Set params

MIN_PARTICIPANTS_PER_COUNTRY <- 400

Read in IAT data.

#d <- haven::read_sav("../data/Race IAT.public.2016.sav") 
#write_feather(d, "../data/Race IAT.public.2016.feather") 

raw_iat_behavioral <- read_feather("../data/Race IAT.public.2016.feather") %>%
  rename(mixed_countryres = countryres) %>%
  mutate(mixed_countryres = trimws(mixed_countryres)) %>%
  select(D_biep.White_Good_all, birthsex, mixed_countryres, PCT_error_3467, 
         Mn_RT_all_3467, Mn_RT_all_3, Mn_RT_all_4, Mn_RT_all_6, 
         Mn_RT_all_7, anes1, N_ERROR_3, N_ERROR_4,
         N_ERROR_6, N_ERROR_7, N_3, N_4, N_6, N_7, raceomb_002, raceomb, raceombmulti) %>%
    rename(overall_iat_D_score = D_biep.White_Good_all) %>%
  data.table()
country_key <- read_csv("../../../data/other/mixed_countryres_to_country_res_key.csv")

raw_iat_behavioral_complete <- raw_iat_behavioral %>%
                                filter(!is.na(mixed_countryres), 
                                       mixed_countryres != ".",
                                       mixed_countryres != "nu",  # not clear what this refers to (it's lower case and not in codebook- Niue seems unlikely)
                                       !is.na(overall_iat_D_score)) %>%
                                left_join(country_key) %>%
                                select(-mixed_countryres)
                                      

country_ns <- raw_iat_behavioral_complete %>%
  left_join(country_key) %>%
  count(countryres)  %>%
  filter(n >= MIN_PARTICIPANTS_PER_COUNTRY) %>%
  arrange(-n) %>%
  filter(countryres != "NA")

raw_iat_behavioral_complete_dense_country <- raw_iat_behavioral_complete %>%
  filter(countryres %in% country_ns$countryres)
# same exclusions as Nosek, Banjali, & Greenwald (2002), pg. 104. 
iat_behavioral <- raw_iat_behavioral_complete_dense_country %>%
  filter(Mn_RT_all_3467 <= 1500, # RTs
         Mn_RT_all_3 <= 1800,
         Mn_RT_all_4 <= 1800,
         Mn_RT_all_6 <= 1800,
         Mn_RT_all_7 <= 1800) %>%
  filter(N_ERROR_3/N_3 <=.25, # errors
         N_ERROR_4/N_4 <=.25,
         N_ERROR_6/N_6 <=.25,
         N_ERROR_7/N_7 <=.25)

There are 47 with at least 400 participants. This includes 1002080 participants.

Country means

country_means_race <- iat_behavioral %>%
  group_by(countryres) %>%
  multi_boot_standard(col = "overall_iat_D_score")

Plot

country_means_race_with_name <- country_means_race %>%
  mutate(country_name = countrycode(countryres,
                                    "iso2c",
                                    "country.name")) %>%
   mutate(country_name = replace(country_name, 
                                countryres == "UK", "UK")) 

ggplot(country_means_race_with_name, aes(x = reorder(country_name, mean), 
                                 y = mean)) +
  geom_bar(stat = "identity", position = "dodge") +
  ggtitle("IAT race bias (2005 - 2016 data)") +
  ylab("IAT race bias") +
  xlab("Country") +
  geom_linerange(aes(ymin = ci_lower, ymax = ci_upper),  
                 position = position_dodge(width = .9)) +
  theme_bw() +
  theme(axis.text.x = element_text(angle = 60, hjust = 1),
        legend.position = "none") 

map_world <- map_data(map = "world") %>%
  mutate(country_name = region) %>%
  mutate(country_name = fct_recode(country_name,
    `United States of America` = "USA", 
    `Russian Federation` = "Russia",
    `Republic of Korea` = "South Korea")) 

map_data <- left_join(map_world, country_means_race_with_name) %>%
  rename(race_bias = mean) %>%
  filter(lat > -57, 
         long > -165) 

ggplot(map_data, aes(x = long, y = lat, 
                           group = group, 
                           fill = race_bias)) + 
  scale_fill_gradient2(midpoint = mean(map_data$race_bias, na.rm = T), 
                       low = "blue", high = "red") +
  #scale_fill_continuous(low ="white", high = "blue") + 
  geom_polygon(color = "black", size = .1) +
  #geom_polygon() +
  theme_bw()

Diversity

Migration index (bigger values -> less diverse)

diversity = read_csv("../data/countryDiversity_parsed.txt") %>%
  rename(homogeneity = diag) %>%
  mutate(country_name = countrycode(wbcode,
                                    "wb_api3c",
                                    "country.name"),
         country_code = countrycode(wbcode,
                                    "wb_api3c",
                                    "iso2c"))

map_data <- left_join(map_world, diversity, by = c("country_name")) %>%
  filter(lat > -57, 
         long > -165)

ggplot(map_data, aes(x = long, y = lat, 
                           group = group, 
                           fill = homogeneity)) + 
  scale_fill_gradient2(
                       low = "blue", high = "red") +
  #scale_fill_continuous(low ="white", high = "blue") + 
  geom_polygon(color = "black", size = .1) +
  #geom_polygon() +
  theme_bw()

both <- left_join(country_means_race, diversity, by = c("countryres" = "country_code")) %>%
  ungroup()

ggplot(both, aes(x = homogeneity, y = mean)) +
  ylab("race bias") +
  geom_text(aes(label = countryres)) +
  geom_smooth(method = "lm")+
  theme_bw()

cor.test(both$homogeneity, both$mean, method = "spearman")
## 
##  Spearman's rank correlation rho
## 
## data:  both$homogeneity and both$mean
## S = 10915, p-value = 0.1317
## alternative hypothesis: true rho is not equal to 0
## sample estimates:
##      rho 
## 0.230792