foodloss_sensitivity.Rmd

---
title: "Daniel food loss revision"
author: "Johannes Piipponen"
date: "`Spring 2023"
output: html_document
---

Code written by Johannes Piipponen

# Get and explore data

```{r}
# Load necessary libraries
library(tidyverse)
library(fixest)

# Read the data
df_raw <- read_csv("data_johannes.csv")

# Add a new column ln_gni as the natural logarithm of gni
df_raw <- df_raw %>% 
  mutate(ln_gni = log(gni))

# Explore data: number of countries, crops and years
length(unique(df_raw$country)) # 81 countries
unique(df_raw$year) %>% sort() # years 2000-2020



# Group data by country and count the number of cases
df_raw %>% 
  group_by(country) %>% 
  summarise(Number_of_cases = n()) %>% 
  arrange(Number_of_cases)

# Create a list of countries with more than 80 observations
#my_cntrylist_over80 <- 
my_countries_obs <-
  df_raw %>% 
  group_by(country) %>% 
  summarise(Number_of_cases = n()) %>% 
 # filter(Number_of_cases > 80) %>% 
  arrange(Number_of_cases)# %>%   pull(country)


```






# Sensitivity analysis

The following code performs a regression analysis using all available data and then explores the sensitivity of the model by excluding individual countries from the analysis.

When all countries are included in the model, it produces specific values for the intercept and beta1 coefficients. However, if we exclude a particular country (e.g., Finland), these coefficient values may change. The objective is to determine the magnitude of these changes.

The ultimate goal is to create a graph that visualizes how the regression results vary when we exclude one country at a time.

Explanation for f_exclude_countries function:
The f_exclude_countries function accepts the country to be excluded as an argument (country_tobe_excluded). It runs the feols model with the specified formula and filters the dataset so that the specified country is not included. After fitting the model, the function uses broom::tidy() to extract the regression results and selects the relevant columns (term, estimate, and p-value). Lastly, the function adds a new column, excluded_country, to indicate the country that was excluded in that particular model run.


```{r}
# Create a function to run feols excluding one country at a time
f_exclude_countries <- function(country_tobe_excluded) {
  feols(loss_percentage1 ~ ln_gni + agriShr + agriEmp + msch + 
           electricityRural + phoneSub + polStab + export + regionalName + as.factor(year),
         cluster = ~ regionalName, 
         data = df_raw %>% filter(country != country_tobe_excluded)) %>%
    broom::tidy() %>%
    dplyr::select(term, estimate, p.value) %>%
    mutate(excluded_country = country_tobe_excluded)
}

# Example function calls
f_exclude_countries("Angola")
f_exclude_countries("Burundi")

# Use the function for all countries
my_countries <- unique(df_raw$country)

# Apply the function to all countries and store the results
results <- map_df(my_countries, f_exclude_countries)

# Widen the results and rename p-value columns ---> not sure if this is needed
results_wider <- results %>%
  pivot_wider(names_from = term, values_from = c(estimate, p.value)) %>%
  rename(
    pval_intercept = `p.value_(Intercept)`,
    pval_ln_gni = `p.value_ln_gni`,
    pval_agriShr = `p.value_agriShr`,
    pval_agriEmp = `p.value_agriEmp`,
    pval_msch = `p.value_msch`,
    pval_electricityRural = `p.value_electricityRural`,
    pval_phoneSub = `p.value_phoneSub`,
    pval_polStab = `p.value_polStab`,
    pval_export = `p.value_export`
  )


```


# Save the original regression

```{r}
library(flextable)
library(modelsummary)

model <-   feols(loss_percentage1 ~ ln_gni + agriShr + agriEmp + msch + 
                   electricityRural + phoneSub + polStab + export + regionalName + as.factor(year),
                 cluster = ~ regionalName, 
                 data = df_raw)



# Create the table for the feols model
model_table <- modelsummary(model, stars = T, output = "flextable")

# uncomment this
#save_as_docx(model_table, path = ("selected_results.docx"))
```


# Plot the results

# Sensitivity like originally but when number of cases is used to define X axis

```{r}
results_arranged <- map_df(unique(my_countries_obs$country), f_exclude_countries)

results_wider_arranged <- results_arranged %>%
  pivot_wider(names_from = term, values_from = c(estimate, p.value)) %>%
  rename(
    pval_intercept = `p.value_(Intercept)`,
    pval_ln_gni = `p.value_ln_gni`,
    pval_agriShr = `p.value_agriShr`,
    pval_agriEmp = `p.value_agriEmp`,
    pval_msch = `p.value_msch`,
    pval_electricityRural = `p.value_electricityRural`,
    pval_phoneSub = `p.value_phoneSub`,
    pval_polStab = `p.value_polStab`,
    pval_export = `p.value_export`
  )


results_long_arranged <- results_wider_arranged %>%
  pivot_longer(
    cols = starts_with("estimate_"),
    names_to = "variable",
    values_to = "value"
  ) %>% 
  dplyr::select(excluded_country, variable, value) %>%
  drop_na()



# Select only some variables
results_long_arranged <- results_long_arranged %>% 
  filter(variable %in% c("estimate_(Intercept)", "estimate_ln_gni",                                 
 "estimate_agriShr",                                  
  "estimate_agriEmp",                                  
  "estimate_msch",                                     
  "estimate_electricityRural",                         
  "estimate_phoneSub",                                 
  "estimate_polStab",                                  
  "estimate_export"   ))


# Draw a combined plot
library(scico)

plt_combined_arranged <- ggplot(results_long_arranged, aes(x = factor(excluded_country, levels = unique(excluded_country)), y = value, color = variable, group = variable)) +
  geom_line(size = 1.2) + 
  facet_grid(variable ~ .,
             scales = "free_y",
             labeller = as_labeller(c("estimate_(Intercept)" = "Intercept",
                                      estimate_ln_gni = "Log GNI",
                                      estimate_agriShr = "Agri Share",
                                      estimate_agriEmp = "Agri Emp",
                                      estimate_msch = "Mean Sch",
                                      estimate_electricityRural = "Elec Acc",
                                      estimate_phoneSub = "Phone Sub",
                                      estimate_polStab = "Pol Stab",
                                      estimate_export = "Exports"))) +
  theme_gray() +
  theme(axis.text.x = element_text(angle = 90, hjust = 1, size = 6),
        axis.text.y = element_text(size = 8),
        strip.text = element_text(size = 8),
        legend.position = "none") +
  scale_color_scico_d(palette = "roma") +
  labs(title = "Impact of excluding a country on regression output",
       x = "Excluded country countries with fewer observations         countries with more observations",
       y = "Estimate")


plt_combined_arranged

# uncomment this

# ggsave("sensitivity_plot_x_axis_based_on_N_obs.png", plot = plt_combined_arranged, dpi = 300, width = 300, height = 225, units = "mm")
# 
# ggsave("sensitivity_plot_x_axis_based_on_N_obs.pdf", plot = plt_combined_arranged)


```


# Same but regional variables 

11] "estimate_regionalNameCentral & South Asia"         
[12] "estimate_regionalNameEast & Southeast Asia"        
[13] "estimate_regionalNameEurope & North America"       
[14] "estimate_regionalNameLatin America & the Caribbean"
[15] "estimate_regionalNameMiddle East & North Africa"   
[16] "estimate_regionalNameSub-Saharan Africa" 

```{r}
# Prepare data for plotting
results_long_reg <- results_wider %>%
  pivot_longer(
    cols = starts_with("estimate_regional"),
    names_to = "variable",
    values_to = "value"
  ) %>% 
  dplyr::select(excluded_country, variable, value)

# Draw a combined plot
plt_combined_reg <- ggplot(results_long_reg, aes(x = excluded_country, y = value, color = variable)) +
  geom_point() +
  facet_grid(variable ~ .,
             scales = "free_y",
             labeller = as_labeller(c("estimate_regionalNameCentral & South Asia" = "Cent & S Asia",
                                      "estimate_regionalNameEast & Southeast Asia" = "E & SE Asia",
                                      "estimate_regionalNameEurope & North America" = "Eur & N Am",
                                      "estimate_regionalNameLatin America & the Caribbean" = "Lat Am & Car",
                                      "estimate_regionalNameMiddle East & North Africa" = "MEast & N Af",
                                      "estimate_regionalNameSub-Saharan Africa" = "Sub-Sah Africa"))) +
  theme_classic() +
  theme(axis.text.x = element_text(angle = 45, hjust = 1),
    legend.position = "none") +
  labs(title = "Impact of excluding a country on regression output",
    x = "Excluded Country",
    y = "Estimate")


plt_combined_reg

# ggsave("sensitivity_plot_reg.png", plot = plt_combined_reg, dpi = 300, width = 300, height = 225, units = "mm")
```

# Sensitivity plot for years

```{r}
results_long_year <- results_wider %>%
  pivot_longer(
    cols = starts_with("estimate_as.factor"),
    names_to = "variable",
    values_to = "value"
  ) %>% 
  dplyr::select(excluded_country, variable, value)



plt_combined_year <- ggplot(results_long_year, aes(x = excluded_country, y = value, color = variable)) +
  geom_point() +
  facet_grid(variable ~ ., scales = "free_y", labeller = as_labeller(c("estimate_as.factor(year)2001" = "2001",
                                                                      "estimate_as.factor(year)2002" = "2002",
                                                                      "estimate_as.factor(year)2003" = "2003",
                                                                      "estimate_as.factor(year)2004" = "2004",
                                                                      "estimate_as.factor(year)2005" = "2005",
                                                                      "estimate_as.factor(year)2006" = "2006",
                                                                      "estimate_as.factor(year)2007" = "2007",
                                                                      "estimate_as.factor(year)2008" = "2008",
                                                                      "estimate_as.factor(year)2009" = "2009",
                                                                      "estimate_as.factor(year)2010" = "2010",
                                                                      "estimate_as.factor(year)2011" = "2011",
                                                                      "estimate_as.factor(year)2012" = "2012",
                                                                      "estimate_as.factor(year)2013" = "2013",
                                                                      "estimate_as.factor(year)2014" = "2014",
                                                                      "estimate_as.factor(year)2015" = "2015",
                                                                      "estimate_as.factor(year)2016" = "2016",
                                                                      "estimate_as.factor(year)2017" = "2017",
                                                                      "estimate_as.factor(year)2018" = "2018",
                                                                      "estimate_as.factor(year)2019" = "2019",
                                                                      "estimate_as.factor(year)2020" = "2020"))) +
  theme_classic() +
  theme(axis.text.x = element_text(angle = 45, hjust = 1),
    legend.position = "none") +
  labs(title = "Impact of excluding a country on regression output",
    x = "Excluded Country",
    y = "Estimate")


plt_combined_year

# ggsave("sensitivity_plot_year.png", plot = plt_combined_year, dpi = 300, width = 300, height = 225, units = "mm")
```




#