Using data relevant to countries around the world, we will explore data science tools to tell complex stories through data visualization. We will consider data on infant mortality in class, and for the report you are welcome to use any data that can be plotted by country.
Using data from the World Bank or a complex data source of your own choosing,
create a Tableau dashboard interactive visual presentation of our world in data
explain your figures to class in an oral presentation by the group
Case for Class: Infant Mortality
Infant death is defined as the death of a live-born infant during the first year of life. The infant mortality rate is the number of infant deaths per 1000 live births. The infant mortality rate, which is relatively easily measured even in resource-poor settings, is often used as a marker of the overall health of a society because of the overlap in causes of infant mortality and factors that are likely to influence the health of an overall population, such as general living conditions, poverty and socioeconomic status, quality of medical care, and rates of illness. Infant mortality rates can be compared across different areas or across time to evaluate the impact of public health program or to determine where new investments in health are needed.
Changes in US Infant Mortality over Time
Infant Mortality in US, 1915-2007 (NCHS)
What do you think about the way the authors chose to display time in the plot?
Disparities in Infant Mortality Rate
While overall infant mortality rates in the US have declined over time, major disparities in infant mortality rates exist, shown here as a function of race and ethnicity.
Infant Mortality Worldwide
Worldmapper created a visualization of infant mortality worldwide in 2002, rescaling each country by its contribution to infant deaths worldwide.
2002 Contributions to Worldwide Infant Mortality
Mapping Infant Mortality
In order to create our map of infant mortality around the world, we will need to identify the following data.
Infant mortality data for time period of interest by country
Country mapping information and link to infant mortality data
Extracting Latitude and Longitude of Country Centroids
First, we will obtain latitude and longitude information for the center of each country. We can then link that to infant mortality rates for each country.
First we’ll combine the demographic data on infant mortality and population size because these come from the same source and consequently should have fewer problems with matching.
IMPop =merge(infMortDF, popDF, by ="ctry", all =FALSE)
Missed Merges
In the merge of infMortDF and popDF, some rows were excluded from the resulting data frame. Determine which rows in one data frame did not find a match in the other.
Check to see if all countries in the infant mortality dataset are also in the population size dataset:
which((!infMortDF$ctry %in%popDF$ctry))
## integer(0)
Missed Merges
Which countries are in the population data set but not the infant mortality data set?
## [1] AX KT CK UC DX FK VT KV MJ NE NF PC TB RN SV TL
## 272 Levels: AA AC AE AF AG AJ AL AM AN AO AQ AR AS AT AU AV AX AY ... ZI
Appendix D of the CIA World Factbook contains country codes: AX (Akrotiri, a special UK overseas territory near Cyprus), KT (Christmas Island), CK (Cocos/Keeling Islands), UC (Curacao), DX (Dhekelia, see AX), FK (Falkland Islands), VT (Vatican City) , KV (Kosovo), MJ (Montenegro), NE (Niue), NF (Norfolk Island), PC (Pitcairn Islands), TB (St. Barthelemy), RN (St. Martin), SV (Svalbard), TL (Tokelau)
First, we’ll begin by selecting the colors for our plot. We’ll select only 5 colors, and consequently 5 levels of mortality, because it’s difficult for our eyes to distinguish among more than 5 to 7 colors.
Hmm, the good news is that the right tail is not very fat. However, this may not be a great choice for displaying the data.
Categorizing Mortality Rate
We want our colors to be more evenly distributed than they are based on the default. One option is to put 20% of the data in each color group, though that may lead to cutpoints at odd locations.
quantile(allCtryData$infMort, probs =seq(0, 1, by =0.2))
The infant mortality rate is not the only factor in decision-making. We are more likely to intervene when more lives are at risk. Let’s examine population size.
hist(allCtryData$pop, breaks =20, main ="",
xlab ="Population Size")
Large variance – let’s try a scale transformation.
Population Size
hist(sqrt(allCtryData$pop), breaks =20, main ="",
xlab ="sqrt(Population Size)")
That’s much better. We’ll work with some function of square root of population size (further scalng to be sure the radius of each point is not too large for our plot).
Initial Plot
Now we are ready to plot the mortality rates around the world. Size of the points will be a function of population size, while color will be a function of the mortality rate (darker=higher).
# Create maplibrary(maps)
world =map(database ="world", fill =TRUE, col="light grey")
# Add discs, circles sets radius of discs, fg outlines discs and bg fills them insymbols(allCtryData$longitude, allCtryData$latitude, add =TRUE,
circles=sqrt(allCtryData$pop)/4000, inches =FALSE,
fg = cols[InfMortDiscrete], bg = cols[InfMortDiscrete])
# Add legendlegend(x =-150, y =0, title ="Infant Mortality",
legend =levels(InfMortDiscrete), fill = cols, cex =0.8)
Infant Mortality Worldwide
Hmm, what’s wrong with this picture?
Correcting Radius Size
The small countries have symbols that are so tiny that we can barely see them, if at all, and we certainly can’t see their colors (check out the Caribbean). We will fix this by having a minimum radius, so any country with a population below a certain value is still visible.
# Set minimum radius size
rads =pmax(sqrt(allCtryData$pop)/4000, 1)
# Fix Discs
world =map(database ="world", fill =TRUE, col="light grey")
symbols(allCtryData$longitude, allCtryData$latitude, add =TRUE,
circles= rads, inches =FALSE,
fg = cols[InfMortDiscrete], bg = cols[InfMortDiscrete])
legend(x =-150, y =0, title ="Infant Mortality",
legend =levels(InfMortDiscrete), fill = cols, cex =0.8)
Correcting Radius Size
Further Corrections
Looking at our plot we can see that something is still very wrong! Some European country has a huge radius. China’s population size is also incorrect, and does Russia really have no infant mortality?
We probably made an error in the merge step somehow. Here was our code.
For the latitude/longitude files, we used ISO 3166 country codes, set by the International Organization for Standardization. Did the CIA Factbook use the same codes?
The R data frame codeMapDF (extracted from the CIA factbook) contains the crosswalk between the CIA and ISO country coding and can be used to conduct a corrected merge of the data.
# Fix Data Mergingload("data/codeMapDF.RData")
latlonCtry =merge(latlonDF, codeMapDF, by.x ="code", by.y ="iso", all =FALSE)
IMpopCtry =merge(IMPop, codeMapDF, by.x="ctry", by.y ="cia", all=FALSE)
allCtryData <-merge(latlonCtry, IMpopCtry, by="name")
names(allCtryData)[1] <- "Country"save(allCtryData, file="data/allCtryData.RData") # save for Tableau tutorial# Check Data
allCtryData[allCtryData$ctry %in%c("CH", "SZ", "GB", "UK", "NU","NI","RI","RS"),
c(1,6,10,4,5,8,9)]
## Country cia iso latitude longitude infMort pop
## 43 China CH CN 35 105.00 14.79 1355692576
## 70 Gabon GB GA -1 11.75 47.03 1672597
## 141 Nicaragua NU NI 13 -85.00 20.36 5848641
## 143 Nigeria NI NG 10 8.00 74.09 177155754
## 159 Russia RS RU 60 100.00 7.08 142470272
## 171 Serbia RI RS 44 21.00 6.16 7209764
## 186 Switzerland SZ CH 47 8.00 3.73 8061516
## 203 United Kingdom UK GB 54 -2.00 4.44 63742977
Fixing Map
InfMortDiscrete =cut(allCtryData$infMort,
breaks =c(0, 10, 25, 50, 75, 150))
# Redo Map
world =map(database ="world", fill =TRUE, col="light grey")
rads =pmax(sqrt(allCtryData$pop)/4000, 1)
symbols(allCtryData$longitude, allCtryData$latitude, add =TRUE,
circles= rads, inches =FALSE,
fg = cols[InfMortDiscrete], bg = cols[InfMortDiscrete])
legend(x =-150, y =0, title ="Infant Mortality",
legend =levels(InfMortDiscrete), fill = cols, cex =0.8)
Fixing Map
In case you’re curious!
This map includes the countries for which there is latitude and longitude, even if there is no demographic information.
# Remerge the data
allCtryData <-merge(latlonCtry, IMpopCtry, by="name", all.x=TRUE, all.y =FALSE)
# Update the colors & add NA as a level
cols =brewer.pal(9, "YlOrRd")[c(1, 2, 4, 6, 7, 9)]
InfMortDiscrete =cut(allCtryData$infMort,
breaks =c(0, 10, 25, 50, 75, 150))
InfMortDiscrete <-addNA(InfMortDiscrete)
# Add radius size for NA
rads =pmax(sqrt(allCtryData$pop)/4000, 1)
for(i in1:length(rads)){
if(is.na(rads[i])){rads[i] <-1}
}
Revised Map
world =map(database ="world", fill =TRUE, col="light grey")
symbols(allCtryData$longitude, allCtryData$latitude, add =TRUE,
circles= rads, inches =FALSE,
fg = cols[InfMortDiscrete], bg = cols[InfMortDiscrete])
legend(x =-150, y =0, title ="Infant Mortality",
legend =levels(InfMortDiscrete), fill = cols, cex =0.8)
