########################################################
# PATENT CITATION ANALYSIS

library(mgcv)
library(mgcViz)

df <- read.csv("patent_sample_50k.csv")

events <- df[,c('rec_pub_year','lag','rec_outd','jac_sim','cumu_cit_rec','tfe','sim')]
non_events <- df[,c('rec_pub_year_2','lag_2','rec_outd_2','jac_sim_2','cumu_cit_rec_2','tfe_2','sim_2')]

rec_pub_year <- cbind(events[,1],non_events[,1])
lag <- cbind(events[,2],non_events[,2])
rec_outd <- cbind(log(events[,3]),log(non_events[,3]))#log transform
jac_sim <- cbind(events[,4],non_events[,4])
cumu_cit_rec <- cbind(log(events[,5]),log(non_events[,5]))#log transform
tfe <- cbind(log(events[,6]+1),log(non_events[,6]+1))#log(x+1) transform
sim <- cbind(events[,7],non_events[,7])

y <- rep(1,nrow(events))

W <- matrix(NA,nrow = nrow(events),ncol=2)
W[,1] <- 1; W[,2] <- -1

model <- gam(formula = y ~ -1 + 
               s(rec_pub_year, by=W)+
               s(lag,by=W)+
               #s(rec_outd,by=W)+
               #s(jac_sim,by=W)+
               #s(cumu_cit_rec,by=W)+
               #s(tfe,by=W)+
               s(sim,by=W),
             family="binomial"(link = 'logit'),
             method="REML")
plot(model)


model_viz <- getViz(model)
plot(model_viz,selec=1)+theme_get()



remdat<-read.csv2("manufacturing.csv")
dim(remdat)
head(remdat)
attach(remdat)


n<-nrow(remdat)
ones<-rep(1,n)
#set up data structure

# reciprocity
T<-cbind(r4a,non.r4a)
I<-cbind(ones,-ones)

# sender and receiver effects
S<-cbind(as.factor(s),as.factor(non.s))
R<-cbind(as.factor(r),as.factor(non.r))

manu.rem<-gam(ones~ -1 + s(T,by=I) + s(S,by=I,bs="re")+
                s(R,by=I,bs="re"), family=binomial)

summary(manu.rem)