In [11]:
library(data.table)
In [12]:
library(fmsb)
In [13]:
library(ggplot2)
In [14]:
longitudinal_edits <- read.table("../../../results/misalignment_and_edits_3_22_18_post_processed.txt", header=TRUE, sep="\t")
In [15]:
head(longitudinal_edits, n=60)
In [16]:
longitudinal_edits$human_non_bot_like_edit = longitudinal_edits$human_edit - longitudinal_edits$human_bot_like_edit;
longitudinal_edits$anon_non_bot_like_edit = longitudinal_edits$anon_edit - longitudinal_edits$anon_bot_like_edit;
longitudinal_edits$semi_automated_non_bot_like_edit = longitudinal_edits$semi_automated_edit - longitudinal_edits$semi_automated_bot_like_edit;
longitudinal_edits$total_edits = longitudinal_edits$bot_edit + longitudinal_edits$semi_automated_edit + longitudinal_edits$human_edit + longitudinal_edits$anon_edit
longitudinal_edits$bot_edit_prop = longitudinal_edits$bot_edit / longitudinal_edits$total_edits
longitudinal_edits$semi_automated_bot_like_edit_prop = longitudinal_edits$semi_automated_bot_like_edit / longitudinal_edits$total_edits
longitudinal_edits$human_bot_like_edit_prop = longitudinal_edits$human_bot_like_edit / longitudinal_edits$total_edits
longitudinal_edits$anon_bot_like_edit_prop = longitudinal_edits$anon_bot_like_edit / longitudinal_edits$total_edits
longitudinal_edits$human_non_bot_like_edit_prop = longitudinal_edits$human_non_bot_like_edit / longitudinal_edits$total_edits
longitudinal_edits$anon_non_bot_like_edit_prop = longitudinal_edits$anon_non_bot_like_edit / longitudinal_edits$total_edits
longitudinal_edits$semi_automated_non_bot_like_edit_prop = longitudinal_edits$semi_automated_non_bot_like_edit / longitudinal_edits$total_edits
longitudinal_edits$semi_automated_edit_prop = longitudinal_edits$semi_automated_edit / longitudinal_edits$total_edits
longitudinal_edits$human_bot_like_over_human_edit_prop = longitudinal_edits$human_bot_like_edit / longitudinal_edits$human_edit
In [17]:
head(longitudinal_edits, n=60)
In [18]:
# detach(longitudinal_edits)
attach(longitudinal_edits)
In [56]:
agent_type_regression <- lm(change_of_mean_error ~ scale(bot_edit) +
scale(semi_automated_edit) +
scale(human_non_bot_like_edit) +
scale(anon_non_bot_like_edit) +
scale(human_bot_like_edit) +
scale(anon_bot_like_edit)
);
# scale(semi_automated_bot_like_edit) +
# scale(semi_automated_non_bot_like_edit)
# NOTE: When replacing "semi_automated_edit" with the two above, semi_automated_non_bot is significatn with a positive value.
# How do bot-like behaviors influence changes in misalignment
In [57]:
summary(agent_type_regression)
In [42]:
agent_type_regression <- lm(change_of_mean_error ~ scale(bot_edit_prop) +
scale(semi_automated_edit_prop) +
scale(anon_non_bot_like_edit_prop) +
scale(human_bot_like_edit_prop) +
scale(anon_bot_like_edit_prop)
);
# scale(human_non_bot_like_edit_prop)
# scale(semi_automated_bot_like_edit_prop) +
# How do bot-like behaviors influence changes in misalignment
In [43]:
summary(agent_type_regression)
In [ ]:
In [ ]:
In [ ]:
In [ ]:
In [11]:
summary(lm(bot_edits ~
scale(semi_automated_edits) +
scale(non_bot_edits) +
scale(anon_edits) +
scale(under_five_seconds) +
scale(five_to_ten_seconds) +
scale(ten_to_twenty_seconds) +
scale(twenty_to_one_hundred_seconds) +
scale(over_one_hundred_seconds))
)
#VIF is 1.37
In [64]:
summary(lm(semi_automated_edits ~ scale(bot_edits) +
scale(non_bot_edits) +
scale(anon_edits) +
scale(under_five_seconds) +
scale(five_to_ten_seconds) +
scale(ten_to_twenty_seconds) +
scale(twenty_to_one_hundred_seconds) +
scale(over_one_hundred_seconds))
)
#Vif is 2.22
In [65]:
summary(lm(non_bot_edits ~ scale(bot_edits) +
scale(semi_automated_edits) +
scale(anon_edits) +
scale(under_five_seconds) +
scale(five_to_ten_seconds) +
scale(ten_to_twenty_seconds) +
scale(twenty_to_one_hundred_seconds) +
scale(over_one_hundred_seconds))
)
#vif is 2.83
In [66]:
summary(lm(anon_edits ~ scale(bot_edits) +
scale(semi_automated_edits) +
scale(non_bot_edits) +
scale(under_five_seconds) +
scale(five_to_ten_seconds) +
scale(ten_to_twenty_seconds) +
scale(twenty_to_one_hundred_seconds) +
scale(over_one_hundred_seconds))
)
#Vif is 1.31
In [83]:
summary(lm(under_five_seconds ~ scale(bot_edits) +
scale(semi_automated_edits) +
scale(non_bot_edits) +
scale(anon_edits) +
scale(five_to_ten_seconds) +
scale(ten_to_twenty_seconds) +
scale(twenty_to_one_hundred_seconds) +
scale(over_one_hundred_seconds))
)
#Vif is 1.05
In [68]:
summary(lm(five_to_ten_seconds ~ scale(bot_edits) +
scale(semi_automated_edits) +
scale(non_bot_edits) +
scale(anon_edits) +
scale(under_five_seconds) +
scale(ten_to_twenty_seconds) +
scale(twenty_to_one_hundred_seconds) +
scale(over_one_hundred_seconds))
)
#Vif is 1.02
In [69]:
summary(lm(ten_to_twenty_seconds ~ scale(bot_edits) +
scale(semi_automated_edits) +
scale(non_bot_edits) +
scale(anon_edits) +
scale(under_five_seconds) +
scale(five_to_ten_seconds) +
scale(twenty_to_one_hundred_seconds) +
scale(over_one_hundred_seconds))
)
#vif is 1.22
In [70]:
summary(lm(twenty_to_one_hundred_seconds ~ scale(bot_edits) +
scale(semi_automated_edits) +
scale(non_bot_edits) +
scale(anon_edits) +
scale(under_five_seconds) +
scale(five_to_ten_seconds) +
scale(ten_to_twenty_seconds) +
scale(over_one_hundred_seconds))
)
#vif is 7.33
In [71]:
summary(lm(over_one_hundred_seconds ~ scale(bot_edits) +
scale(semi_automated_edits) +
scale(non_bot_edits) +
scale(anon_edits) +
scale(under_five_seconds) +
scale(five_to_ten_seconds) +
scale(ten_to_twenty_seconds) +
scale(twenty_to_one_hundred_seconds))
)
# vif is 5.28
In [ ]:
In [119]:
independent_and_dependent_variables = data.table(bot_edits = bot_edits, semi_automated_edits = semi_automated_edits, non_bot_edits = non_bot_edits, anon_edits = anon_edits, difference_in_alignment_with_previous = difference_in_alignment_with_previous, under_five_seconds = under_five_seconds, five_to_ten_seconds_proportion = five_to_ten_seconds_proportion, ten_to_twenty_proportion = ten_to_twenty_proportion, twenty_to_one_hundred_seconds_proportion = twenty_to_one_hundred_seconds_proportion, over_one_hundred_seconds_proportion = over_one_hundred_seconds_proportion)
In [73]:
edit_type_regression_without_anon <- lm(difference_in_alignment_with_previous ~ scale(bot_edits) + scale(semi_automated_edits) + scale(non_bot_edits))
In [74]:
anon_residuals = data.frame(month=as.Date(paste(yyyymm, "01", sep=""), format="%Y%m%d"), anon_edits = anon_edits, residuals= edit_type_regression_without_anon$residuals)
In [75]:
summary(anon_residuals)
In [76]:
ggplot(anon_residuals, aes(x=month, y=scale(residuals))) + geom_bar(stat="identity") + geom_line(aes(y=scale(anon_edits)))
In [77]:
hist(scale(anon_residuals$residuals)- scale(anon_residuals$anon_edits))
In [78]:
plot(scale(anon_residuals$residuals), scale(anon_residuals$anon_edits))
In [120]:
cor(independent_and_dependent_variables, method="spearman")
In [80]:
VIF(edit_type_regression)
In [81]:
qqnorm(edit_type_regression$residuals)
In [ ]:
In [ ]:
In [ ]:
In [82]:
names(edit_type_regression)
In [37]:
length(longitudinal_edits$human_bot_like_over_human_edit_prop[longitudinal_edits$human_bot_like_over_human_edit_prop > .05])
In [26]:
ggplot(longitudinal_edits,
aes(x=human_bot_like_over_human_edit_prop)) +
geom_histogram(bins=100);
In [25]:
mean(longitudinal_edits$human_bot_like_over_human_edit_prop)
In [27]:
summary(longitudinal_edits$human_bot_like_over_human_edit_prop)