Analysis on Twitter data

Loading the libraries and Twitter data

rm(list=ls()) 
#setwd("D:/emre/SkyDrive/makale/Twitter/Incivility/Revisit/")
library("foreign")
library("sjmisc")
library("sjPlot")
library("lessR")

## 
## lessR 3.5.0      feedback: gerbing@pdx.edu      web: lessRstats.com
## -------------------------------------------------------------------
## 1. Read a text, Excel, SPSS, SAS or R data file from your computer
##    into the mydata data table, enter:  mydata <- Read()  
## 2. For a list of help topics and functions, enter:  Help()
## 3. Use theme function for global settings, Ex: theme(colors="gray")

## 
## Attaching package: 'lessR'

## The following object is masked from 'package:sjmisc':
## 
##     rec

library("pander")
twitter_data <- read.dta("D:/emre/SkyDrive/makale/Twitter/Incivility/Revisit/TwitterKasimRS12N.dta")

Cleaning data and formating the variables

Contains.Link <- twitter_data$Link
Contains.Retweet <- twitter_data$Retweet
Candidate.Age<- twitter_data$Aday_yas
Candidate.Party <- twitter_data$c_party
Previously.Candidate <- twitter_data$c_prevcandidate
Contains.Link[Contains.Link > 1] <- 1
twitter_data$Retweet[twitter_data$Retweet > 1] <- 1
Candidate.Age[Candidate.Age < 25] <- NA

Incivility <- ordered(twitter_data$Incivility, levels=c(0,1), labels=c("No", "Yes"))

Candidate.Party.CHP <- relevel(Candidate.Party, ref = "CHP")
Candidate.Party.MHP <- relevel(Candidate.Party, ref = "MHP")
Candidate.Party.AKP <- relevel(Candidate.Party, ref = "AKP")
Candidate.Party.HDP <- relevel(Candidate.Party, ref = "HDP")

Twitter specific model

twitter <- glm(Incivility ~  Candidate.Party+ Candidate.Age + 
                  Previously.Candidate + Contains.Retweet+
                  Contains.Link, data= twitter_data, family = binomial)
panderOptions("digits", 2)
pander(twitter)

Fitting generalized (binomial/logit) linear model: Incivility ~ Candidate.Party + Candidate.Age + Previously.Candidate + Contains.Retweet + Contains.Link (continued below)
	Estimate	Std. Error	z value
Candidate.PartyCHP	0.51	0.15	3.5
Candidate.PartyMHP	1.8	0.11	16
Candidate.PartyHDP	1	0.12	8.6
Candidate.Age	0.016	0.0052	3
Previously.CandidateYes	-0.32	0.084	-3.8
Contains.Retweet	0.88	0.085	10
Contains.Link	-0.2	0.083	-2.4
(Intercept)	-4.4	0.27	-16

	Pr(>\|z\|)
Candidate.PartyCHP	0.00054
Candidate.PartyMHP	5.3e-58
Candidate.PartyHDP	0.0000000000000000061
Candidate.Age	0.0024
Previously.CandidateYes	0.00015
Contains.Retweet	0.00000000000000000000000021
Contains.Link	0.018
(Intercept)	2.3e-57

#With Tone
#twitterT <- glm(Incivility ~ Tone + Candidate.Party+ Candidate.Age + 
#                 Previously.Candidate + Contains.Retweet+
#                 Contains.Link, family = binomial)

#Party Comparisons
#twitter <- glm(Incivility ~ Candidate.Party.CHP+ Candidate.Age + 
                 #Previously.Candidate + Contains.Retweet+
                 #Contains.Link, family = binomial)

#twitter <- glm(Incivility ~ Candidate.Party.MHP+ Candidate.Age + 
                #Previously.Candidate + Contains.Retweet+
                 #Contains.Link, family = binomial)

Descriptive tables of Twitter data

sjp.frq(twitter_data$Tone, show.prc = T)

sjp.frq(twitter_data$Incivility, show.prc = T)

t1 <- sjt.glm(twitter, show.se = T, show.r2 = T,  use.viewer = TRUE)

## Waiting for profiling to be done...

pander(t1)

## Warning in pander.default(t1): No pander.method for "sjTable", reverting to
## default.No pander.method for "sjtglm", reverting to default.

page.style:

page.content:

	Incivility
	OR	CI	std. Error	p
(Intercept)	0.01	0.01 – 0.02	0.27	<.001
Candidate.Party
CHP	1.66	1.24 – 2.21	0.15	.001
MHP	6.01	4.84 – 7.50	0.11	<.001
HDP	2.78	2.21 – 3.51	0.12	<.001
Candidate.Age	1.02	1.01 – 1.03	0.01	.002
Previously.Candidate (Yes)	0.73	0.62 – 0.86	0.08	<.001
Contains.Retweet	2.41	2.04 – 2.85	0.08	<.001
Contains.Link	0.82	0.70 – 0.97	0.08	.018
Observations	11200
Pseudo-R²	R²_CS = .041 R²_N = .111 D = .053

output.complete:

	Incivility
	OR	CI	std. Error	p
(Intercept)	0.01	0.01 – 0.02	0.27	<.001
Candidate.Party
CHP	1.66	1.24 – 2.21	0.15	.001
MHP	6.01	4.84 – 7.50	0.11	<.001
HDP	2.78	2.21 – 3.51	0.12	<.001
Candidate.Age	1.02	1.01 – 1.03	0.01	.002
Previously.Candidate (Yes)	0.73	0.62 – 0.86	0.08	<.001
Contains.Retweet	2.41	2.04 – 2.85	0.08	<.001
Contains.Link	0.82	0.70 – 0.97	0.08	.018
Observations	11200
Pseudo-R²	R²_CS = .041 R²_N = .111 D = .053

knitr:

	Incivility
	OR	CI	std. Error	p
(Intercept)	0.01	0.01 – 0.02	0.27	<.001
Candidate.Party
CHP	1.66	1.24 – 2.21	0.15	.001
MHP	6.01	4.84 – 7.50	0.11	<.001
HDP	2.78	2.21 – 3.51	0.12	<.001
Candidate.Age	1.02	1.01 – 1.03	0.01	.002
Previously.Candidate (Yes)	0.73	0.62 – 0.86	0.08	<.001
Contains.Retweet	2.41	2.04 – 2.85	0.08	<.001
Contains.Link	0.82	0.70 – 0.97	0.08	.018
Observations	11200
Pseudo-R²	R²_CS = .041 R²_N = .111 D = .053

data:

coef.name	estimate1	ci.lo1	ci.hi1	p-value1	se1
(Intercept)	0.01	0.01	0.02	<.001	0.27
Candidate.PartyCHP	1.66	1.24	2.21	.001	0.15
Candidate.PartyMHP	6.01	4.84	7.50	<.001	0.11
Candidate.PartyHDP	2.78	2.21	3.51	<.001	0.12
Candidate.Age	1.02	1.01	1.03	.002	0.01
Previously.CandidateYes	0.73	0.62	0.86	<.001	0.08
Contains.Retweet	2.41	2.04	2.85	<.001	0.08
Contains.Link	0.82	0.70	0.97	.018	0.08

t2<-sjt.xtab(Incivility, Candidate.Party, use.viewer = TRUE,  show.col.prc = T, tdcol.col = "black")
pander(t2)

## Warning in pander.default(t2): No pander.method for "sjTable", reverting to
## default.No pander.method for "sjtxtab", reverting to default.

page.style:

page.content:

Incivility	Candidate.Party				Total
Incivility	AKP	CHP	MHP	HDP	Total
No	5792 97.1 %	2234 95.8 %	1849 85.2 %	2281 91.1 %	12156 93.8 %
Yes	170 2.9 %	98 4.2 %	320 14.8 %	224 8.9 %	812 6.3 %
Total	5962 100 %	2332 100 %	2169 100 %	2505 100 %	12968 100 %
Χ²=432.113 · df=3 · Φ_c=.183 · p<.001

output.complete:

Incivility	Candidate.Party				Total
Incivility	AKP	CHP	MHP	HDP	Total
No	5792 97.1 %	2234 95.8 %	1849 85.2 %	2281 91.1 %	12156 93.8 %
Yes	170 2.9 %	98 4.2 %	320 14.8 %	224 8.9 %	812 6.3 %
Total	5962 100 %	2332 100 %	2169 100 %	2505 100 %	12968 100 %
Χ²=432.113 · df=3 · Φ_c=.183 · p<.001

knitr:

Incivility	Candidate.Party				Total
Incivility	AKP	CHP	MHP	HDP	Total
No	5792 97.1 %	2234 95.8 %	1849 85.2 %	2281 91.1 %	12156 93.8 %
Yes	170 2.9 %	98 4.2 %	320 14.8 %	224 8.9 %	812 6.3 %
Total	5962 100 %	2332 100 %	2169 100 %	2505 100 %	12968 100 %
Χ²=432.113 · df=3 · Φ_c=.183 · p<.001

t3<-sjt.xtab(Incivility, Previously.Candidate, use.viewer = TRUE, show.col.prc = TRUE, tdcol.col = "black")
pander(t3)

## Warning in pander.default(t3): No pander.method for "sjTable", reverting to
## default.No pander.method for "sjtxtab", reverting to default.

page.style:

page.content:

Incivility	Previously.Candidate		Total
Incivility	No	Yes	Total
No	4875 93.1 %	7833 93.8 %	12708 93.6 %
Yes	360 6.9 %	514 6.2 %	874 6.5 %
Total	5235 100 %	8347 100 %	13582 100 %
Χ²=2.643 · df=1 · Φ=.014 · p=.104

output.complete:

Incivility	Previously.Candidate		Total
Incivility	No	Yes	Total
No	4875 93.1 %	7833 93.8 %	12708 93.6 %
Yes	360 6.9 %	514 6.2 %	874 6.5 %
Total	5235 100 %	8347 100 %	13582 100 %
Χ²=2.643 · df=1 · Φ=.014 · p=.104

knitr:

Incivility	Previously.Candidate		Total
Incivility	No	Yes	Total
No	4875 93.1 %	7833 93.8 %	12708 93.6 %
Yes	360 6.9 %	514 6.2 %	874 6.5 %
Total	5235 100 %	8347 100 %	13582 100 %
Χ²=2.643 · df=1 · Φ=.014 · p=.104

t4<-sjt.xtab(Incivility, Contains.Retweet, use.viewer = TRUE, show.col.prc = TRUE, tdcol.col = "black")
pander(t4)

## Warning in pander.default(t4): No pander.method for "sjTable", reverting to
## default.No pander.method for "sjtxtab", reverting to default.

page.style:

page.content:

Incivility	Contains.Retweet		Total
Incivility	0	1	Total
No	9430 95.2 %	3277 89.7 %	12707 93.7 %
Yes	478 4.8 %	378 10.3 %	856 6.3 %
Total	9908 100 %	3655 100 %	13563 100 %
Χ²=136.541 · df=1 · Φ=.101 · p<.001

output.complete:

Incivility	Contains.Retweet		Total
Incivility	0	1	Total
No	9430 95.2 %	3277 89.7 %	12707 93.7 %
Yes	478 4.8 %	378 10.3 %	856 6.3 %
Total	9908 100 %	3655 100 %	13563 100 %
Χ²=136.541 · df=1 · Φ=.101 · p<.001

knitr:

Incivility	Contains.Retweet		Total
Incivility	0	1	Total
No	9430 95.2 %	3277 89.7 %	12707 93.7 %
Yes	478 4.8 %	378 10.3 %	856 6.3 %
Total	9908 100 %	3655 100 %	13563 100 %
Χ²=136.541 · df=1 · Φ=.101 · p<.001

t5<-sjt.xtab(Incivility, Contains.Link, use.viewer = TRUE, show.col.prc = TRUE, tdcol.col = "black")
pander(t5)

## Warning in pander.default(t5): No pander.method for "sjTable", reverting to
## default.No pander.method for "sjtxtab", reverting to default.

page.style:

page.content:

Incivility	Contains.Link		Total
Incivility	0	1	Total
No	5497 93.3 %	6244 94.1 %	11741 93.8 %
Yes	394 6.7 %	390 5.9 %	784 6.2 %
Total	5891 100 %	6634 100 %	12525 100 %
Χ²=3.347 · df=1 · Φ=.017 · p=.067

output.complete:

Incivility	Contains.Link		Total
Incivility	0	1	Total
No	5497 93.3 %	6244 94.1 %	11741 93.8 %
Yes	394 6.7 %	390 5.9 %	784 6.2 %
Total	5891 100 %	6634 100 %	12525 100 %
Χ²=3.347 · df=1 · Φ=.017 · p=.067

knitr:

Incivility	Contains.Link		Total
Incivility	0	1	Total
No	5497 93.3 %	6244 94.1 %	11741 93.8 %
Yes	394 6.7 %	390 5.9 %	784 6.2 %
Total	5891 100 %	6634 100 %	12525 100 %
Χ²=3.347 · df=1 · Φ=.017 · p=.067

Plots of Twitter specific model

sjp.glm(twitter, sort.est= FALSE, title = "Incivility - Twitter Specific Model")

## Waiting for profiling to be done...

sjp.glm(twitter, type="slope")

sjp.glm(twitter, type="slope", show.ci = T)

sjp.glm(twitter, type="eff", show.ci = T)

sjp.glm(twitter, type="pred", show.ci = T, vars=  "Previously.Candidate")

Analysis on Newspaper data

Loading data

np_data <- read.dta("D:/emre/SkyDrive/makale/Twitter/Incivility/Revisit/GazeteKasimRS12.dta")

## Warning in `levels<-`(`*tmp*`, value = if (nl == nL) as.character(labels)
## else paste0(labels, : duplicated levels in factors are deprecated

Formating data

Newspaper.ideology <- np_data$np_pro
Message.sender <- np_data$sender_gov
Time.focus <- np_data$time_focus
Message.content <- np_data$content

Incivility <- ordered(np_data$incivility, levels=c(0,1), labels=c("No", "Yes"))


Newspaper.ideology.c <- relevel(np_data$np_pro, ref = "main stream")

Newspaper Specific Model

np <- glm(incivility ~ Newspaper.ideology.c  + Message.sender + 
            Time.focus + Message.content, data= np_data, family = binomial)
panderOptions("digits", 2)
pander(np)

Fitting generalized (binomial/logit) linear model: incivility ~ Newspaper.ideology.c + Message.sender + Time.focus + Message.content (continued below)
	Estimate	Std. Error	z value
Newspaper.ideology.canti-government	-0.093	0.27	-0.34
Newspaper.ideology.cpro-government	-0.17	0.34	-0.51
Message.sendergovernment	-0.96	0.28	-3.4
Time.focusRetrospective	0.97	0.19	5.1
Message.contentPolicy	-1.6	0.28	-5.6
(Intercept)	0.66	0.31	2.1

	Pr(>\|z\|)
Newspaper.ideology.canti-government	0.73
Newspaper.ideology.cpro-government	0.61
Message.sendergovernment	0.0006
Time.focusRetrospective	0.00000033
Message.contentPolicy	0.00000002
(Intercept)	0.032

Descriptive tables of Newspaper data

n1<-sjt.glm(np, use.viewer = TRUE, show.se = T, show.r2 = T)

## Waiting for profiling to be done...

pander(n1)

## Warning in pander.default(n1): No pander.method for "sjTable", reverting to
## default.No pander.method for "sjtglm", reverting to default.

page.style:

page.content:

	incivility
	OR	CI	std. Error	p
(Intercept)	1.94	1.07 – 3.61	0.31	.032
Newspaper.ideology.c
Newspaper.ideology.canti-government	0.91	0.53 – 1.57	0.27	.735
Newspaper.ideology.cpro-government	0.84	0.44 – 1.65	0.34	.607
Message.sendergovernment	0.38	0.22 – 0.66	0.28	.001
Time.focusRetrospective	2.65	1.82 – 3.85	0.19	<.001
Message.contentPolicy	0.21	0.12 – 0.36	0.28	<.001
Observations	655
Pseudo-R²	R²_CS = .150 R²_N = .210 D = .159

output.complete:

	incivility
	OR	CI	std. Error	p
(Intercept)	1.94	1.07 – 3.61	0.31	.032
Newspaper.ideology.c
Newspaper.ideology.canti-government	0.91	0.53 – 1.57	0.27	.735
Newspaper.ideology.cpro-government	0.84	0.44 – 1.65	0.34	.607
Message.sendergovernment	0.38	0.22 – 0.66	0.28	.001
Time.focusRetrospective	2.65	1.82 – 3.85	0.19	<.001
Message.contentPolicy	0.21	0.12 – 0.36	0.28	<.001
Observations	655
Pseudo-R²	R²_CS = .150 R²_N = .210 D = .159

knitr:

	incivility
	OR	CI	std. Error	p
(Intercept)	1.94	1.07 – 3.61	0.31	.032
Newspaper.ideology.c
Newspaper.ideology.canti-government	0.91	0.53 – 1.57	0.27	.735
Newspaper.ideology.cpro-government	0.84	0.44 – 1.65	0.34	.607
Message.sendergovernment	0.38	0.22 – 0.66	0.28	.001
Time.focusRetrospective	2.65	1.82 – 3.85	0.19	<.001
Message.contentPolicy	0.21	0.12 – 0.36	0.28	<.001
Observations	655
Pseudo-R²	R²_CS = .150 R²_N = .210 D = .159

data:

Table continues below
coef.name	estimate1	ci.lo1	ci.hi1
(Intercept)	1.94	1.07	3.61
Newspaper.ideology.canti-government	0.91	0.53	1.57
Newspaper.ideology.cpro-government	0.84	0.44	1.65
Message.sendergovernment	0.38	0.22	0.66
Time.focusRetrospective	2.65	1.82	3.85
Message.contentPolicy	0.21	0.12	0.36

p-value1	se1
.032	0.31
.735	0.27
.607	0.34
.001	0.28
<.001	0.19
<.001	0.28

Plots of Newspaper Specific Model

sjp.glm(np, sort.est= FALSE, title = "Incivility - Newspaper Specific Model")

## Waiting for profiling to be done...

———Comparison Models———

Comparison Model Newspapers

np_comp <- glm(incivility ~  Message.sender + 
                 Time.focus + Message.content, data= np_data, family = binomial)
#Table
sjt.glm(np_comp, use.viewer = TRUE, show.se = T, show.r2 = T)

## Waiting for profiling to be done...

panderOptions("digits", 2)
pander(np_comp)

Fitting generalized (binomial/logit) linear model: incivility ~ Message.sender + Time.focus + Message.content
	Estimate	Std. Error	z value	Pr(>\|z\|)
Message.sendergovernment	-1	0.2	-5.2	0.00000019
Time.focusRetrospective	0.96	0.19	5.2	0.00000025
Message.contentPolicy	-1.6	0.27	-5.9	0.0000000034
(Intercept)	0.62	0.28	2.2	0.025

#Plot
sjp.glm(np_comp, sort.est= FALSE, title = "Incivility - Newspapers Comparison Model")

## Waiting for profiling to be done...

sjp.glm(np_comp, type="slope", show.ci = T)

sjp.glm(np_comp, type="pred", show.ci = T, vars=  "Message.content")

Comparison Model Twitter

Message.Sender <- twitter_data$cpgroup
Time.focus <- twitter_data$Odak
twt_comp <- glm(Incivility ~ Message.Sender + Time.focus + twitter_data$MessageContent, data= twitter_data, family = binomial)
panderOptions("digits", 2)
pander(twt_comp)

Fitting generalized (binomial/logit) linear model: Incivility ~ Message.Sender + Time.focus + twitter_data$MessageContent (continued below)
	Estimate	Std. Error	z value
Message.SenderGovernment	-1.4	0.096	-14
Time.focusRetrospective	3.6	0.41	8.8
twitter_data$MessageContentPolicy	1.5	0.088	17
(Intercept)	-5.8	0.41	-14

	Pr(>\|z\|)
Message.SenderGovernment	3.4e-46
Time.focusRetrospective	0.0000000000000000016
twitter_data$MessageContentPolicy	1.3e-61
(Intercept)	1.2e-44

#Table
sjt.glm(twt_comp, use.viewer = TRUE, show.se = T, show.r2 = T)

## Waiting for profiling to be done...

#Plot
sjp.glm(twt_comp, sort.est= FALSE, "Incivility - Twitter Comparison Model")

## Warning: Invalid `type` argument. Defaulting to `dots`.

## Waiting for profiling to be done...

Incivility on Twitter and Newspapers

Analysis on Twitter data

Analysis on Newspaper data