# Supplementary Material 2.3: Script for conducting the exploratory and confirmatory factor analsyes reported in the paper and the latent variable path analysis / SEM (author: Stefano Guidi)

# cfa e efa cyberbullying 
library(lavaan)
library(psych)
library(GPArotation)
library(semPlot)

# load data from Rdata format (categorical predictors and outcomes are already factors)
load(file = "S2_Dataset.RData")

# SM2_data_regression_SEM this will be the name of the loaded object (a dataframe)

# Note: data are the same used and loaded in script SM2_2_regression.R  so there is no need to run this line if already loaded

# determine train and test samples 
set.seed(17061973)
indici_train = sort(sample(nrow(SM2_data_regression_SEM), nrow(SM2_data_regression_SEM)*.3))
dati_train <- SM2_data_regression_SEM[indici_train, ]
dati_test <- SM2_data_regression_SEM[-indici_train, ]

# EFA su subset ----
# convert to numeric 
items_cyb_action_num_train <-  dati_train %>% 
  dplyr::select(8:13) %>% 
  mutate(across(everything(), as.numeric)) # %>% View()
#items_cyb_action_num_train %>% View()
#names(items_cyb_action_num_train)

# determine the number of factors
p_anal <- fa.parallel(items_cyb_action_num_train, cor = "poly") # 2 fattori o 1 componenti
p_anal$nfact

fa.parallel(items_cyb_action_num_train, cor = "poly", fm = "ml") # 2 fattori o 1 componenti
VSS(items_cyb_action_num_train, cor = "poly") # 1 o 2 fattori


# 2 fattori
FA_cyb_action_train <- fa(items_cyb_action_num_train, 2, rotate="oblimin", cor="poly") 
summary(FA_cyb_action_train)
print(FA_cyb_action_train, cut = 0.3)

# cross validation -----

extractFactors <- function(df){
  indici_train = sort(sample(nrow(df), nrow(df)*.3))
  df <- df[indici_train, ]
  df <- df %>% 
    dplyr::select(8:13) %>% 
    mutate(across(everything(), as.numeric)) 
  EFA <- fa(df, 2, rotate="oblimin", cor="poly") 
  n_fact <- fa.parallel(df, cor = "poly")
  return(
    list(loadings = EFA$loadings,
         n_fact = n_fact$nfact
         ))
  #print(EFA$loadings, cut = 0.4)
}
prova <- extractFactors(SM2_data_regression_SEM)
prova$n_fact

set.seed(29122014) # al primo loop viene un errore al modello 16. 
n_fattori_estratti <- tibble()
for (i in 1:100){
  risultati <- extractFactors(SM2_data_regression_SEM)
  cat("\nModel:", i, sep = "")
  print(risultati$loadings, cut = 0.35)
  n_fattori_estratti <- bind_rows(
    n_fattori_estratti,
    tibble(n = i, 
           n_factors = risultati$n_fact)
  )
}
frq(n_fattori_estratti$n_factors)

write_excel_csv2(n_fattori_estratti, file = "n_fattori.csv")

findNumberOfFactors <- function(df){
  indici_train = sort(sample(nrow(df), nrow(df)*.3))
  df <- df[indici_train, ]
  df <- df %>% 
    dplyr::select(8:13) %>% 
    mutate(across(everything(), as.numeric)) 
  EFA <- fa(df, 2, rotate="oblimin", cor="poly") 
  return(EFA$loadings)
  #print(EFA$loadings, cut = 0.4)
}
prova <- extractFactors(SM2_data_regression_SEM)

# CFA su test subset ----
modello_cyberbullying_action_2_factors_test <- '
cyberbullying_1 =~ CybBul1 + CybBul2 + CybBul4 
cyberbullying_2 =~ CybBul3 + CybBul5 + CybBul6
'
#is.ordered(dati_ileana_2$CybBul1)
cfa.cyberbullying_involvement_action_2_factors_test <- cfa(
  modello_cyberbullying_action_2_factors_test, 
  dati_test %>% mutate_at(.vars = 8:13, as.ordered),
  ordered = str_c("CybBul", 1:6, sep = "")
) #  as.numeric
summary(cfa.cyberbullying_involvement_action_2_factors_test, fit.measures=TRUE, standardized = TRUE) # 

# CFA modello 1 fattore  (Bauman) su test -----
modello_cyberbullying_action_1_factor_test <- '
cyberbullying =~ CybBul1 + CybBul2 + CybBul3 + CybBul4 + CybBul5 + CybBul6
'
#is.ordered(dati_ileana_2$CybBul1)
cfa.cyberbullying_involvement_action_1_factor_test <- cfa(
  modello_cyberbullying_action_1_factor_test, 
  dati_test %>% mutate_at(.vars = 8:13, as.ordered),
  ordered = str_c("CybBul", 1:6, sep = "")#%>% View()
) #  as.numeric
summary(cfa.cyberbullying_involvement_action_1_factor_test, fit.measures=TRUE, standardized = TRUE) # 

tibble(
  measure = names(fitMeasures(cfa.cyberbullying_involvement_action_1_factor_test)),
  single_factor = round(fitMeasures(cfa.cyberbullying_involvement_action_1_factor_test), digits = 3),
  two_Factor = round(fitMeasures(cfa.cyberbullying_involvement_action_2_factors_test), digits = 3)
) %>% View() # 2 factor model has better fit on the test data.


# conclusions: the 2 factor model suggested by the EFA on the train dataset had a  better fit than the single factor model on the test dataset


# SEM personality observed ----
# Definizione modelli ----
# modello con mediazione di MD su cyb1 e 2, e parametri effetti diretti, indiretti e totali
modello_cyberbullying_action_sem_2_factors_full_no_personality <- '
cyberbullying_1 =~ CybBul1 + CybBul2 + CybBul4 
cyberbullying_2 =~ CybBul3 + CybBul5 + CybBul6
Moral_disengagement =~ MD1 + MD2 + MD3 + MD4 + MD5 + MD6 + MD7 + MD8

cyberbullying_1 ~  a1*agreeableness + c1*conscientiousness +  n1*nevroticism  + o1*openness + e1*extraversion + Nervoso_senza_internet  + md1*Moral_disengagement 

cyberbullying_2 ~  a2*agreeableness + c2*conscientiousness + n2*nevroticism + o2*openness  +  e2*extraversion +  Nervoso_senza_internet + md2*Moral_disengagement 
#  
Moral_disengagement ~ amd*agreeableness + cmd*conscientiousness + nmd*nevroticism + omd*openness +  emd*extraversion # 


# indirect effects
a_md_1 := amd*md1
c_md_1 := cmd*md1
n_md_1 := nmd*md1
o_md_1 := omd*md1
e_md_1 := emd*md1

a_md_2 := amd*md2
c_md_2 := cmd*md2
n_md_2 := nmd*md2
o_md_2 := omd*md2
e_md_2 := emd*md2

# total effects
a_1 := a_md_1 + a1
c_1 := c_md_1 + c1
n_1 := n_md_1 + n1
o_1 := o_md_1 + o1
e_1 := e_md_1 + e1

a_2 := a_md_2 + a2
c_2 := c_md_2 + c2
n_2 := n_md_2 + n2
o_2 := o_md_2 + o2
e_2 := e_md_2 + e2

# differenze effetti su fattori
df_a := a1 - a2
df_c := c1 - c2
df_n := n1 - n2
df_o := o1 - o2
df_e := e1 - e2
df_md := md1 - md2
'

# con effetti indiretti stimati permettendo coefficienti path diversi per ogni gruppi nelle analisi multigruppo
modello_cyberbullying_action_sem_2_factors_full_no_personality_2 <- '
cyberbullying_1 =~ CybBul1 + CybBul2 + CybBul4 
cyberbullying_2 =~ CybBul3 + CybBul5 + CybBul6
Moral_disengagement =~ MD1 + MD2 + MD3 + MD4 + MD5 + MD6 + MD7 + MD8

cyberbullying_1 ~  c(a1a,a1b)*agreeableness + c(c1a,c1b)*conscientiousness +  c(n1a,n1b)*nevroticism  + c(o1a,o1b)*openness + c(e1a,e1b)*extraversion + c(nrv1a,nrv1b)*Nervoso_senza_internet  + c(md1a,md1b)*Moral_disengagement 

cyberbullying_2 ~  c(a2a,a2b)*agreeableness + c(c2a,c2b)*conscientiousness + c(n2a,n2b)*nevroticism + c(o2a,o2b)*openness  + c(e2a,e2b)*extraversion +  c(nrv2a,nrv2b)*Nervoso_senza_internet + c(md2a,md2b)*Moral_disengagement 
#  
Moral_disengagement ~ c(amd_a,amd_b)*agreeableness + c(cmd_a,cmd_b)*conscientiousness + c(nmd_a,nmd_b)*nevroticism + c(omd_a,omd_b)*openness +  c(emd_a,emd_b)*extraversion # 

# moderator effects
mod_md_cy1 := md1b - md1a
mod_a_cy1 := a1b - a1a
mod_c_cy1 := c1b - c1a
mod_n_cy1 := n1b - n1a
mod_o_cy1 := o1b - o1a
mod_e_cy1 := e1b - e1a
mod_nrv_cy1 := nrv1b - nrv1a

mod_md_cy2 := md2b - md2a
mod_a_cy2 := a2b - a2a
mod_c_cy2 := c2b - c2a
mod_n_cy2 := n2b - n2a
mod_o_cy2 := o2b - o2a
mod_e_cy2 := e2b - e2a
mod_nrv_cy2 := nrv2b - nrv2a

mod_a_md := amd_b - amd_a
mod_c_md := cmd_b - cmd_a
mod_n_md := nmd_b - nmd_a
mod_o_md := omd_b - omd_a
mod_e_md := emd_b - emd_a

# indirect effects
a_md_1a := amd_a*md1a
c_md_1a := cmd_a*md1a
n_md_1a := nmd_a*md1a
o_md_1a := omd_a*md1a
e_md_1a := emd_a*md1a

a_md_1b := amd_b*md1b
c_md_1b := cmd_b*md1b
n_md_1b := nmd_b*md1b
o_md_1b := omd_b*md1b
e_md_1b := emd_b*md1b

a_md_2a := amd_a*md2a
c_md_2a := cmd_a*md2a
n_md_2a := nmd_a*md2a
o_md_2a := omd_a*md2a
e_md_2a := emd_a*md2a

a_md_2b := amd_b*md2b
c_md_2b := cmd_b*md2b
n_md_2b := nmd_b*md2b
o_md_2b := omd_b*md2b
e_md_2b := emd_b*md2b

# total effects
a_1a := a_md_1a + a1a
c_1a := c_md_1a + c1a
n_1a := n_md_1a + n1a
o_1a := o_md_1a + o1a
e_1a := e_md_1a + e1a

a_1b := a_md_1b + a1b
c_1b := c_md_1b + c1b
n_1b := n_md_1b + n1b
o_1b := o_md_1b + o1b
e_1b := e_md_1b + e1b

a_2a := a_md_2a + a2a
c_2a := c_md_2a + c2a
n_2a := n_md_2a + n2a
o_2a := o_md_2a + o2a
e_2a := e_md_2a + e2a

a_2b := a_md_2b + a2b
c_2b := c_md_2b + c2b
n_2b := n_md_2b + n2b
o_2b := o_md_2b + o2b
e_2b := e_md_2b + e2b

# moderated mediation
mod_ind_n_md_cy1 := n_md_1a - n_md_1b
mod_ind_n_md_cy2 := n_md_2a - n_md_2b
mod_ind_c_md_cy1 := c_md_1a - c_md_1b
mod_ind_c_md_cy2 := c_md_2a - c_md_2b
mod_ind_e_md_cy1 := e_md_1a - e_md_1b
mod_ind_e_md_cy2 := e_md_2a - e_md_2b
mod_ind_o_md_cy1 := o_md_1a - o_md_1b
mod_ind_o_md_cy2 := o_md_2a - o_md_2b
'


# senza parametri effetti indiretti per analisi multigruppo 
modello_cyberbullying_action_sem_2_factors_full_no_personality_3 <- '
cyberbullying_1 =~ CybBul1 + CybBul2 + CybBul4 
cyberbullying_2 =~ CybBul3 + CybBul5 + CybBul6
Moral_disengagement =~ MD1 + MD2 + MD3 + MD4 + MD5 + MD6 + MD7 + MD8

cyberbullying_1 ~ agreeableness + conscientiousness +  nevroticism  + openness + extraversion + Nervoso_senza_internet  + 
Moral_disengagement 

cyberbullying_2 ~  agreeableness + conscientiousness + nevroticism + openness  +  extraversion + Nervoso_senza_internet + Moral_disengagement 
#  
Moral_disengagement ~ agreeableness + conscientiousness + nevroticism + openness +  extraversion # 
'
dati_SEM_full[indici_completi_all, ] %>% filter(!is.na(Nervoso_senza_internet)) %>% names()

# Single group SEM (reported in revision) ----
sem.cyberbullying_involvement_action_2_factors_full_no_personality <- sem(
  modello_cyberbullying_action_sem_2_factors_full_no_personality, 
  dati_SEM_full[indici_completi_all, ] %>% filter(!is.na(Nervoso_senza_internet)) %>% 
    mutate(nevroticism = 6 - nevroticism),
  ordered = c(str_c("CybBul", 1:6, sep = ""), str_c("MD", 1:8, sep = ""), str_c("BFI", 1:10, sep = ""))
)
summary(sem.cyberbullying_involvement_action_2_factors_full_no_personality, fit.measures=TRUE, standardized = TRUE, rsquare = TRUE) # very god fit

fitMeasures(sem.cyberbullying_involvement_action_2_factors_full_no_personality, fit.indices)
fitMeasures(sem.cyberbullying_involvement_action_2_factors_full_no_personality)

sem.cyberbullying_involvement_action_2_factors_full_no_personality_b <- sem(
  modello_cyberbullying_action_sem_2_factors_full_no_personality, 
  dati_SEM_full[indici_completi_all, ] %>% filter(!is.na(Nervoso_senza_internet)) %>% 
    mutate(nevroticism = 6 - nevroticism),
  ordered = c(str_c("CybBul", 1:6, sep = ""), str_c("MD", 1:8, sep = ""), str_c("BFI", 1:10, sep = ""))
)
summary(sem.cyberbullying_involvement_action_2_factors_full_no_personality, fit.measures=TRUE, standardized = TRUE, rsquare = TRUE) # very god fit

fitMeasures(sem.cyberbullying_involvement_action_2_factors_full_no_personality_b, fit.indices)
fitMeasures(sem.cyberbullying_involvement_action_2_factors_full_no_personality_b)
# strano che sia negativo l'effetto del neuroticismo
cor.test(dati_SEM_full$nevroticism, dati_SEM_full$MD_mean)

# test indirect effect of conscientiousness on cyb via internet addiction.
modello_cyberbullying_action_sem_2_factors_full_no_personality_c <- paste0(
  modello_cyberbullying_action_sem_2_factors_full_no_personality,
  '
  Nervoso_senza_internet ~ conscientiousness
  '
)
sem.cyberbullying_involvement_action_2_factors_full_no_personality_c <- sem(
  modello_cyberbullying_action_sem_2_factors_full_no_personality_c, 
  dati_SEM_full[indici_completi_all, ] %>% filter(!is.na(Nervoso_senza_internet)) %>% 
    mutate(nevroticism = 6 - nevroticism),
  ordered = c(str_c("CybBul", 1:6, sep = ""), str_c("MD", 1:8, sep = ""), str_c("BFI", 1:10, sep = ""))
)
summary(sem.cyberbullying_involvement_action_2_factors_full_no_personality_c, fit.measures=TRUE, standardized = TRUE, rsquare = TRUE) # very god fit

fitMeasures(sem.cyberbullying_involvement_action_2_factors_full_no_personality_c, fit.indices)

# Multigruppo GENERE  ----
sem.cyberbullying_involvement_action_2_factors_full_no_personality_gender <- sem(
  modello_cyberbullying_action_sem_2_factors_full_no_personality_3, 
  dati_SEM_full[indici_completi_all, ] %>% filter(!is.na(Nervoso_senza_internet)),
  ordered = c(str_c("CybBul", 1:6, sep = ""), str_c("MD", 1:8, sep = ""), str_c("BFI", 1:10, sep = "")),
  group = "Genere"
)
#  168 parameters
summary(sem.cyberbullying_involvement_action_2_factors_full_no_personality_gender, fit.measures=TRUE, standardized = TRUE, rsquare = TRUE) # very god fit

fitMeasures(sem.cyberbullying_involvement_action_2_factors_full_no_personality_gender, fit.indices) # so so
fitMeasures(sem.cyberbullying_involvement_action_2_factors_full_no_personality_gender)

sem.cyberbullying_involvement_action_2_factors_full_no_personality_gender1 <- sem(
  modello_cyberbullying_action_sem_2_factors_full_no_personality_3, 
  dati_SEM_full[indici_completi_all, ] %>% filter(!is.na(Nervoso_senza_internet)),
  ordered = c(str_c("CybBul", 1:6, sep = ""), str_c("MD", 1:8, sep = ""), str_c("BFI", 1:10, sep = "")),
  group = "Genere", group.equal = c("loadings")
)
#  168 parameters 11 equality constraints
summary(sem.cyberbullying_involvement_action_2_factors_full_no_personality_gender1, fit.measures=TRUE, standardized = TRUE, rsquare = TRUE) # ok fit
fitMeasures(sem.cyberbullying_involvement_action_2_factors_full_no_personality_gender1, fit.indices) # so so, not better
anova(sem.cyberbullying_involvement_action_2_factors_full_no_personality_gender, sem.cyberbullying_involvement_action_2_factors_full_no_personality_gender1) # ns weak invariance holds

# refit with moderators parameters (emotioinal stability) REPORTED in plot in SM 4
sem.cyberbullying_involvement_action_2_factors_full_no_personality_gender2mod <- sem(
  modello_cyberbullying_action_sem_2_factors_full_no_personality_2, 
  dati_SEM_full[indici_completi_all, ] %>% filter(!is.na(Nervoso_senza_internet)) %>% 
    mutate(nevroticism = 6 - nevroticism),
  ordered = c(str_c("CybBul", 1:6, sep = ""), str_c("MD", 1:8, sep = ""), str_c("BFI", 1:10, sep = "")),
  group = "Genere", group.equal = c("loadings", "intercepts")
) # 185 parameters 61 equality constraints !!! con "intercepts", 11 constr. con "means" "thresholds"
summary(sem.cyberbullying_involvement_action_2_factors_full_no_personality_gender2mod, fit.measures=TRUE, standardized = TRUE, rsquare = TRUE) 
# differenze medie di cyb 1 (male > f) e moral disengagement (m > f)
fitMeasures(sem.cyberbullying_involvement_action_2_factors_full_no_personality_gender2mod, fit.indices) # better
# moderazione di gender sull'effetto di o -> cyb1 (-) e di e -> cy2 (+)

anova(sem.cyberbullying_involvement_action_2_factors_full_no_personality_gender, sem.cyberbullying_involvement_action_2_factors_full_no_personality_gender1, sem.cyberbullying_involvement_action_2_factors_full_no_personality_gender2mod) #! REPORTED IN SM 4

# Multigruppo TEMPO SUI SOCIAL  ----
# unconstrained model
sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo <- sem(
  modello_cyberbullying_action_sem_2_factors_full_no_personality_3, 
  dati_SEM_full[indici_completi_all, ] %>% filter(!is.na(Nervoso_senza_internet)) %>% filter(Tempo_sui_social != "Meno di 1 ora al giorno"),
  ordered = c(str_c("CybBul", 1:6, sep = ""), str_c("MD", 1:8, sep = ""), str_c("BFI", 1:10, sep = "")),
  group = "Tempo_sui_social"
)
#  168 parameters
summary(sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo, fit.measures=TRUE, standardized = TRUE, rsquare = TRUE) #  quite god fit
fitMeasures(sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo, fit.indices) # quite god fit, configural invariance holds
fitMeasures(sem.cyberbullying_involvement_action_2_factors_full_no_personality_gender)

# constrained loadings (test weak invariance)
sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo1 <- sem(
  modello_cyberbullying_action_sem_2_factors_full_no_personality_3, 
  dati_SEM_full[indici_completi_all, ] %>% filter(!is.na(Nervoso_senza_internet)) %>% filter(Tempo_sui_social != "Meno di 1 ora al giorno"),
  ordered = c(str_c("CybBul", 1:6, sep = ""), str_c("MD", 1:8, sep = ""), str_c("BFI", 1:10, sep = "")),
  group = "Tempo_sui_social", group.equal = c("loadings")
)
#  168 parameters, 11 contraints
summary(sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo1, fit.measures=TRUE, standardized = TRUE, rsquare = TRUE) #  quite god fit
fitMeasures(sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo1, fit.indices) # quite god fit, same as before
anova(sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo, sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo1) # NS, weak invariance holds

# constrained loadings and intercepts (test strong invariance)
sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo2 <- sem(
  modello_cyberbullying_action_sem_2_factors_full_no_personality_3, 
  dati_SEM_full[indici_completi_all, ] %>% filter(!is.na(Nervoso_senza_internet)) %>% filter(Tempo_sui_social != "Meno di 1 ora al giorno"),
  ordered = c(str_c("CybBul", 1:6, sep = ""), str_c("MD", 1:8, sep = ""), str_c("BFI", 1:10, sep = "")),
  group = "Tempo_sui_social", group.equal = c("loadings", "intercepts")
)
#  185 parameters, 61 constraints
summary(sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo2, fit.measures=TRUE, standardized = TRUE, rsquare = TRUE) #  quite god fit, interecette LV non diverse tra i gruppi.
# non differenze significative medie di cyb 1, 2 o  moral disengagement (m > f)

fitMeasures(sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo2, fit.indices) # better fit
anova(sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo, sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo1, sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo2) # NS,ma c'è un problema nel confronto, il modello 2 ha più parametri del modello 1, non meno come dovrebbe essere se è annidato

# refit with moderators (neuroticismo è in realtà emotional stability)
sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo2mod <- sem(
  modello_cyberbullying_action_sem_2_factors_full_no_personality_2, 
  dati_SEM_full[indici_completi_all, ] %>% filter(!is.na(Nervoso_senza_internet)) %>% filter(Tempo_sui_social != "Meno di 1 ora al giorno") %>% 
    mutate(nevroticism = 6 - nevroticism),
  ordered = c(str_c("CybBul", 1:6, sep = ""), str_c("MD", 1:8, sep = ""), str_c("BFI", 1:10, sep = "")),
  group = "Tempo_sui_social", group.equal = c("loadings", "intercepts")
)
#  185 parameters, 61 constraints
summary(sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo2mod, fit.measures=TRUE, standardized = TRUE, rsquare = TRUE) #  quite god fit, intercette LV non diverse tra i gruppi.
# non differenze significative medie di cyb 1, 2 o  moral disengagement (m > f)

fitMeasures(sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo2mod, fit.indices) 

anova(sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo, sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo1, sem.cyberbullying_involvement_action_2_factors_full_no_personality_tempo2mod) # NS,ma c'è un problema nel confronto, il modello 2 ha più parametri del modello 1, non meno come dovrebbe essere se è annidato

# Multigruppo N PROFILI ----
# unconstrained model
sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili <- sem(
  modello_cyberbullying_action_sem_2_factors_full_no_personality_3, 
  dati_SEM_full[indici_completi_all, ] %>% filter(!is.na(Nervoso_senza_internet)) %>% filter(Tempo_sui_social != "Meno di 1 ora al giorno"),
  ordered = c(str_c("CybBul", 1:6, sep = ""), str_c("MD", 1:8, sep = ""), str_c("BFI", 1:10, sep = "")),
  group = "N_profili_social"
)
summary(sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili, fit.measures=TRUE, standardized = TRUE, rsquare = TRUE) #  quite god fit
fitMeasures(sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili, fit.indices) #  decent  fit

# loadings
sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili1 <- sem(
  modello_cyberbullying_action_sem_2_factors_full_no_personality_3, 
  dati_SEM_full[indici_completi_all, ] %>% filter(!is.na(Nervoso_senza_internet)) %>% filter(Tempo_sui_social != "Meno di 1 ora al giorno"),
  ordered = c(str_c("CybBul", 1:6, sep = ""), str_c("MD", 1:8, sep = ""), str_c("BFI", 1:10, sep = "")),
  group = "N_profili_social", group.equal = c("loadings")
)
summary(sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili1, fit.measures=TRUE, standardized = TRUE, rsquare = TRUE) #  quite god fit
fitMeasures(sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili1, fit.indices) 
anova(sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili, sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili1) # NS 

# loadings + intercepts
sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili2 <- sem(
  modello_cyberbullying_action_sem_2_factors_full_no_personality_3, 
  dati_SEM_full[indici_completi_all, ] %>% filter(!is.na(Nervoso_senza_internet)) %>% filter(Tempo_sui_social != "Meno di 1 ora al giorno"),
  ordered = c(str_c("CybBul", 1:6, sep = ""), str_c("MD", 1:8, sep = ""), str_c("BFI", 1:10, sep = "")),
  group = "N_profili_social", group.equal = c("loadings", "intercepts")
)
summary(sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili2, fit.measures=TRUE, standardized = TRUE, rsquare = TRUE) #  quite god fit
# moral disengagement > 
fitMeasures(sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili2, fit.indices) 
anova(
  sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili, 
  sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili1, 
  sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili2) # NS 

# refit for moderated mediation analysis (REPORTED in PLOT in SM 4)
sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili2mod <- sem(
  modello_cyberbullying_action_sem_2_factors_full_no_personality_2, 
  dati_SEM_full[indici_completi_all, ] %>% filter(!is.na(Nervoso_senza_internet)) %>% filter(Tempo_sui_social != "Meno di 1 ora al giorno")%>% 
    mutate(nevroticism = 6 - nevroticism),
  ordered = c(str_c("CybBul", 1:6, sep = ""), str_c("MD", 1:8, sep = ""), str_c("BFI", 1:10, sep = "")),
  group = "N_profili_social", group.equal = c("loadings", "intercepts")
)
summary(sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili2mod, fit.measures=TRUE, standardized = TRUE, rsquare = TRUE) #  quite god fit
# moral disengagement > 
fitMeasures(sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili2mod, fit.indices) 

anova(
  sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili, 
  sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili1, 
  sem.cyberbullying_involvement_action_2_factors_full_no_personality_profili2mod) # NS