01_Data_Setup.Rmd

---
title: "Data Preparation - Timing, Development and Diffusion of Work-Injury Laws Globally: Industrialization, Nation States and Institutions"
output: html_document
---

Nate Breznau
Felix Lanver
University of Bremen

We setup the data to analyze both the first law (often employer liability), and the first social insurance law (with the suffix *_socins*)


```{r setup}

# Clean up
rm(list = ls(all = T))


pacman::p_load('netdiffuseR','stargazer','tidyverse','dplyr','countrycode','sandwich','lmtest','Hmisc')

# stargazer 3 (define manually to make code faster)
# source("http://www.barkhof.uni-bremen.de/~mwindzio/modified_stargazer.R")

stargazer3 <- function(model, odds.ratios = F, origin_model = NULL, ...) {
  if(!("list" %in% class(model))) model <- list(model)
  if(!("list" %in% class(origin_model))) origin_model <- list(origin_model) 
  if (odds.ratios) {
    coefOR2 <- lapply(model, function(x) exp(x[,1]))
    #seOR2 <- lapply(model, function(x) exp(x[,1]) * exp(x[,2]))
    p2 <- lapply(model, function(x) x[,4])
    obs <- lapply(origin_model, function(x) nobs(x))
    LL <- lapply(origin_model, function(x) round(logLik(x),3))
    AIC <- lapply(origin_model, function(x) round(AIC(x),3))
    stargazer(model, coef = coefOR2,  p = p2,
              star.char = c("+", "*", "**", "***"),
              star.cutoffs = c(.1, .05, .01, .001),
              notes = "+ p < 0.1; * p < 0.05; ** p < 0.01; *** p < 0.001",
              notes.append = FALSE,
              add.lines=list(c("Observations", unlist(obs)),
                             c("Log Likelihood", unlist(LL)),
                             c("Akaike Inf. Crit.", unlist(AIC))), ...)
  } else {
    stargazer(model, ...)
  }
}

```

## Get Data

**Networkdata sources**

http://www.barkhof.uni-bremen.de/~mwindzio/edgelist_cultural_spheres.zip
http://www.barkhof.uni-bremen.de/~mwindzio/COW_trade_consolidated.zip
http://www.barkhof.uni-bremen.de/~mwindzio/trade_existence_dummy.csv
http://www.barkhof.uni-bremen.de/~mwindzio/capital_distance_edgelist.zip
http://www.barkhof.uni-bremen.de/~mwindzio/vdem_gdp_pc_INTERPOLATED_consolidated.csv
http://www.barkhof.uni-bremen.de/~mwindzio/vdem_regime__INTERPOLATED_consolidated.csv

This one stays zipped as GitHub has a 100 MB max filesize
http://www.barkhof.uni-bremen.de/~mwindzio/capital_distance_edgelist.zip


## 1. Network Data

```{r windzio_data, warning=F, message=F}
#--- 1. Networks here ---------------


#----- 1a) cultural spheres ---------

edgelist_cultural_spheres <- read.csv("data/edgelist_cultural_spheres.csv")
# head(edgelist_cultural_spheres)
dim(edgelist_cultural_spheres)

# - ids should be factor variables
edgelist_cultural_spheres$ego_id <- as.factor(edgelist_cultural_spheres$ego_id)
edgelist_cultural_spheres$alter_id <- as.factor(edgelist_cultural_spheres$alter_id)
# head(edgelist_cultural_spheres)


#----- 1b) trade network -----------------------------------

trade_dyadic_consolidated <- read.csv("data/COW_trade_consolidated.csv")

trade_dyadic <- trade_dyadic_consolidated
dim(trade_dyadic)
# head(trade_dyadic)

summary(trade_dyadic$smoothtotrade)

ddat <- trade_dyadic[is.na(trade_dyadic$smoothtotrade)==FALSE,]
summary(ddat$smoothtotrade)
d <- density(ddat$smoothtotrade)
plot(d)
polygon(d, col="red", border="blue") 

names(trade_dyadic)[names(trade_dyadic)=="iso3_ego"] <- "ego_id"
names(trade_dyadic)[names(trade_dyadic)=="iso3_alter"] <- "alter_id"
# head(trade_dyadic)

# - ids should be factor variables
trade_dyadic$ego_id <- as.factor(trade_dyadic$ego_id)
trade_dyadic$alter_id <- as.factor(trade_dyadic$alter_id)


# - NA to 0 and log positive values
trade_dyadic$smoothtotrade[trade_dyadic$smoothtotrade < 0] <- 0
trade_dyadic$smoothtotrade[is.na(trade_dyadic$smoothtotrade)] <- 0

summary(trade_dyadic$smoothtotrade)
# head(trade_dyadic)

trade_dyadic$log_value <- ifelse(trade_dyadic$smoothtotrade > 0, log(trade_dyadic$smoothtotrade), 0)
summary(trade_dyadic$log_value)

# We realized that with only using a log function to smooth out the much skewed trade values we created negative values. That in itself would be not such a huge problem, however since we are dealing with network weights a negative weight does not only make no sense theoretically, it actually distorts the calculation of the exposure variable a lot. We end up with values outside the range of 0 – 1. To circumvent that we changed the script in line 109 in a way that we make sure to not end up with negative log_values:

trade_dyadic$log_value <- ifelse(trade_dyadic$smoothtotrade > 0, log(trade_dyadic$smoothtotrade + 1), 0) 
summary(trade_dyadic$log_value)

# -- distribution is still extreme  
d <- density(trade_dyadic$log_value)
plot(d)
polygon(d, col="red", border="blue")
trade_edgelist <- trade_dyadic

# -- we want this to be "independence" therefore, we use our own data to create this variable.

A02_workinj <- read.csv("data/A02_workinj.csv", header = TRUE)

# These data have missing cases, fix by using the empty frame

A02_workinj <- select(A02_workinj, country_name, cow_code, iso3, year, independence)

# Replace missing cow_codes
A02_workinj$cow2 <- countrycode(A02_workinj$iso3, 'iso3c','cown')

# Find mismatches
# A02_workinj2 <- subset(A02_workinj, A02_workinj$cow_code != A02_workinj$cow2)
# Fix
A02_workinj <- A02_workinj %>%
  mutate(cow_code = ifelse(cow2==817, 816, cow_code),
         cow_code = ifelse(is.na(cow_code), cow2, cow_code),
         cow_code = ifelse(country_name == "Serbia", 345, cow_code),
         iso3 = ifelse(cow_code == 816, "VNM", ifelse(cow_code == 345, "SRB", iso3)))
```

The GWIP  presents us with a conundrum regarding treatment of the former Soveit Union states. Many were independent states on paper, but were under the centralized control of the Russian Communist party. Socialism is conceptually the same as social insurance when it comes to protections for workers. They do not pay into an insurance system; however, they are subjected to total state employment control. This means that their needs should be taken care of, as if they were in a social insurance system. Therefore, the first laws occur starting in 1922 for non-Russian Soviet states, as they form and/or join the USSR. However, the official independence of these states comes after the dissolution of the USSR in 1991. This means that we either recode the first laws to 1991 or record their independence dates to 1922, or before, depending on the case. We run our models with both versions of the recode but settle on the pre-USSR independence coding as these were national groups aiming to form states and joined the USSR as part of this process or through Russian force. This issue only applies to Belarus, Kyrgyzstan, Croatia and Slovenia.  

```{r fixes, warning=F, message=F}
A02_workinj <- A02_workinj %>%
  mutate(independence = ifelse(iso3 == "BLR" | iso3 == "KGZ", 1922, ifelse(iso3 == "HRV" | iso3 == "SVN", 1937, independence)))

existence <- select(A02_workinj, iso3, year, independence)

colnames(existence) <- c("iso3", "year", "existence")

# now make 0/1 for existence
existence_c <- existence %>%
  group_by(iso3) %>%
  dplyr::summarize(existence = max(existence, na.rm = T))

existence <- select(existence, -c(existence))

existence <- left_join(existence, existence_c, by = "iso3")

existence <- existence %>%
         mutate(exist = ifelse(year < existence, 0, 1))





A02_workinj <- select(A02_workinj, country_name, cow_code, iso3, year)
```

Collier and Messick (1975) argue that percent labor force in agriculture is key to the introduction of work-injury law. This measure should be an indicator of industrialization/modernization. Therefore we import the Banks time-series data here. It is available for a smaller set of countries, but provides a robustness check.

```{r pct_ag}
pct_ag <- read_csv(file = "data/CNTSDATA_2020_Ag_only.csv")

pct_ag <- select(pct_ag, iso3, year, pct_ag_01)

# linear interpolate non-complete-missing series
pct_ag <- pct_ag %>%
  group_by(iso3) %>%
  filter(any(!is.na(pct_ag_01))) %>%
  subset(iso3 != "ARE") %>% # have only one value
  subset(iso3 != "BGD") %>%
  subset(iso3 != "BWA") %>%
  subset(iso3 != "CHN") %>%
  subset(iso3 != "CMR") %>%
  subset(iso3 != "FJI") %>%
  subset(iso3 != "GAB") %>%
  subset(iso3 != "GUY") %>%
  subset(iso3 != "MWI") %>%
  mutate(pct_ag_01_i = zoo::na.approx(pct_ag_01, na.rm = F),
         pct_ag_01_i = zoo::na.fill(pct_ag_01_i, "extend")) %>%
  ungroup()

# put together
A02_workinj <- left_join(A02_workinj, pct_ag, by = c("iso3", "year")) 



```



```{r fixes2, warning=F, message=F}
#----- 1c) colonial ties from CEPI --

# - this is only a cross-sectional network ! 
# - we think about using a better one
# - if so, you will have to switch to "undirected=FALSE"
# - when creating the diffnet object. But don't care at the moment! 

colonial_ties_edgelist <- read.csv("data/colonial_ties_edgelist.csv")
head(colonial_ties_edgelist)

names(colonial_ties_edgelist)[names(colonial_ties_edgelist)=="iso_o"] <- "ego_id"
names(colonial_ties_edgelist)[names(colonial_ties_edgelist)=="iso_d"] <- "alter_id"

# - ids should be factor variables
colonial_ties_edgelist$ego_id <- as.factor(colonial_ties_edgelist$ego_id)
colonial_ties_edgelist$alter_id <- as.factor(colonial_ties_edgelist$alter_id)
# head(colonial_ties_edgelist)
unique(colonial_ties_edgelist$ego_id)
unique(colonial_ties_edgelist$year)
table(colonial_ties_edgelist$colony)
# head(colonial_ties_edgelist)

#----------------------------------------------------------
#----- 1d) spatial distance/ prox. of capitals from CEPI --
#----------------------------------------------------------
# - this is only a cross-sectional network ! 


capital_distance_edgelist <- read.csv(unz("data/capital_distance_edgelist.zip", "capital_distance_edgelist.csv"))



# --  we use log of proximity
capital_distance_edgelist$log_dist <- log(capital_distance_edgelist$capdist)
log(7593.042) # => 8.934988, ok
capital_distance_edgelist$log_proximity <- 1/(capital_distance_edgelist$log_dist)
1/8.934988 # => 0.1119196,  ok
# head(capital_distance_edgelist)
# capital_distance_edgelist <- subset(capital_distance_edgelist, select = -c(log_proximity) )
capital_distance_edgelist$proximity <- 1/(capital_distance_edgelist$capdist)

capital_distance_edgelist <- subset(capital_distance_edgelist,  select=c(ego, alter, year,proximity))

names(capital_distance_edgelist)[names(capital_distance_edgelist)=="ego"] <- "ego_id"
names(capital_distance_edgelist)[names(capital_distance_edgelist)=="alter"] <- "alter_id"

# - ids should be factor variables
capital_distance_edgelist$ego_id <- as.factor(capital_distance_edgelist$ego_id)
capital_distance_edgelist$alter_id <- as.factor(capital_distance_edgelist$alter_id)


```

## 2. Explanatory Variables

```{r explanatory_vars}

#--- 2. explanatory variables -------


# --- we just use gdp and regime-type in this introduction

# GDP from vdem data
gdp <- read.csv("data/vdem_gdp_pc_INTERPOLATED_consolidated.csv", header = TRUE) # actor attributes
# head(gdp) 

names(gdp)[names(gdp)=="filled_vdem_gdp_interpol"] <- "gdp" # it's just gdp

# regime type from vdem data
regime <- read.csv("data/vdem_regime__INTERPOLATED_consolidated.csv", header = TRUE) # actor attributes
# head(regime) 
regime <- subset(regime, select=c(iso3, year,regime))

#-- merge here ----
covar_base <- subset(gdp, select=c(iso3, year,gdp)) 
# head(covar_base)  

covar_base <- merge(covar_base, regime,by=c("iso3","year")) 
# head(covar_base)

# table(covar_base$year) # - every year on board, for 164 countries
# table(covar_base$iso3) # - every country on board, for 131 years
```

## 3. Wor-Injury Policy

From GWIP ([Breznau and Lanver, 2020](https://doi.org/10.7910/DVN/IVKYIE))

Data are converted into longitudinal format (all years for all countries)

There were some discrepancies with the data so we update that which the team prepared for us

Fill in former Serbian republics at some point (basically as 'Serbia' in the GWIP), for now they need to be dropped

```{r sfb_vars}

# ----- Project A 02 -----------------

# Introduction of work regulations
# we already load this data above to create the 'existence' variable.
# A02_workinj <- read.csv("data/A02_workinj.csv", header = TRUE)



# unique(A02_workinj$cow_code) 165 OK!

# Make sure gwip data is up to date
gwip <- read.csv("data/gwip_v1.csv")
gwip <- select(gwip, -c(country_name))

A02_workinj <- left_join(A02_workinj, gwip, by = "cow_code")

# Drop missing countries
A02_workinj <- subset(A02_workinj, !is.na(labor_workinjury_firstlaw))

# Merge



A02_workinj <- A02_workinj %>%
mutate(labor_workinjury_firstlaw = ifelse(labor_workinjury_firstlaw < 1880, 1880, labor_workinjury_firstlaw), # Start with 1880 for FIRST LAW
       labor_workinjury_first_socins = ifelse(labor_workinjury_first_socins < 1884, 1884, labor_workinjury_first_socins), # Start with 1884 for FIRST SOCINS
       labor_workinjury_first_socins = ifelse(labor_workinjury_first_socins < labor_workinjury_firstlaw_bluecollar_fullcoverage, labor_workinjury_firstlaw_bluecollar_fullcoverage, labor_workinjury_first_socins), # ensure that the social insurance law being coded applies to all blue-collar sectors (measured with error, but best possible approximation combining information from these two variables)
       labor_workinjury_first_socins = ifelse(labor_workinjury_first_socins > 2010, 2010, labor_workinjury_first_socins) # trim to end of series at 2010
       )



covar_ <- data.frame(covar_base)
head(covar_)

A02_2 <- data.frame(subset(  A02_workinj, select = c(iso3,labor_workinjury_firstlaw,labor_workinjury_first_socins,year,pct_ag_01_i)  ))

# here we have to drop 13 countries because we do not have relevant work-injury data
covar_x <- merge(covar_, A02_2,by=c("iso3","year"))
# head(covar_)

# FIRST LAW
names(covar_)[names(covar_)=="labor_workinjury_firstlaw"] <- "toa_workinj_A02"

# FIRST SOC INSURANCE
names(covar_)[names(covar_)=="labor_workinjury_first_socins"] <- "toa_workinj_A02_socins"
# head(covar_)


covar_ <- merge(covar_, existence,by=c("iso3","year")) 
# head(covar_)

covar <- covar_
dim(covar)
unique(covar$iso3)
unique(covar$year)
unique(edgelist_cultural_spheres$year)

# head(covar)
# - combined covar object for all explanatory variables
# - covar$gdp & covar$regime



```

## 4. netdiffuseR setup
  

keep in mind which networks we have
  -     >>edgelists<<: 
  - trade_edgelist                => trade network
  - edgelist_cultural_spheres     => cultural spheres
  - colonial_ties_edgelist        => colonial ties
  - capital_distance_edgelist    => d = distance , prox. = 1/d
  - make netdiffuse object: combine edgelist with covariate object
  - begin with cultural spheres

### FIRST LAW


```{r 4_netdiff_culture, warning = F, message = F}

# cultural spheres --

dim(edgelist_cultural_spheres)
diffnet_culture <- edgelist_to_diffnet(
  edgelist = edgelist_cultural_spheres[,1:2], # As usual, a two column dataset [,1:2]
  w        = edgelist_cultural_spheres$weight,
  t0       = edgelist_cultural_spheres$year,  # An integer vector with starting point of spell
  t1       = edgelist_cultural_spheres$year,  # An integer vector with the endpoint of spell
  undirected = TRUE,        # undirected network
  dat      = covar,         # Attributes dataset
  idvar    = "iso3",                  
  toavar   = "toa_workinj_A02",          # -- !! here just change time-of-adoption !! -- #
  timevar  = "year",
  keep.isolates = TRUE    # Keeping isolates (if there's any)
)
# diffnet_culture
# plot(diffnet_culture)
# summary(diffnet_culture)

# -------------------------------------------------------------------------
# -- 5. get the respective exposure variables and time-of-adoption (toa) f
# -- generate exposure and add to diffnet object
diffnet_culture[["lag_w_expo_culture"]] <- exposure(diffnet_culture, valued = T, lags = 1)#
diffnet_culture[["lag_expo_culture"]] <- exposure(diffnet_culture, valued = F, lags = 1)
diffnet_culture[["adopted"]]  <- toa_mat(diffnet_culture)$cumadopt

table(data.frame(diffnet_culture)$adopted) # don't care about no. adopted here
summary(data.frame(diffnet_culture)$lag_w_expo_culture)
# head(data.frame(diffnet_culture)) # exposure is NA for 1st 164 cases => 1880
# tail(data.frame(diffnet_culture))
```


```{r 4_netdiff_colonial, warning = F, message = F}
# -------------------------
# - 4b) colonial ties -----
# -------------------------
# - here comes the core: creating the diffnet object
#head(colonial_ties_edgelist)
dim(colonial_ties_edgelist)
diffnet_colony <- edgelist_to_diffnet(
  edgelist = colonial_ties_edgelist[,1:2], # As usual, a two column dataset [,1:2]
  w        = colonial_ties_edgelist$weight_decay_exp,
  t0       = colonial_ties_edgelist$year,  # An integer vector with starting point of spell
  t1       = colonial_ties_edgelist$year,  # An integer vector with the endpoint of spell
  undirected = FALSE,        # !!! change for >directed< network !!! => we provide it later
  dat      = covar,         # Attributes dataset
  idvar    = "iso3",                  
  toavar   = "toa_workinj_A02",          # -- !! here just change time-of-adoption !! -- #
  timevar  = "year",
  keep.isolates = TRUE    # Keeping isolates (if there's any)
)
# diffnet_colony
# plot(diffnet_colony)
# summary(diffnet_colony)

# -------------------------------------------------------------------------
# -- 5. get the respective exposure variables and time-of-adoption (toa)
# compute exposure: Netdiffuser automatically identifies whether the input is dynamic or not.
diffnet_colony[["w_expo_colony"]] <- exposure(diffnet_colony, valued = T)
diffnet_colony[["lag_expo_colony"]] <- exposure(diffnet_colony, valued = F)

diffnet_colony[["adopted"]]  <- toa_mat(diffnet_colony)$cumadopt

diffnet_colony[["non_normalized_w_expo_colony"]] <- exposure(diffnet_colony, valued = T, normalized = FALSE)

# diffnet_colony
# summary(diffnet_colony)
summary(data.frame(diffnet_colony)$w_expo_colony)
```


```{r 4_netdiff_trade, warning = F, message = F}
# -------------------------
# - 4c) trade network -----
# -------------------------
# - here comes the core: creating the diffnet object
# head(trade_edgelist)
dim(trade_edgelist)
diffnet_trade <- edgelist_to_diffnet(
  edgelist = trade_edgelist[,1:2], # As usual, a two column dataset [,1:2]
  t0       = trade_edgelist$year,  # An integer vector with starting point of spell
  t1       = trade_edgelist$year,  # An integer vector with the endpoint of spell
  w        = trade_edgelist$log_value,
  undirected = TRUE,        # undirected network
  dat      = covar,         # Attributes dataset
  idvar    = "iso3",                  
  toavar   = "toa_workinj_A02",          # -- !! here just change time-of-adoption !! -- #
  timevar  = "year",
  keep.isolates = TRUE    # Keeping isolates (if there's any)
)
# diffnet_trade
# plot(diffnet_trade)
# summary(diffnet_trade)


# -------------------------------------------------------------------------
# -- 5. get the respective exposure variables and time-of-adoption (toa)
# compute exposure: Netdiffuser automatically identifies whether the input is dynamic or not.
diffnet_trade[["lag_w_expo_trade"]] <- exposure(diffnet_trade, valued = T, lags = 1)
diffnet_trade[["lag_expo_trade"]] <- exposure(diffnet_trade, valued = F, lags = 1)

diffnet_trade[["adopted"]]  <- toa_mat(diffnet_trade)$cumadopt

# head(data.frame(diffnet_trade))
# tail(data.frame(diffnet_trade))
summary(data.frame(diffnet_trade)$lag_w_expo_trade)
```


```{r 4_netdiff_spatial, warning = F, message = F}
# -----------------------------
# - 4d) spatial proximity -----
# -----------------------------
# - here comes the core: creating the diffnet object
# head(capital_distance_edgelist)
dim(capital_distance_edgelist)
diffnet_proximity <- edgelist_to_diffnet(
  edgelist = capital_distance_edgelist[,1:2], # As usual, a two column dataset [,1:2]
  t0       = capital_distance_edgelist$year,  # An integer vector with starting point of spell
  t1       = capital_distance_edgelist$year,  # An integer vector with the endpoint of spell
  w        = capital_distance_edgelist$proximity,
  undirected = TRUE,        # undirected network
  dat      = covar,         # Attributes dataset
  idvar    = "iso3",                  
  toavar   = "toa_workinj_A02",          # -- !! here just change time-of-adoption !! -- #
  timevar  = "year",
  keep.isolates = TRUE    # Keeping isolates (if there's any)
)
# diffnet_proximity

# plot(diffnet_proximity)

 # -----------------------------------------------------------------------
diffnet_proximity[["lag_w_expo_proximity"]] <- exposure(diffnet_proximity, valued = T, lags = 1)

diffnet_proximity[["lag_expo_proximity"]] <- exposure(diffnet_proximity, valued = F, lags = 1)

diffnet_proximity[["adopted"]]  <- toa_mat(diffnet_proximity)$cumadopt

# diffnet_proximity

summary(data.frame(diffnet_proximity)$lag_expo_proximity)
# head(data.frame(diffnet_proximity))
```

```{r 4_culture_adoptiontable_socins, warning = F, message = F}
# - save first diffnet object as data as data frame
diff_data_culture <- as.data.frame(diffnet_culture)
diff_data_culture <- select(diff_data_culture, -c(toa_workinj_A02_socins))

table(diff_data_culture$adopted) # don't care about n. of cases

```


### SOCIAL INSURANCE

```{r 4_netdiff_culture_socins, warning = F, message = F}
# -------------------------
# - 4a) cultural spheres --
# -------------------------

diffnet_culture_socins <- edgelist_to_diffnet(
  edgelist = edgelist_cultural_spheres[,1:2], # As usual, a two column dataset [,1:2]
  w        = edgelist_cultural_spheres$weight,
  t0       = edgelist_cultural_spheres$year,  # An integer vector with starting point of spell
  t1       = edgelist_cultural_spheres$year,  # An integer vector with the endpoint of spell
  undirected = TRUE,        # undirected network
  dat      = covar,         # Attributes dataset
  idvar    = "iso3",                  
  toavar   = "toa_workinj_A02_socins",          
  timevar  = "year",
  keep.isolates = TRUE    # Keeping isolates (if there's any)
)


# -------------------------------------------------------------------------

diffnet_culture_socins[["lag_w_expo_culture"]] <- exposure(diffnet_culture_socins, valued = T, lags = 1)#
diffnet_culture_socins[["lag_expo_culture"]] <- exposure(diffnet_culture_socins, valued = F, lags = 1)
diffnet_culture_socins[["adopted"]]  <- toa_mat(diffnet_culture_socins)$cumadopt

table(data.frame(diffnet_culture_socins)$adopted) # don't care about no. adopted here
summary(data.frame(diffnet_culture_socins)$lag_w_expo_culture)

```


```{r 4_netdiff_colonial, warning = F, message = F}
# -------------------------
# - 4b) colonial ties -----
# -------------------------

diffnet_colony_socins <- edgelist_to_diffnet(
  edgelist = colonial_ties_edgelist[,1:2], # As usual, a two column dataset [,1:2]
  w        = colonial_ties_edgelist$weight_decay_exp,
  t0       = colonial_ties_edgelist$year,  # An integer vector with starting point of spell
  t1       = colonial_ties_edgelist$year,  # An integer vector with the endpoint of spell
  undirected = FALSE,        # !!! change for >directed< network !!! => we provide it later
  dat      = covar,         # Attributes dataset
  idvar    = "iso3",                  
  toavar   = "toa_workinj_A02_socins",          # -- !! here just change time-of-adoption !! -- #
  timevar  = "year",
  keep.isolates = TRUE    # Keeping isolates (if there's any)
)


# -------------------------------------------------------------------------

diffnet_colony_socins[["w_expo_colony"]] <- exposure(diffnet_colony_socins, valued = T)
diffnet_colony_socins[["lag_expo_colony"]] <- exposure(diffnet_colony_socins, valued = F)

diffnet_colony_socins[["adopted"]]  <- toa_mat(diffnet_colony_socins)$cumadopt

diffnet_colony_socins[["non_normalized_w_expo_colony"]] <- exposure(diffnet_colony_socins, valued = T, normalized = FALSE)

summary(data.frame(diffnet_colony_socins)$w_expo_colony)
```


```{r 4_netdiff_trade, warning = F, message = F}
# -------------------------
# - 4c) trade network -----
# -------------------------

diffnet_trade_socins <- edgelist_to_diffnet(
  edgelist = trade_edgelist[,1:2], # As usual, a two column dataset [,1:2]
  t0       = trade_edgelist$year,  # An integer vector with starting point of spell
  t1       = trade_edgelist$year,  # An integer vector with the endpoint of spell
  w        = trade_edgelist$log_value,
  undirected = TRUE,        # undirected network
  dat      = covar,         # Attributes dataset
  idvar    = "iso3",                  
  toavar   = "toa_workinj_A02_socins",          # -- !! here just change time-of-adoption !! -- #
  timevar  = "year",
  keep.isolates = TRUE    # Keeping isolates (if there's any)
)



# -------------------------------------------------------------------------

diffnet_trade_socins[["lag_w_expo_trade"]] <- exposure(diffnet_trade_socins, valued = T, lags = 1)
diffnet_trade_socins[["lag_expo_trade"]] <- exposure(diffnet_trade_socins, valued = F, lags = 1)

diffnet_trade_socins[["adopted"]]  <- toa_mat(diffnet_trade_socins)$cumadopt


summary(data.frame(diffnet_trade)$lag_w_expo_trade)
```


```{r 4_netdiff_spatial, warning = F, message = F}
# -----------------------------
# - 4d) spatial proximity -----
# -----------------------------
# - here comes the core: creating the diffnet object
# head(capital_distance_edgelist)
dim(capital_distance_edgelist)
diffnet_proximity_socins <- edgelist_to_diffnet(
  edgelist = capital_distance_edgelist[,1:2], # As usual, a two column dataset [,1:2]
  t0       = capital_distance_edgelist$year,  # An integer vector with starting point of spell
  t1       = capital_distance_edgelist$year,  # An integer vector with the endpoint of spell
  w        = capital_distance_edgelist$proximity,
  undirected = TRUE,        # undirected network
  dat      = covar,         # Attributes dataset
  idvar    = "iso3",                  
  toavar   = "toa_workinj_A02_socins",          # -- !! here just change time-of-adoption !! -- #
  timevar  = "year",
  keep.isolates = TRUE    # Keeping isolates (if there's any)
)
# diffnet_proximity

# plot(diffnet_proximity)


 # -----------------------------------------------------------------------
diffnet_proximity_socins[["lag_w_expo_proximity"]] <- exposure(diffnet_proximity_socins, valued = T, lags = 1)

diffnet_proximity_socins[["lag_expo_proximity"]] <- exposure(diffnet_proximity_socins, valued = F, lags = 1)

diffnet_proximity_socins[["adopted"]]  <- toa_mat(diffnet_proximity_socins)$cumadopt

# diffnet_proximity

summary(data.frame(diffnet_proximity_socins)$lag_expo_proximity)
# head(data.frame(diffnet_proximity))
```


```{r 4_culture_adoptiontable_socins, warning = F, message = F}
# - save first diffnet object as data as data frame
diff_data_culture_socins <- as.data.frame(diffnet_culture_socins)
diff_data_culture_socins <- select(diff_data_culture_socins, -c(toa_workinj_A02))

table(diff_data_culture_socins$adopted) # don't care about n. of cases

```



## 5. Combine Data

### Time Dummies

```{r six}
# -----------------------------------------------------------------
# -----------------------------------------------------------------
# - 6. create process-time control, define the duration of episode
# -----------------------------------------------------------------

# FIRST LAW
diff_data_culture <- diff_data_culture %>%
mutate(t = per - 1880,
       t0_22 = ifelse(t < 23, 1, 0), # stage one Europe, emergence (1880-1902)
       t23_48 = ifelse(t >= 23 & t < 50, 1, 0), # 2 Russian/Soviet Revolution + WWI (1903-1928)
       t49_74 = ifelse(t >= 50 & t < 75, 1, 0), # WW2 + Marshall Plan (1929-1954)
       t75_99 = ifelse(t >= 75 & t < 100, 1, 0),
       t100_130 = ifelse(t >= 100, 1, 0))


#SOC INS
diff_data_culture_socins <- diff_data_culture_socins %>%
mutate(t = per - 1880,
       t0_22 = ifelse(t < 23, 1, 0), # stage one Europe, emergence (1880-1902)
       t23_48 = ifelse(t >= 23 & t < 50, 1, 0), # 2 Russian/Soviet Revolution + WWI (1903-1928)
       t49_74 = ifelse(t >= 50 & t < 75, 1, 0), # WW2 + Marshall Plan (1929-1955)
       t75_99 = ifelse(t >= 75 & t < 100, 1, 0),
       t100_130 = ifelse(t >= 100, 1, 0))
```

### Merge

```{r seven}
# -----------------------------------------------------------

diff_data <- diff_data_culture
diff_data_socins <- diff_data_culture_socins

diffnet_colony_df <- data.frame(diffnet_colony)
diffnet_trade_df <- data.frame(diffnet_trade)
diffnet_proximity_df <- data.frame(diffnet_proximity)
diffnet_culture_df <- data.frame(diffnet_culture)

diffnet_colony_df_socins <- data.frame(diffnet_colony_socins)
diffnet_trade_df_socins <- data.frame(diffnet_trade_socins)
diffnet_proximity_df_socins <- data.frame(diffnet_proximity_socins)
diffnet_culture_df_socins <- data.frame(diffnet_culture_socins)


# -------------------------------------------------------
# -------------- merge everything together ---------------
# -------------------------------------------------------
# --- merge id is now "id",  automatically generated in netdiffuseR objects

# FIRST LAW
d <- subset(  diffnet_colony_df, select = c(id,per,w_expo_colony, lag_expo_colony, non_normalized_w_expo_colony))
diff_data <- merge(diff_data, d,by=c("id","per")) 
d <- subset(  diffnet_trade_df, select = c(id,per,lag_w_expo_trade, lag_expo_trade)  )
diff_data <- merge(diff_data, d,by=c("id","per")) 
d <- subset(  diffnet_proximity_df, select = c(id,per,lag_w_expo_proximity, lag_expo_proximity)  )
diff_data <- merge(diff_data, d,by=c("id","per")) 

# SOC INS
d <- subset(  diffnet_colony_df_socins, select = c(id,per,w_expo_colony, lag_expo_colony, non_normalized_w_expo_colony)  )
diff_data_socins <- merge(diff_data_socins, d,by=c("id","per")) 
d <- subset(  diffnet_trade_df_socins, select = c(id,per,lag_w_expo_trade, lag_expo_trade)  )
diff_data_socins <- merge(diff_data_socins, d,by=c("id","per")) 
d <- subset(  diffnet_proximity_df_socins, select = c(id,per,lag_w_expo_proximity, lag_expo_proximity)  )
diff_data_socins <- merge(diff_data_socins, d,by=c("id","per")) 

# fix GDP
diff_data$gdp10000 <- diff_data$gdp/10000 # in 10,000 USD
diff_data_socins$gdp10000 <- diff_data_socins$gdp/10000 # in 10,000 USD
```

Original datafile source

http://www.barkhof.uni-bremen.de/~mwindzio/clusterID_n.csv


```{r cluster_id}
cluster_id <- read.csv("data/clusterID_n.csv", header = TRUE)
# head(cluster_id) 
cluster_id <- cluster_id[,c(1,2,5)]

names(cluster_id) <- c("id", "per", "cluster_id")
diff_data <- merge(diff_data, cluster_id,by=c("id","per"))
diff_data_socins <- merge(diff_data_socins, cluster_id,by=c("id","per"))

```

## 6. Summary Stats

### Threshold Adoption comparison

```{r adoption plot, warning = F, message = F}
plot1 <- classify(diffnet_trade, include_censored = TRUE)
ftable(plot1)

plot2 <- classify(diffnet_colony, include_censored = TRUE)
ftable(plot2)

```

### Network Distributions

```{r summarydist}
# - distribution of network, range and "density/edge weight" => 164 NAs for 1880
summary(diff_data$lag_w_expo_culture)
#summary(diff_data$w_expo_colony)
#summary(diff_data$lag_w_expo_trade)
#summary(diff_data$lag_w_expo_proximity)

#summary(diff_data$adopted) # - create new for each project specific-toa


```

```{r summarydist_socins}
summary(diff_data_socins$lag_w_expo_culture)
```


## 7. Save Data
```{r savepoint}
# drop missing cases
diff_data <- diff_data[diff_data$existence != "-Inf",]

diff_data_socins <- diff_data_socins[diff_data_socins$existence != "-Inf",]

rm(trade_dyadic, trade_dyadic_consolidated, trade_edgelist, diff_data_culture, capital_distance_edgelist, d, ddat, edgelist_cultural_spheres, existence, gdp, regime, colonial_ties_edgelist)

save.image(file = "data/.Rdata")
```