Data_Analysis_H1_H2.Rmd

---
title: "Data_Analysis_H1_H2"
author: "Susanne_Rehorst"
date: "25-5-2022"
output: html_document
---

# Data Analysis H1 & H2
In the documents Data_Processing_Regulations.Rmd and Data_Processing_Metrics.Rmd, the data was cleaned and processed. In this markdown, analysis will be performed on this processed data for Hypothesis 1 and Hypothesis 2 (quantity, readability and linguistic characteristics).

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```

Load or install packages:
```{r}
library(dplyr)
library(stringr)
library(ggplot2)
library(ggpubr)
```

Set Working Directory
```{r}
setwd("~/Thesis")
```

## Load in the data
There are 3 data files used for this analysis. The first one is Descriptive_Data.RData, which contains the meta data for each regulation document and descriptive statistics, like the amount of words and the readability of the regulation document. The second data file that is used is Processed_Data.RData. This data file contains the regulation data, split into tokens. The third data file that is used is Metrics_Data.RData. This file contains the text describing ESG metrics. This data is in a tokenized data format. The metrics data is used to benchmark the content of the regulations document (measuring how many ESG topics are covered by the regulations).
```{r}
load("Descriptive_Data.RData")
load("Processed_Data.RData")
load("Metrics_Data.RData")
```

# Hypothesis 1: Quantity
The first hypothesis tests the differences in quantity and readability between mandatory and voluntary regulatory documents. 

# Quantity

## Merging total words
First of all, Descriptive_Data.RData is used to inspect the quantity of words and sentences. Especially ,the quantity of words needs to be adjusted, as the program qualifies some words as non-words. This is not correct, but this is caused by very long sentences without punctuation ending marks (e.g., periods, question mark). To fix this, a merge is done on the words and non-words. 
```{r}
Descriptive_Data$totalwords <- rowSums(Descriptive_Data[,c(4,5)])
```

## Merging on document and group level
In order to compare regulations documents, there is merging on 2 levels. Level 1 merges documents that belong to the same region, have the same compliance (Mandatory or Voluntary) and apply to the same topic (E, S, G or a combination thereof) and refer to each other (by amendment or indirectly in the text). By merging them, this controls for the fact that in some regions, regulations of the same type and that refer to each other are scattered across multiple documents. This would imply that the quantity is lower while this is actually not the case. 

Level 2 Merges even further, by merging level 1 documents that already apply to the same region, have the same compliance, refer to each other, REGARDLESS if they have the same topic. Referring means that document refer to each other because one regulation amends the other OR if one documents indirectly refers to the other document in the text. By also looking at group level, there can be controlled for regions that show their ESG regulations in 1 document and regions that publish their E, S, G regulations at different sources, even though they refer to each other and should be used in conjunction with each other. 

## Level 1 merging
In order to merge on level 1, a merge ID variable is created to uniquely identify groups.
```{r}
# Create the merge ID variable
names <- c("GROUP_ID","DOC_TYPE","M")
Descriptive_Data[,names] <- lapply(Descriptive_Data[,names], as.character)
Descriptive_Data$MERGE_ID <- paste(Descriptive_Data$GROUP_TYPE, Descriptive_Data$GROUP_ID,Descriptive_Data$DOC_TYPE,Descriptive_Data$M)

# Set variables back to factor
names <- c("MERGE_ID","DOC_ID","GROUP_ID","GROUP_TYPE","DOC_TYPE","M","REGION")
Descriptive_Data[,names] <- lapply(Descriptive_Data[,names], factor)

# Get merged data by grouping on MERGE_ID
DOC_Quantity <- Descriptive_Data %>% group_by(MERGE_ID) %>%
  summarize(totalwords=sum(totalwords),
            sents=sum(sents))

# Split the MERGE_ID variable
DOC_Quantity[c('GROUP_TYPE', 'GROUP_ID', 'DOC_TYPE', 'M')] <- str_split_fixed(DOC_Quantity$MERGE_ID,' ',4)

# Create new merge variable
DOC_Quantity$MERGE_ID <- paste(DOC_Quantity$DOC_TYPE,DOC_Quantity$M)

# Set variables to factor
names <- c("MERGE_ID","GROUP_TYPE", "GROUP_ID","DOC_TYPE","M")
DOC_Quantity[,names] <- lapply(DOC_Quantity[,names], factor)

# Remove Group ID 42, 43, 44, as they do not contain ESG regulations (only key ESG metrics)
DOC_Quantity <- DOC_Quantity[DOC_Quantity$GROUP_ID!="42" & DOC_Quantity$GROUP_ID!="43" & DOC_Quantity$GROUP_ID!="44", ]
DOC_Quantity <- droplevels(DOC_Quantity)
```

Get the summary statistics for the aggregated data set. These summary statistics show the average length of mandatory and voluntary documents, split by group type. Of course, ESG should only be compared to ESG as it is likely that these documents are larger. 
```{r}
DOC_Stats <- DOC_Quantity %>%                              
  group_by(MERGE_ID) %>% 
  summarize(N=length(MERGE_ID),
            w_min = min(totalwords),
            w_q1 = quantile(totalwords, 0.25),
            w_median = median(totalwords),
            w_mean = mean(totalwords),
            w_q3 = quantile(totalwords, 0.75),
            w_max = max(totalwords),
            w_SD=sd(totalwords),
            s_min = min(sents),
            s_q1 = quantile(sents, 0.25),
            s_median = median(sents),
            s_mean = mean(sents),
            s_q3 = quantile(sents, 0.75),
            s_max = max(sents),
            s_SD=sd(sents))

DOC_Stats
```

## Level 2 merging
Next, analysis is performed on group level. These summary statistics give more insights into group level. The average amount of words and sentences can give an indication if there is a difference between the amount of mandatory and voluntary regulations.

```{r}
# Group the data by Group_ID
Group_Quantity <- Descriptive_Data %>%
  group_by(GROUP_ID, GROUP_TYPE, M) %>%
  summarize(totalwords=sum(totalwords),
            sentences=sum(sents)) %>% ungroup()

# Remove Group ID 42, 43, 44, as they do not contain ESG regulations (only key ESG metrics)
Group_Quantity <- Group_Quantity[Group_Quantity$GROUP_ID!="42" & Group_Quantity$GROUP_ID!="43" & Group_Quantity$GROUP_ID!="44", ]
Group_Quantity <- droplevels(Group_Quantity)

# Get summary statistics 
Group_Stats <- Group_Quantity %>%                              
  group_by(GROUP_TYPE, M) %>% 
  summarize(
            w_min = min(totalwords),
            w_q1 = quantile(totalwords, 0.25),
            w_median = median(totalwords),
            w_mean = mean(totalwords),
            w_q3 = quantile(totalwords, 0.75),
            w_max = max(totalwords),
            w_SD=sd(totalwords),
            s_min = min(sentences),
            s_q1 = quantile(sentences, 0.25),
            s_median = median(sentences),
            s_mean = mean(sentences),
            s_q3 = quantile(sentences, 0.75),
            s_max = max(sentences),
            s_SD=sd(sentences))

Group_Stats
```

Besides, it is calculated what percentage of ESG group documents is made out of pure governance documents by combining information from DOC_Quantity and GROUP_Quantity. Initially, corporate governance regulations were the first ones that were introduced. So it is interesting to see if they still have a large weight in the current ESG regulations. Also, the country IDs are merged with the descriptive data, to examine differences on a country level. 
```{r}
# Create GROUP_ID, REGION identifier
Country_ID <- Descriptive_Data[,c(50,59)]
Country_ID <- distinct(Country_ID)

DOC_Quantity <- merge(DOC_Quantity, Country_ID, by.x = "GROUP_ID", by.y = "GROUP_ID", all.x = TRUE, all.y = FALSE)
Group_Quantity <- merge(Group_Quantity, Country_ID, by.x = "GROUP_ID", by.y = "GROUP_ID", all.x = TRUE, all.y = FALSE)
```

```{r}
# Calculate percentage of G documents in ESG groups, by region
G_Percentage <- DOC_Quantity %>% filter(M=="1" & GROUP_TYPE=="ESG") %>% group_by(GROUP_ID) %>% mutate(
  stotalwords=sum(totalwords),
  p_wgovernance=ifelse(DOC_TYPE=="G",(totalwords/stotalwords),0),
  stotalsents=sum(sents),
  p_sgovernance=ifelse(DOC_TYPE=="G",(sents/stotalsents),0),
) 

# Create summary table, by Region by doc type
Q_Region_DOC <- DOC_Quantity %>% group_by(REGION, DOC_TYPE) %>% summarise(
   w_total = sum(totalwords),
   s_toal = sum(sents))

# Create summary table, by Region by group type
Q_Region_GROUP <- Group_Quantity %>% group_by(REGION, GROUP_TYPE, M) %>% summarise(
   w_total = sum(totalwords),
   s_toal = sum(sentences))

G_Percentage
Q_Region_DOC
Q_Region_GROUP

```


## Test if differences in quantity are significant between mandatory and voluntary regulations
Test if differences in quantity are significant between mandatory and voluntary regulations. Also, log of totalwords and sentences is taken to create 2 new variables. This is done as a check, in case the quantity of words and sentences is heavily distributed to the  right. An indication of this can already be given if the median is lower than the mean quantity. Another way to check this is with a histogram. 
```{r}
# Plots overall
hist(Group_Quantity$totalwords)
hist(Group_Quantity$sentences)
plot(density(Group_Quantity$totalwords))
plot(density(Group_Quantity$sentences))

# Plots comparing distribution of mandatory vs voluntary
plot1<- ggplot(DOC_Quantity, aes(x = sents, fill = M)) + 
  geom_density(alpha = 0.5) +
  theme(text=element_text(size=12,  family="serif"))

plot2 <- ggplot(DOC_Quantity, aes(x = totalwords, fill = M)) + 
  geom_density(alpha = 0.5) +
  theme(text=element_text(size=12,  family="serif"))

plot3<- ggplot(ESG_Group_Quantity, aes(x = log(sentences), fill=M)) + 
  geom_density(alpha = 0.5) + 
  labs(y= "density", x = "number of sentences") +
  theme(text=element_text(size=15,  family="serif")) +
  ggtitle("4.1a Sentence distribution ESG reporting standards")

plot4 <- ggplot(ESG_Group_Quantity, aes(x =log(totalwords), fill=M)) + 
  geom_density(alpha = 0.5) +
  labs(y= "density", x = "number of words") +
  theme(text=element_text(size=15,  family="serif")) + 
  ggtitle("4.1b Word distribution ESG reporting standards")

ggarrange(plot1, plot2, plot3, plot4)
ggarrange(plot3, plot4)
```

Regression models are used to see if difference in average quantity between mandatory and voluntary documents is significant. This is done on a group level, as groups are the least influenced by scattering of documents across different sources. The first 4 models look at document level and control for the different doc type. The models 5 to 8 look at group level and control for group type. The models 9 to 12 look at group level and do a subset analysis to only compare mandatory and voluntary ESG. 
```{r}
# Create log transformed variables
Group_Quantity$logwords <- log(Group_Quantity$totalwords)
Group_Quantity$logsents <- log(Group_Quantity$sentences)
DOC_Quantity$logwords <- log(DOC_Quantity$totalwords)
DOC_Quantity$logsents <- log(DOC_Quantity$sents)

# Significance between M and V (doc level)
model1 <- lm(totalwords ~ M+DOC_TYPE, data = DOC_Quantity)
model2 <- lm(sents ~ M+DOC_TYPE, data = DOC_Quantity)

# Log version
model3 <- lm(logwords ~ M+DOC_TYPE, data = DOC_Quantity)
model4 <- lm(logsents ~ M+DOC_TYPE, data = DOC_Quantity)

# Significance between M and V (group level)
model5 <- lm(totalwords ~ M+GROUP_TYPE, data = Group_Quantity)
model6 <- lm(sentences ~ M+GROUP_TYPE, data = Group_Quantity)

# Log version
model7 <- lm(logwords ~ M+GROUP_TYPE, data = Group_Quantity)
model8 <- lm(logsents ~ M+GROUP_TYPE, data = Group_Quantity)

# Significance between M and V ESG regulations

# Create a sub sample
ESG_Group_Quantity <- Group_Quantity[Group_Quantity$GROUP_TYPE=="ESG", ]

# Significance between M and V (group level)
model9 <- lm(totalwords ~ M, data = ESG_Group_Quantity)
model10 <- lm(sentences ~ M, data = ESG_Group_Quantity)

# Log version
model11 <- lm(logwords ~ M, data = ESG_Group_Quantity)
model12 <- lm(logsents ~ M, data = ESG_Group_Quantity)

# Results
summary(model1)
summary(model2)
summary(model3)
summary(model4)
summary(model5)
summary(model6)
summary(model7)
summary(model8)
summary(model9)
summary(model10)
summary(model11)
summary(model12)
```

# H2a Linguistic characteristics: Readability

## Correct for non-words
This section analyzes the differences in readability score between groups. First of all, there are 7 out of of 70 cases removed where the amount of non-words > 0. When non-words >0, this is an indication that the software was not able to correctly identify sentences and words. This is due to long lists where terms are summed up, but do not contain a punctuation ending mark. As a consequence, readability scores use the wrong amount of words and sentences and are therefore not correct.  
```{r}
Readability <- Descriptive_Data[Descriptive_Data$GROUP_ID!="2"& Descriptive_Data$GROUP_ID!="4"& Descriptive_Data$GROUP_ID!="17"& Descriptive_Data$GROUP_ID!="20"& Descriptive_Data$GROUP_ID!="21"& Descriptive_Data$GROUP_ID!="23"& Descriptive_Data$GROUP_ID!="31", ]

Readability <- Readability[Readability$GROUP_ID!="42" & Readability$GROUP_ID!="43" & Readability$GROUP_ID!="44", ]
```

## Average readability between documents: Mandatory vs Voluntary documents
First, average readability between mandatory and voluntary documents is examined, irrespective of ESG type. For Quantity, this could not be done, because ESG automatically contains more words or sentences than an E or G document. But readability formulas are not influenced by sizes. For this analysis, 4 formulas are utilized. Each formula has its own characteristics. As a robustness check, it is examined if results are similar across different readability formulas.  
```{r}
Readability_Total <- Readability %>%
  group_by(M) %>% 
  summarize(average_gl=mean(gl),
            average_ari=mean(ari),
            average_smog=mean(smog),
            average_cl=mean(cl),
            med_gl=median(gl),
            med_ari=median(ari),
            med_smog=median(smog),
            med_cl=median(cl))%>% ungroup()

Readability_Total
```

The difference in readability score between mandatory and voluntary documents is tested on significance. Also, the distribution is inspected to see if log transformation is needed. 
```{r}
# Plots comparing distribution of mandatory vs voluntary
plot1<- ggplot(Readability, aes(x = gl, fill = M)) + 
  geom_density(alpha = 0.5) + labs(y= "density", x = "FK Readability Score") + ggtitle("4.2a Distribution FK Score - Group Level 1")
  theme(text=element_text(size=12,  family="serif"))

plot2 <- ggplot(Readability, aes(x = ari, fill = M)) + 
  geom_density(alpha = 0.5) + labs(y= "density", x = "AR Readability Index") + ggtitle("4.2b Distribution AR Index - Group Level 1")
  theme(text=element_text(size=12,  family="serif"))

plot3 <- ggplot(Readability, aes(x = cl, fill = M)) + 
  geom_density(alpha = 0.5) + labs(y= "density", x = "CL Readability Index") + ggtitle("4.2c Distribution CL Index - Group Level 1")
  theme(text=element_text(size=12,  family="serif"))

plot4<- ggplot(Group_Readability, aes(x = gl, fill = M)) + 
  geom_density(alpha = 0.5) + labs(y= "density", x = "FK Readability Score") + ggtitle("4.2d Distribution FK Score - Group Level 2")
  theme(text=element_text(size=12,  family="serif"))

plot5 <- ggplot(Group_Readability, aes(x = ari, fill = M)) + 
  geom_density(alpha = 0.5) + labs(y= "density", x = "AR Readability Index") + ggtitle("4.2e Distribution AR Index - Group Level 2")
  theme(text=element_text(size=12,  family="serif"))

plot6 <- ggplot(Group_Readability, aes(x = cl, fill = M)) + 
  geom_density(alpha = 0.5) + labs(y= "density", x = "CL Readability Index") + ggtitle("4.2f Distribution CL Index - Group Level 2")
  theme(text=element_text(size=12,  family="serif"))

plot11<- ggplot(Readability, aes(x = log(gl), fill = M)) + 
  geom_density(alpha = 0.5) + labs(y= "density", x = "FK Readability Score") + ggtitle("4.2a Distribution FK Score - Group Level 1")
  theme(text=element_text(size=12,  family="serif"))

plot21 <- ggplot(Readability, aes(x = log(ari), fill = M)) + 
  geom_density(alpha = 0.5) + labs(y= "density", x = "AR Readability Index") + ggtitle("4.2b Distribution AR Index - Group Level 1")
  theme(text=element_text(size=12,  family="serif"))

plot31 <- ggplot(Readability, aes(x = log(cl), fill = M)) + 
  geom_density(alpha = 0.5) + labs(y= "density", x = "CL Readability Index") + ggtitle("4.2c Distribution CL Index - Group Level 1")
  theme(text=element_text(size=12,  family="serif"))

plot41<- ggplot(Group_Readability, aes(x = log(gl), fill = M)) + 
  geom_density(alpha = 0.5) + labs(y= "density", x = "FK Readability Score") + ggtitle("4.2d Distribution FK Score - Group Level 2")
  theme(text=element_text(size=12,  family="serif"))

plot51 <- ggplot(Group_Readability, aes(x = log(ari), fill = M)) + 
  geom_density(alpha = 0.5) + labs(y= "density", x = "AR Readability Index") + ggtitle("4.2e Distribution AR Index - Group Level 2")
  theme(text=element_text(size=12,  family="serif"))

plot61 <- ggplot(Group_Readability, aes(x = log(cl), fill = M)) + 
  geom_density(alpha = 0.5) + labs(y= "density", x = "CL Readability Index") + ggtitle("4.2f Distribution CL Index - Group Level 2")
  theme(text=element_text(size=12,  family="serif"))
  
  
ggarrange(plot1, plot2, plot3, plot4, plot5, plot6)
ggarrange(plot11, plot21, plot31, plot41, plot51,plot61)
```

Regression models: 
```{r}
# Log transform variables
Readability$log_gl <- log(Readability$gl)
Readability$log_ari <- log(Readability$ari)
Readability$log_smog <- log(Readability$smog)
Readability$log_cl <- log(Readability$cl)

# Test for significance
model10 <- lm(gl~M, data = Readability)
model11 <- lm(log_gl~M, data = Readability)

model12 <- lm(ari~M, data = Readability)
model13 <- lm(log_ari~M, data = Readability)

model14 <- lm(smog~M, data = Readability)
model15 <- lm(log_smog~M, data = Readability)

model16 <- lm(cl~M, data = Readability)
model17 <- lm(log_cl~M, data = Readability)

summary(model10)
summary(model11)
summary(model12)
summary(model13)
summary(model14)
summary(model15)
summary(model16)
summary(model17)
```

## Average readability between documents: Mandatory or Voluntary, subset by DOC_TYPE (doc level)
After examining the average readability score between mandatory and voluntary, this difference is examined in more detail by making a distinction in DOC_TYPE. So, if it is an ESG, G, or E document.  
```{r}
Readability_DOC <- Readability %>%
  group_by(DOC_TYPE, M) %>% 
  summarize(average_gl=mean(gl),
            average_ari=mean(ari),
            average_smog=mean(smog),
            average_cl=mean(cl),
            med_gl=median(gl),
            med_ari=median(ari),
            med_smog=median(smog),
            med_cl=median(cl))%>% ungroup()

Readability_DOC
```

Again, the significance between certain groups on doc level is examined. The differences that are examined are ESG(M)-ESG(V), ESG(M)-G(M), ESG(V)-E(V).
```{r}
# Log transform variables
Readability$log_gl <- log(Readability$gl)
Readability$log_ari <- log(Readability$ari)
Readability$log_smog <- log(Readability$smog)
Readability$log_cl <- log(Readability$cl)

# Create sub-samples
ESG_Readability_DOC <- Readability[Readability$DOC_TYPE=="ESG", ]
M_Readability_DOC <- Readability[Readability$M=="1" & Readability$DOC_TYPE=="ESG" | Readability$M=="1" & Readability$DOC_TYPE=="G", ]
V_Readability_DOC <- Readability[Readability$M=="0" & Readability$DOC_TYPE=="ESG" | Readability$M=="0" & Readability$DOC_TYPE=="E", ]

# Test for significance - ESG Mandatory vs ESG Voluntary
model101 <- lm(gl~M, data = ESG_Readability_DOC)
model111 <- lm(log_gl~M, data = ESG_Readability_DOC)

model121 <- lm(ari~M, data = ESG_Readability_DOC)
model131 <- lm(log_ari~M, data = ESG_Readability_DOC)

model141 <- lm(smog~M, data = ESG_Readability_DOC)
model151 <- lm(log_smog~M, data = ESG_Readability_DOC)

model161 <- lm(cl~M, data = ESG_Readability_DOC)
model171 <- lm(log_cl~M, data = ESG_Readability_DOC)

summary(model101)
summary(model111)
summary(model121)
summary(model131)
summary(model141)
summary(model151)
summary(model161)
summary(model171)

# Test for significance - ESG Mandatory vs Governance Mandatory
model102 <- lm(gl~DOC_TYPE, data = M_Readability_DOC)
model112 <- lm(log_gl~DOC_TYPE, data = M_Readability_DOC)

model122 <- lm(ari~DOC_TYPE, data = M_Readability_DOC)
model132 <- lm(log_ari~DOC_TYPE, data = M_Readability_DOC)

model142 <- lm(smog~DOC_TYPE, data = M_Readability_DOC)
model152 <- lm(log_smog~DOC_TYPE, data = M_Readability_DOC)

model162 <- lm(cl~DOC_TYPE, data = M_Readability_DOC)
model172 <- lm(log_cl~DOC_TYPE, data = M_Readability_DOC)

summary(model102)
summary(model112)
summary(model122)
summary(model132)
summary(model142)
summary(model152)
summary(model162)
summary(model172)


# Test for significance - ESG Voluntary vs Environmental Voluntary
model103 <- lm(gl~DOC_TYPE, data = V_Readability_DOC)
model113 <- lm(log_gl~DOC_TYPE, data = V_Readability_DOC)

model123 <- lm(ari~DOC_TYPE, data = V_Readability_DOC)
model133 <- lm(log_ari~DOC_TYPE, data = V_Readability_DOC)

model143 <- lm(smog~DOC_TYPE, data = V_Readability_DOC)
model153 <- lm(log_smog~DOC_TYPE, data = V_Readability_DOC)

model163 <- lm(cl~DOC_TYPE, data = V_Readability_DOC)
model173 <- lm(log_cl~DOC_TYPE, data = V_Readability_DOC)

summary(model103)
summary(model113)
summary(model123)
summary(model133)
summary(model143)
summary(model153)
summary(model163)
summary(model173)
```

## Average Readability between groups: Mandatory vs Voluntary, subset by Group Type (group level)
As a final part of the analysis, the differences between mandatory and voluntary are examined, while sub-setting at group type. The concept of grouping documents is already explained in the Quantity analysis part of this document. By looking at average readability within a group, the overall readability of documents within the same group is examined. So, a conclusion can be drawn on how readability is overall for a company when they need to apply to regulations applying to a certain region. Therefore, this group analysis also allows to draw conclusions on regional differences.  
```{r}
# Group the data by Group_ID
Group_Readability <- Readability %>%
  group_by(GROUP_ID, GROUP_TYPE, M) %>%
  summarize(gl=mean(gl),
            ari=mean(ari),
            smog=mean(smog),
            cl=mean(cl))%>% ungroup()

# Remove Group ID 42, 43, 44, as they do not contain ESG regulations (only key ESG metrics)
Group_Readability <- Group_Readability[Group_Readability$GROUP_ID!="42" & Group_Readability$GROUP_ID!="43" & Group_Readability$GROUP_ID!="44", ]
Group_Readability <- droplevels(Group_Readability)

# Get summary statistics by M
Group_R_Stats_total <- Group_Readability %>%                              
  group_by(M) %>% 
  summarize(med_gl=median(gl),
            med_ari=median(ari),
            med_smog=median(smog),
            med_cl=median(cl),
            average_gl=mean(gl),
            average_ari=mean(ari),
            average_smog=mean(smog),
            average_cl=mean(cl)) %>% ungroup()

Group_R_Stats_total

# Get summary statistics by M and Group Type
Group_R_Stats <- Group_Readability %>%                              
  group_by(GROUP_TYPE, M) %>% 
  summarize(med_gl=median(gl),
            med_ari=median(ari),
            med_smog=median(smog),
            med_cl=median(cl),
            average_gl=mean(gl),
            average_ari=mean(ari),
            average_smog=mean(smog),
            average_cl=mean(cl)) %>% ungroup()

Group_R_Stats
```

The region is added to the results on group level. This allows to examine if differences are driven by regional differences.  
```{r}
Group_Readability <- merge(Group_Readability, Country_ID, by.x = "GROUP_ID", by.y = "GROUP_ID", all.x = TRUE, all.y = FALSE)
```

```{r}
# Create summary table, by Region by doc type
Readability_Region_DOC <- Readability %>% group_by(REGION, DOC_TYPE) %>% summarise(
            average_gl=mean(gl),
            average_ari=mean(ari),
            average_smog=mean(smog),
            average_cl=mean(cl))

# Create summary table, by Region by group type
Readability_Region_GROUP <- Group_Readability %>% group_by(REGION, GROUP_TYPE, M) %>% summarise(
            average_gl=mean(gl),
            average_ari=mean(ari),
            average_smog=mean(smog),
            average_cl=mean(cl))

# Create summary table, by Region by M
Readability_Region <- Group_Readability %>% group_by(REGION, M) %>% summarise(
            average_gl=mean(gl),
            average_ari=mean(ari),
            average_smog=mean(smog),
            average_cl=mean(cl))

Readability_Region_DOC
Readability_Region_GROUP
Readability_Region

```

Lastly, the significance is tested on a group level for readability scores. Difference is examined between mandatory and voluntary. Also, a sub-set analysis is done to test if differences are significant between certain groups. 
```{r}
# Log transform variables
Group_Readability$log_gl <- log(Group_Readability$gl)
Group_Readability$log_ari <- log(Group_Readability$ari)
Group_Readability$log_smog <- log(Group_Readability$smog)
Group_Readability$log_cl <- log(Group_Readability$cl)

# Test for significance - M vs V
model1011 <- lm(gl~M, data = Group_Readability)
model1111 <- lm(log_gl~M, data = Group_Readability)

model1211 <- lm(ari~M, data = Group_Readability)
model1311 <- lm(log_ari~M, data = Group_Readability)

model1411 <- lm(smog~M, data = Group_Readability)
model1511 <- lm(log_smog~M, data = Group_Readability)

model1611 <- lm(cl~M, data = Group_Readability)
model1711 <- lm(log_cl~M, data = Group_Readability)

summary(model1011)
summary(model1111)
summary(model1211)
summary(model1311)
summary(model1411)
summary(model1511)
summary(model1611)
summary(model1711)

# Create sub sample
ESG_Group_Readability <- Group_Readability[Group_Readability$GROUP_TYPE=="ESG", ]
M_Group_Readability <- Group_Readability[Group_Readability$M=="1" & Group_Readability$GROUP_TYPE=="ESG" | Group_Readability$M=="1" & Group_Readability$GROUP_TYPE=="G", ]
V_Group_Readability <- Group_Readability[Group_Readability$M=="0" & Group_Readability$GROUP_TYPE=="ESG" | Group_Readability$M=="0" & Group_Readability$GROUP_TYPE=="E", ]

# Test for significance - ESG M vs ESG V
modelx101 <- lm(gl~M, data = ESG_Group_Readability)
modelx111 <- lm(log_gl~M, data = ESG_Group_Readability)

modelx121 <- lm(ari~M, data = ESG_Group_Readability)
modelx131 <- lm(log_ari~M, data = ESG_Group_Readability)

modelx141 <- lm(smog~M, data = ESG_Group_Readability)
modelx151 <- lm(log_smog~M, data = ESG_Group_Readability)

modelx161 <- lm(cl~M, data = ESG_Group_Readability)
modelx171 <- lm(log_cl~M, data = ESG_Group_Readability)

summary(modelx101)
summary(modelx111)
summary(modelx121)
summary(modelx131)
summary(modelx141)
summary(modelx151)
summary(modelx161)
summary(modelx171)

# Test for significance - ESG M vs G M
modelx102 <- lm(gl~GROUP_TYPE, data = M_Group_Readability)
modelx112 <- lm(log_gl~GROUP_TYPE, data = M_Group_Readability)

modelx122 <- lm(ari~GROUP_TYPE, data = M_Group_Readability)
modelx132 <- lm(log_ari~GROUP_TYPE, data = M_Group_Readability)

modelx142 <- lm(smog~GROUP_TYPE, data = M_Group_Readability)
modelx152 <- lm(log_smog~GROUP_TYPE, data = M_Group_Readability)

modelx162 <- lm(cl~GROUP_TYPE, data = M_Group_Readability)
modelx172 <- lm(log_cl~GROUP_TYPE, data = M_Group_Readability)

summary(modelx102)
summary(modelx112)
summary(modelx122)
summary(modelx132)
summary(modelx142)
summary(modelx152)
summary(modelx162)
summary(modelx172)


# Test for significance - ESG V vs E V
modelx103 <- lm(gl~GROUP_TYPE, data = V_Group_Readability)
modelx113 <- lm(log_gl~GROUP_TYPE, data = V_Group_Readability)

modelx123 <- lm(ari~GROUP_TYPE, data = V_Group_Readability)
modelx133 <- lm(log_ari~GROUP_TYPE, data = V_Group_Readability)

modelx143 <- lm(smog~GROUP_TYPE, data = V_Group_Readability)
modelx153 <- lm(log_smog~GROUP_TYPE, data = V_Group_Readability)

modelx163 <- lm(cl~GROUP_TYPE, data = V_Group_Readability)
modelx173 <- lm(log_cl~GROUP_TYPE, data = V_Group_Readability)

summary(modelx103)
summary(modelx113)
summary(modelx123)
summary(modelx133)
summary(modelx143)
summary(modelx153)
summary(modelx163)
summary(modelx173)
```

#H2b & H2c. Linguistic Characteristics - Restrictiveness
H2b and H2c aim to test differences in linguistic characteristics between texts. This is done by counting the amount of certain words. It is expected that the average amount of restrictive words, action words, negative and positive strong words, and exemption words is significantly higher among mandatory documents. On the other hand, it is expected that the average amount of advice words, example words and framework words is significantly higher among voluntary documents. It is expected that voluntary documents have a more guiding character and use less direct and restrictive language. Regarding international frameworks, it is examined which one is most referred to.  

# Count relative amount of words
As a first part of the analysis, the total number of words per word category is divided by the total number of words. This prevents a bias where large texts already contain a larger number of category words. Afterwards, this number is multiplied by 1000 for better interpretation. So, the amount of category words per 1000 words is the new measure that has been created. 
```{r}
R_Data <- Descriptive_Data

R_Data <- R_Data[R_Data$GROUP_ID!="42" & R_Data$GROUP_ID!="43" & R_Data$GROUP_ID!="44", ]

R_Data$negative_strong <- 1000* R_Data$negative_strong/R_Data$totalwords
R_Data$positive_strong <-  1000* R_Data$positive_strong/R_Data$totalwords
R_Data$exemption <-  1000* R_Data$exemption/R_Data$totalwords
R_Data$example <-  1000* R_Data$example/R_Data$totalwords
R_Data$restriction <-  1000* R_Data$restriction/R_Data$totalwords
R_Data$advice <-  1000* R_Data$advice/R_Data$totalwords
R_Data$action<-  1000* R_Data$action/R_Data$totalwords
R_Data$framework<-  1000* R_Data$framework/R_Data$totalwords
```

Variable transformation is also done to see which of the 5 largest international voluntary frameworks (cdp cdsb gri iirc sasb) are mentioned the most across the current voluntary reporting standards. Also, the mentioning of sustainable development goals (SDGs) is counted and the number that TCFD is mentioned (TCFD is a special framework for climate related financial disclosure).
```{r}
R_Data$cdp <-  1000* R_Data$cdp/R_Data$totalwords
R_Data$cdsb <-  1000* R_Data$cdsb/R_Data$totalwords
R_Data$gri <-  1000* R_Data$gri/R_Data$totalwords
R_Data$sasb<-  1000* R_Data$sasb/R_Data$totalwords
R_Data$tcfd <-  1000* R_Data$tcfd/R_Data$totalwords
R_Data$iirc <-  1000* R_Data$iirc/R_Data$totalwords
R_Data$sdg <-  1000* R_Data$sdg/R_Data$totalwords
```

## Average linguistic characteristics between documents: Mandatory vs Voluntary documents
First the difference between mandatory and voluntary documents is examined, irrespective of ESG type. 
```{r}
R_Total <- R_Data %>%
  group_by(M) %>% 
  summarize(average_negative_strong=mean(negative_strong),
            average_positive_strong=mean(positive_strong),
            average_exemption=mean(exemption),
            average_example=mean(example),
            average_restriction=mean(restriction),
            average_advice=mean(advice),
            average_action=mean(action),
            average_framework=mean(framework),
            median_negative_strong=median(negative_strong),
            median_positive_strong=median(positive_strong),
            median_exemption=median(exemption),
            median_example=median(example),
            median_restriction=median(restriction),
            median_advice=median(advice),
            median_action=median(action),
            median_framework=median(framework)) %>% ungroup()

R_Total
```

The numbers can also be visualized in a bar chart, where average (relative) frequency per 1000 words is shown per each word category, per Mandatory/Voluntary. Also, the distribution of mentioning of frameworks across voluntary regulations can be visualized in a bar chart. 
```{r}
# Average frequency per 1000 words, per word category, by M
ggplot(R_Total, aes(factor(M), average_restriction, fill=M)) + 
  geom_bar(stat="identity", position = "dodge") + 
  scale_fill_brewer(palette = "Set1")

# Table with average mentioning of framework
R_Data_frameworks <- R_Data %>% filter(M=="0" & DOC_TYPE=="E" | M=="0" & DOC_TYPE=="ESG") %>% group_by(DOC_TYPE) %>%
  summarise(cdp=mean(cdp),
            cdsb=mean(cdsb),
            gri=mean(gri),
            sasb=mean(sasb),
            tcfd=mean(tcfd),
            iirc=mean(iirc),
            sdg=mean(sdg))

R_Data_frameworks

# Shape data framework to long format
library(reshape2)
frameworks_long <- melt(R_Data_frameworks, id.vars=c("DOC_TYPE"))
frameworks_long$variable <- as.factor(frameworks_long$variable)

# Make a bar chart
ggplot(frameworks_long, aes(factor(variable), value, fill=DOC_TYPE)) + 
  geom_bar(stat="identity", position = "dodge") + 
  scale_fill_brewer(palette = "Set2") + 
  labs(y= "Frequency per 1000 words", x = "Frameworks") +
  theme(text=element_text(size=12,  family="serif"))
```
The difference in linguistic characteristics between mandatory and voluntary documents is tested on significance. Also, the distribution is inspected to see if log transformation is needed. 
```{r}
# Plots comparing distribution of mandatory vs voluntary
plot1<- ggplot(R_Data, aes(x = negative_strong , fill = M)) + 
  geom_density(alpha = 0.5) +
  theme(text=element_text(size=12,  family="serif"))

plot2 <- ggplot(R_Data, aes(x = positive_strong, fill = M)) + 
  geom_density(alpha = 0.5) +
  theme(text=element_text(size=12,  family="serif"))

plot3<- ggplot(R_Data, aes(x = exemption, fill = M)) + 
  geom_density(alpha = 0.5) +
  theme(text=element_text(size=12,  family="serif"))

plot4 <- ggplot(R_Data, aes(x = action, fill = M)) + 
  geom_density(alpha = 0.5) +
  theme(text=element_text(size=12,  family="serif"))

plot5 <- ggplot(R_Data, aes(x = restriction, fill = M)) + 
  geom_density(alpha = 0.5) +
  theme(text=element_text(size=12,  family="serif"))

ggarrange(plot1, plot2, plot3, plot4, plot5)

plot6<- ggplot(R_Data, aes(x = example, fill = M)) + 
  geom_density(alpha = 0.5) +
  theme(text=element_text(size=12,  family="serif"))

plot7 <- ggplot(R_Data, aes(x = advice, fill = M)) + 
  geom_density(alpha = 0.5) +
  theme(text=element_text(size=12,  family="serif"))

plot8<- ggplot(R_Data, aes(x = framework, fill = M)) + 
  geom_density(alpha = 0.5) +
  theme(text=element_text(size=12,  family="serif"))

ggarrange(plot6, plot7, plot8)
```

The difference between groups is also tested for significance by performing a regression analysis. 
```{r}
# Test for significance

# Overall
rmodel10 <- lm(negative_strong~M, data = R_Data)
rmodel11 <- lm(positive_strong~M, data = R_Data)
rmodel12 <- lm(exemption~M, data = R_Data)
rmodel13 <- lm(example~M, data = R_Data)
rmodel14 <- lm(restriction~M, data = R_Data)
rmodel15 <- lm(advice~M, data = R_Data)
rmodel16 <- lm(action~M, data = R_Data)
rmodel17 <- lm(framework~M, data = R_Data)

summary(rmodel10)
summary(rmodel11)
summary(rmodel12)
summary(rmodel13)
summary(rmodel14)
summary(rmodel15)
summary(rmodel16)
summary(rmodel17)

```

## Average linguistic characteristics between documents: Mandatory or Voluntary, subset by DOC_TYPE (ESG)
After examining the average linguistic score between mandatory and voluntary, this difference is examined in more detail by making a distinction in DOC_TYPE. So, if it is an ESG, G, or E document. Because the relative frequency is used, differences in document size do not influence the measurement. 
```{r}
R_DOC <- R_Data %>%
  group_by(DOC_TYPE, M) %>% 
  summarize(average_negative_strong=mean(negative_strong),
            average_positive_strong=mean(positive_strong),
            average_exemption=mean(exemption),
            average_example=mean(example),
            average_restriction=mean(restriction),
            average_advice=mean(advice),
            average_action=mean(action),
            average_framework=mean(framework),
            average_cdp=mean(cdp),
            average_cdsb=mean(cdsb),
            average_gri=mean(gri),
            average_sasb=mean(sasb),
            average_tcfd=mean(tcfd),
            average_iirc=mean(iirc),
            average_sdg=mean(sdg),
            median_negative_strong=median(negative_strong),
            median_positive_strong=median(positive_strong),
            median_exemption=median(exemption),
            median_example=median(example),
            median_restriction=median(restriction),
            median_advice=median(advice),
            median_action=median(action),
            median_framework=median(framework),
            median_cdp=median(cdp),
            median_cdsb=median(cdsb),
            median_gri=median(gri),
            median_sasb=median(sasb),
            median_tcfd=median(tcfd),
            median_iirc=median(iirc),
            median_sdg=median(sdg))%>% ungroup()

R_DOC
```

Again, the significance between certain groups on doc level is examined. The differences that are examined are ESG(M)-ESG(V), ESG(M)-G(M), ESG(V)-E(V).
```{r}
# Create sub-sample
ESG_R_DOC <- R_Data[R_Data$DOC_TYPE=="ESG", ]
M_R_DOC <- R_Data[R_Data$M=="1" & R_Data$DOC_TYPE=="ESG" | R_Data$M=="1" & 
                    R_Data$DOC_TYPE=="G", ]
V_R_DOC <- R_Data[R_Data$M=="0" & R_Data$DOC_TYPE=="ESG" | R_Data$M=="0" & 
                    R_Data$DOC_TYPE=="E", ]

# ESG mandatory vs ESG Voluntary #########################
ermodel10 <- lm(negative_strong~M, data = ESG_R_DOC)
ermodel11 <- lm(positive_strong~M, data = ESG_R_DOC)
ermodel12 <- lm(exemption~M, data = ESG_R_DOC)
ermodel13 <- lm(example~M, data = ESG_R_DOC)
ermodel14 <- lm(restriction~M, data = ESG_R_DOC)
ermodel15 <- lm(advice~M, data = ESG_R_DOC)
ermodel16 <- lm(action~M, data = ESG_R_DOC)
ermodel17 <- lm(framework~M, data = ESG_R_DOC)

summary(ermodel10)
summary(ermodel11)
summary(ermodel12)
summary(ermodel13)
summary(ermodel14)
summary(ermodel15)
summary(ermodel16)
summary(ermodel17)

# ESG mandatory vs G Mandatory #########################
ermodelx10 <- lm(negative_strong~DOC_TYPE, data = M_R_DOC)
ermodelx11 <- lm(positive_strong~DOC_TYPE, data = M_R_DOC)
ermodelx12 <- lm(exemption~DOC_TYPE, data = M_R_DOC)
ermodelx13 <- lm(example~DOC_TYPE, data = M_R_DOC)
ermodelx14 <- lm(restriction~DOC_TYPE, data = M_R_DOC)
ermodelx15 <- lm(advice~DOC_TYPE, data = M_R_DOC)
ermodelx16 <- lm(action~DOC_TYPE, data = M_R_DOC)
ermodelx17 <- lm(framework~DOC_TYPE, data = M_R_DOC)

summary(ermodelx10)
summary(ermodelx11)
summary(ermodelx12)
summary(ermodelx13)
summary(ermodelx14)
summary(ermodelx15)
summary(ermodelx16)
summary(ermodelx17)

# ESG voluntary vs E voluntary #########################
ermodelyx10 <- lm(negative_strong~DOC_TYPE, data = V_R_DOC)
ermodelyx11 <- lm(positive_strong~DOC_TYPE, data = V_R_DOC)
ermodelyx12 <- lm(exemption~DOC_TYPE, data = V_R_DOC)
ermodelyx13 <- lm(example~DOC_TYPE, data = V_R_DOC)
ermodelyx14 <- lm(restriction~DOC_TYPE, data = V_R_DOC)
ermodelyx15 <- lm(advice~DOC_TYPE, data = V_R_DOC)
ermodelyx16 <- lm(action~DOC_TYPE, data = V_R_DOC)
ermodelyx17 <- lm(framework~DOC_TYPE, data = V_R_DOC)

#frameworks separated
fermodelyx11 <- lm(cdp~DOC_TYPE, data = V_R_DOC)
fermodelyx12 <- lm(cdsb~DOC_TYPE, data = V_R_DOC)
fermodelyx13 <- lm(gri~DOC_TYPE, data = V_R_DOC)
fermodelyx14 <- lm(sasb~DOC_TYPE, data = V_R_DOC)
fermodelyx15 <- lm(tcfd~DOC_TYPE, data = V_R_DOC)
fermodelyx16 <- lm(iirc~DOC_TYPE, data = V_R_DOC)
fermodelyx17 <- lm(sdg~DOC_TYPE, data = V_R_DOC)

summary(ermodelyx10)
summary(ermodelyx11)
summary(ermodelyx12)
summary(ermodelyx13)
summary(ermodelyx14)
summary(ermodelyx15)
summary(ermodelyx16)
summary(ermodelyx17)

summary(fermodelyx11)
summary(fermodelyx12)
summary(fermodelyx13)
summary(fermodelyx14)
summary(fermodelyx15)
summary(fermodelyx16)
summary(fermodelyx17)
```

## Average Linguistic Characteristics: Mandatory vs Voluntary, subset by Group Type 
As a final part of the analysis, the differences between mandatory and voluntary are examined, while sub-setting at group type. The concept of grouping documents is already explained in the Quantity analysis part of this document. By looking at averages within a group, the overall linguistic characteristics of documents can be examined. So, a conclusion can be drawn on how restrictive mandatory ESG documentation compared to voluntary ESG documentation is.
```{r}
# Group the data by Group_ID
Group_R <- R_Data %>%
  group_by(GROUP_ID, GROUP_TYPE, M) %>%
  summarize(negative_strong=sum(negative_strong),
            positive_strong=sum(positive_strong),
            exemption=sum(exemption),
            example=sum(example),
            restriction=sum(restriction),
            advice=sum(advice),
            action=sum(action),
            framework=sum(framework),
            cdp=sum(cdp),
            cdsb=sum(cdsb),
            gri=sum(gri),
            sasb=sum(sasb),
            tcfd=sum(tcfd),
            iirc=sum(iirc),
            sdg=sum(sdg))%>% ungroup()

# Get summary statistics by group type and by M
Group_Restrictions_Stats <- Group_R %>%                              
  group_by(GROUP_TYPE, M) %>% 
  summarize(average_negative_strong=mean(negative_strong),
            average_positive_strong=mean(positive_strong),
            average_exemption=mean(exemption),
            average_example=mean(example),
            average_restriction=mean(restriction),
            average_advice=mean(advice),
            average_action=mean(action),
            average_framework=mean(framework),
            average_cdp=mean(cdp),
            average_cdsb=mean(cdsb),
            average_gri=mean(gri),
            average_sasb=mean(sasb),
            average_tcfd=mean(tcfd),
            average_iirc=mean(iirc),
            average_sdg=mean(sdg),
            median_negative_strong=median(negative_strong),
            median_positive_strong=median(positive_strong),
            median_exemption=median(exemption),
            median_example=median(example),
            median_restriction=median(restriction),
            median_advice=median(advice),
            median_action=median(action),
            median_framework=median(framework),
            median_cdp=median(cdp),
            median_cdsb=median(cdsb),
            median_gri=median(gri),
            median_sasb=median(sasb),
            median_tcfd=median(tcfd),
            median_iirc=median(iirc),
            median_sdg=median(sdg)) %>% ungroup()

# Get summary statistics by M
Group_Restrictions_total <- Group_R %>%                              
  group_by(M) %>% 
  summarize(average_negative_strong=mean(negative_strong),
            average_positive_strong=mean(positive_strong),
            average_exemption=mean(exemption),
            average_example=mean(example),
            average_restriction=mean(restriction),
            average_advice=mean(advice),
            average_action=mean(action),
            average_framework=mean(framework),
            average_cdp=mean(cdp),
            average_cdsb=mean(cdsb),
            average_gri=mean(gri),
            average_sasb=mean(sasb),
            average_tcfd=mean(tcfd),
            average_iirc=mean(iirc),
            average_sdg=mean(sdg),
            median_negative_strong=median(negative_strong),
            median_positive_strong=median(positive_strong),
            median_exemption=median(exemption),
            median_example=median(example),
            median_restriction=median(restriction),
            median_advice=median(advice),
            median_action=median(action),
            median_framework=median(framework),
            median_cdp=median(cdp),
            median_cdsb=median(cdsb),
            median_gri=median(gri),
            median_sasb=median(sasb),
            median_tcfd=median(tcfd),
            median_iirc=median(iirc),
            median_sdg=median(sdg)) %>% ungroup()

Group_Restrictions_Stats
Group_Restrictions_total
```

The region is added to the results on doc level and group level. This allows to examine if differences are driven by regional differences. 
```{r}
Group_R <- merge(Group_R, Country_ID, by.x = "GROUP_ID", by.y = "GROUP_ID", all.x = TRUE, all.y = FALSE)
```

```{r}
# Create summary table, by Region by doc type
R_Region_DOC <- R_Data %>% group_by(REGION, DOC_TYPE) %>% summarise(
            average_negative_strong=mean(negative_strong),
            average_positive_strong=mean(positive_strong),
            average_exemption=mean(exemption),
            average_example=mean(example),
            average_restriction=mean(restriction),
            average_advice=mean(advice),
            average_action=mean(action),
            average_framework=mean(framework))

# Create summary table, by Region by group type by M
R_Region_Group <- Group_R %>% group_by(REGION, GROUP_TYPE, M) %>% summarise(
            average_negative_strong=mean(negative_strong),
            average_positive_strong=mean(positive_strong),
            average_exemption=mean(exemption),
            average_example=mean(example),
            average_restriction=mean(restriction),
            average_advice=mean(advice),
            average_action=mean(action),
            average_framework=mean(framework))

R_Region_DOC
R_Region_Group
```

Some bar charts were created to see difference in frameworks between E voluntary and ESG voluntary on a group level. Also, bar chart is added that looks at the other linguistic categories, grouped by mandatory or voluntary. 
```{r}
# Division of framework types, compared between E and ESG voluntary (group level)
R_Groupdata_frameworks <- Group_R %>% filter(M=="0") %>% group_by(GROUP_TYPE) %>%
  summarise(cdp=mean(cdp),
            cdsb=mean(cdsb),
            gri=mean(gri),
            sasb=mean(sasb),
            tcfd=mean(tcfd),
            iirc=mean(iirc),
            sdg=mean(sdg))

R_Groupdata_frameworks

# Shape data framework to long format
frameworks_longG <- melt(R_Groupdata_frameworks, id.vars=c("GROUP_TYPE"))
frameworks_longG$variable <- as.factor(frameworks_longG$variable)

# Make a bar chart
ggplot(frameworks_longG, aes(factor(variable), value, fill=GROUP_TYPE)) + 
  geom_bar(stat="identity", position = "dodge") + 
  scale_fill_brewer(palette = "Set2") + 
  labs(y= "Frequency per 1000 words", x = "Frameworks") +
  theme(text=element_text(size=12,  family="serif"))

# Average frequency of linguistic characteristics across Mandatory and Voluntary (group level)
R_Groupdata_types <- Group_R %>% group_by(M) %>%
  summarise( negative_strong=mean(negative_strong),
            positive_strong=mean(positive_strong),
            exemption=mean(exemption),
            example=mean(example),
            restriction=mean(restriction),
            advice=mean(advice),
            action=mean(action),
            framework=mean(framework))

R_Groupdata_types

# Shape data framework to long format
frameworks_longT <- melt(R_Groupdata_types, id.vars=c("M"))
frameworks_longT$variable <- as.factor(frameworks_longT$variable)

# Make a bar chart
ggplot(frameworks_longT, aes(factor(variable), value, fill=M)) + 
  geom_bar(stat="identity", position = "dodge") + 
  scale_fill_brewer(palette = "Set2") + 
  labs(y= "Frequency per 1000 words", x = "categories") +
  theme(text=element_text(size=14,  family="serif")) + theme(axis.text.x=element_text(angle = 45, hjust = 1))
```

The difference between groups is also tested for significance by performing a regression analysis.   
```{r}
# Test for significance - M vs V
rmodelq10 <- lm(negative_strong~M, data = Group_R)
rmodelq11 <- lm(positive_strong~M, data = Group_R)
rmodelq12 <- lm(exemption~M, data = Group_R)
rmodelq13 <- lm(example~M, data = Group_R)
rmodelq14 <- lm(restriction~M, data = Group_R)
rmodelq15 <- lm(advice~M, data = Group_R)
rmodelq16 <- lm(action~M, data = Group_R)
rmodelq17 <- lm(framework~M, data = Group_R)

summary(rmodelq10)
summary(rmodelq11)
summary(rmodelq12)
summary(rmodelq13)
summary(rmodelq14)
summary(rmodelq15)
summary(rmodelq16)
summary(rmodelq17)

# Test for significance > sub-samples
# Create sub-sample
ESG_RES_DOC <- Group_R[Group_R$GROUP_TYPE=="ESG", ]
M_RES_DOC <- Group_R[Group_R$M=="1" & Group_R$GROUP_TYPE=="ESG" | Group_R$M=="1" & Group_R$GROUP_TYPE=="G", ]
V_RES_DOC <- Group_R[Group_R$M=="0" & Group_R$GROUP_TYPE=="ESG" | Group_R$M=="0" & Group_R$GROUP_TYPE=="E", ]

# ESG mandatory vs ESG Voluntary #########################
# Overall
ermodelw10 <- lm(negative_strong~M, data = ESG_RES_DOC)
ermodelw11 <- lm(positive_strong~M, data = ESG_RES_DOC)
ermodelw12 <- lm(exemption~M, data = ESG_RES_DOC)
ermodelw13 <- lm(example~M, data = ESG_RES_DOC)
ermodelw14 <- lm(restriction~M, data = ESG_RES_DOC)
ermodelw15 <- lm(advice~M, data = ESG_RES_DOC)
ermodelw16 <- lm(action~M, data = ESG_RES_DOC)
ermodelw17 <- lm(framework~M, data = ESG_RES_DOC)

summary(ermodelw10)
summary(ermodelw11)
summary(ermodelw12)
summary(ermodelw13)
summary(ermodelw14)
summary(ermodelw15)
summary(ermodelw16)
summary(ermodelw17)

# ESG mandatory vs G Mandatory #########################
# Overall
ermodeldx10 <- lm(negative_strong~GROUP_TYPE, data = M_RES_DOC)
ermodeldx11 <- lm(positive_strong~GROUP_TYPE, data = M_RES_DOC)
ermodeldx12 <- lm(exemption~GROUP_TYPE, data = M_RES_DOC)
ermodeldx13 <- lm(example~GROUP_TYPE, data = M_RES_DOC)
ermodeldx14 <- lm(restriction~GROUP_TYPE, data = M_RES_DOC)
ermodeldx15 <- lm(advice~GROUP_TYPE, data = M_RES_DOC)
ermodeldx16 <- lm(action~GROUP_TYPE, data = M_RES_DOC)
ermodeldx17 <- lm(framework~GROUP_TYPE, data = M_RES_DOC)

summary(ermodeldx10)
summary(ermodeldx11)
summary(ermodeldx12)
summary(ermodeldx13)
summary(ermodeldx14)
summary(ermodeldx15)
summary(ermodeldx16)
summary(ermodeldx17)

# ESG voluntary vs E voluntary #########################
# Overall
ermodelkyx10 <- lm(negative_strong~GROUP_TYPE, data = V_RES_DOC)
ermodelkyx11 <- lm(positive_strong~GROUP_TYPE, data = V_RES_DOC)
ermodelkyx12 <- lm(exemption~GROUP_TYPE, data = V_RES_DOC)
ermodelkyx13 <- lm(example~GROUP_TYPE, data = V_RES_DOC)
ermodelkyx14 <- lm(restriction~GROUP_TYPE, data = V_RES_DOC)
ermodelkyx15 <- lm(advice~GROUP_TYPE, data = V_RES_DOC)
ermodelkyx16 <- lm(action~GROUP_TYPE, data = V_RES_DOC)
ermodelkyx17 <- lm(framework~GROUP_TYPE, data = V_RES_DOC)

#frameworks separated
fermodelkyx11 <- lm(cdp~GROUP_TYPE, data = V_RES_DOC)
fermodelkyx12 <- lm(cdsb~GROUP_TYPE, data = V_RES_DOC)
fermodelkyx13 <- lm(gri~GROUP_TYPE, data = V_RES_DOC)
fermodelkyx14 <- lm(sasb~GROUP_TYPE, data = V_RES_DOC)
fermodelkyx15 <- lm(tcfd~GROUP_TYPE, data = V_RES_DOC)
fermodelkyx16 <- lm(iirc~GROUP_TYPE, data = V_RES_DOC)
fermodelkyx17 <- lm(sdg~GROUP_TYPE, data = V_RES_DOC)

summary(ermodelkyx10)
summary(ermodelkyx11)
summary(ermodelkyx12)
summary(ermodelkyx13)
summary(ermodelkyx14)
summary(ermodelkyx15)
summary(ermodelkyx16)
summary(ermodelkyx17)

summary(fermodelkyx11)
summary(fermodelkyx12)
summary(fermodelkyx13)
summary(fermodelkyx14)
summary(fermodelkyx15)
summary(fermodelkyx16)
summary(fermodelkyx17)
```