Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/theeconomist/oecd_regional_inequality

Regional inequality data and analysis
https://github.com/theeconomist/oecd_regional_inequality

Last synced: 6 days ago
JSON representation

Regional inequality data and analysis

Awesome Lists containing this project

README 

        
---

title: "OECD regional inequality 2018"

output: github_document

---



A primer on regional inequality and data



Set up, load in libraries



```{r}

#path <- your.path.to.our.repo

#setwd(path)



#libs

libs <- c("tidyverse", "dtplyr", "data.table", "OECD", "knitr", "countrycode")

lapply(libs, require, character.only=T)



```



The OECD data we're going to work with is available here: https://stats.oecd.org/Index.aspx?DataSetCode=REGION_DEMOGR#



These files are massive and complex. We've downloaded the data in advance and made it available in the repo's "inputs" folder.



Let's read in the file and inspect the vars



```{r}

reg.dat <- fread("inputs/oecd_raw_dat.csv")



#Inspect the data, by code and descriptor var pairs

reg.dat[, .N, .(TL, Territory.Level.and.Typology)] #TL level

reg.dat[, .N, .(REG_ID, Region)] #Region

reg.dat[, .N, .(SERIES, SNA.Classification)] #Series

reg.dat[, .N, .(VAR, Indicator)] #Indicator

reg.dat[, .N, .(MEAS, Measure)] #Measure

reg.dat[, .N, .(POS, Position)] #Position

reg.dat[, .N, .(TIME, Year)] %>% head() #time

reg.dat[, .N, .(Unit.Code, Unit)] #units

reg.dat[, .N, .(PowerCode.Code, PowerCode)] #powercode

reg.dat[, .N, .(Reference.Period.Code, Reference.Period)] #ref period

reg.dat[, .N, .(Flag.Codes, Flags)] #flags



#We also need a names lookup file that we created

oecd.names <- fread("inputs/oecd_names_lookup.csv")



```



First let's explore all the measures in our dataset, for UK TL3 (small) areas alone



```{r}

#---------- Which measure is best?  ----------------



#Check out the GDP measures

reg.dat[grepl("3", TL),] %>%

  .[, .(REG_ID, Region, VAR, Indicator, MEAS, Measure, Year, Value)] %>% .[, .N, .(VAR, MEAS, Measure)]



#Total GDP at TL3 (d1)

reg.dat[VAR == "GDP" & MEAS == "REAL_PPP" & grepl("3", TL),] %>%

  .[grepl("UK", REG_ID) & Year == 2016,] %>%

  .[, .(REG_ID, Region, Indicator, Measure, Year, Value)] %>%

  arrange(-Value) -> d1 #%>% kable(.)

head(d1) %>% kable



#GDP per person at TL3

reg.dat[VAR == "GDP" & MEAS == "PC_REAL_PPP" & grepl("3", TL),] %>%

  .[grepl("UK", REG_ID) & Year == 2016,] %>%

  .[, .(REG_ID, Region, Indicator, Measure, Year, Value)] %>%

  arrange(-Value) -> d2 #%>% kable(.)

head(d2) %>% kable



#Population by region (implicit)

full_join(d1, d2, by = c("REG_ID", "Region")) %>%

  .[, `:=`(Measure = "Population", Indicator = "Pop", Year = 2016, Value = Value.x * 10^6 / Value.y)] %>%

  .[, .(REG_ID, Region, Indicator, Measure, Year, Value)] %>%

  arrange(Value) -> d3

head(d3) %>% kable



#Work-based employment

reg.dat[VAR == "EMP_IND_TOTAL" & grepl("3", TL),] %>%

  .[grepl("UK", REG_ID) & Year == 2016,] %>%

  .[, .(REG_ID, Region, Indicator, Measure, Year, Value)] %>%

  arrange(-Value) -> d4

head(d4) %>% kable



#GDP per person employed

full_join(d1, d4, by = c("REG_ID", "Region")) %>%

  .[, `:=`(Measure = "GDP per person employed", Indicator = "GDP per person", Year = 2016, Value = Value.x * 10^6 / Value.y)] %>%

  .[, .(REG_ID, Region, Indicator, Measure, Year, Value)] %>%

  arrange(-Value) -> d5

head(d5) %>% kable



#Regional Gross Value Added (GDP ex. subsidies)

reg.dat[VAR == "GVA_IND_TOTAL" & MEAS == "PW_REAL_PPP" & grepl("3", TL),] %>%

  .[grepl("UK", REG_ID) & Year == 2016,] %>%

  .[, .(REG_ID, Region, Indicator, Measure, Year, Value)] %>%

  arrange(-Value) -> d6 #%>% kable(.)

head(d6) %>% kable

#n.b not available for all TL3 regions



rbindlist(l = list(d1, d2, d3, d4, d5, d6), fill=T) -> dat.uk

head(dat.uk) %>% kable



```



With this data gathered, let's see how things look in Camden & City of London versus the Isle of Angelsey: 



```{r}

#Grab Camden & Anglesey

dat.uk[grepl("Camden", Region),] %>% kable

dat.uk[grepl("Torbay", Region),] %>% kable



#Comparing measures

dat.uk[REG_ID %in% c("UKL11", "UKI31"),] %>%

  dcast(Measure ~ Region, value.var = "Value") %>%

  rename(Camden.City=2, Anglesey=3) %>%

  mutate(mutiple = round(Camden.City / Anglesey, 3)) %>% kable(.)



```



The table above shows descriptive the values for each var, and the multiples between them. 



Now, let's produce the same variables for every OECD country. 



First we'll being with GDP per person: 



```{r}



#--------- Get GDP per person for different geographic areas ---------



#National 

reg.dat[VAR == "GDP" & MEAS == "PC_REAL_PPP" & TL == 1,] %>%

  .[!REG_ID %in% c("OECD"), .(REG_ID, Region, Measure, Year, Value)] %>%

  left_join(., rename(oecd.names, REG_ID = oecd.nat.code), by = "REG_ID") %>%

  .[, .(iso3c, REG_ID, Region, Measure, Year, Value)] -> dat2.nat



#GDP per person at TL2 

reg.dat[VAR == "GDP" & MEAS == "PC_REAL_PPP" & grepl("2", TL),] %>%

  .[, .(REG_ID, Region, Measure, Year, Value)] %>%

  .[, oecd.imp.code := substr(REG_ID, 1, 2)] %>%

  left_join(., oecd.names, by = "oecd.imp.code") -> dat2.tl2



#GDP per person at TL3 

reg.dat[VAR == "GDP" & MEAS == "PC_REAL_PPP" & grepl("3", TL),] %>%

  .[, .(REG_ID, Region, Measure, Year, Value)] %>%

  .[, oecd.imp.code := substr(REG_ID, 1, 2)] %>%

  left_join(., oecd.names, by = "oecd.imp.code") -> dat2.tl3



```



With the main data isolated, dat2, for national, tl2 and tl3, let's contstruct the plot data



First, the richest areas in each TL3 (TL2s for USA)



```{r}



#--------- Richest regions ---------



print("Richest regions, latest data")

dat2.tl3[, .SD[which.max(Value)], by=.(iso3c, Year)] %>%

  .[, .SD[which.max(Year)], by=.(iso3c)] %>% 

  .[, geo := "richest.region"] %>%

  .[, .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  arrange(iso3c) -> d2.richest.latest



#Adding richest region in USA (TL2) to TL3 regions

dat2.tl2[, .SD[which.max(Value)], by=.(iso3c, Year)] %>%

  .[, .SD[which.max(Year)], by=.(iso3c)] %>% 

  .[, geo := "richest.region"] %>%

  .[iso3c == "USA", .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  bind_rows(., d2.richest.latest) %>% 

  arrange(iso3c) -> d2.richest.latest

d2.richest.latest %>% kable



print("Richest regions in 2000, or earliest avail.")

dat2.tl3[REG_ID %in% d2.richest.latest$REG_ID,] %>%

  .[Year >= 2000, .SD[which.min(Year)], by=.(iso3c)] %>%

  .[, geo := "richest.region"] %>%

  .[, .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  arrange(iso3c) -> d2.richest.earliest



#Adding in richest region in USA (TL2) in 2000 to TL3 regions in 2000

dat2.tl2[REG_ID %in% d2.richest.latest$REG_ID,] %>%

  .[Year >= 2000, .SD[which.min(Year)], by=.(iso3c)] %>%

  .[, geo := "richest.region"] %>%

  .[iso3c == "USA", .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  bind_rows(., d2.richest.earliest) %>% 

  arrange(iso3c) -> d2.richest.earliest

d2.richest.earliest %>% kable



```



And repeat the steps above for the poorest regions



```{r}



#--------- Poorest regions ------------



print("Poorest regions, latest data")

dat2.tl3[!REG_ID %in% c("FRY50", "FRY30"), ] %>% #removing France's overseas departments

  .[, .SD[which.min(Value)], by=.(iso3c, Year)] %>% 

  .[, .SD[which.max(Year)], by=.(iso3c)] %>% 

  .[, geo := "poorest.region"] %>%

  .[, .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  arrange(iso3c) -> d2.poorest.latest



#Adding richest region in USA (TL2) to TL3 regions

dat2.tl2[, .SD[which.min(Value)], by=.(iso3c, Year)] %>%

  .[, .SD[which.max(Year)], by=.(iso3c)] %>% 

  .[, geo := "poorest.region"] %>%

  .[iso3c == "USA", .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  bind_rows(., d2.poorest.latest) %>% 

  arrange(iso3c) -> d2.poorest.latest

d2.poorest.latest %>% kable



print("Poorest regions in 2000, or earliest avail")

dat2.tl3[REG_ID %in% d2.poorest.latest$REG_ID,] %>%

  .[Year >= 2000, .SD[which.min(Year)], by=.(iso3c)] %>%

  .[, geo := "poorest.region"] %>%

  .[, .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  arrange(iso3c) -> d2.poorest.earliest



#Adding in poorest region in USA (TL2) in 2000 to TL3 regions in 2000

dat2.tl2[REG_ID %in% d2.poorest.latest$REG_ID,] %>%

  .[Year >= 2000, .SD[which.min(Year)], by=.(iso3c)] %>%

  .[, geo := "poorest.region"] %>%

  .[iso3c == "USA", .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  bind_rows(., d2.poorest.earliest) %>% 

  arrange(iso3c) -> d2.poorest.earliest

d2.poorest.earliest %>% kable



```



And then the national averages, being careful to match the same years above: 



```{r}



#--------- National averages (matching years with richest and poorest above) ---------



print("Nat average, match year to latest year in richest/poorest")

dat2.nat[iso3c %in% d2.richest.latest$iso3c,]  %>%

  inner_join(., d2.richest.latest[,.(iso3c, Year)], by = c("iso3c", "Year")) %>%

  .[, geo := "national.av"] %>%

  .[, .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  arrange(iso3c) -> d2.nat.latest

d2.nat.latest %>% kable



print("Nat average, 2000 or earliest avail.")

dat2.nat[iso3c %in% d2.richest.earliest$iso3c,]  %>%

  inner_join(., d2.richest.earliest[,.(iso3c, Year)], by = c("iso3c", "Year")) %>%

  .[, geo := "national.av"] %>%

  .[, .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  arrange(iso3c) -> d2.nat.earliest

d2.nat.earliest %>% kable



```



Now gather the data together (bind) and plot it (roughly)



```{r}



#Bind data

rbindlist(l = list(d2.nat.latest, d2.nat.earliest, 

                   d2.richest.latest, d2.richest.earliest, 

                   d2.poorest.latest, d2.poorest.earliest)) %>%

  left_join(., oecd.names, by = "iso3c") %>%

  dcast(country + iso3c + Year ~ geo, value.var = "Value") %>%

  mutate(indx.poorest.region = poorest.region / national.av * 100, 

         indx.richest.region = richest.region / national.av * 100) -> rich.poor.1

rich.poor.1 %>% kable



#The countries in our original chart

plot.iso <- c("GBR", "DEU", "USA", "FRA", "KOR", "ITA", "JPA", "ESP", "SWE")



#Plot it

rich.poor.1 %>% 

  filter(iso3c %in% plot.iso) %>%

  select(country, Year, poorest=indx.poorest.region, richest=indx.richest.region) %>% 

  melt(id.vars = c("country", "Year"), variable.name = "geo") -> plot.dat.1

plot.dat.1 %>% kable

write_csv(plot.dat.1, "outputs/plot.dat.1.csv")

  

ggplot(plot.dat.1, aes(x=value, y=1, colour=Year)) + geom_point() + geom_text(aes(label=geo, vjust=-1)) +

  facet_wrap(~country, ncol=1) + xlim(0, 1200) + theme_minimal() +

  theme(axis.title.y=element_blank(), axis.text.y=element_blank(), axis.ticks.y=element_blank()) + 

  theme(legend.position="top") + ggtitle("Richest v poorest regions") -> plot1

print(plot1)

```



Now, as the Medium piece discusses while in context of the original Economist piece

discussing regional inequality (https://www.economist.com/briefing/2017/10/21/globalisation-has-marginalised-many-regions-in-the-rich-world) 

the measure is useful. It has since been taken out of context. 



So, let's introduce an alternative measure, keeping the richest and poorest regions (for ease of communication) but dividing a region's GDP not by 

its population but the number of employees working there. 



```{r}



#------- Total GDP ----------



#National 

reg.dat[VAR == "GDP" & MEAS == "REAL_PPP" & TL == 1,] %>%

  .[!REG_ID %in% c("OECD"), .(REG_ID, Region, Measure, Year, Value)] %>%

  left_join(., rename(oecd.names, REG_ID = oecd.nat.code), by = "REG_ID") %>%

  .[, .(iso3c, REG_ID, Region, Measure, Year, Value)] -> dat1.nat



#TL2 

reg.dat[VAR == "GDP" & MEAS == "REAL_PPP" & grepl("2", TL),] %>%

  .[, .(REG_ID, Region, Measure, Year, Value)] %>%

  .[, oecd.imp.code := substr(REG_ID, 1, 2)] %>%

  left_join(., oecd.names, by = "oecd.imp.code") -> dat1.tl2



#TL3 

reg.dat[VAR == "GDP" & MEAS == "REAL_PPP" & grepl("3", TL),] %>%

  .[, .(REG_ID, Region, Measure, Year, Value)] %>%

  .[, oecd.imp.code := substr(REG_ID, 1, 2)] %>%

  left_join(., oecd.names, by = "oecd.imp.code") -> dat1.tl3



#------- Work-based employment ---------



#National 

reg.dat[VAR == "EMP_IND_TOTAL" & TL == 1,] %>%

  .[!REG_ID %in% c("OECD"), .(REG_ID, Region, Measure, Year, Value)] %>%

  left_join(., rename(oecd.names, REG_ID = oecd.nat.code), by = "REG_ID") %>%

  .[, .(iso3c, REG_ID, Region, Measure, Year, Value)] -> dat4.nat



#TL2 

reg.dat[VAR == "EMP_IND_TOTAL" & grepl("2", TL),] %>%

  .[, .(REG_ID, Region, Measure, Year, Value)] %>%

  .[, oecd.imp.code := substr(REG_ID, 1, 2)] %>%

  left_join(., oecd.names, by = "oecd.imp.code") -> dat4.tl2



#TL3 

reg.dat[VAR == "EMP_IND_TOTAL" & grepl("3", TL),] %>%

  .[, .(REG_ID, Region, Measure, Year, Value)] %>%

  .[, oecd.imp.code := substr(REG_ID, 1, 2)] %>%

  left_join(., oecd.names, by = "oecd.imp.code") -> dat4.tl3



#------ GDP per person employed (implicit) --------



#National 

full_join(dat1.nat, dat4.nat, by = c("iso3c", "REG_ID", "Region", "Year")) %>%

  .[, `:=`(Measure = "GDP per person employed", Value = Value.x * 10^6 / Value.y)] %>%

  .[!is.na(Value), .(iso3c, REG_ID, Region, Measure, Year, Value)]  -> dat5.nat



#TL2

full_join(dat1.tl2, dat4.tl2, by = c("iso3c", "REG_ID", "Region", "Year")) %>%

  .[, `:=`(Measure = "GDP per person employed", Value = Value.x * 10^6 / Value.y)] %>%

  .[!is.na(Value), .(iso3c, REG_ID, Region, Measure, Year, Value)] -> dat5.tl2



#TL3

full_join(dat1.tl3, dat4.tl3, by = c("iso3c", "REG_ID", "Region", "Year")) %>%

  .[, `:=`(Measure = "GDP per person employed", Value = Value.x * 10^6 / Value.y)] %>%

  .[!is.na(Value), .(iso3c, REG_ID, Region, Measure, Year, Value)] -> dat5.tl3



```



Now we have the data we need, we'll breeze through the next steps as they repeat those for the GDP per person measure above. 



```{r}



#------- Richest regions, GDP per person --------- 



#Richest regions, latest data

dat5.tl3[, .SD[which.max(Value)], by=.(iso3c, Year)] %>%

  .[, .SD[which.max(Year)], by=.(iso3c)] %>% 

  .[, geo := "richest.region"] %>%

  .[, .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  arrange(iso3c) -> d5.richest.latest



#Adding richest region in USA (TL2) to TL3 regions

dat5.tl2[, .SD[which.max(Value)], by=.(iso3c, Year)] %>%

  .[, .SD[which.max(Year)], by=.(iso3c)] %>% 

  .[, geo := "richest.region"] %>%

  .[iso3c == "USA", .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  bind_rows(., d5.richest.latest) %>% 

  arrange(iso3c) -> d5.richest.latest

d5.richest.latest %>% kable



#Richest regions in 2000, or earliest avail. 

dat5.tl3[REG_ID %in% d5.richest.latest$REG_ID,] %>%

  .[Year >= 2000, .SD[which.min(Year)], by=.(iso3c)] %>%

  .[, geo := "richest.region"] %>%

  .[, .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  arrange(iso3c) -> d5.richest.earliest



#Adding in richest region in USA (TL2) in 2000 to TL3 regions in 2000

dat5.tl2[REG_ID %in% d5.richest.latest$REG_ID,] %>%

  .[Year >= 2000, .SD[which.min(Year)], by=.(iso3c)] %>%

  .[, geo := "richest.region"] %>%

  .[iso3c == "USA", .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  bind_rows(., d5.richest.earliest) %>% 

  arrange(iso3c) -> d5.richest.earliest

d5.richest.earliest %>% kable



#--------- Poorest regions, GDP per person employed ------------



#Poorest regions, latest data

dat5.tl3[!REG_ID %in% c("FRY50", "FRY30"), ] %>% #removing France's overseas departments

  .[, .SD[which.min(Value)], by=.(iso3c, Year)] %>% 

  .[, .SD[which.max(Year)], by=.(iso3c)] %>% 

  .[, geo := "poorest.region"] %>%

  .[, .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  arrange(iso3c) -> d5.poorest.latest



#Adding richest region in USA (TL2) to TL3 regions

dat5.tl2[, .SD[which.min(Value)], by=.(iso3c, Year)] %>%

  .[, .SD[which.max(Year)], by=.(iso3c)] %>% 

  .[, geo := "poorest.region"] %>%

  .[iso3c == "USA", .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  bind_rows(., d5.poorest.latest) %>% 

  arrange(iso3c) -> d5.poorest.latest

d5.poorest.latest %>% kable



#Poorest regions in 2000, or earliest avail. 

dat5.tl3[REG_ID %in% d5.poorest.latest$REG_ID,] %>%

  .[Year >= 2000, .SD[which.min(Year)], by=.(iso3c)] %>%

  .[, geo := "poorest.region"] %>%

  .[, .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  arrange(iso3c) -> d5.poorest.earliest



#Adding in poorest region in USA (TL2) in 2000 to TL3 regions in 2000

dat5.tl2[REG_ID %in% d5.poorest.latest$REG_ID,] %>%

  .[Year >= 2000, .SD[which.min(Year)], by=.(iso3c)] %>%

  .[, geo := "poorest.region"] %>%

  .[iso3c == "USA", .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  bind_rows(., d5.poorest.earliest) %>% 

  arrange(iso3c) -> d5.poorest.earliest

d5.poorest.earliest %>% kable



#------- Nationl averages, GDP per person employed, matching years above ------- 



#Nat average, match year to latest year in richest/poorest 

dat5.nat[iso3c %in% d5.richest.latest$iso3c,]  %>%

  inner_join(., d5.richest.latest[,.(iso3c, Year)], by = c("iso3c", "Year")) %>%

  .[, geo := "national.av"] %>%

  .[, .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  arrange(iso3c) -> d5.nat.latest

d5.nat.latest %>% kable



#Nat average, 2000 or earliest avail.

dat5.nat[iso3c %in% d5.richest.earliest$iso3c,]  %>%

  inner_join(., d5.richest.earliest[,.(iso3c, Year)], by = c("iso3c", "Year")) %>%

  .[, geo := "national.av"] %>%

  .[, .(iso3c, geo, REG_ID, Region, Measure, Year, Value)] %>%

  arrange(iso3c) -> d5.nat.earliest

d5.nat.earliest %>% kable



```



Again, let's bind the data together and plot our new measure. 



```{r}



#------- Bind data together and plot  -------- 



rbindlist(l = list(d5.nat.latest, d5.nat.earliest, 

                   d5.richest.latest, d5.richest.earliest, 

                   d5.poorest.latest, d5.poorest.earliest)) %>%

  left_join(., oecd.names, by = "iso3c") %>%

  dcast(country + iso3c + Year ~ geo, value.var = "Value") %>%

  mutate(indx.poorest.region = poorest.region / national.av * 100, 

         indx.richest.region = richest.region / national.av * 100) -> rich.poor.2

rich.poor.2 %>% kable



#The countries in our original chart

plot.iso <- c("GBR", "DEU", "USA", "FRA", "KOR", "ITA", "JPA", "ESP", "SWE")



#Plot it

rich.poor.2 %>% 

  filter(iso3c %in% plot.iso) %>%

  select(country, Year, poorest=indx.poorest.region, richest=indx.richest.region) %>% 

  melt(id.vars = c("country", "Year"), variable.name = "geo") -> plot.dat.2

plot.dat.2 %>% kable

write_csv(plot.dat.2, "outputs/plot.dat.2.csv")



ggplot(plot.dat.2, aes(x=value, y=1, colour=Year)) + geom_point() + geom_text(aes(label=geo, vjust=-1)) +

  facet_wrap(~country, ncol=1) + theme_minimal() +

  theme(axis.title.y=element_blank(), axis.text.y=element_blank(), axis.ticks.y=element_blank()) + 

  theme(legend.position="top") + ggtitle("Richest v poorest regions, GDP per person employed") -> plot2

print(plot2)



```