1_Data_tidying.R

########################################################################################################################
## GOSSIP IN HUNGARIAN FIRMS
## Data tidying (1)
## R script written by Jose Luis Estevez (Masaryk University)
## Date: February 26th, 2022
########################################################################################################################

# R PACKAGES REQUIRED
library(ggplot2)

# DATA LOADING AND DATA TIDYING
rm(list=ls())
load('data.RData')

# Leaders per setting (fixing this variable which, in the original data set, is inaccurate)
for(i in 1:nrow(attributes)){
  if(!is.na(stringr::str_detect(attributes$hr_job[[i]],' '))){
    if(stringr::str_detect(attributes$hr_job[[i]],c('Manager|Chief|Director|director|leader'))){
      attributes$hr_leader[[i]] <- 1
    }
  }
}

attributes$woman[is.na(attributes$woman)] <- 0 # We know those two did not report their gender were male

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

# GENERAL INFORMATION
(organisation_ID <- unique(na.omit(attributes$group))) # 9 units or firms/departments
(employee_ID <- attributes$responder) # N=225, but 2 duplicated employees: 1F6032 and 1F6033
(networks_available <- unique(networks$network_questionID_EN)) # 43 relational variables per unit available 
# "received_personal_info" refers to the gossip, so it was excluded from the set of relational variables (42 variables)
networks <- networks[networks$network_questionID_EN != "received_personal_info",]
networks_available <- networks_available[-42]

# Fixing duplicated employees in all three objects: attributes, networks, and gossip
'%!in%' <- function(x,y)!('%in%'(x,y))
attributes <- attributes[attributes$responder %!in% c('1F6032_F106b','1F6033_F106b'),] # N=223
attributes$responder[122] <- '1F6032'
attributes$responder[123] <- '1F6033'

duplicated_actors <- c('1F6032_F106a','1F6032_F106b','1F6033_F106a','1F6033_F106b')

for(i in 1:nrow(networks)){
  if(networks$sender[i] %in% duplicated_actors){
    networks$sender[i] <- substr(networks$sender[i],1,6)
  }
  if(networks$receiver[i] %in% duplicated_actors){
    networks$receiver[i] <- substr(networks$receiver[i],1,6)
  }
}

for(i in 1:nrow(gossip)){
  if(gossip$sender[i] %in% duplicated_actors){
    gossip$sender[i] <- substr(gossip$sender[i],1,6)
  }
  if(gossip$receiver[i] %in% duplicated_actors){
    gossip$receiver[i] <- substr(gossip$receiver[i],1,6)
  }
  if(gossip$target[i] %in% duplicated_actors){
    gossip$target[i] <- substr(gossip$target[i],1,6)
  }
}

# Respondents missing
apply(attributes[,],2,is.na)*1 -> missing_data
attributes$respondent_missing <- as.vector(apply(missing_data,1,sum)) # items non-responded
attributes$respondent_missing <- attributes$respondent_missing > 50
(missing_respondents <- attributes[attributes$respondent_missing == TRUE,'responder']) # 18 non-respondents

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

# CREATION OF GOSSIP-CUBES
# Subjects' IDs by organisation: for the dimensions of matrices and cubes
org_subject <- vector('list',length=length(organisation_ID))
names(org_subject) <- organisation_ID

for(i in seq_along(org_subject)){
  org_subject[[i]] <- attributes[attributes$group == names(org_subject)[i],]$responder
}

# Exclusion of gossip triples when not all three parties (sender, receiver, target) are part of the same network
valid_triplets <- vector(length=nrow(gossip))

for(i in 1:nrow(gossip)){
  if(gossip$sender[i] %in% org_subject[[gossip$group[[i]]]] &
     gossip$receiver[i] %in% org_subject[[gossip$group[[i]]]] &
     gossip$target[i] %in% org_subject[[gossip$group[[i]]]])
    valid_triplets[i] <- TRUE
}
gossip <- gossip[valid_triplets,]

# Any duplicated?
any(duplicated(gossip[,c('sender','receiver','target')],)) 
which(duplicated(gossip[,c('sender','receiver','target')],))
gossip[c(598,599,676,677,770,771),] # neutral gossips to positive or negative
gossip <- gossip[-c(599,677,770),]

# Creation of gossip cubes (sender, receiver, target)
gossip_cube <- org_subject
for(i in seq_along(gossip_cube)){
  gossip_cube[[i]] <- array(NA,dim=c(rep(length(org_subject[[i]]),3)),
                            dimnames=list(org_subject[[i]],org_subject[[i]],org_subject[[i]]))
}

# Allocation of the gossip triplets in the cubes
for(i in 1:nrow(gossip)){
  gossip_cube[[gossip$group[i]]][gossip$sender[i],gossip$receiver[i],gossip$target[i]] <- gossip$info[i]
}

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

# CREATION OF RELATIONAL MATRICES
# Exclusion of ties between different networks
valid_ties <- vector(length=nrow(networks))

for(i in 1:nrow(networks)){
  if(networks$sender[i] %in% org_subject[[networks$group[[i]]]] &
     networks$receiver[i] %in% org_subject[[networks$group[[i]]]])
    valid_ties[i] <- TRUE
}
networks <- networks[valid_ties,]

# Creation of matrices (relational data)
networks_mtx <- org_subject

for(i in seq_along(networks_mtx)){
  networks_mtx[[i]] <- vector('list',length=length(networks_available))
  names(networks_mtx[[i]]) <- networks_available
  for(j in seq_along(networks_mtx[[i]])){
    networks_mtx[[i]][[j]] <- array(0,dim=c(rep(length(org_subject[[i]]),2)),
                                    dimnames=list(org_subject[[i]],org_subject[[i]]))
    # Allocation of missing data
    diag(networks_mtx[[i]][[j]]) <- NA # diagonal
    for(k in 1:nrow(networks_mtx[[i]][[j]])){
      if(rownames(networks_mtx[[i]][[j]])[k] %in% missing_respondents){
        networks_mtx[[i]][[j]][k,] <- NA # respondents missing
      }
    }
  }
}

# Allocation of the ties into the matrices
networks$tie <- 1
for(i in 1:nrow(networks)){
  gro <- networks$group[i]
  mtx <- networks$network_questionID_EN[i]
  row <- networks$sender[i]
  col <- networks$receiver[i]
  networks_mtx[[gro]][[mtx]][row,col] <- networks$tie[i]
}

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

# CHECKING THE GOSSIP CUBE
gos <- gos_pos <- gos_neg <- gossip_cube

for(x in seq_along(gossip_cube)){
  for(i in 1:nrow(gossip_cube[[x]])){
    for(j in 1:nrow(gossip_cube[[x]])){
      for(k in 1:nrow(gossip_cube[[x]])){
        if(i != j & i != k & j != k){
          # the gossip cube (either positive or negative gossip, but not neutral)
          if(!is.na(gossip_cube[[x]][i,j,k]) & (gossip_cube[[x]][i,j,k] == 1 | gossip_cube[[x]][i,j,k] == -1)){
            gos[[x]][i,j,k] <- 1
          }else{
            gos[[x]][i,j,k] <- 0
          }
          # the positive gossip cube
          if(!is.na(gossip_cube[[x]][i,j,k]) & gossip_cube[[x]][i,j,k] == 1){
            gos_pos[[x]][i,j,k] <- 1
          }else{
            gos_pos[[x]][i,j,k] <- 0
          }
          # the negative gossip cube
          if(!is.na(gossip_cube[[x]][i,j,k]) & gossip_cube[[x]][i,j,k] == -1){
            gos_neg[[x]][i,j,k] <- 1
          }else{
            gos_neg[[x]][i,j,k] <- 0
          }
        }
      }
    }
  }
}

# Allocation of missing data (if a respondent was missing, all sender-target ties involving this respondent are missing)
for(x in seq_along(gossip_cube)){
  for(i in rownames(gossip_cube[[x]])){
    if(i %in% missing_respondents){
      gos[[x]][,i,] <- gos_pos[[x]][,i,] <- gos_neg[[x]][,i,] <-  gos[[x]][,i,]*NA
    }
  }
}

# How many triad in the data set? How many participants reported gossip?
rec_sen <- gos
rec_sen_pos <- gos_pos
rec_sen_neg <- gos_neg

for(x in seq_along(rec_sen)){
  rec_sen[[x]] <- apply(rec_sen[[x]],c(2,1),max,na.rm=TRUE) # receiver-sender matrix
  rec_sen_pos[[x]] <- apply(rec_sen_pos[[x]],c(2,1),max,na.rm=TRUE)
  rec_sen_neg[[x]] <- apply(rec_sen_neg[[x]],c(2,1),max,na.rm=TRUE)
  diag(rec_sen[[x]]) <- 0
  diag(rec_sen_pos[[x]]) <- 0
  diag(rec_sen_neg[[x]]) <- 0
  rec_sen[[x]] <- rowSums(rec_sen[[x]])
  rec_sen_pos[[x]] <- rowSums(rec_sen_pos[[x]])
  rec_sen_neg[[x]] <- rowSums(rec_sen_neg[[x]])
}

gossip_sum <- as.data.frame(matrix(NA,nrow=length(rec_sen),ncol=8,
                                   dimnames=list(names(rec_sen),c('gossip','pos_gossip','neg_gossip','triads',
                                                                  'reporters','reporters_pos','reporters_neg','nodes'))))

for(x in seq_along(rec_sen)){
  gossip_sum[x,'gossip'] <- sum(gos[[x]],na.rm=TRUE) # gossip triads
  gossip_sum[x,'pos_gossip'] <- sum(gos_pos[[x]],na.rm=TRUE) # positive gossip triads
  gossip_sum[x,'neg_gossip'] <- sum(gos_neg[[x]],na.rm=TRUE) # negative gossip triads
  gossip_sum[x,'triads'] <- sum(!is.na(gos[[x]])) # potential gossip triads
  gossip_sum[x,'reporters'] <- sum(rec_sen[[x]] != 0) # number of gossip reporters
  gossip_sum[x,'reporters_pos'] <- sum(rec_sen_pos[[x]] != 0) # number of pos. gossip reporters
  gossip_sum[x,'reporters_neg'] <- sum(rec_sen_neg[[x]] != 0) # number of neg. gossip reporters 
  gossip_sum[x,'nodes'] <- length(rec_sen[[x]]) # nodes in the network
}

gossip_sum

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

# SOME DESCRIPTIVES OF THE GOSSIP DATA
desc_receiver <- vector('list',length=length(gossip_cube))
for(i in seq_along(gossip_cube)){
  desc_receiver[[i]] <- as.data.frame(matrix(NA,nrow=nrow(gossip_cube[[i]]),ncol=4))
  rownames(desc_receiver[[i]]) <- rownames(gossip_cube[[i]])
  colnames(desc_receiver[[i]]) <- c('network','dyads','dyads_pos','dyads_neg')
  desc_receiver[[i]]$network <- names(gossip_cube)[i]
}

for(x in seq_along(desc_receiver)){
  for(i in 1:nrow(desc_receiver[[x]])){
    # Receiver-specific 
    desc_receiver[[x]]$dyads[i] <- sum(gos[[x]][,i,],na.rm=TRUE)
    desc_receiver[[x]]$dyads_pos[i] <- sum(gos_pos[[x]][,i,],na.rm=TRUE)
    desc_receiver[[x]]$dyads_neg[i] <- sum(gos_neg[[x]][,i,],na.rm=TRUE)
  }
}

desc_receiver <- do.call('rbind',desc_receiver)
desc_receiver$subject <- rownames(desc_receiver)

# Visualisations
grid.background <- theme_bw()+
  theme(plot.background=element_blank(),panel.grid.minor=element_blank(),panel.border=element_blank())+
  theme(axis.line=element_line(color='black'))+
  theme(strip.text.x=element_text(colour='white',face='bold'))+
  theme(strip.background=element_rect(fill='black'))+
  theme(axis.text.x = element_blank())

jpeg(filename='Receiver-specific gossip.jpeg',width=11,height=6,units='in',res=1000)
ggplot(data=desc_receiver)+
  geom_point(aes(x=subject,y=dyads_neg),size=2)+
  geom_point(aes(x=subject,y=dyads_neg),size=1,colour='red')+
  geom_point(aes(x=subject,y=dyads_pos),size=2)+
  geom_point(aes(x=subject,y=dyads_pos),size=1,colour='chartreuse')+
  facet_wrap(~network,nrow=3,scales='free')+
  xlab('Respondent')+ylab('Sender-target dyads reported')+
  grid.background
dev.off()

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

# Need to exclude networks F106b-c-d (gossip reported by very few individuals)
excl <- c('F106b','F106c','F106d')
attributes <- attributes[attributes$group %!in% excl,]
missing_respondents <- missing_respondents[missing_respondents %in% attributes$responder]
gossip <- gossip[gossip$group %!in% excl,]
gossip_cube <- gossip_cube[organisation_ID %!in% excl]
gos <- gos[organisation_ID %!in% excl]
gos_pos <- gos_pos[organisation_ID %!in% excl]
gos_neg <- gos_neg[organisation_ID %!in% excl]
gossip_sum <- gossip_sum[rownames(gossip_sum) %!in% excl,]
networks_mtx <- networks_mtx[organisation_ID %!in% excl]
org_subject <- org_subject[organisation_ID %!in% excl]
organisation_ID <- organisation_ID[organisation_ID %!in% excl]

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

# Removal of unnecessary objects
rm(duplicated_actors);rm(missing_data);rm(valid_triplets);rm(valid_ties);rm(employee_ID);rm(networks);rm(gro);rm(mtx)
rm(row);rm(col);rm(rec_sen);rm(rec_sen_pos);rm(rec_sen_neg);rm(desc_receiver);rm(gos);rm(excl);rm(i);rm(j);rm(k);rm(x)
rm(gossip);rm(gossip_cube);rm(grid.background)

# Save image
save.image('tidieddata.RData')