http://cssamsterdam.github.io/

THE EFFECT OF SOCIAL MEDia IN school shootings

J. Garcia-Bernardo, H. Qi, J. M. Shultz, A. M. Cohen, N. F. Johnson, P. S. Dodds

Javier García-Bernardo

University of Amsterdam

April 8th, 2019

December 14th, 2012 : Sandy Hook Attack

Since then: 217 school shootings (95 fatal)

Visualization: https://everytownresearch.org/school-shootings
Visualization: https://everytownresearch.org/school-shootings

There is no accurate or useful profile of students who engaged in targeted school violence

Visualization: https://everytownresearch.org/school-shootings

MOTIVATION: attacks cluster in time

MOTIVATION: attacks cluster in time

MOTIVATION: attacks cluster in time

we wanted to understand why they cluster:

- Collected data on school shootings from different sources

- Collected all tweets mentioning "school shooting"

- Look at how social media influences school shootings

- Model the frequency of attacks to understand the behaviour of the attackers.

spoilers:

- Social media chatter increases the risk of new attacks

- Those induced attacks are different

SOCIAL MEDIA AFFECTS the propensity of new attacks

Copycat effect

modeling the different types of attack

FIRST model

RED QUEEN VERSuS BLUE KING

N. F. Johnson, et al., Scientific reports 3, 3463 (2013).

Idea:

  • There is a competition between the attackers and the prevention forces.
  • The attacks accelerate or decelerate depending on the balance of forces.

 

What can we do this this model:

  • Predict the time until the next attack
  • Understand what type of attacks we can expect (characteristics, severity)

Attacks that happen before they "should" are different

modeling THE LOcation and timing

HAWKES PROCESS

D. Marsan, O. Lengliné, Science 319, 1076 (2008).

A. G. Hawkes, D. Oakes, Journal of Applied Probability pp. 493–503 (1974).

Idea:

  • Each attack increases the probability of new attacks.
    • But that effect may reduce over time or distance
  • We can compare the real data with "null models", where attacks happen at random.

 

What can we do this this model:

  • Understand where to focus prevention efforts                                                                            
  • Quantify the time until things go back to normal

second model

distance component

distance component

time component

  • The concept of school shooter in the US is well established.
  • Popular media matters too: "to go NBK on a school"
  • Schools shooters do not need to be creative.
  • We show  limited contagion, but only for a extreme group of people:
    • School shootings cluster in time, but not in space.
    • School shootings get media attention, which in turn affect the probability of new attacks (copycat effect).
    • We can classify the behaviour of the attackers
      • Induced attacks: Low mortality and suicide rate, happening in the two weeks after an attack.
      • Other: High mortality and suicide rate, happening often after periods of inactivity.
Paper:arxiv: 1506.06305

corpnet.uva.nl

@javiergb_com

@uvaCORPNET

javiergb.com

corpnet@uva.nl

garcia@uva.nl

This presentation: slides.com/jgarciab/ss19

CONCLUSIONS

cssAmsterdam.github.io

Made with Slides.com