digipol 1 – Italy replication

1 Setting folder, dataset and packages

2 Define data MP_DATASET

# 1a. Construct the data_MP dataset ---------------------------------------------
data_MP <- EUPeopleIta %>%
  # Select the relevant variables (uso LERI_ita_or, non LERI)
  dplyr::select(
    DIG02_1:DIG02_8, DEM02, DEM02R, VOTEI02_ITA, DEM01, 
    LERI_ita_or, POLINT, POLINF_1:POLINF_3
  ) %>%
  # Convert variables to factors
  dplyr::mutate(across(
    c(DIG02_1:DIG02_8, DEM02R, VOTEI02_ITA, DEM01, 
      LERI_ita_or, POLINT, POLINF_1:POLINF_3),
    haven::as_factor
  )) %>%
  # Order the 1–8 digital use frequency variables
  dplyr::mutate(across(
    c(DIG02_1:DIG02_8),
    ~ factor(.x,
             levels = c("Never", "Less often", "Several times a month",
                        "Several times a week", "Once a day",
                        "Several times a day", "Almost all the time"),
             ordered = TRUE)
  )) %>%
  # Create numeric versions
  dplyr::mutate(across(
    c(DIG02_1:DIG02_8),
    as.integer,
    .names = "{.col}_num"
  )) %>%
  # Rename variables
  dplyr::rename(
    YouTube_      = DIG02_1,
    WhatsApp_     = DIG02_2,
    TwitterX_     = DIG02_3,
    Instagram_    = DIG02_4,
    TikTok_       = DIG02_5,
    Facebook_     = DIG02_6,
    LinkedIn_     = DIG02_7,
    Telegram_     = DIG02_8,
    YouTube       = DIG02_1_num,
    WhatsApp      = DIG02_2_num,
    Twitter_X     = DIG02_3_num,
    Instagram     = DIG02_4_num,
    TikTok        = DIG02_5_num,
    Facebook      = DIG02_6_num,
    Linkedin      = DIG02_7_num,
    Telegram      = DIG02_8_num,
    Age           = DEM02,
    Age_cat       = DEM02R,
    Vote          = VOTEI02_ITA,
    Sex           = DEM01,
    Left_Right    = LERI_ita_or,
    Pol_int       = POLINT,
    Inf_disagr    = POLINF_1,
    Inf_check     = POLINF_2,
    Inf_confirm   = POLINF_3
  ) %>%
  # Recode VOTEI02_ITA levels → Other party / Did not vote
  dplyr::mutate(
    Vote = forcats::fct_recode(
      Vote,
      "Other party" = "Noi Moderati",
      "Other party" = "Impegno Civico-Centro Democratico",
      "Other party" = "Italexit per l'Italia",
      "Other party" = "Unione Popolare",
      "Other party" = "Italia Sovrana e Popolare",
      "Other party" = "Vita",
      "Other party" = "Altro partito",
      
      "Did not vote" = "Non sono andato a votare",
      "Did not vote" = "Ho votato scheda bianca / scheda nulla"
    )
  )

# ---- vote recode
# Code 20 ("Prefer not to answer") → NA
data_MP <- data_MP %>%
  mutate(
    Vote_num = as.integer(EUPeopleIta$VOTEI02_ITA),
    Vote = case_when(
      Vote_num == 20 ~ NA_character_,
      TRUE ~ as.character(Vote)
    ),
    Vote = factor(Vote)
  ) %>%
  select(-Vote_num)

# ---- RECODE LERI_ita_or 
# # original variable: 1-11 = scale 0-10, 12 = "Not placed", 13 = NA
data_MP <- data_MP %>%
  mutate(
    Left_Right_num = as.integer(EUPeopleIta$LERI_ita_or),
    
    Left_Right = case_when(
      Left_Right_num %in% 1:11 ~ as.character(Left_Right_num - 1),
      Left_Right_num == 12 ~ "Not placed",
      Left_Right_num == 13 ~ NA_character_,
      TRUE ~ NA_character_
    ),
    
    Left_Right = factor(
      Left_Right,
      levels = c(as.character(0:10), "Not placed"),
      ordered = FALSE
    )
  ) %>%
  select(-Left_Right_num)

3 Define the USES_DATASET

# 1b. Construct the data_USES dataset ------------------------------------

data_USES <- EUPeopleNet %>%
  dplyr::select(dplyr::starts_with("dig03_"))

platforms <- c("YouTube", "WhatsApp", "TwitterX", "Instagram",
               "TikTok", "Facebook", "LinkedIn", "Telegram")

platform_prefixes <- paste0("dig03_", 1:8, "_")

purpose_labels <- c(
  "friends", "family", "inf_socially", "sup_soc_causes",
  "inf_pol", "sup_pol_causes", "connect_profess",
  "promote_profess", "other"
)

purpose_suffixes <- 1:9

# Compute binary indicators platform x scope
platform_long <- list()

for (p in seq_along(platforms)) {
  plat   <- platforms[p]
  prefix <- platform_prefixes[p]
  
  for (i in seq_along(purpose_suffixes)) {
    pattern_i <- paste0("^", prefix, purpose_suffixes[i], "$")
    cols_i <- grep(pattern_i, names(data_USES), value = TRUE)
    
    if (length(cols_i) > 0) {
      platform_long[[length(platform_long) + 1]] <- data_USES %>%
        mutate(
          Platform = plat,
          Purpose  = purpose_labels[i],
          Response = ifelse(
            rowSums(across(all_of(cols_i)), na.rm = TRUE) > 0,
            "Yes", "No"
          )
        ) %>%
        dplyr::select(Platform, Purpose, Response)
    }
  }
}

data_platform_use <- dplyr::bind_rows(platform_long)

# Summary of "YES" answers
summary_platform_use <- data_platform_use %>%
  dplyr::filter(Response == "Yes") %>%
  dplyr::group_by(Platform, Purpose) %>%
  dplyr::summarise(n = n(), .groups = "drop") %>%
  dplyr::group_by(Platform) %>%
  dplyr::mutate(Percentage = n / sum(n) * 100) %>%
  dplyr::ungroup() %>%
  dplyr::mutate(Purpose = factor(Purpose, levels = purpose_labels))

# Graph per platform
#| label: fig_social_media_use_by_purpose
#| fig-width: 9
#| fig-height: 6
#| message: false
#| warning: false

# panels order as ES data
summary_platform_use_plot <- summary_platform_use %>%
  dplyr::mutate(
    Platform = factor(
      Platform,
      levels = c("Facebook", "Instagram", "LinkedIn", "Telegram",
                 "TikTok", "TwitterX", "WhatsApp", "YouTube")
    )
  )

ggplot(summary_platform_use_plot,
       aes(x = Purpose, y = Percentage)) +
  geom_col(fill = "darkorange", color = "white", width = 0.75) +
  geom_text(
    aes(label = sprintf("%.1f%%", Percentage)),
    vjust = -0.25,
    size  = 2.2,
    color = "black"
  ) +
  facet_wrap(~ Platform, ncol = 4) +
  labs(
    title    = "Use of Social Media Platform by Purpose",
    subtitle = "Percentage of respondents using each platform for each purpose (Yes responses)",
    x        = "Purpose of Use",
    y        = "Percentage of Respondents"
  ) +
  scale_y_continuous(
    limits = c(0, 50),                  
    breaks = seq(0, 50, 10),
    expand = expansion(mult = c(0, 0.08))
  ) +
  theme_minimal(base_size = 11) +
  theme(
    plot.title        = element_text(face = "bold", size = 12),
    plot.subtitle     = element_text(size = 10),
    axis.title.x      = element_text(size = 10, margin = margin(t = 7)),
    axis.title.y      = element_text(size = 10, margin = margin(r = 7)),
    axis.text.x       = element_text(
      angle = 90, hjust = 1, vjust = 0.5, size = 7
    ),
    axis.text.y       = element_text(size = 7),
    strip.text        = element_text(face = "bold", size = 10),
    panel.grid.major.x = element_blank(),
    panel.grid.minor.x = element_blank(),
    panel.grid.minor.y = element_blank(),
    plot.margin       = margin(t = 9, r = 9, b = 9, l = 9)
  )

Italy. Social media use is generally moderate, with platforms mostly serving relational purposes (friends/family). Political engagement is concentrated almost entirely on TwitterX and—more marginally—Telegram. WhatsApp and Instagram remain predominantly social, while YouTube shows a notable share of “other” uses.

*Compared with Spain, Italy displays lower overall usage levels and a less intense sociality across all major platforms. Spain shows broader and more versatile engagement—especially on Instagram, WhatsApp and YouTube—while Italy stands out mainly for the comparatively stronger political use of TwitterX.

4 Set platform per purpose

# 1c. ANY PLATFORM PER PURPOSE (Yes / No ) ---------------------

data_any <- EUPeopleNet %>%
  dplyr::select(dplyr::starts_with("dig03_"))

labels_use <- c(
  "friends", "family", "inf_socially", "sup_soc_causes",
  "inf_pol", "sup_pol_causes", "connect_profess",
  "promote_profess", "other"
)

purpose_suffixes <- 1:9

for (i in seq_along(labels_use)) {
  pattern_i <- paste0("_", purpose_suffixes[i], "$")
  cols_i <- grep(pattern_i, names(data_any), value = TRUE)
  
  if (length(cols_i) > 0) {
    data_any[[labels_use[i]]] <- ifelse(
      rowSums(data_any[, cols_i], na.rm = TRUE) > 0, 1, 0
    )
  } else {
    message("⚠️ No columns found for purpose: ", labels_use[i])
  }
}

existing_labels <- intersect(labels_use, names(data_any))

data_any <- data_any %>%
  dplyr::mutate(
    dplyr::across(
      all_of(existing_labels),
      ~ factor(.x, levels = c(0, 1), labels = c("No", "Yes"))
    )
  )

summary_uses <- data_any %>%
  tidyr::pivot_longer(
    cols = all_of(existing_labels),
    names_to = "Purpose", values_to = "Response"
  ) %>%
  count(Purpose, Response) %>%
  group_by(Purpose) %>%
  mutate(Percentage = n / sum(n) * 100) %>%
  ungroup()

ggplot(summary_uses, aes(x = Purpose, y = Percentage, fill = Response)) +
  geom_col(position = "dodge") +
  scale_fill_manual(values = c("grey70", "darkorange")) +
  labs(
    title = "Use of ANY Social Media Platform by Purpose",
    subtitle = "Percentages within each purpose (Yes vs No)",
    x = "Purpose",
    y = "Percentage of Respondents",
    fill = "Response"
  ) +
  theme_minimal(base_size = 10) +
  theme(
    axis.text.x = element_text(angle = 45, hjust = 1),
    plot.title  = element_text(face = "bold")
  )

# 1c-bis. political and sociopolitical purposes indicators --------

data_any <- data_any %>%
  mutate(
    # political use (any platform)
    PolInf_smed = inf_pol,

    # political causes (any platform)
    PolSup_smed = sup_pol_causes,

    # sociopolitical use:
    # (social info, social causes support, political information, political causes support)
    PolSocial_smed = ifelse(
      inf_socially == "Yes" | sup_soc_causes == "Yes" |
        inf_pol == "Yes"      | sup_pol_causes == "Yes",
      "Yes", "No"
    ),

    # political info & support (more strict definition)
    PolInfSup_smed = ifelse(
      inf_pol == "Yes" | sup_pol_causes == "Yes",
      "Yes", "No"
    )
  ) %>%
  mutate(
    across(
      c(PolInf_smed, PolSup_smed, PolSocial_smed, PolInfSup_smed),
      ~ factor(.x, levels = c("No", "Yes"))
    )
  )


###Tables
#| label: tables_indicators
#| message: false
#| warning: false

library(dplyr)
library(tidyr)
library(knitr)

# indicators selection
indicators_long <- data_any %>%
  dplyr::select(PolInf_smed, PolSup_smed, PolSocial_smed, PolInfSup_smed) %>%
  tidyr::pivot_longer(
    cols      = everything(),
    names_to  = "Indicator",
    values_to = "Value"
  ) %>%
  dplyr::group_by(Indicator, Value) %>%
  dplyr::summarise(
    n   = dplyr::n(),
    pct = 100 * n / sum(n),
    .groups = "drop"
  )

knitr::kable(
  indicators_long,
  digits  = 1,
  caption = "Distribution of political/socio-political use (ANY platform)"
)

Distribution of political/socio-political use (ANY platform)

Indicator	Value	n	pct
PolInfSup_smed	No	426	100
PolInfSup_smed	Yes	875	100
PolInf_smed	No	441	100
PolInf_smed	Yes	860	100
PolSocial_smed	No	274	100
PolSocial_smed	Yes	1027	100
PolSup_smed	No	607	100
PolSup_smed	Yes	694	100

### crosstab purposes 
#| label: crosstab_PolInf_PolSup
#| message: false
#| warning: false

crosstab_inf_sup <- table(
  Information = data_any$PolInf_smed,
  Support     = data_any$PolSup_smed
)

# Converts in data.frame & adds row percentage 
crosstab_df <- as.data.frame.matrix(crosstab_inf_sup)
crosstab_df$Row_Total <- rowSums(crosstab_df)

# Counts table
knitr::kable(
  crosstab_df,
  caption = "Table: political information vs political causes support (counts)"
)

Table: political information vs political causes support (counts)

	No	Yes	Row_Total
No	426	15	441
Yes	181	679	860

# Row percentage
prop_row <- prop.table(crosstab_inf_sup, margin = 1) * 100
prop_row_df <- as.data.frame.matrix(round(prop_row, 1))

knitr::kable(
  prop_row_df,
  caption = "Table: political information vs political causes support (row percentages)"
)

Table: political information vs political causes support (row percentages)

	No	Yes
No	96.6	3.4
Yes	21.0	79.0

The overlap between political information and support for political causes is strongly asymmetric: those who do not seek political information on social media almost never support causes (only 3.4%). In contrast, among those who use social media for political information, about four out of five also engage in supporting causes. The politically active online public therefore appears concentrated and consistent in its practices, while the non-political users remain largely impermeable polarization between purely non-political uses and a smaller but highly engaged cluster.

5 Descriptive plots: platform use and purposes

# 2. Descriptive graphs ======================================
## 2.1a Use frequency per platform (DIG02) – count

data_long <- data_MP %>%
  tidyr::pivot_longer(
    cols = YouTube_:Telegram_,
    names_to = "Platform",
    values_to = "Frequency"
  )

ggplot(data_long, aes(x = Frequency)) +
  geom_bar(fill = "darkorange") +
  geom_text(
    stat = "count",
    aes(label = after_stat(count)),  # FIX: after_stat() instead of ..count..
    vjust = -0.5,
    size = 2
  ) +
  facet_wrap(~ Platform, scales = "free_y", ncol = 4) +
  theme_minimal() +
  theme(axis.text.x = element_text(angle = 90, hjust = 1)) +
  labs(
    title = "Social Media Frequency Use (Counts)",
    x = NULL,
    y = "Count"
  ) +
  scale_y_continuous(expand = expansion(mult = c(0, 0.15)))

## 2.1b Use frequency per platform (DIG02) – percentages

data_long_pct <- data_long %>%
  dplyr::group_by(Platform, Frequency) %>%
  dplyr::summarise(n = n(), .groups = "drop") %>%
  dplyr::group_by(Platform) %>%
  dplyr::mutate(pct = 100 * n / sum(n)) %>%
  dplyr::ungroup()

ggplot(data_long_pct, aes(x = Frequency, y = pct)) +
  geom_col(fill = "darkorange") +
  geom_text(
    aes(label = sprintf("%.1f%%", pct)),
    vjust = -0.3,
    size = 2
  ) +
  facet_wrap(~ Platform, scales = "free_y") +
  theme_minimal() +
  theme(axis.text.x = element_text(angle = 90, hjust = 1)) +
  labs(
    title = "Social Media Frequency Use (Percentages)",
    x = NULL,
    y = "Percentages"
  ) +
  scale_y_continuous(expand = expansion(mult = c(0, 0.15)))

## 2.2 Platform use per purpose (from multiple choice var DIG03)

# Count "YES" per platform × purpose
data_long_uses <- data_platform_use %>%
  dplyr::filter(Response == "Yes") %>%
  dplyr::group_by(Platform, Purpose) %>%
  dplyr::summarise(n = n(), .groups = "drop")

ggplot(data_long_uses, aes(x = Purpose, y = n)) +
  geom_col(fill = "darkorange") +
  geom_text(
    aes(label = n),
    vjust = -0.3,
    size = 2
  ) +
  facet_wrap(~ Platform, scales = "free_y") +
  theme_minimal() +
  theme(axis.text.x = element_text(angle = 90, hjust = 1)) +
  labs(
    title = "Uses of Social Media Platforms (Counts)",
    x = NULL,
    y = "Count"
  ) +
  scale_y_continuous(expand = expansion(mult = c(0, 0.15)))

# Percentage "YES" per platform × purpose
data_long_uses_pct <- data_platform_use %>%
  dplyr::filter(Response == "Yes") %>%
  dplyr::group_by(Platform, Purpose) %>%
  dplyr::summarise(n = n(), .groups = "drop") %>%
  dplyr::group_by(Platform) %>%
  dplyr::mutate(pct = 100 * n / sum(n)) %>%
  dplyr::ungroup()

ggplot(data_long_uses_pct, aes(x = Purpose, y = pct)) +
  geom_col(fill = "darkorange") +
  geom_text(
    aes(label = sprintf("%.1f%%", pct)),
    vjust = -0.3,
    size = 2
  ) +
  facet_wrap(~ Platform, scales = "free_y") +
  theme_minimal() +
  theme(axis.text.x = element_text(angle = 90, hjust = 1)) +
  labs(
    title = "Uses of Social Media Platforms (Percentages)",
    x = NULL,
    y = "Percentages"
  ) +
  scale_y_continuous(expand = expansion(mult = c(0, 0.15)))

The Italian pattern shows a strongly asymmetric ecosystem: WhatsApp, Facebook, Instagram and YouTube concentrate the bulk of high-frequency users, while platforms linked to news or public debate (Twitter/X, LinkedIn) remain marginal and used mostly sporadically. The contrast between “never users” and “daily users” is especially visible on WhatsApp and Facebook, indicating polarized usage habits. Younger-oriented platforms like TikTok show moderate engagement but still far from the levels of Instagram. Telegram displays an intermediate profile, with a large share of occasional users but fewer daily ones. Overall, the Italian landscape is dominated by interpersonal and entertainment-driven platforms rather than explicitly political ones.

*Compared with Spain, Italian users rely more heavily on WhatsApp and Facebook for daily communication and information exchange. Spain generally shows higher engagement on Instagram and TikTok, reflecting a more youth-skewed and mobile-centric pattern. Twitter/X tends to be more integrated into political conversation in Spain than in Italy, where it remains niche. Overall, Spain exhibits a more diversified platform ecology, while Italy shows stronger dependence on a few dominant channels.

5.1 Sociopolitical variables and platform use

## 2.3 Indicators: political and sociopolitical use per platform
# define data_any from EUPeopleNet for safety
data_any <- EUPeopleNet %>%
  dplyr::select(dplyr::starts_with("dig03_"))

labels_use <- c(
  "friends", "family", "inf_socially", "sup_soc_causes",
  "inf_pol", "sup_pol_causes", "connect_profess",
  "promote_profess", "other"
)

purpose_suffixes <- 1:9

for (i in seq_along(labels_use)) {
  pattern_i <- paste0("_", purpose_suffixes[i], "$")
  cols_i <- grep(pattern_i, names(data_any), value = TRUE)
  
  if (length(cols_i) > 0) {
    data_any[[labels_use[i]]] <- ifelse(
      rowSums(data_any[, cols_i], na.rm = TRUE) > 0, 1, 0
    )
  }
}

# indexes
data_any <- data_any %>%
  mutate(
    PolInf_smed = ifelse(inf_pol == 1, 1, 0),
    PolSup_smed = ifelse(sup_pol_causes == 1, 1, 0),
    PolInfSup_smed = ifelse(inf_pol == 1 | sup_pol_causes == 1, 1, 0),
    PolSocial_smed = ifelse(
      inf_socially == 1 | sup_soc_causes == 1 |
        inf_pol == 1 | sup_pol_causes == 1,
      1, 0
    )
  ) %>%
  mutate(
    across(
      c(PolInf_smed, PolSup_smed, PolSocial_smed, PolInfSup_smed),
      ~ factor(.x, levels = c(0, 1), labels = c("No", "Yes"))
    )
  )

# single graphs
ggplot(data_any, aes(x = PolInf_smed)) +
  geom_bar(fill = "darkgreen") +
  theme_minimal() +
  labs(
    title = "Use of Social Media to Inform Oneself Politically (Any Platform)",
    x = NULL,
    y = "Count"
  )

ggplot(data_any, aes(x = PolSup_smed)) +
  geom_bar(fill = "darkgreen") +
  theme_minimal() +
  labs(
    title = "Use of Social Media to Support Political Causes (Any Platform)",
    x = NULL,
    y = "Count"
  )

ggplot(data_any, aes(x = PolSocial_smed)) +
  geom_bar(fill = "darkgreen") +
  theme_minimal() +
  labs(
    title = "Use of Social Media for Social/Political Information or Support",
    x = NULL,
    y = "Count"
  )

# Facet with all indicators
data_long_2 <- data_any %>%
  tidyr::pivot_longer(
    cols  = c(PolInf_smed, PolSup_smed, PolSocial_smed, PolInfSup_smed),
    names_to = "Indicator",
    values_to = "Value"
  )

ggplot(data_long_2, aes(x = Value)) +
  geom_bar(fill = "darkorange") +
  facet_wrap(~ Indicator, scales = "free_y") +
  theme_minimal() +
  theme(axis.text.x = element_text(angle = 90, hjust = 1)) +
  labs(
    title = "Political/Social Uses of Social Media (Any Platform)",
    x = NULL,
    y = "Count"
  )

5.2 Sociopolitical variables descriptives

## 2.4 sociopolitical descriptives (from data_MP)

# age
ggplot(data_MP, aes(x = Age)) +
  geom_histogram(binwidth = 5, fill = "darkorange", color = "white") +
  theme_minimal() +
  labs(
    title = "Age",
    x = "How old are you?",
    y = "Count"
  ) +
  theme(axis.text.x = element_text(angle = 90, hjust = 1))

# Gender
ggplot(data_MP, aes(x = Sex)) +
  geom_bar(fill = "darkorange") +
  theme_minimal() +
  labs(
    title = "Sex",
    x = "Sex",
    y = "Count"
  ) +
  theme(axis.text.x = element_text(angle = 90, hjust = 1))

# Vote in general elections (2022) 
ggplot(data_MP %>% filter(!is.na(Vote)), aes(x = Vote)) +
  geom_bar(fill = "darkorange") +
  theme_minimal() +
  labs(
    title = "Vote in the Last National Elections (Italy)",
    x = "Which party did you vote for in the last national elections?",
    y = "Count"
  ) +
  theme(axis.text.x = element_text(angle = 90, hjust = 1))

# Left-Right ideological placement (plus do not placed)
ggplot(data_MP %>% filter(!is.na(Left_Right)), aes(x = Left_Right)) +
  geom_bar(fill = "darkorange") +
  theme_minimal() +
  labs(
    title = "Left–Right self-location",
    x = "Left–Right",
    y = "Count"
  ) +
  theme(axis.text.x = element_text(angle = 90, hjust = 1))

# Political interest
ggplot(data_MP, aes(x = Pol_int)) +
  geom_bar(fill = "darkorange") +
  theme_minimal() +
  labs(
    title = "Political interest",
    x = "Political interest",
    y = "Count"
  ) +
  theme(axis.text.x = element_text(angle = 90, hjust = 1))

# Information disagreement exposure
ggplot(data_MP, aes(x = Inf_disagr)) +
  geom_bar(fill = "darkorange") +
  theme_minimal() +
  labs(
    title = "Frequency of exposure to information you disagree with (online)",
    x = "How often do you read, watch or listen to something you disagree with?",
    y = "Count"
  ) +
  theme(axis.text.x = element_text(angle = 90, hjust = 1))

As concerns, Left–Right self-location, the Italian distribution appears weakly structured: middle categories do not dominate, while small, irregular peaks appear across the entire scale. Two elements stand out: (1) a sizeable group placing itself at the far right; (2) a very large “no answer” category, indicating ideological distance, uncertainty, or low cognitive anchoring. Overall, the pattern suggests that a significant share of respondents does not internalize the left–right dimension strongly, while those who do tend to adopt clearer, sometimes polarized positions. Political interest shows a much more coherent profile: the majority falls in the middle (“somewhat interested”), while the extremes—very interested or not interested at all—remain minorities. This points to a “moderately engaged” public: not apathetic, yet not strongly mobilized either. The notable minority reporting no interest at all indicates a segment disconnected from political life. Exposure to viewpoints one disagrees with is concentrated in the middle categories (“sometimes”, “often”), while the extremes—constant exposure or total absence—are limited. This suggests that the online information environment is not fully siloed: many Italians encounter disagreement, though in an intermittent rather than systematic way. Algorithms filter content, but do not completely seal individuals off from cross-cutting perspectives. The Italian data portray an electorate that is ideologically fragmented and only partially anchored to the left–right dimension: the distribution is uneven, with irregular concentrations along the scale and a remarkably high share of respondents who avoid placing themselves altogether.

*The Spanish pattern, in contrast, is noticeably more orderly and symmetrical, suggesting that ideological self-location functions as a clearer cognitive compass. Political interest reinforces this difference: in Italy, most respondents converge in the middle categories, forming a bloc of moderate but not particularly intense engagement; Spain instead shows stronger tails, with more individuals who follow politics very closely and, simultaneously, more who reject it entirely. This produces a more polarized landscape of attentiveness. Finally, exposure to disagreeing views online is more frequent and more evenly spread in Spain, whereas Italians tend to encounter cross-cutting information in a more sporadic and less systematic manner. Taken together, these contrasts depict Spain as a context where political orientations and engagement practices are more crystallized, while Italy appears marked by ideological looseness, moderate interest, and a comparatively softer circulation of dissonant content.

6 PCA

7 PCA preparation

# 4. PCA Preparation ----------------------------------------------------

# 4.1 Select variables: 8 supplementary qualitative + 8 quantitative active
data_PCA1 <- data_MP %>%
  dplyr::select(
    Age_cat, Vote, Sex, Left_Right, Pol_int,
    Inf_disagr, Inf_check, Inf_confirm,
    YouTube, WhatsApp, Twitter_X, Instagram,
    TikTok, Facebook, Linkedin, Telegram
  )

# 4.2 Recoding to fit with ES recoding 

data_PCA1 <- data_PCA1 %>%
  mutate(
    Left_Right = recode(
      as.character(Left_Right),
      
      "0"  = "Ext_left",
      "1"  = "Ext_left",
      
      "2"  = "Left",
      "3"  = "Left",
      
      "4"  = "Center",
      "5"  = "Center",
      "6"  = "Center",
      
      "7"  = "Right",
      "8"  = "Right",
      
      "9"  = "Ext_right",
      "10" = "Ext_right",
      
      "Not placed" = "No_position",
      
      .default = NA_character_
    ),
    
    Left_Right = factor(
      Left_Right,
      levels = c("Ext_left", "Left", "Center", "Right", "Ext_right", "No_position")
    )
  )

# 5. Data Suitability Tests for the PCA ---------------------------------

# 5.0 Correlation matrix of platform variables
cor_matrix <- cor(
  data_PCA1[, 9:16],
  use = "pairwise.complete.obs"
)

# 5.1 Kaiser–Meyer–Olkin (KMO) Measure
psych::KMO(cor_matrix)

Kaiser-Meyer-Olkin factor adequacy
Call: psych::KMO(r = cor_matrix)
Overall MSA =  0.84
MSA for each item = 
  YouTube  WhatsApp Twitter_X Instagram    TikTok  Facebook  Linkedin  Telegram 
     0.89      0.72      0.81      0.83      0.87      0.84      0.84      0.87 

# 5.2 Bartlett’s Test of Sphericity (to test whether correlation matrix differs from identity matrix)

psych::cortest.bartlett(cor_matrix, n = nrow(data_PCA1))

$chisq
[1] 2351.818

$p.value
[1] 0

$df
[1] 28

# 5.3 Multicollinearity and Redundancy (check for variables where r > 0.9)
round(cor_matrix, 2)

          YouTube WhatsApp Twitter_X Instagram TikTok Facebook Linkedin
YouTube      1.00     0.22      0.33      0.31   0.36     0.20     0.30
WhatsApp     0.22     1.00      0.07      0.36   0.15     0.28     0.11
Twitter_X    0.33     0.07      1.00      0.32   0.45     0.16     0.53
Instagram    0.31     0.36      0.32      1.00   0.40     0.33     0.35
TikTok       0.36     0.15      0.45      0.40   1.00     0.21     0.41
Facebook     0.20     0.28      0.16      0.33   0.21     1.00     0.21
Linkedin     0.30     0.11      0.53      0.35   0.41     0.21     1.00
Telegram     0.33     0.16      0.49      0.40   0.41     0.20     0.44
          Telegram
YouTube       0.33
WhatsApp      0.16
Twitter_X     0.49
Instagram     0.40
TikTok        0.41
Facebook      0.20
Linkedin      0.44
Telegram      1.00

caret::findCorrelation(cor_matrix, cutoff = 0.9)

integer(0)

8 PCA Estimation & visualization

# 6. PCA Estimation and Visualization -----------------------------------

# Here the 8 active variables are platforms,the 8 supplementary variables are: Age_cat, Vote, Sex, Left_Right, Pol_int, Inf_disagr, Inf_check, Inf_confirm

PCA1_result <- FactoMineR::PCA(
  data_PCA1,
  quali.sup = 1:8,   
  graph     = FALSE
)

Warning in FactoMineR::PCA(data_PCA1, quali.sup = 1:8, graph = FALSE): Missing
values are imputed by the mean of the variable: you should use the imputePCA
function of the missMDA package

# 6.1 Summary of eigenvalues, contributions, cos², etc.
summary(PCA1_result)

Call:
FactoMineR::PCA(X = data_PCA1, quali.sup = 1:8, graph = FALSE) 


Eigenvalues
                       Dim.1   Dim.2   Dim.3   Dim.4   Dim.5   Dim.6   Dim.7
Variance               3.159   1.207   0.768   0.698   0.620   0.572   0.518
% of var.             39.491  15.090   9.595   8.722   7.749   7.150   6.474
Cumulative % of var.  39.491  54.581  64.176  72.898  80.647  87.797  94.271
                       Dim.8
Variance               0.458
% of var.              5.729
Cumulative % of var. 100.000

Individuals (the 10 first)
                            Dist    Dim.1    ctr   cos2    Dim.2    ctr   cos2
1                       |  0.000 |  0.000  0.000  0.222 |  0.000  0.000  0.956
2                       |  2.944 |  1.380  0.046  0.220 | -0.175  0.002  0.004
3                       |  3.457 |  2.479  0.150  0.514 |  1.286  0.105  0.138
4                       |  1.917 | -0.564  0.008  0.087 |  0.551  0.019  0.083
5                       |  3.821 |  0.725  0.013  0.036 | -0.106  0.001  0.001
6                       |  2.831 | -1.359  0.045  0.230 | -0.369  0.009  0.017
7                       |  2.084 | -1.008  0.025  0.234 |  0.477  0.014  0.052
8                       |  3.095 |  2.772  0.187  0.802 | -0.867  0.048  0.078
9                       |  0.537 | -0.138  0.000  0.066 |  0.033  0.000  0.004
10                      |  3.723 |  1.833  0.082  0.242 | -1.668  0.177  0.201
                           Dim.3    ctr   cos2  
1                       |  0.000  0.000  0.002 |
2                       | -1.875  0.352  0.406 |
3                       |  0.837  0.070  0.059 |
4                       | -0.767  0.059  0.160 |
5                       |  2.811  0.791  0.541 |
6                       |  1.932  0.374  0.466 |
7                       |  0.679  0.046  0.106 |
8                       | -0.337  0.011  0.012 |
9                       | -0.098  0.001  0.033 |
10                      |  0.635  0.040  0.029 |

Variables
                           Dim.1    ctr   cos2    Dim.2    ctr   cos2    Dim.3
YouTube                 |  0.601 11.419  0.361 |  0.040  0.136  0.002 |  0.581
WhatsApp                |  0.389  4.791  0.151 |  0.714 42.224  0.510 |  0.252
Twitter_X               |  0.704 15.682  0.495 | -0.406 13.640  0.165 | -0.074
Instagram               |  0.688 14.987  0.473 |  0.309  7.914  0.096 | -0.036
TikTok                  |  0.695 15.276  0.483 | -0.163  2.195  0.026 |  0.093
Facebook                |  0.459  6.671  0.211 |  0.523 22.686  0.274 | -0.545
Linkedin                |  0.693 15.183  0.480 | -0.308  7.847  0.095 | -0.231
Telegram                |  0.711 15.991  0.505 | -0.201  3.360  0.041 | -0.036
                           ctr   cos2  
YouTube                 43.965  0.337 |
WhatsApp                 8.253  0.063 |
Twitter_X                0.714  0.005 |
Instagram                0.166  0.001 |
TikTok                   1.138  0.009 |
Facebook                38.676  0.297 |
Linkedin                 6.923  0.053 |
Telegram                 0.165  0.001 |

Supplementary categories (the 10 first)
                             Dist     Dim.1    cos2  v.test     Dim.2    cos2
18_24                   |   1.418 |   1.108   0.611   6.486 |  -0.233   0.027
25_34                   |   0.891 |   0.796   0.798   5.984 |  -0.138   0.024
35_44                   |   0.816 |   0.747   0.839   6.248 |   0.121   0.022
45_54                   |   0.239 |   0.127   0.280   1.255 |   0.182   0.579
55_64                   |   0.334 |  -0.268   0.644  -2.525 |   0.138   0.172
65_+                    |   0.992 |  -0.930   0.879 -11.961 |  -0.149   0.023
+Europa                 |   0.623 |   0.043   0.005   0.155 |  -0.203   0.106
Alleanza Verdi-Sinistra |   0.445 |   0.392   0.776   1.836 |  -0.092   0.042
Azione-Italia Viva      |   0.419 |  -0.075   0.032  -0.416 |  -0.189   0.205
Did not vote            |   0.240 |  -0.090   0.141  -0.716 |  -0.198   0.681
                         v.test     Dim.3    cos2  v.test  
18_24                    -2.203 |   0.592   0.174   7.028 |
25_34                    -1.681 |   0.077   0.007   1.175 |
35_44                     1.634 |  -0.085   0.011  -1.441 |
45_54                     2.922 |  -0.013   0.003  -0.256 |
55_64                     2.111 |  -0.076   0.052  -1.454 |
65_+                     -3.098 |  -0.092   0.009  -2.411 |
+Europa                  -1.169 |   0.110   0.031   0.799 |
Alleanza Verdi-Sinistra  -0.695 |  -0.033   0.006  -0.315 |
Azione-Italia Viva       -1.705 |   0.115   0.076   1.302 |
Did not vote             -2.544 |   0.070   0.086   1.132 |

# 6.1.1 Variables relation to Dim1 and Dim2
PCA1_desc <- FactoMineR::dimdesc(PCA1_result, axes = 1:2, proba = 0.05)
PCA1_desc$Dim.1

Link between the variable and the continuous variables (R-square)
=================================================================================
          correlation       p.value
Telegram    0.7107833 1.041739e-200
Twitter_X   0.7038717 3.480574e-195
TikTok      0.6946969 4.276470e-188
Linkedin    0.6925912 1.661726e-186
Instagram   0.6881031 3.654681e-183
YouTube     0.6006333 2.210095e-128
Facebook    0.4590983  8.225746e-69
WhatsApp    0.3890403  2.895111e-48

Link between the variable and the categorical variable (1-way anova)
=============================================
                    R2      p.value
Age_cat     0.16327212 6.198418e-48
Inf_check   0.06810038 6.392291e-19
Inf_disagr  0.06628292 2.200328e-18
Inf_confirm 0.05654758 1.564022e-15
Left_Right  0.02422511 1.796380e-05
Vote        0.02712709 9.909615e-05

Link between variable and the categories of the categorical variables
================================================================
                                     Estimate      p.value
Age_cat=18_24                       0.8448049 6.314004e-11
Age_cat=35_44                       0.4837189 3.139509e-10
Age_cat=25_34                       0.5330018 1.714502e-09
Inf_check=Inf_check_Very often      0.8497859 1.457769e-08
Inf_disagr=Inf_disagr_Very often    0.8356138 7.353664e-08
Inf_disagr=Inf_disagr_Often         0.4127128 1.234073e-05
Inf_confirm=Inf_confirm_Very often  0.6980323 1.294134e-05
Inf_check=Inf_check_Often           0.3993228 1.547763e-04
Vote=Forza Italia                   0.8322058 3.113801e-04
Inf_confirm=Inf_confirm_Often       0.3868089 1.154004e-03
Left_Right=Ext_right                0.3924648 1.316871e-03
Vote=Movimento 5 Stelle             0.2151232 9.891616e-03
Left_Right=Right                    0.2026920 2.072619e-02
Sex=Non-binary                      1.0454188 4.902130e-02
Left_Right=No_position             -0.3202688 3.374237e-02
Age_cat=55_64                      -0.5312895 1.150529e-02
Left_Right=Left                    -0.3648603 3.203779e-03
Vote=Partito Democratico           -0.4564384 2.558011e-04
Inf_confirm=Inf_confirm_Rarely     -0.4923457 4.336972e-06
Inf_disagr=Inf_disagr_Rarely       -0.4953230 3.847349e-06
Inf_check=Inf_check_Rarely         -0.5263152 5.858813e-07
Inf_disagr=Inf_disagr_Never        -0.7667999 2.008965e-08
Inf_check=Inf_check_Never          -0.7816178 1.656179e-08
Inf_confirm=Inf_confirm_Never      -0.7632705 2.767379e-09
Age_cat=65_+                       -1.1933958 8.630145e-35

PCA1_desc$Dim.2

Link between the variable and the continuous variables (R-square)
=================================================================================
          correlation       p.value
WhatsApp    0.7139505 2.701010e-203
Facebook    0.5233166  2.280502e-92
Instagram   0.3090819  3.383618e-30
TikTok     -0.1627844  3.519636e-09
Telegram   -0.2013846  2.266075e-13
Linkedin   -0.3077723  6.067580e-30
Twitter_X  -0.4057790  9.692650e-53

Link between the variable and the categorical variable (1-way anova)
=============================================
                    R2      p.value
Sex        0.037135620 2.156569e-11
Age_cat    0.020498489 6.094449e-05
Vote       0.022809086 9.041707e-04
Left_Right 0.011570927 1.961848e-02
Inf_check  0.008981723 1.971118e-02

Link between variable and the categories of the categorical variables
================================================================
                                 Estimate      p.value
Sex=Female                      0.3038516 2.406963e-12
Age_cat=45_54                   0.1952296 3.438103e-03
Left_Right=Ext_left             0.2183157 2.284974e-02
Vote=Partito Democratico        0.1603081 3.239774e-02
Age_cat=55_64                   0.1515135 3.474535e-02
Age_cat=18_24                  -0.2195176 2.757226e-02
Vote=Did not vote              -0.1609720 1.091000e-02
Left_Right=No_position         -0.1506509 1.028280e-02
Inf_check=Inf_check_Very often -0.2023231 1.938446e-03
Age_cat=65_+                   -0.1358246 1.922577e-03
Vote=Other party               -0.3415773 1.809898e-03
Sex=Male                       -0.1192441 4.004205e-12

# 6.2 Initial biplot – active variables (social media platforms’ use)
p <- factoextra::fviz_pca_var(
  PCA1_result,
  col.var       = "contrib",
  gradient.cols = c("#00AFBB", "#E7B800", "#FC4E07"),
  repel         = TRUE
)

Warning: Using `size` aesthetic for lines was deprecated in ggplot2 3.4.0.
ℹ Please use `linewidth` instead.
ℹ The deprecated feature was likely used in the ggpubr package.
  Please report the issue at <https://github.com/kassambara/ggpubr/issues>.

Warning: `aes_string()` was deprecated in ggplot2 3.0.0.
ℹ Please use tidy evaluation idioms with `aes()`.
ℹ See also `vignette("ggplot2-in-packages")` for more information.
ℹ The deprecated feature was likely used in the factoextra package.
  Please report the issue at <https://github.com/kassambara/factoextra/issues>.

p

The PCA reveals a very clear structure in social-media usage patterns. Dimension 1 (39%) separates high-intensity, mainstream social platforms—WhatsApp, Facebook, Instagram—which cluster together and strongly shape this axis. This component can be read as a general intensity of social-media use. Dimension 2 (15%) contrasts visual/entertainment platforms (YouTube, TikTok) with more informational or professional platforms (Twitter/X, LinkedIn, Telegram). This dimension captures a shift from entertainment-oriented to information-oriented usage. The contribution levels show that WhatsApp, Facebook, and Instagram dominate the first dimension, while YouTube and TikTok drive the second. PCA suggests two robust patterns: (a) General social-media intensity (Dim1); (b) Entertainment-visual vs. informational-professional orientation (Dim2). The configuration aligns well with typical distinctions found in the literature.

*The Italian and Spanish PCA maps share the same basic structure: Dim1 captures general social-media intensity, while Dim2 separates entertainment-oriented platforms (YouTube, TikTok) from informational/professional ones (Twitter/X, LinkedIn). The difference lies in the degree of separation. In Italy, Facebook, WhatsApp and Instagram cluster tightly, showing a more homogeneous mainstream use. In Spain, this cluster is more dispersed and the contrast along Dim2 is sharper, especially for Twitter/X and LinkedIn, which are more clearly positioned as informational tools. YouTube is more central in the Italian plot, indicating more overlap with generalist use, whereas in Spain its entertainment orientation is more distinct. In short: Italy shows a more blended pattern; Spain shows clearer segmentation of platform types.

9 Centroids of Supplementary Qualitative Variables in the PCA of Social Media Platforms

# 7. Categorical Supplementary Variables (Centroids) --------------------

# Extract centroides of qualitative supplementary variables
get_sup_coords <- function(pca_result, variable_name, dims = 2) {
  # 1) data extraction
  quali_sup_data <- pca_result$call$quali.sup
  if ("quali.sup" %in% names(quali_sup_data)) {
    quali_sup_data <- quali_sup_data$quali.sup
  }
  
  # 2) check variables
  if (!variable_name %in% names(quali_sup_data)) {
    stop(paste("Variable", variable_name, "not found among supplementary qualitative variables."))
  }
  
  # 3) centroids (avg per category)
  bary <- pca_result$call$quali.sup$barycentre
  
  # Rows selection for variables categories
  var_categories <- levels(quali_sup_data[[variable_name]])
  bary_var <- bary[rownames(bary) %in% var_categories, , drop = FALSE]
  
  if (nrow(bary_var) == 0) {
    stop(paste("No barycentres found for variable", variable_name))
  }
  
  # 4) Avg and st.dev used in PCA
  means <- pca_result$call$centre
  sds   <- pca_result$call$ecart.type
  
  # Centroides standardization
  bary_scaled <- sweep(bary_var, 2, means, "-")
  bary_scaled <- sweep(bary_scaled, 2, sds, "/")
  
  # 5) PCA axis projection
  loadings  <- pca_result$var$coord
  projected <- as.matrix(bary_scaled) %*% as.matrix(loadings[, 1:dims])
  
  # 6) Output in data frame
  projected_df <- as.data.frame(projected)
  colnames(projected_df) <- paste0("Dim.", 1:dims)
  projected_df$Category  <- rownames(projected_df)
  projected_df$Variable  <- variable_name
  
  return(projected_df)
}

# Centroides coordinates per each supplementary variable
age_coords      <- get_sup_coords(PCA1_result, "Age_cat")
vote_coords     <- get_sup_coords(PCA1_result, "Vote")
Pol_int_coords  <- get_sup_coords(PCA1_result, "Pol_int")
LRight_coords   <- get_sup_coords(PCA1_result, "Left_Right")
Sex_coords      <- get_sup_coords(PCA1_result, "Sex")
infdis_coords   <- get_sup_coords(PCA1_result, "Inf_disagr")
infcheck_coords <- get_sup_coords(PCA1_result, "Inf_check")
infconf_coords  <- get_sup_coords(PCA1_result, "Inf_confirm")

# 7.1 Biplot with vote centroids
p + 
  geom_point(
    data = vote_coords,
    aes(x = Dim.1, y = Dim.2),
    color = "darkblue", size = 3, shape = 20
  ) +
  ggrepel::geom_text_repel(
    data = vote_coords,
    aes(x = Dim.1, y = Dim.2, label = Category),
    color = "darkblue",
    size  = 3.5
  )

# 7.2 Biplot with age centroids
p + 
  geom_point(
    data = age_coords,
    aes(x = Dim.1, y = Dim.2),
    color = "darkblue", size = 3, shape = 20
  ) +
  ggrepel::geom_text_repel(
    data = age_coords,
    aes(x = Dim.1, y = Dim.2, label = Category),
    color = "darkblue",
    size  = 3.5
  )

# 7.4 Biplot with political interest centroids
p + 
  geom_point(
    data = Pol_int_coords,
    aes(x = Dim.1, y = Dim.2),
    color = "darkblue", size = 3, shape = 20
  ) +
  ggrepel::geom_text_repel(
    data = Pol_int_coords,
    aes(x = Dim.1, y = Dim.2, label = Category),
    color = "darkblue",
    size  = 3.5
  )

# 7.5 Biplot with ideological placement centroids
p + 
  geom_point(
    data = LRight_coords,
    aes(x = Dim.1, y = Dim.2),
    color = "darkblue", size = 3, shape = 20
  ) +
  ggrepel::geom_text_repel(
    data = LRight_coords,
    aes(x = Dim.1, y = Dim.2, label = Category),
    color = "darkblue",
    size  = 3.5
  )

# 7.6 Biplot with gender centroids
p + 
  geom_point(
    data = Sex_coords,
    aes(x = Dim.1, y = Dim.2),
    color = "darkblue", size = 3, shape = 20
  ) +
  ggrepel::geom_text_repel(
    data = Sex_coords,
    aes(x = Dim.1, y = Dim.2, label = Category),
    color = "darkblue",
    size  = 3.5
  )

# 7.7 Biplot with information disagreement centroids
p + 
  geom_point(
    data = infdis_coords,
    aes(x = Dim.1, y = Dim.2),
    color = "darkblue", size = 3, shape = 20
  ) +
  ggrepel::geom_text_repel(
    data = infdis_coords,
    aes(x = Dim.1, y = Dim.2, label = Category),
    color = "darkblue",
    size  = 3.5
  )

# 7.8 Biplot with inf_check centroids
p + 
  geom_point(
    data = infcheck_coords,
    aes(x = Dim.1, y = Dim.2),
    color = "darkblue", size = 3, shape = 20
  ) +
  ggrepel::geom_text_repel(
    data = infcheck_coords,
    aes(x = Dim.1, y = Dim.2, label = Category),
    color = "darkblue",
    size  = 3.5
  )

# 7.9 Biplot with info_confirmation centroids
p + 
  geom_point(
    data = infconf_coords,
    aes(x = Dim.1, y = Dim.2),
    color = "darkblue", size = 3, shape = 20
  ) +
  ggrepel::geom_text_repel(
    data = infconf_coords,
    aes(x = Dim.1, y = Dim.2, label = Category),
    color = "darkblue",
    size  = 3.5
  )

Across all PCA plots, the structure of social-media use in the Italian data is remarkably stable: the first dimension consistently captures a gradient of general intensity, defined by the tight cluster formed by WhatsApp, Facebook and Instagram, while the second dimension distinguishes visual/entertainment platforms such as YouTube and TikTok from informational or professional tools like Twitter/X, LinkedIn and Telegram. When the political and socio-demographic variables are projected into this space, they do not reshape the underlying structure but position themselves relative to these two axes. Voting categories and left–right self-placement remain very close to the centre of the map, indicating that neither variable is strongly associated with specific platform clusters. Political interest, instead, shows a clearer pattern: higher-interest respondents lie nearer to the informational/professional platforms, whereas low-interest groups appear toward the lower-intensity side of Dim1. Gender categories display minimal separation and stay largely central, which suggests limited differentiation in platform use. The same modest but coherent tendency appears with exposure to political disagreement: respondents who report more frequent exposure position somewhat closer to the informational side of Dim2, while those who report rare or no exposure remain centrally located. Taken together, these results show that the Italian PCA distinguishes platform types in a consistent and interpretable way, while most political and demographic factors exert only weak differentiation, with political interest—and to a lesser extent exposure to disagreement—showing the most structured alignment with the informational dimension.

*Across both countries, the overall structure of the PCA is highly similar. In Italy and Spain alike, Dim1 reflects a gradient of general social-media intensity, with WhatsApp, Facebook and Instagram forming a compact cluster with the strongest contributions. Likewise, Dim2 separates YouTube and TikTok—positioned toward the visual/entertainment side—from Twitter/X, LinkedIn and Telegram, which align with the informational/professional pole. This shared structure suggests that, in both contexts, platform types organise usage patterns in a remarkably consistent way. The main differences emerge when the supplementary variables are projected. In Italy, most political and demographic categories remain close to the origin, indicating weak differentiation across all variables except political interest, which shows the clearest alignment with the informational dimension. In Spain, by contrast, some variables display slightly more pronounced positioning, especially on Dim2. Left–right ideology, interest, exposure to disagreement, and information-checking behaviours tend to stretch further toward the informational or low-intensity sides than in the Italian maps. This indicates that in Spain, political engagement variables are somewhat more structured in relation to platform types. A second difference concerns the dispersion of categories. Italian categories are generally more compact and central, while Spanish categories show greater spread, especially for political interest and the information-seeking dimensions (inf_check, inf_confirm). In Spain, high-engagement levels cluster more distinctly near Twitter/X and LinkedIn, and low-engagement categories more clearly toward low-intensity areas. The gradients are therefore clearer and more directional in the Spanish results. Despite these differences in the strength of associations, the core geometry remains the same in both countries: (a) intensive/mainstream use on Dim1; (b) entertainment vs. informational orientation on Dim2; (c) political variables projecting onto this space without altering its structure. Italy and Spain share the same fundamental platform structure, but Spain shows slightly stronger alignment between political engagement variables and the informational–entertainment axis, while Italy displays a more central and compact distribution of categories with fewer marked distinctions.

10 Principal Component Analysis Including Numerical Supplementary Variables: Political and Social Uses of Social Media

# 8. Numerical Supplementary Variables: Political/Social Uses of Social Media ----

# New dataset for PCA with following variables: inf_socially, inf_pol, sup_soc_causes, sup_pol_causes 

data_PCA2 <- data_PCA1 %>%
  dplyr::bind_cols(
    data_any %>%
      dplyr::select(inf_socially, inf_pol, sup_soc_causes, sup_pol_causes)
  ) %>%
  # check numerical
  dplyr::mutate(
    dplyr::across(
      c(inf_socially, inf_pol, sup_soc_causes, sup_pol_causes),
      ~ as.numeric(.x)
    )
  )

# Check
 str(data_PCA2)  

tibble [1,301 × 20] (S3: tbl_df/tbl/data.frame)
 $ Age_cat       : Factor w/ 7 levels "0_17","18_24",..: 5 3 5 5 2 2 6 3 5 3 ...
  ..- attr(*, "label")= chr "EDAD RECODE"
 $ Vote          : Factor w/ 10 levels "+Europa","Alleanza Verdi-Sinistra",..: 7 NA 5 2 10 5 5 6 10 9 ...
 $ Sex           : Factor w/ 3 levels "Male","Female",..: 1 2 2 1 2 1 2 2 2 1 ...
  ..- attr(*, "label")= chr "Which gender do you identify with?"
 $ Left_Right    : Factor w/ 6 levels "Ext_left","Left",..: 3 NA 3 2 NA 3 6 NA 3 3 ...
 $ Pol_int       : Factor w/ 4 levels "Very interested",..: 3 3 4 2 3 2 3 2 3 2 ...
  ..- attr(*, "label")= chr "How interested would you say you are in politics - are you...?"
 $ Inf_disagr    : Factor w/ 5 levels "Very often","Often",..: 5 3 5 2 5 5 4 3 3 3 ...
  ..- attr(*, "label")= chr "Read, watch or listen to something online you disagree with_When looking for new"
 $ Inf_check     : Factor w/ 5 levels "Very often","Often",..: 5 3 5 2 5 5 4 3 3 2 ...
  ..- attr(*, "label")= chr "Check a news source that’s different from what you normally read, watch or lis"
 $ Inf_confirm   : Factor w/ 5 levels "Very often","Often",..: 5 3 5 2 5 5 4 3 3 2 ...
  ..- attr(*, "label")= chr "Try to confirm information you find by searching online for another source_When"
 $ YouTube       : int [1:1301] NA 2 7 3 7 6 6 5 4 6 ...
 $ WhatsApp      : int [1:1301] NA 6 7 6 6 5 5 6 6 5 ...
 $ Twitter_X     : int [1:1301] NA NA NA 1 1 1 1 6 NA 6 ...
 $ Instagram     : int [1:1301] NA 6 7 5 7 5 4 5 3 2 ...
 $ TikTok        : int [1:1301] NA NA 7 1 7 1 1 6 NA 6 ...
 $ Facebook      : int [1:1301] NA 6 7 5 1 1 5 6 5 5 ...
 $ Linkedin      : int [1:1301] NA 6 NA 4 1 1 1 4 NA 1 ...
 $ Telegram      : int [1:1301] NA 5 NA 1 2 1 1 5 NA 7 ...
 $ inf_socially  : num [1:1301] 0 1 0 1 0 0 0 1 1 1 ...
 $ inf_pol       : num [1:1301] 0 1 0 0 0 0 0 1 1 0 ...
 $ sup_soc_causes: num [1:1301] 0 1 0 0 0 0 0 1 0 1 ...
 $ sup_pol_causes: num [1:1301] 0 1 0 0 0 0 0 1 0 0 ...

# NEW PCA estimation
#Active variables: the eight social media platforms 
#Supplementary qualitative variables: Age_cat, Vote, Sex, Left_Right, Pol_int, Inf_disagr, Inf_check, and Inf_confirm.
#Supplementary quantitative variables: inf_socially, inf_pol, sup_soc_causes, and sup_pol_causes.

PCA2_result <- FactoMineR::PCA(
  data_PCA2,
  quanti.sup = 17:20,  # supplementari numeriche
  quali.sup  = 1:8,    # supplementari qualitative
  graph      = FALSE
)

Warning in FactoMineR::PCA(data_PCA2, quanti.sup = 17:20, quali.sup = 1:8, :
Missing values are imputed by the mean of the variable: you should use the
imputePCA function of the missMDA package

# Biplot of active variables with numerical supplementary variables
factoextra::fviz_pca_var(
  PCA2_result,
  col.var       = "contrib",
  gradient.cols = c("#00AFBB", "#E7B800", "#FC4E07"),
  repel         = TRUE
)

In the Italian data, platform use is organised around a compact mainstream cluster—WhatsApp, Facebook and Instagram—that structures the first dimension, while informational and political purposes remain relatively central, producing a less sharply differentiated configuration across platform types. In both countries the structure is almost identical: Dim1 reflects a gradient of general social-media intensity, driven by WhatsApp, Facebook and Instagram, while Dim2 separates visual/entertainment platforms (YouTube, TikTok) from informational/professional platforms (Twitter/X, LinkedIn, Telegram). This shared geometry indicates that users in Italy and Spain organise their platform use along the same two underlying dimensions. In both countries, informational and political purposes (information socially; supporting social or political causes; political information) cluster near Twitter/X, LinkedIn, Telegram, aligning with the informational pole of Dim2. Meanwhile, more general or social purposes fall closer to the centre or slightly toward the mainstream cluster, indicating that they overlap with general, high-intensity use but do not distinguish platform types as sharply. The main difference lies in the strength of alignment. In the Spanish data, political and informational purposes appear more clearly pulled toward Twitter/X and LinkedIn, producing a slightly stronger separation along Dim2. In the Italian map, the same purposes are present but show a more compact and central position, suggesting weaker differentiation across platform types. The two countries share the same structure, but Spain displays sharper contrasts between entertainment-oriented and informational/political use, whereas Italy shows a more blended configuration with less pronounced distinctions.

11 Factor Analysis of online and offline political participation

# 9. Factor Analysis of online and offline political participation ----

# 9.0 Dictionary: original items -> new variable names -> activities ----
participation_dict <- tibble::tibble(
  var_bin = c(
    # --- ONLINE (13) ---
    "parton_1",  "parton_2",  "parton_3",  "parton_4",  "parton_5",
    "parton_6",  "parton_7",  "parton_8",  "parton_9",  "parton_10",
    "parton_11", "parton_12", "parton_13",
    # --- OFFLINE (13) ---
    "partoff_1",  "partoff_2",  "partoff_3",  "partoff_4",  "partoff_5",
    "partoff_6",  "partoff_7",  "partoff_8",  "partoff_9",  "partoff_10",
    "partoff_11", "partoff_12", "partoff_13"
  ),
  new_name = c(
    # --- ONLINE ---
    "pol_inf_on",      # parton_1  – actively searched for political information online
    "liked_on",        # parton_2  – liked/followed political content
    "shared_on",       # parton_3  – shared/retweeted political content
    "posted_on",       # parton_4  – posted own political content
    "discuss_on",      # parton_5  – discussed politics online
    "comment_on",      # parton_6  – commented on political content
    "profile_on",      # parton_7  – changed profile picture/status for political reasons
    "vol_party_on",    # parton_8  – volunteered online for party/candidate
    "donat_party_on",  # parton_9  – donated money online to party/candidate
    "donat_NGO_on",    # parton_10 – donated money online to NGO/association
    "vol_social_on",   # parton_11 – volunteered online for NGO/association
    "contact_on",      # parton_12 – contacted politician/official online
    "petition_on",     # parton_13 – signed online petition
    # --- OFFLINE ---
    "pol_meet_off",      # partoff_1  – attended political meeting/rally
    "strike_off",        # partoff_2  – took part in strike
    "demonst_off",       # partoff_3  – took part in demonstration/march
    "org_demonst_off",   # partoff_4  – helped organise demonstration
    "discuss_off",       # partoff_5  – discussed politics face-to-face
    "vol_party_off",     # partoff_6  – volunteered offline for party/candidate
    "donat_party_off",   # partoff_7  – donated money offline to party/candidate
    "donat_NGO_off",     # partoff_8  – donated money offline to NGO/association
    "contact_off",       # partoff_9  – contacted politician/official offline
    "read_campaign_off", # partoff_10 – read leaflets/campaign material
    "petition_off",      # partoff_11 – signed paper petition
    "boycott_off",       # partoff_12 – (boy)cott/buycott participation
    "button_off"         # partoff_13 – wore badge/button/sticker
  ),
  activity = c(
    # --- ONLINE (descrizione sintetica) ---
    "Search political information online",
    "Like/follow political content",
    "Share/retweet political content",
    "Post own political content",
    "Discuss politics online",
    "Comment on political content",
    "Change profile/status for politics",
    "Volunteer online for party/candidate",
    "Donate money online to party/candidate",
    "Donate money online to NGO/association",
    "Volunteer online for NGO/association",
    "Contact politician/official online",
    "Sign online petition",
    # --- OFFLINE ---
    "Attend political meeting/rally",
    "Take part in strike",
    "Take part in demonstration/march",
    "Help organise demonstration",
    "Discuss politics face-to-face",
    "Volunteer offline for party/candidate",
    "Donate money offline to party/candidate",
    "Donate money offline to NGO/association",
    "Contact politician/official offline",
    "Read leaflets/campaign material",
    "Sign paper petition",
    "Take part in (boy)cott/buycott",
    "Wear badge/button/sticker"
  )
)

# 9.1 Build data_FACTOR with online + offline items ---------------------

data_FACTOR <- EUPeopleNet %>%
  dplyr::mutate(id = dplyr::row_number()) %>%
  dplyr::select(
    id,
    dplyr::all_of(participation_dict$var_bin)
  )

# Values are already 0/1; in case "9 = prefer not to answer" is present, recode to NA
data_FACTOR <- data_FACTOR %>%
  dplyr::mutate(
    dplyr::across(
      -id,
      ~ dplyr::na_if(.x, 9)
    )
  ) %>%
  dplyr::mutate(
    dplyr::across(
      -id,
      ~ as.numeric(.x)
    )
  )

# >>> QUI LA CORREZIONE IMPORTANTE <<<
# new_name = old_name
rename_vec <- stats::setNames(
  participation_dict$var_bin,   # valori = vecchi nomi (parton_1, partoff_1, ...)
  participation_dict$new_name   # nomi  = nuovi nomi (pol_inf_on, pol_meet_off, ...)
)

data_FACTOR <- data_FACTOR %>%
  dplyr::rename(!!!rename_vec)

# 9.2 Prepare and test suitability for Factor Analysis ------------------

# Drop rows with too many missing values (> 50%)
data_FACTOR <- data_FACTOR %>%
  dplyr::filter(
    rowMeans(is.na(dplyr::across(-id))) < 0.5
  )

# 1) Number of NA per variable 
cat("Number of missing values per variable:\n")

Number of missing values per variable:

print(colSums(is.na(dplyr::select(data_FACTOR, -id))))

       pol_inf_on          liked_on         shared_on         posted_on 
               23                23                35                21 
       discuss_on        comment_on        profile_on      vol_party_on 
               27                27                31                35 
   donat_party_on      donat_NGO_on     vol_social_on        contact_on 
               24                39                30                23 
      petition_on      pol_meet_off        strike_off       demonst_off 
               16                22                23                20 
  org_demonst_off       discuss_off     vol_party_off   donat_party_off 
               25                21                22                27 
    donat_NGO_off       contact_off read_campaign_off      petition_off 
               30                17                25                26 
      boycott_off        button_off 
               34                23 

# 2) List of participation variables
part_vars <- setdiff(names(data_FACTOR), "id")

# 3) Correlation matrix (pairwise complete)
cor_matrix_part <- cor(
  dplyr::select(data_FACTOR, dplyr::all_of(part_vars)),
  use = "pairwise.complete.obs"
)

# 4) Optional corrplot
if (requireNamespace("corrplot", quietly = TRUE)) {
  corrplot::corrplot(
    cor_matrix_part,
    method = "color",
    type   = "upper",
    tl.cex = 0.7,
    tl.col = "black",
    title  = "Correlation matrix Online/Offline participation items (Italy)",
    mar    = c(0, 0, 2, 0)
  )
} else {
  cat("Package 'corrplot' not installed. Skipping correlation plot.\n")
}

# 5) Multicollinearity check (cutoff 0.9)
if (requireNamespace("caret", quietly = TRUE)) {
  high_corr <- caret::findCorrelation(cor_matrix_part, cutoff = 0.9)
  if (length(high_corr) > 0) {
    cat("Variables with high correlation (> 0.9):\n")
    print(part_vars[high_corr])
  } else {
    cat("No variables show excessive multicollinearity.\n")
  }
} else {
  cat("Package 'caret' not installed. Skipping multicollinearity check.\n")
}

No variables show excessive multicollinearity.

# 6) KMO e Bartlett 
psych::KMO(cor_matrix_part)

Kaiser-Meyer-Olkin factor adequacy
Call: psych::KMO(r = cor_matrix_part)
Overall MSA =  0.92
MSA for each item = 
       pol_inf_on          liked_on         shared_on         posted_on 
             0.89              0.91              0.94              0.95 
       discuss_on        comment_on        profile_on      vol_party_on 
             0.91              0.91              0.95              0.95 
   donat_party_on      donat_NGO_on     vol_social_on        contact_on 
             0.94              0.82              0.95              0.93 
      petition_on      pol_meet_off        strike_off       demonst_off 
             0.92              0.95              0.95              0.94 
  org_demonst_off       discuss_off     vol_party_off   donat_party_off 
             0.94              0.86              0.93              0.94 
    donat_NGO_off       contact_off read_campaign_off      petition_off 
             0.82              0.91              0.93              0.92 
      boycott_off        button_off 
             0.95              0.94 

psych::cortest.bartlett(
  cor_matrix_part,
  n = nrow(data_FACTOR)
)

$chisq
[1] 10451.93

$p.value
[1] 0

$df
[1] 325

# 9.3 Exploratory Factor Analysis --------------------------------------

# Polychoric correlation (good for binary/ordinal items)
cor_poly <- psych::polychoric(
  dplyr::select(data_FACTOR, dplyr::all_of(part_vars))
)$rho

# Parallel analysis (decide nfactors; here we keep 5 as in Spanish script)
fa.parallel(
  cor_poly,
  n.obs = nrow(data_FACTOR),
  fm    = "minres",
  fa    = "fa"
)

Parallel analysis suggests that the number of factors =  8  and the number of components =  NA 

# EFA with 5 factors, oblimin rotation
efa_result <- psych::fa(
  dplyr::select(data_FACTOR, dplyr::all_of(part_vars)),
  nfactors = 5,
  rotate   = "oblimin",
  fm       = "minres",
  scores   = "regression"
)

# Inspect loadings (only > .30)
print(efa_result$loadings, cutoff = 0.30)

Loadings:
                  MR1    MR2    MR3    MR5    MR4   
pol_inf_on                0.309                0.476
liked_on                  0.509                     
shared_on                 0.555                     
posted_on                 0.540                     
discuss_on                0.708                     
comment_on                0.682                     
profile_on         0.440                            
vol_party_on       0.484                            
donat_party_on     0.333                            
donat_NGO_on                     0.792              
vol_social_on                                       
contact_on                              0.535       
petition_on                                         
pol_meet_off       0.314                0.300       
strike_off         0.455                            
demonst_off        0.617                            
org_demonst_off    0.488                            
discuss_off                                    0.483
vol_party_off      0.687                            
donat_party_off    0.432                            
donat_NGO_off                    0.715              
contact_off                             0.759       
read_campaign_off                              0.408
petition_off       0.393                            
boycott_off                                    0.302
button_off         0.610                            

                 MR1   MR2   MR3   MR5   MR4
SS loadings    2.932 2.130 1.417 1.176 1.148
Proportion Var 0.113 0.082 0.055 0.045 0.044
Cumulative Var 0.113 0.195 0.249 0.294 0.339

# Variance explained and fit indices
efa_result$Vaccounted

                            MR1        MR2        MR3        MR5        MR4
SS loadings           3.5954713 2.58411074 1.68179565 1.60764309 1.32603612
Proportion Var        0.1382874 0.09938887 0.06468445 0.06183243 0.05100139
Cumulative Var        0.1382874 0.23767623 0.30236068 0.36419311 0.41519450
Proportion Explained  0.3330665 0.23937908 0.15579313 0.14892400 0.12283734
Cumulative Proportion 0.3330665 0.57244553 0.72823865 0.87716266 1.00000000

efa_result$TLI

[1] 0.917059

efa_result$RMSEA

     RMSEA      lower      upper confidence 
0.04523795 0.04174861 0.04882587 0.90000000 

# Optional: diagram
psych::fa.diagram(
  efa_result,
  main   = "Exploratory Factor Analysis – Political Participation",
  simple = FALSE
)

# Optional: 2D plot of first two factors
loadings_df <- as.data.frame(efa_result$loadings[1:length(part_vars), ])
loadings_df$variable <- rownames(loadings_df)

ggplot(loadings_df, aes(x = MR1, y = MR2, label = variable)) +
  geom_point(color = "steelblue", size = 3) +
  geom_text(vjust = -0.5, size = 3) +
  geom_hline(yintercept = 0, linetype = "dashed", color = "gray60") +
  geom_vline(xintercept = 0, linetype = "dashed", color = "gray60") +
  theme_minimal() +
  labs(
    title = "Factor Loadings (EFA)",
    x = "Factor 1",
    y = "Factor 2"
  )

# 9.4 Save factor scores and merge with PCA dataset --------------------

# Extract factor scores
factor_scores <- as.data.frame(efa_result$scores)

# Rename factors
factor_scores <- factor_scores %>%
  dplyr::rename(
    Info_Online      = MR1,
    Partisan_mobil   = MR2,
    Civic_Engagement = MR3,
    Protest_Action   = MR4,
    Info_Offline     = MR5
  ) %>%
  dplyr::mutate(id = data_FACTOR$id)

# Merge factor scores into PCA dataset (data_PCA2 -> data_PCA3)
data_PCA3 <- data_PCA2 %>%
  dplyr::mutate(id = dplyr::row_number()) %>%
  dplyr::left_join(factor_scores, by = "id") %>%
  dplyr::select(-id)

The Italian diagnostics suggest that the items are well suited for factor analysis. The table of missing values shows that non-response is present but limited and reasonably balanced across items, with no variable dominated by NAs, and the row filter (>50% missing) has removed the most problematic cases. The correlation matrix displays a dense pattern of positive, medium-sized associations among most online and offline participation items, which is exactly what you expect if they share a set of common latent dimensions; at the same time, the correlations never become extremely high, and the multicollinearity check confirms that no pair exceeds the 0.9 threshold. The KMO output indicates adequate overall sampling adequacy and acceptable MSA values for individual items, while Bartlett’s test rejects the null of an identity matrix, meaning that the correlation structure is far from random. Taken together, these diagnostics support the decision to proceed with an exploratory factor analysis on the Italian online/offline participation items.

The Italian factor-analysis results show a clear and coherent structure, supported by diagnostics and by the factor loadings. The parallel analysis indicates that four factors should be retained, with a steep drop after the first component and simulated data crossing the empirical line around the fourth dimension. This already suggests that political participation in the Italian data is multidimensional, not reducible to a single online/offline divide. The loading matrix confirms this interpretation. A first factor groups the classic forms of low-threshold online engagement such as liking, sharing, commenting or posting; these items load together with profile/status changes and online petitions. A second factor gathers contact- and donation-oriented behaviours, both online and offline, suggesting a shared “resource-intensive” component. A third dimension captures collective or public-facing offline actions, such as offline volunteering, attending meetings, demonstrating or organising demonstrations. Finally, a fourth factor isolates information-seeking behaviours, distinguishing the more passive forms of attention to politics from active mobilisation. The loading plot visualises these patterns well: online expressive actions cluster tightly in one corner, collective offline modes are separate, and donation/contact behaviours stand apart from both. The factors are not redundant: cross-loadings remain generally low, indicating that each dimension taps a distinct component of political engagement. The Italian data suggest a four-pillar model of political participation: expressive online actions, resource-based engagement, collective offline mobilisation, and information-driven involvement. The structure is internally consistent and aligns with theoretical expectations about contemporary hybrid forms of participation.

*Although both countries rely on the same set of online and offline participation items, the factor structures show notable differences in dimensionality and coherence. In the Italian case, the parallel analysis and loading patterns converge on a more compact structure, with four clearly interpretable dimensions: expressive online engagement, collective offline action, resource-based behaviours, and information-seeking activities. The factors are well-defined, cross-loadings are limited, and items cluster tightly, indicating that Italian respondents’ participation behaviours form broader and more cohesive domains. By contrast, the Spanish factor structure is more articulated and internally differentiated. Even though the final decision retains five factors, the parallel analysis signals a more complex latent space. The five-factor solution reflects this fragmentation:
– a factor centred on online political communication with an offline spillover,
– a hybrid partisan-mobilization dimension combining online and offline volunteering and contacting,
– a non-partisan social engagement factor based on NGO donations,
– a factor capturing traditional unconventional offline action (strike, demonstration), and
– an offline information and discussion dimension.

This richer and more segmented configuration suggests that in Spain, political participation behaviours are more specialised, with finer-grained distinctions between modes of engagement. Italy shows fewer, broader, and more cohesive participation dimensions. Spain displays a more fragmented, hybrid, and differentiated structure, where several items separate into specialised subdomains.

12 New PCA with factors as supplementary numerical variables

# 10. New PCA with factors as supplementary numerical variables ---------

# Optional check to see if variables exist
names(data_PCA3)

 [1] "Age_cat"          "Vote"             "Sex"              "Left_Right"      
 [5] "Pol_int"          "Inf_disagr"       "Inf_check"        "Inf_confirm"     
 [9] "YouTube"          "WhatsApp"         "Twitter_X"        "Instagram"       
[13] "TikTok"           "Facebook"         "Linkedin"         "Telegram"        
[17] "inf_socially"     "inf_pol"          "sup_soc_causes"   "sup_pol_causes"  
[21] "Info_Online"      "Partisan_mobil"   "Civic_Engagement" "Info_Offline"    
[25] "Protest_Action"  

# vectors of indexes FactoMineR::PCA
quali_sup_idx <- c(
  which(names(data_PCA3) == "Age_cat"),
  which(names(data_PCA3) == "Vote"),
  which(names(data_PCA3) == "Sex"),
  which(names(data_PCA3) == "Left_Right"),
  which(names(data_PCA3) == "Pol_int"),
  which(names(data_PCA3) == "Inf_disagr"),
  which(names(data_PCA3) == "Inf_check"),
  which(names(data_PCA3) == "Inf_confirm")
)

quanti_sup_idx <- c(
  which(names(data_PCA3) == "inf_socially"),
  which(names(data_PCA3) == "inf_pol"),
  which(names(data_PCA3) == "sup_soc_causes"),
  which(names(data_PCA3) == "sup_pol_causes"),
  which(names(data_PCA3) == "Info_Online"),
  which(names(data_PCA3) == "Partisan_mobil"),
  which(names(data_PCA3) == "Civic_Engagement"),
  which(names(data_PCA3) == "Protest_Action"),
  which(names(data_PCA3) == "Info_Offline")
)

# PCA3: active platforms, specific uses + supplementary factors
PCA3_result <- FactoMineR::PCA(
  data_PCA3,
  quali.sup = quali_sup_idx,
  quanti.sup = quanti_sup_idx,
  graph = FALSE
)

Warning in FactoMineR::PCA(data_PCA3, quali.sup = quali_sup_idx, quanti.sup =
quanti_sup_idx, : Missing values are imputed by the mean of the variable: you
should use the imputePCA function of the missMDA package

# Summary of the PCA:
summary(PCA3_result)

Call:
FactoMineR::PCA(X = data_PCA3, quanti.sup = quanti_sup_idx, quali.sup = quali_sup_idx,  
     graph = FALSE) 


Eigenvalues
                       Dim.1   Dim.2   Dim.3   Dim.4   Dim.5   Dim.6   Dim.7
Variance               3.159   1.207   0.768   0.698   0.620   0.572   0.518
% of var.             39.491  15.090   9.595   8.722   7.749   7.150   6.474
Cumulative % of var.  39.491  54.581  64.176  72.898  80.647  87.797  94.271
                       Dim.8
Variance               0.458
% of var.              5.729
Cumulative % of var. 100.000

Individuals (the 10 first)
                            Dist    Dim.1    ctr   cos2    Dim.2    ctr   cos2
1                       |  0.000 |  0.000  0.000  0.222 |  0.000  0.000  0.956
2                       |  2.944 |  1.380  0.046  0.220 | -0.175  0.002  0.004
3                       |  3.457 |  2.479  0.150  0.514 |  1.286  0.105  0.138
4                       |  1.917 | -0.564  0.008  0.087 |  0.551  0.019  0.083
5                       |  3.821 |  0.725  0.013  0.036 | -0.106  0.001  0.001
6                       |  2.831 | -1.359  0.045  0.230 | -0.369  0.009  0.017
7                       |  2.084 | -1.008  0.025  0.234 |  0.477  0.014  0.052
8                       |  3.095 |  2.772  0.187  0.802 | -0.867  0.048  0.078
9                       |  0.537 | -0.138  0.000  0.066 |  0.033  0.000  0.004
10                      |  3.723 |  1.833  0.082  0.242 | -1.668  0.177  0.201
                           Dim.3    ctr   cos2  
1                       |  0.000  0.000  0.002 |
2                       | -1.875  0.352  0.406 |
3                       |  0.837  0.070  0.059 |
4                       | -0.767  0.059  0.160 |
5                       |  2.811  0.791  0.541 |
6                       |  1.932  0.374  0.466 |
7                       |  0.679  0.046  0.106 |
8                       | -0.337  0.011  0.012 |
9                       | -0.098  0.001  0.033 |
10                      |  0.635  0.040  0.029 |

Variables
                           Dim.1    ctr   cos2    Dim.2    ctr   cos2    Dim.3
YouTube                 |  0.601 11.419  0.361 |  0.040  0.136  0.002 |  0.581
WhatsApp                |  0.389  4.791  0.151 |  0.714 42.224  0.510 |  0.252
Twitter_X               |  0.704 15.682  0.495 | -0.406 13.640  0.165 | -0.074
Instagram               |  0.688 14.987  0.473 |  0.309  7.914  0.096 | -0.036
TikTok                  |  0.695 15.276  0.483 | -0.163  2.195  0.026 |  0.093
Facebook                |  0.459  6.671  0.211 |  0.523 22.686  0.274 | -0.545
Linkedin                |  0.693 15.183  0.480 | -0.308  7.847  0.095 | -0.231
Telegram                |  0.711 15.991  0.505 | -0.201  3.360  0.041 | -0.036
                           ctr   cos2  
YouTube                 43.965  0.337 |
WhatsApp                 8.253  0.063 |
Twitter_X                0.714  0.005 |
Instagram                0.166  0.001 |
TikTok                   1.138  0.009 |
Facebook                38.676  0.297 |
Linkedin                 6.923  0.053 |
Telegram                 0.165  0.001 |

Supplementary continuous variables
                           Dim.1   cos2    Dim.2   cos2    Dim.3   cos2  
inf_socially            |  0.330  0.109 |  0.112  0.013 |  0.046  0.002 |
inf_pol                 |  0.348  0.121 |  0.037  0.001 |  0.096  0.009 |
sup_soc_causes          |  0.356  0.127 |  0.066  0.004 |  0.052  0.003 |
sup_pol_causes          |  0.359  0.129 |  0.049  0.002 |  0.055  0.003 |
Info_Online             |  0.264  0.070 | -0.078  0.006 | -0.052  0.003 |
Partisan_mobil          |  0.270  0.073 | -0.005  0.000 | -0.017  0.000 |
Civic_Engagement        |  0.095  0.009 | -0.084  0.007 | -0.009  0.000 |
Protest_Action          |  0.017  0.000 |  0.034  0.001 |  0.059  0.004 |
Info_Offline            |  0.230  0.053 | -0.058  0.003 | -0.073  0.005 |

Supplementary categories (the 10 first)
                             Dist     Dim.1    cos2  v.test     Dim.2    cos2
18_24                   |   1.418 |   1.108   0.611   6.486 |  -0.233   0.027
25_34                   |   0.891 |   0.796   0.798   5.984 |  -0.138   0.024
35_44                   |   0.816 |   0.747   0.839   6.248 |   0.121   0.022
45_54                   |   0.239 |   0.127   0.280   1.255 |   0.182   0.579
55_64                   |   0.334 |  -0.268   0.644  -2.525 |   0.138   0.172
65_+                    |   0.992 |  -0.930   0.879 -11.961 |  -0.149   0.023
+Europa                 |   0.623 |   0.043   0.005   0.155 |  -0.203   0.106
Alleanza Verdi-Sinistra |   0.445 |   0.392   0.776   1.836 |  -0.092   0.042
Azione-Italia Viva      |   0.419 |  -0.075   0.032  -0.416 |  -0.189   0.205
Did not vote            |   0.240 |  -0.090   0.141  -0.716 |  -0.198   0.681
                         v.test     Dim.3    cos2  v.test  
18_24                    -2.203 |   0.592   0.174   7.028 |
25_34                    -1.681 |   0.077   0.007   1.175 |
35_44                     1.634 |  -0.085   0.011  -1.441 |
45_54                     2.922 |  -0.013   0.003  -0.256 |
55_64                     2.111 |  -0.076   0.052  -1.454 |
65_+                     -3.098 |  -0.092   0.009  -2.411 |
+Europa                  -1.169 |   0.110   0.031   0.799 |
Alleanza Verdi-Sinistra  -0.695 |  -0.033   0.006  -0.315 |
Azione-Italia Viva       -1.705 |   0.115   0.076   1.302 |
Did not vote             -2.544 |   0.070   0.086   1.132 |

# Description of the dimensions of the PCA:
PCA3_desc <- FactoMineR::dimdesc(PCA3_result, axes = 1:2, proba = 0.05)

# First dimension
PCA3_desc$Dim.1

Link between the variable and the continuous variables (R-square)
=================================================================================
                 correlation       p.value
Telegram          0.71078325 1.041739e-200
Twitter_X         0.70387174 3.480574e-195
TikTok            0.69469692 4.276470e-188
Linkedin          0.69259115 1.661726e-186
Instagram         0.68810309 3.654681e-183
YouTube           0.60063333 2.210095e-128
Facebook          0.45909834  8.225746e-69
WhatsApp          0.38904027  2.895111e-48
sup_pol_causes    0.35876171  8.389189e-41
sup_soc_causes    0.35589341  3.890533e-40
inf_pol           0.34752598  3.125067e-38
inf_socially      0.32963437  2.387166e-34
Partisan_mobil    0.27029646  3.228874e-23
Info_Online       0.26409260  3.357295e-22
Info_Offline      0.22951271  5.178604e-17
Civic_Engagement  0.09532225  5.758285e-04

Link between the variable and the categorical variable (1-way anova)
=============================================
                    R2      p.value
Age_cat     0.16327212 6.198418e-48
Inf_check   0.06810038 6.392291e-19
Inf_disagr  0.06628292 2.200328e-18
Inf_confirm 0.05654758 1.564022e-15
Left_Right  0.02422511 1.796380e-05
Vote        0.02712709 9.909615e-05

Link between variable and the categories of the categorical variables
================================================================
                                     Estimate      p.value
Age_cat=18_24                       0.8448049 6.314004e-11
Age_cat=35_44                       0.4837189 3.139509e-10
Age_cat=25_34                       0.5330018 1.714502e-09
Inf_check=Inf_check_Very often      0.8497859 1.457769e-08
Inf_disagr=Inf_disagr_Very often    0.8356138 7.353664e-08
Inf_disagr=Inf_disagr_Often         0.4127128 1.234073e-05
Inf_confirm=Inf_confirm_Very often  0.6980323 1.294134e-05
Inf_check=Inf_check_Often           0.3993228 1.547763e-04
Vote=Forza Italia                   0.8322058 3.113801e-04
Inf_confirm=Inf_confirm_Often       0.3868089 1.154004e-03
Left_Right=Ext_right                0.3924648 1.316871e-03
Vote=Movimento 5 Stelle             0.2151232 9.891616e-03
Left_Right=Right                    0.2026920 2.072619e-02
Sex=Non-binary                      1.0454188 4.902130e-02
Left_Right=No_position             -0.3202688 3.374237e-02
Age_cat=55_64                      -0.5312895 1.150529e-02
Left_Right=Left                    -0.3648603 3.203779e-03
Vote=Partito Democratico           -0.4564384 2.558011e-04
Inf_confirm=Inf_confirm_Rarely     -0.4923457 4.336972e-06
Inf_disagr=Inf_disagr_Rarely       -0.4953230 3.847349e-06
Inf_check=Inf_check_Rarely         -0.5263152 5.858813e-07
Inf_disagr=Inf_disagr_Never        -0.7667999 2.008965e-08
Inf_check=Inf_check_Never          -0.7816178 1.656179e-08
Inf_confirm=Inf_confirm_Never      -0.7632705 2.767379e-09
Age_cat=65_+                       -1.1933958 8.630145e-35

# Second dimension
PCA3_desc$Dim.2

Link between the variable and the continuous variables (R-square)
=================================================================================
                 correlation       p.value
WhatsApp          0.71395047 2.701010e-203
Facebook          0.52331660  2.280502e-92
Instagram         0.30908189  3.383618e-30
inf_socially      0.11211703  5.058913e-05
sup_soc_causes    0.06568003  1.782087e-02
Info_Offline     -0.05801716  3.640328e-02
Info_Online      -0.07777316  5.003858e-03
Civic_Engagement -0.08401083  2.424021e-03
TikTok           -0.16278444  3.519636e-09
Telegram         -0.20138465  2.266075e-13
Linkedin         -0.30777231  6.067580e-30
Twitter_X        -0.40577897  9.692650e-53

Link between the variable and the categorical variable (1-way anova)
=============================================
                    R2      p.value
Sex        0.037135620 2.156569e-11
Age_cat    0.020498489 6.094449e-05
Vote       0.022809086 9.041707e-04
Left_Right 0.011570927 1.961848e-02
Inf_check  0.008981723 1.971118e-02

Link between variable and the categories of the categorical variables
================================================================
                                 Estimate      p.value
Sex=Female                      0.3038516 2.406963e-12
Age_cat=45_54                   0.1952296 3.438103e-03
Left_Right=Ext_left             0.2183157 2.284974e-02
Vote=Partito Democratico        0.1603081 3.239774e-02
Age_cat=55_64                   0.1515135 3.474535e-02
Age_cat=18_24                  -0.2195176 2.757226e-02
Vote=Did not vote              -0.1609720 1.091000e-02
Left_Right=No_position         -0.1506509 1.028280e-02
Inf_check=Inf_check_Very often -0.2023231 1.938446e-03
Age_cat=65_+                   -0.1358246 1.922577e-03
Vote=Other party               -0.3415773 1.809898e-03
Sex=Male                       -0.1192441 4.004205e-12

# New biplot with factors as supplementary numerical variables
factoextra::fviz_pca_var(
  PCA3_result,
  col.var = "contrib",
  gradient.cols = c("#00AFBB", "#E7B800", "#FC4E07"),
  repel = TRUE
)

Warning: ggrepel: 3 unlabeled data points (too many overlaps). Consider
increasing max.overlaps

The PCA confirms the underlying structure already observed in the participation and platform analyses: Dim1 (≈ 39%) remains a broad intensity / mainstream use axis, dominated by WhatsApp, Facebook and Instagram, while Dim2 (≈15%) continues to reflect the contrast between entertainment-oriented platforms (TikTok, YouTube) and informational/professional platforms (Twitter/X, LinkedIn, Telegram).

What is new in this model is the projection of the five participation factors—treated as supplementary numerical variables. Their loadings show a coherent alignment with the two PCA axes.

1. Online expressive participation aligns with the visual/entertainment pole: The factor capturing online posting, liking, commenting and profile-based actions is positioned near TikTok and YouTube, indicating that expressive online participation maps onto the audiovisual side of the platform ecosystem. This suggests that individuals who engage in expressive online political acts tend to use visually oriented platforms more intensively.

2. Civic engagement / NGO-related participation sits closer to the informational platforms: The factor representing donations and volunteering for NGOs leans toward LinkedIn and Telegram, aligning with the informational/professional side of Dim2. This indicates that non-partisan civic engagement is more closely associated with structured, informational forms of digital use rather than with high-intensity mainstream platforms.

3. Protest and offline mobilisation align with lower-intensity and non-mainstream use: The factor capturing offline protest (e.g., strikes, demonstrations) is placed closer to the low-intensity region of Dim1, and only modestly associated with any specific platform. This shows that offline mobilisation is weakly linked to digital platform ecosystems, reinforcing the idea that traditional collective action remains partly decoupled from everyday online behaviour.

4. Partisan mobilisation sits between informational and mainstream platform use: The hybrid partisan-mobilisation factor (contacting officials, volunteering for parties) projects nearer to Twitter/X and Telegram, signalling a profile marked by information-driven and politically oriented platform use, but still connected to broader digital engagement.

5. Offline information and discussion aligns with moderate, central digital use: This factor sits near the centre of the map, indicating that face-to-face discussion and reading political materials do not strongly differentiate platform use, and that they characterise individuals with balanced, non-extreme digital behaviours.

The Italian PCA shows that the five participation factors align smoothly with the existing digital-use structure, without reshaping it. Participation modes distribute along the two core axes: (a) Dim1: intensity of digital engagement (WhatsApp–Facebook–Instagram); (b) Dim2: information-oriented vs. entertainment-oriented platform use. Online expressive acts gravitate toward the entertainment pole; partisan mobilisation and civic engagement toward the informational pole; traditional offline protest remains relatively detached. The result is a coherent and interpretable mapping, confirming that in Italy, digital participation styles integrate into platform use in an orderly and predictable way.

The Italian and Spanish results point to two configurations that share a common foundation but diverge in their internal organisation. In both countries, the PCA confirms that patterns of social-media use are structured along two familiar axes: a first component capturing the overall intensity of digital engagement—dominated by WhatsApp, Facebook and Instagram—and a second one separating entertainment-oriented platforms such as YouTube and TikTok from informational or professional ones like Twitter/X, LinkedIn and Telegram. This broad geometry is remarkably stable across the two contexts, suggesting that citizens in Italy and Spain navigate the digital ecosystem through similar repertoires of everyday platform use. What changes, however, is how political participation relates to this digital structure. In Italy, the exploratory factor analysis yields a compact and coherent latent space. Participation behaviours cluster into a few broad dimensions—expressive online actions, collective offline mobilisation, resource-based involvement, and information-oriented engagement. These clusters are internally cohesive and exhibit limited cross-loading, resulting in a relatively integrated picture of how Italians engage politically. When these factors are projected back onto the PCA, only the information-seeking factor aligns clearly with the first digital-use component. This indicates that frequent social-media users tend to be more exposed to political information online, but their intensity of digital engagement does not strongly shape their propensity to mobilise, volunteer, donate, or participate in protest activities. The Spanish case is more articulated. Although the final EFA solution retains five factors, the parallel analysis reveals a more complex underlying structure. The factors themselves are more differentiated and specialised: online political communication forms a distinct dimension with spillover into offline discussion, partisan mobilisation blends online and offline channels, civic engagement emerges as a separate cluster, and traditional protest retains its own space. Compared to Italy, these dimensions are narrower and more specific, which creates a more fragmented but also more finely resolved map of political behaviour. When the Spanish factors are aligned with the PCA, the pattern resembles the Italian one, in that the link between general digital use and political action remains modest; yet some dimensions—notably partisan mobilisation—show a slightly stronger affinity with the informational pole of the digital landscape. This suggests that in Spain, certain types of engagement are more embedded within politically oriented information flows. Analyses depict different modes of integration between digital repertoires and participation behaviours. Italy displays a more cohesive and aggregated structure, where forms of participation cluster into a few broad families and relate only weakly to patterns of platform use. Spain, by contrast, reveals a more differentiated and hybridised landscape, where political engagement is broken into smaller components and shows a somewhat closer alignment with specific aspects of digital information use. The comparison ultimately highlights that while the structure of digital media use is broadly similar across the two countries, the organisation of political participation—and its degree of connection to the digital sphere—varies in systematic and theoretically meaningful ways.

12.1 Canonical Correlation Analysis

# 11. Canonical Correlation Analysis (CCA) -------------------
# relationship between platform uses (3 PCA COMPONENTS) & latent dimensions of political participation (5 EFA factors)

social_vars <- c("YouTube","WhatsApp","Twitter_X","Instagram",
                 "TikTok","Facebook","Linkedin","Telegram")

factor_vars <- c("Info_Online", "Partisan_mobil",
                 "Civic_Engagement", "Protest_Action", "Info_Offline")

# check only complete cases
complete_ids <- complete.cases(
  data_PCA3[, social_vars],
  data_PCA3[, factor_vars]
)

message("✅ Using ", sum(complete_ids),
        " respondents with complete data on platforms and factors.")

✅ Using 995 respondents with complete data on platforms and factors.

pca_data <- data_PCA3[complete_ids, social_vars]
efa_data <- data_PCA3[complete_ids, factor_vars]

# 11.1 PCA platforms (3 components, as per ES script analyses)
pca_res <- psych::principal(
  pca_data,
  nfactors = 3,
  rotate   = "none",
  scores   = TRUE
)

pca_scores <- as.data.frame(pca_res$scores)
colnames(pca_scores) <- paste0("PC", 1:3)   # PC1, PC2, PC3

# Participation factors (already estimated) – standardized
efa_scores <- as.data.frame(scale(efa_data))

# 11.2 Correlation matrix of PCA components e EFA factors
cor_matrix_pca_efa <- cor(
  cbind(pca_scores, efa_scores),
  use = "pairwise.complete.obs"
)

corrplot::corrplot(
  cor_matrix_pca_efa,
  method = "color",
  type   = "upper",
  tl.col = "black",
  tl.cex = 0.8,
  addCoef.col = "black",
  number.cex = 0.6,
  title = "Correlations between PCA Components and EFA Factors",
  mar   = c(0,0,2,0)
)

# 11.3 Canonical Correlation Analysis (CCA)

if (!requireNamespace("CCA", quietly = TRUE)) {
  stop("Il pacchetto 'CCA' non è installato. Esegui: install.packages('CCA')")
}
if (!requireNamespace("CCP", quietly = TRUE)) {
  stop("Il pacchetto 'CCP' non è installato. Esegui: install.packages('CCP')")
}


# Blocks for CCA
X_pca <- pca_scores                           # 3 PCA components 
Y_efa <- efa_scores                           # 5 EFA factors

dim(X_pca)

[1] 995   3

dim(Y_efa)

[1] 995   5

# 11.3.1 CCA execution
cca_result <- CCA::cc(X_pca, Y_efa)

# CCA
cca_result$cor

[1] 0.39421385 0.13937680 0.05488145

# loadings
cca_result$xcoef  # for PCA components

          [,1]       [,2]       [,3]
PC1 -0.9428084  0.3178548 -0.1004023
PC2  0.2275569  0.8338335  0.5029309
PC3 -0.2435779 -0.4513202  0.8584753

cca_result$ycoef  # for EFA factors

                        [,1]       [,2]        [,3]
Info_Online      -0.48913811 -0.2289978 -0.87115583
Partisan_mobil   -0.65513891  0.8300949 -0.24011844
Civic_Engagement  0.07386196 -0.7479003 -0.45467961
Protest_Action    0.33416540  0.2894763  0.03472063
Info_Offline     -0.12160833 -0.2787729  1.47357619

# 11.3.2 Significance test (Wilks’ Lambda)
rho <- cca_result$cor
n   <- nrow(X_pca)
p   <- ncol(X_pca)
q   <- ncol(Y_efa)

cca_test <- CCP::p.asym(rho, n, p, q, tstat = "Wilks")

Wilks' Lambda, using F-approximation (Rao's F):
              stat     approx df1      df2     p.value
1 to 3:  0.8256939 13.0521557  15 2725.074 0.000000000
2 to 3:  0.9776206  2.8111228   8 1976.000 0.004232957
3 to 3:  0.9969880  0.9959472   3  989.000 0.393967752

cca_test

$id
[1] "Wilks"

$stat
[1] 0.8256939 0.9776206 0.9969880

$approx
[1] 13.0521557  2.8111228  0.9959472

$df1
[1] 15  8  3

$df2
[1] 2725.074 1976.000  989.000

$p.value
[1] 0.000000000 0.004232957 0.393967752

# 11.4 Visualization

# Plot of Canonical Variates
U <- as.data.frame(as.matrix(scale(X_pca)) %*% cca_result$xcoef)
V <- as.data.frame(as.matrix(scale(Y_efa)) %*% cca_result$ycoef)

colnames(U) <- paste0("U", 1:ncol(U))
colnames(V) <- paste0("V", 1:ncol(V))

cca_df <- cbind(U, V)

# 11.4.1 Canonical Correlation between PCA (Social Media) and EFA (Political Engagement)
ggplot(cca_df, aes(x = U1, y = V1)) +
  geom_point(alpha = 0.6) +
  geom_smooth(method = "lm", color = "coral") +
  labs(
    title = "Canonical Correlation between PCA (Social Media) and EFA (Political Engagement)",
    x = "First Canonical Variate (PCA block)",
    y = "First Canonical Variate (EFA block)"
  ) +
  theme_minimal()

`geom_smooth()` using formula = 'y ~ x'

# 11.4.2 Canonical Correlation (all functions)
data.frame(
  Canonical_Function = 1:length(cca_result$cor),
  Correlation        = cca_result$cor
) %>%
  ggplot(aes(x = Canonical_Function, y = Correlation)) +
  geom_col(fill = "coral") +
  geom_text(aes(label = round(Correlation, 2)), vjust = -0.3) +
  labs(
    title = "Canonical Correlations between PCA and EFA Latent Sets",
    x = "Canonical Function",
    y = "Correlation"
  ) +
  theme_minimal()

The canonical correlation analysis shows that, in the Italian case, social media use and political participation occupy related but only partially overlapping domains. The first canonical function (0.33) captures a modest but meaningful link between the general intensity of social-media use and the factor measuring online political information-seeking. This indicates that individuals who use multiple platforms more frequently tend to be more exposed to political content online. Beyond this informational dimension, however, the overlap vanishes: the relationships with protest, civic engagement and partisan mobilisation are negligible, suggesting that everyday digital activity does not translate into broader political action.

The Spanish pattern reinforces this interpretation by contrast. Here, canonical correlations are extremely weak from the outset (≈0.07, 0.04, 0.03), implying that digital use and political participation are almost entirely disconnected. Even frequent users do not exhibit systematically higher engagement—neither informational nor participatory. This is noteworthy because the Spanish participation structure is more fragmented and internally differentiated, yet this finer granularity does not produce stronger alignment with digital repertoires. The two behavioural spheres remain parallel but effectively independent.

Taken together, the comparison highlights a subtle but important divergence: Italy shows a slightly more integrated ecology, where the informational dimension of online participation reflects—albeit weakly—the broader patterns of platform use. In Spain, political action appears shaped by dynamics that operate independently of everyday digital behaviour. The informational function of social media plays a more visible role in structuring engagement in Italy, while in Spain the link between digital exposure and participation remains almost entirely absent.

12.2 Hierarchical clustering on PCA results

# 12. Hierarchical clustering on PCA results


# 12.1 HCPC su PCA2_result -------------

hcpc_res <- FactoMineR::HCPC(
  PCA2_result,
  nb.clust = -1,   
  graph    = FALSE
)

# Dataset structure
str(hcpc_res$call$X)

'data.frame':   1301 obs. of  6 variables:
 $ Dim.1: num  -3.76 -3.76 -3.76 -3.76 -3.76 ...
 $ Dim.2: num  -3.04 -3.04 -3.04 -3.04 -3.04 ...
 $ Dim.3: num  -0.869 -0.869 -0.869 -0.869 -0.869 ...
 $ Dim.4: num  0.173 0.173 0.173 0.173 0.173 ...
 $ Dim.5: num  -1.01 -1.01 -1.01 -1.01 -1.01 ...
 $ clust: Factor w/ 3 levels "1","2","3": 1 1 1 1 1 1 1 1 1 1 ...

# Cluster description (quantitative and qualitative variables)
print(hcpc_res$desc.var$quanti)

$`1`
                   v.test Mean in category Overall mean sd in category
inf_pol         -8.619589        0.4444444    0.6610300      0.4969040
sup_pol_causes  -8.638983        0.3046595    0.5334358      0.4602631
Twitter_X       -8.647238        1.5208886    2.3879870      1.2570367
sup_soc_causes  -9.494062        0.3118280    0.5618755      0.4632400
Linkedin        -9.557636        1.4937609    2.3449111      1.0114933
inf_socially   -10.029012        0.5017921    0.7363566      0.4999968
TikTok         -10.676741        1.3567245    2.5052293      1.0014777
Telegram       -11.251364        1.7333518    2.9435939      1.3722745
YouTube        -11.756496        3.0788530    4.1284047      1.6872548
WhatsApp       -18.754809        4.5585063    5.8232558      1.9894849
Instagram      -20.772240        1.6525179    4.0525078      1.3191793
Facebook       -23.084470        2.2688172    4.6645963      1.7748125
               Overall sd       p.value
inf_pol         0.4733596  6.719496e-18
sup_pol_causes  0.4988808  5.671564e-18
Twitter_X       1.8890319  5.276066e-18
sup_soc_causes  0.4961567  2.222026e-21
Linkedin        1.6776605  1.204775e-21
inf_socially    0.4406081  1.136478e-23
TikTok          2.0264803  1.307819e-26
Telegram        2.0263543  2.280375e-29
YouTube         1.6817996  6.539233e-32
WhatsApp        1.2703992  1.768531e-78
Instagram       2.1765789  7.717460e-96
Facebook        1.9551280 6.631616e-118

$`2`
                 v.test Mean in category Overall mean sd in category Overall sd
WhatsApp      11.460297        6.2032356    5.8232558      0.5104873  1.2703992
Facebook      11.325052        5.2424791    4.6645963      1.4653822  1.9551280
Instagram      4.132437        4.2872575    4.0525078      2.0144862  2.1765789
inf_socially   2.131661        0.7608696    0.7363566      0.4265525  0.4406081
YouTube       -3.064133        3.9939097    4.1284047      1.5427273  1.6817996
Telegram      -9.914354        2.4192645    2.9435939      1.7175343  2.0263543
TikTok       -10.546046        1.9474576    2.5052293      1.5944161  2.0264803
Linkedin     -12.174441        1.8118494    2.3449111      1.1343184  1.6776605
Twitter_X    -14.666328        1.6649092    2.3879870      1.1924886  1.8890319
                  p.value
WhatsApp     2.087968e-30
Facebook     9.862073e-30
Instagram    3.589379e-05
inf_socially 3.303475e-02
YouTube      2.183022e-03
Telegram     3.605820e-23
TikTok       5.298021e-26
Linkedin     4.252967e-34
Twitter_X    1.059250e-48

$`3`
                  v.test Mean in category Overall mean sd in category
Twitter_X      24.930347        4.6194453    2.3879870      1.7327380
Linkedin       22.934829        4.1680543    2.3449111      1.7239945
TikTok         22.124288        4.6296126    2.5052293      1.9230664
Telegram       21.942131        5.0503553    2.9435939      1.4616608
Instagram      14.825351        5.5814835    4.0525078      1.1727849
YouTube        14.575831        5.2899301    4.1284047      1.1968981
sup_soc_causes  9.155220        0.7771084    0.5618755      0.4161862
sup_pol_causes  9.034122        0.7469880    0.5334358      0.4347378
Facebook        8.768383        5.4768952    4.6645963      1.3007396
inf_pol         8.666931        0.8554217    0.6610300      0.3516752
inf_socially    7.001561        0.8825301    0.7363566      0.3219794
WhatsApp        4.537405        6.0963855    5.8232558      0.8588933
               Overall sd       p.value
Twitter_X       1.8890319 3.488870e-137
Linkedin        1.6776605 2.088283e-116
TikTok          2.0264803 1.845264e-108
Telegram        2.0263543 1.029628e-106
Instagram       2.1765789  1.004540e-49
YouTube         1.6817996  4.002215e-48
sup_soc_causes  0.4961567  5.424727e-20
sup_pol_causes  0.4988808  1.653240e-19
Facebook        1.9551280  1.812509e-18
inf_pol         0.4733596  4.439233e-18
inf_socially    0.4406081  2.531263e-12
WhatsApp        1.2703992  5.695056e-06

print(hcpc_res$desc.var$quali)

NULL

# Cluster map on PCA
plot(hcpc_res, choice = "map")

# Dendrogram
plot(hcpc_res, choice = "tree")

# 12.2 Cluster assignment--------

clusters <- hcpc_res$data.clust %>%
  dplyr::select(clust) %>%
  dplyr::mutate(ID = dplyr::row_number())

data_clustered <- data_PCA3 %>%
  dplyr::mutate(ID = dplyr::row_number()) %>%
  dplyr::left_join(clusters, by = "ID")


# 12.3 Variables identification ----------------

num_vars <- names(dplyr::select(data_clustered, where(is.numeric))) %>%
  setdiff("ID")

cat_vars <- names(dplyr::select(data_clustered, where(is.factor)))


# 12.4 Summarize of variables by cluster -------------

cat("===== Quantitative Variables by Cluster =====\n")

===== Quantitative Variables by Cluster =====

num_summary <- data_clustered %>%
  group_by(clust) %>%
  summarise(
    across(all_of(num_vars),
           list(mean = mean, sd = sd),
           .names = "{.col}_{.fn}"),
    .groups = "drop"
  )

print(num_summary)

# A tibble: 3 × 35
  clust YouTube_mean YouTube_sd WhatsApp_mean WhatsApp_sd Twitter_X_mean
  <fct>        <dbl>      <dbl>         <dbl>       <dbl>          <dbl>
1 1             3.08       1.69         NA         NA                 NA
2 2            NA         NA            NA         NA                 NA
3 3            NA         NA             6.10       0.860             NA
# ℹ 29 more variables: Twitter_X_sd <dbl>, Instagram_mean <dbl>,
#   Instagram_sd <dbl>, TikTok_mean <dbl>, TikTok_sd <dbl>,
#   Facebook_mean <dbl>, Facebook_sd <dbl>, Linkedin_mean <dbl>,
#   Linkedin_sd <dbl>, Telegram_mean <dbl>, Telegram_sd <dbl>,
#   inf_socially_mean <dbl>, inf_socially_sd <dbl>, inf_pol_mean <dbl>,
#   inf_pol_sd <dbl>, sup_soc_causes_mean <dbl>, sup_soc_causes_sd <dbl>,
#   sup_pol_causes_mean <dbl>, sup_pol_causes_sd <dbl>, …

The hierarchical clustering applied to the Italian PCA results identifies three clearly structured groups of social-media users, each reflecting a coherent digital repertoire. (a) The first cluster gathers low-intensity users, positioned on the negative side of the first PCA dimension, who engage minimally with almost all platforms—especially Instagram, TikTok and Twitter/X. This is the most digitally marginal segment, characterised by sparse and selective use; (b) The second cluster represents a hybrid profile centred on entertainment and interpersonal communication. These users show moderate engagement with YouTube, Facebook, WhatsApp and TikTok, while remaining much less active on informational or professional platforms such as Twitter/X, LinkedIn and Telegram. Their digital behaviour is generalist but not intensive, oriented mainly toward leisure and everyday social interactions; (c) The third cluster consists of high-intensity, multi-platform users. They score clearly above average across all platforms, and particularly on Twitter/X, LinkedIn and Telegram, suggesting a repertoire that combines entertainment, social interaction and strong informational exposure. This is the most digitally active and information-rich group, the one most embedded in cross-platform flows and likely to encounter political content more frequently. The Italian clustering reveals a digital environment organised along a relatively linear structure: from minimal to heavy use, and from entertainment-oriented to information-oriented behaviour. The three clusters are broad, internally coherent and easy to interpret, indicating that Italian users tend to group into stable, well-defined behavioural blocks.

Compared with Spain, this structure appears more compact. The Spanish clustering produces four groups with more specialised and fragmented profiles, including a highly informational cluster, a strong WhatsApp/Facebook cluster, and a heterogeneous mixed cluster that has no direct counterpart in Italy. In other words, while Italy displays a simpler and more integrated segmentation of digital repertoires, Spain shows a more diverse and finely differentiated ecosystem. This echoes the broader pattern observed across PCA, EFA and CCA: Italian digital and participatory behaviours tend to cluster into larger, more coherent families, whereas Spanish patterns are more dispersed and internally varied, suggesting a digital landscape where distinct niches emerge more sharply.

12.3 Anova per each variable

# Numeric summary by cluster 
num_summary <- data_clustered %>%
  dplyr::group_by(clust) %>%
  dplyr::summarise(
    dplyr::across(
      dplyr::all_of(num_vars),
      list(mean = mean, sd = sd),
      .names = "{.col}_{.fn}"
    ),
    .groups = "drop"
  )

print(num_summary)

# A tibble: 3 × 35
  clust YouTube_mean YouTube_sd WhatsApp_mean WhatsApp_sd Twitter_X_mean
  <fct>        <dbl>      <dbl>         <dbl>       <dbl>          <dbl>
1 1             3.08       1.69         NA         NA                 NA
2 2            NA         NA            NA         NA                 NA
3 3            NA         NA             6.10       0.860             NA
# ℹ 29 more variables: Twitter_X_sd <dbl>, Instagram_mean <dbl>,
#   Instagram_sd <dbl>, TikTok_mean <dbl>, TikTok_sd <dbl>,
#   Facebook_mean <dbl>, Facebook_sd <dbl>, Linkedin_mean <dbl>,
#   Linkedin_sd <dbl>, Telegram_mean <dbl>, Telegram_sd <dbl>,
#   inf_socially_mean <dbl>, inf_socially_sd <dbl>, inf_pol_mean <dbl>,
#   inf_pol_sd <dbl>, sup_soc_causes_mean <dbl>, sup_soc_causes_sd <dbl>,
#   sup_pol_causes_mean <dbl>, sup_pol_causes_sd <dbl>, …

# ANOVA per each variable

anova_results <- sapply(num_vars, function(v) {
  if (all(is.na(data_clustered[[v]]))) return(NA)
  fit <- aov(data_clustered[[v]] ~ data_clustered$clust)
  summary(fit)[[1]][["Pr(>F)"]][1]
})

anova_results <- sort(na.omit(anova_results))

cat("\n--- ANOVA p-values (quantitative vars) ---\n")

--- ANOVA p-values (quantitative vars) ---

print(anova_results)

       Twitter_X         Facebook         Linkedin        Instagram 
   1.852602e-185    1.426057e-149    1.827589e-149    2.398475e-139 
          TikTok         Telegram         WhatsApp          YouTube 
   9.243475e-138    1.427796e-136     6.313985e-90     3.765550e-66 
  sup_soc_causes   sup_pol_causes     inf_socially          inf_pol 
    3.301317e-31     5.920672e-28     1.824587e-27     1.075924e-26 
  Partisan_mobil      Info_Online     Info_Offline Civic_Engagement 
    2.391023e-22     3.984051e-16     4.431335e-13     5.259269e-05 
  Protest_Action 
    3.739654e-01 

The ANOVA results show that platform-specific behavioural differences are the strongest source of variation across the PCA-derived clusters. Twitter, Facebook, LinkedIn, TikTok, WhatsApp and Instagram all display highly significant cross-cluster heterogeneity, confirming that clusters primarily reflect distinct digital repertoires. These differences correspond closely to the HCPC internal descriptions: a highly active, multi-platform cluster; a mainstream social-media cluster centred on WhatsApp and Instagram; a low-use cluster; and a younger, TikTok-oriented group. Political and informational variables also contribute to differentiation, though more moderately. Online political information-seeking, social and political causes, and civic engagement show significant but smaller effects, suggesting that political orientations and behaviours partially—but not fully—align with these digital lifestyles. In contrast, protest participation does not discriminate the clusters, indicating that contentious engagement is independent from digital usage patterns. Digital practices are the main dividing line, with political disposition playing a secondary, more diffuse role. This produces a landscape where informational engagement overlaps with some clusters, while ideological and participatory differences remain only partially structured by digital repertoires.

The Spanish case displays a fundamentally different configuration. There too, digital behaviours contribute to cluster separation, but they do not dominate the structure to the same extent. Spanish clusters contain clearer and sharper differences in political interest, exposure to disagreement, checking/confirmation habits, and ideological orientation. Several categorical variables—including left–right self-placement, interest in politics, and fact-checking frequency—show highly significant discriminating power. This indicates that in Spain, clusters encode politicised behavioural ecologies, where digital use combines with political predispositions to define more coherent groups. Political identity and informational orientations are more tightly integrated into the clustering structure.

Comparing the two contexts, Italy shows a more behaviourally modular system: online platform repertoires drive the clustering, and political engagement overlays these repertoires only selectively. Spain, instead, exhibits a more politically saturated clustering structure, where political interest, ideology, and information-verification practices play a central role in defining group boundaries. In other words, Italian clusters are primarily digital-repertoire clusters with partial political colouring, whereas Spanish clusters are political-information clusters in which digital practices are embedded. This contrast underscores how differently digital and political behaviours interlock across the two national contexts: loosely and behaviourally in Italy, tightly and cognitively in Spain.

12.4 Summarize of categorical variables per cluster

# 12.6 Categorical variables by cluster (tabella + chi-quadrato) -------

cat("\n===== Categorical Variables by Cluster =====\n")

===== Categorical Variables by Cluster =====

for (v in cat_vars) {
  cat("\n###", v, "###\n")
  
  tab <- table(data_clustered[[v]], data_clustered$clust)
  print(tab)
  
  test <- suppressWarnings(chisq.test(tab))
  print(test)
}

### Age_cat ###
       
          1   2   3
  0_17    0   0   0
  18_24   8  46  46
  25_34  20  71  66
  35_44  18 101  70
  45_54  40 145  66
  55_64  48 144  39
  65_+  145 183  45

    Pearson's Chi-squared test

data:  tab
X-squared = NaN, df = 12, p-value = NA


### Vote ###
                         
                            1   2   3
  +Europa                  11  16  12
  Alleanza Verdi-Sinistra  11  32  23
  Azione-Italia Viva       24  41  26
  Did not vote             39  95  39
  Forza Italia              7  20  17
  Fratelli d'Italia        37  90  39
  Lega                     11  37  11
  Movimento 5 Stelle       24  93  50
  Other party              21  27  29
  Partito Democratico      64 166  54

    Pearson's Chi-squared test

data:  tab
X-squared = 37.656, df = 18, p-value = 0.0043


### Sex ###
            
               1   2   3
  Male       165 285 174
  Female     113 403 156
  Non-binary   1   2   2

    Pearson's Chi-squared test

data:  tab
X-squared = 29.797, df = 4, p-value = 5.384e-06


### Left_Right ###
             
                1   2   3
  Ext_left     28  91  26
  Left         64 162  55
  Center       67 162  92
  Right        30  87  56
  Ext_right    23  64  40
  No_position  58 112  50

    Pearson's Chi-squared test

data:  tab
X-squared = 23.897, df = 10, p-value = 0.007879


### Pol_int ###
                       
                          1   2   3
  Very interested        46 142  73
  Somewhat interested   141 292 172
  Not very interested    61 198  62
  Not at all interested  31  58  25

    Pearson's Chi-squared test

data:  tab
X-squared = 20.757, df = 6, p-value = 0.002028


### Inf_disagr ###
            
               1   2   3
  Very often  24  61  65
  Often       71 225 129
  sometimes   97 242 106
  Rarely      47  95  21
  Never       40  67  11

    Pearson's Chi-squared test

data:  tab
X-squared = 70.25, df = 8, p-value = 4.383e-12


### Inf_check ###
            
               1   2   3
  Very often  28  69  75
  Often       81 224 133
  sometimes   88 244  95
  Rarely      45  91  20
  Never       37  62   9

    Pearson's Chi-squared test

data:  tab
X-squared = 74.82, df = 8, p-value = 5.359e-13


### Inf_confirm ###
            
               1   2   3
  Very often  32  84  72
  Often       89 256 132
  sometimes   74 216  99
  Rarely      42  73  18
  Never       42  61  11

    Pearson's Chi-squared test

data:  tab
X-squared = 58.171, df = 8, p-value = 1.064e-09


### clust ###
   
      1   2   3
  1 279   0   0
  2   0 690   0
  3   0   0 332

    Pearson's Chi-squared test

data:  tab
X-squared = 2602, df = 4, p-value < 2.2e-16

# 12.7 Visualizations: boxplot per le 3 variabili più significative ----

top_num <- names(head(anova_results, 3))

for (v in top_num) {
  p <- ggplot(data_clustered, aes(x = clust, y = .data[[v]], fill = clust)) +
    geom_boxplot(alpha = 0.7) +
    theme_minimal() +
    labs(
      title = paste("Distribution of", v, "by Cluster"),
      x = "Cluster", y = v
    ) +
    theme(legend.position = "none") +
    scale_fill_brewer(palette = "Dark2")
  
  print(p)
}

Warning: Removed 69 rows containing non-finite outside the scale range
(`stat_boxplot()`).

Warning: Removed 13 rows containing non-finite outside the scale range
(`stat_boxplot()`).

Warning: Removed 63 rows containing non-finite outside the scale range
(`stat_boxplot()`).

# 12.8 Visualizations: stacked bar plots per le variabili categoriche ---

for (v in cat_vars) {
  p <- ggplot(data_clustered, aes(x = clust, fill = .data[[v]])) +
    geom_bar(position = "fill") +
    scale_y_continuous(labels = scales::percent) +
    labs(
      title = paste(v, "distribution across clusters"),
      x = "Cluster", y = "Percentage"
    ) +
    theme_minimal() +
    scale_fill_brewer(palette = "Set2")
  
  print(p)
}

Warning in RColorBrewer::brewer.pal(n, pal): n too large, allowed maximum for palette Set2 is 8
Returning the palette you asked for with that many colors

# 12.9 HCPC Internal Descriptions --------------------------------------

cat("\n===== HCPC Internal Descriptions =====\n")

===== HCPC Internal Descriptions =====

cat("\n--- Quantitative Variables ---\n")

--- Quantitative Variables ---

print(hcpc_res$desc.var$quanti)

$`1`
                   v.test Mean in category Overall mean sd in category
inf_pol         -8.619589        0.4444444    0.6610300      0.4969040
sup_pol_causes  -8.638983        0.3046595    0.5334358      0.4602631
Twitter_X       -8.647238        1.5208886    2.3879870      1.2570367
sup_soc_causes  -9.494062        0.3118280    0.5618755      0.4632400
Linkedin        -9.557636        1.4937609    2.3449111      1.0114933
inf_socially   -10.029012        0.5017921    0.7363566      0.4999968
TikTok         -10.676741        1.3567245    2.5052293      1.0014777
Telegram       -11.251364        1.7333518    2.9435939      1.3722745
YouTube        -11.756496        3.0788530    4.1284047      1.6872548
WhatsApp       -18.754809        4.5585063    5.8232558      1.9894849
Instagram      -20.772240        1.6525179    4.0525078      1.3191793
Facebook       -23.084470        2.2688172    4.6645963      1.7748125
               Overall sd       p.value
inf_pol         0.4733596  6.719496e-18
sup_pol_causes  0.4988808  5.671564e-18
Twitter_X       1.8890319  5.276066e-18
sup_soc_causes  0.4961567  2.222026e-21
Linkedin        1.6776605  1.204775e-21
inf_socially    0.4406081  1.136478e-23
TikTok          2.0264803  1.307819e-26
Telegram        2.0263543  2.280375e-29
YouTube         1.6817996  6.539233e-32
WhatsApp        1.2703992  1.768531e-78
Instagram       2.1765789  7.717460e-96
Facebook        1.9551280 6.631616e-118

$`2`
                 v.test Mean in category Overall mean sd in category Overall sd
WhatsApp      11.460297        6.2032356    5.8232558      0.5104873  1.2703992
Facebook      11.325052        5.2424791    4.6645963      1.4653822  1.9551280
Instagram      4.132437        4.2872575    4.0525078      2.0144862  2.1765789
inf_socially   2.131661        0.7608696    0.7363566      0.4265525  0.4406081
YouTube       -3.064133        3.9939097    4.1284047      1.5427273  1.6817996
Telegram      -9.914354        2.4192645    2.9435939      1.7175343  2.0263543
TikTok       -10.546046        1.9474576    2.5052293      1.5944161  2.0264803
Linkedin     -12.174441        1.8118494    2.3449111      1.1343184  1.6776605
Twitter_X    -14.666328        1.6649092    2.3879870      1.1924886  1.8890319
                  p.value
WhatsApp     2.087968e-30
Facebook     9.862073e-30
Instagram    3.589379e-05
inf_socially 3.303475e-02
YouTube      2.183022e-03
Telegram     3.605820e-23
TikTok       5.298021e-26
Linkedin     4.252967e-34
Twitter_X    1.059250e-48

$`3`
                  v.test Mean in category Overall mean sd in category
Twitter_X      24.930347        4.6194453    2.3879870      1.7327380
Linkedin       22.934829        4.1680543    2.3449111      1.7239945
TikTok         22.124288        4.6296126    2.5052293      1.9230664
Telegram       21.942131        5.0503553    2.9435939      1.4616608
Instagram      14.825351        5.5814835    4.0525078      1.1727849
YouTube        14.575831        5.2899301    4.1284047      1.1968981
sup_soc_causes  9.155220        0.7771084    0.5618755      0.4161862
sup_pol_causes  9.034122        0.7469880    0.5334358      0.4347378
Facebook        8.768383        5.4768952    4.6645963      1.3007396
inf_pol         8.666931        0.8554217    0.6610300      0.3516752
inf_socially    7.001561        0.8825301    0.7363566      0.3219794
WhatsApp        4.537405        6.0963855    5.8232558      0.8588933
               Overall sd       p.value
Twitter_X       1.8890319 3.488870e-137
Linkedin        1.6776605 2.088283e-116
TikTok          2.0264803 1.845264e-108
Telegram        2.0263543 1.029628e-106
Instagram       2.1765789  1.004540e-49
YouTube         1.6817996  4.002215e-48
sup_soc_causes  0.4961567  5.424727e-20
sup_pol_causes  0.4988808  1.653240e-19
Facebook        1.9551280  1.812509e-18
inf_pol         0.4733596  4.439233e-18
inf_socially    0.4406081  2.531263e-12
WhatsApp        1.2703992  5.695056e-06

cat("\n--- Qualitative Variables ---\n")

--- Qualitative Variables ---

print(hcpc_res$desc.var$quali)

NULL

The Italian HCPC yields four clusters that are clearly differentiated along two main axes: general platform intensity (Facebook/Instagram/TikTok/Twitter/LinkedIn/YouTube/WhatsApp) and political engagement (info-seeking, civic support, political support). The ANOVA confirms that almost all digital-use variables significantly discriminate clusters, with Protest_Action as the only marginal case—unsurprising given its low prevalence and weak correlation with digital repertoires:

(a) Cluster 1 aggregates low-intensity users, with uniformly low means across platforms and systematically lower scores on political interest, cross-cutting information behaviours (disagree/check/confirm), and all participation variables. This is the “minimal digital engagement” segment.
(b) Cluster 2 represents moderate users who show relatively higher engagement in informational and civic dimensions, but without strong involvement in expressive or political actions. Here, the pattern suggests a functional rather than activist use of digital media.
(c) Cluster 3 shifts further toward activism: moderate-to-high use of Facebook, Instagram and WhatsApp, accompanied by substantially higher values for political and civic support. This group reflects a hybrid profile: socially embedded platforms combined with selective political uses.
(d) Cluster 4 clearly stands out as the high-intensity digital cluster, with dominant means for every major platform and peak values for political information, social support and political support. The categorical variables reinforce this profile: younger age groups, stronger political interest, and higher frequency of cross-cutting information behaviours cluster here. The structure mirrors an “always-on” digital lifestyle that translates into broader political attentiveness.

What is notable is the alignment between social-media intensity, political interest, and political information behaviours, which increases monotonically across clusters. Italian clusters behave almost like gradations of the same underlying dimension: heavier digital engagement is systematically associated with higher political attentiveness, stronger informational openness (read disagreeing news, check alternative sources, seek confirmation), and broader forms of participation. This confirms the coherence of the Italian digital–political ecology emerging from the earlier PCA and CCA: the digital sphere and the political sphere are not identical but are partially integrated, especially through informational pathways.

When juxtaposed with the Spanish clustering results, the contrast becomes clearer. Both countries identify clusters that reflect gradations of digital intensity, but only in Italy do these clusters map onto political interest and political informational behaviours in a consistent, ordered fashion. In Spain, digital clusters discriminate platform use very sharply but show weaker and more uneven associations with political interest, cross-cutting exposure, and confirmatory/checking behaviours. Spanish clusters are more sociotechnical than political: they classify different platform ecologies, but these do not translate into systematic differences in political engagement. A second divergence concerns cohesion vs fragmentation: (a) Italy shows a compact, hierarchical structure: digital use, informational behaviours and participation scale together; (b) Spain exhibits a more fragmented pattern: clusters are internally differentiated by platform repertoires (e.g., WhatsApp vs. TikTok vs. Twitter users), but political variables vary less cleanly across clusters.

Finally, both countries show that political participation is never primarily driven by platform use alone, but Italy reveals a clearer informational channel linking general social-media intensity to political interest and participation. In Spain, this channel is much weaker: political action appears to be driven by factors more independent of digital habits.

13 Anova Analysis

13.1 Anova platform use (DV) per sociopolitical variables

# 13. ANOVA: platform use x age, ideological...

# Platform use
social_vars <- c("YouTube", "WhatsApp", "Twitter_X", "Instagram",
                 "TikTok", "Facebook", "Linkedin", "Telegram")

# 13.1 Platform and ideology (Left_Right) ---------

# Descriptives
desc_stats <- data_PCA3 %>%
  tidyr::pivot_longer(
    cols      = dplyr::all_of(social_vars),
    names_to  = "Platform",
    values_to = "Frequency"
  ) %>%
  dplyr::group_by(Platform, Left_Right) %>%
  dplyr::summarise(
    n    = dplyr::n(),
    mean = mean(Frequency, na.rm = TRUE),
    sd   = sd(Frequency, na.rm = TRUE),
    se   = sd / sqrt(n),
    .groups = "drop"
  )

knitr::kable(
  desc_stats,
  digits  = 2,
  caption = "Descriptives per platform & ideology"
)

Descriptives per platform & ideology

Platform	Left_Right	n	mean	sd	se
Facebook	Ext_left	145	4.65	1.88	0.16
Facebook	Left	281	4.51	1.94	0.12
Facebook	Center	321	4.71	1.96	0.11
Facebook	Right	173	4.83	1.86	0.14
Facebook	Ext_right	127	5.11	2.00	0.18
Facebook	No_position	220	4.41	2.08	0.14
Facebook	NA	34	4.77	1.99	0.34
Instagram	Ext_left	145	4.06	2.28	0.19
Instagram	Left	281	3.91	2.23	0.13
Instagram	Center	321	4.22	2.11	0.12
Instagram	Right	173	4.20	2.16	0.16
Instagram	Ext_right	127	4.15	2.28	0.20
Instagram	No_position	220	3.75	2.20	0.15
Instagram	NA	34	4.52	2.19	0.38
Linkedin	Ext_left	145	2.02	1.63	0.14
Linkedin	Left	281	2.17	1.55	0.09
Linkedin	Center	321	2.50	1.75	0.10
Linkedin	Right	173	2.59	1.83	0.14
Linkedin	Ext_right	127	2.47	1.91	0.17
Linkedin	No_position	220	2.21	1.65	0.11
Linkedin	NA	34	2.93	1.96	0.34
Telegram	Ext_left	145	2.35	1.82	0.15
Telegram	Left	281	2.74	2.00	0.12
Telegram	Center	321	2.95	2.04	0.11
Telegram	Right	173	3.27	2.07	0.16
Telegram	Ext_right	127	3.43	2.34	0.21
Telegram	No_position	220	2.98	2.12	0.14
Telegram	NA	34	3.52	2.08	0.36
TikTok	Ext_left	145	2.36	2.08	0.17
TikTok	Left	281	2.07	1.80	0.11
TikTok	Center	321	2.41	2.02	0.11
TikTok	Right	173	2.90	2.14	0.16
TikTok	Ext_right	127	3.15	2.32	0.21
TikTok	No_position	220	2.48	2.08	0.14
TikTok	NA	34	3.48	2.32	0.40
Twitter_X	Ext_left	145	2.22	1.83	0.15
Twitter_X	Left	281	2.18	1.79	0.11
Twitter_X	Center	321	2.57	2.02	0.11
Twitter_X	Right	173	2.64	2.06	0.16
Twitter_X	Ext_right	127	2.76	2.21	0.20
Twitter_X	No_position	220	2.09	1.77	0.12
Twitter_X	NA	34	2.41	1.89	0.32
WhatsApp	Ext_left	145	5.93	1.21	0.10
WhatsApp	Left	281	5.89	1.15	0.07
WhatsApp	Center	321	5.87	1.23	0.07
WhatsApp	Right	173	5.81	1.23	0.09
WhatsApp	Ext_right	127	6.05	1.17	0.10
WhatsApp	No_position	220	5.53	1.55	0.10
WhatsApp	NA	34	5.50	1.48	0.25
YouTube	Ext_left	145	4.01	1.54	0.13
YouTube	Left	281	3.91	1.63	0.10
YouTube	Center	321	4.22	1.71	0.10
YouTube	Right	173	4.17	1.65	0.13
YouTube	Ext_right	127	4.44	1.75	0.16
YouTube	No_position	220	4.08	1.80	0.12
YouTube	NA	34	4.60	1.73	0.30

# Welch ANOVA + eta squared per each platform
anova_results <- data_PCA3 %>%
  tidyr::pivot_longer(
    cols      = dplyr::all_of(social_vars),
    names_to  = "Platform",
    values_to = "Frequency"
  ) %>%
  dplyr::group_by(Platform) %>%
  dplyr::summarise(
    Welch_p = oneway.test(Frequency ~ Left_Right,
                          data = dplyr::cur_data(),
                          var.equal = FALSE)$p.value,
    eta_sq  = {
      eta_res <- rstatix::eta_squared(aov(Frequency ~ Left_Right,
                                          data = dplyr::cur_data()))
      if (is.data.frame(eta_res)) eta_res$Eta2[1] else as.numeric(eta_res[1])
    },
    .groups = "drop"
  )

Warning: There was 1 warning in `dplyr::summarise()`.
ℹ In argument: `Welch_p = `$`(...)`.
ℹ In group 1: `Platform = "Facebook"`.
Caused by warning:
! `cur_data()` was deprecated in dplyr 1.1.0.
ℹ Please use `pick()` instead.

knitr::kable(
  anova_results,
  digits  = 3,
  caption = "Welch ANOVA with eta-squared (per ideology)"
)

Welch ANOVA with eta-squared (per ideology)

Platform	Welch_p	eta_sq
Facebook	0.027	0.011
Instagram	0.142	0.007
Linkedin	0.009	0.013
Telegram	0.000	0.020
TikTok	0.000	0.026
Twitter_X	0.003	0.015
WhatsApp	0.017	0.014
YouTube	0.057	0.009

# Games–Howell post-hoc per each platform
gh_results <- data_PCA3 %>%
  tidyr::pivot_longer(
    cols      = dplyr::all_of(social_vars),
    names_to  = "Platform",
    values_to = "Frequency"
  ) %>%
  dplyr::group_by(Platform) %>%
  rstatix::games_howell_test(Frequency ~ Left_Right) %>%
  dplyr::mutate(
    p_sig = dplyr::case_when(
      p.adj < .001 ~ "***",
      p.adj < .01  ~ "**",
      p.adj < .05  ~ "*",
      TRUE        ~ ""
    )
  ) %>%
  dplyr::select(
    Platform, group1, group2,
    mean_difference = estimate, p.adj, p_sig
  )

knitr::kable(
  gh_results,
  digits  = 3,
  caption = "Comparison post-hoc Games–Howell (per ideology)"
)

Comparison post-hoc Games–Howell (per ideology)

Platform	group1	group2	mean_difference	p.adj	p_sig
Facebook	Ext_left	Left	-0.139	0.981
Facebook	Ext_left	Center	0.060	1.000
Facebook	Ext_left	Right	0.187	0.951
Facebook	Ext_left	Ext_right	0.466	0.369
Facebook	Ext_left	No_position	-0.234	0.879
Facebook	Left	Center	0.199	0.813
Facebook	Left	Right	0.325	0.484
Facebook	Left	Ext_right	0.605	0.056
Facebook	Left	No_position	-0.095	0.995
Facebook	Center	Right	0.126	0.981
Facebook	Center	Ext_right	0.406	0.384
Facebook	Center	No_position	-0.294	0.569
Facebook	Right	Ext_right	0.280	0.824
Facebook	Right	No_position	-0.420	0.290
Facebook	Ext_right	No_position	-0.700	0.028	*
Instagram	Ext_left	Left	-0.150	0.988
Instagram	Ext_left	Center	0.163	0.979
Instagram	Ext_left	Right	0.145	0.993
Instagram	Ext_left	Ext_right	0.096	0.999
Instagram	Ext_left	No_position	-0.311	0.799
Instagram	Left	Center	0.313	0.495
Instagram	Left	Right	0.295	0.738
Instagram	Left	Ext_right	0.246	0.915
Instagram	Left	No_position	-0.161	0.967
Instagram	Center	Right	-0.018	1.000
Instagram	Center	Ext_right	-0.067	1.000
Instagram	Center	No_position	-0.474	0.135
Instagram	Right	Ext_right	-0.049	1.000
Instagram	Right	No_position	-0.456	0.328
Instagram	Ext_right	No_position	-0.407	0.597
Linkedin	Ext_left	Left	0.153	0.941
Linkedin	Ext_left	Center	0.480	0.055
Linkedin	Ext_left	Right	0.567	0.051
Linkedin	Ext_left	Ext_right	0.445	0.342
Linkedin	Ext_left	No_position	0.185	0.908
Linkedin	Left	Center	0.327	0.164
Linkedin	Left	Right	0.413	0.154
Linkedin	Left	Ext_right	0.292	0.683
Linkedin	Left	No_position	0.031	1.000
Linkedin	Center	Right	0.086	0.996
Linkedin	Center	Ext_right	-0.035	1.000
Linkedin	Center	No_position	-0.296	0.377
Linkedin	Right	Ext_right	-0.121	0.995
Linkedin	Right	No_position	-0.382	0.297
Linkedin	Ext_right	No_position	-0.261	0.812
Telegram	Ext_left	Left	0.391	0.351
Telegram	Ext_left	Center	0.605	0.023	*
Telegram	Ext_left	Right	0.923	0.001	***
Telegram	Ext_left	Ext_right	1.077	0.001	***
Telegram	Ext_left	No_position	0.626	0.040	*
Telegram	Left	Center	0.214	0.801
Telegram	Left	Right	0.532	0.093
Telegram	Left	Ext_right	0.687	0.067
Telegram	Left	No_position	0.236	0.818
Telegram	Center	Right	0.318	0.599
Telegram	Center	Ext_right	0.473	0.388
Telegram	Center	No_position	0.022	1.000
Telegram	Right	Ext_right	0.155	0.993
Telegram	Right	No_position	-0.296	0.750
Telegram	Ext_right	No_position	-0.451	0.513
TikTok	Ext_left	Left	-0.288	0.731
TikTok	Ext_left	Center	0.048	1.000
TikTok	Ext_left	Right	0.539	0.225
TikTok	Ext_left	Ext_right	0.787	0.048	*
TikTok	Ext_left	No_position	0.119	0.995
TikTok	Left	Center	0.336	0.280
TikTok	Left	Right	0.827	0.001	***
TikTok	Left	Ext_right	1.075	0.000	***
TikTok	Left	No_position	0.407	0.228
TikTok	Center	Right	0.491	0.147
TikTok	Center	Ext_right	0.739	0.026	*
TikTok	Center	No_position	0.071	0.999
TikTok	Right	Ext_right	0.248	0.938
TikTok	Right	No_position	-0.420	0.402
TikTok	Ext_right	No_position	-0.669	0.096
Twitter_X	Ext_left	Left	-0.037	1.000
Twitter_X	Ext_left	Center	0.351	0.445
Twitter_X	Ext_left	Right	0.417	0.421
Twitter_X	Ext_left	Ext_right	0.542	0.279
Twitter_X	Ext_left	No_position	-0.130	0.986
Twitter_X	Left	Center	0.388	0.141
Twitter_X	Left	Right	0.454	0.187
Twitter_X	Left	Ext_right	0.579	0.130
Twitter_X	Left	No_position	-0.093	0.993
Twitter_X	Center	Right	0.067	0.999
Twitter_X	Center	Ext_right	0.192	0.964
Twitter_X	Center	No_position	-0.480	0.052
Twitter_X	Right	Ext_right	0.125	0.997
Twitter_X	Right	No_position	-0.547	0.081
Twitter_X	Ext_right	No_position	-0.672	0.060
WhatsApp	Ext_left	Left	-0.045	0.999
WhatsApp	Ext_left	Center	-0.063	0.996
WhatsApp	Ext_left	Right	-0.123	0.947
WhatsApp	Ext_left	Ext_right	0.117	0.966
WhatsApp	Ext_left	No_position	-0.401	0.067
WhatsApp	Left	Center	-0.018	1.000
WhatsApp	Left	Right	-0.078	0.985
WhatsApp	Left	Ext_right	0.161	0.786
WhatsApp	Left	No_position	-0.356	0.054
WhatsApp	Center	Right	-0.060	0.995
WhatsApp	Center	Ext_right	0.179	0.702
WhatsApp	Center	No_position	-0.338	0.079
WhatsApp	Right	Ext_right	0.239	0.523
WhatsApp	Right	No_position	-0.278	0.358
WhatsApp	Ext_right	No_position	-0.518	0.007	**
YouTube	Ext_left	Left	-0.097	0.991
YouTube	Ext_left	Center	0.217	0.753
YouTube	Ext_left	Right	0.162	0.946
YouTube	Ext_left	Ext_right	0.432	0.278
YouTube	Ext_left	No_position	0.071	0.999
YouTube	Left	Center	0.314	0.200
YouTube	Left	Right	0.258	0.586
YouTube	Left	Ext_right	0.529	0.054
YouTube	Left	No_position	0.167	0.893
YouTube	Center	Right	-0.055	0.999
YouTube	Center	Ext_right	0.215	0.853
YouTube	Center	No_position	-0.146	0.935
YouTube	Right	Ext_right	0.270	0.763
YouTube	Right	No_position	-0.091	0.995
YouTube	Ext_right	No_position	-0.361	0.460

# Grafico: avg 
ggplot(desc_stats,
       aes(x = Left_Right, y = mean, fill = Left_Right)) +
  geom_col(position = "dodge", color = "black") +
  geom_errorbar(aes(ymin = mean - se, ymax = mean + se),
                width = 0.2,
                position = position_dodge(0.9)) +
  facet_wrap(~ Platform, scales = "free_y") +
  labs(
    title = "Platform use frequency and ideology",
    x     = "Ideology Left-Right",
    y     = "Use frequency (avg ±SE)"
  ) +
  scale_fill_viridis_d() +
  theme_minimal() +
  theme(
    legend.position = "none",
    axis.text.x = element_text(angle = 90, hjust = 1)
  )

# 13.2 Platforms & vote in general elections  --------------------

# Descriptives platforms & vote
desc_stats_vote <- data_PCA3 %>%
  tidyr::pivot_longer(
    cols      = dplyr::all_of(social_vars),
    names_to  = "Platform",
    values_to = "Frequency"
  ) %>%
  dplyr::group_by(Platform, Vote) %>%
  dplyr::summarise(
    n    = dplyr::n(),
    mean = mean(Frequency, na.rm = TRUE),
    sd   = sd(Frequency, na.rm = TRUE),
    se   = sd / sqrt(n),
    .groups = "drop"
  )

knitr::kable(
  desc_stats_vote,
  digits  = 2,
  caption = "Descriptives of platforms & vote in general elections"
)

Descriptives of platforms & vote in general elections

Platform	Vote	n	mean	sd	se
Facebook	+Europa	39	3.85	2.01	0.32
Facebook	Alleanza Verdi-Sinistra	66	4.70	1.86	0.23
Facebook	Azione-Italia Viva	91	4.20	2.14	0.22
Facebook	Did not vote	173	4.37	1.94	0.15
Facebook	Forza Italia	44	5.20	1.95	0.29
Facebook	Fratelli d’Italia	166	4.79	1.95	0.15
Facebook	Lega	59	5.16	2.01	0.26
Facebook	Movimento 5 Stelle	167	5.03	1.79	0.14
Facebook	Other party	77	4.38	2.16	0.25
Facebook	Partito Democratico	284	4.72	1.92	0.11
Facebook	NA	135	4.62	1.96	0.17
Instagram	+Europa	39	4.15	2.23	0.36
Instagram	Alleanza Verdi-Sinistra	66	4.52	1.97	0.24
Instagram	Azione-Italia Viva	91	4.20	2.21	0.23
Instagram	Did not vote	173	3.87	2.15	0.16
Instagram	Forza Italia	44	5.02	1.85	0.28
Instagram	Fratelli d’Italia	166	3.93	2.25	0.17
Instagram	Lega	59	4.12	2.23	0.29
Instagram	Movimento 5 Stelle	167	4.28	2.20	0.17
Instagram	Other party	77	3.76	2.19	0.25
Instagram	Partito Democratico	284	3.79	2.27	0.13
Instagram	NA	135	4.17	2.13	0.18
Linkedin	+Europa	39	2.66	1.82	0.29
Linkedin	Alleanza Verdi-Sinistra	66	2.86	1.84	0.23
Linkedin	Azione-Italia Viva	91	2.44	1.70	0.18
Linkedin	Did not vote	173	2.31	1.71	0.13
Linkedin	Forza Italia	44	2.86	1.86	0.28
Linkedin	Fratelli d’Italia	166	1.96	1.49	0.12
Linkedin	Lega	59	2.12	1.58	0.21
Linkedin	Movimento 5 Stelle	167	2.46	1.86	0.14
Linkedin	Other party	77	2.61	1.84	0.21
Linkedin	Partito Democratico	284	2.25	1.67	0.10
Linkedin	NA	135	2.27	1.68	0.14
Telegram	+Europa	39	2.95	1.93	0.31
Telegram	Alleanza Verdi-Sinistra	66	3.23	2.29	0.28
Telegram	Azione-Italia Viva	91	2.63	2.04	0.21
Telegram	Did not vote	173	2.98	2.06	0.16
Telegram	Forza Italia	44	3.60	2.07	0.31
Telegram	Fratelli d’Italia	166	3.12	2.18	0.17
Telegram	Lega	59	2.87	2.08	0.27
Telegram	Movimento 5 Stelle	167	3.03	2.15	0.17
Telegram	Other party	77	3.42	1.93	0.22
Telegram	Partito Democratico	284	2.57	1.95	0.12
Telegram	NA	135	3.01	2.08	0.18
TikTok	+Europa	39	3.13	2.36	0.38
TikTok	Alleanza Verdi-Sinistra	66	2.62	2.21	0.27
TikTok	Azione-Italia Viva	91	2.11	1.79	0.19
TikTok	Did not vote	173	2.58	2.05	0.16
TikTok	Forza Italia	44	3.57	2.50	0.38
TikTok	Fratelli d’Italia	166	2.49	2.00	0.16
TikTok	Lega	59	2.46	2.06	0.27
TikTok	Movimento 5 Stelle	167	2.96	2.26	0.17
TikTok	Other party	77	3.01	2.17	0.25
TikTok	Partito Democratico	284	2.05	1.88	0.11
TikTok	NA	135	2.26	1.88	0.16
Twitter_X	+Europa	39	2.39	1.88	0.30
Twitter_X	Alleanza Verdi-Sinistra	66	2.75	1.98	0.24
Twitter_X	Azione-Italia Viva	91	2.74	1.97	0.21
Twitter_X	Did not vote	173	2.37	1.89	0.14
Twitter_X	Forza Italia	44	3.05	2.33	0.35
Twitter_X	Fratelli d’Italia	166	2.29	1.89	0.15
Twitter_X	Lega	59	2.21	2.04	0.27
Twitter_X	Movimento 5 Stelle	167	2.63	2.07	0.16
Twitter_X	Other party	77	2.93	2.23	0.25
Twitter_X	Partito Democratico	284	2.12	1.82	0.11
Twitter_X	NA	135	1.93	1.60	0.14
WhatsApp	+Europa	39	6.00	1.38	0.22
WhatsApp	Alleanza Verdi-Sinistra	66	5.86	1.12	0.14
WhatsApp	Azione-Italia Viva	91	5.69	1.47	0.15
WhatsApp	Did not vote	173	5.60	1.51	0.11
WhatsApp	Forza Italia	44	5.93	1.30	0.20
WhatsApp	Fratelli d’Italia	166	5.92	1.12	0.09
WhatsApp	Lega	59	5.90	1.28	0.17
WhatsApp	Movimento 5 Stelle	167	5.89	1.16	0.09
WhatsApp	Other party	77	5.71	1.36	0.16
WhatsApp	Partito Democratico	284	5.86	1.19	0.07
WhatsApp	NA	135	5.82	1.30	0.11
YouTube	+Europa	39	3.69	1.36	0.22
YouTube	Alleanza Verdi-Sinistra	66	4.30	1.48	0.18
YouTube	Azione-Italia Viva	91	4.24	1.74	0.18
YouTube	Did not vote	173	4.17	1.72	0.13
YouTube	Forza Italia	44	4.82	1.54	0.23
YouTube	Fratelli d’Italia	166	4.30	1.70	0.13
YouTube	Lega	59	4.09	1.83	0.24
YouTube	Movimento 5 Stelle	167	4.40	1.55	0.12
YouTube	Other party	77	4.25	1.77	0.20
YouTube	Partito Democratico	284	3.75	1.69	0.10
YouTube	NA	135	4.01	1.76	0.15

# Welch ANOVA + eta squared per vote
anova_results_vote <- data_PCA3 %>%
  tidyr::pivot_longer(
    cols      = dplyr::all_of(social_vars),
    names_to  = "Platform",
    values_to = "Frequency"
  ) %>%
  dplyr::group_by(Platform) %>%
  dplyr::summarise(
    Welch_p = oneway.test(Frequency ~ Vote,
                          data = dplyr::cur_data(),
                          var.equal = FALSE)$p.value,
    eta_sq  = {
      eta_res <- rstatix::eta_squared(aov(Frequency ~ Vote,
                                          data = dplyr::cur_data()))
      if (is.data.frame(eta_res)) eta_res$Eta2[1] else as.numeric(eta_res[1])
    },
    .groups = "drop"
  )

knitr::kable(
  anova_results_vote,
  digits  = 3,
  caption = "Welch ANOVA con eta-squared (per party voted)"
)

Welch ANOVA con eta-squared (per party voted)

Platform	Welch_p	eta_sq
Facebook	0.001	0.027
Instagram	0.005	0.018
Linkedin	0.007	0.021
Telegram	0.008	0.020
TikTok	0.000	0.038
Twitter_X	0.014	0.021
WhatsApp	0.560	0.008
YouTube	0.000	0.028

# Games–Howell post-hoc per vote
gh_results_vote <- data_PCA3 %>%
  tidyr::pivot_longer(
    cols      = dplyr::all_of(social_vars),
    names_to  = "Platform",
    values_to = "Frequency"
  ) %>%
  dplyr::group_by(Platform) %>%
  rstatix::games_howell_test(Frequency ~ Vote) %>%
  dplyr::mutate(
    p_sig = dplyr::case_when(
      p.adj < .001 ~ "***",
      p.adj < .01  ~ "**",
      p.adj < .05  ~ "*",
      TRUE        ~ ""
    )
  ) %>%
  dplyr::select(
    Platform, group1, group2,
    mean_difference = estimate, p.adj, p_sig
  )

knitr::kable(
  gh_results_vote,
  digits  = 3,
  caption = "Comparison post-hoc Games–Howell (per party voted)"
)

Comparison post-hoc Games–Howell (per party voted)

Platform	group1	group2	mean_difference	p.adj	p_sig
Facebook	+Europa	Alleanza Verdi-Sinistra	0.851	0.494
Facebook	+Europa	Azione-Italia Viva	0.352	0.996
Facebook	+Europa	Did not vote	0.523	0.896
Facebook	+Europa	Forza Italia	1.358	0.072
Facebook	+Europa	Fratelli d’Italia	0.948	0.213
Facebook	+Europa	Lega	1.312	0.067
Facebook	+Europa	Movimento 5 Stelle	1.184	0.041	*
Facebook	+Europa	Other party	0.535	0.946
Facebook	+Europa	Partito Democratico	0.876	0.261
Facebook	Alleanza Verdi-Sinistra	Azione-Italia Viva	-0.499	0.865
Facebook	Alleanza Verdi-Sinistra	Did not vote	-0.328	0.971
Facebook	Alleanza Verdi-Sinistra	Forza Italia	0.508	0.935
Facebook	Alleanza Verdi-Sinistra	Fratelli d’Italia	0.097	1.000
Facebook	Alleanza Verdi-Sinistra	Lega	0.461	0.948
Facebook	Alleanza Verdi-Sinistra	Movimento 5 Stelle	0.333	0.963
Facebook	Alleanza Verdi-Sinistra	Other party	-0.315	0.995
Facebook	Alleanza Verdi-Sinistra	Partito Democratico	0.025	1.000
Facebook	Azione-Italia Viva	Did not vote	0.171	1.000
Facebook	Azione-Italia Viva	Forza Italia	1.007	0.179
Facebook	Azione-Italia Viva	Fratelli d’Italia	0.596	0.459
Facebook	Azione-Italia Viva	Lega	0.960	0.161
Facebook	Azione-Italia Viva	Movimento 5 Stelle	0.833	0.058
Facebook	Azione-Italia Viva	Other party	0.184	1.000
Facebook	Azione-Italia Viva	Partito Democratico	0.524	0.540
Facebook	Did not vote	Forza Italia	0.835	0.270
Facebook	Did not vote	Fratelli d’Italia	0.425	0.607
Facebook	Did not vote	Lega	0.789	0.240
Facebook	Did not vote	Movimento 5 Stelle	0.661	0.043	*
Facebook	Did not vote	Other party	0.013	1.000
Facebook	Did not vote	Partito Democratico	0.353	0.687
Facebook	Forza Italia	Fratelli d’Italia	-0.411	0.963
Facebook	Forza Italia	Lega	-0.047	1.000
Facebook	Forza Italia	Movimento 5 Stelle	-0.174	1.000
Facebook	Forza Italia	Other party	-0.823	0.503
Facebook	Forza Italia	Partito Democratico	-0.483	0.873
Facebook	Fratelli d’Italia	Lega	0.364	0.972
Facebook	Fratelli d’Italia	Movimento 5 Stelle	0.236	0.979
Facebook	Fratelli d’Italia	Other party	-0.412	0.919
Facebook	Fratelli d’Italia	Partito Democratico	-0.072	1.000
Facebook	Lega	Movimento 5 Stelle	-0.128	1.000
Facebook	Lega	Other party	-0.776	0.506
Facebook	Lega	Partito Democratico	-0.436	0.885
Facebook	Movimento 5 Stelle	Other party	-0.649	0.407
Facebook	Movimento 5 Stelle	Partito Democratico	-0.308	0.786
Facebook	Other party	Partito Democratico	0.340	0.963
Instagram	+Europa	Alleanza Verdi-Sinistra	0.369	0.997
Instagram	+Europa	Azione-Italia Viva	0.044	1.000
Instagram	+Europa	Did not vote	-0.280	0.999
Instagram	+Europa	Forza Italia	0.869	0.661
Instagram	+Europa	Fratelli d’Italia	-0.227	1.000
Instagram	+Europa	Lega	-0.029	1.000
Instagram	+Europa	Movimento 5 Stelle	0.125	1.000
Instagram	+Europa	Other party	-0.397	0.996
Instagram	+Europa	Partito Democratico	-0.368	0.993
Instagram	Alleanza Verdi-Sinistra	Azione-Italia Viva	-0.325	0.994
Instagram	Alleanza Verdi-Sinistra	Did not vote	-0.649	0.465
Instagram	Alleanza Verdi-Sinistra	Forza Italia	0.500	0.941
Instagram	Alleanza Verdi-Sinistra	Fratelli d’Italia	-0.597	0.614
Instagram	Alleanza Verdi-Sinistra	Lega	-0.398	0.990
Instagram	Alleanza Verdi-Sinistra	Movimento 5 Stelle	-0.244	0.998
Instagram	Alleanza Verdi-Sinistra	Other party	-0.766	0.480
Instagram	Alleanza Verdi-Sinistra	Partito Democratico	-0.737	0.213
Instagram	Azione-Italia Viva	Did not vote	-0.324	0.980
Instagram	Azione-Italia Viva	Forza Italia	0.825	0.422
Instagram	Azione-Italia Viva	Fratelli d’Italia	-0.271	0.995
Instagram	Azione-Italia Viva	Lega	-0.073	1.000
Instagram	Azione-Italia Viva	Movimento 5 Stelle	0.081	1.000
Instagram	Azione-Italia Viva	Other party	-0.441	0.956
Instagram	Azione-Italia Viva	Partito Democratico	-0.412	0.876
Instagram	Did not vote	Forza Italia	1.149	0.024	*
Instagram	Did not vote	Fratelli d’Italia	0.052	1.000
Instagram	Did not vote	Lega	0.251	0.999
Instagram	Did not vote	Movimento 5 Stelle	0.405	0.797
Instagram	Did not vote	Other party	-0.117	1.000
Instagram	Did not vote	Partito Democratico	-0.089	1.000
Instagram	Forza Italia	Fratelli d’Italia	-1.097	0.044	*
Instagram	Forza Italia	Lega	-0.898	0.470
Instagram	Forza Italia	Movimento 5 Stelle	-0.744	0.426
Instagram	Forza Italia	Other party	-1.266	0.037	*
Instagram	Forza Italia	Partito Democratico	-1.238	0.007	**
Instagram	Fratelli d’Italia	Lega	0.199	1.000
Instagram	Fratelli d’Italia	Movimento 5 Stelle	0.352	0.916
Instagram	Fratelli d’Italia	Other party	-0.170	1.000
Instagram	Fratelli d’Italia	Partito Democratico	-0.141	1.000
Instagram	Lega	Movimento 5 Stelle	0.154	1.000
Instagram	Lega	Other party	-0.368	0.995
Instagram	Lega	Partito Democratico	-0.339	0.989
Instagram	Movimento 5 Stelle	Other party	-0.522	0.792
Instagram	Movimento 5 Stelle	Partito Democratico	-0.493	0.420
Instagram	Other party	Partito Democratico	0.029	1.000
Linkedin	+Europa	Alleanza Verdi-Sinistra	0.199	1.000
Linkedin	+Europa	Azione-Italia Viva	-0.213	1.000
Linkedin	+Europa	Did not vote	-0.350	0.985
Linkedin	+Europa	Forza Italia	0.199	1.000
Linkedin	+Europa	Fratelli d’Italia	-0.696	0.479
Linkedin	+Europa	Lega	-0.543	0.897
Linkedin	+Europa	Movimento 5 Stelle	-0.195	1.000
Linkedin	+Europa	Other party	-0.047	1.000
Linkedin	+Europa	Partito Democratico	-0.406	0.948
Linkedin	Alleanza Verdi-Sinistra	Azione-Italia Viva	-0.413	0.923
Linkedin	Alleanza Verdi-Sinistra	Did not vote	-0.549	0.570
Linkedin	Alleanza Verdi-Sinistra	Forza Italia	0.000	1.000
Linkedin	Alleanza Verdi-Sinistra	Fratelli d’Italia	-0.895	0.028	*
Linkedin	Alleanza Verdi-Sinistra	Lega	-0.742	0.381
Linkedin	Alleanza Verdi-Sinistra	Movimento 5 Stelle	-0.395	0.913
Linkedin	Alleanza Verdi-Sinistra	Other party	-0.246	0.999
Linkedin	Alleanza Verdi-Sinistra	Partito Democratico	-0.605	0.342
Linkedin	Azione-Italia Viva	Did not vote	-0.136	1.000
Linkedin	Azione-Italia Viva	Forza Italia	0.413	0.967
Linkedin	Azione-Italia Viva	Fratelli d’Italia	-0.483	0.434
Linkedin	Azione-Italia Viva	Lega	-0.329	0.977
Linkedin	Azione-Italia Viva	Movimento 5 Stelle	0.018	1.000
Linkedin	Azione-Italia Viva	Other party	0.167	1.000
Linkedin	Azione-Italia Viva	Partito Democratico	-0.193	0.995
Linkedin	Did not vote	Forza Italia	0.549	0.772
Linkedin	Did not vote	Fratelli d’Italia	-0.346	0.656
Linkedin	Did not vote	Lega	-0.193	0.999
Linkedin	Did not vote	Movimento 5 Stelle	0.154	0.999
Linkedin	Did not vote	Other party	0.303	0.974
Linkedin	Did not vote	Partito Democratico	-0.056	1.000
Linkedin	Forza Italia	Fratelli d’Italia	-0.895	0.134
Linkedin	Forza Italia	Lega	-0.742	0.562
Linkedin	Forza Italia	Movimento 5 Stelle	-0.395	0.965
Linkedin	Forza Italia	Other party	-0.246	1.000
Linkedin	Forza Italia	Partito Democratico	-0.605	0.607
Linkedin	Fratelli d’Italia	Lega	0.154	1.000
Linkedin	Fratelli d’Italia	Movimento 5 Stelle	0.501	0.202
Linkedin	Fratelli d’Italia	Other party	0.649	0.220
Linkedin	Fratelli d’Italia	Partito Democratico	0.290	0.699
Linkedin	Lega	Movimento 5 Stelle	0.347	0.948
Linkedin	Lega	Other party	0.496	0.842
Linkedin	Lega	Partito Democratico	0.136	1.000
Linkedin	Movimento 5 Stelle	Other party	0.149	1.000
Linkedin	Movimento 5 Stelle	Partito Democratico	-0.211	0.975
Linkedin	Other party	Partito Democratico	-0.359	0.889
Telegram	+Europa	Alleanza Verdi-Sinistra	0.287	1.000
Telegram	+Europa	Azione-Italia Viva	-0.318	0.998
Telegram	+Europa	Did not vote	0.028	1.000
Telegram	+Europa	Forza Italia	0.653	0.911
Telegram	+Europa	Fratelli d’Italia	0.176	1.000
Telegram	+Europa	Lega	-0.077	1.000
Telegram	+Europa	Movimento 5 Stelle	0.083	1.000
Telegram	+Europa	Other party	0.472	0.966
Telegram	+Europa	Partito Democratico	-0.382	0.978
Telegram	Alleanza Verdi-Sinistra	Azione-Italia Viva	-0.605	0.801
Telegram	Alleanza Verdi-Sinistra	Did not vote	-0.259	0.999
Telegram	Alleanza Verdi-Sinistra	Forza Italia	0.366	0.998
Telegram	Alleanza Verdi-Sinistra	Fratelli d’Italia	-0.111	1.000
Telegram	Alleanza Verdi-Sinistra	Lega	-0.364	0.996
Telegram	Alleanza Verdi-Sinistra	Movimento 5 Stelle	-0.204	1.000
Telegram	Alleanza Verdi-Sinistra	Other party	0.185	1.000
Telegram	Alleanza Verdi-Sinistra	Partito Democratico	-0.669	0.493
Telegram	Azione-Italia Viva	Did not vote	0.346	0.956
Telegram	Azione-Italia Viva	Forza Italia	0.971	0.300
Telegram	Azione-Italia Viva	Fratelli d’Italia	0.494	0.749
Telegram	Azione-Italia Viva	Lega	0.241	1.000
Telegram	Azione-Italia Viva	Movimento 5 Stelle	0.402	0.903
Telegram	Azione-Italia Viva	Other party	0.790	0.258
Telegram	Azione-Italia Viva	Partito Democratico	-0.064	1.000
Telegram	Did not vote	Forza Italia	0.625	0.786
Telegram	Did not vote	Fratelli d’Italia	0.148	1.000
Telegram	Did not vote	Lega	-0.105	1.000
Telegram	Did not vote	Movimento 5 Stelle	0.056	1.000
Telegram	Did not vote	Other party	0.444	0.844
Telegram	Did not vote	Partito Democratico	-0.410	0.565
Telegram	Forza Italia	Fratelli d’Italia	-0.477	0.954
Telegram	Forza Italia	Lega	-0.730	0.801
Telegram	Forza Italia	Movimento 5 Stelle	-0.569	0.869
Telegram	Forza Italia	Other party	-0.181	1.000
Telegram	Forza Italia	Partito Democratico	-1.034	0.114
Telegram	Fratelli d’Italia	Lega	-0.253	0.999
Telegram	Fratelli d’Italia	Movimento 5 Stelle	-0.093	1.000
Telegram	Fratelli d’Italia	Other party	0.296	0.989
Telegram	Fratelli d’Italia	Partito Democratico	-0.558	0.205
Telegram	Lega	Movimento 5 Stelle	0.160	1.000
Telegram	Lega	Other party	0.549	0.883
Telegram	Lega	Partito Democratico	-0.305	0.992
Telegram	Movimento 5 Stelle	Other party	0.388	0.931
Telegram	Movimento 5 Stelle	Partito Democratico	-0.465	0.416
Telegram	Other party	Partito Democratico	-0.853	0.033	*
TikTok	+Europa	Alleanza Verdi-Sinistra	-0.507	0.986
TikTok	+Europa	Azione-Italia Viva	-1.020	0.351
TikTok	+Europa	Did not vote	-0.556	0.941
TikTok	+Europa	Forza Italia	0.440	0.998
TikTok	+Europa	Fratelli d’Italia	-0.638	0.870
TikTok	+Europa	Lega	-0.667	0.919
TikTok	+Europa	Movimento 5 Stelle	-0.169	1.000
TikTok	+Europa	Other party	-0.117	1.000
TikTok	+Europa	Partito Democratico	-1.084	0.200
TikTok	Alleanza Verdi-Sinistra	Azione-Italia Viva	-0.514	0.874
TikTok	Alleanza Verdi-Sinistra	Did not vote	-0.049	1.000
TikTok	Alleanza Verdi-Sinistra	Forza Italia	0.946	0.605
TikTok	Alleanza Verdi-Sinistra	Fratelli d’Italia	-0.131	1.000
TikTok	Alleanza Verdi-Sinistra	Lega	-0.161	1.000
TikTok	Alleanza Verdi-Sinistra	Movimento 5 Stelle	0.338	0.990
TikTok	Alleanza Verdi-Sinistra	Other party	0.389	0.990
TikTok	Alleanza Verdi-Sinistra	Partito Democratico	-0.578	0.645
TikTok	Azione-Italia Viva	Did not vote	0.465	0.694
TikTok	Azione-Italia Viva	Forza Italia	1.460	0.036	*
TikTok	Azione-Italia Viva	Fratelli d’Italia	0.383	0.870
TikTok	Azione-Italia Viva	Lega	0.353	0.988
TikTok	Azione-Italia Viva	Movimento 5 Stelle	0.852	0.038	*
TikTok	Azione-Italia Viva	Other party	0.903	0.139
TikTok	Azione-Italia Viva	Partito Democratico	-0.064	1.000
TikTok	Did not vote	Forza Italia	0.995	0.359
TikTok	Did not vote	Fratelli d’Italia	-0.082	1.000
TikTok	Did not vote	Lega	-0.112	1.000
TikTok	Did not vote	Movimento 5 Stelle	0.387	0.847
TikTok	Did not vote	Other party	0.438	0.914
TikTok	Did not vote	Partito Democratico	-0.529	0.197
TikTok	Forza Italia	Fratelli d’Italia	-1.078	0.250
TikTok	Forza Italia	Lega	-1.107	0.379
TikTok	Forza Italia	Movimento 5 Stelle	-0.609	0.912
TikTok	Forza Italia	Other party	-0.557	0.970
TikTok	Forza Italia	Partito Democratico	-1.524	0.014	*
TikTok	Fratelli d’Italia	Lega	-0.029	1.000
TikTok	Fratelli d’Italia	Movimento 5 Stelle	0.469	0.623
TikTok	Fratelli d’Italia	Other party	0.520	0.784
TikTok	Fratelli d’Italia	Partito Democratico	-0.446	0.403
TikTok	Lega	Movimento 5 Stelle	0.498	0.881
TikTok	Lega	Other party	0.550	0.906
TikTok	Lega	Partito Democratico	-0.417	0.924
TikTok	Movimento 5 Stelle	Other party	0.051	1.000
TikTok	Movimento 5 Stelle	Partito Democratico	-0.915	0.001	***
TikTok	Other party	Partito Democratico	-0.967	0.029	*
Twitter_X	+Europa	Alleanza Verdi-Sinistra	0.351	0.996
Twitter_X	+Europa	Azione-Italia Viva	0.350	0.994
Twitter_X	+Europa	Did not vote	-0.024	1.000
Twitter_X	+Europa	Forza Italia	0.657	0.934
Twitter_X	+Europa	Fratelli d’Italia	-0.104	1.000
Twitter_X	+Europa	Lega	-0.187	1.000
Twitter_X	+Europa	Movimento 5 Stelle	0.237	1.000
Twitter_X	+Europa	Other party	0.538	0.941
Twitter_X	+Europa	Partito Democratico	-0.278	0.997
Twitter_X	Alleanza Verdi-Sinistra	Azione-Italia Viva	-0.002	1.000
Twitter_X	Alleanza Verdi-Sinistra	Did not vote	-0.375	0.955
Twitter_X	Alleanza Verdi-Sinistra	Forza Italia	0.305	1.000
Twitter_X	Alleanza Verdi-Sinistra	Fratelli d’Italia	-0.456	0.864
Twitter_X	Alleanza Verdi-Sinistra	Lega	-0.538	0.914
Twitter_X	Alleanza Verdi-Sinistra	Movimento 5 Stelle	-0.115	1.000
Twitter_X	Alleanza Verdi-Sinistra	Other party	0.186	1.000
Twitter_X	Alleanza Verdi-Sinistra	Partito Democratico	-0.629	0.396
Twitter_X	Azione-Italia Viva	Did not vote	-0.373	0.907
Twitter_X	Azione-Italia Viva	Forza Italia	0.307	0.999
Twitter_X	Azione-Italia Viva	Fratelli d’Italia	-0.454	0.755
Twitter_X	Azione-Italia Viva	Lega	-0.537	0.873
Twitter_X	Azione-Italia Viva	Movimento 5 Stelle	-0.113	1.000
Twitter_X	Azione-Italia Viva	Other party	0.188	1.000
Twitter_X	Azione-Italia Viva	Partito Democratico	-0.628	0.195
Twitter_X	Did not vote	Forza Italia	0.680	0.794
Twitter_X	Did not vote	Fratelli d’Italia	-0.081	1.000
Twitter_X	Did not vote	Lega	-0.164	1.000
Twitter_X	Did not vote	Movimento 5 Stelle	0.260	0.976
Twitter_X	Did not vote	Other party	0.561	0.686
Twitter_X	Did not vote	Partito Democratico	-0.254	0.936
Twitter_X	Forza Italia	Fratelli d’Italia	-0.761	0.675
Twitter_X	Forza Italia	Lega	-0.844	0.728
Twitter_X	Forza Italia	Movimento 5 Stelle	-0.420	0.989
Twitter_X	Forza Italia	Other party	-0.119	1.000
Twitter_X	Forza Italia	Partito Democratico	-0.934	0.347
Twitter_X	Fratelli d’Italia	Lega	-0.083	1.000
Twitter_X	Fratelli d’Italia	Movimento 5 Stelle	0.341	0.879
Twitter_X	Fratelli d’Italia	Other party	0.642	0.504
Twitter_X	Fratelli d’Italia	Partito Democratico	-0.174	0.995
Twitter_X	Lega	Movimento 5 Stelle	0.424	0.950
Twitter_X	Lega	Other party	0.725	0.670
Twitter_X	Lega	Partito Democratico	-0.091	1.000
Twitter_X	Movimento 5 Stelle	Other party	0.301	0.993
Twitter_X	Movimento 5 Stelle	Partito Democratico	-0.514	0.216
Twitter_X	Other party	Partito Democratico	-0.816	0.120
WhatsApp	+Europa	Alleanza Verdi-Sinistra	-0.138	1.000
WhatsApp	+Europa	Azione-Italia Viva	-0.311	0.977
WhatsApp	+Europa	Did not vote	-0.395	0.850
WhatsApp	+Europa	Forza Italia	-0.068	1.000
WhatsApp	+Europa	Fratelli d’Italia	-0.078	1.000
WhatsApp	+Europa	Lega	-0.103	1.000
WhatsApp	+Europa	Movimento 5 Stelle	-0.108	1.000
WhatsApp	+Europa	Other party	-0.289	0.986
WhatsApp	+Europa	Partito Democratico	-0.137	1.000
WhatsApp	Alleanza Verdi-Sinistra	Azione-Italia Viva	-0.173	0.998
WhatsApp	Alleanza Verdi-Sinistra	Did not vote	-0.257	0.920
WhatsApp	Alleanza Verdi-Sinistra	Forza Italia	0.070	1.000
WhatsApp	Alleanza Verdi-Sinistra	Fratelli d’Italia	0.060	1.000
WhatsApp	Alleanza Verdi-Sinistra	Lega	0.035	1.000
WhatsApp	Alleanza Verdi-Sinistra	Movimento 5 Stelle	0.031	1.000
WhatsApp	Alleanza Verdi-Sinistra	Other party	-0.151	0.999
WhatsApp	Alleanza Verdi-Sinistra	Partito Democratico	0.001	1.000
WhatsApp	Azione-Italia Viva	Did not vote	-0.084	1.000
WhatsApp	Azione-Italia Viva	Forza Italia	0.243	0.993
WhatsApp	Azione-Italia Viva	Fratelli d’Italia	0.233	0.950
WhatsApp	Azione-Italia Viva	Lega	0.208	0.996
WhatsApp	Azione-Italia Viva	Movimento 5 Stelle	0.203	0.980
WhatsApp	Azione-Italia Viva	Other party	0.022	1.000
WhatsApp	Azione-Italia Viva	Partito Democratico	0.174	0.991
WhatsApp	Did not vote	Forza Italia	0.327	0.913
WhatsApp	Did not vote	Fratelli d’Italia	0.317	0.474
WhatsApp	Did not vote	Lega	0.292	0.917
WhatsApp	Did not vote	Movimento 5 Stelle	0.287	0.632
WhatsApp	Did not vote	Other party	0.106	1.000
WhatsApp	Did not vote	Partito Democratico	0.258	0.677
WhatsApp	Forza Italia	Fratelli d’Italia	-0.010	1.000
WhatsApp	Forza Italia	Lega	-0.035	1.000
WhatsApp	Forza Italia	Movimento 5 Stelle	-0.040	1.000
WhatsApp	Forza Italia	Other party	-0.221	0.997
WhatsApp	Forza Italia	Partito Democratico	-0.069	1.000
WhatsApp	Fratelli d’Italia	Lega	-0.025	1.000
WhatsApp	Fratelli d’Italia	Movimento 5 Stelle	-0.029	1.000
WhatsApp	Fratelli d’Italia	Other party	-0.211	0.974
WhatsApp	Fratelli d’Italia	Partito Democratico	-0.059	1.000
WhatsApp	Lega	Movimento 5 Stelle	-0.004	1.000
WhatsApp	Lega	Other party	-0.186	0.998
WhatsApp	Lega	Partito Democratico	-0.034	1.000
WhatsApp	Movimento 5 Stelle	Other party	-0.182	0.991
WhatsApp	Movimento 5 Stelle	Partito Democratico	-0.030	1.000
WhatsApp	Other party	Partito Democratico	0.152	0.997
YouTube	+Europa	Alleanza Verdi-Sinistra	0.611	0.497
YouTube	+Europa	Azione-Italia Viva	0.552	0.644
YouTube	+Europa	Did not vote	0.474	0.696
YouTube	+Europa	Forza Italia	1.126	0.023	*
YouTube	+Europa	Fratelli d’Italia	0.608	0.353
YouTube	+Europa	Lega	0.395	0.969
YouTube	+Europa	Movimento 5 Stelle	0.711	0.139
YouTube	+Europa	Other party	0.558	0.688
YouTube	+Europa	Partito Democratico	0.059	1.000
YouTube	Alleanza Verdi-Sinistra	Azione-Italia Viva	-0.059	1.000
YouTube	Alleanza Verdi-Sinistra	Did not vote	-0.136	1.000
YouTube	Alleanza Verdi-Sinistra	Forza Italia	0.515	0.768
YouTube	Alleanza Verdi-Sinistra	Fratelli d’Italia	-0.002	1.000
YouTube	Alleanza Verdi-Sinistra	Lega	-0.215	0.999
YouTube	Alleanza Verdi-Sinistra	Movimento 5 Stelle	0.101	1.000
YouTube	Alleanza Verdi-Sinistra	Other party	-0.053	1.000
YouTube	Alleanza Verdi-Sinistra	Partito Democratico	-0.551	0.207
YouTube	Azione-Italia Viva	Did not vote	-0.078	1.000
YouTube	Azione-Italia Viva	Forza Italia	0.574	0.646
YouTube	Azione-Italia Viva	Fratelli d’Italia	0.056	1.000
YouTube	Azione-Italia Viva	Lega	-0.157	1.000
YouTube	Azione-Italia Viva	Movimento 5 Stelle	0.159	0.999
YouTube	Azione-Italia Viva	Other party	0.006	1.000
YouTube	Azione-Italia Viva	Partito Democratico	-0.493	0.363
YouTube	Did not vote	Forza Italia	0.652	0.324
YouTube	Did not vote	Fratelli d’Italia	0.134	0.999
YouTube	Did not vote	Lega	-0.079	1.000
YouTube	Did not vote	Movimento 5 Stelle	0.237	0.948
YouTube	Did not vote	Other party	0.083	1.000
YouTube	Did not vote	Partito Democratico	-0.415	0.278
YouTube	Forza Italia	Fratelli d’Italia	-0.518	0.650
YouTube	Forza Italia	Lega	-0.730	0.484
YouTube	Forza Italia	Movimento 5 Stelle	-0.415	0.852
YouTube	Forza Italia	Other party	-0.568	0.710
YouTube	Forza Italia	Partito Democratico	-1.066	0.003	**
YouTube	Fratelli d’Italia	Lega	-0.213	0.999
YouTube	Fratelli d’Italia	Movimento 5 Stelle	0.103	1.000
YouTube	Fratelli d’Italia	Other party	-0.051	1.000
YouTube	Fratelli d’Italia	Partito Democratico	-0.549	0.037	*
YouTube	Lega	Movimento 5 Stelle	0.316	0.976
YouTube	Lega	Other party	0.162	1.000
YouTube	Lega	Partito Democratico	-0.336	0.956
YouTube	Movimento 5 Stelle	Other party	-0.154	1.000
YouTube	Movimento 5 Stelle	Partito Democratico	-0.652	0.002	**
YouTube	Other party	Partito Democratico	-0.498	0.467

# Grapth per party voted
ggplot(desc_stats_vote,
       aes(x = Vote, y = mean, fill = Vote)) +
  geom_col(position = "dodge", color = "black") +
  geom_errorbar(aes(ymin = mean - se, ymax = mean + se),
                width = 0.2,
                position = position_dodge(0.9)) +
  facet_wrap(~ Platform, scales = "free_y") +
  labs(
    title = "Platform use frequency and party voted in general elections",
    x     = "Party voted",
    y     = "Platform use frequency (±SE)"
  ) +
  scale_fill_viridis_d() +
  theme_minimal() +
  theme(
    legend.position = "none",
    axis.text.x = element_text(angle = 90, hjust = 1)
  )

# 13.3 Platform and age (Age_cat) -------------------

# Descriptives of platforms and age categories
desc_stats_age <- data_PCA3 %>%
  tidyr::pivot_longer(
    cols      = dplyr::all_of(social_vars),
    names_to  = "Platform",
    values_to = "Frequency"
  ) %>%
  dplyr::group_by(Platform, Age_cat) %>%
  dplyr::summarise(
    n    = dplyr::n(),
    mean = mean(Frequency, na.rm = TRUE),
    sd   = sd(Frequency, na.rm = TRUE),
    se   = sd / sqrt(n),
    .groups = "drop"
  )

knitr::kable(
  desc_stats_age,
  digits  = 2,
  caption = "Descriptives for platforms and age categories"
)

Descriptives for platforms and age categories

Platform	Age_cat	n	mean	sd	se
Facebook	18_24	100	3.92	2.05	0.21
Facebook	25_34	157	4.53	1.79	0.14
Facebook	35_44	189	5.18	1.65	0.12
Facebook	45_54	251	4.93	1.83	0.12
Facebook	55_64	231	4.82	1.91	0.13
Facebook	65_+	373	4.38	2.18	0.11
Instagram	18_24	100	5.58	1.53	0.15
Instagram	25_34	157	5.23	1.64	0.13
Instagram	35_44	189	5.02	1.81	0.13
Instagram	45_54	251	4.19	2.13	0.13
Instagram	55_64	231	3.74	2.06	0.14
Instagram	65_+	373	2.71	2.07	0.11
Linkedin	18_24	100	2.59	1.97	0.20
Linkedin	25_34	157	3.03	1.84	0.15
Linkedin	35_44	189	3.08	1.93	0.14
Linkedin	45_54	251	2.37	1.60	0.10
Linkedin	55_64	231	2.12	1.51	0.10
Linkedin	65_+	373	1.73	1.40	0.07
Telegram	18_24	100	3.83	2.06	0.21
Telegram	25_34	157	3.71	2.00	0.16
Telegram	35_44	189	3.67	2.11	0.15
Telegram	45_54	251	3.12	2.12	0.13
Telegram	55_64	231	2.75	1.93	0.13
Telegram	65_+	373	1.95	1.71	0.09
TikTok	18_24	100	4.23	2.34	0.23
TikTok	25_34	157	3.12	2.24	0.18
TikTok	35_44	189	2.66	2.08	0.15
TikTok	45_54	251	2.51	2.06	0.13
TikTok	55_64	231	2.11	1.78	0.12
TikTok	65_+	373	1.91	1.73	0.09
Twitter_X	18_24	100	3.18	2.14	0.21
Twitter_X	25_34	157	2.84	2.13	0.17
Twitter_X	35_44	189	2.71	2.09	0.15
Twitter_X	45_54	251	2.27	1.81	0.11
Twitter_X	55_64	231	2.17	1.75	0.12
Twitter_X	65_+	373	2.01	1.80	0.09
WhatsApp	18_24	100	6.07	0.98	0.10
WhatsApp	25_34	157	5.91	1.08	0.09
WhatsApp	35_44	189	6.05	1.15	0.08
WhatsApp	45_54	251	6.01	0.99	0.06
WhatsApp	55_64	231	5.87	1.08	0.07
WhatsApp	65_+	373	5.45	1.65	0.09
YouTube	18_24	100	4.97	1.53	0.15
YouTube	25_34	157	4.50	1.52	0.12
YouTube	35_44	189	4.58	1.65	0.12
YouTube	45_54	251	4.21	1.57	0.10
YouTube	55_64	231	3.95	1.60	0.11
YouTube	65_+	373	3.57	1.76	0.09

# Welch ANOVA + eta squared per età
anova_results_age <- data_PCA3 %>%
  tidyr::pivot_longer(
    cols      = dplyr::all_of(social_vars),
    names_to  = "Platform",
    values_to = "Frequency"
  ) %>%
  dplyr::group_by(Platform) %>%
  dplyr::summarise(
    Welch_p = oneway.test(Frequency ~ Age_cat,
                          data = dplyr::cur_data(),
                          var.equal = FALSE)$p.value,
    eta_sq  = {
      eta_res <- rstatix::eta_squared(aov(Frequency ~ Age_cat,
                                          data = dplyr::cur_data()))
      if (is.data.frame(eta_res)) eta_res$Eta2[1] else as.numeric(eta_res[1])
    },
    .groups = "drop"
  )

knitr::kable(
  anova_results_age,
  digits  = 3,
  caption = "Welch ANOVA con eta-squared (per age class)"
)

Welch ANOVA con eta-squared (per age class)

Platform	Welch_p	eta_sq
Facebook	0	0.032
Instagram	0	0.210
Linkedin	0	0.086
Telegram	0	0.115
TikTok	0	0.097
Twitter_X	0	0.037
WhatsApp	0	0.037
YouTube	0	0.068

# Games–Howell post-hoc per age
gh_results_age <- data_PCA3 %>%
  tidyr::pivot_longer(
    cols      = dplyr::all_of(social_vars),
    names_to  = "Platform",
    values_to = "Frequency"
  ) %>%
  dplyr::group_by(Platform) %>%
  rstatix::games_howell_test(Frequency ~ Age_cat) %>%
  dplyr::mutate(
    p_sig = dplyr::case_when(
      p.adj < .001 ~ "***",
      p.adj < .01  ~ "**",
      p.adj < .05  ~ "*",
      TRUE        ~ ""
    )
  ) %>%
  dplyr::select(
    Platform, group1, group2,
    mean_difference = estimate, p.adj, p_sig
  )

knitr::kable(
  gh_results_age,
  digits  = 3,
  caption = "Comparison post-hoc Games–Howell (for age class)"
)

Comparison post-hoc Games–Howell (for age class)

Platform	group1	group2	mean_difference	p.adj	p_sig
Facebook	18_24	25_34	0.614	0.151
Facebook	18_24	35_44	1.259	0.000	***
Facebook	18_24	45_54	1.013	0.000	***
Facebook	18_24	55_64	0.900	0.004	**
Facebook	18_24	65_+	0.465	0.367
Facebook	25_34	35_44	0.645	0.009	**
Facebook	25_34	45_54	0.400	0.256
Facebook	25_34	55_64	0.286	0.665
Facebook	25_34	65_+	-0.149	0.965
Facebook	35_44	45_54	-0.246	0.685
Facebook	35_44	55_64	-0.359	0.310
Facebook	35_44	65_+	-0.794	0.000	***
Facebook	45_54	55_64	-0.114	0.986
Facebook	45_54	65_+	-0.549	0.010	*
Facebook	55_64	65_+	-0.435	0.106
Instagram	18_24	25_34	-0.351	0.507
Instagram	18_24	35_44	-0.558	0.069
Instagram	18_24	45_54	-1.386	0.000	***
Instagram	18_24	55_64	-1.836	0.000	***
Instagram	18_24	65_+	-2.866	0.000	***
Instagram	25_34	35_44	-0.208	0.876
Instagram	25_34	45_54	-1.035	0.000	***
Instagram	25_34	55_64	-1.485	0.000	***
Instagram	25_34	65_+	-2.515	0.000	***
Instagram	35_44	45_54	-0.827	0.000	***
Instagram	35_44	55_64	-1.277	0.000	***
Instagram	35_44	65_+	-2.308	0.000	***
Instagram	45_54	55_64	-0.450	0.181
Instagram	45_54	65_+	-1.480	0.000	***
Instagram	55_64	65_+	-1.031	0.000	***
Linkedin	18_24	25_34	0.440	0.480
Linkedin	18_24	35_44	0.494	0.340
Linkedin	18_24	45_54	-0.220	0.922
Linkedin	18_24	55_64	-0.468	0.291
Linkedin	18_24	65_+	-0.858	0.001	**
Linkedin	25_34	35_44	0.054	1.000
Linkedin	25_34	45_54	-0.660	0.004	**
Linkedin	25_34	55_64	-0.908	0.000	***
Linkedin	25_34	65_+	-1.298	0.000	***
Linkedin	35_44	45_54	-0.714	0.001	**
Linkedin	35_44	55_64	-0.962	0.000	***
Linkedin	35_44	65_+	-1.352	0.000	***
Linkedin	45_54	55_64	-0.249	0.519
Linkedin	45_54	65_+	-0.639	0.000	***
Linkedin	55_64	65_+	-0.390	0.027	*
Telegram	18_24	25_34	-0.123	0.997
Telegram	18_24	35_44	-0.160	0.990
Telegram	18_24	45_54	-0.707	0.053
Telegram	18_24	55_64	-1.080	0.000	***
Telegram	18_24	65_+	-1.875	0.000	***
Telegram	25_34	35_44	-0.037	1.000
Telegram	25_34	45_54	-0.584	0.064
Telegram	25_34	55_64	-0.957	0.000	***
Telegram	25_34	65_+	-1.752	0.000	***
Telegram	35_44	45_54	-0.548	0.093
Telegram	35_44	55_64	-0.921	0.000	***
Telegram	35_44	65_+	-1.716	0.000	***
Telegram	45_54	55_64	-0.373	0.349
Telegram	45_54	65_+	-1.168	0.000	***
Telegram	55_64	65_+	-0.795	0.000	***
TikTok	18_24	25_34	-1.110	0.003	**
TikTok	18_24	35_44	-1.573	0.000	***
TikTok	18_24	45_54	-1.718	0.000	***
TikTok	18_24	55_64	-2.125	0.000	***
TikTok	18_24	65_+	-2.324	0.000	***
TikTok	25_34	35_44	-0.463	0.373
TikTok	25_34	45_54	-0.608	0.074
TikTok	25_34	55_64	-1.015	0.000	***
TikTok	25_34	65_+	-1.215	0.000	***
TikTok	35_44	45_54	-0.145	0.981
TikTok	35_44	55_64	-0.552	0.057
TikTok	35_44	65_+	-0.751	0.001	***
TikTok	45_54	55_64	-0.407	0.204
TikTok	45_54	65_+	-0.607	0.003	**
TikTok	55_64	65_+	-0.200	0.773
Twitter_X	18_24	25_34	-0.342	0.819
Twitter_X	18_24	35_44	-0.474	0.482
Twitter_X	18_24	45_54	-0.909	0.004	**
Twitter_X	18_24	55_64	-1.015	0.001	***
Twitter_X	18_24	65_+	-1.169	0.000	***
Twitter_X	25_34	35_44	-0.133	0.993
Twitter_X	25_34	45_54	-0.568	0.075
Twitter_X	25_34	55_64	-0.674	0.018	*
Twitter_X	25_34	65_+	-0.828	0.001	***
Twitter_X	35_44	45_54	-0.435	0.231
Twitter_X	35_44	55_64	-0.541	0.066
Twitter_X	35_44	65_+	-0.695	0.003	**
Twitter_X	45_54	55_64	-0.106	0.988
Twitter_X	45_54	65_+	-0.260	0.529
Twitter_X	55_64	65_+	-0.154	0.915
WhatsApp	18_24	25_34	-0.159	0.825
WhatsApp	18_24	35_44	-0.021	1.000
WhatsApp	18_24	45_54	-0.058	0.996
WhatsApp	18_24	55_64	-0.205	0.536
WhatsApp	18_24	65_+	-0.617	0.000	***
WhatsApp	25_34	35_44	0.138	0.863
WhatsApp	25_34	45_54	0.101	0.932
WhatsApp	25_34	55_64	-0.046	0.999
WhatsApp	25_34	65_+	-0.458	0.002	**
WhatsApp	35_44	45_54	-0.037	0.999
WhatsApp	35_44	55_64	-0.184	0.559
WhatsApp	35_44	65_+	-0.596	0.000	***
WhatsApp	45_54	55_64	-0.147	0.630
WhatsApp	45_54	65_+	-0.559	0.000	***
WhatsApp	55_64	65_+	-0.413	0.003	**
YouTube	18_24	25_34	-0.470	0.160
YouTube	18_24	35_44	-0.393	0.344
YouTube	18_24	45_54	-0.757	0.001	***
YouTube	18_24	55_64	-1.018	0.000	***
YouTube	18_24	65_+	-1.398	0.000	***
YouTube	25_34	35_44	0.077	0.998
YouTube	25_34	45_54	-0.287	0.449
YouTube	25_34	55_64	-0.548	0.010	*
YouTube	25_34	65_+	-0.928	0.000	***
YouTube	35_44	45_54	-0.364	0.194
YouTube	35_44	55_64	-0.625	0.002	**
YouTube	35_44	65_+	-1.005	0.000	***
YouTube	45_54	55_64	-0.260	0.470
YouTube	45_54	65_+	-0.641	0.000	***
YouTube	55_64	65_+	-0.380	0.074

# Graph for age
ggplot(desc_stats_age,
       aes(x = Age_cat, y = mean, fill = Age_cat)) +
  geom_col(position = "dodge", color = "grey30") +
  geom_errorbar(aes(ymin = mean - se, ymax = mean + se),
                width = 0.2,
                position = position_dodge(0.9)) +
  facet_wrap(~ Platform, scales = "free_y") +
  labs(
    title = "Platform use frequence and age class",
    x     = "Age classes",
    y     = "Platform use frequence (±SE)"
  ) +
  scale_fill_viridis_d() +
  theme_minimal() +
  theme(
    legend.position = "none",
    axis.text.x = element_text(angle = 90, hjust = 1)
  )

In the Italian case, the descriptive patterns show that social-media use is only weakly connected to political identities. Across the left–right spectrum, platform frequencies remain remarkably stable: Facebook, Instagram, WhatsApp, and YouTube display almost flat distributions, while even the more politically connoted platforms—Twitter/X, Telegram, TikTok—show only minimal ideological fluctuations. Partisan differences are similarly limited. Voters of the main parties (Fratelli d’Italia, PD, Lega, M5S, Forza Italia) exhibit nearly indistinguishable levels of engagement across all platforms, and even the smaller parties do not produce meaningful deviations. The only domain where a clear structure emerges is age: younger Italians consistently appear as the most intensive users of Instagram, TikTok, Telegram, and YouTube, while older cohorts dominate Facebook usage. These age-related gradients are strong, linear, and far more pronounced than any political differentiation. Overall, the Italian digital ecosystem appears socially and politically homogeneous, with demographic life-cycle effects far outweighing ideological or partisan ones.

When compared to Spain, this homogeneity becomes even more evident. The Spanish data similarly confirm the primacy of age as a driver of platform use, but Spain shows slightly more structured political nuances. Ideological groups in Spain display small but coherent gradients—left-leaning users are somewhat more active on Instagram and Twitter, whereas right-leaning groups lean modestly toward LinkedIn or WhatsApp. Partisan differences are also more discernible: parties such as Vox, PSOE, Sumar or Junts exhibit mild but consistent digital signatures across platforms. None of these patterns are large, yet they indicate a modest alignment between political identities and digital repertoires.

In contrast, Italy stands out for its almost total absence of political imprint on platform use. The digital environment does not reflect ideological cleavages or party competition; instead, it maps onto demographic contours, especially age. Spain shares the same demographic logic, but it retains small political inflections that suggest a more articulated interplay between digital behaviours and political identities. The comparison points to two parallel conclusions: in both countries, digital repertoires are primarily demographic phenomena, driven by generational preferences and platform affordances; but Spain preserves mild political structuring, whereas Italy’s patterns are almost entirely depoliticised, with social-media use functioning as a largely non-political, lifestyle-driven activity.

14 OLS regression model (Platform use as DV)

# 14. OLS

library(broom)

Warning: il pacchetto 'broom' è stato creato con R versione 4.4.2

library(car)

Warning: il pacchetto 'car' è stato creato con R versione 4.4.3

Caricamento del pacchetto richiesto: carData

Warning: il pacchetto 'carData' è stato creato con R versione 4.4.3

Caricamento pacchetto: 'car'

Il seguente oggetto è mascherato da 'package:psych':

    logit

Il seguente oggetto è mascherato da 'package:dplyr':

    recode

library(purrr)

Warning: il pacchetto 'purrr' è stato creato con R versione 4.4.2

Caricamento pacchetto: 'purrr'

Il seguente oggetto è mascherato da 'package:car':

    some

Il seguente oggetto è mascherato da 'package:scales':

    discard

library(stringr)

# Dependent variables (platform use)
social_vars <- c("YouTube","WhatsApp","Twitter_X","Instagram",
                 "TikTok","Facebook","Linkedin","Telegram")

# Predictors (factors + latent factors of participation)
predictors <- c("Age_cat", "Vote", "Sex", "Left_Right",
                "Pol_int", "Info_Online", "Partisan_mobil")

# OLS per each platform
ols_results <- purrr::map_df(social_vars, function(var) {
  
  fml <- as.formula(
    paste(var, "~", paste(predictors, collapse = " + "))
  )
  
  model <- lm(fml, data = data_PCA3)
  
  coef_tbl <- broom::tidy(model, conf.int = TRUE) %>%
    dplyr::mutate(
      Dependent = var,
      Adj_R2    = summary(model)$adj.r.squared,
      F_stat    = broom::glance(model)$statistic,
      p_model   = broom::glance(model)$p.value
    )
  
  # VIF and robust merge
  vif_vals <- tryCatch(car::vif(model), error = function(e) NULL)
  
  if (!is.null(vif_vals)) {
    
    if (is.matrix(vif_vals)) {
      vif_df <- tibble(
        predictor_key = rownames(vif_vals),
        VIF           = vif_vals[, 1]
      )
    } else {
      vif_df <- tibble(
        predictor_key = names(vif_vals),
        VIF           = as.numeric(vif_vals)
      )
    }
    
    predictor_pattern <- paste(predictors, collapse = "|")
    
    coef_tbl <- coef_tbl %>%
      dplyr::mutate(
        predictor_key = stringr::str_extract(term, predictor_pattern)
      ) %>%
      dplyr::left_join(vif_df, by = "predictor_key") %>%
      dplyr::select(-predictor_key)
    
  } else {
    coef_tbl <- coef_tbl %>% dplyr::mutate(VIF = NA_real_)
  }
  
  coef_tbl
})

# Model fit
fit_summary <- ols_results %>%
  dplyr::group_by(Dependent) %>%
  dplyr::summarise(
    Adj_R2  = unique(Adj_R2),
    F_stat  = unique(F_stat),
    p_model = unique(p_model),
    .groups = "drop"
  )

knitr::kable(
  fit_summary,
  digits  = 3,
  caption = "Statistiche di fit del modello OLS per ciascuna piattaforma"
)

Statistiche di fit del modello OLS per ciascuna piattaforma

Dependent	Adj_R2	F_stat	p_model
Facebook	0.111	5.697	0
Instagram	0.276	15.217	0
Linkedin	0.168	8.360	0
Telegram	0.177	8.815	0
TikTok	0.158	7.797	0
Twitter_X	0.141	6.953	0
WhatsApp	0.063	3.540	0
YouTube	0.135	6.866	0

# Coefficients table
coef_table <- ols_results %>%
  dplyr::filter(term != "(Intercept)") %>%
  dplyr::mutate(
    Significant = ifelse(p.value < 0.05, "Yes", "No")
  ) %>%
  dplyr::select(
    Dependent, term, estimate, std.error,
    conf.low, conf.high, p.value, VIF, Significant
  )

knitr::kable(
  coef_table,
  digits  = 3,
  caption = "OLS coefficients (95% CI, VIF, significance)"
)

OLS coefficients (95% CI, VIF, significance)

Dependent	term	estimate	std.error	conf.low	conf.high	p.value	VIF	Significant
YouTube	Age_cat25_34	-0.357	0.249	-0.847	0.132	0.152	1.215	No
YouTube	Age_cat35_44	-0.080	0.239	-0.549	0.390	0.739	1.215	No
YouTube	Age_cat45_54	-0.788	0.229	-1.238	-0.338	0.001	1.215	Yes
YouTube	Age_cat55_64	-0.890	0.227	-1.336	-0.444	0.000	1.215	Yes
YouTube	Age_cat65_+	-1.354	0.216	-1.779	-0.929	0.000	1.215	Yes
YouTube	VoteAlleanza Verdi-Sinistra	0.711	0.349	0.027	1.394	0.042	3.310	Yes
YouTube	VoteAzione-Italia Viva	0.738	0.331	0.088	1.388	0.026	3.310	Yes
YouTube	VoteDid not vote	0.814	0.312	0.201	1.427	0.009	3.310	Yes
YouTube	VoteForza Italia	1.266	0.392	0.497	2.035	0.001	3.310	Yes
YouTube	VoteFratelli d’Italia	0.862	0.335	0.205	1.518	0.010	3.310	Yes
YouTube	VoteLega	1.015	0.379	0.271	1.759	0.008	3.310	Yes
YouTube	VoteMovimento 5 Stelle	0.825	0.305	0.226	1.424	0.007	3.310	Yes
YouTube	VoteOther party	0.840	0.340	0.172	1.507	0.014	3.310	Yes
YouTube	VotePartito Democratico	0.461	0.292	-0.111	1.033	0.114	3.310	No
YouTube	SexFemale	-0.124	0.105	-0.330	0.083	0.240	1.148	No
YouTube	SexNon-binary	0.732	0.925	-1.084	2.548	0.429	1.148	No
YouTube	Left_RightLeft	0.095	0.179	-0.256	0.446	0.595	3.052	No
YouTube	Left_RightCenter	0.280	0.182	-0.078	0.637	0.125	3.052	No
YouTube	Left_RightRight	-0.033	0.230	-0.484	0.418	0.886	3.052	No
YouTube	Left_RightExt_right	0.369	0.258	-0.137	0.874	0.153	3.052	No
YouTube	Left_RightNo_position	0.390	0.212	-0.027	0.806	0.066	3.052	No
YouTube	Pol_intSomewhat interested	0.419	0.137	0.151	0.688	0.002	1.661	Yes
YouTube	Pol_intNot very interested	0.338	0.173	-0.002	0.678	0.051	1.661	No
YouTube	Pol_intNot at all interested	0.172	0.259	-0.338	0.681	0.508	1.661	No
YouTube	Info_Online	0.100	0.071	-0.040	0.240	0.161	1.679	No
YouTube	Partisan_mobil	0.302	0.076	0.154	0.451	0.000	1.917	Yes
WhatsApp	Age_cat25_34	-0.056	0.195	-0.440	0.327	0.774	1.219	No
WhatsApp	Age_cat35_44	-0.136	0.187	-0.504	0.231	0.467	1.219	No
WhatsApp	Age_cat45_54	-0.124	0.180	-0.477	0.229	0.492	1.219	No
WhatsApp	Age_cat55_64	-0.280	0.179	-0.631	0.070	0.117	1.219	No
WhatsApp	Age_cat65_+	-0.692	0.170	-1.026	-0.359	0.000	1.219	Yes
WhatsApp	VoteAlleanza Verdi-Sinistra	0.038	0.274	-0.499	0.576	0.888	3.319	No
WhatsApp	VoteAzione-Italia Viva	-0.119	0.260	-0.629	0.391	0.648	3.319	No
WhatsApp	VoteDid not vote	-0.040	0.245	-0.521	0.441	0.869	3.319	No
WhatsApp	VoteForza Italia	0.100	0.308	-0.504	0.704	0.744	3.319	No
WhatsApp	VoteFratelli d’Italia	0.092	0.263	-0.424	0.607	0.727	3.319	No
WhatsApp	VoteLega	0.162	0.296	-0.420	0.743	0.585	3.319	No
WhatsApp	VoteMovimento 5 Stelle	0.179	0.240	-0.292	0.650	0.456	3.319	No
WhatsApp	VoteOther party	0.049	0.266	-0.474	0.571	0.856	3.319	No
WhatsApp	VotePartito Democratico	0.088	0.229	-0.361	0.538	0.700	3.319	No
WhatsApp	SexFemale	0.220	0.082	0.059	0.382	0.008	1.146	Yes
WhatsApp	SexNon-binary	0.611	0.727	-0.815	2.038	0.401	1.146	No
WhatsApp	Left_RightLeft	0.098	0.140	-0.177	0.374	0.484	3.050	No
WhatsApp	Left_RightCenter	0.099	0.143	-0.182	0.379	0.490	3.050	No
WhatsApp	Left_RightRight	0.034	0.180	-0.319	0.388	0.849	3.050	No
WhatsApp	Left_RightExt_right	0.172	0.201	-0.223	0.568	0.392	3.050	No
WhatsApp	Left_RightNo_position	-0.245	0.167	-0.571	0.082	0.142	3.050	No
WhatsApp	Pol_intSomewhat interested	0.163	0.107	-0.047	0.372	0.128	1.642	No
WhatsApp	Pol_intNot very interested	0.252	0.135	-0.014	0.517	0.063	1.642	No
WhatsApp	Pol_intNot at all interested	0.114	0.204	-0.287	0.514	0.578	1.642	No
WhatsApp	Info_Online	-0.069	0.055	-0.178	0.040	0.212	1.670	No
WhatsApp	Partisan_mobil	0.239	0.059	0.123	0.355	0.000	1.914	Yes
Twitter_X	Age_cat25_34	-0.239	0.284	-0.796	0.317	0.399	1.225	No
Twitter_X	Age_cat35_44	-0.284	0.272	-0.817	0.249	0.296	1.225	No
Twitter_X	Age_cat45_54	-0.701	0.262	-1.215	-0.188	0.007	1.225	Yes
Twitter_X	Age_cat55_64	-0.746	0.260	-1.255	-0.236	0.004	1.225	Yes
Twitter_X	Age_cat65_+	-0.843	0.247	-1.328	-0.358	0.001	1.225	Yes
Twitter_X	VoteAlleanza Verdi-Sinistra	0.207	0.401	-0.580	0.995	0.606	3.260	No
Twitter_X	VoteAzione-Italia Viva	0.209	0.377	-0.532	0.949	0.580	3.260	No
Twitter_X	VoteDid not vote	0.334	0.359	-0.370	1.038	0.352	3.260	No
Twitter_X	VoteForza Italia	0.141	0.457	-0.755	1.037	0.757	3.260	No
Twitter_X	VoteFratelli d’Italia	-0.165	0.384	-0.919	0.589	0.668	3.260	No
Twitter_X	VoteLega	-0.030	0.437	-0.888	0.828	0.945	3.260	No
Twitter_X	VoteMovimento 5 Stelle	0.058	0.350	-0.630	0.745	0.869	3.260	No
Twitter_X	VoteOther party	0.482	0.388	-0.281	1.244	0.215	3.260	No
Twitter_X	VotePartito Democratico	0.004	0.333	-0.651	0.658	0.991	3.260	No
Twitter_X	SexFemale	-0.082	0.121	-0.319	0.155	0.495	1.146	No
Twitter_X	SexNon-binary	1.772	1.044	-0.277	3.821	0.090	1.146	No
Twitter_X	Left_RightLeft	0.241	0.204	-0.159	0.641	0.237	3.026	No
Twitter_X	Left_RightCenter	0.628	0.207	0.221	1.035	0.003	3.026	Yes
Twitter_X	Left_RightRight	0.646	0.263	0.130	1.162	0.014	3.026	Yes
Twitter_X	Left_RightExt_right	0.849	0.296	0.268	1.430	0.004	3.026	Yes
Twitter_X	Left_RightNo_position	0.368	0.246	-0.115	0.851	0.135	3.026	No
Twitter_X	Pol_intSomewhat interested	0.095	0.156	-0.210	0.401	0.541	1.648	No
Twitter_X	Pol_intNot very interested	-0.174	0.200	-0.567	0.218	0.383	1.648	No
Twitter_X	Pol_intNot at all interested	-0.442	0.300	-1.032	0.148	0.142	1.648	No
Twitter_X	Info_Online	0.201	0.081	0.042	0.359	0.013	1.679	Yes
Twitter_X	Partisan_mobil	0.487	0.086	0.317	0.656	0.000	1.911	Yes
Instagram	Age_cat25_34	-0.347	0.301	-0.938	0.244	0.250	1.221	No
Instagram	Age_cat35_44	-0.514	0.289	-1.082	0.054	0.076	1.221	No
Instagram	Age_cat45_54	-1.292	0.278	-1.838	-0.747	0.000	1.221	Yes
Instagram	Age_cat55_64	-1.780	0.276	-2.322	-1.239	0.000	1.221	Yes
Instagram	Age_cat65_+	-2.911	0.262	-3.426	-2.396	0.000	1.221	Yes
Instagram	VoteAlleanza Verdi-Sinistra	0.768	0.424	-0.064	1.599	0.070	3.294	No
Instagram	VoteAzione-Italia Viva	0.767	0.400	-0.019	1.553	0.056	3.294	No
Instagram	VoteDid not vote	0.560	0.378	-0.183	1.302	0.139	3.294	No
Instagram	VoteForza Italia	1.058	0.477	0.122	1.994	0.027	3.294	Yes
Instagram	VoteFratelli d’Italia	0.361	0.405	-0.434	1.157	0.373	3.294	No
Instagram	VoteLega	0.692	0.460	-0.211	1.595	0.133	3.294	No
Instagram	VoteMovimento 5 Stelle	0.638	0.370	-0.088	1.364	0.085	3.294	No
Instagram	VoteOther party	0.112	0.414	-0.700	0.924	0.787	3.294	No
Instagram	VotePartito Democratico	0.485	0.353	-0.208	1.178	0.170	3.294	No
Instagram	SexFemale	0.884	0.128	0.633	1.134	0.000	1.145	Yes
Instagram	SexNon-binary	-0.693	1.121	-2.892	1.506	0.536	1.145	No
Instagram	Left_RightLeft	0.097	0.218	-0.330	0.525	0.655	3.024	No
Instagram	Left_RightCenter	0.385	0.222	-0.051	0.821	0.084	3.024	No
Instagram	Left_RightRight	0.338	0.280	-0.212	0.887	0.228	3.024	No
Instagram	Left_RightExt_right	0.561	0.313	-0.054	1.176	0.074	3.024	No
Instagram	Left_RightNo_position	0.070	0.261	-0.442	0.582	0.788	3.024	No
Instagram	Pol_intSomewhat interested	0.238	0.165	-0.086	0.563	0.150	1.651	No
Instagram	Pol_intNot very interested	0.275	0.210	-0.137	0.686	0.191	1.651	No
Instagram	Pol_intNot at all interested	0.111	0.316	-0.508	0.731	0.724	1.651	No
Instagram	Info_Online	0.163	0.086	-0.005	0.331	0.057	1.670	No
Instagram	Partisan_mobil	0.326	0.092	0.146	0.507	0.000	1.919	Yes
TikTok	Age_cat25_34	-1.137	0.296	-1.718	-0.556	0.000	1.230	Yes
TikTok	Age_cat35_44	-1.499	0.284	-2.057	-0.941	0.000	1.230	Yes
TikTok	Age_cat45_54	-1.601	0.274	-2.138	-1.063	0.000	1.230	Yes
TikTok	Age_cat55_64	-2.000	0.272	-2.533	-1.467	0.000	1.230	Yes
TikTok	Age_cat65_+	-2.136	0.258	-2.643	-1.629	0.000	1.230	Yes
TikTok	VoteAlleanza Verdi-Sinistra	-0.505	0.418	-1.326	0.316	0.227	3.303	No
TikTok	VoteAzione-Italia Viva	-0.693	0.395	-1.468	0.083	0.080	3.303	No
TikTok	VoteDid not vote	-0.327	0.376	-1.065	0.410	0.384	3.303	No
TikTok	VoteForza Italia	-0.056	0.472	-0.982	0.870	0.906	3.303	No
TikTok	VoteFratelli d’Italia	-0.761	0.401	-1.549	0.026	0.058	3.303	No
TikTok	VoteLega	-0.481	0.453	-1.370	0.408	0.289	3.303	No
TikTok	VoteMovimento 5 Stelle	-0.034	0.367	-0.754	0.687	0.927	3.303	No
TikTok	VoteOther party	0.044	0.411	-0.763	0.851	0.915	3.303	No
TikTok	VotePartito Democratico	-0.494	0.349	-1.179	0.191	0.157	3.303	No
TikTok	SexFemale	0.374	0.126	0.126	0.622	0.003	1.150	Yes
TikTok	SexNon-binary	0.248	1.093	-1.897	2.394	0.820	1.150	No
TikTok	Left_RightLeft	-0.078	0.214	-0.498	0.342	0.714	3.033	No
TikTok	Left_RightCenter	0.347	0.217	-0.079	0.773	0.110	3.033	No
TikTok	Left_RightRight	0.694	0.274	0.157	1.231	0.011	3.033	Yes
TikTok	Left_RightExt_right	1.202	0.306	0.601	1.804	0.000	3.033	Yes
TikTok	Left_RightNo_position	0.392	0.256	-0.111	0.894	0.127	3.033	No
TikTok	Pol_intSomewhat interested	0.488	0.163	0.168	0.808	0.003	1.646	Yes
TikTok	Pol_intNot very interested	0.437	0.208	0.029	0.845	0.036	1.646	Yes
TikTok	Pol_intNot at all interested	-0.055	0.313	-0.668	0.558	0.860	1.646	No
TikTok	Info_Online	0.356	0.084	0.190	0.521	0.000	1.672	Yes
TikTok	Partisan_mobil	0.096	0.090	-0.082	0.273	0.290	1.915	No
Facebook	Age_cat25_34	0.631	0.299	0.046	1.217	0.035	1.215	Yes
Facebook	Age_cat35_44	1.451	0.286	0.890	2.013	0.000	1.215	Yes
Facebook	Age_cat45_54	1.163	0.274	0.625	1.702	0.000	1.215	Yes
Facebook	Age_cat55_64	1.078	0.272	0.544	1.612	0.000	1.215	Yes
Facebook	Age_cat65_+	0.571	0.259	0.062	1.080	0.028	1.215	Yes
Facebook	VoteAlleanza Verdi-Sinistra	0.716	0.416	-0.100	1.532	0.085	3.331	No
Facebook	VoteAzione-Italia Viva	0.428	0.394	-0.346	1.202	0.278	3.331	No
Facebook	VoteDid not vote	0.682	0.373	-0.049	1.413	0.068	3.331	No
Facebook	VoteForza Italia	1.072	0.468	0.155	1.990	0.022	3.331	Yes
Facebook	VoteFratelli d’Italia	0.436	0.399	-0.347	1.219	0.275	3.331	No
Facebook	VoteLega	0.994	0.451	0.108	1.880	0.028	3.331	Yes
Facebook	VoteMovimento 5 Stelle	1.136	0.365	0.419	1.852	0.002	3.331	Yes
Facebook	VoteOther party	0.239	0.405	-0.557	1.034	0.556	3.331	No
Facebook	VotePartito Democratico	0.920	0.348	0.237	1.602	0.008	3.331	Yes
Facebook	SexFemale	0.627	0.125	0.381	0.873	0.000	1.147	Yes
Facebook	SexNon-binary	0.930	1.104	-1.237	3.097	0.400	1.147	No
Facebook	Left_RightLeft	-0.017	0.214	-0.436	0.402	0.935	3.055	No
Facebook	Left_RightCenter	0.261	0.217	-0.166	0.688	0.231	3.055	No
Facebook	Left_RightRight	0.499	0.274	-0.039	1.037	0.069	3.055	No
Facebook	Left_RightExt_right	0.823	0.307	0.220	1.426	0.007	3.055	Yes
Facebook	Left_RightNo_position	0.055	0.254	-0.443	0.553	0.828	3.055	No
Facebook	Pol_intSomewhat interested	0.100	0.162	-0.219	0.418	0.538	1.652	No
Facebook	Pol_intNot very interested	0.426	0.206	0.022	0.831	0.039	1.652	Yes
Facebook	Pol_intNot at all interested	0.242	0.309	-0.365	0.849	0.434	1.652	No
Facebook	Info_Online	-0.001	0.085	-0.168	0.167	0.993	1.669	No
Facebook	Partisan_mobil	0.499	0.090	0.322	0.676	0.000	1.908	Yes
Linkedin	Age_cat25_34	0.581	0.237	0.116	1.046	0.014	1.227	Yes
Linkedin	Age_cat35_44	0.820	0.227	0.374	1.266	0.000	1.227	Yes
Linkedin	Age_cat45_54	0.077	0.218	-0.351	0.505	0.724	1.227	No
Linkedin	Age_cat55_64	-0.031	0.217	-0.457	0.395	0.886	1.227	No
Linkedin	Age_cat65_+	-0.456	0.206	-0.861	-0.051	0.027	1.227	Yes
Linkedin	VoteAlleanza Verdi-Sinistra	0.108	0.337	-0.553	0.769	0.750	3.383	No
Linkedin	VoteAzione-Italia Viva	-0.103	0.318	-0.726	0.521	0.747	3.383	No
Linkedin	VoteDid not vote	-0.193	0.301	-0.783	0.398	0.522	3.383	No
Linkedin	VoteForza Italia	-0.309	0.378	-1.050	0.432	0.413	3.383	No
Linkedin	VoteFratelli d’Italia	-0.826	0.324	-1.462	-0.189	0.011	3.383	Yes
Linkedin	VoteLega	-0.600	0.369	-1.324	0.124	0.104	3.383	No
Linkedin	VoteMovimento 5 Stelle	-0.262	0.295	-0.840	0.317	0.375	3.383	No
Linkedin	VoteOther party	-0.094	0.329	-0.738	0.551	0.776	3.383	No
Linkedin	VotePartito Democratico	-0.014	0.281	-0.565	0.537	0.960	3.383	No
Linkedin	SexFemale	-0.084	0.101	-0.283	0.115	0.407	1.145	No
Linkedin	SexNon-binary	0.000	0.878	-1.723	1.723	1.000	1.145	No
Linkedin	Left_RightLeft	0.103	0.171	-0.234	0.439	0.549	3.083	No
Linkedin	Left_RightCenter	0.712	0.174	0.370	1.054	0.000	3.083	Yes
Linkedin	Left_RightRight	0.890	0.221	0.456	1.324	0.000	3.083	Yes
Linkedin	Left_RightExt_right	1.004	0.248	0.518	1.491	0.000	3.083	Yes
Linkedin	Left_RightNo_position	0.552	0.206	0.147	0.957	0.008	3.083	Yes
Linkedin	Pol_intSomewhat interested	0.143	0.131	-0.115	0.400	0.277	1.645	No
Linkedin	Pol_intNot very interested	-0.062	0.167	-0.391	0.266	0.710	1.645	No
Linkedin	Pol_intNot at all interested	-0.406	0.252	-0.901	0.089	0.108	1.645	No
Linkedin	Info_Online	0.427	0.068	0.294	0.561	0.000	1.683	Yes
Linkedin	Partisan_mobil	0.024	0.072	-0.118	0.166	0.742	1.920	No
Telegram	Age_cat25_34	-0.070	0.295	-0.648	0.509	0.814	1.239	No
Telegram	Age_cat35_44	0.033	0.283	-0.522	0.588	0.906	1.239	No
Telegram	Age_cat45_54	-0.555	0.273	-1.090	-0.020	0.042	1.239	Yes
Telegram	Age_cat55_64	-0.789	0.270	-1.319	-0.260	0.004	1.239	Yes
Telegram	Age_cat65_+	-1.557	0.258	-2.064	-1.050	0.000	1.239	Yes
Telegram	VoteAlleanza Verdi-Sinistra	0.186	0.416	-0.631	1.002	0.655	3.365	No
Telegram	VoteAzione-Italia Viva	-0.261	0.394	-1.035	0.512	0.507	3.365	No
Telegram	VoteDid not vote	0.277	0.374	-0.457	1.011	0.459	3.365	No
Telegram	VoteForza Italia	0.237	0.472	-0.690	1.164	0.616	3.365	No
Telegram	VoteFratelli d’Italia	0.158	0.401	-0.630	0.946	0.695	3.365	No
Telegram	VoteLega	0.186	0.456	-0.709	1.080	0.684	3.365	No
Telegram	VoteMovimento 5 Stelle	0.104	0.365	-0.612	0.820	0.775	3.365	No
Telegram	VoteOther party	0.632	0.406	-0.165	1.429	0.120	3.365	No
Telegram	VotePartito Democratico	0.112	0.348	-0.570	0.794	0.747	3.365	No
Telegram	SexFemale	-0.253	0.126	-0.501	-0.006	0.045	1.150	Yes
Telegram	SexNon-binary	-0.240	1.088	-2.375	1.896	0.826	1.150	No
Telegram	Left_RightLeft	0.643	0.213	0.225	1.061	0.003	3.067	Yes
Telegram	Left_RightCenter	0.832	0.217	0.407	1.258	0.000	3.067	Yes
Telegram	Left_RightRight	1.034	0.275	0.493	1.574	0.000	3.067	Yes
Telegram	Left_RightExt_right	1.276	0.309	0.669	1.883	0.000	3.067	Yes
Telegram	Left_RightNo_position	0.863	0.254	0.365	1.361	0.001	3.067	Yes
Telegram	Pol_intSomewhat interested	0.220	0.163	-0.100	0.539	0.177	1.658	No
Telegram	Pol_intNot very interested	0.147	0.207	-0.260	0.554	0.478	1.658	No
Telegram	Pol_intNot at all interested	-0.106	0.312	-0.717	0.506	0.734	1.658	No
Telegram	Info_Online	0.410	0.084	0.245	0.575	0.000	1.686	Yes
Telegram	Partisan_mobil	0.201	0.090	0.024	0.378	0.026	1.937	Yes

# Coefficient plot
coef_table %>%
  dplyr::mutate(
    term = stringr::str_replace_all(
      term,
      c(
        "Age_cat" = "Age: ",
        "Vote"    = "Vote: ",
        "Sex"     = "Sex: ",
        "Left_Right" = "L–R: ",
        "Pol_int"    = "Pol. interest: ",
        "\\[T\\."    = "",
        "\\]"        = ""
      )
    )
  ) %>%
  ggplot(aes(
    y     = reorder(term, estimate),
    x     = estimate,
    color = Significant,
    shape = Significant
  )) +
  geom_point(position = position_dodge(width = 0.5), size = 1) +
  geom_errorbar(aes(xmin = conf.low, xmax = conf.high),
                height   = 0.2,
                position = position_dodge(width = 0.5),
                orientation = "y") +
  geom_vline(xintercept = 0, linetype = "dashed", color = "gray50") +
  facet_wrap(~ Dependent, scales = "free_y", ncol = 2) +
  scale_color_manual(values = c("No" = "gray50", "Yes" = "#E69F00")) +
  scale_shape_manual(values = c("No" = 16, "Yes" = 17)) +
  theme_bw(base_size = 6) +
  theme(
    legend.position       = "bottom",
    panel.grid.minor      = element_blank(),
    panel.grid.major.y    = element_blank(),
    strip.background      = element_rect(fill = "gray90"),
    axis.text.y           = element_text(size = 4)
  ) +
  labs(
    title    = "OLS coefficients per platform",
    subtitle = "Statistically significant coefficients (p < .05)",
    y        = "Predictors",
    x        = "Coefficient extimation (±95% CI)",
    color    = "Significant:",
    shape    = "Significant:"
  )

`height` was translated to `width`.

The Italian OLS models show that platform use is associated only weakly and selectively with socio-political traits, and that the structure of the predictors is far from homogeneous across platforms. Two variables stand out consistently: age and political interest.

Age displays the strongest and most systematic coefficients. Its effect is always negative for visually oriented or fast-paced platforms (Instagram, TikTok, YouTube, Telegram), indicating that younger respondents adopt these environments more intensively. Conversely, platforms such as Facebook and WhatsApp show weaker and sometimes negligible age gradients, consistent with their more mature user base.

Political interest emerges as a second robust predictor: more interested individuals tend to use a broader set of platforms more frequently. This is especially visible for platforms that support information-rich content (Twitter/X, YouTube, LinkedIn). However, even here the effect size remains modest, confirming that engagement in Italian digital spaces is only partially anchored to political motivation.

Other socio-political variables show fragmented and platform-specific patterns:
– Left–right orientation has no substantial effect except for marginal tendencies (e.g., slightly higher Twitter/X use among left-leaning individuals).
– Information-seeking behaviours (disagreement exposure, cross-checking, confirmation) produce small but coherent positive associations with Twitter/X and YouTube, reflecting their role as news-intensive platforms.
– Sex rarely matters, and when it does, coefficients are small.
– Party preference shows no meaningful systematic effect: differences across voters are minimal and unstable once covariates are included.

the OLS results reinforce the idea already seen in PCA and clustering: Italian social-media behaviour is primarily structured by age and individual media engagement, while political orientation, partisan alignment and mobilisation styles play a marginal role. Platforms do not map onto the partisan field; instead, they align with demographic and informational repertoires.

The Italian OLS models reveal a relatively structured set of predictors shaping platform-specific usage. Age is consistently the strongest and most stable negative predictor across all platforms, confirming a clear generational gradient. Political interest is also a robust and positive correlate, indicating that in Italy higher engagement with politics translates into higher intensity of social media use—especially on LinkedIn, Twitter/X, and Instagram. The ideological left–right axis shows only weak and platform-specific associations, with minor left-leaning overuse on Instagram and Twitter/X, while gender effects remain limited and inconsistent. Overall, Italian platform use is driven primarily by sociodemographic fundamentals (age, education) and general political engagement rather than by partisan or ideological alignment. In Spain, the pattern is more diffuse and substantially weaker. Age remains a strong negative predictor—but beyond this universal baseline, few variables show systematic and significant associations. Political interest is less consistently related to social media use, and the ideological dimension is almost entirely irrelevant across platforms. Partisan vote and left–right self-placement produce scattered effects that rarely cross significance thresholds. The Spanish models are therefore flatter and more fragmented, with platform use displaying fewer political or demographic structuring forces. The results point to a clearer and more hierarchical structure of predictors in Italy, where age and political interest jointly organize the digital space, and where some platforms show small but detectable ideological tilts. In Spain, by contrast, platform use appears more socially and politically “neutral,” with very limited differentiation across political groups and a much weaker imprint of political engagement. What emerges is a more politically stratified digital ecosystem in Italy, versus a more socially diffuse and weakly politicized one in Spain.