IL LCR Regression
IL Political Extremism
Amir Freund
2025-08-24
# Read the data from indices_table.txt
df <- as.data.frame(readRDS("Israel Survey/data/il_pe.RDS"))
wave_var <- "Wave"
wave_order <- c("First", "Second", "Third", "Fourth", "Fifth", "Sixth")
community_var <- "pe_left_center_right"
community_order <- c("left", "center", "right") # Set "center" as the reference level since it is the 'neutral' level.
dimensions_order <- c("Overall", "Cognitive", "Behavioral", "Social")
# Calculate the Political Extremism Gauge Indices
indices_result <- af_gauge_indices(df, pop_var1 = wave_var, comm_var1 = community_var,
threshold_type = "MAD", k_factor = 1.5)
df <- indices_result$df
# Convert data to a more manageable format for analysis
df <- df %>%
mutate(Wave = factor(Wave, levels = wave_order)) %>%
mutate(!!sym(community_var) := factor(!!sym(community_var), levels = community_order)) %>%
mutate(event_occurred = factor(1, levels = c(0, 1))) # used for pairwise regression
# Print Event Table
event_table <- data.frame(
event_name = c("Inland Terror", "Bennet Gov. Fall", "Judicial Reform", "Gallant Dismissal", "Oct. 7th War"),
waves = c("1-2", "2-3", "3-4", "4-5", "5-6"),
type = c("Security", "Political", "Political", "Political", "Security"),
stringsAsFactors = FALSE
)
gt(event_table) %>%
tab_header(
title = md("**Event Table**"),
)| Event Table | ||
| event_name | waves | type |
|---|---|---|
| Inland Terror | 1-2 | Security |
| Bennet Gov. Fall | 2-3 | Political |
| Judicial Reform | 3-4 | Political |
| Gallant Dismissal | 4-5 | Political |
| Oct. 7th War | 5-6 | Security |
# add event_type and event_result info to df based on the event_table
df <- af_add_event_info(df, wave_var, wave_order, event_table, community_var)
# Create the wave list in the form
wave_list <- list()
for(i in 1:nrow(event_table)) {
wave_range <- as.numeric(unlist(strsplit(event_table$waves[i], "-")))
wave_list[[event_table$event_name[i]]] <- c(wave_order[wave_range[1]],
wave_order[wave_range[2]])
}
# Create demographics regression formula part
demographics <- c("gender", "age_group") # , "education"
d_fmla <- paste(demographics, collapse = "+")
# Set display names for regression results
display_names <- list(
"Wave" = "Wave",
"post_event1" = "Inland Terror",
"post_event2" = "Bennet Gov. Fall",
"post_event3" = "Judicial Reform",
"post_event4" = "Gallant Dismissal",
"post_event5" = "Oct. 7th War",
"immediate_event1" = "Inland Terror",
"immediate_event2" = "Bennet Gov. Fall",
"immediate_event3" = "Judicial Reform",
"immediate_event4" = "Gallant Dismissal",
"immediate_event5" = "Oct. 7th War",
"pe_left_center_right" = "Political Orientation",
"event_occurred" = "Event Occured",
"event_type" = "Event Type",
"gender" = "Gender",
"age_group" = "Age Group",
"education" = "Education"
)
coef_names <- c(
"event_occurred1" = "Event Occurred",
"pe_left_center_rightcenter" = "Political Orientation[center]",
"pe_left_center_rightright" = "Political Orientation[right]",
"event_occurred1:pe_left_center_rightcenter" = "Event Occurred : Political Orientation[center]",
"event_occurred1:pe_left_center_rightright" = "Event Occurred : Political Orientation[right]"
)1 Individual Extremism Level
The comprehensive analysis across all three dimensions of political extremism reveals a complex landscape of differential responsiveness to destabilizing events, providing robust support for the theoretical framework while highlighting the multifaceted nature of political radicalization.
Hypothesis 1 receives unequivocal confirmation, as the three extremism dimensions demonstrated fundamentally different response patterns: cognitive extremism showed dynamic, event-specific changes with significant effects in three of five events; behavioral extremism exhibited more limited responsiveness with significant effects in only two events and generally smaller effect sizes; while social extremism remained remarkably stable with no significant event effects despite massive baseline political orientation differences.
Hypothesis 2 receives partial but important support, with political orientation significantly moderating event effects in cognitive extremism (3 of 5 events) and behavioral extremism (2 of 5 events), though notably failing to moderate any effects in social extremism despite substantial baseline differences.
Hypothesis 3 receives mixed and dimension-specific support: the expectation that all events would uniformly increase extremism (3.1, 3.5) was largely unsupported across dimensions, with security-focused events (Inland Terror, October 7th) showing minimal impact; institutional/governmental events (Bennett Government Fall, Judicial Reform, Gallant Dismissal) produced more complex patterns of increases and decreases that varied by both political orientation and extremism dimension, suggesting that threat perceptions and responses are highly contextual and dimension-specific rather than uniform across all forms of political extremism.
These findings underscore that political extremism is not a monolithic construct but rather comprises distinct psychological and behavioral components that respond differentially to various types of political stressors, with important implications for understanding radicalization processes and developing targeted intervention strategies.
1.1 Create Panel Data
# Create Panel Dataset
df1 <- df$respondent_id[df$Wave == "Third"]
df2 <- df$respondent_id[df$Wave == "Fourth"]
panel_respondents <- intersect(df1, df2)
panel_df <- df %>%
filter(Wave %in% c("Third", "Fourth")) %>%
filter(respondent_id %in% panel_respondents) %>%
mutate(event_occurred = factor(case_when(
Wave == "Third" ~ 0,
Wave == "Fourth" ~ 1
), levels = c(0, 1)))
p_data <- plm::pdata.frame(panel_df, index = c("respondent_id", "event_occurred"))
plm::pdim(p_data)Balanced Panel: n = 671, T = 2, N = 1342
1.2 Create Panel Logistic Models
# Logistic regression with the response variable being a binary indicator of whether the respondent
# is part of the more extremist group (counted in the Extremism Level index).
#
# The model asks: "How do event_occurred, political_orientation, and their interaction affect the odds of i_cel = 1,
# controlling for age_group and accounting for unobserved individual-level characteristics?"
mp1 <- clogit(i_cel ~ event_occurred * pe_left_center_right + age_group + cluster(respondent_id),
data = panel_df, method = "approximate")
mp2 <- clogit(i_bel ~ event_occurred * pe_left_center_right + age_group + cluster(respondent_id),
data = panel_df, method = "approximate")
mp3_no_interaction <- clogit(i_sel ~ event_occurred + pe_left_center_right + age_group + cluster(respondent_id),
data = panel_df, method = "approximate")
mp4 <- clogit(i_oel ~ event_occurred * pe_left_center_right + age_group + cluster(respondent_id),
data = panel_df, method = "approximate")1.3 Cognitive Dimension
The logistic regression analyses examining cognitive extremism (ideology dimension) across five destabilizing events reveal nuanced patterns of political radicalization that vary considerably by both event type and political orientation. The Bennett Government Fall emerges as the most polarizing event, demonstrating a substantial main effect (OR = 3.125, p < 0.001) alongside significant interaction effects, particularly for right-wing respondents who showed dramatically reduced extremism likelihood (OR = 0.307, p < 0.001) while center-wing respondents exhibited marginally significant reductions (OR = 0.453, p < 0.05). The Judicial Reform produced contrasting dynamics, with center-wing respondents showing significantly increased extremism (OR = 1.661, p < 0.01) while right-wing respondents remained statistically unchanged, suggesting asymmetric threat perceptions across the political spectrum. The Gallant Dismissal generated a moderate negative main effect (OR = 0.420, p < 0.01) but was characterized by divergent responses, with right-wing respondents showing marginally significant increases in extremism (OR = 1.923, p < 0.05). Notably, both the Inland Terror attack and October 7th War, despite their severity, failed to produce significant main effects or interaction effects on cognitive extremism, indicating that these security-focused events may not substantially alter ideological positioning. Demographic controls consistently showed that center and right-wing orientations were associated with lower baseline extremism compared to left-wing respondents, while older age groups (60+) demonstrated consistently higher extremism across most events.
These findings provide mixed support for the theoretical framework, with Hypothesis 1 receiving strong confirmation as different extremism dimensions respond non-uniformly to socio-political events, evidenced by the highly variable effect sizes and significance patterns across events.
Hypothesis 2 is partially supported, as political orientation significantly moderated event effects in three of five cases (Bennett Government Fall, Judicial Reform, and Gallant Dismissal), though the moderation patterns were more complex than anticipated.
Hypothesis 3 receives nuanced support: contrary to expectation 3.1, Inland Terror showed no significant impact on any political group; expectation 3.2 was partially confirmed as the Bennett Government Fall reduced right-wing extremism but did not increase left/center extremism; expectation 3.3 was partially supported with increased center-wing extremism during Judicial Reform but no corresponding right-wing reduction; expectation 3.4 was partially confirmed with increased right-wing extremism following Gallant Dismissal; and expectation 3.5 was not supported as October 7th showed no significant effects despite its presumed severity.
formula_str <- paste("i_cel ~ event_occurred * pe_left_center_right","+", d_fmla)
models <- af_wave_pair_regression(df, wave_var = wave_var, wave_list = wave_list,
formula_str = formula_str, regression_type = "Logit")
models[[3]] <- mp1 # Override with panel model
coef_table <- af_coef_and_ci_table(models, coef_names)
af_coef_and_ci_plot(coef_table, xpose = TRUE, title = "Cognitive Dimension",
exp_ctrl = c(TRUE, TRUE, TRUE, TRUE, TRUE))
notes <- af_create_regression_notes(df, models = models, display_names = display_names, is_exp = TRUE,
show_significance = TRUE, significance_levels = c(0.05, 0.01, 0.001))
cov_labels <- af_cov_names(df, models, display_names)
af_stargazer(models = models, cov_labels = cov_labels, coef_exp = TRUE, notes = notes,
title = "Individual Extremist ~ Event Occurred x Political Orientation + Demographics")| Dependent variable: | |||||
| i_cel | i_cel | i_cel | |||
| logistic | conditional | logistic | |||
| logistic | |||||
| Inland Terror | Bennet Gov. Fall | Judicial Reform | Gallant Dismissal | Oct. 7th War | |
| (1) | (2) | (3) | (4) | (5) | |
| Event Occured[1] | 0.969 | 3.125*** | 0.964 | 0.420** | 1.456 |
| (0.194) | (0.246) | (0.220) | (0.276) | (0.247) | |
| Political Orientation[center] | 0.277*** | 0.209*** | 0.217*** | 0.208*** | 0.330*** |
| (0.196) | (0.208) | (0.257) | (0.287) | (0.208) | |
| Political Orientation[right] | 0.526*** | 0.400*** | 0.282*** | 0.138*** | 0.283*** |
| (0.160) | (0.166) | (0.200) | (0.259) | (0.191) | |
| Gender[Female] | 0.513*** | 0.468*** | 0.402*** | 0.521*** | |
| (0.094) | (0.112) | (0.119) | (0.105) | ||
| Age Group[31–45] | 0.932 | 0.912 | 0.887 | 0.875 | 1.073 |
| (0.124) | (0.151) | (0.172) | (0.155) | (0.150) | |
| Age Group[46–60] | 1.107 | 1.196 | 1.256 | 1.112 | 1.292 |
| (0.129) | (0.156) | (0.173) | (0.156) | (0.150) | |
| Age Group[60plus] | 1.323* | 1.684** | 1.848*** | 1.808** | 1.844*** |
| (0.140) | (0.160) | (0.175) | (0.183) | (0.163) | |
| Event Occured[1] × Political Orientation[center] | 0.765 | 0.453* | 1.661** | 1.518 | 1.061 |
| (0.284) | (0.353) | (0.339) | (0.355) | (0.306) | |
| Event Occured[1] × Political Orientation[right] | 0.751 | 0.307*** | 0.995 | 1.923* | 0.783 |
| (0.228) | (0.286) | (0.278) | (0.322) | (0.285) | |
| Observations | 3,215 | 2,493 | 1,342 | 2,221 | 2,638 |
| R2 | 0.069 | ||||
| Note: * p < 0.050; ** p < 0.010; *** p < 0.001. The reference category for Event Occured is ‘0’. The reference category for Political Orientation is ‘left’. The reference category for Gender is ‘Male’. The reference category for Age Group is ‘18–30’. Standard errors in parentheses. The coefficients of models 1 and 2 and 3 and 4 and 5 are exponentiated. | |||||
1.4 Behavioral Dimension
The logistic regression analyses examining behavioral extremism (violence dimension) reveal markedly different patterns compared to cognitive extremism, with generally weaker main effects but notable baseline political orientation differences and specific interaction patterns. The Inland Terror attack produced a significant negative main effect (OR = 0.521, p < 0.01), indicating reduced behavioral extremism across the population, but this effect was significantly moderated by right-wing orientation, with right-wing respondents showing increased behavioral extremism (OR = 1.853, p < 0.01). The Bennett Government Fall demonstrated distinctive baseline effects, with right-wing respondents showing substantially higher behavioral extremism (OR = 2.319, p < 0.001), though the event itself and its interactions showed no significant effects. The Judicial Reform revealed complex dynamics with right-wing respondents exhibiting elevated baseline extremism (OR = 1.824, p < 0.05) but significantly reduced extremism following the event (OR = 0.515, p < 0.05), suggesting that this institutional challenge may have dampened violent tendencies among right-wing constituents. Neither the Gallant Dismissal nor October 7th War produced significant main effects or interaction effects, indicating limited impact on behavioral extremism. Demographic patterns consistently showed that female respondents exhibited significantly lower behavioral extremism across all events (OR ranging from 0.637 to 0.699, all p < 0.001), while older age groups generally demonstrated reduced behavioral extremism, contrasting sharply with the cognitive dimension findings.
These behavioral extremism findings provide additional complexity to the theoretical framework, offering continued support for Hypothesis 1 as the behavioral dimension responds differently to events compared to the cognitive dimension, with generally smaller effect sizes and different significance patterns.
Hypothesis 2 receives partial support, with significant moderation effects observed in two of five events (Inland Terror and Judicial Reform), though the moderation patterns differ substantially from cognitive extremism.
Hypothesis 3 receives limited support in this dimension: expectation 3.1 is contradicted as Inland Terror decreased rather than increased extremism overall, though right-wing groups did show increased extremism; expectations 3.2 and 3.3 are not supported as neither the Bennett Government Fall nor Judicial Reform increased left-wing extremism, with the latter actually reducing right-wing behavioral extremism; expectation 3.4 is not supported as Gallant Dismissal showed no significant effects; and expectation 3.5 is not supported as October 7th demonstrated no significant impact on behavioral extremism across any political orientation.
formula_str <- paste("i_bel ~ event_occurred * pe_left_center_right","+", d_fmla)
models <- af_wave_pair_regression(df, wave_var = wave_var, wave_list = wave_list,
formula_str = formula_str, regression_type = "Logit")
models[[3]] <- mp2 # Override with panel model
coef_table <- af_coef_and_ci_table(models, coef_names)
af_coef_and_ci_plot(coef_table, xpose = TRUE, title = "Behavioral Dimension",
exp_ctrl = c(TRUE, TRUE, TRUE, TRUE, TRUE))
notes <- af_create_regression_notes(df, models = models, display_names = display_names, is_exp = TRUE,
show_significance = TRUE, significance_levels = c(0.05, 0.01, 0.001))
cov_labels <- af_cov_names(df, models, display_names)
af_stargazer(models = models, cov_labels = cov_labels, coef_exp = TRUE, notes = notes,
title = "Individual Extremist ~ Event Occurred x Political Orientation + Demographics")| Dependent variable: | |||||
| i_bel | i_bel | i_bel | |||
| logistic | conditional | logistic | |||
| logistic | |||||
| Inland Terror | Bennet Gov. Fall | Judicial Reform | Gallant Dismissal | Oct. 7th War | |
| (1) | (2) | (3) | (4) | (5) | |
| Event Occured[1] | 0.521** | 0.810 | 1.557 | 1.343 | 1.289 |
| (0.211) | (0.302) | (0.344) | (0.286) | (0.248) | |
| Political Orientation[center] | 0.878 | 1.283 | 1.451 | 1.228 | 0.881 |
| (0.171) | (0.194) | (0.310) | (0.284) | (0.191) | |
| Political Orientation[right] | 1.250 | 2.319*** | 1.824* | 0.952 | 0.770 |
| (0.154) | (0.176) | (0.290) | (0.259) | (0.178) | |
| Gender[Female] | 0.699*** | 0.682*** | 0.693*** | 0.637*** | |
| (0.077) | (0.089) | (0.095) | (0.086) | ||
| Age Group[31–45] | 0.848 | 0.776* | 0.891 | 0.988 | 0.928 |
| (0.098) | (0.114) | (0.121) | (0.118) | (0.113) | |
| Age Group[46–60] | 0.695*** | 0.577*** | 0.422*** | 0.575*** | 0.606*** |
| (0.106) | (0.126) | (0.163) | (0.129) | (0.121) | |
| Age Group[60plus] | 0.615*** | 0.670** | 0.878 | 0.810 | 0.687** |
| (0.122) | (0.134) | (0.147) | (0.160) | (0.139) | |
| Event Occured[1] × Political Orientation[center] | 1.461 | 1.571 | 0.769 | 0.718 | 1.181 |
| (0.258) | (0.351) | (0.394) | (0.343) | (0.290) | |
| Event Occured[1] × Political Orientation[right] | 1.853** | 1.111 | 0.515* | 0.816 | 0.735 |
| (0.233) | (0.322) | (0.366) | (0.315) | (0.272) | |
| Observations | 3,215 | 2,493 | 1,342 | 2,221 | 2,638 |
| R2 | 0.031 | ||||
| Note: * p < 0.050; ** p < 0.010; *** p < 0.001. The reference category for Event Occured is ‘0’. The reference category for Political Orientation is ‘left’. The reference category for Gender is ‘Male’. The reference category for Age Group is ‘18–30’. Standard errors in parentheses. The coefficients of models 1 and 2 and 3 and 4 and 5 are exponentiated. | |||||
2 Dimensions
2.1 Create Panel Models (plm)
# The within estimator will automatically drop the main effects of time-invariant variables
# (like pe_left_center_right, gender, age_group) but keeps the interactions event_occurred * pe_left_center_right.
mp1 <- plm(pe_ideology ~ event_occurred * pe_left_center_right + age_group, data = p_data, model = "within")
mp2 <- plm(pe_violence ~ event_occurred * pe_left_center_right + age_group, data = p_data, model = "within")
mp3 <- plm(pe_intolerance ~ event_occurred * pe_left_center_right + age_group, data = p_data, model = "within")
mp4 <- plm(pe_overall ~ event_occurred * pe_left_center_right + age_group, data = p_data, model = "within")2.2 Cognitive Dimension
formula_str <- paste("pe_ideology ~ event_occurred * pe_left_center_right","+", d_fmla)
models <- af_wave_pair_regression(df, wave_var = wave_var, wave_list = wave_list,
formula_str = formula_str, regression_type = "OLS")
models[[3]] <- mp1 # Override with panel model
coef_table <- af_coef_and_ci_table(models, coef_names)
af_coef_and_ci_plot(coef_table, xpose = TRUE, title = "Cognitive Dimension")
notes <- af_create_regression_notes(df, models = models, display_names = display_names,
show_significance = TRUE, significance_levels = c(0.05, 0.01, 0.001))
cov_labels <- af_cov_names(df, models, display_names)
af_stargazer(models = models, cov_labels = cov_labels, notes = notes,
title = "Cognitive Political Extremism ~ Event Occurred x Political Orientation + Demographics")| Dependent variable: | |||||
| pe_ideology | |||||
| OLS | panel | OLS | |||
| linear | |||||
| Inland Terror | Bennet Gov. Fall | Judicial Reform | Gallant Dismissal | Oct. 7th War | |
| (1) | (2) | (3) | (4) | (5) | |
| Event Occured[1] | 0.041 | 0.459* | 0.256 | -0.386 | 0.295 |
| (0.168) | (0.194) | (0.176) | (0.227) | (0.211) | |
| Political Orientation[center] | -1.256*** | -1.320*** | 0.347 | -1.667*** | -1.026*** |
| (0.147) | (0.131) | (0.303) | (0.225) | (0.158) | |
| Political Orientation[right] | -0.767*** | -1.222*** | 0.214 | -1.745*** | -1.156*** |
| (0.134) | (0.121) | (0.330) | (0.204) | (0.147) | |
| Gender[Female] | -0.570*** | -0.587*** | -0.621*** | -0.519*** | |
| (0.066) | (0.068) | (0.075) | (0.070) | ||
| Age Group[31–45] | -0.067 | -0.188* | 0.123 | -0.097 | -0.030 |
| (0.086) | (0.089) | (1.006) | (0.097) | (0.095) | |
| Age Group[46–60] | -0.007 | -0.028 | 1.036 | 0.112 | 0.200* |
| (0.092) | (0.095) | (1.299) | (0.100) | (0.097) | |
| Age Group[60plus] | 0.302** | 0.383*** | 0.535*** | 0.525*** | |
| (0.103) | (0.102) | (0.128) | (0.114) | ||
| Event Occured[1] × Political Orientation[center] | -0.068 | -0.360 | -0.284 | 0.634* | -0.240 |
| (0.207) | (0.234) | (0.211) | (0.273) | (0.246) | |
| Event Occured[1] × Political Orientation[right] | -0.467* | -0.543* | -0.233 | 0.570* | -0.289 |
| (0.189) | (0.214) | (0.190) | (0.249) | (0.229) | |
| Observations | 3,215 | 2,493 | 1,342 | 2,221 | 2,638 |
| R2 | 0.080 | 0.121 | 0.006 | 0.096 | 0.080 |
| Adjusted R2 | 0.077 | 0.117 | -1.007 | 0.092 | 0.077 |
| Note: * p < 0.050; ** p < 0.010; *** p < 0.001. The reference category for Event Occured is ‘0’. The reference category for Political Orientation is ‘left’. The reference category for Gender is ‘Male’. The reference category for Age Group is ‘18–30’. Standard errors in parentheses. | |||||
2.3 Behavioral Dimension
formula_str <- paste("pe_violence ~ event_occurred * pe_left_center_right","+", d_fmla)
models <- af_wave_pair_regression(df, wave_var = wave_var, wave_list = wave_list,
formula_str = formula_str, regression_type = "OLS")
models[[3]] <- mp2 # Override with panel model
coef_table <- af_coef_and_ci_table(models, coef_names)
af_coef_and_ci_plot(coef_table, xpose = TRUE, title = "Behavioral Dimension")
notes <- af_create_regression_notes(df, models = models, display_names = display_names,
show_significance = TRUE, significance_levels = c(0.05, 0.01, 0.001))
cov_labels <- af_cov_names(df, models, display_names)
af_stargazer(models = models, cov_labels = cov_labels, notes = notes,
title = "Behavioral Political Extremism ~ Event Occurred x Political Orientation + Demographics")| Dependent variable: | |||||
| pe_violence | |||||
| OLS | panel | OLS | |||
| linear | |||||
| Inland Terror | Bennet Gov. Fall | Judicial Reform | Gallant Dismissal | Oct. 7th War | |
| (1) | (2) | (3) | (4) | (5) | |
| Event Occured[1] | -0.031 | -0.122 | 0.114 | 0.156 | 0.326** |
| (0.086) | (0.114) | (0.102) | (0.123) | (0.113) | |
| Political Orientation[center] | 0.105 | 0.152* | 0.036 | 0.193 | 0.029 |
| (0.075) | (0.077) | (0.176) | (0.121) | (0.085) | |
| Political Orientation[right] | 0.165* | 0.353*** | -0.027 | 0.092 | -0.013 |
| (0.069) | (0.071) | (0.192) | (0.110) | (0.079) | |
| Gender[Female] | -0.176*** | -0.192*** | -0.275*** | -0.269*** | |
| (0.034) | (0.040) | (0.040) | (0.037) | ||
| Age Group[31–45] | -0.131** | -0.154** | 0.512 | -0.004 | -0.059 |
| (0.044) | (0.052) | (0.585) | (0.052) | (0.051) | |
| Age Group[46–60] | -0.180*** | -0.256*** | 0.560 | -0.266*** | -0.205*** |
| (0.047) | (0.056) | (0.756) | (0.054) | (0.052) | |
| Age Group[60plus] | -0.283*** | -0.309*** | -0.193** | -0.254*** | |
| (0.053) | (0.060) | (0.069) | (0.061) | ||
| Event Occured[1] × Political Orientation[center] | 0.048 | 0.110 | 0.032 | -0.166 | -0.205 |
| (0.106) | (0.138) | (0.122) | (0.147) | (0.132) | |
| Event Occured[1] × Political Orientation[right] | 0.195* | 0.042 | -0.192 | -0.113 | -0.332** |
| (0.097) | (0.126) | (0.111) | (0.135) | (0.123) | |
| Observations | 3,215 | 2,493 | 1,342 | 2,221 | 2,638 |
| R2 | 0.034 | 0.043 | 0.019 | 0.039 | 0.033 |
| Adjusted R2 | 0.031 | 0.040 | -0.981 | 0.035 | 0.030 |
| Note: * p < 0.050; ** p < 0.010; *** p < 0.001. The reference category for Event Occured is ‘0’. The reference category for Political Orientation is ‘left’. The reference category for Gender is ‘Male’. The reference category for Age Group is ‘18–30’. Standard errors in parentheses. | |||||
2.4 Social Dimension
formula_str <- paste("pe_intolerance ~ event_occurred * pe_left_center_right","+", d_fmla)
models <- af_wave_pair_regression(df, wave_var = wave_var, wave_list = wave_list,
formula_str = formula_str, regression_type = "OLS")
models[[3]] <- mp3 # Override with panel model
coef_table <- af_coef_and_ci_table(models, coef_names)
af_coef_and_ci_plot(coef_table, xpose = TRUE, title = "Social Dimension")
notes <- af_create_regression_notes(df, models = models, display_names = display_names,
show_significance = TRUE, significance_levels = c(0.05, 0.01, 0.001))
cov_labels <- af_cov_names(df, models, display_names)
af_stargazer(models = models, cov_labels = cov_labels, notes = notes,
title = "Social Political Extremism ~ Event Occurred x Political Orientation + Demographics")| Dependent variable: | |||||
| pe_intolerance | |||||
| OLS | panel | OLS | |||
| linear | |||||
| Inland Terror | Bennet Gov. Fall | Judicial Reform | Gallant Dismissal | Oct. 7th War | |
| (1) | (2) | (3) | (4) | (5) | |
| Event Occured[1] | -0.118 | -0.051 | -0.113 | 0.273 | -0.005 |
| (0.128) | (0.167) | (0.167) | (0.197) | (0.181) | |
| Political Orientation[center] | 0.814*** | 0.687*** | 0.062 | 0.939*** | 0.679*** |
| (0.112) | (0.113) | (0.289) | (0.195) | (0.135) | |
| Political Orientation[right] | 1.559*** | 1.807*** | -0.101 | 1.738*** | 1.449*** |
| (0.103) | (0.104) | (0.314) | (0.176) | (0.126) | |
| Gender[Female] | 0.002 | 0.089 | -0.018 | 0.046 | |
| (0.050) | (0.058) | (0.065) | (0.060) | ||
| Age Group[31–45] | 0.215** | 0.186* | 0.728 | 0.136 | 0.188* |
| (0.066) | (0.077) | (0.957) | (0.084) | (0.081) | |
| Age Group[46–60] | 0.337*** | 0.257** | 2.183 | 0.341*** | 0.310*** |
| (0.070) | (0.082) | (1.237) | (0.086) | (0.083) | |
| Age Group[60plus] | 0.461*** | 0.458*** | 0.398*** | 0.386*** | |
| (0.079) | (0.088) | (0.111) | (0.097) | ||
| Event Occured[1] × Political Orientation[center] | -0.121 | 0.241 | 0.038 | -0.266 | 0.068 |
| (0.158) | (0.202) | (0.200) | (0.237) | (0.211) | |
| Event Occured[1] × Political Orientation[right] | 0.244 | -0.135 | 0.063 | -0.294 | -0.128 |
| (0.145) | (0.185) | (0.181) | (0.216) | (0.196) | |
| Observations | 3,215 | 2,493 | 1,342 | 2,221 | 2,638 |
| R2 | 0.170 | 0.175 | 0.009 | 0.118 | 0.091 |
| Adjusted R2 | 0.168 | 0.172 | -1.001 | 0.115 | 0.087 |
| Note: * p < 0.050; ** p < 0.010; *** p < 0.001. The reference category for Event Occured is ‘0’. The reference category for Political Orientation is ‘left’. The reference category for Gender is ‘Male’. The reference category for Age Group is ‘18–30’. Standard errors in parentheses. | |||||
3 Multicollinearity (VIF) Test
formula_str <- paste("pe_overall ~ Wave * pe_left_center_right","+", d_fmla)
model <- lm(formula_str, data=df, na.action = na.omit)
result <- af_vif_test(model, interactions = TRUE)
print(result$interpretation)[1] “We use VIF to test for multicollinearity. The results indicate that all adjusted GVIF values are ≤ 2.is not a concern.adjusted GVIF value: 1.01.”
| Multicollinearity Assessment (with Interactions) | |
| Predictor | Adjusted GVIF1 |
|---|---|
| Wave | 1.002 |
| pe_left_center_right | 1.002 |
| gender | 1.010 |
| age_group | 1.010 |
| 1 Values > 2 indicate problematic multicollinearity | |
4 Robustness
4.1 Impact of Occurred Events on the Dimensions
mp1 <- plm(pe_ideology ~ event_occurred , data = p_data, model = "within")
mp2 <- plm(pe_violence ~ event_occurred , data = p_data, model = "within")
mp3 <- plm(pe_intolerance ~ event_occurred , data = p_data, model = "within")
mp4 <- plm(pe_overall ~ event_occurred , data = p_data, model = "within")formula_str <- "pe_ideology ~ event_occurred"
models <- af_wave_pair_regression(df, wave_var = wave_var, wave_list = wave_list,
formula_str = formula_str, regression_type = "OLS")
models[[3]] <- mp1 # Override with panel model
notes <- af_create_regression_notes(df, models = models, display_names = display_names,
show_significance = TRUE, significance_levels = c(0.05, 0.01, 0.001))
cov_labels <- af_cov_names(df, models, display_names)
af_stargazer(models = models, cov_labels = cov_labels, notes = notes,
title = "Cognitive Political Extremism ~ Event Occurred")| Dependent variable: | |||||
| pe_ideology | |||||
| OLS | panel | OLS | |||
| linear | |||||
| Inland Terror | Bennet Gov. Fall | Judicial Reform | Gallant Dismissal | Oct. 7th War | |
| (1) | (2) | (3) | (4) | (5) | |
| Event Occured[1] | -0.210** | -0.047 | 0.035 | 0.106 | 0.121 |
| (0.068) | (0.074) | (0.054) | (0.084) | (0.073) | |
| Observations | 3,215 | 2,493 | 1,342 | 2,221 | 2,638 |
| R2 | 0.003 | 0.0002 | 0.001 | 0.001 | 0.001 |
| Adjusted R2 | 0.003 | -0.0002 | -1.000 | 0.0003 | 0.001 |
| Note: * p < 0.050; ** p < 0.010; *** p < 0.001. The reference category for Event Occured is ‘0’. Standard errors in parentheses. | |||||
formula_str <- "pe_violence ~ event_occurred"
models <- af_wave_pair_regression(df, wave_var = wave_var, wave_list = wave_list,
formula_str = formula_str, regression_type = "OLS")
models[[3]] <- mp2 # Override with panel model
notes <- af_create_regression_notes(df, models = models, display_names = display_names,
show_significance = TRUE, significance_levels = c(0.05, 0.01, 0.001))
cov_labels <- af_cov_names(df, models, display_names)
af_stargazer(models = models, cov_labels = cov_labels, notes = notes,
title = "Behavioral Political Extremism ~ Event Occurred")| Dependent variable: | |||||
| pe_violence | |||||
| OLS | panel | OLS | |||
| linear | |||||
| Inland Terror | Bennet Gov. Fall | Judicial Reform | Gallant Dismissal | Oct. 7th War | |
| (1) | (2) | (3) | (4) | (5) | |
| Event Occured[1] | 0.075* | -0.038 | 0.005 | 0.054 | 0.034 |
| (0.034) | (0.042) | (0.032) | (0.044) | (0.038) | |
| Observations | 3,215 | 2,493 | 1,342 | 2,221 | 2,638 |
| R2 | 0.001 | 0.0003 | 0.00004 | 0.001 | 0.0003 |
| Adjusted R2 | 0.001 | -0.0001 | -1.001 | 0.0002 | -0.0001 |
| Note: * p < 0.050; ** p < 0.010; *** p < 0.001. The reference category for Event Occured is ‘0’. Standard errors in parentheses. | |||||
formula_str <- "pe_intolerance ~ event_occurred"
models <- af_wave_pair_regression(df, wave_var = wave_var, wave_list = wave_list,
formula_str = formula_str, regression_type = "OLS")
models[[3]] <- mp3 # Override with panel model
notes <- af_create_regression_notes(df, models = models, display_names = display_names,
show_significance = TRUE, significance_levels = c(0.05, 0.01, 0.001))
cov_labels <- af_cov_names(df, models, display_names)
af_stargazer(models = models, cov_labels = cov_labels, notes = notes,
title = "Social Political Extremism ~ Event Occurred")| Dependent variable: | |||||
| pe_intolerance | |||||
| OLS | panel | OLS | |||
| linear | |||||
| Inland Terror | Bennet Gov. Fall | Judicial Reform | Gallant Dismissal | Oct. 7th War | |
| (1) | (2) | (3) | (4) | (5) | |
| Event Occured[1] | 0.0004 | -0.001 | -0.057 | 0.000 | 0.000 |
| (0.055) | (0.066) | (0.052) | (0.073) | (0.063) | |
| Observations | 3,215 | 2,493 | 1,342 | 2,221 | 2,638 |
| R2 | 0.000 | 0.00000 | 0.002 | 0.000 | 0.000 |
| Adjusted R2 | -0.0003 | -0.0004 | -0.998 | -0.0005 | -0.0004 |
| Note: * p < 0.050; ** p < 0.010; *** p < 0.001. The reference category for Event Occured is ‘0’. Standard errors in parentheses. | |||||
formula_str <- "pe_overall ~ event_occurred"
models <- af_wave_pair_regression(df, wave_var = wave_var, wave_list = wave_list,
formula_str = formula_str, regression_type = "OLS")
models[[3]] <- mp4 # Override with panel model
notes <- af_create_regression_notes(df, models = models, display_names = display_names,
show_significance = TRUE, significance_levels = c(0.05, 0.01, 0.001))
cov_labels <- af_cov_names(df, models, display_names)
af_stargazer(models = models, cov_labels = cov_labels, notes = notes,
title = "Overall Political Extremism ~ Event Occurred")| Dependent variable: | |||||
| pe_overall | |||||
| OLS | panel | OLS | |||
| linear | |||||
| Inland Terror | Bennet Gov. Fall | Judicial Reform | Gallant Dismissal | Oct. 7th War | |
| (1) | (2) | (3) | (4) | (5) | |
| Event Occured[1] | -0.078* | -0.013 | -0.006 | 0.050 | 0.057 |
| (0.035) | (0.039) | (0.030) | (0.044) | (0.038) | |
| Observations | 3,215 | 2,493 | 1,342 | 2,221 | 2,638 |
| R2 | 0.002 | 0.00004 | 0.0001 | 0.001 | 0.001 |
| Adjusted R2 | 0.001 | -0.0004 | -1.001 | 0.0001 | 0.001 |
| Note: * p < 0.050; ** p < 0.010; *** p < 0.001. The reference category for Event Occured is ‘0’. Standard errors in parentheses. | |||||
1.5 Social Dimension
The social dimension of wave 4 does not have any left politically-oriented people in the extremism tail, i.e., i_sel == 0 for all respondents. This means we have perfect separation, and the regression will provide wrong results (inf). To resolve this, the interaction term was removed from pairs 3 & 4.
The logistic regression analyses examining social extremism (intolerance dimension) reveal the most pronounced baseline political orientation effects across all dimensions, with consistently massive differences between left-wing and other political groups, though event effects remain largely non-significant. Center-wing respondents demonstrated dramatically higher baseline social extremism compared to left-wing respondents across all events (OR ranging from 1.831 to 12.048, with four of five reaching statistical significance), while right-wing respondents showed even more extreme baseline differences (OR ranging from 9.447 to 32.068, all p < 0.001). Notably, the Judicial Reform period exhibited the most extreme baseline disparities, with center-wing respondents showing 12-fold higher odds (OR = 12.048, p < 0.05) and right-wing respondents showing over 32-fold higher odds (OR = 32.068, p < 0.001) of social extremism compared to left-wing respondents. Despite these substantial baseline differences, none of the five events produced significant main effects on social extremism, and crucially, none of the interaction terms reached statistical significance, indicating that political orientation differences in social extremism remain stable regardless of destabilizing events. Demographic controls showed minimal effects, with only marginal age-related patterns and no consistent gender effects, contrasting sharply with the other extremism dimensions.
These social extremism findings provide the strongest support for Hypothesis 1, demonstrating that different extremism dimensions respond fundamentally differently to socio-political events, with social intolerance showing remarkable stability despite massive baseline political differences.
Hypothesis 2 receives no support in this dimension, as political orientation failed to moderate any event effects despite the substantial baseline differences between political groups.
Hypothesis 3 receives no support across all expectations (3.1-3.5), as none of the five destabilizing events produced significant changes in social extremism within any political orientation group, suggesting that intolerance levels remain relatively fixed even during periods of significant political upheaval. This dimensional stability contrasts markedly with the dynamic responses observed in cognitive and behavioral extremism, highlighting the distinct nature of social intolerance as potentially more trait-like than responsive to external political stimuli.