replication_main_socialmedia_oss.Rmd

### Generate Figure 1 in the paper
```{r}
library(readr)
library(ggplot2)
library(ggpubr)

year_aggreagte_media_com_R = read_csv('./data/year_aggreagte_media_com_R_no_so.csv') 
year_aggreagte_media_com_Python = read_csv('./data/year_aggreagte_media_com_Python_no_so.csv') 

legend_font_size = 22
title_size = 20
tick_size = 18
axis_label_size = 22
plot_line_size = 2

p1 = ggplot(year_aggreagte_media_com_R, aes(x=year)) + 
  geom_line(aes(y = tw_post_count, colour  = "Twitter"), size = plot_line_size) + 
  geom_line(aes(y = hn_post_count, colour ="HackerNews"), size = plot_line_size) + 
    geom_line(aes(y = rd_post_count, colour ="Reddit"), size = plot_line_size) +
  scale_y_continuous(trans = "log10") + 
  scale_colour_manual("", 
                      breaks = c("Twitter", "HackerNews", "Reddit"),
                      values = c("blue", "red", "orange"))+
 ylab("Count (log scale)")+
  ggtitle("Number of posts for R projects")+
  theme(plot.title = element_text(hjust = 0.5,size = title_size),
        legend.text = element_text(size=legend_font_size),
        axis.title.x = element_blank(),
        axis.text=element_text(size=tick_size),
        axis.title=element_text(size=axis_label_size))



p2 = ggplot(year_aggreagte_media_com_R, aes(x=year)) + 
  geom_line(aes(y = tw_author_count, colour  = "Twitter"), size = plot_line_size) + 
  geom_line(aes(y = hn_author_count, colour ="HackerNews"), size = plot_line_size) + 
    geom_line(aes(y = rd_author_count, colour ="Reddit"), size = plot_line_size) +
  scale_y_continuous(trans = "log10") + 
  scale_colour_manual("", 
                      breaks = c("Twitter", "HackerNews", "Reddit"),
                      values = c("blue", "red", "orange"))+
 ylab("Count (log scale)")+
  ggtitle("Number of users posting for R projects")+
  theme(plot.title = element_text(hjust = 0.5,size = title_size),
        legend.text = element_text(size=legend_font_size),
        axis.title.y = element_blank(),
        axis.title.x = element_blank(),
        axis.text=element_text(size=tick_size),
        axis.title=element_text(size=axis_label_size))


p3 = ggplot(year_aggreagte_media_com_R, aes(x=year)) + 
  geom_line(aes(y = tw_repo_count, colour  = "Twitter"), size = plot_line_size) + 
  geom_line(aes(y = hn_repo_count, colour ="HackerNews"), size = plot_line_size) + 
    geom_line(aes(y = rd_repo_count, colour ="Reddit"), size = plot_line_size) +
  scale_y_continuous(trans = "log10") + 
  scale_colour_manual("", 
                      breaks = c("Twitter", "HackerNews", "Reddit"),
                      values = c("blue", "red", "orange"))+
 ylab("Count (log scale)")+
  ggtitle("Number of R projects mentioned")+
  theme(plot.title = element_text(hjust = 0.5,size = title_size),
        legend.text = element_text(size=legend_font_size),
        axis.title.y = element_blank(),
        axis.title.x = element_blank(),
        axis.text=element_text(size=tick_size),
        axis.title=element_text(size=axis_label_size))




p4 = ggplot(year_aggreagte_media_com_Python, aes(x=year)) + 
  geom_line(aes(y = tw_post_count, colour  = "Twitter"), size = plot_line_size) + 
  geom_line(aes(y = hn_post_count, colour ="HackerNews"), size = plot_line_size) + 
    geom_line(aes(y = rd_post_count, colour ="Reddit"), size = plot_line_size) +
  scale_y_continuous(trans = "log10") + 
  scale_colour_manual("", 
                      breaks = c("Twitter", "HackerNews", "Reddit"),
                      values = c("blue", "red", "orange"))+
 ylab("Count (log scale)")+
  ggtitle("Number of posts for Python projects")+
  theme(plot.title = element_text(hjust = 0.5,size = title_size),
        legend.text = element_text(size=legend_font_size),
        axis.text=element_text(size=tick_size),
        axis.title=element_text(size=axis_label_size))



p5 = ggplot(year_aggreagte_media_com_Python, aes(x=year)) + 
  geom_line(aes(y = tw_author_count, colour  = "Twitter"), size = plot_line_size) + 
  geom_line(aes(y = hn_author_count, colour ="HackerNews"), size = plot_line_size) + 
    geom_line(aes(y = rd_author_count, colour ="Reddit"), size = plot_line_size) +
  scale_y_continuous(trans = "log10") + 
  scale_colour_manual("", 
                      breaks = c("Twitter", "HackerNews", "Reddit"),
                      values = c("blue", "red", "orange"))+
 ylab("Count (log scale)")+
  ggtitle("Number of users posting for Python projects")+
  theme(plot.title = element_text(hjust = 0.5,size = title_size),
        legend.text = element_text(size=legend_font_size),
        axis.title.y = element_blank(),
        axis.text=element_text(size=tick_size),
        axis.title=element_text(size=axis_label_size))


p6 = ggplot(year_aggreagte_media_com_Python, aes(x=year)) + 
  geom_line(aes(y = tw_repo_count, colour  = "Twitter"), size = plot_line_size) + 
  geom_line(aes(y = hn_repo_count, colour ="HackerNews"), size = plot_line_size) + 
    geom_line(aes(y = rd_repo_count, colour ="Reddit"), size = plot_line_size) + 
  scale_y_continuous(trans = "log10") + 
  scale_colour_manual("", 
                      breaks = c("Twitter", "HackerNews", "Reddit"),
                      values = c("blue", "red", "orange"))+
 ylab("Count (log scale)")+
  ggtitle("Number of Python projects mentioned")+
  theme(plot.title = element_text(hjust = 0.5,size = title_size),
        legend.text = element_text(size=legend_font_size),
        axis.title.y = element_blank(),
        axis.text=element_text(size=tick_size),
        axis.title=element_text(size=axis_label_size))

ggarrange(p1, p2, p3, p4, p5, p6, ncol=3, nrow=2, common.legend = TRUE, legend="right") 

ggsave(file="./figure/social_media_use.pdf", width=21.5, height=9, dpi=300)
```

### Generate Figure 2 in the paper
```{r}
library(tidyr)
legend_font_size = 28
title_size = 32
tick_size = 28
x_axis_label_size = 32
y_axis_label_size = 18
plot_line_size = 3
legend_title_size = 32
coef_hn_rd_so_R = 15
coef_hn_rd_so_Python = 6

# R community

df_first_year = read_csv('./data/media_post_relative_time_com_R_no_so.csv')
df_first_year = subset(df_first_year, creation_year > 2010 & (creation_year < 2018 | project_create_before_201806_flag == TRUE) &
                                      project_history_count >= 365 ) # only keep projects which we have at least one-year of history, and created after year 2010, because early days have few projects

df_repo_has_media_by_year = setNames(aggregate(cbind(df_first_year$mention_on_tw_first_year_flag, df_first_year$mention_on_hn_first_year_flag, df_first_year$mention_on_rd_first_year_flag), by = list(df_first_year$creation_year), FUN = mean), c("year", "Twitter", 'HackerNews', 'Reddit'))

df_repo_has_media_by_year$HackerNews = df_repo_has_media_by_year$HackerNews * coef_hn_rd_so_R
df_repo_has_media_by_year$Reddit = df_repo_has_media_by_year$Reddit * coef_hn_rd_so_R

df_repo_has_media_by_year_long <- df_repo_has_media_by_year %>% 
  pivot_longer(
    cols = c('Twitter','HackerNews','Reddit'), 
    names_to = "media",
    values_to = "ratio_mentioned"
)

df_repo_has_media_by_year_long <- df_repo_has_media_by_year_long %>%
  mutate( media=factor(media,levels=c("Twitter", "HackerNews", "Reddit", "Stargazer")) )

p1_R = ggplot(df_repo_has_media_by_year_long,aes(x=year, y=ratio_mentioned, group=media, color=media))+
  geom_line(size = plot_line_size) +
    scale_colour_manual(values = c('Twitter' = "blue",'HackerNews' = "red", 'Reddit' = "orange"))+
    scale_y_continuous(

    # Features of the first axis
    name = "Ratio of projects mentioned on TW",

    # Add a second axis and specify its features
    sec.axis = sec_axis(~.*1.0 / coef_hn_rd_so_R, name="Ratio of projects mentioned on others")
  ) +
  ggtitle("R projects")+
  xlab('year')+
  theme(plot.title = element_text(hjust = 0.5,size = title_size),
        legend.text = element_text(size=legend_font_size),
        axis.text=element_text(size=tick_size),
        axis.title.x=element_text(size=x_axis_label_size),
        axis.title.y=element_text(size=y_axis_label_size),
        legend.title=element_text(size=legend_title_size))

# Python

df_first_year = read_csv('./data/media_post_relative_time_com_Python_no_so.csv')
df_first_year = subset(df_first_year, creation_year > 2010 & (creation_year < 2018 | project_create_before_201806_flag == TRUE) &
                                      project_history_count >= 365 ) # only keep projects which we have at least one-year of history, and created after year 2010, because early days have few projects


df_repo_has_media_by_year = setNames(aggregate(cbind(df_first_year$mention_on_tw_first_year_flag, df_first_year$mention_on_hn_first_year_flag, df_first_year$mention_on_rd_first_year_flag), by = list(df_first_year$creation_year), FUN = mean), c("year", "Twitter", 'HackerNews', 'Reddit'))

df_repo_has_media_by_year$HackerNews = df_repo_has_media_by_year$HackerNews * coef_hn_rd_so_Python
df_repo_has_media_by_year$Reddit = df_repo_has_media_by_year$Reddit * coef_hn_rd_so_Python

df_repo_has_media_by_year_long <- df_repo_has_media_by_year %>% 
  pivot_longer(
    cols = c('Twitter','HackerNews','Reddit'), 
    names_to = "media",
    values_to = "ratio_mentioned"
)

df_repo_has_media_by_year_long <- df_repo_has_media_by_year_long %>%
  mutate( media=factor(media,levels=c("Twitter", "HackerNews", "Reddit", "Stargazer")) )

p2_Python = ggplot(df_repo_has_media_by_year_long,aes(x=year, y=ratio_mentioned, group=media, color=media))+
  geom_line(size = plot_line_size) +
    scale_colour_manual(values = c('Twitter' = "blue",'HackerNews' = "red", 'Reddit' = "orange"))+
    scale_y_continuous(

    # Features of the first axis
    name = "Ratio of projects mentioned on TW",

    # Add a second axis and specify its features
    sec.axis = sec_axis(~.*1.0 / coef_hn_rd_so_Python, name="Ratio of projects mentioned on others")
  ) +
  ggtitle("Python projects")+
  xlab('year')+
  theme(plot.title = element_text(hjust = 0.5,size = title_size),
        legend.text = element_text(size=legend_font_size),
        axis.text=element_text(size=tick_size),
        axis.title.x=element_text(size=x_axis_label_size),
        axis.title.y=element_text(size=y_axis_label_size),
        legend.title=element_text(size=legend_title_size))

ggarrange(p1_R, p2_Python, nrow=2, common.legend = TRUE, legend="right")

ggsave(file="./figure/ratio_being_posted.pdf", width=16, height=12, dpi=300)

```

### Generate Figure 3 in the paper
```{r}
library(tidyr)
library(readr)
library(ggplot2)
library(dplyr)
library(ggpubr)
legend_font_size = 20
title_size = 32
y_tick_size = 22
x_tick_size = 32
axis_label_size = 24
node_size = 4
coef_hn_rd_so = 20
coef_hn_rd_Python = 4

longitudional_peak_media_R = read_csv('./data/media_post_relative_time_com_R_no_so.csv')
longitudional_peak_devstar_R = read_csv('./data/dev_star_activity_com_R_complete_period.csv')
df_R = merge(longitudional_peak_media_R, longitudional_peak_devstar_R, by = 'repo_slug')

df_R = df_R[c('repo_slug', 'first_rd_time', 'first_hn_time', 'first_tw_time', 'first_stargazer_time')]


df_R_long <- df_R %>% 
  pivot_longer(
    cols = c('first_rd_time','first_hn_time', 'first_tw_time', 'first_stargazer_time'), 
    names_to = "type",
    values_to = "first_appear_time"
)

df_R_long[df_R_long$type == "first_rd_time",]$type <- 'Reddit'
df_R_long[df_R_long$type == "first_hn_time",]$type <- 'HackerNews'
df_R_long[df_R_long$type == "first_tw_time",]$type <- 'Twitter'
df_R_long[df_R_long$type == "first_stargazer_time",]$type <- 'Stargazer'
df_R_long[df_R_long$first_appear_time == -1,]$first_appear_time <- NA

df_R_long <- df_R_long %>%
  mutate( type=factor(type,levels=c("Twitter", "HackerNews", "Reddit", "Stargazer")) )


p1_R = ggplot(df_R_long, aes(x=type, y=first_appear_time)) + 
  geom_violin() +
  stat_summary(fun.data=mean_cl_boot, fun.args=list(conf.int=0.95),
                 geom="pointrange", color="red") + ylab('First appear time (in days)') + xlab('')+
  ggtitle("R projects")+
  theme(plot.title = element_text(hjust = 0.5,size = title_size),
        legend.text = element_text(size=legend_font_size),
        axis.text.x=element_text(size=x_tick_size),
        axis.text.y=element_text(size=y_tick_size),
        axis.title=element_text(size=axis_label_size))


longitudional_peak_media_Python = read_csv('./data/media_post_relative_time_com_Python_no_so.csv')
longitudional_peak_devstar_Python = read_csv('./data/dev_star_activity_com_Python_complete_period.csv')
                                                        
df_Python = merge(longitudional_peak_media_Python, longitudional_peak_devstar_Python, by = 'repo_slug')

df_Python = df_Python[c('repo_slug', 'first_rd_time', 'first_hn_time', 'first_tw_time', 'first_stargazer_time')]


df_Python_long <- df_Python %>% 
  pivot_longer(
    cols = c('first_rd_time','first_hn_time', 'first_tw_time', 'first_stargazer_time'), 
    names_to = "type",
    values_to = "first_appear_time"
)
df_Python_long[df_Python_long$type == "first_rd_time",]$type <- 'Reddit'
df_Python_long[df_Python_long$type == "first_hn_time",]$type <- 'HackerNews'
df_Python_long[df_Python_long$type == "first_tw_time",]$type <- 'Twitter'
df_Python_long[df_Python_long$type == "first_stargazer_time",]$type <- 'Stargazer'
df_Python_long[df_Python_long$first_appear_time == -1,]$first_appear_time <- NA

df_Python_long <- df_Python_long %>%
  mutate( type=factor(type,levels=c("Twitter", "HackerNews", "Reddit", "Stargazer")) )

p2_Python = ggplot(df_Python_long, aes(x=type, y=first_appear_time)) + 
  geom_violin() +
  stat_summary(fun.data=mean_cl_boot, fun.args=list(conf.int=0.95),
                 geom="pointrange", color="red") + ylab('First appear time (in days)') + xlab('')+
  ggtitle("Python projects")+
  theme(plot.title = element_text(hjust = 0.5,size = title_size),
        legend.text = element_text(size=legend_font_size),
        axis.text.x=element_text(size=x_tick_size),
        axis.text.y=element_text(size=y_tick_size),
        axis.title=element_text(size=axis_label_size))

ggarrange(p1_R, p2_Python, ncol=1, nrow=2) 

ggsave(file="./figure/first_appear_time.pdf", width=16, height=12, dpi=300)

df_R_long_tw = subset(df_R_long, type == 'Stargazer')
summary(df_R_long_tw$first_appear_time)

df_Python_long_tw = subset(df_Python_long, type == 'Stargazer')
summary(df_Python_long_tw$first_appear_time)

```


### Generate Figure 4 in the paper
```{r}
tick_size = 32
x_axis_label_size = 30
y_axis_label_size = 22
legend_font_size = 28
legend_title_size = 32
coef_hn_rd_so = 50
coef_hn_rd_Python = 20
time_shift_R = read_csv('./data/time_shift_of_platforms_com_R_no_so.csv')
tw_range_R = c(0.97, 1)
other_range_R = c(0, 0.03)
b_R <- diff(tw_range_R)/diff(other_range_R)
a_R <- tw_range_R[1] - b_R*other_range_R[1]


tw_range_Python = c(0.85, 1)
other_range_Python = c(0, 0.05)
b_Python <- diff(tw_range_Python)/diff(other_range_Python)
a_Python <- tw_range_Python[1] - b_Python*other_range_Python[1]


time_shift_R_tw = subset(time_shift_R, media == 'Twitter')
time_shift_R_hn = subset(time_shift_R, media == 'HackerNews')
time_shift_R_hn$average_ratio = time_shift_R_hn$average_ratio * b_R + a_R
time_shift_R_rd = subset(time_shift_R, media == 'Reddit')
time_shift_R_rd$average_ratio = time_shift_R_rd$average_ratio * b_R + a_R


time_shift_R_processed = rbind(time_shift_R_tw, time_shift_R_hn, time_shift_R_rd)

time_shift_R_processed <- time_shift_R_processed %>%
  mutate( time_group=factor(time_group,levels=c("first", "second", "third", "forth")) )

time_shift_R_processed <- time_shift_R_processed %>%
  mutate( media=factor(media,levels=c("Twitter", "HackerNews", "Reddit")) )

p1_R = ggplot(time_shift_R_processed,aes(x=time_group, y=average_ratio, group=media, color=media))+
  geom_line(size = plot_line_size) +
    scale_colour_manual(values = c('Twitter' = "blue",'HackerNews' = "red", 'Reddit' = "orange"))+
    scale_y_continuous(

    # Features of the first axis
    name = "Ratio of posts on TW",

    # Add a second axis and specify its features
    # sec.axis = sec_axis(~.*1.0 / coef_hn_rd_so, name="Ratio of posts on HN, RD, SO")
    sec.axis = sec_axis(~ (. - a_R)/b_R, name="Ratio of posts on others")

  )+
  xlab('Time of post appearance (quantile)')+
  theme(plot.title = element_text(hjust = 0.5,size = title_size),
        legend.text = element_text(size=legend_font_size),
        axis.text=element_text(size=tick_size),
        axis.title.x=element_text(size=x_axis_label_size),
        axis.title.y=element_text(size=y_axis_label_size),
        legend.title=element_text(size=legend_title_size))+
  ggtitle('R projects')


time_shift_Python = read_csv('./data/time_shift_of_platforms_com_Python_no_so.csv')



time_shift_Python_tw = subset(time_shift_Python, media == 'Twitter')
time_shift_Python_hn = subset(time_shift_Python, media == 'HackerNews')
time_shift_Python_hn$average_ratio = time_shift_Python_hn$average_ratio * b_Python + a_Python
time_shift_Python_rd = subset(time_shift_Python, media == 'Reddit')
time_shift_Python_rd$average_ratio = time_shift_Python_rd$average_ratio * b_Python + a_Python


time_shift_Python_processed = rbind(time_shift_Python_tw, time_shift_Python_hn, time_shift_Python_rd)

time_shift_Python_processed <- time_shift_Python_processed %>%
  mutate( time_group=factor(time_group,levels=c("first", "second", "third", "forth")) )

time_shift_Python_processed <- time_shift_Python_processed %>%
  mutate( media=factor(media,levels=c("Twitter", "HackerNews", "Reddit")) )

p2_Python = ggplot(time_shift_Python_processed,aes(x=time_group, y=average_ratio, group=media, color=media))+
  geom_line(size = plot_line_size) +
    scale_colour_manual(values = c('Twitter' = "blue",'HackerNews' = "red", 'Reddit' = "orange"))+
    scale_y_continuous(

    # Features of the first axis
    name = "Ratio of posts on TW",

    # Add a second axis and specify its features
    # sec.axis = sec_axis(~.*1.0 / coef_hn_rd_Python, name="Ratio of posts on HN, RD, SO")
    sec.axis = sec_axis(~ (. - a_Python)/b_Python, name="Ratio of posts on others")

  )+
  xlab('Time of post appearance (quantile)')+
  theme(plot.title = element_text(hjust = 0.5,size = title_size),
        legend.text = element_text(size=legend_font_size),
        axis.text=element_text(size=tick_size),
        axis.title.x=element_text(size=x_axis_label_size),
        axis.title.y=element_text(size=y_axis_label_size),
        legend.title=element_text(size=legend_title_size))+
  ggtitle('Python projects')

ggarrange(p1_R, p2_Python, ncol=1, nrow=2, common.legend = TRUE, legend="right") 

ggsave(file="./figure/time_shift_focus.pdf", width=12, height=9, dpi=300)
```




### Generate Figure 5 in the paper, step 1, data process
```{r}
library(DescTools)
library(dplyr)
library(boot)
library(hash)
library(ggplot2)
library(gridExtra)
library(relaimpo)
library(gdata)
library(tidyverse)

community_name = 'Python' # R or Python

longitudional_peak_media = read_csv(sprintf('./data/media_post_relative_time_com_%s_no_so.csv', community_name))


longitudional_peak_devstar = read_csv(sprintf('./data/dev_star_activity_com_%s_complete_period.csv', community_name))
df_raw = merge(longitudional_peak_media, longitudional_peak_devstar, by = 'repo_slug')
                                                                                  
df_data = setNames(df_raw[c('repo_slug', 'one_hop_author_within_owner_count', 'two_hop_author_within_owner_count', 'three_hop_author_within_owner_count', 'total_author_at_the_time_count', 'new_author_one_hop_count', 'new_author_two_hop_count', 'new_author_three_hop_count', 'new_author_fourormore_hop_count', 'stargazer_one_hop_count', 'stargazer_two_hop_count', 'stargazer_three_hop_count', 'stargazer_fourormore_hop_count', 'rd_post_count', 'hn_post_count', 'tw_post_count', 'project_history_count')],
                   
c('repo_slug', 'one_hop_gh_user_count', 'two_hop_gh_user_count', 'three_hop_gh_user_count', 'total_gh_user_at_the_time_count', 'one_hop_dev_count', 'two_hop_dev_count', 'three_hop_dev_count', 'fourormore_hop_dev_count', 'one_hop_stargazer_count', 'two_hop_stargazer_count', 'three_hop_stargazer_count', 'fourormore_hop_stargazer_count', 'rd_count', 'hn_count',  'tw_count', 'project_history_count'))

df_data$fourormore_hop_user_count = df_data$total_gh_user_at_the_time_count -  df_data$one_hop_gh_user_count - df_data$two_hop_gh_user_count - df_data$three_hop_gh_user_count

if (community_name == 'R'){
data_no_outlier = subset(df_data, tw_count < exp(4)
                                &  hn_count < exp(1)
                                &  rd_count < exp(1)
                                &  one_hop_stargazer_count < exp(2)
                                &  two_hop_stargazer_count < exp(3)
                                &  three_hop_stargazer_count < exp(3)
                                &  fourormore_hop_stargazer_count < exp(6)
                                &  one_hop_dev_count < exp(2)
                                &  two_hop_dev_count < exp(1.5)
                                &  three_hop_dev_count < exp(1.5)
                                &  fourormore_hop_dev_count < exp(3)
                                &  one_hop_gh_user_count < exp(8)
                                &  two_hop_gh_user_count < exp(12)
                                &  fourormore_hop_user_count > exp(13)
                                &  project_history_count > exp(2))
}else{

data_no_outlier = subset(df_data, tw_count < exp(5)
                                 & hn_count < exp(2)
                                 & rd_count < exp(2)
                                 & one_hop_stargazer_count < exp(2)
                                 & two_hop_stargazer_count < exp(4)
                                 & three_hop_stargazer_count < exp(4)
                                 & fourormore_hop_stargazer_count < exp(8)
                                 & one_hop_dev_count < exp(2)
                                 & two_hop_dev_count < exp(2)
                                 & three_hop_dev_count < exp(3) 
                                 & fourormore_hop_dev_count < exp(4)
                                 & one_hop_gh_user_count < exp(8)
                                 & two_hop_gh_user_count < exp(12)
                                 & fourormore_hop_user_count >= exp(10))

}
hist(log(df_data$tw_count+1))
hist(log(df_data$hn_count+1))
hist(log(df_data$rd_count+1))

hist(log(df_data$one_hop_stargazer_count+1))
hist(log(df_data$two_hop_stargazer_count+1))
hist(log(df_data$three_hop_stargazer_count+1))
hist(log(df_data$fourormore_hop_stargazer_count+1))

hist(log(df_data$one_hop_dev_count+1))
hist(log(df_data$two_hop_dev_count+1))
hist(log(df_data$three_hop_dev_count+1))
hist(log(df_data$fourormore_hop_dev_count+1))

hist(log(df_data$one_hop_gh_user_count+1))
hist(log(df_data$two_hop_gh_user_count+1))
hist(log(df_data$three_hop_gh_user_count+1))
hist(log(df_data$fourormore_hop_user_count+1))

hist(log(df_data$project_history_count+1))

# log scale the number of posts
data_no_outlier$tw = log(data_no_outlier$tw_count + 1)
data_no_outlier$hn = log(data_no_outlier$hn_count + 1)
data_no_outlier$rd = log(data_no_outlier$rd_count + 1)


```



### Generate Figure 5 in the paper, step 2, regression for stargazer
```{r}
legend_font_size = 28
title_size = 20
hjust_length = 1
y_tick_size = 18
x_tick_size = 26
axis_label_size = 22
legend_title_size = 32
width_middle = 1.0
width_adjust = 0.225
height_adjust = 0
one_hop_color = 'red'
two_hop_color = 'coral'
three_hop_color = 'gold'
fourormore_hop_color = 'darkgoldenrod'

significance_level_text_size = 10


generate_label <- function(p){
  label = ""
  if (p < 0.001)
  {
    label = '***'
  } else if (p < 0.01){
    label = '**'
  } else if (p < 0.05){
    label = '*'
  }else{
    label = ""
  }
  return(label)
}

one_hop_star = lm(log(one_hop_stargazer_count + 1) ~ 
                   + tw
                   + hn
                   + rd
                   + log(one_hop_gh_user_count + 1)
                   + log(two_hop_gh_user_count + 1)
                   + log(three_hop_gh_user_count + 1)
                   + log(fourormore_hop_user_count + 1)
                   + log(project_history_count + 1)
                   , data = data_no_outlier)

summary(one_hop_star)
bootresults = boot.relimp(one_hop_star, b = 100)
ci<-booteval.relimp(bootresults)
lower_percentile = c(ci$lmg.lower[1], ci$lmg.lower[2], ci$lmg.lower[3])
upper_percentile = c(ci$lmg.upper[1], ci$lmg.upper[2], ci$lmg.upper[3])
mean_est = c(ci$est['tw.lmg'], ci$est['hn.lmg'], ci$est['rd.lmg'])
type_name = c('TW', 'HN', 'RD')
df_ci_onehop_star = data.frame(lower_percentile, upper_percentile,mean_est,type_name)

two_hop_star = lm(log(two_hop_stargazer_count + 1) ~ 
                   + tw
                   + hn
                   + rd
                   + log(one_hop_gh_user_count + 1)
                   + log(two_hop_gh_user_count + 1)
                   + log(three_hop_gh_user_count + 1)
                   + log(fourormore_hop_user_count + 1)
                   + log(project_history_count + 1)
                   , data = data_no_outlier)

summary(two_hop_star)
bootresults = boot.relimp(two_hop_star, b = 100)
ci<-booteval.relimp(bootresults)
lower_percentile = c(ci$lmg.lower[1], ci$lmg.lower[2], ci$lmg.lower[3])
upper_percentile = c(ci$lmg.upper[1], ci$lmg.upper[2], ci$lmg.upper[3])
mean_est = c(ci$est['tw.lmg'], ci$est['hn.lmg'], ci$est['rd.lmg'])
type_name = c('TW', 'HN', 'RD')
df_ci_twohop_star = data.frame(lower_percentile, upper_percentile,mean_est,type_name)

three_hop_star = lm(log(three_hop_stargazer_count + 1) ~ 
                   + tw
                   + hn
                   + rd
                   + log(one_hop_gh_user_count + 1)
                   + log(two_hop_gh_user_count + 1)
                   + log(three_hop_gh_user_count + 1)
                   + log(fourormore_hop_user_count + 1)
                   + log(project_history_count + 1)
                   , data = data_no_outlier)

summary(three_hop_star)
bootresults = boot.relimp(three_hop_star, b = 100)
ci<-booteval.relimp(bootresults)

lower_percentile = c(ci$lmg.lower[1], ci$lmg.lower[2], ci$lmg.lower[3])
upper_percentile = c(ci$lmg.upper[1], ci$lmg.upper[2], ci$lmg.upper[3])
mean_est = c(ci$est['tw.lmg'], ci$est['hn.lmg'], ci$est['rd.lmg'])
type_name = c('TW', 'HN', 'RD')
df_ci_threehop_star = data.frame(lower_percentile, upper_percentile,mean_est,type_name)


fourormore_hop_star = lm(log(fourormore_hop_stargazer_count + 1) ~ 
                   + tw
                   + hn
                   + rd
                   + log(one_hop_gh_user_count + 1)
                   + log(two_hop_gh_user_count + 1)
                   + log(three_hop_gh_user_count + 1)
                   + log(fourormore_hop_user_count + 1)
                   + log(project_history_count + 1)
                   , data = data_no_outlier)

summary(fourormore_hop_star)

bootresults = boot.relimp(fourormore_hop_star, b = 100)
ci<-booteval.relimp(bootresults)

lower_percentile = c(ci$lmg.lower[1], ci$lmg.lower[2], ci$lmg.lower[3])
upper_percentile = c(ci$lmg.upper[1], ci$lmg.upper[2], ci$lmg.upper[3])
mean_est = c(ci$est['tw.lmg'], ci$est['hn.lmg'], ci$est['rd.lmg'])
type_name = c('TW', 'HN', 'RD')
df_ci_fourormore_star = data.frame(lower_percentile, upper_percentile,mean_est,type_name)



combined_table = combine(df_ci_onehop_star, df_ci_twohop_star,df_ci_threehop_star, df_ci_fourormore_star)
combined_table$user_distance = ""
combined_table[combined_table$source == 'df_ci_onehop_star',]$user_distance <- 'one hop'
combined_table[combined_table$source == 'df_ci_twohop_star',]$user_distance <- 'two hop'
combined_table[combined_table$source == 'df_ci_threehop_star',]$user_distance <- 'three hop'
combined_table[combined_table$source == 'df_ci_fourormore_star',]$user_distance <- 'four or more hop'

combined_table$type_name <- factor(combined_table$type_name, levels = c('TW', 'HN', 'RD', 'SO'))
combined_table$user_distance <- factor(combined_table$user_distance, levels = c('one hop', 'two hop', 'three hop', 'four or more hop'))
combined_table_tw = subset(combined_table, type_name == 'TW')
combined_table_hn = subset(combined_table, type_name == 'HN')
combined_table_rd = subset(combined_table, type_name == 'RD')

tw_onehop_star_p = summary(one_hop_star)$coefficient['tw', ][4]
tw_twohop_star_p = summary(two_hop_star)$coefficient['tw', ][4]
tw_threehop_star_p = summary(three_hop_star)$coefficient['tw', ][4]
tw_fourormore_star_p = summary(fourormore_hop_star)$coefficient['tw', ][4]

tw_onehop_star_label = generate_label(tw_onehop_star_p)
tw_twohop_star_label = generate_label(tw_twohop_star_p)
tw_threehop_star_label = generate_label(tw_threehop_star_p)
tw_fourormore_star_label = generate_label(tw_fourormore_star_p)

p1_tw_star = ggplot(combined_table_tw, aes(fill=user_distance, y=mean_est, x=type_name)) + 
    geom_bar(position="dodge", stat="identity")+
scale_fill_manual('user_distance', values = c('one hop' = one_hop_color, 'two hop' = two_hop_color, 'three hop' = three_hop_color, 'four or more hop' = fourormore_hop_color))+
      geom_errorbar(aes(ymin=lower_percentile, ymax=upper_percentile), width=.2,
                 position=position_dodge(.9)) +
  ylab('Ratio of variance explained')+
  theme(plot.title = element_text(hjust = 0.5,size = title_size),
        legend.text = element_text(size=legend_font_size),
        axis.text.y=element_text(size=y_tick_size),
        axis.text.x=element_text(size=x_tick_size),
        axis.title=element_text(size=axis_label_size),
        axis.title.x = element_blank(),
        legend.title=element_text(size=legend_title_size))+
  geom_text(aes(x = c(width_middle - 1.5 *width_adjust, width_middle - 0.5 * width_adjust, width_middle + 0.5 * width_adjust, width_middle + 1.5 * width_adjust), label = c(tw_onehop_star_label, tw_twohop_star_label, tw_threehop_star_label, tw_fourormore_star_label)), vjust = height_adjust, colour = "black", size = significance_level_text_size)+
guides(fill=guide_legend("user distance"))+ 
  scale_x_discrete(labels = c('Twitter'))



hn_onehop_star_p = summary(one_hop_star)$coefficient['hn', ][4]
hn_twohop_star_p = summary(two_hop_star)$coefficient['hn', ][4]
hn_threehop_star_p = summary(three_hop_star)$coefficient['hn', ][4]
hn_fourormore_hop_star_p = summary(fourormore_hop_star)$coefficient['hn', ][4]

hn_onehop_star_label = generate_label(hn_onehop_star_p)
hn_twohop_star_label = generate_label(hn_twohop_star_p)
hn_threehop_star_label = generate_label(hn_threehop_star_p)
hn_fourormore_hop_star_label = generate_label(hn_fourormore_hop_star_p)



p2_hn_star = ggplot(combined_table_hn, aes(fill=user_distance, y=mean_est, x=type_name)) + 
    geom_bar(position="dodge", stat="identity")+
scale_fill_manual('user_distance', values = c('one hop' = one_hop_color, 'two hop' = two_hop_color, 'three hop' = three_hop_color, 'four or more hop' = fourormore_hop_color))+
    geom_errorbar(aes(ymin=lower_percentile, ymax=upper_percentile), width=.2,
                 position=position_dodge(.9)) +
  theme(plot.title = element_text(hjust = 0.5,size = title_size),
        legend.text = element_text(size=legend_font_size),
        axis.text.y=element_text(size=y_tick_size),
        axis.text.x=element_text(size=x_tick_size),
        axis.title=element_text(size=axis_label_size),
        axis.title.y = element_blank(),
        axis.title.x = element_blank(),
        legend.title=element_text(size=legend_title_size))+
  geom_text(aes(x = c(width_middle - 1.5 *width_adjust, width_middle - 0.5 * width_adjust, width_middle + 0.5 * width_adjust, width_middle + 1.5 * width_adjust), label = c(hn_onehop_star_label, hn_twohop_star_label, hn_threehop_star_label, hn_fourormore_hop_star_label)), vjust = height_adjust, colour = "black", size = significance_level_text_size)+ 
  scale_x_discrete(labels = c('HackerNews'))




rd_onehop_star_p = summary(one_hop_star)$coefficient['rd', ][4]
rd_twohop_star_p = summary(two_hop_star)$coefficient['rd', ][4]
rd_threehop_star_p = summary(three_hop_star)$coefficient['rd', ][4]
rd_fourormorehop_star_p = summary(fourormore_hop_star)$coefficient['rd', ][4]

rd_onehop_star_label = generate_label(rd_onehop_star_p)
rd_twohop_star_label = generate_label(rd_twohop_star_p)
rd_threehop_star_label = generate_label(rd_threehop_star_p)
rd_fourormore_star_label = generate_label(rd_fourormorehop_star_p)


p3_rd_star = ggplot(combined_table_rd, aes(fill=user_distance, y=mean_est, x=type_name)) + 
    geom_bar(position="dodge", stat="identity")+
scale_fill_manual('user_distance', values = c('one hop' = one_hop_color, 'two hop' = two_hop_color, 'three hop' = three_hop_color, 'four or more hop' = fourormore_hop_color))+
    geom_errorbar(aes(ymin=lower_percentile, ymax=upper_percentile), width=.2,
                 position=position_dodge(.9)) +
  theme(plot.title = element_text(hjust = 0.5,size = title_size),
        legend.text = element_text(size=legend_font_size),
        axis.text.y=element_text(size=y_tick_size),
        axis.text.x=element_text(size=x_tick_size),
        axis.title=element_text(size=axis_label_size),
        axis.title.y = element_blank(),
        axis.title.x = element_blank(),
        legend.title=element_text(size=legend_title_size))+
  geom_text(aes(x = c(width_middle - 1.5 *width_adjust, width_middle - 0.5 * width_adjust, width_middle + 0.5 * width_adjust, width_middle + 1.5 * width_adjust), label = c(rd_onehop_star_label, rd_twohop_star_label, rd_threehop_star_label, rd_fourormore_star_label)), vjust = height_adjust, colour = "black", size = significance_level_text_size)+ 
  scale_x_discrete(labels = c('Reddit'))



```

### Generate Figure 5 in the paper, step 3, regression for new developer
```{r}
library(ggpubr)
one_hop_dev = lm(log(one_hop_dev_count + 1) ~ 
                   + tw
                   + hn
                   + rd
                   + log(one_hop_gh_user_count + 1)
                   + log(two_hop_gh_user_count + 1)
                   + log(three_hop_gh_user_count + 1)
                   + log(fourormore_hop_user_count + 1)
                   + log(project_history_count + 1)
                   , data = data_no_outlier)

summary(one_hop_dev)
bootresults = boot.relimp(one_hop_dev, b = 100)
ci<-booteval.relimp(bootresults)
lower_percentile = c(ci$lmg.lower[1], ci$lmg.lower[2], ci$lmg.lower[3])
upper_percentile = c(ci$lmg.upper[1], ci$lmg.upper[2], ci$lmg.upper[3])
mean_est = c(ci$est['tw.lmg'], ci$est['hn.lmg'], ci$est['rd.lmg'])
type_name = c('TW', 'HN', 'RD')
df_ci_onehop_dev = data.frame(lower_percentile, upper_percentile,mean_est,type_name)

two_hop_dev = lm(log(two_hop_dev_count + 1) ~
                   + tw
                   + hn
                   + rd
                   + log(one_hop_gh_user_count + 1)
                   + log(two_hop_gh_user_count + 1)
                   + log(three_hop_gh_user_count + 1)
                   + log(fourormore_hop_user_count + 1)
                   + log(project_history_count + 1)
                   , data = data_no_outlier)

summary(two_hop_dev)
bootresults = boot.relimp(two_hop_dev, b = 100)
ci<-booteval.relimp(bootresults)
lower_percentile = c(ci$lmg.lower[1], ci$lmg.lower[2], ci$lmg.lower[3])
upper_percentile = c(ci$lmg.upper[1], ci$lmg.upper[2], ci$lmg.upper[3])
mean_est = c(ci$est['tw.lmg'], ci$est['hn.lmg'], ci$est['rd.lmg'])
type_name = c('TW', 'HN', 'RD')
df_ci_twohop_dev = data.frame(lower_percentile, upper_percentile,mean_est,type_name)

three_hop_dev = lm(log(three_hop_dev_count + 1) ~ 
                   + tw
                   + hn
                   + rd
                   + log(one_hop_gh_user_count + 1)
                   + log(two_hop_gh_user_count + 1)
                   + log(three_hop_gh_user_count + 1)
                   + log(fourormore_hop_user_count + 1)
                   + log(project_history_count + 1)
                   , data = data_no_outlier)

summary(three_hop_dev)
bootresults = boot.relimp(three_hop_dev, b = 100)
ci<-booteval.relimp(bootresults)
lower_percentile = c(ci$lmg.lower[1], ci$lmg.lower[2], ci$lmg.lower[3])
upper_percentile = c(ci$lmg.upper[1], ci$lmg.upper[2], ci$lmg.upper[3])
mean_est = c(ci$est['tw.lmg'], ci$est['hn.lmg'], ci$est['rd.lmg'])
type_name = c('TW', 'HN', 'RD')
df_ci_three_hop_dev = data.frame(lower_percentile, upper_percentile,mean_est,type_name)

fourormore_hop_dev = lm(log(fourormore_hop_dev_count + 1) ~ 
                   + tw
                   + hn
                   + rd
                   + log(one_hop_gh_user_count + 1)
                   + log(two_hop_gh_user_count + 1)
                   + log(three_hop_gh_user_count + 1)
                   + log(fourormore_hop_user_count + 1)
                   + log(project_history_count + 1)
                   , data = data_no_outlier)

summary(fourormore_hop_dev)
bootresults = boot.relimp(fourormore_hop_dev, b = 100)
ci<-booteval.relimp(bootresults)
lower_percentile = c(ci$lmg.lower[1], ci$lmg.lower[2], ci$lmg.lower[3])
upper_percentile = c(ci$lmg.upper[1], ci$lmg.upper[2], ci$lmg.upper[3])
mean_est = c(ci$est['tw.lmg'], ci$est['hn.lmg'], ci$est['rd.lmg'])
type_name = c('TW', 'HN', 'RD')
df_ci_fourormore_hop_dev = data.frame(lower_percentile, upper_percentile,mean_est,type_name)

combined_table = combine(df_ci_onehop_dev,df_ci_twohop_dev, df_ci_three_hop_dev, df_ci_fourormore_hop_dev)

combined_table$user_distance = ""
combined_table[combined_table$source == 'df_ci_onehop_dev',]$user_distance <- 'one hop'
combined_table[combined_table$source == 'df_ci_twohop_dev',]$user_distance <- 'two hop'
combined_table[combined_table$source == 'df_ci_three_hop_dev',]$user_distance <- 'three hop'
combined_table[combined_table$source == 'df_ci_fourormore_hop_dev',]$user_distance <- 'four or more hop'
combined_table$type_name <- factor(combined_table$type_name, levels = c('TW', 'HN', 'RD'))
combined_table$user_distance <- factor(combined_table$user_distance, levels = c('one hop', 'two hop', 'three hop', 'four or more hop'))
combined_table_tw = subset(combined_table, type_name == 'TW')
combined_table_hn = subset(combined_table, type_name == 'HN')
combined_table_rd = subset(combined_table, type_name == 'RD')

tw_onehop_dev_p = summary(one_hop_dev)$coefficient['tw', ][4]
tw_twohop_dev_p = summary(two_hop_dev)$coefficient['tw', ][4]
tw_threehop_dev_p = summary(three_hop_dev)$coefficient['tw', ][4]
tw_fourormore_hop_dev_p = summary(fourormore_hop_dev)$coefficient['tw', ][4]

tw_onehop_dev_label = generate_label(tw_onehop_dev_p)
tw_twohop_dev_label = generate_label(tw_twohop_dev_p)
tw_threehop_dev_label = generate_label(tw_threehop_dev_p)
tw_fourormore_hop_dev_label = generate_label(tw_fourormore_hop_dev_p)


p4_tw_dev = ggplot(combined_table_tw, aes(fill=user_distance, y=mean_est, x=type_name)) + 
    geom_bar(position="dodge", stat="identity")+
scale_fill_manual('user_distance', values = c('one hop' = one_hop_color, 'two hop' = two_hop_color, 'three hop' = three_hop_color, 'four or more hop' = fourormore_hop_color ))+
      geom_errorbar(aes(ymin=lower_percentile, ymax=upper_percentile), width=.2,
                 position=position_dodge(.9)) +
  ylab('Ratio of variance explained')+
  theme(plot.title = element_text(hjust = 0.5,size = title_size),
        legend.text = element_text(size=legend_font_size),
        axis.text.y=element_text(size=y_tick_size),
        axis.text.x=element_text(size=x_tick_size),
        axis.title=element_text(size=axis_label_size),
        axis.title.x = element_blank(),
        legend.title=element_text(size=legend_title_size))+
     geom_text(aes(x = c(width_middle - 1.5 *width_adjust, width_middle - 0.5 * width_adjust, width_middle + 0.5 * width_adjust, width_middle + 1.5 * width_adjust), label = c(tw_onehop_dev_label, tw_twohop_dev_label, tw_threehop_dev_label, tw_fourormore_hop_dev_label)), vjust = height_adjust, colour = "black", size = significance_level_text_size)+ 
     guides(fill=guide_legend("user distance"))+ 
  scale_x_discrete(labels = c( 'Twitter'))


hn_onehop_dev_p = summary(one_hop_dev)$coefficient['hn', ][4]
hn_twohop_dev_p = summary(two_hop_dev)$coefficient['hn', ][4]
hn_threehop_dev_p = summary(three_hop_dev)$coefficient['hn', ][4]

hn_fourormore_hop_dev_p = summary(fourormore_hop_dev)$coefficient['hn', ][4]

hn_onehop_dev_label = generate_label(hn_onehop_dev_p)
hn_twohop_dev_label = generate_label(hn_twohop_dev_p)
hn_threehop_dev_label = generate_label(hn_threehop_dev_p)

hn_fourormore_hop_dev_label = generate_label(hn_fourormore_hop_dev_p)


p5_hn_dev = ggplot(combined_table_hn, aes(fill=user_distance, y=mean_est, x=type_name)) + 
    geom_bar(position="dodge", stat="identity")+
scale_fill_manual('user_distance', values = c('one hop' = one_hop_color, 'two hop' = two_hop_color, 'three hop' = three_hop_color, 'four or more hop' = fourormore_hop_color ))+
      geom_errorbar(aes(ymin=lower_percentile, ymax=upper_percentile), width=.2,
                 position=position_dodge(.9)) +
  theme(plot.title = element_text(hjust = 0.5,size = title_size),
        legend.text = element_text(size=legend_font_size),
        axis.text.y=element_text(size=y_tick_size),
        axis.text.x=element_text(size=x_tick_size),
        axis.title=element_text(size=axis_label_size),
        axis.title.y = element_blank(),
        axis.title.x = element_blank(),
        legend.title=element_text(size=legend_title_size))+
     geom_text(aes(x = c(width_middle - 1.5 *width_adjust, width_middle - 0.5 * width_adjust, width_middle + 0.5 * width_adjust, width_middle + 1.5 * width_adjust), label = c(hn_onehop_dev_label, hn_twohop_dev_label,  hn_threehop_dev_label, hn_fourormore_hop_dev_label)), vjust = height_adjust, colour = "black", size = significance_level_text_size)+ 
  scale_x_discrete(labels = c( 'HackerNews'))


rd_onehop_dev_p = summary(one_hop_dev)$coefficient['rd', ][4]
rd_twohop_dev_p = summary(two_hop_dev)$coefficient['rd', ][4]
rd_threehop_dev_p = summary(three_hop_dev)$coefficient['rd', ][4]
rd_fourormore_hop_dev_p = summary(fourormore_hop_dev)$coefficient['rd', ][4]

rd_onehop_dev_label = generate_label(rd_onehop_dev_p)
rd_twohop_dev_label = generate_label(rd_twohop_dev_p)
rd_threehop_dev_label = generate_label(rd_threehop_dev_p)
rd_fourormore_hop_dev_label = generate_label(rd_fourormore_hop_dev_p)

p6_rd_dev = ggplot(combined_table_rd, aes(fill=user_distance, y=mean_est, x=type_name)) + 
    geom_bar(position="dodge", stat="identity")+
scale_fill_manual('user_distance', values = c('one hop' = one_hop_color, 'two hop' = two_hop_color, 'three hop' = three_hop_color, 'four or more hop' = fourormore_hop_color ))+
      geom_errorbar(aes(ymin=lower_percentile, ymax=upper_percentile), width=.2,
                 position=position_dodge(.9)) +
  theme(plot.title = element_text(hjust = 0.5,size = title_size),
        legend.text = element_text(size=legend_font_size),
        axis.text.y=element_text(size=y_tick_size),
        axis.text.x=element_text(size=x_tick_size),
        axis.title=element_text(size=axis_label_size),
        axis.title.y = element_blank(),
        axis.title.x = element_blank(),
        legend.title=element_text(size=legend_title_size))+
     geom_text(aes(x = c(width_middle - 1.5 *width_adjust, width_middle - 0.5 * width_adjust, width_middle + 0.5 * width_adjust, width_middle + 1.5 * width_adjust), label = c(rd_onehop_dev_label, rd_twohop_dev_label, rd_threehop_dev_label, rd_fourormore_hop_dev_label)), vjust = height_adjust, colour = "black", size = significance_level_text_size)+ 
  scale_x_discrete(labels = c( 'Reddit'))



p_star = ggarrange(p1_tw_star, p2_hn_star,p3_rd_star, ncol=3, nrow=1, common.legend = TRUE, legend="right") 
p_star = annotate_figure(p_star, top = text_grob("Stargazer", 
               color = "black", face = "bold", size = 36, hjust = hjust_length))

p_dev = ggarrange(p4_tw_dev, p5_hn_dev,p6_rd_dev, ncol=3, nrow=1, common.legend = TRUE, legend="right") 
p_dev = annotate_figure(p_dev, top = text_grob("Contributor", 
               color = "black", face = "bold", size = 36, hjust = hjust_length))

ggarrange(p_star, p_dev, ncol=1, nrow=2, common.legend = TRUE, legend="right")


ggsave(file=sprintf('./figure/association_%s.pdf', community_name), width=18, height=9, dpi=300)

```
### Generate table for regression result (not shown in the paper because of space limit)
```{r}

library(stargazer)
stargazer(one_hop_star, two_hop_star, three_hop_star, fourormore_hop_star,one_hop_dev, two_hop_dev, three_hop_dev, fourormore_hop_dev, omit = c('Constant'), digits = 2, omit.stat = c('ser', 'f', 'rsq'), star.cutoffs = c(.05, .01, .001), star.char = c("*", "**", "***"),
          covariate.labels = c('Twitter post (log)', 'HackerNews post (log)', 'Reddit post (log)', 'One-hop developer (log)', 'Two-hop developer (log)', 'Three-hop developer (log)', 'Four or more-hop developer (log)', 'Project age (log)'))


```





### Generate Figure 6 in the paper
```{r}

legend_font_size = 28
title_size = 40
tick_size = 32
axis_label_size = 28
legend_title_size = 32
plot_line_size = 3
coef_hn_rd_so_star = 1
coef_hn_rd_so_dev = 1

onehop_dev_explained_tw = subset(df_ci_onehop_dev, type_name == 'TW')$mean_est[1] / sum(df_ci_onehop_dev$mean_est)
onehop_dev_explained_hn = subset(df_ci_onehop_dev, type_name == 'HN')$mean_est[1] / sum(df_ci_onehop_dev$mean_est) * coef_hn_rd_so_dev
onehop_dev_explained_rd = subset(df_ci_onehop_dev, type_name == 'RD')$mean_est[1] / sum(df_ci_onehop_dev$mean_est) * coef_hn_rd_so_dev

relative_importance = c(onehop_dev_explained_tw, onehop_dev_explained_hn, onehop_dev_explained_rd)
hop = c('one', 'one', 'one')
media = c('Twitter', 'HackerNews', 'Reddit')
user_type = c('dev', 'dev', 'dev')
df_onehop_dev_ri = data.frame(relative_importance, hop,media, user_type)

two_hop_dev_explained_tw = subset(df_ci_twohop_dev, type_name == 'TW')$mean_est[1] / sum(df_ci_twohop_dev$mean_est)
two_hop_dev_explained_hn = subset(df_ci_twohop_dev, type_name == 'HN')$mean_est[1] / sum(df_ci_twohop_dev$mean_est) * coef_hn_rd_so_dev 
two_hop_dev_explained_rd = subset(df_ci_twohop_dev, type_name == 'RD')$mean_est[1] / sum(df_ci_twohop_dev$mean_est) * coef_hn_rd_so_dev

relative_importance = c(two_hop_dev_explained_tw, two_hop_dev_explained_hn, two_hop_dev_explained_rd)
hop = c('two', 'two', 'two')
media = c('Twitter', 'HackerNews', 'Reddit')
user_type = c('dev', 'dev', 'dev')
df_two_hop_dev_ri = data.frame(relative_importance, hop,media, user_type)



three_hop_dev_explained_tw = subset(df_ci_three_hop_dev, type_name == 'TW')$mean_est[1] / sum(df_ci_three_hop_dev$mean_est)
three_hop_dev_explained_hn = subset(df_ci_three_hop_dev, type_name == 'HN')$mean_est[1] / sum(df_ci_three_hop_dev$mean_est) * coef_hn_rd_so_dev 
three_hop_dev_explained_rd = subset(df_ci_three_hop_dev, type_name == 'RD')$mean_est[1] / sum(df_ci_three_hop_dev$mean_est) * coef_hn_rd_so_dev


relative_importance = c(three_hop_dev_explained_tw, three_hop_dev_explained_hn, three_hop_dev_explained_rd)
hop = c('three', 'three', 'three')
media = c('Twitter', 'HackerNews', 'Reddit')
user_type = c('dev', 'dev', 'dev')
df_three_hop_dev_ri = data.frame(relative_importance, hop,media, user_type)


fourormore_hop_dev_explained_tw = subset(df_ci_fourormore_hop_dev, type_name == 'TW')$mean_est[1] / sum(df_ci_fourormore_hop_dev$mean_est)
fourormore_hop_dev_explained_hn = subset(df_ci_fourormore_hop_dev, type_name == 'HN')$mean_est[1] / sum(df_ci_fourormore_hop_dev$mean_est) * coef_hn_rd_so_dev 
fourormore_hop_dev_explained_rd = subset(df_ci_fourormore_hop_dev, type_name == 'RD')$mean_est[1] / sum(df_ci_fourormore_hop_dev$mean_est) * coef_hn_rd_so_dev

relative_importance = c(fourormore_hop_dev_explained_tw, fourormore_hop_dev_explained_hn, fourormore_hop_dev_explained_rd)
hop = c('four or more', 'four or more', 'four or more')
media = c('Twitter', 'HackerNews', 'Reddit')
user_type = c('dev', 'dev', 'dev')
df_fourormore_hop_dev_ri = data.frame(relative_importance, hop,media, user_type)




onehop_star_explained_tw = subset(df_ci_onehop_star, type_name == 'TW')$mean_est[1] / sum(df_ci_onehop_star$mean_est)
onehop_star_explained_hn = subset(df_ci_onehop_star, type_name == 'HN')$mean_est[1] / sum(df_ci_onehop_star$mean_est) * coef_hn_rd_so_star
onehop_star_explained_rd = subset(df_ci_onehop_star, type_name == 'RD')$mean_est[1] / sum(df_ci_onehop_star$mean_est) * coef_hn_rd_so_star

relative_importance = c(onehop_star_explained_tw, onehop_star_explained_hn, onehop_star_explained_rd)
hop = c('one', 'one', 'one')
media = c('Twitter', 'HackerNews', 'Reddit')
user_type = c('star', 'star', 'star')
df_onehop_star_ri = data.frame(relative_importance, hop,media, user_type)



twohop_star_explained_tw = subset(df_ci_twohop_star, type_name == 'TW')$mean_est[1] / sum(df_ci_twohop_star$mean_est)
twohop_star_explained_hn = subset(df_ci_twohop_star, type_name == 'HN')$mean_est[1] / sum(df_ci_twohop_star$mean_est) * coef_hn_rd_so_star
twohop_star_explained_rd = subset(df_ci_twohop_star, type_name == 'RD')$mean_est[1] / sum(df_ci_twohop_star$mean_est) * coef_hn_rd_so_star

relative_importance = c(twohop_star_explained_tw, twohop_star_explained_hn, twohop_star_explained_rd)
hop = c('two', 'two', 'two')
media = c('Twitter', 'HackerNews', 'Reddit')
user_type = c('star', 'star', 'star')
df_twohop_star_ri = data.frame(relative_importance, hop,media, user_type)



threehop_star_explained_tw = subset(df_ci_threehop_star, type_name == 'TW')$mean_est[1] / sum(df_ci_threehop_star$mean_est)
threehop_star_explained_hn = subset(df_ci_threehop_star, type_name == 'HN')$mean_est[1] / sum(df_ci_threehop_star$mean_est) * coef_hn_rd_so_star
threehop_star_explained_rd = subset(df_ci_threehop_star, type_name == 'RD')$mean_est[1] / sum(df_ci_threehop_star$mean_est) * coef_hn_rd_so_star


relative_importance = c(threehop_star_explained_tw, threehop_star_explained_hn, threehop_star_explained_rd)
hop = c('three', 'three', 'three')
media = c('Twitter', 'HackerNews', 'Reddit')
user_type = c('star', 'star', 'star')
df_threehop_star_ri = data.frame(relative_importance, hop,media, user_type)



fourormore_hop_star_explained_tw = subset(df_ci_fourormore_star, type_name == 'TW')$mean_est[1] / sum(df_ci_fourormore_star$mean_est)
fourormore_hop_star_explained_hn = subset(df_ci_fourormore_star, type_name == 'HN')$mean_est[1] / sum(df_ci_fourormore_star$mean_est) * coef_hn_rd_so_star
fourormore_hop_star_explained_rd = subset(df_ci_fourormore_star, type_name == 'RD')$mean_est[1] / sum(df_ci_fourormore_star$mean_est) * coef_hn_rd_so_star


relative_importance = c(fourormore_hop_star_explained_tw, fourormore_hop_star_explained_hn, fourormore_hop_star_explained_rd)
hop = c('four or more', 'four or more', 'four or more')
media = c('Twitter', 'HackerNews', 'Reddit')
user_type = c('star', 'star', 'star')
df_fourormore_hop_star_ri = data.frame(relative_importance, hop,media, user_type)



df = rbind(df_onehop_dev_ri,df_two_hop_dev_ri, df_three_hop_dev_ri, df_fourormore_hop_dev_ri, df_onehop_star_ri, df_twohop_star_ri, df_threehop_star_ri, df_fourormore_hop_star_ri)

df$hop <- factor(df$hop, levels = c('one','two', 'three', 'four or more'))
df$media <- factor(df$media, levels = c('Twitter', 'HackerNews', 'Reddit'))

df_star = subset(df, user_type == 'star')
df_dev = subset(df, user_type == 'dev')

p1_star = ggplot(df_star,aes(x=hop, y=relative_importance, group=media, color=media))+
  geom_line(size = plot_line_size) +
    scale_colour_manual(values = c('Twitter' = "blue",'HackerNews' = "red", 'Reddit' = "orange"))+
    scale_y_continuous(

    # Features of the first axis
    name = "Ratio of variance explained",

    # Add a second axis and specify its features
    # sec.axis = sec_axis(~.*1.0 / coef_hn_rd_so_star, name="Ratio of variance explained for HN, RD, SO")
  )+
  xlab('Stargazer network distance (in hops)')+
  theme(plot.title = element_text(hjust = 0.5,size = title_size, face="bold"),
        legend.text = element_text(size=legend_font_size),
        axis.text=element_text(size=tick_size),
        axis.title=element_text(size=axis_label_size),
        legend.title=element_text(size=legend_title_size)) +
   ggtitle('Stargazer')

p1_dev = ggplot(df_dev,aes(x=hop, y=relative_importance, group=media, color=media))+
  geom_line(size = plot_line_size) +
    scale_colour_manual(values = c('Twitter' = "blue",'HackerNews' = "red", 'Reddit' = "orange"))+
    scale_y_continuous(

    # Features of the first axis
    name = "Ratio of variance explained",

    # Add a second axis and specify its features
    # sec.axis = sec_axis(~.*1.0 / coef_hn_rd_so_dev, name="Percentage variance for HN, RD, SO")
  ) +
  # ggtitle("R projects")+
  xlab('New contributor network distance (in hops)')+
  theme(plot.title = element_text(hjust = 0.5,size = title_size, face="bold"),
        legend.text = element_text(size=legend_font_size),
        axis.text=element_text(size=tick_size),
        axis.title=element_text(size=axis_label_size),
        legend.title=element_text(size=legend_title_size))+
   ggtitle('Contributor')

ggarrange(p1_star, p1_dev, ncol=2, nrow=1, common.legend = TRUE, legend="right") 

ggsave(file=sprintf('./figure/relative_importance_%s.pdf', community_name), width=24, height=6, dpi=300)

```