A frequent tradeoff can be thought of as truthfulness vs. simplicity.
In other words, you will need to balance:
If you drop outliers, for example, your chart’s readability will almost surely improve. But it could be less truthful.
If you label a subset of your observations, then arguably some information “is lost”, unless you post your data.
library(tidyverse)
# Get fiscal data
imf <- read_csv("data_macro/imf-fiscalmonitor-apr2023.csv")
# Reformat data to wide and keep only the latest year
imf_wide2022 <- imf %>% select(variable,country,`2022`) %>%
pivot_wider(names_from = variable, values_from = `2022`)
# Merge in inflation data:
inf <- read_csv("data_macro/inflation_WDI.csv")
inf2022 <- inf %>% filter(year==2022)
econ2022 <- left_join(imf_wide2022,inf2022,by="country")You may agree that you should almost ever make graphs that look like this:
When you can make charts like this instead:
The code for both charts is shown below. Notice something?
# Chart 1 (basic version; the draft 0)
econ2022 %>%
filter(country != "Norway") %>%
ggplot(aes(y=inflation,x=`Overall Balance`,label=country)) +
geom_point() +
geom_text()
# Chart 2 (the cleaner version)
econ2022 %>%
filter(country != "Norway") %>%
ggplot(aes(y=inflation,
x=`Overall Balance`)) +
geom_point() +
ggrepel::geom_text_repel(data=econ2022 %>%
filter(country %in% c("Italy",
"Sweden",
"United States")),
aes(label=country),nudge_x=.25,nudge_y=1) +
labs(title = "Economic situation in 2022",
x="Budget balance (% of GDP)", y= "Inflation (%)") +
theme_classic() +
geom_hline(yintercept = 0,linetype=2,color="grey40") +
geom_vline(xintercept = 0,linetype=2,color="grey40")It turns out that whoever the wrote the code decided to just drop a row (and not display Norway). I that OK?
econ2022 %>% select(`Overall Balance`,country) %>%
arrange(desc(`Overall Balance`))# A tibble: 36 × 2
`Overall Balance` country
<dbl> <chr>
1 22.4 Norway
2 2.48 Denmark
3 2.26 Cyprus
4 1.20 Ireland
5 0.670 Sweden
6 0.441 Singapore
7 0.206 Switzerland
8 0.134 Israel
9 -0.242 Luxembourg
10 -0.700 Canada
# ℹ 26 more rowsStarting in 2021, inflation increased in many countries and became a source of serious concern for citizens and politicians. One set of substantive debates dealt with this set of questions: should governments be blamed for excessive spending (and borrowing)? Were fiscal decisions responsible for inflation? Here, we’ll deal with one potential approach to designing visual exhibits which might faciliate some international comparisons.1
Let’s get some data (available via the Github repo)
library(tidyverse)
imf <- read_csv("https://raw.githubusercontent.com/zilinskyjan/DataViz/main/data_macro/imf-fiscalmonitor-apr2023.csv")The following variables are available, all expressed as percentages of GDP:
unique(imf$variable)[1] "General Government Expenditure" "Overall Balance"
[3] "Gross public debt" Let’s view the first 3 rows to get a sense of the structure:
head(imf,3)# A tibble: 3 × 12
variable country `2014` `2015` `2016` `2017` `2018` `2019` `2020` `2021`
<chr> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 General Gover… Austra… 36.9 37.4 37.4 36.9 37.0 39.1 44.6 42.2
2 General Gover… Austria 52.3 51.0 50.1 49.3 48.8 48.6 56.7 56.0
3 General Gover… Belgium 55.6 53.7 53.1 52.0 52.3 51.9 58.9 55.5
# ℹ 2 more variables: `2022` <dbl>, `2023` <dbl>If we wanted each row to correspond to one country-year observation, we would run this snippet:
imf %>% pivot_longer(cols=`2014`:`2023`,
names_to = 'year') # this part is not necessary but it's useful# A tibble: 1,080 × 4
variable country year value
<chr> <chr> <chr> <dbl>
1 General Government Expenditure Australia 2014 36.9
2 General Government Expenditure Australia 2015 37.4
3 General Government Expenditure Australia 2016 37.4
4 General Government Expenditure Australia 2017 36.9
5 General Government Expenditure Australia 2018 37.0
6 General Government Expenditure Australia 2019 39.1
7 General Government Expenditure Australia 2020 44.6
8 General Government Expenditure Australia 2021 42.2
9 General Government Expenditure Australia 2022 38.4
10 General Government Expenditure Australia 2023 39.0
# ℹ 1,070 more rowsBut we’ll focus here on year 2022, so let’s simply create 3 informative columns:
# Reformat data to wide and keep only the latest year
imf_wide2022 <- imf %>% select(variable,country,`2022`) %>%
pivot_wider(names_from = variable, values_from = `2022`)imf_wide2022# A tibble: 36 × 4
country General Government Expendit…¹ `Overall Balance` `Gross public debt`
<chr> <dbl> <dbl> <dbl>
1 Australia 38.4 -3.30 55.7
2 Austria 52.4 -3.31 77.8
3 Belgium 53.7 -4.32 105.
4 Canada 41.5 -0.700 107.
5 Croatia 45.7 -0.943 67.5
6 Cyprus 39.9 2.26 86.5
7 Czechia 44.8 -3.59 42.3
8 Denmark 49.2 2.48 29.7
9 Estonia 40.2 -1.15 17.2
10 Finland 54.0 -1.86 74.8
# ℹ 26 more rows
# ℹ abbreviated name: ¹`General Government Expenditure`Finally, we want to add inflation data to our fiscal data:
# Merge in inflation data:
inf <- read_csv("data_macro/inflation_WDI.csv")
inf2022 <- inf %>% filter(year==2022)
econ2022 <- left_join(imf_wide2022,inf2022,by="country")So we will work with the econ2022 data object for a moment.
This chart displays the relationship between public budget deficits and inflation.
# Make a simple plot:
econ2022 %>%
filter(country != "Norway") %>%
ggplot(aes(y=inflation,
x=`Overall Balance`,
label=country)) +
geom_point() +
geom_text() +
labs(title = "Economic situation in 2022",
x="Overall budget balance (% of GDP)", y= "Inflation (%)")
A few things to notice:
geom_text() which uses what we placed inside aes(... label=country).The same data can be displayed this way; ggrepel::geom_text_repel() is helpful in this context:
econ2022 %>%
filter(country != "Norway") %>%
ggplot(aes(y=inflation,
x=`Overall Balance`,
label=country)) +
geom_point() +
ggrepel::geom_text_repel() +
labs(title = "Economic situation in 2022",
x="Overall budget balance (% of GDP)", y= "Inflation (%)")
Or you can highlight a subset of subset of observations relevant for your analysis.
Let’s create a vector of country names:
subset <- c("Italy","Sweden","United States")We’ll want to insert data=econ2022 %>% filter(country %in% subset) into geom_text_repel:
scatter1 <- econ2022 %>%
filter(country != "Norway") %>%
ggplot(aes(y=inflation,
x=`Overall Balance`)) +
geom_point() +
ggrepel::geom_text_repel(data=econ2022 %>%
filter(country %in% subset),
aes(label=country),nudge_x=.75,nudge_y=.75) +
labs(title = "Economic situation in 2022",
x="Overall budget balance (% of GDP)", y= "Inflation (%)") +
theme_classic()
scatter1
Adding a layer and selecting attributes for a specific subset of points is a general, more widely applicable, approach:
scatter1 +
geom_point(data=econ2022 %>% filter(country %in% subset),
color="red")
Note also that I also that it could have been tempting to label the x-axis as showing “Public deficit” because we almost always talk about deficits. But that would have been misleading, given that only negative values on the x-axis would have denoted the deficit.
We see that deficit spending is not informative: moderately high inflation was common across OECD countries; neither large deficits, nor budget surpluses, were prognostic of better/worse outcomes.
What about total government spending? We can check:
econ2022 %>%
ggplot(aes(y=inflation,
x=`General Government Expenditure`)) +
geom_point() +
ggrepel::geom_text_repel(data=econ2022 %>%
filter(country %in% c("Italy",
"Sweden",
"United States")),
aes(label=country),nudge_y=.5,nudge_x=-.5) +
labs(title = "Economic situation in 2022",
x="Public spending (% of GDP)", y= "Inflation (%)") 
This would seem to suggest that higher (government) spending is not necessarily associate with faster price growth.
To be sure, a more careful analysis here would require looking it changes in government expenditures.
Here a few ways to show how sharply inflation increased between 2021 in 2022 in most OECD countries:
inf %>% filter(iso3c %in% econ2022$iso3c) %>%
filter(year>=2021) %>%
mutate(avg = mean(inflation), .by = country) %>%
ggplot(aes(x=inflation,
color= factor(year),
y=fct_reorder(country,avg))) +
geom_point() +
ggrepel::geom_text_repel(aes(label=round(inflation,1)),
show.legend = FALSE) +
scale_color_brewer(palette = 2,type = "qual") +
labs(x="Inflation (%)",y="",color="Year")
inf %>% filter(iso3c %in% econ2022$iso3c) %>%
filter(year>=2021) %>%
mutate(avg = mean(inflation), .by = country) %>%
ggplot(aes(x=inflation,
color= factor(year),
y=fct_reorder(country,avg))) +
geom_line(aes(group=country),color="grey50") +
geom_point(size=2) +
scale_color_brewer(palette = 2,type = "qual") +
labs(x="Inflation (%)",y="",color="Year")
Above we simply added a geom_line(aes(group=country),color="grey50") and made sure that the line was plotted before the points were added for each country (for aesthetic reasons).
Or you could make a bar chart:
inf %>% filter(iso3c %in% econ2022$iso3c) %>%
filter(year>=2021) %>%
mutate(avg = mean(inflation), .by = country) %>%
ggplot(aes(x=inflation,
fill= factor(year),
y=fct_reorder(country,avg))) +
geom_col(position = position_dodge()) +
scale_fill_brewer(palette = 3,type = "qual") +
labs(x="Inflation (%)",y="",fill="Year",title = "Inflation in OECD countries")
Note that we:
color to fill within aes()labs()scale_color_brewer to scale_fill_brewergeom_point() with geom_col(position = position_dodge())Use the provided datasets to explore whether an increase in government spending predicts (current or future) inflation.
Consider the following issues and provide brief justifications:
But also remember that evidence of this kind can inform factual debates but it wouldn’t settle the debate.↩︎