#### Code for survival analysis presented in
#### A test of the “genetic rescue” technique using bottlenecked donor populations
#### Heber S, Briskie JV and Apiolaza LA
####
#### Sol Heber
#### School of Biological Sciences, University of Canterbury
#### Private Bag 4800, Christchurch, New Zealand
#### Sol.Heber@pg.canterbury.ac.nz
####
####
#### Code cleaned on 2011-10-07

####### Packages needed for analysis
library(survival) # Therneau, T. (2011)
library(multcomp) # Hothorn et al. (2008)

####### Read data files
survival<-read.table(file="S5_survival.txt",header=T)
attach(survival)
names(survival)

####### Fit model with exponential error distribution:
model1<-survreg(Surv(time,ci)~cross,dist="exponential")

####### Fit model with Weibull error distribution:
model2<-survreg(Surv(time,ci)~cross)

####### Compare fit:
anova(model1,model2)

####### Simplyfiying model by combining hybrid F2 from hybrid-hybrid and hybrid-inbred matings:
model3<-survreg(Surv(time,ci)~cross2) 
anova(model2,model3)
summary(model3)

####### Compare survival rates between all group pairs:
crosscomp<-glht(model3,linfct=mcp(cross2='Tukey'))
summary(crosscomp)
plot(crosscomp)

####### Calculate adjusted confidence intervals:
adjusted.z<-abs(qnorm("p-value"/2)) #abs for absolute gets rid of +/- sign as not important in t- and z-values
std.error<-"Estimate"/"adjusted.z"
lower95%CI<-"Estimate"-(std.error*1.96)
upper95%CI<-"Estimate"+(std.error*1.96)

####### Very significant p-values are reported for example as <0.001. Therefore extract the exact p-values to be used in above calculations of adujusted CIs:
summary(crosscomp)$test$pvalues

####### Graph with survival curves:
plot(survfit(Surv(time,ci)~cross2),lty=c(1:6),ylab="Survivorship",xlab="Time")

####### Predict life span for each group separately:
tapply(predict(model3,type="response"),cross2,mean)