Loading, setting up
tmc <- readRDS("tm-c.rds")
tme <- readRDS("tm-e.rds")
library(lme4)
library(igraph)
library(tidyverse)
library(ergm)
Conversing model
s <- tmc %>%
select(sender, sender_profile, sender_n_tweets)
r <- tmc %>%
select(receiver, receiver_profile, receiver_n_tweets)
names(r) <- names(s)
vertices_c <- bind_rows(s, r) %>%
distinct() %>%
filter(!is.na(sender_profile))
names(vertices_c) <- c("participant", "profile_code", "n_tweets")
edges_c <- tmc %>%
select(sender, receiver, tie, weight_l) %>%
filter(tie == 1)
g <- graph_from_data_frame(edges_c, vertices = vertices_c)
V(g)$n_tweets <- scale(V(g)$n_tweets)
g <- igraph::simplify(g, remove.loops=T)
gg <- intergraph::asNetwork(g)
model1 <- ergm(gg ~ edges + mutual + nodecov("n_tweets") + nodemix("profile_code", base = 25),
control = control.ergm(MCMC.burnin=50000, MCMC.interval=5000, MCMLE.maxit=30)) # unclear/unclear
summary(model1)
##
## ==========================
## Summary of model fit
## ==========================
##
## Formula: gg ~ edges + mutual + nodecov("n_tweets") + nodemix("profile_code",
## base = 25)
##
## Iterations: 6 out of 30
##
## Monte Carlo MLE Results:
## Estimate Std. Error
## edges -5.184600 0.247888
## mutual 3.959900 0.056389
## nodecov.n_tweets 0.313171 0.004341
## mix.profile_code.Administration.Administration 0.895671 0.254624
## mix.profile_code.Other.Administration 0.932937 0.268038
## mix.profile_code.Policy/Research.Administration 1.079581 0.264026
## mix.profile_code.Teacher.Administration 0.749287 0.254627
## mix.profile_code.Unclear.Administration 0.714088 0.282336
## mix.profile_code.Administration.Other 0.478922 0.274329
## mix.profile_code.Other.Other 0.467628 0.283149
## mix.profile_code.Policy/Research.Other 1.048855 0.270811
## mix.profile_code.Teacher.Other 0.425297 0.258580
## mix.profile_code.Unclear.Other 0.153155 0.335300
## mix.profile_code.Administration.Policy/Research 0.390549 0.272812
## mix.profile_code.Other.Policy/Research 0.404394 0.283248
## mix.profile_code.Policy/Research.Policy/Research 0.791467 0.267401
## mix.profile_code.Teacher.Policy/Research 0.215017 0.260940
## mix.profile_code.Unclear.Policy/Research 0.359864 0.299270
## mix.profile_code.Administration.Teacher 0.641437 0.256569
## mix.profile_code.Other.Teacher 0.709844 0.261554
## mix.profile_code.Policy/Research.Teacher 0.739890 0.257393
## mix.profile_code.Teacher.Teacher 0.638359 0.248299
## mix.profile_code.Unclear.Teacher 0.588830 0.262507
## mix.profile_code.Administration.Unclear -0.473839 0.309148
## mix.profile_code.Other.Unclear 0.086665 0.323034
## mix.profile_code.Policy/Research.Unclear 0.058613 0.313227
## mix.profile_code.Teacher.Unclear 0.052929 0.273030
## MCMC % p-value
## edges 0 < 1e-04 ***
## mutual 0 < 1e-04 ***
## nodecov.n_tweets 1 < 1e-04 ***
## mix.profile_code.Administration.Administration 0 0.000436 ***
## mix.profile_code.Other.Administration 0 0.000500 ***
## mix.profile_code.Policy/Research.Administration 0 < 1e-04 ***
## mix.profile_code.Teacher.Administration 0 0.003254 **
## mix.profile_code.Unclear.Administration 0 0.011433 *
## mix.profile_code.Administration.Other 0 0.080847 .
## mix.profile_code.Other.Other 0 0.098633 .
## mix.profile_code.Policy/Research.Other 0 0.000108 ***
## mix.profile_code.Teacher.Other 0 0.100024
## mix.profile_code.Unclear.Other 0 0.647836
## mix.profile_code.Administration.Policy/Research 0 0.152269
## mix.profile_code.Other.Policy/Research 0 0.153378
## mix.profile_code.Policy/Research.Policy/Research 0 0.003078 **
## mix.profile_code.Teacher.Policy/Research 0 0.409935
## mix.profile_code.Unclear.Policy/Research 0 0.229182
## mix.profile_code.Administration.Teacher 0 0.012418 *
## mix.profile_code.Other.Teacher 0 0.006649 **
## mix.profile_code.Policy/Research.Teacher 0 0.004046 **
## mix.profile_code.Teacher.Teacher 0 0.010143 *
## mix.profile_code.Unclear.Teacher 0 0.024891 *
## mix.profile_code.Administration.Unclear 0 0.125345
## mix.profile_code.Other.Unclear 0 0.788481
## mix.profile_code.Policy/Research.Unclear 0 0.851563
## mix.profile_code.Teacher.Unclear 0 0.846287
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Null Deviance: 329461 on 237656 degrees of freedom
## Residual Deviance: 35766 on 237629 degrees of freedom
##
## AIC: 35820 BIC: 36100 (Smaller is better.)
d <- plogis(coef(model1)[['edges']] + coef(model1)[['mutual']] + coef(model1)[['nodecov.n_tweets']] + coef(model1)[4:27]) %>% as.data.frame() %>%
rownames_to_column("var") %>%
mutate(var = str_sub(var, start = 18))
b <- broom::tidy(model1) %>% slice(4:27) %>% filter(p.value < .05) %>% mutate(term = str_sub(term, start = 18)) %>% dplyr::select(term, p.value) %>% arrange(p.value) %>% mutate(char = "*")
names(d) <- c('term', 'probability')
d %>%
left_join(b) %>%
ggplot(aes(x = reorder(term, probability), y = probability, label = char)) +
geom_col() +
geom_text(color = "red", size = 5) +
coord_flip() +
theme_bw() +
labs(title = "Probability of individuals 'conversing' (replying to or mentioning the other) on #NGSSchat",
caption = "Groups with effects that significantly differ from those in the unclear group conversing are denoted with a red asterisk.") +
theme(text = element_text(family = "Times")) +
xlab(NULL) +
ylab("Probability")
Endorsing model
s <- tme %>%
select(sender, sender_profile, sender_n_tweets)
r <- tme %>%
select(receiver, receiver_profile, receiver_n_tweets)
names(r) <- names(s)
vertices_c <- bind_rows(s, r) %>%
distinct() %>%
filter(!is.na(sender_profile))
names(vertices_c) <- c("participant", "profile_code", "n_tweets")
edges_c <- tme %>%
select(sender, receiver, tie) %>%
filter(tie == 1)
g <- graph_from_data_frame(edges_c, vertices = vertices_c)
V(g)$n_tweets <- scale(V(g)$n_tweets)
g <- igraph::simplify(g, remove.loops=T)
gg <- intergraph::asNetwork(g)
model1 <- ergm(gg ~ edges + mutual + nodecov("n_tweets") + nodemix("profile_code", base = 25),
control = control.ergm(MCMC.burnin=50000, MCMC.interval=5000, MCMLE.maxit=20))
summary(model1)
##
## ==========================
## Summary of model fit
## ==========================
##
## Formula: gg ~ edges + mutual + nodecov("n_tweets") + nodemix("profile_code",
## base = 25)
##
## Iterations: 8 out of 20
##
## Monte Carlo MLE Results:
## Estimate Std. Error
## edges -4.317123 0.161282
## mutual 2.760495 0.039200
## nodecov.n_tweets 0.298927 0.004408
## mix.profile_code.Administration.Administration 0.820694 0.164170
## mix.profile_code.Other.Administration 0.658389 0.179474
## mix.profile_code.Policy/Research.Administration 0.701413 0.173678
## mix.profile_code.Teacher.Administration 0.765018 0.165059
## mix.profile_code.Unclear.Administration 0.360331 0.191984
## mix.profile_code.Administration.Other 0.767197 0.175290
## mix.profile_code.Other.Other 0.483380 0.186299
## mix.profile_code.Policy/Research.Other 0.735126 0.174324
## mix.profile_code.Teacher.Other 0.680445 0.166009
## mix.profile_code.Unclear.Other 0.277491 0.206567
## mix.profile_code.Administration.Policy/Research 0.822523 0.175781
## mix.profile_code.Other.Policy/Research 0.690024 0.184777
## mix.profile_code.Policy/Research.Policy/Research 0.748822 0.168633
## mix.profile_code.Teacher.Policy/Research 0.769509 0.167902
## mix.profile_code.Unclear.Policy/Research 0.510383 0.198943
## mix.profile_code.Administration.Teacher 0.617942 0.167860
## mix.profile_code.Other.Teacher 0.586895 0.170250
## mix.profile_code.Policy/Research.Teacher 0.349516 0.167627
## mix.profile_code.Teacher.Teacher 0.668027 0.161622
## mix.profile_code.Unclear.Teacher 0.422050 0.170723
## mix.profile_code.Administration.Unclear 0.121134 0.193555
## mix.profile_code.Other.Unclear 0.319310 0.200854
## mix.profile_code.Policy/Research.Unclear 0.097968 0.196849
## mix.profile_code.Teacher.Unclear 0.276962 0.179725
## MCMC % p-value
## edges 0 < 1e-04 ***
## mutual 0 < 1e-04 ***
## nodecov.n_tweets 1 < 1e-04 ***
## mix.profile_code.Administration.Administration 0 < 1e-04 ***
## mix.profile_code.Other.Administration 0 0.000244 ***
## mix.profile_code.Policy/Research.Administration 0 < 1e-04 ***
## mix.profile_code.Teacher.Administration 0 < 1e-04 ***
## mix.profile_code.Unclear.Administration 0 0.060536 .
## mix.profile_code.Administration.Other 0 < 1e-04 ***
## mix.profile_code.Other.Other 0 0.009469 **
## mix.profile_code.Policy/Research.Other 0 < 1e-04 ***
## mix.profile_code.Teacher.Other 0 < 1e-04 ***
## mix.profile_code.Unclear.Other 0 0.179161
## mix.profile_code.Administration.Policy/Research 0 < 1e-04 ***
## mix.profile_code.Other.Policy/Research 0 0.000188 ***
## mix.profile_code.Policy/Research.Policy/Research 0 < 1e-04 ***
## mix.profile_code.Teacher.Policy/Research 0 < 1e-04 ***
## mix.profile_code.Unclear.Policy/Research 0 0.010304 *
## mix.profile_code.Administration.Teacher 0 0.000232 ***
## mix.profile_code.Other.Teacher 0 0.000566 ***
## mix.profile_code.Policy/Research.Teacher 0 0.037063 *
## mix.profile_code.Teacher.Teacher 0 < 1e-04 ***
## mix.profile_code.Unclear.Teacher 0 0.013432 *
## mix.profile_code.Administration.Unclear 0 0.531421
## mix.profile_code.Other.Unclear 0 0.111889
## mix.profile_code.Policy/Research.Unclear 0 0.618708
## mix.profile_code.Teacher.Unclear 0 0.123310
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Null Deviance: 341738 on 246512 degrees of freedom
## Residual Deviance: 67815 on 246485 degrees of freedom
##
## AIC: 67869 BIC: 68150 (Smaller is better.)
d <- plogis(coef(model1)[['edges']] + coef(model1)[['mutual']] + coef(model1)[['nodecov.n_tweets']] + coef(model1)[4:27]) %>% as.data.frame() %>%
rownames_to_column("var") %>%
mutate(var = str_sub(var, start = 18))
b <- broom::tidy(model1) %>% slice(4:27) %>% filter(p.value < .05) %>% mutate(term = str_sub(term, start = 18)) %>% dplyr::select(term, p.value) %>% arrange(p.value) %>% mutate(char = "*")
names(d) <- c('term', 'probability')
d %>%
left_join(b) %>%
# bind_rows(data.frame())
ggplot(aes(x = reorder(term, probability), y = probability, label = char)) +
geom_col() +
geom_text(color = "red", size = 5) +
coord_flip() +
theme_bw() +
labs(title = "Probability of individuals 'endorsing' (favoriting or retweeting the other) on #NGSSchat",
caption = "Groups with effects that significantly differ from those in the unclear group conversing are denoted with a red asterisk. With mean number of tweets sent to the chat.") +
theme(text = element_text(family = "Times")) +
xlab(NULL) +
ylab("Probability")
