Wednesday, February 17, 2016

Visualizing Anti-Shia violence in Pakistan Part I: Has Anti-Shia Violence increased in Pakistan?

I have been playing around with the Tashaddud sectarian violence dataset intermittently between thesis related analyses for the past year or so. What is really special about this dataset is that by tracking Anti-Shia violence using newspaper articles from the DAWN archive, it includes city/town specific data on (mostly) Anti-Shia killings and attacks including the initial number of ppl killed (as published by the newspaper, an underestimate as many of the injured may die in the days/weeks/months following an attack), and injured (overestimate as many injured may die in the days/weeks/months following an attack). Additionally, this dataset also provides details about the religious affiliation of the attacker (Sunni or Shia) and some of the personal details about the professions of the victims. While I have been spending most of my time trying to add information about the type of attack (Assassination, Mass Shooting, Sectarian Clash, Bomb Blast, Suicide Blast, Arson). location (street, Imambargah, shop, medical facility, marketplace, education facility, home, religious procession) and the occupation of the victims (doctor, lawyer, politician..etc) from the DAWN articles references in the dataset, I had some time recently to analyse general trends in Anti-Shia violence.

Brief note: In addition to visualizing the trends in the data, I also conducted statistical analyses using generalized linear models. Model selection was based on the minimum adequate model criterion using AICc to determine which statistical distribution to use for null hypothesis significant testing (either poisson, quasi-poisson or negative binomial) and log likelihood ratio tests to determine which explanatory variables were important in explaining the variation in response. Details about the analyses and relevant code will be provided in an upcoming post. I also used conditional inference regression trees based on machine learning to determine if there were any thresholds in explanatory variables explaining Anti-Shia violence using Bonferonni adjusted p-values. The word significance in this post denotes statistical significance (p < 0.05).

Brief note #2: Anti Shia violence data from Tashaddud was only available for a nine year period from 2001 to 2013. I added data for the number of ppl killed and injured for 2014 from the United States Commission on International Religious Freedom 2014 report on Pakistan.  Before plotting any data or performing any analysis, I removed any data for violent attacks by Shias on LeJ/ASWJ/Sunni Tehreek members (108 of 723 data points).

Trend #1: The number of Anti-Shia attacks has increased significantly from 2001 to 2014 (p<0 .0001, dispersion="1.24).

Trend #2: The number of Pakistani Shias killed has increased significantly (p<0.0001, dispersion = 0.704). However, there was no trend detected in the number of people killed per attack (p = 0.904), suggesting that there has been no change in the deadliness of Anti-Shia violence during this 14 year period.

Trend #3: The number of Anti-Shia attacks increased after 2007 (p =0.01, predicted effect size = 48.3%. Units for y axes on boxplots are number of attacks.

Trend #4: The number of Pakistani Shias killed also increased after 2007(p= 0.004, predicted effect size = 179%). Units for y axes on boxplots are number of people killed. 
Coming up: Visualizing Anti-Shia violence in Pakistan Part II: Where do Pakistani Shias experience the most Anti-Shia Violence?