# Oscar data -- "winner" as time-dependent covariate
---------------------------------------------------------  

# FULL DATA for Winners versus nominated

ds=read.table("WinNomR.txt",header=T); summary(ds)

# extract info just on nominees (and create 1 new record for each) 

 did.not.win = ds[is.na(ds$age.1st.win),]; 

 did.not.win$winner   = 0
 did.not.win$age.from = did.not.win$age.1st.nom
 did.not.win$age.to   = did.not.win$age.last
 did.not.win[1:5,]

# extract info just on winners (and create 2 new records for each)
#                           [just 1 record if won on first nomination ]

  won=ds[!is.na(ds$age.1st.win),] ;

  won.1st.record          = won[won$age.1st.win > won$age.1st.nom,]
 
  won.1st.record$winner   = 0
  won.1st.record$age.from = won.1st.record$age.1st.nom
  won.1st.record$age.to   = won.1st.record$age.1st.win
  won.1st.record$dead     = 0
  won.1st.record[1:5,]

  won.2nd.record          = won
 
  won.2nd.record$winner   = 1
  won.2nd.record$age.from = won.2nd.record$age.1st.win
  won.2nd.record$age.to   = won.2nd.record$age.last
  won.2nd.record[1:5,]

# join the records

  ds.split.records = rbind(won.1st.record, won.2nd.record, did.not.win);
  ds.split.records = ds.split.records[order(ds.split.records$identity),];
  ds.split.records[1:15,]

# clean up
   
  remove("won.1st.record", "won.2nd.record", "won", "did.not.win" ); ls();
  
require(survival)
  
# using Surv(start.time, stop.time, event.idicator) format...
  
# 2 (#680 and # 729) died the same age at which they were nominated, so
# add 1/10th of a year to last age, so age.to > age.from

t.dep.model = coxph(Surv(age.from,age.to + 0.1,dead) ~ winner, 
                    data = ds.split.records );

summary(t.dep.model);

# add 2 other (both fixed) variates  male + year.born + 

##############

 model with winner as FIXED variate

 ds$w=!is.na(ds$age.1st.win)
 
t.indep.model = coxph(Surv(age.last,dead) ~ male + year.born + w, data = ds )

summary(t.indep.model)