on a campus where resource officers were present
• 72% of the shootings occurred on campuses where at least 25% of the student-body was eligible for a reduced-priced or free lunch program—an indication of general poverty

Statistical Inferences

Going further and unwinding some of the data’s deeper complexities, we uncovered more interesting relationships. Attempting to predict the number of casualties in a given school shooting using the other factors in the data, the following can be concluded:

Poverty: The greater the percentage of impoverished students in the school, the fewer casualties are suffered in a given shooting. When you look at how school-shooting events cluster together—which shootings share many common characteristics with one another—most of the schools in the data were impoverished generally. But these impoverished schools saw few casualties per shooting, usually 0 to 2. The schools that suffered the most casualties generally possessed a student body that was more economically well-off. Think of Parkland, Columbine, and Sandy Hook.

The data seems to suggest that while poorer schools are more likely to see a higher number of low-casualty shootings, schools which are better-off economically attract shooters who want to gain notoriety by causing as much harm as possible. This could serve as one explanation for why casualties are generally higher in economically better schools, but there are plenty of other reasons why this could be.

Resource Officer: When a resource officer is on campus during a shooting the data shows an average 3.5 more casualties occur when other factors are held constant. This was a surprising finding and maybe unbelievable to some, but the results were always statistically significant, no matter how the data was sliced. In deriving this result, around 30 other factors were accounted and controlled for—factors such as enrollment, type of weapon used in the shooting, age of the shooter, percent of poverty in the school (and by proxy, violent-crime rate), etc.

Having said that, I’m not convinced that a resource officer on campus leads to more casualties necessarily (although we don’t rule it out). Correlation does not necessarily mean causation.The evidence for such a relationship is strong, and therefore, we must challenge ourselves to look beyond the factors controlled for in the model that may explain why we see this relationship. Keeping this in mind can stretch us to think more deeply about the issue and to consider different ideas.

Weapon Type: The models suggested that the weapon used by the shooter made a significant difference in the number of casualties. In particular, the most prolific weapon was a rifle—any kind of rifle. Rifles caused on average 6 more casualties when used in a school shooting compared to when some other weapon was used, controlling for all other factors in the data. This finding was also statistically significant, so, like the relationship illustrated with resource officers, it’s hard to say that it’s only due to chance that we see this occur.

Illegally Obtained Weapons: The data loosely suggested that fewer casualties were caused when a weapon was obtained illegally. This was a statistically insignificant finding, but we thought it important to note as this is a hotly contested issue. We would warn against drawing a strong conclusion either way; it’s safer to say that we need more evidence before fully understanding how the legality of a weapon affects shootings. Nevertheless, according to the best guess the data can offer, this is the relationship that existed.

To recapitulate:

• More impoverished schools see fewer casualties per school shooting, all else constant.
• Schools with resource officers on campus see more casualties per school shooting, all else constant.
• Shooters who use rifles cause more casualties, all else constant.
• The legality of the weapon used was inconclusive as far as how it affects shootings, but the models hinted that fewer casualties occur when the weapon is obtained illegally.  

The Bottom Line

So,  what are the practical implications? Since we have some numerical and statistical output available to us, what should we do about it? My answer: use it to think more deeply. We don’t have to make knee-jerk reactions on these issues based on what our favorite political parties are saying about it. There are some things that we can only ever guess at. But there are other things we can discover by using statistical methods or reading about how such methods are being applied to these data. The methods won’t settle all debates, but they will reveal the complicated nature of some of the variables. And that’s a start.