FictionEro - Affective Impact of Fiction

Data Preparation

library(tidyverse)
library(easystats)
library(patchwork)
library(ggside)
library(glmmTMB)

options(mc.cores = parallel::detectCores(),
        brms.backend = "cmdstanr",
        width = 300)

dfsub <- read.csv("../data/data_participants.csv")
df <- read.csv("../data/data.csv") |> 
  right_join(
    select(dfsub, Participant, Sample, Language, Mobile, starts_with("Feedback_"), 
           BAIT_Visual, BAIT_Text, AI_Knowledge, SexualActivity, 
           GAAIS_Negative, GAAIS_Positive, Porn, COPS_Frequency_2,
           Screen_Size,
           -Feedback_Comments),
    by = "Participant"
    ) |> 
  datawizard::rescale(select=c("Valence"), range=c(-1, 1), to=c(0, 1)) |> 
  mutate(Condition = fct_relevel(Condition, "Photograph", "AI-Generated"),
         Relevance =  fct_relevel(Relevance, "Relevant", "Irrelevant", "Non-erotic"),
         Sex =  fct_relevel(Sex, "Male", "Female"),
         PornFrequency = as.numeric(as.factor(COPS_Frequency_2)),
         SexualActivity_num = as.numeric(as.factor(SexualActivity)),
         ConditionBelief = case_when(
           Condition == "Photograph" & Realness > 0.5 ~ "True",
           Condition == "AI-Generated" & Realness < 0.5 ~ "True",
           .default = "False"
         )) |> 
  mutate(across(starts_with("Feedback_"), function(x) {fct_relevel(x, "False", "True")}))

results_table <- function(model, effects="fixed", filter=NULL) {
  if("marginaleffects" %in% class(model)) {
    model |> 
      parameters::parameters() |> 
      as.data.frame() |> 
      select(-Parameter, -SE, -S, z=Statistic) |> 
      insight::format_table() |> 
      parameters::display()
  } else {
    display(parameters::parameters(model, effects=effects, keep=filter))
  }
}

Manipulation Check

This section is about selecting the best model and making a manipulation check to make sure that the emotional appraisal were sensitive to the main manipulation (Neutral vs. Emotional).

Arousal

The Arousal data is composed of a mixture of extreme responses (0 and 1) and non-extreme responses between 0 and 1, lending itself to ordered beta model (Kubinec, 2022), designed explicitly for bounded data such as survey slider scales. It is a more parsimonious evolution of the zero-and-one inflated beta (ZOIB) regression model (see this tutorial) that jointly models the probability of extreme responses, the probability of 1 vs. 0, and the mean and spread of the rest of the responses.

arousal <- df |> 
  mutate(Arousal_Extreme = ifelse(Arousal %in% c(0, 1), "Extreme", "Not extreme")) |>
  ggplot(aes(x=Arousal, fill=Relevance)) +
  geom_histogram(aes(color=Arousal_Extreme), position = "identity", bins=80) +
  facet_grid(Relevance~Sex, scales="free_y") +
  scale_color_manual(values=c("red", "white"), guide="none") +
  scale_fill_manual(values=c("Relevant"="#E91E63", "Irrelevant"="#FF9800", "Non-erotic"="#03A9F4"), guide="none") +
  scale_y_sqrt(expand = c(0, 0)) +
  labs(title = "Arousal") +
  theme_abyss() +
  theme(plot.title = element_text(hjust = 0.5, face="bold"))

arousal

# Get a sense of Random Factors
m1 <- glmmTMB::glmmTMB(Arousal ~ Sex / Relevance + (1|Participant) + (1|Item), data=df)
m2 <- glmmTMB::glmmTMB(Arousal ~ Sex / Relevance + (1+Relevance|Participant) + (1|Item), data=df)
m3 <- glmmTMB::glmmTMB(Arousal ~ Sex / Relevance + (1|Participant) + (1+Sex|Item), data=df)
m4 <- glmmTMB::glmmTMB(Arousal ~ Sex / Relevance + (1|Participant) + (1+Relevance|Item), data=df)
m5 <- glmmTMB::glmmTMB(Arousal ~ Sex / Relevance + (1|Participant) + (1+Sex / Relevance|Item), data=df)
m6 <- glmmTMB::glmmTMB(Arousal ~ Sex / Relevance + (1+Relevance|Participant) + (1+Sex|Item), data=df)
m7 <- glmmTMB::glmmTMB(Arousal ~ Sex / Relevance + (1+Relevance|Participant) + (1+Relevance|Item), data=df)


rbind(
  mutate(performance::icc(m1, by_group=TRUE), Model = "m1"),
  data.frame(Group = "Conditional R2", ICC = performance::r2(m1)$R2_conditional, Model = "m1"),
  mutate(performance::icc(m2, by_group=TRUE), Model = "m2"),
  data.frame(Group = "Conditional R2", ICC = performance::r2(m2)$R2_conditional, Model = "m2"),
  mutate(performance::icc(m3, by_group=TRUE), Model = "m3"),
  data.frame(Group = "Conditional R2", ICC = performance::r2(m3)$R2_conditional, Model = "m3"),
  mutate(performance::icc(m4, by_group=TRUE), Model = "m4"),
  data.frame(Group = "Conditional R2", ICC = performance::r2(m4)$R2_conditional, Model = "m4"),
  mutate(performance::icc(m5, by_group=TRUE), Model = "m5"),
  data.frame(Group = "Conditional R2", ICC = performance::r2(m5)$R2_conditional, Model = "m5"),
  mutate(performance::icc(m6, by_group=TRUE), Model = "m6"),
  data.frame(Group = "Conditional R2", ICC = performance::r2(m6)$R2_conditional, Model = "m6"),
  mutate(performance::icc(m7, by_group=TRUE), Model = "m7"),
  data.frame(Group = "Conditional R2", ICC = performance::r2(m7)$R2_conditional, Model = "m7")
) |>
  ggplot(aes(x = Model, y = ICC, color=Group)) +
  geom_line(aes(group=Group))

test_bf(m6, m1, m2, m3, m4, m5, m7)
compare_performance(m2, m7)

# It seems like the full model, with the effect of relevance on participants and items, is the best.

This model evaluates the effects of Sex and Relevance on Arousal, accounting for random variability due to participants and items (i.e., random effects).

The analysis revealed that on average women rate relevant stimuli lower in arousal compared to men. Additionally, both genders tend to rate irrelevant and non-erotic stimuli lower in arousal then relevant stimuli, with males showing a stronger reduction for irrelevant stimuli and a comparable one for non-erotic stimuli. Furthermore, females rate irrelevant stimuli higher in arousal than males, whereas both genders rate non-erotic images similarly in terms of arousal.

There is notable variability in arousal ratings across participants, with a greater diversity in responses to irrelevant stimuli compared to relevant ones.There is a strong positive correlation between changes in arousal ratings for irrelevant stimuli and non-erotic stimuli. This suggests that participants who exhibit significant increases or decreases in arousal when exposed to irrelevant stimuli tend to exhibit similar changes when exposed to non-erotic stimuli, indicating consistent trends in arousal levels. There is significant variability in baseline arousal ratings across different items, indicating that different items elicit varying baseline levels of arousal.

m_a <- glmmTMB::glmmTMB(Arousal ~ Sex / Relevance + (Relevance | Participant) + (1|Item),
                      data=df, family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))
                      # control = glmmTMB::glmmTMBControl(optimizer=stats::optim, # nlminb, optim
                      #                                   optArgs = list(method = "BFGS"),  #Nelder-Mead, BFGS, CG
                      #                                   parallel = 8))

results_table(m_a)

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
(Intercept)	0.52	0.09	(0.36, 0.69)	6.16	< .001
Sex (Female)	-1.21	0.13	(-1.46, -0.96)	-9.64	< .001
Sex (Male) × RelevanceIrrelevant	-3.62	0.13	(-3.88, -3.36)	-27.06	< .001
Sex (Female) × RelevanceIrrelevant	-0.28	0.15	(-0.56, 7.41e-03)	-1.91	0.056
Sex (Male) × RelevanceNon-erotic	-3.67	0.13	(-3.92, -3.42)	-28.94	< .001
Sex (Female) × RelevanceNon-erotic	-2.29	0.14	(-2.56, -2.02)	-16.39	< .001

estimate_relation(m_a) |> 
  ggplot(aes(x=Relevance, y=Predicted)) +
  geom_pointrange(aes(ymin=CI_low, ymax=CI_high, color=Relevance), position=position_dodge(width=0.5)) +
  scale_color_manual(values=c("Relevant"="#E91E63", "Irrelevant"="#FF9800", "Non-erotic"="#03A9F4"), guide="none") +
  labs(y="Arousal") +
  facet_wrap(~Sex) +
  theme_abyss()

Enticement

The same type of model seems warranted for enticement.

enticiment <- df |> 
  mutate(Enticement_Extreme = ifelse(Enticement %in% c(0, 1), "Extreme", "Not extreme")) |>
  ggplot(aes(x=Enticement, fill=Relevance)) +
  geom_histogram(aes(color=Enticement_Extreme), position = "identity", bins=80) +
  facet_grid(Relevance~Sex, scales="free_y") +
  scale_color_manual(values=c("red", "white"), guide="none") +
  scale_fill_manual(values=c("Relevant"="#E91E63", "Irrelevant"="#FF9800", "Non-erotic"="#03A9F4"), guide="none") +
  scale_y_sqrt(expand = c(0, 0)) +
  labs(title = "Enticement") +
  theme_abyss() +
  theme(plot.title = element_text(hjust = 0.5, face="bold"))

enticiment

This model evaluates the effects of Sex and Relevance on Enticement ratings, accounting for random variability due to participants and items (i.e., random effects).

In general, women rate relevant images lower in enticement than males. Males rate irrelevant images significantly less enticing than relevant ones, while females do not show a significant difference in enticement ratings between irrelevant and relevant images. Both males and females rate non-erotic images as the least enticing, with a stronger effect for males.

Random effects show considerable participant-to-participant variation in baseline enticement ratings, especially for irrelevant and non-erotic stimuli, compared to relevant stimuli.Participants with higher initial ratings for relevant stimuli tend to show less change in ratings for irrelevant and non-erotic stimuli, though the effect size varies.A notable correlation exists between changes in enticement ratings for irrelevant and non-erotic stimuli, indicating consistent responsiveness across different stimulus types. Lastly, the analysis indicates that different items elicit varying baseline levels of enticement.

m_e <- glmmTMB::glmmTMB(Enticement ~ Sex / Relevance + (Relevance | Participant) + (1|Item),
                      data=df, family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

results_table(m_e)

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
(Intercept)	0.79	0.08	(0.64, 0.94)	10.14	< .001
Sex (Female)	-0.92	0.11	(-1.15, -0.70)	-8.10	< .001
Sex (Male) × RelevanceIrrelevant	-2.19	0.12	(-2.43, -1.96)	-18.23	< .001
Sex (Female) × RelevanceIrrelevant	0.04	0.13	(-0.22, 0.30)	0.31	0.760
Sex (Male) × RelevanceNon-erotic	-2.44	0.12	(-2.67, -2.21)	-20.74	< .001
Sex (Female) × RelevanceNon-erotic	-1.17	0.13	(-1.43, -0.92)	-9.02	< .001

estimate_relation(m_e) |> 
  ggplot(aes(x=Relevance, y=Predicted)) +
  geom_pointrange(aes(ymin=CI_low, ymax=CI_high, color=Relevance), position=position_dodge(width=0.5)) +
  scale_color_manual(values=c("Relevant"="#E91E63", "Irrelevant"="#FF9800", "Non-erotic"="#03A9F4"), guide="none") +
  labs(y="Enticement") +
  facet_wrap(~Sex) +
  theme_abyss()

Valence

The same type of model seems warranted for valence.

valence <- df |> 
  mutate(Valence_Extreme = ifelse(Valence %in% c(0, 1), "Extreme", "Not extreme")) |>
  ggplot(aes(x=Valence, fill=Relevance)) +
  geom_histogram(aes(color=Valence_Extreme), position = "identity", bins=80) +
  facet_grid(Relevance~Sex, scales="free_y") +
  scale_color_manual(values=c("red", "white"), guide="none") +
  scale_fill_manual(values=c("Relevant"="#E91E63", "Irrelevant"="#FF9800", "Non-erotic"="#03A9F4"), guide="none") +
  scale_y_sqrt(expand = c(0, 0)) +
  labs(title = "Valence") +
  theme_abyss() +
  theme(plot.title = element_text(hjust = 0.5, face="bold"))

valence

This model is evaluating the effects of Sex and Relevance on Valence ratings, accounting for random variability due to participants and items (i.e., random effects).

The analysis reveals that women rate relevant stimuli significantly lower in valence compared to men. Male participants consistently rate relevant stimuli higher in valence compared to their ratings of irrelevant stimuli, while females do not show significant differences between these stimuli. Both genders rate non-erotic stimuli lower in valence than relevant stimuli and higher than irrelevant ones, with males showing a more pronounced difference.

Participants show diverse baseline valence ratings, especially for irrelevant and non-erotic stimuli, with greater variability observed compared to relevant stimuli. Participants who rate relevant stimuli higher in valence exhibit less change in valence ratings for irrelevant and non-erotic stimuli, indicating consistent individual responses across different types of stimuli.There is a significant correlation between changes in valence ratings for irrelevant and non-erotic stimuli, indicating consistent responsiveness across different stimulus types. Different items elicit varying baseline levels of valence..

m_v <- glmmTMB::glmmTMB(Valence ~ Sex / Relevance + (Relevance | Participant) + (1|Item),
                      data=df, family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

results_table(m_v)

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
(Intercept)	0.86	0.08	(0.69, 1.02)	10.13	< .001
Sex (Female)	-0.96	0.12	(-1.20, -0.72)	-7.77	< .001
Sex (Male) × RelevanceIrrelevant	-1.64	0.12	(-1.88, -1.41)	-13.49	< .001
Sex (Female) × RelevanceIrrelevant	0.04	0.13	(-0.21, 0.30)	0.33	0.741
Sex (Male) × RelevanceNon-erotic	-0.66	0.12	(-0.89, -0.43)	-5.61	< .001
Sex (Female) × RelevanceNon-erotic	0.47	0.12	(0.23, 0.71)	3.85	< .001

estimate_relation(m_v) |> 
  ggplot(aes(x=Relevance, y=Predicted)) +
  geom_pointrange(aes(ymin=CI_low, ymax=CI_high, color=Relevance), position=position_dodge(width=0.5)) +
  scale_color_manual(values=c("Relevant"="#E91E63", "Irrelevant"="#FF9800", "Non-erotic"="#03A9F4"), guide="none") +
  labs(y="Valence") +
  facet_wrap(~Sex) +
  theme_abyss()

Arousal

plot_descriptive <- function(df, target="Arousal") {
  # By item
  order <- df |>
    mutate(.Item_Order = paste0(Relevance, "_", Sex, "_", Item)) |>
    summarize(mean = mean(.data[[target]], na.rm=TRUE), .by=c("Condition", ".Item_Order")) |> 
    arrange(.Item_Order, desc(Condition)) |> 
    mutate(diff = sign(mean - lag(mean)),
           Relevance = str_extract(.Item_Order, "^[^_]+"),
           .Item_Order = str_extract(.Item_Order, "(?<=_).+")) |>
    filter(Condition == "Photograph") |> 
    arrange(Relevance, diff, mean)
    
  p1 <- df |>
    mutate(.Item_Order = paste0(Sex, "_", Item)) |>
    group_by(.Item_Order, Condition) |>
    ggdist::mean_qi(.data[[target]], .width=0.95) |>
    mutate(Sex = fct_rev(str_extract(.Item_Order, "^[^_]+"))) |>
    left_join(order[c(".Item_Order", "diff")], by=".Item_Order") |>
    mutate(.Item_Order = fct_relevel(.Item_Order, order$.Item_Order)) |>
    # mutate(.average = mean(.data[[target]]), .by="Item") |>
    # mutate(Item = fct_reorder(Item, .average)) |>
    ggplot(aes(y = .Item_Order, x=.data[[target]])) +
    geom_line(aes(color=as.factor(diff)), linewidth=1) +
    geom_point(aes(color=Condition), size=2) +
      
    # ggdist::stat_slabinterval(aes(fill=Condition, color=Condition),
    #                           .width=0.95, normalize="xy", slab_alpha=0.5) +
    # ggdist::stat_histinterval(aes(fill=Condition, color=Condition),
    #                           .width=0.95, normalize="xy", slab_alpha=0.5) +
    scale_color_manual(values=c("AI-Generated"="#2196F3", "Photograph"="#F44336", "1"="#F44336", "-1"="#2196F3"),
                       breaks = c('AI-Generated', 'Photograph')) +
    # scale_fill_manual(values=c("AI-Generated"="#2196F3", "Photograph"="#F44336")) +
    scale_y_discrete(labels = function(x) str_extract(x, "(?<=_).+")) +
    theme_minimal() +
    theme(panel.grid.minor.x = element_blank(),
          # panel.grid.major.x = element_blank(),
          panel.border = element_blank(),
          axis.line.x = element_blank()) +
    labs(title = paste0(target, " by Item"), color="Condition", y="Items") +
    facet_wrap(~Sex, scales="free_y")

  # By participant
  p2 <- df |>
    summarize(.average = mean(.data[[target]]),
              .by=c("Sex", "Participant", "Condition")) |>
    pivot_wider(names_from = Condition, values_from = .average) |>
    mutate(.diff = Photograph - `AI-Generated`) |>
    mutate(Participant = fct_reorder(Participant, .diff)) |>
    ggplot(aes(x = .diff, y=Participant)) +
    geom_bar(aes(fill=.diff), stat = "identity")  +
    scale_fill_gradient2(guide="none", low="red", high="red", mid="blue") +
    labs(title = paste0(target, " by Participant")) +
    facet_wrap(~Sex, ncol=2, scales="free_y") +
    theme_minimal() +
    theme(panel.grid.major.y = element_blank(),
          panel.border = element_blank(),
          axis.line.x = element_blank(),
          axis.text.y = element_blank()) +
    labs(title = paste0(target, " by Participant"), x="Difference (Photograph - AI-Generated)", y="Participants")

  p1 / p2 + plot_layout(heights = c(2, 1))
}

arousal_desc <- plot_descriptive(df, "Arousal")
arousal_desc

Whole Sample

This model examines the effects of Sex, Relevance, and Condition on Arousal, accounting for random variability due to participants and items (i.e., random effects).

Women rate relevant (photograph) stimuli significantly lower in arousal compared to men. Males rate irrelevant (photograph), and non-erotic (photograph), stimuli significantly lower in arousal compared to relevant stimuli. Females also rate non-erotic (photograph) stimuli lower in arousal compared to relevant stimuli. Lastly, males rate AI-Generated images significantly lower in arousal compared to Photograph images.

The analysis also examined the average marginal effect of the variable “Condition” (comparing AI-Generated to Photograph) separately for males and females. For males, the arousal is on average lower for AI-Generated images compared to Photograph images. For females, the arousal is on average higher for AI-Generated images compared to Photograph images; however, this effect is not significant.

m_arousal <- glmmTMB::glmmTMB(Arousal ~ Sex / Relevance / Condition + 
                                (Relevance | Participant) + (1|Item),
                      data=df, family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

results_table(m_arousal)

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
(Intercept)	0.57	0.09	(0.40, 0.73)	6.61	< .001
Sex (Female)	-1.27	0.13	(-1.52, -1.02)	-10.03	< .001
Sex (Male) × RelevanceIrrelevant	-3.67	0.13	(-3.93, -3.40)	-27.22	< .001
Sex (Female) × RelevanceIrrelevant	-0.26	0.15	(-0.55, 0.03)	-1.76	0.079
Sex (Male) × RelevanceNon-erotic	-3.68	0.13	(-3.93, -3.43)	-28.75	< .001
Sex (Female) × RelevanceNon-erotic	-2.30	0.14	(-2.58, -2.03)	-16.25	< .001
Sex (Male) × RelevanceRelevant × ConditionAI-Generated	-0.08	0.02	(-0.12, -0.05)	-4.74	< .001
Sex (Female) × RelevanceRelevant × ConditionAI-Generated	0.04	0.03	(-0.02, 0.09)	1.36	0.173
Sex (Male) × RelevanceIrrelevant × ConditionAI-Generated	0.01	0.03	(-0.04, 0.06)	0.42	0.677
Sex (Female) × RelevanceIrrelevant × ConditionAI-Generated	-4.65e-03	0.03	(-0.06, 0.05)	-0.18	0.859
Sex (Male) × RelevanceNon-erotic × ConditionAI-Generated	-0.07	0.03	(-0.13, -5.91e-03)	-2.15	0.031
Sex (Female) × RelevanceNon-erotic × ConditionAI-Generated	0.06	0.04	(-0.02, 0.14)	1.53	0.127

marginaleffects::avg_slopes(m_arousal, variables="Condition", by="Sex", re.form=NA) |> 
  results_table()

Comparison	Sex	Coefficient	z	p	95% CI
AI-Generated - Photograph	Male	-7.21e-03	-4.86	< .001	[-0.01, 0.00]
AI-Generated - Photograph	Female	3.25e-03	1.22	0.224	[ 0.00, 0.01]

Subsamples

Sample

When examining the random slopes, it was observed that only males significantly rated arousal lower for AI-generated images compared to photographs. Further analysis of the sample subgroups revealed that this effect was predominantly driven by males from the general population, who significantly contributed to the lower arousal ratings for AI-generated images compared to photographs.

Note. Model converged when adding Relevance and Condition as random slopes of each stimuli.

m <- glmmTMB::glmmTMB(Arousal ~ Sample / Sex / Relevance / Condition + (Relevance | Participant) + ((Relevance / Condition)|Item),
                      data=df, 
                      family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; non-positive-definite Hessian matrix. See vignette('troubleshooting')

results_table(m, filter="Condition")

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
Sample (General Population) × SexMale × RelevanceRelevant × ConditionAI-Generated	-0.09	0.02	(-0.13, -0.05)	-4.46	< .001
Sample (Students (Incentivized)) × SexMale × RelevanceRelevant × ConditionAI-Generated	-0.07	0.06	(-0.19, 0.05)	-1.20	0.228
Sample (Students (Non-incentivized)) × SexMale × RelevanceRelevant × ConditionAI-Generated	-0.06	0.05	(-0.17, 0.05)	-1.08	0.279
Sample (General Population) × SexFemale × RelevanceRelevant × ConditionAI-Generated	0.05	0.05	(-0.05, 0.14)	1.02	0.310
Sample (Students (Incentivized)) × SexFemale × RelevanceRelevant × ConditionAI-Generated	-9.43e-03	0.04	(-0.10, 0.08)	-0.21	0.831
Sample (Students (Non-incentivized)) × SexFemale × RelevanceRelevant × ConditionAI-Generated	0.07	0.05	(-0.02, 0.16)	1.61	0.108
Sample (General Population) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	0.02	0.03	(-0.04, 0.08)	0.58	0.561
Sample (Students (Incentivized)) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	-0.08	0.11	(-0.31, 0.14)	-0.74	0.462
Sample (Students (Non-incentivized)) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	0.06	0.09	(-0.13, 0.24)	0.61	0.544
Sample (General Population) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	0.04	0.05	(-0.06, 0.13)	0.76	0.446
Sample (Students (Incentivized)) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	0.01	0.04	(-0.07, 0.10)	0.24	0.809
Sample (Students (Non-incentivized)) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	-0.06	0.05	(-0.15, 0.03)	-1.33	0.184
Sample (General Population) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	-0.06	0.04	(-0.13, 0.02)	-1.42	0.155
Sample (Students (Incentivized)) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	-9.32e-03	0.13	(-0.26, 0.24)	-0.07	0.941
Sample (Students (Non-incentivized)) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	-0.18	0.11	(-0.39, 0.02)	-1.73	0.084
Sample (General Population) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	0.07	0.08	(-0.08, 0.22)	0.86	0.390
Sample (Students (Incentivized)) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	0.04	0.07	(-0.10, 0.19)	0.57	0.570
Sample (Students (Non-incentivized)) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	0.08	0.07	(-0.05, 0.22)	1.24	0.214

Mobile

We investigated the influence of using a mobile device (mobile = TRUE) versus other devices on arousal ratings, considering the effects of condition and relevance.

Males rated relevant AI-generated images significantly lower in arousal compared to relevant photographs, irrespective of whether they used a mobile device or not. However, this effect was more pronounced among male participants who used a mobile device.

We also explored whether the interaction between condition (AI-Generated vs. Photograph) and screen size affected arousal levels. No statistically significant results were found for this interaction.

Note. Model converged when adding Relevance and Condition as random slopes of each stimuli.

m <- glmmTMB::glmmTMB(Arousal ~ Mobile / Sex / Relevance / Condition + (Relevance | Participant) + ((Relevance / Condition)|Item),
                      data=df, 
                      family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; non-positive-definite Hessian matrix. See vignette('troubleshooting')

results_table(m, filter="Condition")

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
Mobile (False) × SexMale × RelevanceRelevant × ConditionAI-Generated	-0.07	0.02	(-0.11, -0.02)	-2.84	0.005
Mobile (True) × SexMale × RelevanceRelevant × ConditionAI-Generated	-0.13	0.03	(-0.19, -0.07)	-4.22	< .001
Mobile (False) × SexFemale × RelevanceRelevant × ConditionAI-Generated	0.05	0.03	(-0.01, 0.11)	1.51	0.131
Mobile (True) × SexFemale × RelevanceRelevant × ConditionAI-Generated	6.81e-03	0.05	(-0.09, 0.11)	0.13	0.894
Mobile (False) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	9.14e-03	0.04	(-0.07, 0.08)	0.24	0.810
Mobile (True) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	0.02	0.04	(-0.06, 0.10)	0.55	0.580
Mobile (False) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	-0.01	0.03	(-0.08, 0.05)	-0.45	0.651
Mobile (True) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	0.02	0.05	(-0.08, 0.12)	0.41	0.678
Mobile (False) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	-0.06	0.05	(-0.16, 0.03)	-1.29	0.197
Mobile (True) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	-0.06	0.05	(-0.16, 0.03)	-1.36	0.175
Mobile (False) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	0.04	0.05	(-0.06, 0.15)	0.75	0.452
Mobile (True) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	0.11	0.07	(-0.03, 0.25)	1.61	0.108

m <- glmmTMB::glmmTMB(Arousal ~ Sex / Relevance / (Condition * Screen_Size) + (Relevance | Participant) + ((Relevance/Condition)|Item),
                      data=standardize(df, select="Screen_Size"), 
                      family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; non-positive-definite Hessian matrix. See vignette('troubleshooting')

results_table(m, filter="Screen")

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
Sex (Male) × RelevanceRelevant × Screen Size	0.01	0.04	(-0.07, 0.09)	0.29	0.772
Sex (Female) × RelevanceRelevant × Screen Size	0.09	0.07	(-0.05, 0.24)	1.22	0.222
Sex (Male) × RelevanceIrrelevant × Screen Size	0.12	0.08	(-0.05, 0.28)	1.40	0.160
Sex (Female) × RelevanceIrrelevant × Screen Size	0.08	0.15	(-0.21, 0.36)	0.52	0.603
Sex (Male) × RelevanceNon-erotic × Screen Size	0.04	0.07	(-0.10, 0.18)	0.50	0.614
Sex (Female) × RelevanceNon-erotic × Screen Size	0.16	0.13	(-0.09, 0.42)	1.25	0.212
Sex (Male) × RelevanceRelevant × ConditionAI-Generated × Screen Size	-0.02	0.02	(-0.05, 0.01)	-1.23	0.220
Sex (Female) × RelevanceRelevant × ConditionAI-Generated × Screen Size	-0.04	0.03	(-0.11, 0.03)	-1.14	0.252
Sex (Male) × RelevanceIrrelevant × ConditionAI-Generated × Screen Size	0.03	0.03	(-0.02, 0.08)	1.10	0.272
Sex (Female) × RelevanceIrrelevant × ConditionAI-Generated × Screen Size	0.05	0.03	(-0.02, 0.12)	1.44	0.151
Sex (Male) × RelevanceNon-erotic × ConditionAI-Generated × Screen Size	-0.01	0.03	(-0.07, 0.05)	-0.35	0.726
Sex (Female) × RelevanceNon-erotic × ConditionAI-Generated × Screen Size	0.08	0.05	(-0.03, 0.18)	1.46	0.144

Moderators

This model explores whether various factors such as attitudes towards AI, knowledge about AI, porn usage, frequency of porn consumption, and sexual activity moderate the relationship between sex, image relevance (specifically AI-generated versus photographs), and arousal ratings.

The analysis revealed a significant interaction effect only for porn usage among females. Specifically, females with higher levels of porn usage rated AI-generated images significantly lower in arousal compared to photographs. No other moderators showed statistically significant effects on arousal ratings across the interactions of sex, condition (AI-Generated vs. Photograph), and the respective moderators.

We also examined whether feedback moderates the relationship between sex, condition, and relevance on arousal, we found the following significant effects:

Findings suggest that feedback about the arousal potential of AI-generated images can significantly moderate arousal ratings, with females being more positively influenced by feedback that AI images are more arousing, and males being more negatively influenced by feedback that AI images are less arousing.

test_moderator <- function(df, outcome="Arousal", moderator="BAIT_Visual") {
  f <- paste0(outcome, " ~ Sex / (Condition * ", moderator, ") + (Condition | Participant) + (1|Item)")
  r <- glmmTMB::glmmTMB(as.formula(f),
                      data=filter(df, Relevance == "Relevant"), 
                      family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8)) |>
    parameters::parameters() |> 
    as.data.frame() |> 
    select(-df_error, -Effects, -Group, -Component, -SE)
  r$Moderator <- moderator
  r$Outcome <- outcome
  insight::format_table(r[c(7, 8), ])
}


moderator_results <- data.frame()
for(m in c("BAIT_Visual", "BAIT_Text", "AI_Knowledge",
           "GAAIS_Positive", "GAAIS_Negative",
           "Porn", "PornFrequency", "SexualActivity_num")) {
  moderator_results <- rbind(moderator_results,
                             test_moderator(df, "Arousal", m))
}

display(moderator_results)

Parameter	Coefficient	95% CI	z	p	Moderator	Outcome
Sex [Male] × ConditionAI-Generated × BAIT Visual	-0.11	[-0.34, 0.11]	-0.97	0.334	BAIT_Visual	Arousal
Sex [Female] × ConditionAI-Generated × BAIT Visual	0.28	[-0.11, 0.66]	1.39	0.164	BAIT_Visual	Arousal
Sex [Male] × ConditionAI-Generated × BAIT Text	0.02	[-0.14, 0.18]	0.26	0.792	BAIT_Text	Arousal
Sex [Female] × ConditionAI-Generated × BAIT Text	0.02	[-0.25, 0.29]	0.16	0.876	BAIT_Text	Arousal
Sex [Male] × ConditionAI-Generated × AI Knowledge	0.03	[-0.01, 0.06]	1.47	0.142	AI_Knowledge	Arousal
Sex [Female] × ConditionAI-Generated × AI Knowledge	9.70e-03	[-0.03, 0.05]	0.44	0.658	AI_Knowledge	Arousal
Sex [Male] × ConditionAI-Generated × GAAIS Positive	0.07	[-0.11, 0.26]	0.79	0.430	GAAIS_Positive	Arousal
Sex [Female] × ConditionAI-Generated × GAAIS Positive	0.21	[-0.08, 0.51]	1.42	0.157	GAAIS_Positive	Arousal
Sex [Male] × ConditionAI-Generated × GAAIS Negative	-0.04	[-0.20, 0.12]	-0.49	0.621	GAAIS_Negative	Arousal
Sex [Female] × ConditionAI-Generated × GAAIS Negative	0.14	[-0.14, 0.42]	0.97	0.331	GAAIS_Negative	Arousal
Sex [Male] × ConditionAI-Generated × Porn	-0.04	[-0.09, 0.02]	-1.39	0.164	Porn	Arousal
Sex [Female] × ConditionAI-Generated × Porn	-0.08	[-0.16, 0.00]	-2.00	0.046	Porn	Arousal
Sex [Male] × ConditionAI-Generated × PornFrequency	-0.01	[-0.04, 0.01]	-0.92	0.358	PornFrequency	Arousal
Sex [Female] × ConditionAI-Generated × PornFrequency	-0.03	[-0.07, 0.01]	-1.48	0.138	PornFrequency	Arousal
Sex [Male] × ConditionAI-Generated × SexualActivity num	2.14e-03	[-0.03, 0.04]	0.12	0.907	SexualActivity_num	Arousal
Sex [Female] × ConditionAI-Generated × SexualActivity num	-5.65e-03	[-0.05, 0.04]	-0.26	0.791	SexualActivity_num	Arousal

feedback_results <- data.frame()
for(m in c("Feedback_CouldDiscriminate", "Feedback_CouldNotDiscriminate",
           "Feedback_AIMoreArousing", "Feedback_AILessArousing", 
           "Feedback_LabelsIncorrect", "Feedback_LabelsReversed")) {
  feedback_results <- rbind(feedback_results,
                            test_moderator(df, "Arousal", m))
}
display(feedback_results)

Parameter	Coefficient	95% CI	z	p	Moderator	Outcome
Sex [Male] × ConditionAI-Generated × Feedback CouldDiscriminateTrue	0.07	[-0.07, 0.22]	0.98	0.328	Feedback_CouldDiscriminate	Arousal
Sex [Female] × ConditionAI-Generated × Feedback CouldDiscriminateTrue	-0.15	[-0.39, 0.09]	-1.24	0.217	Feedback_CouldDiscriminate	Arousal
Sex [Male] × ConditionAI-Generated × Feedback CouldNotDiscriminateTrue	9.14e-03	[-0.07, 0.09]	0.23	0.815	Feedback_CouldNotDiscriminate	Arousal
Sex [Female] × ConditionAI-Generated × Feedback CouldNotDiscriminateTrue	0.03	[-0.08, 0.14]	0.48	0.633	Feedback_CouldNotDiscriminate	Arousal
Sex [Male] × ConditionAI-Generated × Feedback AIMoreArousingTrue	5.61e-03	[-0.13, 0.14]	0.08	0.934	Feedback_AIMoreArousing	Arousal
Sex [Female] × ConditionAI-Generated × Feedback AIMoreArousingTrue	0.16	[ 0.01, 0.30]	2.03	0.042	Feedback_AIMoreArousing	Arousal
Sex [Male] × ConditionAI-Generated × Feedback AILessArousingTrue	-0.16	[-0.28, -0.04]	-2.61	0.009	Feedback_AILessArousing	Arousal
Sex [Female] × ConditionAI-Generated × Feedback AILessArousingTrue	0.18	[ 0.00, 0.37]	1.91	0.056	Feedback_AILessArousing	Arousal
Sex [Male] × ConditionAI-Generated × Feedback LabelsIncorrectTrue	-0.03	[-0.10, 0.05]	-0.70	0.485	Feedback_LabelsIncorrect	Arousal
Sex [Female] × ConditionAI-Generated × Feedback LabelsIncorrectTrue	-0.06	[-0.18, 0.06]	-0.98	0.329	Feedback_LabelsIncorrect	Arousal
Sex [Male] × ConditionAI-Generated × Feedback LabelsReversedTrue	-0.11	[-0.28, 0.07]	-1.20	0.230	Feedback_LabelsReversed	Arousal
Sex [Female] × ConditionAI-Generated × Feedback LabelsReversedTrue	7.56e-03	[-0.20, 0.21]	0.07	0.943	Feedback_LabelsReversed	Arousal

Believed

  m <- glmmTMB::glmmTMB(Arousal ~ Sex / Relevance / Condition + (Relevance | Participant) + ((Relevance/Condition)|Item),
                        data=filter(df, ConditionBelief == "True"), 
                        family=glmmTMB::ordbeta(),
                        control = glmmTMB::glmmTMBControl(parallel = 8))

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; non-positive-definite Hessian matrix. See vignette('troubleshooting')

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; singular convergence (7). See vignette('troubleshooting'), help('diagnose')

  results_table(m)

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
(Intercept)	0.62	0.10	(0.43, 0.81)	6.33	< .001
Sex (Female)	-1.20	0.14	(-1.48, -0.92)	-8.34	< .001
Sex (Male) × RelevanceIrrelevant	-3.66	0.14	(-3.92, -3.39)	-26.80	< .001
Sex (Female) × RelevanceIrrelevant	-0.31	0.15	(-0.59, -0.02)	-2.09	0.036
Sex (Male) × RelevanceNon-erotic	-3.78	0.15	(-4.07, -3.49)	-25.20	< .001
Sex (Female) × RelevanceNon-erotic	-2.39	0.16	(-2.71, -2.07)	-14.69	< .001
Sex (Male) × RelevanceRelevant × ConditionAI-Generated	-0.39	0.04	(-0.47, -0.30)	-9.28	< .001
Sex (Female) × RelevanceRelevant × ConditionAI-Generated	-0.37	0.05	(-0.47, -0.27)	-7.07	< .001
Sex (Male) × RelevanceIrrelevant × ConditionAI-Generated	-0.07	0.06	(-0.18, 0.04)	-1.24	0.215
Sex (Female) × RelevanceIrrelevant × ConditionAI-Generated	-0.17	0.06	(-0.27, -0.06)	-2.99	0.003
Sex (Male) × RelevanceNon-erotic × ConditionAI-Generated	-0.18	0.07	(-0.33, -0.04)	-2.46	0.014
Sex (Female) × RelevanceNon-erotic × ConditionAI-Generated	-0.08	0.10	(-0.27, 0.11)	-0.86	0.389

  moderator_believed <- data.frame()
  for(m in c("BAIT_Visual", "BAIT_Text", "AI_Knowledge",
             "GAAIS_Positive", "GAAIS_Negative",
             "Porn", "PornFrequency", "SexualActivity_num", 
             "Feedback_CouldDiscriminate", "Feedback_CouldNotDiscriminate",
             "Feedback_AIMoreArousing", "Feedback_AILessArousing", 
             "Feedback_LabelsIncorrect", "Feedback_LabelsReversed")) {
    moderator_believed <- rbind(moderator_believed,
                               test_moderator(filter(df, ConditionBelief == "True"), "Arousal", m))
  }
  
  display(filter(moderator_believed, p < .1))

Parameter	Coefficient	95% CI	z	p	Moderator	Outcome
Sex [Female] × ConditionAI-Generated × GAAIS Negative	0.51	[-0.06, 1.07]	1.76	0.078	GAAIS_Negative	Arousal
Sex [Female] × ConditionAI-Generated × Porn	-0.15	[-0.31, 0.00]	-1.92	0.055	Porn	Arousal
Sex [Female] × ConditionAI-Generated × SexualActivity num	-0.09	[-0.17, -0.01]	-2.11	0.035	SexualActivity_num	Arousal
Sex [Male] × ConditionAI-Generated × Feedback AILessArousingTrue	-0.36	[-0.58, -0.15]	-3.35	< .001	Feedback_AILessArousing	Arousal

Enticement

enticement_desc <- plot_descriptive(df, "Enticement")
enticement_desc

Whole Sample

This model examines how Sex, Relevance, and Condition influence Enticement ratings, treating relevance within each participant and items as random effects.

Females consistently rate relevant (photograph) images significantly lower in enticement compared to males. Conversely, men rate irrelevant images (photographs) significantly lower in enticement than relevant ones, although this difference is not observed among females. Both genders rate non-erotic images (photographs) significantly lower in enticement than relevant ones. Interestingly, neither gender shows a significant difference in enticement ratings between AI-generated images and photographs, regardless of relevance.

We also analyzed the average marginal effect of the variable “Condition” (comparing AI-Generated to Photograph) separately for males and females.

For males, enticement ratings are significantly lower on average for AI-Generated images compared to Photograph images. For females, enticement ratings are slightly higher on average for AI-Generated images compared to Photograph images, although this difference is not statistically significant.

Note. Model adding Relevance and Condition as random slopes for each stimuli converged

m_enticement <- glmmTMB::glmmTMB(Enticement ~ Sex / Relevance / Condition + 
                                   (Relevance | Participant) + (1|Item),
                      data=df, family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

results_table(m_enticement)

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
(Intercept)	0.83	0.08	(0.68, 0.99)	10.62	< .001
Sex (Female)	-0.98	0.11	(-1.20, -0.75)	-8.51	< .001
Sex (Male) × RelevanceIrrelevant	-2.23	0.12	(-2.47, -1.99)	-18.45	< .001
Sex (Female) × RelevanceIrrelevant	0.06	0.13	(-0.21, 0.32)	0.43	0.669
Sex (Male) × RelevanceNon-erotic	-2.46	0.12	(-2.69, -2.22)	-20.74	< .001
Sex (Female) × RelevanceNon-erotic	-1.17	0.13	(-1.42, -0.91)	-8.89	< .001
Sex (Male) × RelevanceRelevant × ConditionAI-Generated	-0.08	0.02	(-0.12, -0.05)	-4.68	< .001
Sex (Female) × RelevanceRelevant × ConditionAI-Generated	0.02	0.02	(-0.02, 0.07)	0.99	0.323
Sex (Male) × RelevanceIrrelevant × ConditionAI-Generated	-5.40e-03	0.02	(-0.05, 0.03)	-0.26	0.792
Sex (Female) × RelevanceIrrelevant × ConditionAI-Generated	-9.63e-03	0.03	(-0.06, 0.04)	-0.38	0.700
Sex (Male) × RelevanceNon-erotic × ConditionAI-Generated	-0.05	0.02	(-0.09, -8.37e-03)	-2.34	0.019
Sex (Female) × RelevanceNon-erotic × ConditionAI-Generated	0.01	0.03	(-0.04, 0.07)	0.45	0.654

marginaleffects::avg_slopes(m_enticement, variables="Condition", by="Sex", re.form=NA) |> 
  results_table()

Comparison	Sex	Coefficient	z	p	95% CI
AI-Generated - Photograph	Male	-8.68e-03	-4.40	< .001	[-0.01, 0.00]
AI-Generated - Photograph	Female	1.94e-03	0.58	0.560	[ 0.00, 0.01]

Subsamples

Sample

Here we examined the interaction between Sample (different participant groups), Sex, Relevance, and Condition (AI-Generated vs. Photograph) on enticement ratings.

These findings indicate that in the General Population and Students (Non-incentivized) samples, males show a preference for relevant Photograph images over relevant AI-Generated images in terms of enticement, while females tend to rate relevant AI-Generated images higher in enticement. The differences in enticement ratings between AI-Generated and Photograph images were not significant in other samples or conditions examined.

Note. Model adding Relevance and Condition as random slopes for each stimuli converged.

m <- glmmTMB::glmmTMB(Enticement ~ Sample / Sex / Relevance / Condition + (Relevance | Participant) + ((Relevance/Condition)|Item),
                      data=df, 
                      family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; non-positive-definite Hessian matrix. See vignette('troubleshooting')

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; singular convergence (7). See vignette('troubleshooting'), help('diagnose')

results_table(m, filter="Condition")

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
Sample (General Population) × SexMale × RelevanceRelevant × ConditionAI-Generated	-0.10	0.02	(-0.14, -0.06)	-4.73	< .001
Sample (Students (Incentivized)) × SexMale × RelevanceRelevant × ConditionAI-Generated	-0.07	0.06	(-0.19, 0.05)	-1.16	0.247
Sample (Students (Non-incentivized)) × SexMale × RelevanceRelevant × ConditionAI-Generated	-0.03	0.06	(-0.14, 0.09)	-0.45	0.656
Sample (General Population) × SexFemale × RelevanceRelevant × ConditionAI-Generated	9.39e-03	0.05	(-0.08, 0.10)	0.20	0.841
Sample (Students (Incentivized)) × SexFemale × RelevanceRelevant × ConditionAI-Generated	-0.04	0.04	(-0.12, 0.04)	-1.09	0.275
Sample (Students (Non-incentivized)) × SexFemale × RelevanceRelevant × ConditionAI-Generated	0.12	0.04	(0.03, 0.20)	2.66	0.008
Sample (General Population) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	-6.55e-03	0.02	(-0.05, 0.04)	-0.29	0.771
Sample (Students (Incentivized)) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	-0.09	0.08	(-0.24, 0.07)	-1.11	0.269
Sample (Students (Non-incentivized)) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	0.05	0.07	(-0.09, 0.18)	0.65	0.513
Sample (General Population) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	-6.90e-03	0.05	(-0.10, 0.08)	-0.15	0.882
Sample (Students (Incentivized)) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	-0.04	0.04	(-0.12, 0.04)	-0.90	0.368
Sample (Students (Non-incentivized)) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	0.03	0.04	(-0.06, 0.11)	0.60	0.550
Sample (General Population) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	-0.05	0.02	(-0.10, -5.95e-03)	-2.20	0.028
Sample (Students (Incentivized)) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	0.11	0.08	(-0.05, 0.27)	1.36	0.173
Sample (Students (Non-incentivized)) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	-0.16	0.07	(-0.29, -0.03)	-2.34	0.020
Sample (General Population) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	7.01e-03	0.05	(-0.09, 0.11)	0.14	0.892
Sample (Students (Incentivized)) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	-0.03	0.05	(-0.12, 0.07)	-0.53	0.594
Sample (Students (Non-incentivized)) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	0.06	0.05	(-0.04, 0.15)	1.13	0.260

Mobile

Here, we explored the impact of using a mobile device (mobile = TRUE) versus not using one as a predictor of enticement, alongside condition, sex, and relevance.

We observed that males rated relevant AI-generated images significantly lower in enticement compared to relevant photographs, regardless of whether they used a mobile device or not. However, this effect was more pronounced among males using a mobile device. Additionally, for males, using a mobile device influenced enticement ratings for non-erotic stimuli, where they rated AI-generated non-erotic stimuli lower in enticement compared to non-erotic photographs.

Note. Model adding Relevance and Condition as random slopes for each stimuli converged.

m <- glmmTMB::glmmTMB(Enticement ~ Mobile / Sex / Relevance / Condition + (Relevance | Participant) + ((Relevance/Condition)|Item),
                      data=df, 
                      family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; non-positive-definite Hessian matrix. See vignette('troubleshooting')

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; singular convergence (7). See vignette('troubleshooting'), help('diagnose')

results_table(m, filter="Condition")

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
Mobile (False) × SexMale × RelevanceRelevant × ConditionAI-Generated	-0.07	0.02	(-0.12, -0.03)	-3.19	0.001
Mobile (True) × SexMale × RelevanceRelevant × ConditionAI-Generated	-0.11	0.03	(-0.18, -0.05)	-3.73	< .001
Mobile (False) × SexFemale × RelevanceRelevant × ConditionAI-Generated	0.04	0.03	(-0.02, 0.09)	1.22	0.224
Mobile (True) × SexFemale × RelevanceRelevant × ConditionAI-Generated	-9.76e-03	0.05	(-0.11, 0.09)	-0.19	0.847
Mobile (False) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	-0.05	0.03	(-0.10, 3.65e-03)	-1.82	0.068
Mobile (True) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	0.06	0.03	(-8.07e-03, 0.12)	1.71	0.086
Mobile (False) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	-0.03	0.03	(-0.08, 0.03)	-0.93	0.354
Mobile (True) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	0.05	0.05	(-0.05, 0.15)	1.02	0.306
Mobile (False) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	-0.03	0.03	(-0.09, 0.02)	-1.08	0.280
Mobile (True) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	-0.09	0.04	(-0.16, -0.02)	-2.47	0.014
Mobile (False) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	0.01	0.03	(-0.06, 0.08)	0.31	0.756
Mobile (True) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	0.02	0.06	(-0.09, 0.13)	0.31	0.753

Moderators

We examined whether attitudes towards AI, porn usage and frequency, and sexual activity moderated the relationship between the predictors (sex, relevance, condition) and enticement ratings. The results indicated that no moderator was significant.

Looking at feedback as a moderator, the results showed that enticement ratings for images labeled AI-generated were moderated by the feedback that AI is more arousing, but this was only significant for women. Specifically, women who reported that AI-generated images as more arousing rated these images higher in enticement compared to images labeled as photographs.

Lastly, the interaction between sexual activity and condition on enticement, suggests that for females, the effect of AI-Generated condition on enticing ratings varies with different levels of sexual activity.

for(m in c("BAIT_Visual", "BAIT_Text", "AI_Knowledge",
           "GAAIS_Positive", "GAAIS_Negative",
           "Porn", "PornFrequency", "SexualActivity_num")) {
  moderator_results <- rbind(moderator_results,
                             test_moderator(df, "Enticement", m))
}

display(filter(moderator_results, Outcome == "Enticement"))

Parameter	Coefficient	95% CI	z	p	Moderator	Outcome
Sex [Male] × ConditionAI-Generated × BAIT Visual	0.10	[-0.13, 0.32]	0.84	0.401	BAIT_Visual	Enticement
Sex [Female] × ConditionAI-Generated × BAIT Visual	0.27	[-0.09, 0.63]	1.47	0.143	BAIT_Visual	Enticement
Sex [Male] × ConditionAI-Generated × BAIT Text	0.02	[-0.14, 0.18]	0.28	0.783	BAIT_Text	Enticement
Sex [Female] × ConditionAI-Generated × BAIT Text	0.10	[-0.15, 0.35]	0.78	0.437	BAIT_Text	Enticement
Sex [Male] × ConditionAI-Generated × AI Knowledge	0.02	[-0.02, 0.05]	1.02	0.306	AI_Knowledge	Enticement
Sex [Female] × ConditionAI-Generated × AI Knowledge	2.45e-03	[-0.04, 0.04]	0.12	0.905	AI_Knowledge	Enticement
Sex [Male] × ConditionAI-Generated × GAAIS Positive	0.04	[-0.14, 0.22]	0.41	0.678	GAAIS_Positive	Enticement
Sex [Female] × ConditionAI-Generated × GAAIS Positive	0.23	[-0.04, 0.50]	1.68	0.092	GAAIS_Positive	Enticement
Sex [Male] × ConditionAI-Generated × GAAIS Negative	0.03	[-0.13, 0.20]	0.42	0.675	GAAIS_Negative	Enticement
Sex [Female] × ConditionAI-Generated × GAAIS Negative	0.16	[-0.10, 0.41]	1.20	0.231	GAAIS_Negative	Enticement
Sex [Male] × ConditionAI-Generated × Porn	-0.04	[-0.09, 0.01]	-1.57	0.115	Porn	Enticement
Sex [Female] × ConditionAI-Generated × Porn	-0.07	[-0.14, 0.00]	-1.97	0.049	Porn	Enticement
Sex [Male] × ConditionAI-Generated × PornFrequency	-0.01	[-0.04, 0.01]	-0.82	0.413	PornFrequency	Enticement
Sex [Female] × ConditionAI-Generated × PornFrequency	-4.65e-03	[-0.04, 0.04]	-0.23	0.819	PornFrequency	Enticement
Sex [Male] × ConditionAI-Generated × SexualActivity num	-0.01	[-0.05, 0.02]	-0.63	0.529	SexualActivity_num	Enticement
Sex [Female] × ConditionAI-Generated × SexualActivity num	-0.02	[-0.05, 0.02]	-0.84	0.403	SexualActivity_num	Enticement

for(m in c("Feedback_CouldDiscriminate", "Feedback_CouldNotDiscriminate",
           "Feedback_AIMoreArousing", "Feedback_AILessArousing", 
           "Feedback_LabelsIncorrect", "Feedback_LabelsReversed")) {
  feedback_results <- rbind(feedback_results,
                            test_moderator(df, "Enticement", m))
}
display(filter(feedback_results, Outcome == "Enticement"))

Parameter	Coefficient	95% CI	z	p	Moderator	Outcome
Sex [Male] × ConditionAI-Generated × Feedback CouldDiscriminateTrue	-0.05	[-0.20, 0.10]	-0.66	0.510	Feedback_CouldDiscriminate	Enticement
Sex [Female] × ConditionAI-Generated × Feedback CouldDiscriminateTrue	-0.04	[-0.25, 0.17]	-0.37	0.711	Feedback_CouldDiscriminate	Enticement
Sex [Male] × ConditionAI-Generated × Feedback CouldNotDiscriminateTrue	0.07	[ 0.00, 0.15]	1.85	0.065	Feedback_CouldNotDiscriminate	Enticement
Sex [Female] × ConditionAI-Generated × Feedback CouldNotDiscriminateTrue	0.03	[-0.07, 0.13]	0.54	0.589	Feedback_CouldNotDiscriminate	Enticement
Sex [Male] × ConditionAI-Generated × Feedback AIMoreArousingTrue	0.01	[-0.12, 0.15]	0.20	0.841	Feedback_AIMoreArousing	Enticement
Sex [Female] × ConditionAI-Generated × Feedback AIMoreArousingTrue	0.14	[ 0.00, 0.28]	1.91	0.056	Feedback_AIMoreArousing	Enticement
Sex [Male] × ConditionAI-Generated × Feedback AILessArousingTrue	-0.14	[-0.26, -0.02]	-2.33	0.020	Feedback_AILessArousing	Enticement
Sex [Female] × ConditionAI-Generated × Feedback AILessArousingTrue	0.14	[-0.04, 0.31]	1.56	0.120	Feedback_AILessArousing	Enticement
Sex [Male] × ConditionAI-Generated × Feedback LabelsIncorrectTrue	-0.04	[-0.11, 0.04]	-0.95	0.344	Feedback_LabelsIncorrect	Enticement
Sex [Female] × ConditionAI-Generated × Feedback LabelsIncorrectTrue	-0.10	[-0.21, 0.01]	-1.80	0.071	Feedback_LabelsIncorrect	Enticement
Sex [Male] × ConditionAI-Generated × Feedback LabelsReversedTrue	-0.07	[-0.24, 0.10]	-0.81	0.420	Feedback_LabelsReversed	Enticement
Sex [Female] × ConditionAI-Generated × Feedback LabelsReversedTrue	0.22	[ 0.02, 0.41]	2.19	0.028	Feedback_LabelsReversed	Enticement

m <- glmmTMB::glmmTMB(Arousal ~ Sex / (Condition * SexualActivity_num) + (Condition | Participant) + ((Relevance/Condition)|Item),
                      data=filter(df, ConditionBelief == "True"), 
                      family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; non-positive-definite Hessian matrix. See vignette('troubleshooting')

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; singular convergence (7). See vignette('troubleshooting'), help('diagnose')

results_table(m)

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
(Intercept)	-2.30	0.10	(-2.50, -2.10)	-22.42	< .001
Sex (Female)	0.27	0.14	(-2.13e-04, 0.55)	1.96	0.050
Sex (Male) × ConditionAI-Generated	-0.13	0.06	(-0.26, -6.95e-03)	-2.07	0.039
Sex (Female) × ConditionAI-Generated	-0.06	0.09	(-0.25, 0.12)	-0.68	0.496
Sex (Male) × SexualActivity num	-0.04	0.03	(-0.11, 0.02)	-1.26	0.209
Sex (Female) × SexualActivity num	-0.04	0.04	(-0.11, 0.03)	-1.18	0.237
Sex (Male) × ConditionAI-Generated × SexualActivity num	-0.02	0.03	(-0.07, 0.04)	-0.53	0.595
Sex (Female) × ConditionAI-Generated × SexualActivity num	-0.02	0.03	(-0.08, 0.03)	-0.77	0.441

sex_act_plot<- estimate_relation(m, length=20) |> 
  ggplot(aes(x=SexualActivity_num, y=Predicted)) +
  geom_ribbon(aes(ymin=CI_low, ymax=CI_high, fill=Condition), alpha=0.2) +
  geom_line(aes(color=Condition)) +
  facet_grid(~Sex)

sex_act_plot

Believed

Note. Model adding Relevance and Condition as random slopes for each stimuli converged.

m <- glmmTMB::glmmTMB(Enticement ~ Sex / Relevance / Condition + (Relevance | Participant) + ((Relevance/Condition)|Item),
                      data=filter(df, ConditionBelief == "True"), 
                      family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; non-positive-definite Hessian matrix. See vignette('troubleshooting')

results_table(m)

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
(Intercept)	0.89	0.09	(0.71, 1.07)	9.55	< .001
Sex (Female)	-0.94	0.14	(-1.21, -0.68)	-6.97	< .001
Sex (Male) × RelevanceIrrelevant	-2.23	0.12	(-2.47, -1.99)	-18.21	< .001
Sex (Female) × RelevanceIrrelevant	0.05	0.13	(-0.21, 0.32)	0.39	0.699
Sex (Male) × RelevanceNon-erotic	-2.48	0.14	(-2.74, -2.21)	-18.36	< .001
Sex (Female) × RelevanceNon-erotic	-1.20	0.15	(-1.48, -0.91)	-8.18	< .001
Sex (Male) × RelevanceRelevant × ConditionAI-Generated	-0.40	0.04	(-0.49, -0.31)	-8.99	< .001
Sex (Female) × RelevanceRelevant × ConditionAI-Generated	-0.30	0.05	(-0.40, -0.20)	-5.79	< .001
Sex (Male) × RelevanceIrrelevant × ConditionAI-Generated	-0.13	0.04	(-0.20, -0.05)	-3.37	< .001
Sex (Female) × RelevanceIrrelevant × ConditionAI-Generated	-0.28	0.05	(-0.36, -0.19)	-6.07	< .001
Sex (Male) × RelevanceNon-erotic × ConditionAI-Generated	-0.23	0.05	(-0.32, -0.13)	-4.77	< .001
Sex (Female) × RelevanceNon-erotic × ConditionAI-Generated	-0.12	0.07	(-0.25, 0.01)	-1.74	0.081

moderator_believed <- data.frame()
for(m in c("BAIT_Visual", "BAIT_Text", "AI_Knowledge",
           "GAAIS_Positive", "GAAIS_Negative",
           "Porn", "PornFrequency", "SexualActivity_num", 
           "Feedback_CouldDiscriminate", "Feedback_CouldNotDiscriminate",
           "Feedback_AIMoreArousing", "Feedback_AILessArousing", 
           "Feedback_LabelsIncorrect", "Feedback_LabelsReversed")) {
  moderator_believed <- rbind(moderator_believed,
                             test_moderator(filter(df, ConditionBelief == "True"), "Enticement", m))
}

display(filter(moderator_believed, p < .1))

Parameter	Coefficient	95% CI	z	p	Moderator	Outcome
Sex [Male] × ConditionAI-Generated × BAIT Text	0.30	[-0.02, 0.63]	1.84	0.065	BAIT_Text	Enticement
Sex [Female] × ConditionAI-Generated × Porn	-0.13	[-0.27, 0.01]	-1.81	0.070	Porn	Enticement
Sex [Female] × ConditionAI-Generated × SexualActivity num	-0.07	[-0.14, 0.01]	-1.80	0.072	SexualActivity_num	Enticement
Sex [Male] × ConditionAI-Generated × Feedback AILessArousingTrue	-0.38	[-0.58, -0.18]	-3.73	< .001	Feedback_AILessArousing	Enticement

Valence

valence_desc <- plot_descriptive(df, "Valence")
valence_desc

Whole Sample

This model investigates the effects of Sex, Relevance, and Condition on Valence, accounting for random variability due to participants and items (i.e., random effects).

Females rate relevant images significantly lower in valence than males on average. Males rate irrelevant and non-erotic images significantly lower in valence compared to relevant images. Females rate both irrelevant and non-erotic images higher in valence compared to relevant ones, but this effect is only significant for non-erotic stimuli. Males rate relevant images significantly lower in valence when labeled as AI-generated compared to when labeled as photographs.

Females rate relevant (photograph) images significantly lower in valence compared to males. Males rate irrelevant (photograph) and non-erotic (photograph) images significantly lower in valence compared to relevant images. Females rate non-erotic (photograph) images higher in valence compared to relevant ones. Males rate relevant AI-Generated images significantly lower in valence compared to photographs.

Lastly, we can see that for males, valence ratings are on average significantly lower for AI-Generated images compared to Photograph images. For females, there is no significant effect.

Note. Model adding Relevance and Condition as random slopes for each stimuli converged.

m_valence <- glmmTMB::glmmTMB(Valence ~ Sex / Relevance / Condition + 
                                (Relevance | Participant) + (1|Item),
                      data=df, family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

results_table(m_valence)

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
(Intercept)	0.91	0.09	(0.74, 1.07)	10.65	< .001
Sex (Female)	-1.02	0.12	(-1.26, -0.77)	-8.19	< .001
Sex (Male) × RelevanceIrrelevant	-1.68	0.12	(-1.92, -1.44)	-13.71	< .001
Sex (Female) × RelevanceIrrelevant	0.06	0.13	(-0.20, 0.31)	0.43	0.671
Sex (Male) × RelevanceNon-erotic	-0.68	0.12	(-0.92, -0.45)	-5.82	< .001
Sex (Female) × RelevanceNon-erotic	0.49	0.12	(0.25, 0.73)	3.97	< .001
Sex (Male) × RelevanceRelevant × ConditionAI-Generated	-0.10	0.02	(-0.13, -0.06)	-5.60	< .001
Sex (Female) × RelevanceRelevant × ConditionAI-Generated	0.02	0.02	(-0.02, 0.06)	0.87	0.383
Sex (Male) × RelevanceIrrelevant × ConditionAI-Generated	-0.03	0.02	(-0.06, 6.13e-03)	-1.61	0.108
Sex (Female) × RelevanceIrrelevant × ConditionAI-Generated	-5.90e-03	0.02	(-0.05, 0.04)	-0.26	0.793
Sex (Male) × RelevanceNon-erotic × ConditionAI-Generated	-0.04	0.02	(-0.07, -6.72e-03)	-2.36	0.018
Sex (Female) × RelevanceNon-erotic × ConditionAI-Generated	-0.02	0.02	(-0.07, 0.03)	-0.86	0.389

marginaleffects::avg_slopes(m_valence, variables="Condition", by="Sex", re.form=NA) |> 
  results_table()

Comparison	Sex	Coefficient	z	p	95% CI
AI-Generated - Photograph	Male	-0.01	-5.38	< .001	[-0.02, -0.01]
AI-Generated - Photograph	Female	-4.69e-04	-0.14	0.885	[-0.01, 0.01]

Subsamples

Sample

Examining the effect of sample type as a predictor, males consistently rate relevant AI-generated images significantly lower in valence compared to relevant photographs. Additionally, among non-incentivized students, this pattern persists for males, while females rate relevant AI-generated images significantly higher in valence compared to relevant photographs

Note. Model adding Relevance and Condition as random slopes for each stimuli converged.

m <- glmmTMB::glmmTMB(Valence ~ Sample / Sex / Relevance / Condition + (Relevance | Participant) + ((Relevance/Condition)|Item),
                      data=df, 
                      family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; non-positive-definite Hessian matrix. See vignette('troubleshooting')

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; function evaluation limit reached without convergence (9). See vignette('troubleshooting'), help('diagnose')

results_table(m, filter="Condition")

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
Sample (General Population) × SexMale × RelevanceRelevant × ConditionAI-Generated	-0.12	0.02	(-0.16, -0.08)	-5.52	< .001
Sample (Students (Incentivized)) × SexMale × RelevanceRelevant × ConditionAI-Generated	-0.02	0.06	(-0.13, 0.10)	-0.27	0.784
Sample (Students (Non-incentivized)) × SexMale × RelevanceRelevant × ConditionAI-Generated	-9.87e-03	0.06	(-0.12, 0.10)	-0.18	0.859
Sample (General Population) × SexFemale × RelevanceRelevant × ConditionAI-Generated	0.02	0.04	(-0.07, 0.11)	0.48	0.632
Sample (Students (Incentivized)) × SexFemale × RelevanceRelevant × ConditionAI-Generated	-0.05	0.04	(-0.12, 0.03)	-1.27	0.203
Sample (Students (Non-incentivized)) × SexFemale × RelevanceRelevant × ConditionAI-Generated	0.08	0.04	(4.36e-03, 0.16)	2.07	0.039
Sample (General Population) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	-0.03	0.02	(-0.07, 0.01)	-1.27	0.204
Sample (Students (Incentivized)) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	-0.15	0.07	(-0.28, -0.02)	-2.25	0.025
Sample (Students (Non-incentivized)) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	0.06	0.06	(-0.06, 0.17)	0.96	0.335
Sample (General Population) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	7.95e-03	0.04	(-0.08, 0.09)	0.18	0.856
Sample (Students (Incentivized)) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	-0.02	0.04	(-0.09, 0.05)	-0.54	0.590
Sample (Students (Non-incentivized)) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	-2.71e-03	0.04	(-0.08, 0.08)	-0.07	0.946
Sample (General Population) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	-0.03	0.02	(-0.07, 5.62e-03)	-1.65	0.099
Sample (Students (Incentivized)) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	-0.03	0.06	(-0.14, 0.09)	-0.47	0.638
Sample (Students (Non-incentivized)) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	-0.13	0.05	(-0.24, -0.02)	-2.41	0.016
Sample (General Population) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	-0.05	0.04	(-0.13, 0.04)	-1.10	0.271
Sample (Students (Incentivized)) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	6.70e-03	0.04	(-0.07, 0.08)	0.18	0.856
Sample (Students (Non-incentivized)) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	-0.03	0.04	(-0.11, 0.05)	-0.68	0.493

Mobile

When examining the use of mobile devices as a predictor, men consistently rated the valence lower for relevant AI-generated images compared to photographs, regardless of whether they used a mobile or a standard computer. However, the effect was more pronounced among those using mobile devices.

m <- glmmTMB::glmmTMB(Valence ~ Mobile / Sex / Relevance / Condition + (Relevance | Participant) + ((Relevance/Condition)|Item),
                      data=df, 
                      family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; non-positive-definite Hessian matrix. See vignette('troubleshooting')

results_table(m, filter="Condition")

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
Mobile (False) × SexMale × RelevanceRelevant × ConditionAI-Generated	-0.08	0.02	(-0.12, -0.03)	-3.24	0.001
Mobile (True) × SexMale × RelevanceRelevant × ConditionAI-Generated	-0.14	0.03	(-0.20, -0.08)	-4.58	< .001
Mobile (False) × SexFemale × RelevanceRelevant × ConditionAI-Generated	0.01	0.03	(-0.04, 0.07)	0.53	0.595
Mobile (True) × SexFemale × RelevanceRelevant × ConditionAI-Generated	0.02	0.05	(-0.08, 0.11)	0.37	0.708
Mobile (False) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	-0.04	0.02	(-0.09, 2.95e-03)	-1.84	0.066
Mobile (True) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	-6.41e-04	0.03	(-0.06, 0.06)	-0.02	0.983
Mobile (False) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	-0.01	0.03	(-0.07, 0.04)	-0.40	0.686
Mobile (True) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	8.53e-03	0.05	(-0.08, 0.10)	0.18	0.857
Mobile (False) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	-0.03	0.02	(-0.07, 8.94e-03)	-1.53	0.126
Mobile (True) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	-0.05	0.03	(-0.11, 1.47e-03)	-1.91	0.057
Mobile (False) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	-0.01	0.03	(-0.07, 0.04)	-0.51	0.609
Mobile (True) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	-0.04	0.05	(-0.13, 0.05)	-0.84	0.400

Moderators

Here we look at whether the relationship between sex, condition and relevance and valence is moderated by AI-attitudes, porn frequency and usage, and sexual activity. Females who have more positive attitudes towards AI rated AI-generated images significantly higher in valence compared to images labelled as photographs.

Looking at feedback as a moderator we can see that ales who reported that they could not discriminate between AI-generated and photograph images, rated AI-generated images significantly higher in valence compared to images labelled as photographs.

for(m in c("BAIT_Visual", "BAIT_Text", "AI_Knowledge",
           "GAAIS_Positive", "GAAIS_Negative",
           "Porn", "PornFrequency", "SexualActivity_num")) {
  moderator_results <- rbind(moderator_results,
                             test_moderator(df, "Valence", m))
}

display(filter(moderator_results, Outcome == "Valence"))

Parameter	Coefficient	95% CI	z	p	Moderator	Outcome
Sex [Male] × ConditionAI-Generated × BAIT Visual	0.08	[-0.15, 0.32]	0.72	0.473	BAIT_Visual	Valence
Sex [Female] × ConditionAI-Generated × BAIT Visual	0.30	[-0.06, 0.65]	1.64	0.100	BAIT_Visual	Valence
Sex [Male] × ConditionAI-Generated × BAIT Text	0.10	[-0.06, 0.26]	1.24	0.214	BAIT_Text	Valence
Sex [Female] × ConditionAI-Generated × BAIT Text	0.17	[-0.07, 0.42]	1.42	0.157	BAIT_Text	Valence
Sex [Male] × ConditionAI-Generated × AI Knowledge	5.74e-03	[-0.03, 0.04]	0.33	0.742	AI_Knowledge	Valence
Sex [Female] × ConditionAI-Generated × AI Knowledge	0.03	[-0.01, 0.07]	1.55	0.122	AI_Knowledge	Valence
Sex [Male] × ConditionAI-Generated × GAAIS Positive	0.04	[-0.15, 0.22]	0.41	0.680	GAAIS_Positive	Valence
Sex [Female] × ConditionAI-Generated × GAAIS Positive	0.35	[ 0.09, 0.62]	2.59	0.010	GAAIS_Positive	Valence
Sex [Male] × ConditionAI-Generated × GAAIS Negative	-2.35e-03	[-0.17, 0.16]	-0.03	0.978	GAAIS_Negative	Valence
Sex [Female] × ConditionAI-Generated × GAAIS Negative	0.18	[-0.07, 0.43]	1.38	0.166	GAAIS_Negative	Valence
Sex [Male] × ConditionAI-Generated × Porn	-0.04	[-0.09, 0.01]	-1.56	0.118	Porn	Valence
Sex [Female] × ConditionAI-Generated × Porn	-0.06	[-0.13, 0.01]	-1.76	0.079	Porn	Valence
Sex [Male] × ConditionAI-Generated × PornFrequency	-0.01	[-0.04, 0.01]	-0.88	0.377	PornFrequency	Valence
Sex [Female] × ConditionAI-Generated × PornFrequency	-0.02	[-0.06, 0.02]	-1.04	0.300	PornFrequency	Valence
Sex [Male] × ConditionAI-Generated × SexualActivity num	-0.02	[-0.06, 0.01]	-1.30	0.193	SexualActivity_num	Valence
Sex [Female] × ConditionAI-Generated × SexualActivity num	-9.41e-03	[-0.05, 0.03]	-0.49	0.621	SexualActivity_num	Valence

for(m in c("Feedback_CouldDiscriminate", "Feedback_CouldNotDiscriminate",
           "Feedback_AIMoreArousing", "Feedback_AILessArousing", 
           "Feedback_LabelsIncorrect", "Feedback_LabelsReversed")) {
  feedback_results <- rbind(feedback_results,
                            test_moderator(df, "Valence", m))
}
display(filter(feedback_results, Outcome == "Valence"))

Parameter	Coefficient	95% CI	z	p	Moderator	Outcome
Sex [Male] × ConditionAI-Generated × Feedback CouldDiscriminateTrue	-0.14	[-0.30, 0.01]	-1.87	0.062	Feedback_CouldDiscriminate	Valence
Sex [Female] × ConditionAI-Generated × Feedback CouldDiscriminateTrue	-0.05	[-0.26, 0.15]	-0.50	0.618	Feedback_CouldDiscriminate	Valence
Sex [Male] × ConditionAI-Generated × Feedback CouldNotDiscriminateTrue	0.10	[ 0.02, 0.18]	2.49	0.013	Feedback_CouldNotDiscriminate	Valence
Sex [Female] × ConditionAI-Generated × Feedback CouldNotDiscriminateTrue	0.05	[-0.05, 0.16]	1.03	0.301	Feedback_CouldNotDiscriminate	Valence
Sex [Male] × ConditionAI-Generated × Feedback AIMoreArousingTrue	0.04	[-0.09, 0.18]	0.63	0.529	Feedback_AIMoreArousing	Valence
Sex [Female] × ConditionAI-Generated × Feedback AIMoreArousingTrue	0.11	[-0.03, 0.26]	1.57	0.116	Feedback_AIMoreArousing	Valence
Sex [Male] × ConditionAI-Generated × Feedback AILessArousingTrue	-0.11	[-0.24, 0.01]	-1.82	0.069	Feedback_AILessArousing	Valence
Sex [Female] × ConditionAI-Generated × Feedback AILessArousingTrue	0.08	[-0.10, 0.25]	0.87	0.386	Feedback_AILessArousing	Valence
Sex [Male] × ConditionAI-Generated × Feedback LabelsIncorrectTrue	-0.05	[-0.13, 0.02]	-1.36	0.175	Feedback_LabelsIncorrect	Valence
Sex [Female] × ConditionAI-Generated × Feedback LabelsIncorrectTrue	-0.08	[-0.19, 0.03]	-1.36	0.173	Feedback_LabelsIncorrect	Valence
Sex [Male] × ConditionAI-Generated × Feedback LabelsReversedTrue	0.02	[-0.16, 0.19]	0.20	0.843	Feedback_LabelsReversed	Valence
Sex [Female] × ConditionAI-Generated × Feedback LabelsReversedTrue	0.12	[-0.07, 0.31]	1.24	0.214	Feedback_LabelsReversed	Valence

Believed

m <- glmmTMB::glmmTMB(Valence ~ Sex / Relevance / Condition + (Relevance | Participant) + ((Relevance/Condition)|Item),
                      data=filter(df, ConditionBelief == "True"), 
                      family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; non-positive-definite Hessian matrix. See vignette('troubleshooting')

results_table(m)

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
(Intercept)	0.96	0.09	(0.79, 1.13)	11.02	< .001
Sex (Female)	-0.94	0.13	(-1.19, -0.70)	-7.45	< .001
Sex (Male) × RelevanceIrrelevant	-1.66	0.11	(-1.87, -1.44)	-15.10	< .001
Sex (Female) × RelevanceIrrelevant	0.03	0.12	(-0.21, 0.26)	0.22	0.828
Sex (Male) × RelevanceNon-erotic	-0.72	0.14	(-0.99, -0.45)	-5.22	< .001
Sex (Female) × RelevanceNon-erotic	0.42	0.14	(0.14, 0.70)	2.94	0.003
Sex (Male) × RelevanceRelevant × ConditionAI-Generated	-0.45	0.05	(-0.54, -0.37)	-9.99	< .001
Sex (Female) × RelevanceRelevant × ConditionAI-Generated	-0.33	0.05	(-0.43, -0.23)	-6.53	< .001
Sex (Male) × RelevanceIrrelevant × ConditionAI-Generated	-0.21	0.03	(-0.27, -0.14)	-6.01	< .001
Sex (Female) × RelevanceIrrelevant × ConditionAI-Generated	-0.23	0.04	(-0.32, -0.15)	-5.44	< .001
Sex (Male) × RelevanceNon-erotic × ConditionAI-Generated	-0.17	0.04	(-0.25, -0.10)	-4.70	< .001
Sex (Female) × RelevanceNon-erotic × ConditionAI-Generated	-0.23	0.06	(-0.34, -0.12)	-4.07	< .001

moderator_believed <- data.frame()
for(m in c("BAIT_Visual", "BAIT_Text", "AI_Knowledge",
           "GAAIS_Positive", "GAAIS_Negative",
           "Porn", "PornFrequency", "SexualActivity_num", 
           "Feedback_CouldDiscriminate", "Feedback_CouldNotDiscriminate",
           "Feedback_AIMoreArousing", "Feedback_AILessArousing", 
           "Feedback_LabelsIncorrect", "Feedback_LabelsReversed")) {
  moderator_believed <- rbind(moderator_believed,
                             test_moderator(filter(df, ConditionBelief == "True"), "Valence", m))
}

display(filter(moderator_believed, p < .1))

Parameter	Coefficient	95% CI	z	p	Moderator	Outcome
Sex [Male] × ConditionAI-Generated × BAIT Text	0.37	[ 0.02, 0.72]	2.05	0.040	BAIT_Text	Valence
Sex [Male] × ConditionAI-Generated × SexualActivity num	-0.07	[-0.15, 0.01]	-1.77	0.076	SexualActivity_num	Valence
Sex [Male] × ConditionAI-Generated × Feedback CouldDiscriminateTrue	-0.29	[-0.61, 0.02]	-1.83	0.068	Feedback_CouldDiscriminate	Valence
Sex [Female] × ConditionAI-Generated × Feedback CouldNotDiscriminateTrue	0.22	[ 0.01, 0.42]	2.02	0.043	Feedback_CouldNotDiscriminate	Valence
Sex [Male] × ConditionAI-Generated × Feedback AILessArousingTrue	-0.27	[-0.49, -0.04]	-2.32	0.021	Feedback_AILessArousing	Valence

Emotionality (Combined Factor)

Here we combined the data for arousal, valence and enticing into an emotionality factor.

pca <- principal_components(df[c("Valence", "Arousal", "Enticement")], n = 2)
pca

# Loadings from Principal Component Analysis (no rotation)

Variable   |  PC1 |   PC2 | Complexity
--------------------------------------
Valence    | 0.79 |  0.61 |       1.88
Arousal    | 0.87 | -0.37 |       1.34
Enticement | 0.91 | -0.18 |       1.08

The 2 principal components accounted for 91.45% of the total variance of the original data (PC1 = 73.65%, PC2 = 17.80%).

df$Emotionality <- normalize(predict(pca)$PC1)

emotionality <- df |> 
  mutate(Emotionality_Extreme = ifelse(Emotionality %in% c(0, 1), "Extreme", "Not extreme")) |>
  ggplot(aes(x=Emotionality, fill=Relevance)) +
  geom_histogram(aes(color=Emotionality_Extreme), position = "identity", bins=80) +
  facet_grid(Relevance~Sex, scales="free_y") +
  scale_color_manual(values=c("red", "white"), guide="none") +
  scale_fill_manual(values=c("Relevant"="#E91E63", "Irrelevant"="#FF9800", "Non-erotic"="#03A9F4"), guide="none") +
  scale_y_sqrt(expand = c(0, 0)) +
  labs(title = "Emotionality") +
  theme_abyss() +
  theme(plot.title = element_text(hjust = 0.5, face="bold"))

emotionality

emotionality_desc <- plot_descriptive(df, "Emotionality")
emotionality_desc

Whole Sample

This model looks at the effect of sex, relevance and condition on emotionality, accounting for random variability due to participants and items (i.e., random effects).

These findings suggest that females consistently rate relevant (photograph) images lower in emotionality compared to males. Additionally, males specifically rate irrelevant (photograph) images lower in emotionality, and both genders rate non-erotic (photgraph) images lower in emotionality compared to relevant images. Furthermore, males rate AI-generated relevant images lower in emotionality compared to those labelled as photographs.

For males, emotionality ratings are on average significantly lower for AI-Generated images compared to Photograph images. This indicates that males generally rate AI-Generated images lower in emotionality compared to photographs. For females, there is no significant difference between the mean ratings of AI-Generated and Photograph images in terms of emotionality. This suggests that females do not show a significant preference or difference in emotionality ratings between AI-Generated and Photograph images.

m <- glmmTMB::glmmTMB(Emotionality ~ Sex / Relevance / Condition + (Relevance | Participant) + (1|Item),
                      data=df, family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

results_table(m)

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
(Intercept)	0.81	0.07	(0.67, 0.96)	10.91	< .001
Sex (Female)	-1.21	0.11	(-1.43, -1.00)	-11.04	< .001
Sex (Male) × RelevanceIrrelevant	-2.26	0.11	(-2.47, -2.05)	-21.03	< .001
Sex (Female) × RelevanceIrrelevant	0.03	0.12	(-0.20, 0.26)	0.23	0.820
Sex (Male) × RelevanceNon-erotic	-1.80	0.10	(-1.99, -1.60)	-17.91	< .001
Sex (Female) × RelevanceNon-erotic	-0.48	0.11	(-0.68, -0.27)	-4.48	< .001
Sex (Male) × RelevanceRelevant × ConditionAI-Generated	-0.10	0.01	(-0.13, -0.07)	-6.72	< .001
Sex (Female) × RelevanceRelevant × ConditionAI-Generated	0.02	0.02	(-0.02, 0.06)	0.94	0.348
Sex (Male) × RelevanceIrrelevant × ConditionAI-Generated	-0.02	0.02	(-0.05, 0.01)	-1.07	0.286
Sex (Female) × RelevanceIrrelevant × ConditionAI-Generated	7.95e-05	0.02	(-0.04, 0.04)	3.97e-03	0.997
Sex (Male) × RelevanceNon-erotic × ConditionAI-Generated	-0.03	0.01	(-0.06, 5.16e-04)	-1.93	0.054
Sex (Female) × RelevanceNon-erotic × ConditionAI-Generated	5.20e-03	0.02	(-0.03, 0.04)	0.26	0.795

marginaleffects::avg_slopes(m, variables="Condition", by="Sex", re.form=NA, backend ="emmeans") |> 
  results_table()

Warning: These arguments are not known to be supported for models of class `glmmTMB`: backend. These arguments are known to be valid: re.form, allow.new.levels, zitype. All arguments are still passed to the model-specific prediction function, but users are encouraged to check if the argument is
indeed supported by their modeling package. Please file a request on Github if you believe that an unknown argument should be added to the `marginaleffects` white list of known arguments, in order to avoid raising this warning: https://github.com/vincentarelbundock/marginaleffects/issues
Warning: These arguments are not known to be supported for models of class `glmmTMB`: backend. These arguments are known to be valid: re.form, allow.new.levels, zitype. All arguments are still passed to the model-specific prediction function, but users are encouraged to check if the argument is
indeed supported by their modeling package. Please file a request on Github if you believe that an unknown argument should be added to the `marginaleffects` white list of known arguments, in order to avoid raising this warning: https://github.com/vincentarelbundock/marginaleffects/issues

Comparison	Sex	Coefficient	z	p	95% CI
AI-Generated - Photograph	Male	-9.98e-03	-5.92	< .001	[-0.01, -0.01]
AI-Generated - Photograph	Female	1.87e-03	0.70	0.482	[ 0.00, 0.01]

Subsamples

Sample

This model investigates the influence of sample type (General Population, Incentivized Students, Non-incentivized Students), Sex, Relevance, and Condition (AI-Generated vs. Photograph) on emotionality ratings. Males from the General Population consistently rated AI-Generated images lower in emotionality compared to photographs across relevant conditions. Females from the Non-incentivized Students group showed a significant preference for AI-Generated images in terms of emotionality compared to photographs.

These findings suggest that the perception of emotionality in AI-Generated images varies significantly based on sample type (general population vs. students) and the relevance of the images, particularly for males and non-incentivized female students.

m <- glmmTMB::glmmTMB(Emotionality ~ Sample / Sex / Relevance / Condition + (Relevance | Participant) + ((Relevance/Condition)|Item),
                      data=df, 
                      family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; non-positive-definite Hessian matrix. See vignette('troubleshooting')

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; singular convergence (7). See vignette('troubleshooting'), help('diagnose')

results_table(m, filter="Condition")

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
Sample (General Population) × SexMale × RelevanceRelevant × ConditionAI-Generated	-0.12	0.02	(-0.15, -0.08)	-6.35	< .001
Sample (Students (Incentivized)) × SexMale × RelevanceRelevant × ConditionAI-Generated	-0.09	0.05	(-0.19, 7.93e-03)	-1.80	0.071
Sample (Students (Non-incentivized)) × SexMale × RelevanceRelevant × ConditionAI-Generated	-0.02	0.05	(-0.11, 0.07)	-0.51	0.610
Sample (General Population) × SexFemale × RelevanceRelevant × ConditionAI-Generated	-7.06e-04	0.04	(-0.08, 0.08)	-0.02	0.985
Sample (Students (Incentivized)) × SexFemale × RelevanceRelevant × ConditionAI-Generated	-0.04	0.03	(-0.11, 0.03)	-1.16	0.246
Sample (Students (Non-incentivized)) × SexFemale × RelevanceRelevant × ConditionAI-Generated	0.09	0.04	(0.02, 0.16)	2.62	0.009
Sample (General Population) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	-0.02	0.02	(-0.05, 0.02)	-0.97	0.334
Sample (Students (Incentivized)) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	-0.11	0.06	(-0.23, 0.01)	-1.75	0.080
Sample (Students (Non-incentivized)) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	0.04	0.05	(-0.07, 0.15)	0.74	0.457
Sample (General Population) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	0.02	0.04	(-0.05, 0.10)	0.61	0.540
Sample (Students (Incentivized)) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	-0.02	0.03	(-0.09, 0.04)	-0.73	0.464
Sample (Students (Non-incentivized)) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	9.76e-03	0.03	(-0.06, 0.08)	0.28	0.779
Sample (General Population) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	-0.03	0.02	(-0.06, 7.29e-03)	-1.52	0.129
Sample (Students (Incentivized)) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	2.80e-03	0.05	(-0.10, 0.10)	0.05	0.957
Sample (Students (Non-incentivized)) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	-0.08	0.04	(-0.17, 3.56e-03)	-1.88	0.060
Sample (General Population) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	8.97e-04	0.04	(-0.07, 0.07)	0.02	0.981
Sample (Students (Incentivized)) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	7.45e-03	0.03	(-0.06, 0.07)	0.22	0.823
Sample (Students (Non-incentivized)) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	5.43e-03	0.03	(-0.06, 0.07)	0.16	0.869

Mobile

This analysis explores the influence of using a mobile device to complete the experiment on emotionality ratings, focusing on differences between males and females across relevant and irrelevant conditions of AI-Generated and photograph images.

Males consistently rate emotionality significantly lower for relevant AI-Generated images compared to photograph images, irrespective of whether they use a mobile device or not. They also rate emotionality significantly lower for irrelevant AI-Generated images compared to photograph images when not using a mobile device. For females, there are no significant differences in how they rate emotionality for AI-Generated images based on whether they use a mobile device or not, regardless of relevance.

These findings highlight that the perception of emotionality in AI-Generated images among males is notably affected by the use of mobile devices during the experiment, whereas females show no significant variation in their ratings based on device usage.

m <- glmmTMB::glmmTMB(Emotionality ~ Mobile / Sex / Relevance / Condition + (Relevance | Participant) + ((Relevance/Condition)|Item),
                      data=df, 
                      family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; non-positive-definite Hessian matrix. See vignette('troubleshooting')

results_table(m, filter="Condition")

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
Mobile (False) × SexMale × RelevanceRelevant × ConditionAI-Generated	-0.09	0.02	(-0.13, -0.05)	-4.32	< .001
Mobile (True) × SexMale × RelevanceRelevant × ConditionAI-Generated	-0.13	0.03	(-0.19, -0.08)	-5.15	< .001
Mobile (False) × SexFemale × RelevanceRelevant × ConditionAI-Generated	0.03	0.02	(-0.02, 0.07)	1.04	0.300
Mobile (True) × SexFemale × RelevanceRelevant × ConditionAI-Generated	-0.01	0.04	(-0.09, 0.07)	-0.26	0.794
Mobile (False) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	-0.05	0.02	(-0.09, -9.11e-03)	-2.39	0.017
Mobile (True) × SexMale × RelevanceIrrelevant × ConditionAI-Generated	0.04	0.03	(-0.02, 0.09)	1.34	0.179
Mobile (False) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	-0.01	0.02	(-0.06, 0.04)	-0.46	0.642
Mobile (True) × SexFemale × RelevanceIrrelevant × ConditionAI-Generated	0.04	0.04	(-0.04, 0.12)	0.94	0.347
Mobile (False) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	-0.02	0.02	(-0.06, 0.01)	-1.27	0.203
Mobile (True) × SexMale × RelevanceNon-erotic × ConditionAI-Generated	-0.04	0.03	(-0.09, 9.13e-03)	-1.60	0.111
Mobile (False) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	6.65e-03	0.02	(-0.04, 0.05)	0.29	0.769
Mobile (True) × SexFemale × RelevanceNon-erotic × ConditionAI-Generated	-1.14e-03	0.04	(-0.08, 0.08)	-0.03	0.978

Moderators

Looking at AI-attitudes, porn frequency and usage, sexual activity, and feedback as moderators, the analysis showed that for females, having a positive attitude towards AI significantly increased their emotionality ratings for AI-generated images compared to those who did not have such positive attitudes. Additionally, for females reporting finding AI more arousing lowered their emotionality ratings for AI-generated images compared to those that did not. This effect was significantly reversed for males.

for(m in c("BAIT_Visual", "BAIT_Text", "AI_Knowledge",
           "GAAIS_Positive", "GAAIS_Negative",
           "Porn", "PornFrequency", "SexualActivity_num")) {
  moderator_results <- rbind(moderator_results,
                             test_moderator(df, "Emotionality", m))
}

display(filter(moderator_results, Outcome == "Emotionality"))

Parameter	Coefficient	95% CI	z	p	Moderator	Outcome
Sex [Male] × ConditionAI-Generated × BAIT Visual	0.04	[-0.17, 0.25]	0.34	0.737	BAIT_Visual	Emotionality
Sex [Female] × ConditionAI-Generated × BAIT Visual	0.23	[-0.10, 0.57]	1.37	0.170	BAIT_Visual	Emotionality
Sex [Male] × ConditionAI-Generated × BAIT Text	0.03	[-0.12, 0.18]	0.38	0.701	BAIT_Text	Emotionality
Sex [Female] × ConditionAI-Generated × BAIT Text	0.10	[-0.12, 0.33]	0.90	0.370	BAIT_Text	Emotionality
Sex [Male] × ConditionAI-Generated × AI Knowledge	0.01	[-0.02, 0.04]	0.67	0.504	AI_Knowledge	Emotionality
Sex [Female] × ConditionAI-Generated × AI Knowledge	0.01	[-0.03, 0.05]	0.62	0.536	AI_Knowledge	Emotionality
Sex [Male] × ConditionAI-Generated × GAAIS Positive	0.06	[-0.11, 0.23]	0.71	0.475	GAAIS_Positive	Emotionality
Sex [Female] × ConditionAI-Generated × GAAIS Positive	0.28	[ 0.03, 0.53]	2.18	0.029	GAAIS_Positive	Emotionality
Sex [Male] × ConditionAI-Generated × GAAIS Negative	3.08e-03	[-0.15, 0.15]	0.04	0.968	GAAIS_Negative	Emotionality
Sex [Female] × ConditionAI-Generated × GAAIS Negative	0.13	[-0.11, 0.37]	1.08	0.279	GAAIS_Negative	Emotionality
Sex [Male] × ConditionAI-Generated × Porn	-0.04	[-0.09, 0.01]	-1.52	0.128	Porn	Emotionality
Sex [Female] × ConditionAI-Generated × Porn	-0.06	[-0.13, 0.00]	-1.85	0.065	Porn	Emotionality
Sex [Male] × ConditionAI-Generated × PornFrequency	-0.01	[-0.03, 0.01]	-0.98	0.329	PornFrequency	Emotionality
Sex [Female] × ConditionAI-Generated × PornFrequency	-5.61e-03	[-0.04, 0.03]	-0.30	0.764	PornFrequency	Emotionality
Sex [Male] × ConditionAI-Generated × SexualActivity num	-0.01	[-0.04, 0.02]	-0.65	0.518	SexualActivity_num	Emotionality
Sex [Female] × ConditionAI-Generated × SexualActivity num	-0.03	[-0.06, 0.01]	-1.38	0.168	SexualActivity_num	Emotionality

For females reporting finding AI more arousing lowered their emotionality ratings for AI-generated images compared to those that did not. This effect was significantly reversed for males.

for(m in c("Feedback_CouldDiscriminate", "Feedback_CouldNotDiscriminate",
           "Feedback_AIMoreArousing", "Feedback_AILessArousing", 
           "Feedback_LabelsIncorrect", "Feedback_LabelsReversed")) {
  feedback_results <- rbind(feedback_results,
                            test_moderator(df, "Emotionality", m))
}
display(filter(feedback_results, Outcome == "Emotionality"))

Parameter	Coefficient	95% CI	z	p	Moderator	Outcome
Sex [Male] × ConditionAI-Generated × Feedback CouldDiscriminateTrue	-0.05	[-0.19, 0.08]	-0.76	0.445	Feedback_CouldDiscriminate	Emotionality
Sex [Female] × ConditionAI-Generated × Feedback CouldDiscriminateTrue	-0.16	[-0.36, 0.03]	-1.66	0.096	Feedback_CouldDiscriminate	Emotionality
Sex [Male] × ConditionAI-Generated × Feedback CouldNotDiscriminateTrue	0.05	[-0.02, 0.12]	1.36	0.175	Feedback_CouldNotDiscriminate	Emotionality
Sex [Female] × ConditionAI-Generated × Feedback CouldNotDiscriminateTrue	-8.43e-03	[-0.11, 0.09]	-0.17	0.864	Feedback_CouldNotDiscriminate	Emotionality
Sex [Male] × ConditionAI-Generated × Feedback AIMoreArousingTrue	0.04	[-0.09, 0.16]	0.59	0.555	Feedback_AIMoreArousing	Emotionality
Sex [Female] × ConditionAI-Generated × Feedback AIMoreArousingTrue	0.12	[-0.01, 0.25]	1.79	0.074	Feedback_AIMoreArousing	Emotionality
Sex [Male] × ConditionAI-Generated × Feedback AILessArousingTrue	-0.17	[-0.28, -0.06]	-2.94	0.003	Feedback_AILessArousing	Emotionality
Sex [Female] × ConditionAI-Generated × Feedback AILessArousingTrue	0.14	[-0.03, 0.30]	1.64	0.101	Feedback_AILessArousing	Emotionality
Sex [Male] × ConditionAI-Generated × Feedback LabelsIncorrectTrue	-0.04	[-0.11, 0.03]	-1.23	0.219	Feedback_LabelsIncorrect	Emotionality
Sex [Female] × ConditionAI-Generated × Feedback LabelsIncorrectTrue	-0.06	[-0.16, 0.05]	-1.05	0.294	Feedback_LabelsIncorrect	Emotionality
Sex [Male] × ConditionAI-Generated × Feedback LabelsReversedTrue	-0.05	[-0.21, 0.12]	-0.56	0.578	Feedback_LabelsReversed	Emotionality
Sex [Female] × ConditionAI-Generated × Feedback LabelsReversedTrue	0.09	[-0.09, 0.27]	0.96	0.336	Feedback_LabelsReversed	Emotionality

Believed

m <- glmmTMB::glmmTMB(Emotionality ~ Sex / Relevance / Condition + (Relevance | Participant) + ((Relevance/Condition)|Item),
                      data=filter(df, ConditionBelief == "True"), 
                      family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

Warning in finalizeTMB(TMBStruc, obj, fit, h, data.tmb.old): Model convergence problem; non-positive-definite Hessian matrix. See vignette('troubleshooting')

results_table(m)

Fixed Effects

Parameter	Coefficient	SE	95% CI	z	p
(Intercept)	0.87	0.10	(0.67, 1.07)	8.55	< .001
Sex (Female)	-1.15	0.15	(-1.44, -0.86)	-7.79	< .001
Sex (Male) × RelevanceIrrelevant	-2.27	0.12	(-2.50, -2.03)	-18.55	< .001
Sex (Female) × RelevanceIrrelevant	2.80e-04	0.13	(-0.26, 0.26)	2.14e-03	0.998
Sex (Male) × RelevanceNon-erotic	-1.85	0.12	(-2.08, -1.61)	-15.27	< .001
Sex (Female) × RelevanceNon-erotic	-0.57	0.13	(-0.82, -0.32)	-4.51	< .001
Sex (Male) × RelevanceRelevant × ConditionAI-Generated	-0.46	0.05	(-0.55, -0.36)	-9.49	< .001
Sex (Female) × RelevanceRelevant × ConditionAI-Generated	-0.37	0.05	(-0.48, -0.27)	-7.03	< .001
Sex (Male) × RelevanceIrrelevant × ConditionAI-Generated	-0.15	0.03	(-0.22, -0.08)	-4.36	< .001
Sex (Female) × RelevanceIrrelevant × ConditionAI-Generated	-0.25	0.04	(-0.33, -0.17)	-6.11	< .001
Sex (Male) × RelevanceNon-erotic × ConditionAI-Generated	-0.16	0.03	(-0.22, -0.10)	-5.09	< .001
Sex (Female) × RelevanceNon-erotic × ConditionAI-Generated	-0.13	0.05	(-0.22, -0.04)	-2.93	0.003

moderator_believed <- data.frame()
for(m in c("BAIT_Visual", "BAIT_Text", "AI_Knowledge",
           "GAAIS_Positive", "GAAIS_Negative",
           "Porn", "PornFrequency", "SexualActivity_num", 
           "Feedback_CouldDiscriminate", "Feedback_CouldNotDiscriminate",
           "Feedback_AIMoreArousing", "Feedback_AILessArousing", 
           "Feedback_LabelsIncorrect", "Feedback_LabelsReversed")) {
  moderator_believed <- rbind(moderator_believed,
                             test_moderator(filter(df, ConditionBelief == "True"), "Emotionality", m))
}

display(filter(moderator_believed, p < .1))

Parameter	Coefficient	95% CI	z	p	Moderator	Outcome
Sex [Male] × ConditionAI-Generated × BAIT Text	0.31	[-0.03, 0.64]	1.78	0.076	BAIT_Text	Emotionality
Sex [Female] × ConditionAI-Generated × Porn	-0.12	[-0.27, 0.02]	-1.65	0.099	Porn	Emotionality
Sex [Female] × ConditionAI-Generated × Feedback CouldDiscriminateTrue	-0.36	[-0.75, 0.02]	-1.83	0.067	Feedback_CouldDiscriminate	Emotionality
Sex [Male] × ConditionAI-Generated × Feedback AILessArousingTrue	-0.36	[-0.58, -0.15]	-3.31	< .001	Feedback_AILessArousing	Emotionality

Summary

• Believing erotic images to be AI-generated, regardless of its true nature, leads to a less pleasurable experience (i.e., a less intense and less positive emotion). While the effect is the most noticeable for men (who experience a stronger arousal when exposed to the images), it appears overall very small in magnitude. However, it becomes more pronounced for both genders when controlled for the belief in the images’ nature, and is consistent with participants’ feedback, suggesting an important meta-cognitive component (in the form of explicitly conscious beliefs).
• The items developed (BAIT scale) seem to measure a different dimension to AI-expertise.

Figures

pred1 <- 
  rbind(
  modelbased::estimate_means(m_arousal, by=c("Sex", "Relevance", "Condition"), backend = "emmeans") |>
  as.data.frame() |> 
  mutate(Outcome="Arousal"),
  modelbased::estimate_means(m_enticement, by=c("Sex", "Relevance", "Condition"),backend = "emmeans") |>
  as.data.frame() |> 
  mutate(Outcome="Enticement"),
  modelbased::estimate_means(m_valence, by=c("Sex", "Relevance", "Condition"),backend = "emmeans") |>
  as.data.frame() |> 
  mutate(Outcome="Valence"))

stars1 <- data.frame(label=c("***", "***", "*", "***"),
                    Sex=fct_relevel(c("Male", "Male", "Male", "Male"), "Male"),
                    Relevance=c("Relevant", "Relevant", "Non-erotic", "Relevant"),
                    Outcome=c("Arousal", "Enticement", "Enticement", "Valence"),
                    y=c(0.75, 0.8, 0.2, 0.8))


p1 <- pred1 |>
  mutate(Sex = fct_relevel(Sex, "Male", "Female")) |> 
  ggplot(aes(x=Relevance, y=Proportion)) +
  geom_hline(yintercept=0, color="darkgrey") +
  geom_line(aes(group=Condition, color=Condition), position = position_dodge(width=0.2), alpha=0.8, linewidth=0.2) +
  geom_pointrange(aes(ymin = CI_low, ymax=CI_high, color=Condition), position = position_dodge(width=0.2), size=0.2) +
  geom_text(data=stars1, aes(y=y, label=label), hjust=0.5, size=3) +
  geom_rect(data=data.frame(Type="All trials", Proportion=0.1, Sex=as.factor("Male"), Outcome="Arousal", Relevance="Relevant"), 
            aes(xmin=0.8, xmax=1.2, ymin=0.5, ymax=0.80), color="#8BC34A", alpha=0) +
  geom_rect(data=data.frame(Type="All trials", Proportion=0.1, Sex=as.factor("Female"), Outcome="Arousal", Relevance="Relevant"), 
            aes(xmin=0.8, xmax=1.2, ymin=0.2, ymax=0.5), color="#8BC34A", alpha=0) +
  facet_grid(Outcome~Sex, switch="y") +
  scale_y_continuous(limits=c(0, 1), labels=scales::percent) +
  scale_color_manual(values=c("AI-Generated"="#2196F3", "Photograph"="#F44336")) +
  theme_minimal() +
  theme(axis.title.y = element_blank(),
        axis.title.x = element_blank(),
        axis.text.y = element_text(size = 8),
        strip.placement = "outside",
        strip.background.x = element_rect(fill=c("lightgrey"), color=NA),
        strip.text.x = element_text(size = 10),
        strip.text.y = element_text(size = 10),
        axis.text.x = element_text(angle=25, hjust=1, vjust=1.1, size=9, color="black"),
        legend.text = element_text(size = 10),
        legend.position = "bottom") +
  labs(color="Images presented as:",
       tag = "A")
p1

m1 <- glmmTMB::glmmTMB(Arousal ~ Sex / Condition + (Condition | Participant) + (1|Item),
                      data=filter(df, Relevance=="Relevant"), 
                      family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))
m2 <- glmmTMB::glmmTMB(Arousal ~ Sex / Condition + (Condition | Participant) + (1|Item),
                      data=filter(df, Relevance=="Relevant", ConditionBelief == "True"), 
                      family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))
m3 <- glmmTMB::glmmTMB(Arousal ~ Sex / Condition  + (Condition | Participant) + (1|Item),
                      data=filter(df, Relevance=="Relevant", ConditionBelief == "True", Feedback_AILessArousing=="True"), 
                      family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

stars2 <- data.frame(label=c("***", "***", "***", "***", "*"), 
                    Sex=fct_relevel(c("Male", "Male", "Male", "Female", "Female"), 
                                    "Male", "Female"), 
                    Type=c("All trials", "Believed trials", "'AI was less arousing'", "Believed trials", "'AI was less arousing'"), 
                    y=c(0.76, 0.76, 0.76, 0.45, 0.45))

pred2 <- rbind(
  estimate_relation(m1) |> 
    mutate(Type = "All trials"),
  estimate_relation(m2) |> 
    mutate(Type = "Believed trials"),
  estimate_relation(m3) |> 
    mutate(Type = "'AI was less arousing'")
)

p2 <- pred2 |> 
  mutate(Type = fct_relevel(Type, "All trials", "Believed trials", "'AI was less arousing'")) |> 
  ggplot(aes(x=Type, y=Predicted)) +
  geom_hline(yintercept=0, color="darkgrey") +
  geom_line(aes(group=Condition, color=Condition), position = position_dodge(width=0.2), alpha=0.8, linewidth=0.1) +
  geom_bar(stat="identity", aes(fill=Condition), position = position_dodge(width=0.2), width=0.2, alpha=0.6) +
  geom_pointrange(aes(ymin = CI_low, ymax=CI_high, color=Condition), position = position_dodge(width=0.2), size=0.2) +
  geom_rect(data=data.frame(Type="All trials", Predicted=0.1, Sex=as.factor("Male")), 
            aes(xmin=0.7, xmax=1.3, ymin=0.55, ymax=0.7), color="#8BC34A", alpha=0) +
  geom_rect(data=data.frame(Type="All trials", Predicted=0.1, Sex=as.factor("Female")), 
            aes(xmin=0.7, xmax=1.3, ymin=0.25, ymax=0.4), color="#8BC34A", alpha=0) +
  geom_text(data=stars2, aes(y=y, label=label), hjust=0.5, size=3) +
  facet_grid(~Sex) +
  scale_y_continuous(labels=scales::percent) +
  scale_color_manual(values=c("AI-Generated"="#2196F3", "Photograph"="#F44336")) +
  scale_fill_manual(values=c("AI-Generated"="#2196F3", "Photograph"="#F44336")) +
  labs(y="Arousal",
       tag = "B") +
  theme_minimal() +
  theme(axis.title.x = element_blank(),
        axis.text.x = element_text(angle=25, hjust=1, vjust=1.1, size=9),
        axis.text.y = element_text(size = 8),
        axis.title.y = element_text(size = 10),
        strip.placement = "outside",
        strip.background.x = element_rect(fill=c("lightgrey"), color=NA),
        strip.text.x = element_text(size = 10),
        legend.position = "none")


p1 | p2

m6 <- glm(Feedback_AILessArousing ~ Sex / BAIT_Visual, 
    data=mutate(dfsub, Feedback_AILessArousing = ifelse(Feedback_AILessArousing=="True", 1, 0)), 
    family="binomial")

results_table(m6)

Parameter	Log-Odds	SE	95% CI	z	p
(Intercept)	-0.65	0.99	(-2.66, 1.26)	-0.66	0.510
Sex (Male)	0.37	1.16	(-1.87, 2.69)	0.32	0.747
Sex (Female) × BAIT Visual	-2.35	1.48	(-5.30, 0.54)	-1.59	0.112
Sex (Male) × BAIT Visual	-2.79	0.93	(-4.64, -1.00)	-3.02	0.003

pred3 <- estimate_relation(m6, length=50) 

p3 <- pred3 |> 
  mutate(Sex = fct_relevel(Sex, "Male", "Female")) |> 
  ggplot(aes(x=BAIT_Visual, y=Predicted)) +
  geom_hline(yintercept=0, color="darkgrey") +
  geom_ribbon(aes(ymin=CI_low, ymax=CI_high), alpha=0.3) + 
  geom_line() +
  geom_text(data=data.frame(Sex = as.factor("Male")), aes(x=0.65, y=0.25), label="**") +
  facet_grid(~Sex) +
  scale_y_continuous(labels=scales::percent, expand=c(0, 0)) +
  scale_x_continuous(breaks = c(0.3, 0.5, 0.7, 0.9)) +
  labs(y = "\nProbability of 'AI was less arousing' Feedback\n",
       x = "\nBeliefs about Artificial Information Technology (BAIT)\n'AI algorithms can generate realistic images'",
       tag = "C") +
  theme_minimal() +
  theme(strip.text = element_blank(),
        axis.text.y = element_text(size = 8),
        axis.title.y = element_text(size = 10),
        axis.text.x = element_text(size = 8),
        axis.title.x = element_text(size = 10))


p4 <- patchwork::wrap_plots(p2 / p3 + plot_layout(heights = c(1, 1.6)))
p <- patchwork::wrap_plots(p1 | p4, guides="collect")
p

ggsave("figures/figure1.png", p, width=10, height=8, dpi=300)

m4 <- glmmTMB::glmmTMB(Arousal ~ SexualActivity / Sex / Condition + (Condition | Participant) + (1|Item),
                      data=filter(df, Relevance=="Relevant", ConditionBelief == "True"), 
                      family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))

summary(m4)

pred3 <- estimate_relation(m4) 
pred3 |> 
  mutate(xnum = as.numeric(as.factor(SexualActivity))) |>
  ggplot(aes(x=SexualActivity, y=Predicted)) +
  # geom_ribbon(aes(x=xnum, ymin=CI_low, ymax=CI_high, fill=Condition), alpha=0.5) +
  geom_line(aes(group=Condition, color=Condition), position = position_dodge(width=0.2)) +
  geom_pointrange(aes(ymin = CI_low, ymax=CI_high, color=Condition), position = position_dodge(width=0.2)) +
  facet_grid(~Sex) 

m5 <- glmmTMB::glmmTMB(Arousal ~ Sex / (Condition * GAAIS_Negative) + (Condition | Participant) + (1|Item),
                      data=filter(df, Relevance=="Relevant", ConditionBelief == "True"), 
                      family=glmmTMB::ordbeta(),
                      control = glmmTMB::glmmTMBControl(parallel = 8))
summary(m5)

pred4 <- estimate_relation(m5, length=30) 
pred4 |> 
  ggplot(aes(x=GAAIS_Negative, y=Predicted)) +
  geom_line(aes(group=Condition, color=Condition)) +
  facet_grid(~Sex)