Embedding scores:

EMBEDDING_SCORE_PATH <- "embedding_gender_bias_occupations_gabriel.csv"

embedding_scores <- read_csv(EMBEDDING_SCORE_PATH, 
                             col_names = c("wiki_lang_code", "occupation", "female_targets",
                                           "male_targets", "male_score")) %>%
  filter(!(occupation %in% c("male_targets", "female_targets")))

count(embedding_scores, wiki_lang_code) %>%
  kable()
wiki_lang_code n
de 121
en 125
fr 123 
embedding_scores %>%
  ggplot(aes(x = male_score, fill = wiki_lang_code)) +
  geom_histogram() +
  facet_wrap(~wiki_lang_code) +
  theme_classic() +
  theme(legend.position = "none")

Human ratings:

S1 <- read_csv("gabriel_norms_clean_S1.csv") %>%
  mutate(type = "S1",
         gender = "NA")

S1g <- read_csv("gabriel_norms_clean_S1g.csv") %>%
  mutate(type = "S1g")

S2 <- read_csv("gabriel_norms_clean_S2.csv") %>%
  mutate(type = "S2",
          gender = "NA") %>%
  filter(!is.na(occupation))

S2g <- read_csv("gabriel_norms_clean_S2g.csv") %>%
  mutate(type = "S2g")

human_ratings <- bind_rows(list(S1, S1g, S2, S2g)) 

count(human_ratings, wiki_lang_code, type) %>%
  kable()
wiki_lang_code type n
de S1 126
de S1g 252
de S2 126
de S2g 252
en S1 126
en S1g 252
en S2 126
en S2g 252
fr S1 126
fr S1g 252
fr S2 126
fr S2g 252 
human_ratings %>%
  ggplot(aes(x = human_mean_rating, fill = type)) +
  geom_density(alpha = .4) +
  facet_wrap(~wiki_lang_code) +
  theme_classic() +
  theme(legend.position = "none")

all_scores <- embedding_scores %>%
  select(-female_targets, -male_targets) %>%
  full_join(human_ratings) %>%
  filter(!is.na(human_mean_rating)) # fix aupairs!

all_scores %>%
  ggplot(aes(y = human_mean_rating, x = male_score, 
             color = gender, group = gender)) +
  geom_point(aes(color = wiki_lang_code)) +
  xlab("embedding male score") +
  ylab("human male score") +
  geom_smooth(method = "lm") + 
  facet_grid(type~wiki_lang_code, scale = "free") +
  theme_classic() +
  theme(legend.position = "none")

full_df <- human_ratings %>%
  filter(type %in% c("S1", "S2")) %>%
  rename(human_wiki_lang_code = wiki_lang_code) %>%
  left_join(embedding_scores %>%
  select(-female_targets, -male_targets)  %>%
  rename(embedding_wiki_lang_code = wiki_lang_code)) %>%
  mutate(same_lang = human_wiki_lang_code == embedding_wiki_lang_code)

full_df %>%
  ggplot(aes(y = human_mean_rating, 
             x = male_score, 
             group = embedding_wiki_lang_code,
             color = embedding_wiki_lang_code)) +
  xlab("embedding male score") +
  ylab("human male score") +
  geom_smooth(method = "lm", aes(linetype = same_lang)) + 
  facet_grid(type~human_wiki_lang_code) +
  theme_classic()