R Notebook

rm(list = ls())

Sys.setenv(LANG = "en")
library(xlsx)
library(readr)
library(readxl)
library(tidyverse)
library(stringr)
library(stringi)
library(gghighlight)
library(countrycode)
library(countries)
library(janitor)
library(sf)

library(janitor)
library(sf)
library(ggthemes)
library(here)
library(viridis)
library(ggplot2)
library(formatR)
library(gtools)
library(nnet) 
library(patchwork)
library(ggh4x)

library(plm)
library(broom)
library(ggalluvial)

library(pals)

Load data —————————————————————

Large datasets (move load to different notebook)

aggregates <- read_csv("../aggregated_data/aggregates.csv")
aggregates_year <- read_csv("../aggregated_data/aggregates_year.csv")

Convert vars to factor. read_csv changes original type.

aggregates_year <- aggregates_year %>% 
  mutate(subregion_rf = as_factor(subregion),
        subregion_rf = relevel(subregion_rf, "Northern America"),
        continent = as_factor(continent),
        continent = relevel(continent, "Americas"))

Figure 6. Regression

regr_numeric <- c("population_n",
                  "immig_stock",
                  "immig_relative_stock",
                  "emig_stock",
                  "emig_stock_rel_ctry",
                  "research_GDP",
                  "research",
                  "GDP")

aggregates_mod <- aggregates_year %>% 
    filter(!is.na(ISO3)) %>% 
    filter(pubyear %in% c(1996, 2000, 2005, 2010, 2015, 2020),
           !is.na(GDP) & !is.na(research)) %>% 
    mutate(repr_y_s_2 = as.factor(if_else(repr_y_s %in% c("Fair", "Over"), 0, 1)),
           across(all_of(regr_numeric), ~ .x + 0.000000000001))

Linear probability model.

Functions

tidy_r1 <- function(reg) {
  reg %>% tidy(conf.int = TRUE) %>% 
    filter(!str_detect(term, 'pubyear'), 
           !term %in% c("(Intercept)", "log(population_n)")) %>% 
    mutate(term = str_remove(term, "subregion_rf"),
           term = str_remove(term, "continent"),
           term = case_when(
               term == "log(GDP)" ~  "GDPpc",
               term == "log(research_GDP)" ~  "Research as % of GDPpc",
               term == "gsouthGlobal South" ~ "Global South",
               term == "log(emig_stock)" ~ "Emigration stock",
               term == "log(immig_stock)" ~ "Immigration stock",
               TRUE ~ term),
           term = fct_reorder(term, estimate),
           significant = if_else(p.value <= 0.05, false = 0, 1),
           effects = case_when(
               estimate < 0 ~ "over",
               estimate == 0 ~ "fair",
               estimate > 0 ~ "under"
           ))
  
}

plot_reg <- function(reg_tidy, ptitle) { 
    
    reg_tidy %>%  
        filter(term != "(Intercept)") %>% 
        ggplot(aes(estimate, term)) +
        geom_point(aes(color = factor(significant))) +
        geom_errorbarh(aes(xmin = conf.low, xmax = conf.high, color = factor(significant)), height = 0) +
        geom_vline(xintercept = 0, lty = 2) +
        theme_minimal() +
        theme(axis.text = element_text(size = 7),
              axis.title.x = element_text(size = 6),
              legend.position="none",
              plot.title = element_text(size=8)) +
        labs(title = ptitle,
             y = NULL,
             x = "Regression coefficient") +
        scale_color_manual(values=c("0" = "#434343", "1" = "#6BA2D6"))
}

Depvar: Under-research

• Aim of regression: understand how countries economic resources (GDP and research investment) and geographic regions affects likelihood of being an under-researched country.
• log(GDP) works better than factor version, as it is significant. Income groups are not significant.
• Note migration variables have been removed, as they are part of the dependent variable.
• pubyear transformed to factor.

r1_continent <- lm(as.numeric(repr_y_s_2) ~ as.factor(pubyear) + continent + log(population_n) + log(GDP) + log(research_GDP), aggregates_mod)
r1_region <- lm(as.numeric(repr_y_s_2) ~ as.factor(pubyear) + subregion_rf + log(population_n) + log(GDP) + log(research_GDP), aggregates_mod)

Depvar: Research salience

#Absolute stocks
r2_continent <- lm(log(as.numeric(art_country_y)) ~ as.factor(pubyear) + continent + log(population_n) + log(GDP) + log(immig_stock) + log(emig_stock) + log(research_GDP), aggregates_mod)
r2_region <- lm(log(as.numeric(art_country_y))  ~ as.factor(pubyear) + subregion_rf + log(population_n) + log(GDP) + log(immig_stock) + log(emig_stock) + log(research_GDP), aggregates_mod)

#Relative stocks
r2_continent_rel <- lm(log(as.numeric(art_country_y)) ~ as.factor(pubyear) + continent + log(population_n) + log(GDP) + log(immig_relative_stock) + log(emig_stock_rel_ctry) + log(research_GDP), aggregates_mod)
r2_region_rel <- lm(log(as.numeric(art_country_y))  ~ as.factor(pubyear) + subregion_rf + log(population_n) + log(GDP) + log(immig_relative_stock) + log(emig_stock_rel_ctry) + log(research_GDP), aggregates_mod)

Plots

# Under-research
r1_continent_tidy <- r1_continent %>% tidy_r1 
(r1_continent_plot <- r1_continent_tidy %>% plot_reg("1A - Underrepresentation:\n Continents"))
ggsave("../output_figures/cplot_continent.png", plot = last_plot())

r1_region_tidy <- r1_region %>% tidy_r1
(r1_region_plot <- r1_region_tidy %>% plot_reg("1B - Underrepresentation:\n Subregions"))
ggsave("../output_figures/cplot_subregion.png", plot = last_plot())

# Research salience
r2_continent_tidy <- r2_continent %>% tidy_r1 
(r2_continent_plot <- r2_continent_tidy %>% plot_reg("2A - Salience:\n Continents"))
ggsave("../output_figures/cplot_counts_continent.png", plot = last_plot())

r2_region_tidy <- r2_region %>% tidy_r1 
(r2_region_plot <- r2_region_tidy %>% plot_reg("2B - Salience:\n Subregions"))
ggsave("../output_figures/cplot_counts_subregion.png", plot = last_plot())

# Research salience. Relative.Note the results are exactly the same as with absolute numbers!
r2_continent_rel_tidy <- r2_continent_rel %>% tidy_r1 
(r2_continent_rel_plot <- r2_continent_rel_tidy %>% plot_reg("Model S.1:Continents"))
ggsave("../output_figures/cplot_counts_continent_rel.png", plot = last_plot())

r2_region_rel_tidy <- r2_region_rel %>% tidy_r1 
(r2_region_rel_plot <- r2_region_rel_tidy %>% plot_reg("Model S.2: Subregions"))
ggsave("../output_figures/cplot_counts_subregion_rel.png", plot = last_plot())

Combining plots

r1_continent_plot / r1_region_plot #/ r1_gsouth_plot  

ggsave("../output_figures/cplot_combined.png", width = 6, height = 7, dpi = 500, limitsize = F, plot = last_plot())

r2_continent_plot / r2_region_plot #/ r2_gsouth_plot 

ggsave("../output_figures/cplot_counts_combined.png", width = 6, height = 7, dpi = 500, limitsize = F, plot = last_plot())

(r1_continent_plot + r1_region_plot) / (r2_continent_plot + r2_region_plot) 
ggsave("../output_figures/cplot_all.png", width = 6, height = 7, dpi = 500, limitsize = F, plot = last_plot())