SCORE_PATH <- here("analysis/books/adult_comparision/correlation_analysis_2/tidy_gender_scores.csv")
model_biases <- read_csv(SCORE_PATH)


model_scores <- model_biases %>%
  group_by(model_type, word) %>%
  summarize(male_score = mean(male_score)) 
# Merge with human data
GENDER_NORMS <- here("data/processed/words/gender_ratings_mean.csv")

gender_words <- read_csv(GENDER_NORMS)
gender_norms <- gender_words %>%
  mutate(word = map_chr(word, ~unlist(str_split(.x, " "))[[1]]),
         word = tolower(word),
         word = str_remove_all(word, '[:punct:]')) %>%
  distinct(word, .keep_all = T) %>%
  group_by(word) %>%
  summarize(human_gender_rating  = mean(mean, na.rm = T))


all_scores <- gender_norms %>%
  inner_join(model_scores)
all_scores %>%
  ggplot(aes( x = male_score,
              y = human_gender_rating,
              color = model_type)) +
  geom_point(size = .5, alpha = .2)+
  geom_smooth(method = "lm") +
  theme_classic()

all_scores %>%
  group_by(model_type) %>%
  nest() %>%
  mutate(test = map(data, ~ tidy(cor.test(.x$male_score, 
                                          .x$human_gender_rating)))) %>%
  select(-data) %>%
  unnest()  %>%
  select(1:7) %>%
  mutate(sig = case_when(p.value  < .05  ~ "*", TRUE ~ "")) %>%
  kable()
model_type estimate statistic p.value parameter conf.low conf.high sig
coca -0.4080415 -15.36596 0 1182 -0.4544480 -0.3594261 *
kid -0.3612783 -12.93749 0 1115 -0.4112195 -0.3091748 * 
 gender_norms %>% 
  inner_join(model_biases) %>%
  group_by(model_type, run) %>%
  nest() %>%
  mutate(test = map(data, ~ tidy(cor.test(.x$male_score, 
                                          .x$human_gender_rating)))) %>%
  select(-data) %>%
  unnest()  %>%
  select(1:8) %>%
  mutate(sig = case_when(p.value  < .05  ~ "*", TRUE ~ "")) %>%
  kable()
model_type run estimate statistic p.value parameter conf.low conf.high sig
coca 1 -0.3005056 -9.087952 0 832 -0.3610265 -0.2374641 *
coca 10 -0.3015096 -8.949804 0 801 -0.3631195 -0.2372746 *
coca 2 -0.3352277 -10.244789 0 829 -0.3942478 -0.2734536 *
coca 3 -0.3960748 -12.329411 0 817 -0.4523073 -0.3367057 *
coca 4 -0.2815993 -8.383340 0 816 -0.3435154 -0.2172458 *
coca 5 -0.3375960 -10.307739 0 826 -0.3966052 -0.2758084 *
coca 6 -0.3766578 -11.699999 0 828 -0.4335935 -0.3167266 *
coca 7 -0.3281883 -9.894183 0 811 -0.3881862 -0.2654202 *
coca 8 -0.2932725 -8.874891 0 837 -0.3539302 -0.2301579 *
coca 9 -0.3610774 -11.181957 0 834 -0.4186327 -0.3006332 *
kid 1 -0.3392428 -12.041968 0 1115 -0.3901350 -0.2862842 *
kid 10 -0.3244020 -11.451617 0 1115 -0.3759045 -0.2709014 *
kid 2 -0.3497329 -12.465340 0 1115 -0.4001790 -0.2971738 *
kid 3 -0.3341407 -11.837901 0 1115 -0.3852454 -0.2809926 *
kid 4 -0.3497485 -12.465973 0 1115 -0.4001939 -0.2971900 *
kid 5 -0.3104181 -10.904019 0 1115 -0.3624735 -0.2564319 *
kid 6 -0.3103981 -10.903241 0 1115 -0.3624543 -0.2564112 *
kid 7 -0.3521110 -12.562052 0 1115 -0.4024543 -0.2996444 *
kid 8 -0.3212276 -11.326595 0 1115 -0.3728575 -0.2676147 *
kid 9 -0.3284790 -11.612826 0 1115 -0.3798163 -0.2751246 *

Same words across all models

common_words <- count(all_scores, word) %>%
  filter(n == 2) %>%
  pull(word)

all_scores %>%
  filter(word %in% common_words) %>%
  ggplot(aes( x = male_score,
              y = human_gender_rating,
              color = model_type)) +
  geom_point(size = .5, alpha = .2)+
  geom_smooth(method = "lm") +
  theme_classic()

all_scores %>%
    filter(word %in% common_words) %>%

  group_by(model_type) %>%
  nest() %>%
  mutate(test = map(data, ~ tidy(cor.test(.x$male_score, 
                                          .x$human_gender_rating)))) %>%
  select(-data) %>%
  unnest()  %>%
  select(1:7) %>%
  mutate(sig = case_when(p.value  < .05  ~ "*", TRUE ~ "")) %>%
  kable()
model_type estimate statistic p.value parameter conf.low conf.high sig
coca -0.4562223 -14.95605 0 851 -0.5077967 -0.4013893 *
kid -0.3867795 -12.23535 0 851 -0.4424183 -0.3281747 *

Middle range only

all_scores %>%
  filter(male_score > -.1, male_score < .1) %>%
  ggplot(aes( x = male_score,
              y = human_gender_rating,
              color = model_type)) +
  geom_point(size = .5, alpha = .2)+
  geom_smooth(method = "lm") +
  theme_classic()

all_scores %>%
  filter(male_score > -.1, male_score < .1) %>%

  group_by(model_type) %>%
  nest() %>%
  mutate(test = map(data, ~ tidy(cor.test(.x$male_score, 
                                          .x$human_gender_rating)))) %>%
  select(-data) %>%
  unnest()  %>%
  select(1:7) %>%
  mutate(sig = case_when(p.value  < .05  ~ "*", TRUE ~ "")) %>%
  kable()
model_type estimate statistic p.value parameter conf.low conf.high sig
coca -0.3142719 -11.323480 0 1170 -0.3649659 -0.2617199 *
kid -0.2822794 -9.763462 0 1101 -0.3357123 -0.2270358 *