Data Visualization and Animation with R
This example visualizes data with ggplot2 in many different ways and employs animation techniques with gganimation.
- The foreign direct investment (FDI) data for this example comes from the Financial Times and it is measured as the number of investment projects per year, at the state-level. The other variables employed are the population size and GDP per capita, from the BEA. I subsequently generate a dummy variable to identify right-to-work (RTW) states, meaning states that have RTW laws that prohibit union security agreements between employers and labor unions. The general hypothesis is that RTW states should, in theory, receive more FDI as they have a more “business friendly” environment. The idea is to investigate this hypothesis visually. First, I read in the data from a Stata file. Then we run a simple scatterplot to investigate the relationship between GDP per capita and FDI on RTW and non-RTW states:
library(foreign)
library(ggplot2)
library(ellipsis)
library(gganimate)
library(stringi)
library(gapminder)
library(gifski)
datapd <- read.dta(file.choose())
# We can have a simple scatterplot
scplot <- ggplot(datapd, aes(x=GDPpc, y=Projects)) +
geom_point(aes(col=RTW, size=Pop)) +
labs(subtitle="GDP per capita Vs FDI Projects",
y="FDI Projects",
x="GDP per capita",
title="Scatterplot",
caption = "Source: Financial Times")
plot(scplot)
The main problems with this data is that the cluster at the far right side of the plot is making it difficult to visualize the rest of the data. If we further investigate the cluster we see that these data points, which show a very high GDP per capita and low population size, belong to Washington-DC. We then proceed to eliminate the outlier (DC) from the data and generate the same plot again:
datapd2 <- read.dta(file.choose())
scplot2 <- ggplot(datapd2, aes(x=GDPpc, y=Projects)) +
geom_point(aes(col=RTW, size=Pop)) +
labs(subtitle="GDP per capita Vs FDI Projects",
y="FDI Projects",
x="GDP per capita",
title="Scatterplot",
caption = "Source: Financial Times")
plot(scplot2)
Now we can better visualize the data and investigate the relationship it illustrates. For instance, at first look, it seems non-RTW states have higher GDP per capita and receive more FDI. It also seems populous states (larger circles) receive more FDI, but not necessarily richer states (higher GDP per capita). We can also employ an encircling techinque from the ggalt package to highlight the data points with high FDI inflows:
library(ggalt)
fdi_select <- datapd2[datapd2$Projects > 100 &
datapd2$GDPpc > 50000 &
datapd2$GDPpc < 65000,]
scplot3 <- ggplot(datapd2, aes(x=GDPpc, y=Projects)) +
geom_point(aes(col=RTW, size=Pop)) +
geom_encircle(aes(x=GDPpc, y=Projects),
data=fdi_select,
color="red",
size=2,
expand=0.08) +
labs(subtitle="GDP per capita Vs FDI Projects",
y="FDI Projects",
x="GDP per capita",
title="Scatterplot",
caption = "Source: Financial Times")
plot(scplot3)
Another interesting plot we can explore is the Diverging Bars, which can handle both negative and positive values. The idea is to see which states are above or below the average number of FDI projects in all states. This means we will have to prepare the data before we run the code for the plot:
datapd2$Projects_z <- round((datapd2$Projects - mean(datapd2$Projects))/sd(datapd2$Projects), 2) # compute normalized FDI Projects
datapd2$ab_bel <- ifelse(datapd2$Projects_z< 0, "below", "above") # above / below avg FDI
datapd2 <- datapd2[order(datapd2$Projects_z), ] # sort
And then we can obtain the diverging bars as follows:
ggplot(datapd2, aes(x=State, y=Projects_z, label=Projects_z)) +
geom_bar(stat='identity', aes(fill=datapd2$ab_bel), width=.5) +
scale_fill_manual(name="FDI Projects",
labels = c("Above Average", "Below Average"),
values = c("above"="#00ba38", "below"="#f8766d")) +
labs(subtitle="Normalised FDI Projects",
title= "Diverging Bars") +
coord_flip()
As expected, we see that the large states of California, Texas, and New York receive, by far, the largest amount of investment in the US. In order to better see the ranking of the states that receive the most FDI, we can also order the data and plot an ordered bar chart:
mean_state <- aggregate(datapd2$Projects, by=list(datapd2$State), FUN=mean) # aggregate by year
mean_state <- mean_state[order(mean_state$x), ] # sort
mean_state$Group.1 <- factor(mean_state$Group.1, levels = mean_state$Group.1) # to retain the order in plot.
ggplot(mean_state, aes(x=Group.1, y=x)) +
geom_bar(stat='identity', width=.5, fill="tomato3") +
labs(title="Ordered Bar Chart",
subtitle="State's Avg. FDI Projects",
y="FDI Projects",
x="States",
caption="source: Financial Times") +
theme(axis.text.x = element_text(angle=65, vjust=0.6))
This plot clearly illustrates the gradual increase in FDI inflows between the states and the difference between them and the three most important FDI destinations in the country; California, Texas, and New York.
Going back to our original investigation, we can generate a time-series plot differentiating between RTW and non-RTW states to see their differences over time. Since this is actually panel data, we can aggregate FDI by RTW and year before generating the plot:
rtw <- aggregate(datapd2$Projects, by=list(datapd2$RTW, datapd2$Year), FUN=mean) # aggregate by RTW and year
ggplot(rtw, aes(x=Group.2, y=x)) +
geom_line(aes(color = as.factor(Group.1)), size = 1) +
scale_color_manual(values = c("#00AFBB", "#E7B800")) +
labs(title="Time Series Plot",
subtitle="FDI by RTW States",
y="FDI Projects",
x="Year",
color = "RTW",
caption="source: Financial Times") +
theme_minimal()
The plot suggests non-RTW states have increased their share of FDI since 2003. Ee can also see the same information as an area plot instead:
ggplot(rtw, aes(x=Group.2, y=x)) +
geom_area(aes(fill = as.factor(Group.1)),
alpha = 0.5) +
scale_color_manual(values = c("#00AFBB", "#E7B800")) +
scale_fill_manual(values = c("#00AFBB", "#E7B800")) +
labs(title="Stacked Area Chart",
subtitle="FDI by RTW States",
y="FDI Projects",
x="Year",
fill = "RTW",
caption="source: Financial Times")
The area plot better illustrates the increasing difference between non-RTW and RTW states. Overall, the plots suggest non-RTW states attract more FDI, which goes against the expectation that the more business friendly RTW states would be more attractive to investment.
And lastly, as the pièce de résistance, we can run an animation with gganimate to see the “explosion” in FDI inflows into California, New York, and Texas over time. In this case, we use the USPS code to identify (label) the states in the plot using the option geom_text:
mapping <- aes(x =GDPpc, y = Projects,
frame = Year)
ggplot(datapd2, mapping = mapping) +
geom_point() +
theme_linedraw() +
labs(title = 'Year: {frame_time}', x = 'GDP per capita', y = 'FDI Projects') +
geom_text(aes(label=USPS,vjust=0)) +
transition_time(Year, range = c(2003,2016)) +
ease_aes('linear')
This animation clearly illustrates the increasing difference between these 3 states and the rest since 2003.
We can also remove the USPS code and add a size distinction based on population and a color distinction between RTW and non-RTW state to see the differences evolving over time since 2003
mapping <- aes(x =GDPpc, y = Projects,
size = Pop, color = RTW,
frame = Year)
ggplot(datapd2, mapping = mapping) +
geom_point() +
theme_linedraw() +
labs(title = 'Year: {frame_time}', x = 'GDP per capita', y = 'FDI Projects') +
geom_text(aes(label=USPS,vjust=0)) +
transition_time(Year) +
ease_aes('linear')
Here the animation is not quite as helpful as there does not seem to have a clear difference over time between the 2 groups.
We can try splitting the groups in different side-by-side plots instead with the facet_wrap option:
ggplot(datapd2, mapping = mapping) +
geom_point() +
theme_linedraw() +
facet_wrap(~RTW) +
labs(title = 'Year: {frame_time}', x = 'GDP per capita', y = 'FDI Projects') +
geom_text(aes(label=USPS,vjust=0)) +
transition_time(Year, range = c(2003,2016)) +
ease_aes('linear')
