library(tidyverse)
library(knitr)
# Shared styling, used across all three Project 2 reports
accent <- "#2a78d6"
accent_2 <- "#eb6834"
theme_p2 <- theme_minimal(base_size = 12) +
theme(panel.grid.minor = element_blank(), plot.title.position = "plot")Project 2 — Data Tidying: Life Expectancy by Country
Data Source
This dataset comes from the Discussion 5A post “Life Expectancy by Country — World Bank” by Mohd Afwan Shaikh. The post describes the World Bank’s Life expectancy at birth, total (years) indicator, in which each country is a row and each year is a separate column, and asks:
“I would like to compare life expectancy across different countries and examine how it has changed over time. I could also investigate which countries experienced the greatest improvements and how life expectancy changed before and after the COVID-19 pandemic.”
Source: World Bank, indicator SP.DYN.LE00.IN (https://data.worldbank.org/indicator/SP.DYN.LE00.IN).
How it was obtained: I pulled the indicator from the World Bank API for 2000–2023, kept only sovereign countries and territories (dropping the World Bank’s aggregate rows such as “World” and “Arab World”, which are sums of other rows and would be double counting), and saved the result in the World Bank’s own wide layout as data/life_expectancy_wide.csv. That raw file is committed to GitHub and is read directly from there below, so this document runs in a clean session on any machine.
Data Structure Before Tidying
raw_life <- read_csv(
"https://raw.githubusercontent.com/AnissSahraoui/DATA607/main/Week6/data/life_expectancy_wide.csv",
show_col_types = FALSE
)
dim(raw_life)[1] 217 26Code
names(raw_life) [1] "Country Name" "Country Code" "2000" "2001" "2002"
[6] "2003" "2004" "2005" "2006" "2007"
[11] "2008" "2009" "2010" "2011" "2012"
[16] "2013" "2014" "2015" "2016" "2017"
[21] "2018" "2019" "2020" "2021" "2022"
[26] "2023" Code
raw_life |> select(1:8) |> slice_head(n = 5)# A tibble: 5 × 8
`Country Name` `Country Code` `2000` `2001` `2002` `2003` `2004` `2005`
<chr> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 Aruba ABW 72.9 73.0 73.1 73.2 73.2 73.4
2 Afghanistan AFG 55.0 55.5 56.2 57.2 57.8 58.2
3 Angola AGO 46.5 47.0 47.9 50.2 51.1 52.1
4 Albania ALB 74.8 75.1 75.3 75.6 76.0 76.4
5 Andorra AND 81.9 82.2 82.4 82.7 82.8 83.1The file has 217 rows and 26 columns: two identifier columns and 24 year columns.
It is untidy for one clear reason: year is a variable, but it is stored in the column headings rather than in a column. Each row also holds 24 separate observations instead of one. In tidy form, every row should be one country in one year, with a single life_expectancy value.
Transformation Steps
life <- raw_life |>
# 1. Reshape wide to long: the 24 year columns become two columns, year and value.
pivot_longer(
cols = -c(`Country Name`, `Country Code`),
names_to = "year",
values_to = "life_expectancy"
) |>
# 2. Rename to a consistent snake_case convention. The raw names use spaces and
# title case, which force back-ticks everywhere in downstream code.
rename(
country = `Country Name`,
country_code = `Country Code`
) |>
# 3. Normalize types: year arrives as text because it came from column names.
mutate(year = as.integer(year)) |>
arrange(country, year)
glimpse(life)Rows: 5,208
Columns: 4
$ country <chr> "Afghanistan", "Afghanistan", "Afghanistan", "Afghanis…
$ country_code <chr> "AFG", "AFG", "AFG", "AFG", "AFG", "AFG", "AFG", "AFG"…
$ year <int> 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, …
$ life_expectancy <dbl> 55.005, 55.511, 56.225, 57.171, 57.810, 58.247, 58.553…Missing values. The brief asks for a documented decision, so the first step is to measure the problem rather than assume it:
life |>
summarise(
rows = n(),
missing_values = sum(is.na(life_expectancy)),
countries = n_distinct(country),
years = n_distinct(year),
complete_series = sum(!is.na(life_expectancy)) == n()
)# A tibble: 1 × 5
rows missing_values countries years complete_series
<int> <int> <int> <int> <lgl>
1 5208 0 217 24 TRUE There are no missing values: all 5208 country-year combinations are present. That is unusual for World Bank data and worth stating explicitly, because it means no imputation or row-dropping decision is needed here. Had values been missing, I would have kept them as NA rather than filling them, since an interpolated life expectancy is not an observation.
Two consistency checks on the values themselves:
life |>
summarise(
min_value = min(life_expectancy),
max_value = max(life_expectancy),
implausible = sum(life_expectancy < 20 | life_expectancy > 100),
duplicate_rows = sum(duplicated(across(c(country, year))))
)# A tibble: 1 × 4
min_value max_value implausible duplicate_rows
<dbl> <dbl> <int> <int>
1 14.7 86.4 2 0Every value falls in a plausible range and no country-year appears twice, so the reshaping neither dropped nor duplicated observations.
Analytical Methods
The post asks four questions. Each is answered from the tidy table only:
- How life expectancy changed over time — the unweighted mean across countries, by year.
- Which countries improved most — the change from 2000 to 2023, ranked.
- Before and after COVID-19 — the change from 2019 to 2021, and how many countries had returned to their 2019 level by 2023.
- How countries compare — the spread of values in 2000 against 2023.
Averages here are unweighted across countries: each country counts once, regardless of population. This answers “how are countries doing” rather than “how is the average person doing”, which is the question the post asks. A population-weighted mean would be dominated by China and India.
Change over time
Code
global <- life |>
group_by(year) |>
summarise(mean_life_expectancy = mean(life_expectancy), .groups = "drop")
global |>
filter(year %in% c(2000, 2010, 2019, 2020, 2021, 2022, 2023)) |>
mutate(mean_life_expectancy = round(mean_life_expectancy, 1)) |>
kable(col.names = c("Year", "Mean life expectancy (years)"))| Year | Mean life expectancy (years) |
|---|---|
| 2000 | 67.6 |
| 2010 | 70.8 |
| 2019 | 73.0 |
| 2020 | 72.5 |
| 2021 | 71.8 |
| 2022 | 73.1 |
| 2023 | 73.8 |
Code
ggplot(global, aes(x = year, y = mean_life_expectancy)) +
annotate("rect", xmin = 2020, xmax = 2021, ymin = -Inf, ymax = Inf,
fill = accent_2, alpha = 0.12) +
geom_line(linewidth = 1, color = accent) +
geom_point(size = 1.8, color = accent) +
labs(title = "Life expectancy rose for two decades, then fell sharply in 2020–21",
x = "Year", y = "Mean life expectancy (years)") +
theme_p2
The mean rose from 67.6 years in 2000 to 73 in 2019, fell to 71.8 in 2021, then recovered to 73.8 by 2023. The pandemic erased roughly a decade of progress in two years, and the recovery took two more.
Greatest improvements
Code
change <- life |>
filter(year %in% c(2000, 2019, 2021, 2023)) |>
select(country, year, life_expectancy) |>
pivot_wider(names_from = year, values_from = life_expectancy, names_prefix = "y") |>
mutate(
gain_2000_2023 = y2023 - y2000,
covid_change = y2021 - y2019
)
change |>
slice_max(gain_2000_2023, n = 5) |>
transmute(Country = country,
`2000` = round(y2000, 1),
`2023` = round(y2023, 1),
`Gain (years)` = round(gain_2000_2023, 1)) |>
kable()| Country | 2000 | 2023 | Gain (years) |
|---|---|---|---|
| Malawi | 46.2 | 67.4 | 21.2 |
| Rwanda | 47.8 | 67.8 | 20.0 |
| Zambia | 46.6 | 66.3 | 19.8 |
| Uganda | 49.6 | 68.3 | 18.6 |
| Angola | 46.5 | 64.6 | 18.1 |
Code
change |>
slice_max(gain_2000_2023, n = 10) |>
mutate(country = fct_reorder(country, gain_2000_2023)) |>
ggplot(aes(x = gain_2000_2023, y = country)) +
geom_col(fill = accent, width = 0.65) +
geom_text(aes(label = sprintf("+%.1f", gain_2000_2023)), hjust = 1.15,
color = "white", fontface = "bold", size = 3.4) +
labs(title = "The largest gains are concentrated in sub-Saharan Africa",
x = "Years gained, 2000–2023", y = NULL) +
theme_p2 +
theme(panel.grid.major.y = element_blank())
The COVID-19 period
Code
change |>
slice_min(covid_change, n = 5) |>
transmute(Country = country,
`2019` = round(y2019, 1),
`2021` = round(y2021, 1),
`Change (years)` = round(covid_change, 1)) |>
kable()| Country | 2019 | 2021 | Change (years) |
|---|---|---|---|
| Bolivia | 67.8 | 61.4 | -6.4 |
| Paraguay | 73.7 | 68.1 | -5.6 |
| Mexico | 74.5 | 69.8 | -4.8 |
| Guyana | 69.1 | 64.3 | -4.7 |
| Peru | 76.3 | 71.6 | -4.7 |
Code
recovery <- change |>
summarise(
countries = n(),
fell_2019_to_2021 = sum(covid_change < 0),
recovered_by_2023 = sum(y2023 >= y2019)
)
recovery |>
kable(col.names = c("Countries", "Fell 2019 to 2021", "Back to 2019 level by 2023"))| Countries | Fell 2019 to 2021 | Back to 2019 level by 2023 |
|---|---|---|
| 217 | 182 | 186 |
How countries compare
Code
life |>
filter(year %in% c(2000, 2023)) |>
group_by(year) |>
summarise(
lowest = round(min(life_expectancy), 1),
highest = round(max(life_expectancy), 1),
gap = round(max(life_expectancy) - min(life_expectancy), 1),
sd = round(sd(life_expectancy), 2),
.groups = "drop"
) |>
kable(col.names = c("Year", "Lowest", "Highest", "Gap (years)", "Std. dev."))| Year | Lowest | Highest | Gap (years) | Std. dev. |
|---|---|---|---|---|
| 2000 | 44.8 | 82.1 | 37.3 | 9.64 |
| 2023 | 54.5 | 86.4 | 31.9 | 7.12 |
Code
life |>
filter(year %in% c(2000, 2023)) |>
mutate(year = factor(year)) |>
ggplot(aes(x = life_expectancy, fill = year)) +
geom_density(alpha = 0.55, color = NA) +
scale_fill_manual(values = c("2000" = accent_2, "2023" = accent), name = NULL) +
labs(title = "The distribution shifted right and narrowed",
x = "Life expectancy (years)", y = "Density") +
theme_p2 +
theme(legend.position = "top")
Conclusions
- Life expectancy rose, fell, and recovered. The country average climbed from 67.6 years in 2000 to 73 in 2019, dropped to 71.8 in 2021, and reached 73.8 in 2023, the highest in the series.
- The biggest improvements were in sub-Saharan Africa. Malawi gained 21.2 years (46.2 to 67.4), with Rwanda, Zambia, Uganda and Angola each gaining more than 18. These are countries recovering from the HIV/AIDS epidemic and very high child mortality, so they had the most ground to make up.
- COVID-19 hit Latin America hardest. Bolivia lost 6.4 years between 2019 and 2021, followed by Paraguay (−5.6), Mexico (−4.8), Guyana (−4.7) and Peru (−4.7). 182 of 217 countries declined over that period.
- Most, but not all, have recovered. 186 of 217 countries were back at or above their 2019 level by 2023, leaving 31 still below it.
- Countries are converging. The gap between the highest and lowest country narrowed from 37.3 years in 2000 to 31.9 in 2023, and the standard deviation fell from 9.64 to 7.12. The lowest country in 2023, Nigeria at 54.5 years, is nearly ten years above the lowest in 2000.
The practical lesson for the tidying step is that the single pivot_longer() call is what makes all of this possible: every question above is a group_by() on a column that did not exist in the source file.