Brian Abelson | 2013-05-23
setwd("~/Dropbox/aclu-torture")
options(digits = 2)
par(family = "OpenSans-Semibold")
d <- read.csv("data/d-2013-03-18.csv", stringsAsFactors = F)
n <- nrow(d)
## field description
##
## subject_id the test subject's id
##
## start_page the start of the survey page
##
## end_page the end of the survey page
##
## meta1 Deployment in Iraq?
##
## meta2 Deployment in Afghanistan?
##
## meta3 Deployment in GTMO?
##
## meta4 Deployment elsewhere?
##
## meta5 Adequate training prior?
##
## meta6 Adequate training during?
##
## meta7 Adequate training re: response?
##
## meta8 Additional comments?
##
## tech1 Food/water deprivation
##
## tech2 Clothing deprivation
##
## tech3 Sleep deprivation
##
## tech4 Beating a detainee
##
## tech5 Waterboarding
##
## tech6 Choking/strangling
##
## tech7 Threatening
##
## tech8 Other mistreatment
##
## tech9 Burning
##
## tech10 Prolonged restraint
##
## tech11 Stress positions
##
## tech12 Forced physical exercise
##
## tech13 Electrical shock
##
## tech14 Threatening to shock
##
## tech15 Denial/delay of medical care
##
## tech16 Hooding/blindfolding
##
## tech17 Hot/cold temperatures
##
## tech18 Loud music
##
## tech19 Bright lights or darkness
##
## tech20 Extended isolation
##
## tech21 Duct tape
##
## tech22 Rapid response teams
##
## tech23 Military dogs
##
## tech24 Threatened use of military dogs
##
## tech25 Other animals
##
## tech26 Threatened use of other animals
##
## tech27 Qur'an disrespect
##
## tech28 Shaving
##
## tech29 Woman's clothing
##
## tech30 Sexual conduct
##
## tech31 Holding w/o acknowledging
##
## tech32 Rendition
##
## tech33 Threatened rendition
##
## tech34 Threats to family
##
## tech35 Other treatment (emotional trauma)
##
## tech36 Other religious/sexual mistreatment
##
## tech37 Other mistreatment (general)
##
## act1 End participation b/c of methods?
##
## act2 Other agent end participation?
##
## act3 Report concerns to FBI?
##
## act4 Report concerns to non-FBI?
##
## act5 Ordered not to report?
##
## act6 Retaliation?
##
## act7 Additional comments?
##
## misc1 Multiple techniques on same detainee?
##
## comments Additional comments
## percentage of sample by location of deployments
##
## 0.357594936708861 from iraq
## 0.205696202531646 from afghanistan
## 0.604430379746835 from gitmo
## 0.0411392405063291 from elsewhere
## percentage of sample by number of deployments
## 1 2 3
## 0.81 0.16 0.028
## [1] "did respondent end partipcation?"
## [1] 3.8
## [1] "did others end partipcation?"
## [1] 1.9
## [1] "report to fbi?"
## [1] 14
## [1] "report to non-fbi?"
## [1] 7
## [1] "order not to report?"
## [1] 1.9
## [1] "retaliate?"
## [1] 0.32
###Examples of Responses__
## Ordered not to report:
##
## Did not know what standards were used for interrogations, which interfered with successful interviews. Some were trained after his/her initial deployment, s/he hopes that the differences between LE and military interviews was made clear-- and s/he was encouraged not to question techniques used by other agencies. Heard about sleep deprivation, water treatment, threatening physical harm, and extreme temperatures as typical techniques for interrogation. Heard that beating and stress positions had been used in Afghanistan. Detainee testified to mistreatment of the Koran. Also heard that it would be "mentioned" to detainees that others had been rendered. Witnessed use of bright lights and, seperately, a Sargeant applying perfumed lotion to her hands, and (though view was obstructed) touching detainee's genital area causing him to scream, fall off his stool, and hurt his nose.
##
## Retaliation:
##
## Agent reports that two times s/he found detainees who had been chained in interrogation rooms without food or water for over 24 hours. The detainees had in each case urinated on themselves and in one case the detainee had deficated on himself. In addition, the detainees were subject to extreme cold and extreme heat, respectively. One of the detainees had begun to pull the hair out of his head, and "there was a big pile of hair in front of him on the floor." The detainees also reported having been exposed to loud rap music over the course of "hours or days." Agent also rails against the U.S.'s having allowed Chinese nationals to interrogate the Ouigers and says that "we the Chinese government with family contact information regarding their relatives in China."
set.seed(1234)
res7 <- invisible(lda(d$comments, n_topics = 7, stem_words = TRUE))
res7[[1]][1:7, ]
## topic_1 topic_2 topic_3 topic_4 topic_5 topic_6 topic_7
## 1 secret train interview isol report music prison
## 2 inform respons fbi time guard loud hood
## 3 classifi rapid interrog brig interrog light detent
## 4 indic team personnel hour qur interrog wit
## 5 box heard night subject sent sleep iraq
## 6 mark sleep heard held claim bright facil
## 7 check isol conduct kept person observ incid
Assign Topics to Variables
# lets identify a couple of these and make them variables
d$topic <- res7[[2]]$topic_a
d$classified <- 0
d$classified[d$topic == 1] <- 1
d$gitmo = ifelse(d$meta3 == 1, 1, 0)
table(d$act, d$mult_deploys)
##
## 0 1
## 0 200 51
## 1 55 8
table(d$act, d$classified)
##
## 0 1
## 0 184 69
## 1 44 19
table(d$act, d$gitmo)
##
## 0 1
## 0 100 149
## 1 20 42
m6 <- glm(act ~ gitmo + num_deploys + classified, data = d, family = "binomial")
summary(m6)
##
## Call:
## glm(formula = act ~ gitmo + num_deploys + classified, family = "binomial",
## data = d)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -0.797 -0.739 -0.604 -0.443 1.986
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.942 0.487 -1.93 0.053 .
## gitmo 0.451 0.306 1.47 0.141
## num_deploys -0.666 0.375 -1.78 0.076 .
## classified 0.173 0.314 0.55 0.582
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 310.69 on 310 degrees of freedom
## Residual deviance: 305.21 on 307 degrees of freedom
## (5 observations deleted due to missingness)
## AIC: 313.2
##
## Number of Fisher Scoring iterations: 4
Below 1 means odds are decreased
par(xaxs = "i", yaxs = "i", mai = c(0.5, 0.5, 0.5, 0.5))
barplot(exp(m6$coefficients[2:4]), main = "odds ratios of acting out", col = "darkblue",
ylab = "odds ratios")
abline(h = 1, lty = 2)
